Journal of Pathology

J Pathol 2019; 247: 563–573 INVITED REVIEW

Published online 20 February 2019 in Wiley Online Library
(wileyonlinelibrary.com) DOI: 10.1002/path.5230

Molecular profiling in muscle-invasive bladder cancer: more than

the sum of its parts
Gottfrid Sjödahl1,2* , Chelsea L Jackson3,4 , John MS Bartlett4,5 , D Robert Siemens3,6 and David M Berman3,4*
1 Division of Urological Research, Department of Translational Medicine, Lund University, Lund, Sweden
2 Department of Urology, Skåne University Hospital, Malmö, Sweden
3 Division of Cancer Biology & Genetics, Cancer Research Institute, Queen’s University, Kingston, Ontario, Canada
4 Department of Pathology & Molecular Medicine, Queen’s University, Kingston, Ontario, Canada
5 Diagnostic Development Program, Ontario Institute for Cancer Research, Toronto, Ontario, Canada
6 Department of Urology, Queen’s University, Kingston, Ontario, Canada

*Correspondence to: Gottfrid Sjödahl, Division of Urological Research, Department of Translational Medicine, Lund University, Lund 22100, Sweden
and Department of Urology, Skåne University Hospital, Malmö, Sweden. E-mail: gottfrid.sjodahl@med.lu.se;
Or David Berman, Division of Cancer Biology & Genetics and Department of Pathology & Molecular Medicine, Cancer Research Institute, Queen’s
University, Kingston, Ontario K7L 3N6, Canada. E-mail: bermand@queensu.ca

Abstract
Bladder cancers are biologically and clinically heterogeneous. Recent large-scale transcriptomic profiling studies
focusing on life-threatening muscle-invasive cases have demonstrated a small number of molecularly distinct
clusters that largely explain their heterogeneity. Similar to breast cancer, these clusters reflect intrinsic urothelial
cell-type differentiation programs, including those with luminal and basal cell characteristics. Also like breast
cancer, each cell-based subtype demonstrates a distinct profile with regard to its prognosis and its expression
of therapeutic targets. Indeed, a number of studies suggest subtype-specific differential responses to cytotoxic
chemotherapy and to therapies that inhibit a number of targets, including growth factors (EGFR, ERBB2, FGFR)
and immune checkpoint (PD1, PDL1) inhibitors. Despite burgeoning evidence for important clinical implications,
subtyping has yet to enter into routine clinical practice. Here we review the conceptual basis for intrinsic cell
subtyping in muscle-invasive bladder cancer and discuss evidence behind proposed clinical uses for subtyping as a
prognostic or predictive test. In deliberating barriers to clinical implementation, we review pitfalls associated with
transcriptomic profiling and illustrate a simple immunohistochemistry (IHC)-based subtyping algorithm that may
serve as a faster, less expensive alternative. Envisioned as a research tool that can easily be translated into routine
pathology workflow, IHC-based profiling has the potential to more rapidly establish the utility (or lack thereof) of
cell type profiling in clinical practice.
Copyright © 2019 Pathological Society of Great Britain and Ireland. Published by John Wiley & Sons, Ltd.

Keywords: bladder neoplasms; Shh; PPARG; TGFB; cancer stem cell; LUNDTAX; GATA3; KRT5; TP53; RB1

Received 1 October 2018; Revised 17 December 2018; Accepted 28 December 2018

Conflict of interest statement: Dr Berman has served as a paid scientific advisor to Janssen Pharmaceuticals, which manufactures erdafitinib.

Introduction tumor (TURB-T) specimens represents an optimal

approach to understanding bladder tumor biology, and
Urothelial carcinoma (UC) derives from special- the availability of such samples prior to radical surgery
ized epithelia lining the bladder wall. Using a allows molecular profiling to potentially inform patient
well-established differentiation program, basal stem decisions related to both surgery and neoadjuvant treat-
cells at the stromal interface self-renew and generate ment. As discussed below, the differentiation programs
intermediate and superficial/umbrella cells to maintain and molecular subtypes that operate in invasive bladder
and regenerate normal urothelium as needed. Most tumors relate back to the differentiation states of the
urothelial cancers do not have the ability to invade urothelium, but they also relate forward to pan-cancer
into the submucosal stroma (lamina propria) or bladder cellular and stromal phenotypes. Thus, molecular sub-
wall (muscularis propria), but those that do can display types serve as a context that links the biology of tumor
remarkable diversity in histologic variants, clinical out- development to the potential to influence and stratify
comes, and molecular profiles. Prediction of response patients to improve oncological outcomes. We limit this
to available peri-operative treatment, and the devel- review to describe the state of molecular profiling in
opment of novel means of targeting invasive bladder cancers that invade the muscular bladder wall, com-
cancers, are areas of intense research, where molecular monly referred to as muscle-invasive bladder cancers.
profiling is postulated to be of value. Molecular analysis This focus reflects the scarcity of molecular subtype
of the diagnostic trans-urethral resection of bladder studies identified in more superficial bladder cancers,
Copyright © 2019 Pathological Society of Great Britain and Ireland. J Pathol 2019; 247: 563–573
Published by John Wiley & Sons, Ltd. www.pathsoc.org www.thejournalofpathology.com
564 G Sjödahl et al

