Purpose:

Neuroendocrine (NE) bladder carcinoma is a rare and aggressive variant. Molecular subtyping studies have found that 5% to 15% of muscle-invasive bladder cancer (MIBC) have transcriptomic patterns consistent with NE bladder cancer in the absence of NE histology. The clinical implications of this NE-like subtype have not been explored in depth.

Experimental Design:

Transcriptome-wide expression profiles were generated for MIBC collected from 7 institutions and clinical-use of Decipher Bladder. Using unsupervised clustering, we generated a clustering solution on a prospective training cohort (PTC; n = 175), developed single-sample classifiers to predict NE tumors, and evaluated the resultant models on a testing radical cystectomy (RC) cohort (n = 225). A random forest model was finalized and applied to 5 validation cohorts (n = 1302). Uni- and multivariable survival analyses were used to characterize clinical outcomes.

Results:

In the training cohort (PTC), hierarchical clustering using an 84-gene panel showed a cluster of 8 patients (4.6%) with highly heterogeneous expression of NE markers in the absence of basal or luminal marker expression. NE-like tumors were identified in 1% to 6.6% of cases in validation cohorts. Patients with NE-like tumors had significantly worse 1-year progression-free survival (65% NE-like vs. 82% overall; P = 0.046) and, after adjusting for clinical and pathologic factors, had a 6.4-fold increased risk of all-cause mortality (P = 0.001). IHC confirmed the neuronal character of these tumors.

Conclusions:

A single-patient classifier was developed that identifies patients with histologic urothelial cancer harboring a NE transcriptomic profile. These tumors represent a high-risk subgroup of MIBC, which may require different treatment.

Translational Relevance

Neuroendocrine (NE) bladder cancer is an aggressive variant of bladder cancer with significant metastatic potential and high risk of mortality. Diagnosis of these tumors is currently dependent on morphologic criteria and staining for NE markers. Using machine learning and multiple validation cohorts in different clinical settings, we developed a model that identifies tumors with transcriptomic profiles consistent with NE bladder cancers with an absence of NE features by morphologic criteria. Early and accurate identification of these patients by genomic analysis may improve outcomes through treatment intensification and adaptation of standard treatment regimens.

Neuroendocrine (NE) bladder carcinoma is a rare variant of muscle-invasive bladder cancer (MIBC) representing 0.5% to 1.2% of newly diagnosed cases (1–4). These tumors can be classified into small-cell (SC) and large-cell subtypes based on morphology (4). Both are very aggressive tumors, characterized by rapid progression with high metastatic potential. Patients with NE carcinoma have worse outcomes compared with conventional urothelial carcinoma, with median survival ranging between 9 to 20 months after curative intent in the nonmetastatic setting (2, 3, 5–7).

Molecular subtyping has revealed that MIBC can be coarsely divided into basal-like and luminal-like subtypes, although additional subtypes have been described in more granular models (8–10). Previously, The Cancer Genome Atlas (TCGA) and Lund groups have independently identified NE-like tumors in 5% to 15% of patients with predominantly histologic urothelial carcinoma (11, 12). Consistent with their SE/NE biology, these patients were found to have inferior outcomes (11, 12). Importantly, the majority of these NE-like cases had morphologic characteristics of conventional urothelial carcinoma associated with a NE genotype, although both cohorts included a small number of histologic NE tumors.

Given the high rate of occult metastases and the poor survival associated with SC/NE tumors, neoadjuvant chemotherapy (NAC) followed by either radical cystectomy (RC) or radiation therapy has become the standard of care in nonmetastatic cases (13–16). The chemotherapy regimen is extrapolated from experience with SC carcinoma of the lung and includes cisplatin and etoposide (17). However, given the rarity of NE carcinoma of the bladder, there is little direct evidence to define a standardized therapeutic strategy. The best treatment of patients with tumors that have the appearance of histologic urothelial carcinoma, but NE-like gene expression profiles, has yet to be defined. These patients may be at higher risk of progression and stand to gain significantly from treatment intensification or alternative therapies.

In this study, we used machine learning to develop a single-sample gene expression classifier designed to identify tumors with a NE-like transcriptomic signature in the absence histologic features of NE carcinoma.

Patient populations

The prospective training cohort (PTC) consisted of the deidentified and anonymized gene expression profiles of 175 patients with MIBC from the clinical use of the Decipher Bladder transurethral bladder resection test that were available in the Decipher GRID registry (NCT02609269). Pathologic TNM staging as well as clinical outcome data were not available for this cohort (Table 1). The RC cohort 1 (RC1) used for model testing consisted of 225 patients who underwent RC without any neoadjuvant therapy and had pathologic pT1 disease with nodal involvement or MIBC (18). We have specifically used only tumors with conventional urothelial histology to develop the NE-like classifier, because the objective is to identify NE-like gene expression in tumors lacking true NE histology.

Table 1.

Patient clinical characteristic table of the prospective and retrospective cohorts

TrainingTestingValidation
VariablePTCRC1RC2NAC (20)TMTTCGA (11)LUND (12)Imvigor210 (24)
n  175 225 256 223 136 405 307 195 
NE-like n (%) 8 (4.6) 4 (1.8) 8 (3.1) 3 (1.4) 9 (6.6) 9 (2.2%) 11 (3.6) 2 (1) 
Age Median (IQR) 70 (47–90)a 69 (63–75) 70 (63–77) 62 (56–71) 70 (62–77) 69 (60–76) NA NA 
Gender Female 41 (23.4%) 41 (18.2) 48 (18.8) 69 (30.9) 27 (20.0) 106 (26.2) NA 42 (21.5) 
 Male 134 (76.6%) 184 (81.8) 208 (81.2) 154 (69.1) 108 (80.0) 299 (73.8) NA 153 (78.5) 
Clinical lymph node stage (%) cN0 NA 0 (0.0) 231 (90.2) 140 (62.8) 60 (44.2) NA NA NA 
 cN1-3 NA 0 (0.0) 20 (7.8) 83 (37.2) 2 (1.4) NA NA NA 
 cNx NA 225 (100.0) 5 (2.0) 0 (0.0) 74 (53.2) NA NA NA 
Clinical tumor stage (%) cTis NA 0 (0.0) 1 (0.4) 0 (0.0) 0 (0.0) NA NA NA 
 cTa NA 0 (0.0) 7 (2.7) 0 (0.0) 0 (0.0) NA NA NA 
 cT1 NA 0 (0.0) 22 (8.6) 0 (0.0) 0 (0.0) NA NA NA 
 cT2 NA 0 (0.0) 162 (63.3) 90 (40.4) 96 (70.6) NA NA NA 
 cT3 NA 0 (0.0) 42 (16.4) 90 (40.4) 33 (24.6) NA NA NA 
 cT4 NA 0 (0.0) 21 (8.2) 43 (19.3) 6 (4.4) NA NA NA 
 cTx NA 225 (100.0) 1 (0