a relatively understudied topic. With more studies on contrast, is more frequently employed, despite relatively
non-muscle-invasive bladder cancer on the horizon, weaker evidence of benefit [10].
patient stratification for intravesical therapy will be a
pressing unresolved issue. The target audience for this
Metastatic
review is twofold. For the healthcare professional who Despite best efforts at local control, more than 50%
diagnoses or treats bladder cancer, we aim to provide a of patients with MIBC experience and ultimately die
logical structure of molecular subtypes, an enumeration of distant metastases. Furthermore, 18–25% of cys-
of promising targeted therapies, and a vision of how tectomy patients with muscle-invasive bladder cancer
molecular subtypes could be implemented in routine already have metastatic disease at the time of surgery
pathology. We also aim to provide basic researchers in [11]. Thus, there is a sizeable proportion of patients
molecular pathology with an up-to-date view on current with urgent need for effective systemic therapy. As
biomarkers, and insights into the complex interpretation in the peri-operative setting, guidelines recommend
of molecular subtypes in tissue samples with the cellular cisplatin-based combination chemotherapy for fit
diversity of invasive bladder cancers. To provide clin- patients with metastatic bladder cancer, although only
ical and historical context, we also very briefly review approximately 15% of patients will achieve long-term
the current treatment for the superficial, invasive, and survival greater than 2 years. Beyond chemotherapy,
metastatic stages of bladder cancer, and the two major recent modest successes in immunotherapy trials
pathways of tumor development. with checkpoint inhibitors represent the first significant
improvements in treating metastatic UC in decades [12].
In clinical practice, prognosis and treatment decisions
are driven by histopathologic evaluation of stage and
Stages of urothelial carcinoma
grade as well as clinical features including adverse
imaging findings. There are currently no molecular
Non-muscle-invasive classification strategies in routine use for UC patients,
Urologists typically use the term ‘superficial’ or but biomarker discovery research in chemotherapy,
‘non-muscle-invasive’ bladder cancers (NMIBCs) immunotherapy, and targeted therapy trials is ongoing.
to describe non-invasive UCs and their minimally Early results suggest that molecular profiling will be
invasive counterparts that extend only into the lamina useful for patient stratification, as discussed below.
propria. Superficial cases constitute more than 75% This optimism stems, in part, from other cancer types,
of incident bladder cancer cases. Uncommonly, they where expression, mutation, and/or activation of drug
present as pure flat carcinoma in situ (CIS), which car- targets (e.g. EGFR mutations) in the patient’s tumor is
ries a high risk of progression. Most often, they present used to guide treatment and monitor resistance mech-
as non-invasive papillary UC, which carries a high risk anisms (e.g. KRAS mutations). This molecular informa-
of recurrence but a relatively low risk of progression tion can be paired with information about tumor subtype,
to life-threatening muscle-invasive (MIBC) disease which may provide additional evidence of benefit. While
[1–3]. For high-risk NMIBC, intravesical instillations profiling and classification of NMIBC have been slow,
of bacillus Calmette-Guérin (BCG) effectively reduce recent studies in MIBC have yielded a treasure trove
the risk for progression by stimulating an anti-tumor of information regarding genomic and transcriptomic
immune response [4]. Adverse prognostic factors in subtypes. Consensus between classification schemes has
NMIBC include high pathological grade, large tumor gradually emerged and excitement is building around
size, multifocality, and the presence of CIS [1,3]. implementing these discoveries to benefit patients. Yet
significant challenges remain, particularly with regard
to devising a clinically relevant, biologically valid, and
Muscle-invasive practical rubric for testing patients. Recent work [13,14]
Approximately 15% of non-muscle-invasive cases has highlighted pitfalls in transcriptomic profiling and
evolve into more aggressive and invasive forms that the importance of immunohistochemistry (IHC) assays
extend into the muscular bladder wall, an event that for reconciling how cancer cell-specific gene expression
appears to be required for bladder cancer to acquire relates to RNA-based clustering approaches. Here, we
a lethal phenotype. Bladder cancer is more frequent review the basis, performance, and clinical relevance
in men, and in more developed countries, but local of molecular classification of muscle-invasive UC. We
variation occurs, e.g. due to the high prevalence compare and contrast transcriptomic- and IHC-based
of schistosomiasis in parts of Africa [5]. As specified profiling, and explore the potential for implementing
in clinical guidelines, definitive management of local- these tools in routine clinical practice.
ized MIBC centres on cystectomy and peri-operative
platinum-based chemotherapy regimens. However,
the application of these guidelines and treatment out- Bladder cancer development, from papillary
comes vary widely, from patient to patient and from and flat pathways to molecular subtypes
caregiver to caregiver [6,7]. Despite high-level evidence,
pre-operative neoadjuvant chemotherapy (NACT) is not Decades of work elucidated mutually exclusive, paral-
widely utilized [8,9]. Adjuvant chemotherapy (ACT), in lel molecular pathways that distinguish the pathogenesis
Copyright © 2019 Pathological Society of Great Britain and Ireland. J Pathol 2019; 247: 563–573
Published by John Wiley & Sons, Ltd. www.pathsoc.org www.thejournalofpathology.com
Intrinsic subtypes of bladder cancer 565

of papillary UCs from their flat counterparts [15–17]. we will not discuss them further except to state that the
Papillary carcinomas are distinguished by branching precise frequency by which molecular subtypes occur in
structures that extend into the bladder lumen. At the different pathological stages of bladder cancer remains
molecular level, papillary cancers are enriched for acti- unclear.
vating mutations in the fibroblast growth factor recep- The following description of gene expression signa-
tor 3 (FGFR3) oncogene [18,19], whereas cancers that tures and their relationship with molecular subtype clas-
develop through the flat carcinoma in situ pathway lack sification will be based on the data and conclusions
FGFR3 mutations and are instead enriched for inacti- of the main studies on MIBC [25–30,36]. The differ-
vating mutations in the TP53 and retinoblastoma (RB1) ent MIBC classification systems are largely overlapping
tumor suppressor genes [17]. Intriguingly, the minor- [37] and operate at different levels in a hierarchical struc-
ity of papillary UCs that do progress to MIBC show ture [38]. Here, we briefly describe the biological char-
genomic instability and a propensity for homozygous acteristics that shape the hierarchy of bladder cancer
loss of CDKN2A [20,21], which may serve as an alter- molecular subtypes.
nate means to disable TP53 and RB1. The two tumor
suppressor proteins p16INK4a and p14ARF encoded by
CDKN2A regulate functions of RB and TP53, and thus Luminal-like MIBC
encapsulate a wide range of tumor suppressor function At the top hierarchical level, bladder tumors can be
in one genomic event. divided into luminal-like and non-luminal-like classes
In 2009, two laboratories investigating cancer stem based on the presence or absence of a bimodally
cells discovered that, much like stem cells in benign expressed urothelial differentiation signature. About
urothelium, bladder cancer stem cells localize to the half of MIBCs express this signature, which includes
basal compartment at the stromal interface [22,23]. KRT20, UPK1–3, transcriptional regulators, e.g.
These cells sustain long-term tumor growth and exhibit PPARG, GRHL2–3, ELF3, TBX2-3, and pioneer fac-
mRNA expression profiles that are distinct from their tors GATA3 and FOXA1. While many studies have
more differentiated, less tumorigenic progeny. Impor- grouped luminal-like tumors together, The Cancer
tantly, these studies also showed that a shift toward basal Genome Atlas (TCGA) Consortium and a group in
cell gene signatures is associated with higher stage and Lund, Sweden have subdivided this class of MIBCs
decreased survival. The importance of basal cell dif- further. The Lund classification introduces two such
ferentiation in bladder cancer as an indicator of poor sub-classes termed ‘urothelial-like’ (Uro) and ‘genom-
prognosis was soon confirmed in larger-scale profiling ically unstable’ (GU), which differ in prognosis,
studies [24–30]. and tumor-cell intrinsic genomic properties which,
The use of unsupervised clustering to discover basal respectively, include loss of CDKN2A versus RB1,
and luminal UC subtypes echoes seminal work by alterations of FGFR3 versus ERBB2, and maintained
Perou et al [31] and Sørlie et al [32], who identified versus defective cell–cell adhesion. Most importantly,
intrinsic basal and luminal profiles for breast cancer, immunohistochemical studies [39] have shown that
with poor and initially good outcomes, respectively. only the Uro subtype retains stratification of a basal
Luminal breast cancers, however, exhibit later relapses cell-layer, wherein stem cells had previously been
and lethality. Importantly, intrinsic breast cancer sub- shown to reside [22,23]. In contrast, tumors of the GU
types resonated with scientists and clinicians because subtype, while expressing the urothelial-differentiation
they mapped well to an existing paradigm of treat- gene signature, are in fact poorly differentiated, lack cell
ment, which differs between ER-positive (luminal) and polarity, and are uniformly negative for basal-cell mark-
ER-negative (including basal-like cancers) cancers [33]. ers [40]. The TCGA Consortium introduced another
However, clinical profiling in breast cancer currently three-tier sub-classification of luminal-like tumors:
focuses less on molecular subtyping and more on predic- ‘luminal-papillary’ (Lum-Pap), ‘luminal-infiltrated’
tive biomarkers such as ER, progesterone receptor, and (Lum-Inf), and ‘luminal’ (Lum) [29]. The separation
ERBB2 (a.k.a. HER2) and on residual risk of disease between the TCGA sub-classes differs from that of the
relapse and progression. Thus, RNA-based subtyping Lund group in that it is driven by relatively high levels
has not become mandatory in diagnostic or routine clin- of immune- and stroma-related genes in the Lum-Inf
ical use. It has, however, proven scientifically invaluable and Lum sub-classes, respectively, whereas Lum-Pap
and useful for bridging biological and clinical lexicons. tumors are low in both stromal and immune transcripts.
For this reason, there was considerable fanfare when,
a decade later, analogous subtypes were discovered
in UC. Non-luminal-like MIBCs
Non-luminal-like MIBCs are often casually referred
to as ‘basal-like’, a simplification that clearly
Gene expression signatures in the molecular underestimates the variety of MIBC molecular phe-
subtypes of MIBC notypes. About two-thirds of non-luminal-like MIBCs
show strong expression of basal genes including KRT5,
Since a small number of relevant studies [25,34,35] KRT6, KRT14, CD44, CDH3, and EGFR. This molecu-
indicate that nearly all NMIBC tumors are luminal-like, lar subtype shows relatively high levels of both stromal-
Copyright © 2019 Pathological Society of Great Britain and Ireland. J Pathol 2019; 247: 563–573
Published by John Wiley & Sons, Ltd. www.pathsoc.org www.thejournalofpathology.com
566 G Sjödahl et al

and immune-related transcripts and is further charac-

terized by squamous differentiation, and expression of
genes, e.g. S100A7, SPRR1B, and SPRR3, located in
the epidermal differentiation complex on chromosome
1q21 [41]. These genes govern the differentiation and
function of squamous epithelia and are not expressed
in urothelial basal cells, which is why the dual-label
‘basal/squamous’ (Ba/Sq) is preferred. The similarity
between this subtype and benign basal urothelial cells is
limited to positive expression of a handful of markers,
but unlike the Ba/Sq subtype, normal basal cells retain
expression of urothelial differentiation factors, e.g.
PPARG and GATA3. Further support for the squamous
rather than the basal-like nature of this subtype is lent by
its stronger similarity in integrative pan-cancer studies
with squamous cell carcinomas of the head and neck or
lung than with other bladder cancers or basal-like breast
cancer [42].
In addition, several independent groups have identi-
fied additional tumors that are neither luminal-like
nor Ba/Sq. Kardos et al [43] have described a
‘claudin-low’ subset of advanced urothelial carcinomas
with such a profile, and Dadhania et al [13] described
a ‘double-negative’ subtype, simultaneously lacking
both urothelial and squamous differentiation markers.
Classification according to both Lund and TCGA has
identified a subtype with strong neuroendocrine features
that is enriched for cases with small-cell or neuroen-
docrine variant histology [14,29]. The Lund group has
also described a mesenchymal-like subtype in which
expression of epithelial cell adhesion molecules CDH1
and EPCAM is replaced by mesenchymal markers Figure 1. Therapeutic sensitivities associated with molecular sub-
VIM and ZEB2 [14]. Both neuroendocrine-like and types of muscle-invasive or metastatic bladder cancer. See text for
mesenchymal-like MIBCs also fit the bill as ‘double citations to the relevant literature.
negative’ and ‘claudin-low.’ Thus, these subtypes may
encompass rare entities that are not luminal-like, but are cystectomy preceded by platinum-based neoadjuvant
clearly distinct from the Ba/Sq subtype. The frequency chemotherapy (NACT). Indeed, a significant minority
and exact subdivision of these minor subtypes in a study of patients treated this way achieve durable responses
cohort may depend on inclusion criteria for non-UC and increased cancer-specific survival [44]. However,
variant histology. As with histologic variants, powering a meta-analysis has suggested a marginal absolute
clinical trials to link these minor subtypes to clinical survival benefit for NACT of only 5% at 5 years [45].
outcome may prove challenging, and the utility of The marginal benefit is further diminished by concerns
further sub-classifying non-luminal, non-basal subtypes regarding delayed surgery, the non-selective nature of
remains to be proven. treating all MIBC patients with NACT, and its risk of
serious morbidity. Thus, although trending upward,
Therapeutic opportunities for molecular profiling reported utilization rates of NACT for cystectomy
in muscle-invasive urothelial carcinoma patients rarely exceed 25% [46–48].
Biomarker tests that predict chemoresponse could
address these problems by identifying patients most
Recently published data from the TCGA Consortium likely to benefit. Recent efforts have explored the poten-
reported that 69% of bladder cancers contain potentially tial utility of RNA-based subtype classifiers to serve as
actionable therapeutic targets, which associate with spe- biomarkers of NACT response [36,37,49]. Intriguingly,
cific molecular subtypes [29]. In this section, summa- two retrospective studies indicate that NACT may
rized in Figure 1, we describe current efforts towards
preferentially benefit patients with basal-like subtypes
using molecular profiling as a patient stratification
[36,49]. Unlike historical untreated controls who show
method for clinical trials and ultimately clinical practice.
better survival with luminal subtypes, NACT-treated
patients in these studies showed superior survival rates
Peri-operative cytotoxic chemotherapy in basal-like cancers, although the survival benefit was
The gold-standard management for MIBC patients not reflected in the proportions achieving pathological
with disease confined to the pelvis includes radical downstaging. Overall, data to date provide support
Copyright © 2019 Pathological Society of Great Britain and Ireland. J Pathol 2019; 247: 563–573
Published by John Wiley & Sons, Ltd. www.pathsoc.org www.thejournalofpathology.com
Intrinsic subtypes of bladder cancer 567

for prospective clinical trials to confirm the utility of and the ERBB3 receptors also show subtype-specific
subtyping in choosing patients for NACT. Since the overexpression; ERBB3, like HER2, is highest in
biologic basis of response to chemotherapy could be luminal-like (Uro and, particularly, GU luminal-like
similar in the peri-operative (neoadjuvant and adjuvant) subtypes), whereas EGFR is highest in basal-squamous
setting and at progression to metastatic disease, addi- subtypes [41,57].
tional studies should explore applications of cell-based
subtyping at later stages in cancer progression. RXR-α and PPAR-γ
The retinoid X receptor alpha (RXR-α) and peroxi-
FGFR3 some proliferator-activated receptor gamma (PPAR-γ)
Mutations, gene fusions, and amplification of the gene are central for urothelial differentiation [61]. Genetic
encoding fibroblast growth factor receptor 3 (FGFR3) alterations in RXRA and PPARG have been reported in
are present in approximately 50–80% of superficial up to 40% of MIBC patients [29], supporting a role
bladder cancers and up to 20% of muscle-invasive in bladder carcinogenesis. Furthermore, some studies
bladder cancers [29,50,51]. Early phase clinical trials have shown associations between the use of pioglita-
of FGFR inhibitors such as BGJ398 and erdafitinib zone, a PPAR-γ agonist used in treating diabetes, and
have reported activity in metastatic post-platinum UC an increased risk of bladder cancer [62]. Functionally,
patients [52–54]. Despite high rates of FGFR3 muta- RXR-α heterodimerizes with PPARs, activating tran-
tion in luminal-papillary tumors, clinical trials have not scription of PPAR-regulated genes involved in fatty
incorporated molecular subtyping strategies to deter- acid metabolism [63]. Hotspot mutations in the lig-
mine patient eligibility. Interestingly, there may be an and binding domain of RXR-α (S427F/Y) are enriched
inverse relationship between immunotherapy response in bladder cancers and have been shown to stabilize
and presumed features of FGFR3 inhibitor response, the heterodimerization between RXR-α and PPAR-γ,
enhancing its interaction with co-activators and ini-
since response to atezolizumab was low in the TCGA
tiating downstream transcriptional regulation [63,64].
‘cluster 1’ luminal subtype that includes most FGFR3
Expressing RXR-αS427F in a syngeneic mouse bladder
alterations [55]. Indeed, a preliminary subgroup anal-
tumor model activated the PPAR-γ pathway, reduced
ysis of the erdafitinib study showed a 70% confirmed
levels of tumor-infiltrating CD8+ T cells, and blunted
overall response rate among patients who previously
the effects of anti-CTLA4 immunotherapy treatments
failed immune checkpoint inhibitors [54]. If confirmed,
[64]. Conversely, patients whose cancers show a high
these relationships between response and subtype sug-
density of T cells have shown better anti-tumor immu-
gest that cell-based subtyping will be an efficient means
nity in clinical trials of immune checkpoint inhibitors
of stratifying patients for treatments and, at least in some
targeting CTLA-4 and PD-1 [65]. Both RXRA muta-
instances, decrease the need for more complex molecu-
tions and 3p25 amplifications spanning PPARG are sig-
lar testing.
nificantly enriched in the luminal-like subtypes [29].
Therefore, investigators have proposed that antagoniz-
ERBB2/HER2 ing PPAR-γ/RXR-α may sensitize bladder cancers to
Activating mutations in ERBB2 (also known as HER2) immunotherapies, especially for non-inflamed subsets
have been found in approximately 12% of MIBC that may have an inferior response to immunothera-
cases, with basal-squamous subtypes showing about pies [55,64]. If further evidence supports this proposal,
half the rate of alterations (6%) compared with other PPAR-γ may become a target for predictive biomarkers
subtypes (12–15%) [29]. Genomic amplification of and therapeutic intervention, particularly in luminal sub-
HER2 is common and increases the number of poten- types of MIBC.
tially targetable cases to over 20% [56,57]. However,
response to anti-HER2 therapies have largely yielded PD1/PD-L1 immune checkpoint inhibitors
negative results in muscle-invasive bladder cancer Tumor cells evade the immune system by expressing
[58–60]. A clinical trial conducted with lapatinib, a immunosuppressive signals that block active engage-
dual tyrosine kinase inhibitor for EGFR and ERBB2, ment of cytotoxic, tumor-killing T cells. Immune
saw no improvement in progression-free survival for checkpoint inhibitors in current clinical use for MIBC
232 HER2+ patients with metastatic UC [60]. Other include atezolizumab and nivolumab, which target sig-
therapeutic approaches for HER2+ bladder cancer naling by programmed death-ligand 1 (PD-L1) on tumor
patients include a phase 2 trial of DN24-02, a cellular cells to its cognate receptor, programmed death-1 (PD1)
immunotherapy with HER2-derived antigen linked to on activated T cells [65]. UC has shown dramatic,
granulocyte macrophage colony-stimulating factor, but sometimes durable, responses to these agents [66], even
no significant differences in overall survival or distant in chemotherapy-refractory metastatic disease [67–69].
recurrence-free survival were observed [59]. These However, checkpoint inhibitors are not universally
poor historical response rates to HER2 inhibitors may effective and come at an extraordinary financial cost,
be a function of the requirement for HER2 receptor supporting the need for predictive biomarkers. Factors
heterodimerization with other members of the EGFR such as PD-L1 expression and tumor mutational bur-
receptor family for signaling. Indeed, both the EGFR den have been investigated as predictive biomarkers
Copyright © 2019 Pathological Society of Great Britain and Ireland. J Pathol 2019; 247: 563–573
Published by John Wiley & Sons, Ltd. www.pathsoc.org www.thejournalofpathology.com
568 G Sjödahl et al

Table 1. Response rates to immune checkpoint inhibitors atezolizumab (IMvigor210) and nivolumab (CheckMate275) in patient subsets
including molecular subtype (TCGA I-IV)
Cluster I Cluster II Cluster III Cluster IV
‘Luminal’ ‘Luminal-infiltrated’ ‘Basal-squamous’ ‘Basal’
Luminal 1 Luminal 2 Basal 1 Basal 2
IMvigor210 Objective response rate 10% 34% 16% 20%
PD-L1 immune cell 15% 24% 68% 50%
prevalence (defined by
IC2/3 expression)

PD-L1 tumor cell Low (4%) High (39%), no association

expression with ORR

CD8 T-effector gene Increased relative to cluster I

expression

CheckMate275 Percentage of patients 18% 27% 30 % 15 %

with complete or partial
responses

CXCL9/10 cytokines Low – (NSD) Low – (NSD) High Low – (NSD)

CD8 expression Low Medium High Medium

Highest objective response rate in patients with PD-L1

PD-L1
expression >5%. Not correlated to subtype

Overall: Best responders to PD-L1 immunotherapies

Please see text for citations to the relevant literature. NSD, not significantly different.

for response to PD1/PDL-1-based immunotherapies, Hedgehog and bone morphogenetic proteins

with limited success [65,68,69]. More recent efforts As shown in model systems, signaling between tumor
have investigated the potential of cell-based molecu- and stroma can promote or restrain cancer progression
lar subtypes as effective predictive biomarkers. The [70–72]. For example, in mice, sonic hedgehog (Shh)
IMvigor210 phase 2 clinical trial tested patients who expression from epithelial cells activates expression
progressed after platinum-based chemotherapy for of bone morphogenetic protein (Bmp) in the subjacent
responses to atezolizumab and potential relationships stroma, which then feeds back upon epithelial cells
with molecular subtypes as defined by the TCGA to maintain differentiation of the urothelium [71,72].
clusters I–IV. Patients with immune infiltrated lumi- Engineered ablation of Shh signaling in murine urothe-
nal (cluster II) and basal (cluster IV) tumor subtypes lium downregulates Bmp signaling and accelerates
showed the highest objective response rates (34% and bladder cancer growth and invasion [72], a process that
20%, respectively, compared with 10% and 16% for was inhibited by administration of tacrolimus (FK506).
the non-infiltrated clusters I and III) [55]. Similarly, in Tacrolimus is best known as an immunosuppressive
cisplatin-ineligible patients, atezolizumab showed the agent for organ transplant patients that suppresses
greatest benefit in patients with the immune infiltrated cytokine expression. The drug appears to also stimulate
luminal II subtype [66]. The phase 2 CheckMate275 activation of BMP receptors, which was shown to be
clinical trial tested nivolumab as a second-line ther- required for suppressing UC progression [72]. If these
apy in metastatic urothelial carcinoma patients. The results are tested and confirmed in humans, they may
trial demonstrated notable improvements in response be more applicable to patients with basal/squamous
associated with the basal I and luminal II subtypes, subtype cancers, as these show the lowest expression of
with partial or complete responses in 30% and 27% SHH and BMP7 genes [29,38].
of patients, respectively, in comparison to the 19%
overall response rate [67]. Overall, the success of TGFβ
PD1 and PD-L1 inhibitors in non-tumor-cell intrin- Transforming growth factor beta (TGFB) signals
sic subtypes (Table 1) suggests that the efficacy of through canonical and non-canonical pathways.
these immunotherapies may be related to the tumor Through SMADs, it can activate PI3K–AKT, RHOA,
microenvironment (TME) type, rather than the tumor and MAPK pathways [73,74]. TGFβ signaling appears
cell phenotype. Therefore, reliable relationships have to suppress early cancer formation while promoting
yet to be established between checkpoint inhibitor the progression of more advanced cancers [75]. Of
response and urothelial carcinoma phenotypes and particular interest to bladder cancer treatment are recent
genotypes. data indicating that TGFβ signaling in stroma functions
Copyright © 2019 Pathological Society of Great Britain and Ireland. J Pathol 2019; 247: 563–573
Published by John Wiley & Sons, Ltd. www.pathsoc.org www.thejournalofpathology.com
Intrinsic subtypes of bladder cancer 569

Figure 2. Schematic illustration of how different tumor microenvironment (TME) types may be related to the purity of tumor samples, and
how they both influence RNA-based molecular subtype classification.

to exclude tumor-killing T cells from UCs and serves as expression subtypes were reliably identified using two
a reversible mechanism of resistance to PDL1 targeted IHC markers, GATA3 and KRT5/6 [13,39]. Similar
immunotherapy. Mariathasan et al [70] reported a CD8+ correspondence between RNA- and IHC-profiling
T-cell effector phenotype associated with response to for the basal/squamous subtype was also obtained
atezolizumab in patients with metastatic urothelial by others [41]. The literature therefore strongly indi-
carcinoma. Strikingly, tumors from non-responders cates that luminal-like mRNA subtype(s) are robustly
showed immune cell exclusion, and instead were GATA3-positive on IHC, while non-luminal-like sub-
enriched for a signature indicating TGFβ signaling in types express low or no GATA3. In addition to low
fibroblasts. In a mouse model of breast cancer, addition GATA3, the basal/squamous mRNA subtype is robustly
of TGFβ-neutralizing antibodies to PD-L1 blockade identified by positive staining for KRT5 or KRT5/6.
resulted in increased cytotoxic T-cell infiltration into A straightforward classification that is concordant
tumors and more effective anti-tumor immune responses between mRNA and IHC levels, similar to that observed
[70], providing functional evidence that stromal TGFβ in breast cancer, would be highly desirable; however,
could be an important arbiter of immune response. remaining obstacles complicate direct comparison
Importantly, the same study also reported significantly between transcriptional signature and IHC phenotype.
different rates of response to atezolizumab depending First, there is variation in the content of immune
on the Lund molecular subtype, with highest response and stromal cells both within and between tumor cell
rates in the genomically unstable and lowest in the UroA intrinsic subtypes. This is exemplified schematically
subtype. As with response to checkpoint inhibitors, the in Figure 2, where non-inflamed subtypes tend to have
potential utility of modulating TGFβ is perhaps more high sample purity, which results in tumor cell intrinsic
closely linked to the TME type than to the cancer cell subtype classification. Lum-papillary/UroA tumors
subtype. show consistently low immune and stroma-signatures,
which was confirmed by review of H&E slides [29]
and by IHC for lymphocyte and myeloid cell mark-
Considerations of cell type and sampling in MIBC ers [76]. In contrast, the basal/squamous subtype
molecular subtype classification has a high immune signature and a high density of
tumor-infiltrating immune cells [29,76,77]. This may
Transcript-based biopsy profiling and in situ cell profil- be partly explained by differences in invasion depth,
ing (i.e. IHC) are two fundamentally different methods since the number of infiltrating lymphocytes correlates
for profiling gene expression in tumors. Several publica- positively with tumor stage [78]. Furthermore, the
tions provide evidence that the mRNA-based subtypes process of sample taking itself plays a central role
discussed above correspond to in situ tumor cell intrinsic in the association between different TME types and
immunophenotypes. Luminal-like and basal-like gene tumor cell intrinsic subtypes. By sampling only isolated
Copyright © 2019 Pathological Society of Great Britain and Ireland. J Pathol 2019; 247: 563–573
Published by John Wiley & Sons, Ltd. www.pathsoc.org www.thejournalofpathology.com
570 G Sjödahl et al

parts of tumors, systematic variation in sample purity gene expression panels, and/or immunohistochemistry.
is introduced. For example, research samples collected To date, most studies have relied on large-scale tran-
fresh from an exophytic tumor during trans-urethral scriptomic (>40 000 probe) profiling to subtype UCs
resection of a bladder tumor (TURB-T) may have a [25,26,28,29,36,41]. This approach, while sometimes
different purity than routine pathology specimens from employed clinically [80], suffers from a number of draw-
TURB-T or radical cystectomy specimens. Recently, backs. Because information from fewer than 100 genes
Wang et al [79] demonstrated a positive correlation is needed to make most clinical decisions [81], assays
between stroma- and immune-related signatures in the may be unnecessarily complex, challenging to validate,
TCGA RNA-seq data set. This correlation was no longer computationally intensive, and ill-suited for deployment
apparent after adjusting for tumor purity, suggesting in local hospitals. Thus, they are typically available only
that variation in purity and perhaps sampling cause as a sendout test. Sendout testing takes extra time and
immune-infiltrated tumors to also appear stroma-rich removes information gathering and interpretation from
and vice versa. Despite attempts to control for tumor the local healthcare team, thereby compromising com-
purity, e.g. excluding fresh tumor samples with purity munication and storage of medical records and degrad-
lower than 50% [29], or performing macro-dissection ing the team’s ability to interpret routine and atypical
of tumor-rich area(s) before RNA extraction [14], results and to employ them in research that can yield
large gene expression studies consistently identify at important biologic insights and improve patient care.
least one large cluster of samples defined entirely by Compact gene expression panels such as Oncotype
non-tumor intrinsic immune and/or stromal signals Dx and Mammaprint used for breast cancer typically
[14,26,29,36]. It has been shown that these non-intrinsic incorporate fewer than 100 features [81,82]. In con-
mRNA subtypes, exemplified in the middle right panel trast to whole transcriptome profiling, panels may be
in Figure 2, can be resolved into the other intrinsic sub- more widely adopted by local hospital laboratories,
types by in situ IHC analysis, but only partly by mRNA and perform better with small amounts of nucleic acid,
expression analysis [13,14]. Finally, it is likely that allowing incorporation of diverse genomic features
both true TME type and considerations of purity and such as mutations, copy number variation, and gene
sample-taking methods will be relevant when subtypes expression into a single test. Indications for such test-
of MIBC and NMIBC are to be reconciled. Untreated ing currently focus on the presence of mutations (e.g.
NMIBCs have fewer infiltrating non-tumor cells than EGFR, BRAF) or fusions (e.g. ALK, ROS) that predict
MIBCs [76,78], but a low tumor burden may also result response or resistance (e.g. KRAS) to targeted thera-
in samples of low purity. While these considerations pies. Assays to measure expression of subtype-specific
may seem highly technical at first glance, they indicate mRNAs could easily be incorporated into such panels to
that despite published claims to the contrary, ‘infil- provide efficient testing for patients with UC. However,
trated’ transcript-based subtypes can be unreliable [13]. all of this information comes at a cost. Genomic testing
Non-infiltrated or in situ-based subtypes are likely to be or gene expression panels require expensive equipment
more biologically meaningful and more useful in guid- and reagents, highly specialized personnel, and typi-
ing clinical decision-making. An additional benefit of in cally adds 1–2 weeks to the time required to complete
situ profiling is the possibility of detecting intra-tumor a pathology report.
heterogeneity, which cannot be resolved by transcrip- IHC has been a workhorse in surgical pathology lab-
tomics. We currently do not know the proportion of oratories for over 35 years. While challenging to use
bladder tumors that contain more than one molecular for quantitative purposes, IHC is routinely employed
subtype and whether it is clinically meaningful. We also in cancer diagnosis for qualitative purposes, most com-
do not know if molecular subtypes depend on interac- monly to query cell type differentiation. Thus, IHC is
tions with the microenvironment and to what extent they ideally suited for subtype determination. As an example
change in response to treatment, both of which are areas of a clinically tractable scheme to employ IHC in sub-
that require investigation. If we manage to validate the typing, we suggest here an algorithm to be evaluated
use of informative IHC assays on full TURB-T sections, for IHC-based MIBC molecular subtype classification
future studies will be able to identify which cancers (Figure 3). With multiplexing, the vast majority of cases
show subtype plasticity/heterogeneity and devise a could be subtyped using one or two tissue sections
management strategy based on their clinical behavior. using commercially available antibodies that are certi-
fied for use on routinely processed archival tissues [83].
We anticipate that UC subtyping will become estab-
Envisioning molecular cell type profiling lished as a prognostic and predictive tool and propose
in routine pathology practice the use of IHC for routine subtyping on diagnostic
material from muscle-invasive cases. Molecular subtype
If, as suggested by a wealth of work described above, will then be considered along with the entire clinical
UC subtype influences treatment outcomes, pathologists picture for each patient, including staging, renal func-
will need an efficient and effective method of assess- tion, comorbidities, and other information that influ-
ing and reporting this information to clinicians who ences choice of treatment. Once treatment options are
treat bladder cancer patients. Subtyping options to con- enumerated, additional predictive tests can be tailored
sider include whole transcriptome profiling, compact to these options and ordered on an as-needed basis.
Copyright © 2019 Pathological Society of Great Britain and Ireland. J Pathol 2019; 247: 563–573
Published by John Wiley & Sons, Ltd. www.pathsoc.org www.thejournalofpathology.com
Intrinsic subtypes of bladder cancer 571

Figure 3. Proposed algorithm for immunohistochemical (IHC) subtyping of muscle-invasive bladder cancers (MIBCs). Solid arrows represent
IHC assays required for subtype classification. Dotted arrows represent optional immunohistochemical assays along with histological
features for validation of each subtype. For the three major subclasses, which make up 87% of a consecutive cystectomy series [14],
no more than three antibodies should be required. As GATA3 is localized to the nucleus and KRT5 to the cytoplasm, the two are
amenable to multiplexed assays, indicating that two tissue sections would suffice for sub-classification. (A) KRT5 and GATA3 expression
yields three categories: Uro/GU (urothelial-like and genomically unstable subtypes), basal-SCCL (basal/SCC-like), and Mes-like/NE-like
(mesenchymal-like and neuroendocrine-like). Basal SCCL requires no further testing. Rare tumors lack significant expression of both GATA3
and KRT5 and can be further sub-classified into Mes-like or NE-like subtypes. (B) Sub-classification (left) into Uro or GU on the basis of low
versus moderate or high intensity staining with antibodies against p16. EpCAM IHC (right) sub-classified GATA3− /KRT− MIBCs into NE-like
(positive) and MES-like (negative) subtypes. (C) Options for validation include additional IHC and histologic features. CCND1 IHC is positive
in Uro and negative in GU subtypes. The basal/SCC-like subtype can be confirmed by KRT14 expression and/or squamous morphology. The
Mes-like subtype can be confirmed by vimentin staining and/or sarcomatoid morphology. NE-like subtype can be confirmed by small-cell
morphology or typical neuroendocrine IHC markers such as chromogranin or synaptophysin.

Summary/conclusion (DMB), and University Hospitals Kingston Foundation

(DMB).
Although molecular profiling has not yet formally
shown clinical benefit in MIBC, the retrospective data Author contributions statement
on molecular subtypes and on biomarkers for targeted
therapy are promising. The relationship between GS and DMB conceived the manuscript. All authors
tumor microenvironment, sample purity, mRNA, wrote sections of and edited the manuscript. CLJ
and IHC-based profiling in MIBC is complex. The produced Table 1 and Figures 1 and 3. GS produced
stability of molecular subtypes over time and the Figure 2.
subtype heterogeneity within tumors and patients References
are yet unknown and strongly heterogeneous cases
1. Sylvester RJ, van der Meijden APM, Oosterlinck W, et al. Predicting
should be interpreted with caution. Molecular profiling
recurrence and progression in individual patients with stage Ta T1
by high-throughput RNA-based methods is possible,
bladder cancer using EORTC risk tables: a combined analysis of
but IHC-based classification has the advantage of being 2596 patients from seven EORTC trials. Eur Urol 2006; 49: 466–465
readily available to any pathologist. We envision a world discussion 475–477.
where subtyping is efficiently and effectively deployed 2. Babjuk M, Böhle A, Burger M, et al. EAU guidelines on
to ensure (a) that patients derive maximal benefit from non-muscle-invasive urothelial carcinoma of the bladder: update
advances in clinical and translational research, and (b) 2016. Eur Urol 2017; 71: 447–461.
that such prospectively collected information may 3. van Rhijn BWG, Burger M, Lotan Y, et al. Recurrence and progres-
feed back into research, providing a valuable resource sion of disease in non-muscle-invasive bladder cancer: from epidemi-
for future discoveries. ology to treatment strategy. Eur Urol 2009; 56: 430–442.
4. Kamat AM, Li R, O’Donnell MA, et al. Predicting response to
intravesical bacillus Calmette-Guérin immunotherapy: are we there
Acknowledgements
yet? A systematic review. Eur Urol 2018; 73: 738–748.
5. Antoni S, Ferlay J, Soerjomataram I, et al. Bladder cancer incidence
This work was supported by funding from the Swedish and mortality: a global overview and recent trends. Eur Urol 2017;
Cancer Society (GS), the Ontario Molecular Pathology 71: 96–108.
Research Network (DMB), a program of the Ontario 6. Hermans TJN, Voskuilen CS, van der Heijden MS, et al. Neoadjuvant
Institute for Cancer Research (DMB, JMSB), Bladder treatment for muscle-invasive bladder cancer: the past, the present,
Cancer Canada (DMB), the Cancer Research Society and the future. Urol Oncol 2018; 36: 413–422.

Copyright © 2019 Pathological Society of Great Britain and Ireland. J Pathol 2019; 247: 563–573
Published by John Wiley & Sons, Ltd. www.pathsoc.org www.thejournalofpathology.com
572 G Sjödahl et al

7. Nadal R, Apolo AB. Overview of current and future adjuvant therapy 28. The Cancer Genome Atlas Research Network. Comprehensive
for muscle-invasive urothelial carcinoma. Curr Treat Options Oncol molecular characterization of urothelial bladder carcinoma. Nature
2018; 19: 36. 2014; 507: 315–322.
8. Meeks JJ, Bellmunt J, Bochner BH, et al. A systematic review of 29. Robertson AG, Kim J, Al-Ahmadie H, et al. Comprehensive molec-
neoadjuvant and adjuvant chemotherapy for muscle-invasive bladder ular characterization of muscle-invasive bladder cancer. Cell 2018;
cancer. Eur Urol 2012; 62: 523–533. 171: 540–556.
9. Booth CM, Siemens DR, Peng Y, et al. Delivery of perioperative 30. Biton A, Bernard-Pierrot I, Lou Y, et al. Independent component
chemotherapy for bladder cancer in routine clinical practice. Ann analysis uncovers the landscape of the bladder tumor transcriptome
Oncol 2014; 25: 1783–1788. and reveals insights into luminal and basal subtypes. Cell Rep 2014;
10. Clark PE. Neoadjuvant versus adjuvant chemotherapy for 9: 1235–1245.
muscle-invasive bladder cancer. Expert Rev Anticancer Ther 31. Perou CM, Jeffrey SS, van de Rijn M, et al. Distinctive gene expres-
2009; 9: 821–830. sion patterns in human mammary epithelial cells and breast cancers.
11. Li R, Metcalfe M, Kukreja J, et al. Role of radical cystectomy in Proc Natl Acad Sci U S A 1999; 96: 9212–9217.
non-organ confined bladder cancer: a systematic review. Bladder 32. Sørlie T, Perou CM, Tibshirani R, et al. Gene expression patterns of
Cancer 2018; 4: 31–40. breast carcinomas distinguish tumor subclasses with clinical impli-
12. Godwin JL, Hoffman-Censits J, Plimack E. Recent developments cations. Proc Natl Acad Sci U S A 2001; 98: 10869–10874.
in the treatment of advanced bladder cancer. Urol Oncol 2018; 36: 33. Blows FM, Driver KE, Schmidt MK, et al. Subtyping of breast can-
109–114. cer by immunohistochemistry to investigate a relationship between
13. Dadhania V, Zhang M, Zhang L, et al. Meta-analysis of the luminal subtype and short and long term survival: a collaborative analysis of
and basal subtypes of bladder cancer and the identification of sig- data for 10,159 cases from 12 studies. PLoS Med 2010; 7: e1000279.
nature immunohistochemical markers for clinical use. EBioMedicine 34. Hedegaard J, Lamy P, Nordentoft I, et al. Comprehensive transcrip-
2016; 12: 105–117. tional analysis of early-stage urothelial carcinoma. Cancer Cell 2016;
14. Sjödahl G, Eriksson P, Liedberg F, et al. Molecular classification of 30: 27–42.
urothelial carcinoma: global mRNA classification versus tumour-cell 35. Hurst CD, Alder O, Platt FM, et al. Genomic subtypes of
phenotype classification. J Pathol 2017; 242: 113–125. non-invasive bladder cancer with distinct metabolic profile and
15. Spruck CH, Ohneseit PF, Gonzalez-Zulueta M, et al. Two molecular female gender bias in KDM6A mutation frequency. Cancer Cell
pathways to transitional cell carcinoma of the bladder. Cancer Res 2017; 32: 701–715.e7.
1994; 54: 784–788. 36. Seiler R, Ashab HAD, Erho N, et al. Impact of molecular subtypes in
16. van Rhijn BWG, van der Kwast TH, Vis AN, et al. FGFR3 and muscle-invasive bladder cancer on predicting response and survival
P53 characterize alternative genetic pathways in the pathogenesis of after neoadjuvant chemotherapy. Eur Urol 2017; 72: 544–554.
urothelial cell carcinoma. Cancer Res 2004; 64: 1911–1914. 37. Choi W, Ochoa A, McConkey DJ, et al. Genetic alterations in the
17. Wu X-R. Urothelial tumorigenesis: a tale of divergent pathways. Nat molecular subtypes of bladder cancer: illustration in the Cancer
Rev Cancer 2005; 5: 713–725. Genome Atlas dataset. Eur Urol 2017; 72: 354–365.
18. Cappellen D, De Oliveira C, Ricol D, et al. Frequent activating 38. Aine M, Eriksson P, Liedberg F, et al. Biological determinants of
mutations of FGFR3 in human bladder and cervix carcinomas. Nat bladder cancer gene expression subtypes. Sci Rep 2015; 5: 10957.
Genet 1999; 23: 18–20. 39. Sjödahl G, Lövgren K, Lauss M, et al. Toward a molecular patho-
19. van Rhijn BW, Lurkin I, Radvanyi F, et al. The fibroblast growth logic classification of urothelial carcinoma. Am J Pathol 2013; 183:
factor receptor 3 (FGFR3) mutation is a strong indicator of superficial 681–691.
bladder cancer with low recurrence rate. Cancer Res 2001; 61: 40. Eriksson P, Aine M, Veerla S, et al. Molecular subtypes of urothelial
1265–1268. carcinoma are defined by specific gene regulatory systems. BMC Med
20. Simon R, Bürger H, Brinkschmidt C, et al. Chromosomal aberra- Genomics 2015; 8: 25.
tions associated with invasion in papillary superficial bladder cancer. 41. Rebouissou S, Bernard-Pierrot I, de Reyniès A, et al. EGFR as a
J Pathol 1998; 185: 345–351. potential therapeutic target for a subset of muscle-invasive bladder
21. Rebouissou S, Hérault A, Letouzé E, et al. CDKN2A homozy- cancers presenting a basal-like phenotype. Sci Transl Med 2014; 6:
gous deletion is associated with muscle invasion in FGFR3-mutated 244ra91.
urothelial bladder carcinoma. J Pathol 2012; 227: 315–324. 42. Hoadley KA, Yau C, Wolf DM, et al. Multiplatform analysis of 12
22. He X, Marchionni L, Hansel DE, et al. Differentiation of a highly cancer types reveals molecular classification within and across tissues
tumorigenic basal cell compartment in urothelial carcinoma. Stem of origin. Cell 2014; 158: 929–944.
Cells 2009; 27: 1487–1495. 43. Kardos J, Chai S, Mose LE, et al. Claudin-low bladder tumors are
23. Chan KS, Espinosa I, Chao M, et al. Identification, molecular char- immune infiltrated and actively immune suppressed. JCI Insight
acterization, clinical prognosis, and therapeutic targeting of human 2016; 1: e85902.
bladder tumor-initiating cells. Proc Natl Acad Sci U S A 2009; 106: 44. Gandhi NM, Baras A, Munari E, et al. Gemcitabine and cisplatin
14016–14021. neoadjuvant chemotherapy for muscle-invasive urothelial carcinoma:
24. Volkmer J-P, Sahoo D, Chin RK, et al. Three differentiation states predicting response and assessing outcomes. Urol Oncol 2015; 33:
risk-stratify bladder cancer into distinct subtypes. Proc Natl Acad Sci 204.e1–204.e7.
U S A 2012; 109: 2078–2083. 45. Advanced Bladder Cancer (ABC) Meta-analysis Collaboration.
25. Sjödahl G, Lauss M, Lövgren K, et al. A molecular taxonomy for Neoadjuvant chemotherapy in invasive bladder cancer: update of
urothelial carcinoma. Clin Cancer Res 2012; 18: 3377–3386. a systematic review and meta-analysis of individual patient data
26. Choi W, Porten S, Kim S, et al. Identification of distinct basal advanced bladder cancer (ABC) meta-analysis collaboration. Eur
and luminal subtypes of muscle-invasive bladder cancer with dif- Urol 2005; 48: 202–205 discussion 205–206.
ferent sensitivities to frontline chemotherapy. Cancer Cell 2014; 25: 46. Krabbe L-M, Westerman ME, Margulis V, et al. Changing trends in
152–165. utilization of neoadjuvant chemotherapy in muscle-invasive bladder
27. Damrauer JS, Hoadley KA, Chism DD, et al. Intrinsic subtypes of cancer. Can J Urol 2015; 22: 7865–7875.
high-grade bladder cancer reflect the hallmarks of breast cancer 47. Reardon ZD, Patel SG, Zaid HB, et al. Trends in the use of
biology. Proc Natl Acad Sci U S A 2014; 111: 3110–3115. perioperative chemotherapy for localized and locally advanced

Copyright © 2019 Pathological Society of Great Britain and Ireland. J Pathol 2019; 247: 563–573
Published by John Wiley & Sons, Ltd. www.pathsoc.org www.thejournalofpathology.com
Intrinsic subtypes of bladder cancer 573

muscle-invasive bladder cancer: a sign of changing tides. Eur Urol proliferator-activated receptors to drive urothelial proliferation.
2015; 67: 165–170. Elife 2017; 6: e30862.
48. Zaid HB, Patel SG, Stimson CJ, et al. Trends in the utilization of 64. Korpal M, Puyang X, Jeremy Wu Z, et al. Evasion of immunosurveil-
neoadjuvant chemotherapy in muscle-invasive bladder cancer: results lance by genomic alterations of PPARγ/RXRα in bladder cancer. Nat
from the National Cancer Database. Urology 2014; 83: 75–80. Commun 2017; 8: 103.
49. McConkey DJ, Choi W, Shen Y, et al. A prognostic gene expres- 65. Herbst RS, Soria J-C, Kowanetz M, et al. Predictive correlates of
sion signature in the molecular classification of chemotherapy-naïve response to the anti-PD-L1 antibody MPDL3280A in cancer patients.
urothelial cancer is predictive of clinical outcomes from neoadjuvant Nature 2014; 515: 563–567.
chemotherapy: a phase 2 trial of dose-dense methotrexate, vinblas- 66. Balar AV, Castellano D, O’Donnell PH, et al. First-line pem-
tine, doxorubicin, and cisplatin with bevacizumab in urothelial can- brolizumab in cisplatin-ineligible patients with locally advanced
cer. Eur Urol 2016; 69: 855–862. and unresectable or metastatic urothelial cancer (KEYNOTE-052):
50. Tomlinson DC, Baldo O, Harnden P, et al. FGFR3 protein expression a multicentre, single-arm, phase 2 study. Lancet Oncol 2017; 18:
and its relationship to mutation status and prognostic variables in 1483–1492.
bladder cancer. J Pathol 2007; 213: 91–98. 67. Sharma P, Retz M, Siefker-Radtke A, et al. Nivolumab in metastatic
51. Pietzak EJ, Bagrodia A, Cha EK, et al. Next-generation sequencing urothelial carcinoma after platinum therapy (CheckMate 275): a mul-
of nonmuscle invasive bladder cancer reveals potential biomarkers ticentre, single-arm, phase 2 trial. Lancet Oncol 2017; 18: 312–322.
and rational therapeutic targets. Eur Urol 2017; 72: 952–959. 68. Siefker-Radtke AO, Apolo AB, Bivalacqua TJ, et al. Immunotherapy
52. Pal SK, Rosenberg JE, Hoffman-Censits JH, et al. Efficacy of with checkpoint blockade in the treatment of urothelial carcinoma.
BGJ398, a fibroblast growth factor receptor 1-3 inhibitor, in patients J Urol 2018; 199: 1129–1142.
with previously treated advanced urothelial carcinoma with FGFR3 69. Powles T, Eder JP, Fine GD, et al. MPDL3280A (anti-PD-L1) treat-
alterations. Cancer Discov 2018; 8: 812–821. ment leads to clinical activity in metastatic bladder cancer. Nature
53. Nogova L, Sequist LV, Perez Garcia JM, et al. Evaluation of BGJ398, 2014; 515: 558–562.
a fibroblast growth factor receptor 1-3 kinase inhibitor, in patients 70. Mariathasan S, Turley SJ, Nickles D, et al. TGFβ attenuates tumour
with advanced solid tumors harboring genetic alterations in fibroblast response to PD-L1 blockade by contributing to exclusion of T cells.
growth factor receptors: results of a global phase I, dose-escalation Nature 2018; 554: 544–548.
and dose-expansion study. J Clin Oncol 2017; 35: 157–165. 71. Shin K, Lee J, Guo N, et al. Hedgehog/Wnt feedback supports
54. Siefker-Radtke AO, Necchi A, Park SH, et al. First results from regenerative proliferation of epithelial stem cells in bladder. Nature
the primary analysis population of the phase 2 study of erdafi- 2011; 472: 110–114.
tinib (ERDA; JNJ-42756493) in patients (pts) with metastatic or 72. Shin K, Lim A, Zhao C, et al. Hedgehog signaling restrains blad-
unresectable urothelial carcinoma (mUC) and FGFR alterations der cancer progression by eliciting stromal production of urothelial
(FGFRalt). J Clin Oncol 2018; 36: 4503–4503. differentiation factors. Cancer Cell 2014; 26: 521–533.
55. Rosenberg JE, Hoffman-Censits J, Powles T, et al. Atezolizumab 73. Moustakas A, Heldin C-H. Non-Smad TGF-beta signals. J Cell Sci
in patients with locally advanced and metastatic urothelial carci- 2005; 118: 3573–3584.
noma who have progressed following treatment with platinum-based 74. Moustakas A, Souchelnytskyi S, Heldin CH. Smad regulation in
chemotherapy: a single-arm, multicentre, phase 2 trial. Lancet 2016; TGF-beta signal transduction. J Cell Sci 2001; 114: 4359–4369.
387: 1909–1920. 75. Massagué J. TGFbeta in cancer. Cell 2008; 134: 215–230.
56. Kiss B, Wyatt AW, Douglas J, et al. Her2 alterations in 76. Sjödahl G, Lövgren K, Lauss M, et al. Infiltration of CD3+ and
muscle-invasive bladder cancer: patient selection beyond protein CD68+ cells in bladder cancer is subtype specific and affects the
expression for targeted therapy. Sci Rep 2017; 7: 42713. outcome of patients with muscle-invasive tumors. Urol Oncol 2014;
57. Eriksson P, Sjödahl G, Chebil G, et al. HER2 and EGFR amplifica- 32: 791–797.
tion and expression in urothelial carcinoma occurs in distinct biolog- 77. Hodgson A, Liu SK, Vesprini D, et al. Basal-subtype bladder
ical and molecular contexts. Oncotarget 2017; 8: 48905–48914. tumours show a ‘hot’ immunophenotype. Histopathology 2018; 73:
58. Wülfing C, Machiels J-PH, Richel DJ, et al. A single-arm, multicen- 748–757.
ter, open-label phase 2 study of lapatinib as the second-line treatment 78. Lipponen PK, Eskelinen MJ, Jauhiainen K, et al. Tumour infiltrating
of patients with locally advanced or metastatic transitional cell carci- lymphocytes as an independent prognostic factor in transitional cell
noma. Cancer 2009; 115: 2881–2890. bladder cancer. Eur J Cancer 1992; 29A: 69–75.
59. Bajorin DF, Sharma P, Quinn DI, et al. Phase 2 trial results of DN24- 79. Wang L, Saci A, Szabo PM, et al. EMT- and stroma-related gene
02, a HER2-targeted autologous cellular immunotherapy in HER2+ expression and resistance to PD-1 blockade in urothelial cancer. Nat
urothelial cancer patients (pts). J Clin Oncol 2016; 34: 4513–4513. Commun 2018; 9: 3503.
60. Powles T, Huddart RA, Elliott T, et al. Phase III, double-blind, ran- 80. Erho N, Crisan A, Vergara IA, et al. Discovery and validation of
domized trial that compared maintenance lapatinib versus placebo a prostate cancer genomic classifier that predicts early metastasis
after first-line chemotherapy in patients with human epidermal following radical prostatectomy. PLoS One 2013; 8: e66855.
growth factor receptor 1/2-positive metastatic bladder cancer. J Clin 81. Buyse M, Loi S, van’t Veer L, et al. Validation and clinical utility of
Oncol 2017; 35: 48–55. a 70-gene prognostic signature for women with node-negative breast
61. DeGraff DJ, Cates JM, Mauney JR, et al. When urothelial differen- cancer. J Natl Cancer Inst 2006; 98: 1183–1192.
tiation pathways go wrong: implications for bladder cancer develop- 82. Cronin M, Sangli C, Liu M-L, et al. Analytical validation of the
ment and progression. Urol Oncol 2013; 31: 802–811. Oncotype DX genomic diagnostic test for recurrence prognosis
62. Friedland SN, Leong A, Filion KB, et al. The cardiovascular effects and therapeutic response prediction in node-negative, estrogen
of peroxisome proliferator-activated receptor agonists. Am J Med receptor-positive breast cancer. Clin Chem 2007; 53: 1084–1091.
2012; 125: 126–133. 83. Sjödahl G. Molecular subtype profiling of urothelial carcinoma using
63. Halstead AM, Kapadia CD, Bolzenius J, et al. Bladder- a subtype-specific immunohistochemistry panel. Methods Mol Biol
cancer-associated mutations in RXRA activate peroxisome 2018; 1655: 53–64.

Copyright © 2019 Pathological Society of Great Britain and Ireland. J Pathol 2019; 247: 563–573
Published by John Wiley & Sons, Ltd. www.pathsoc.org www.thejournalofpathology.com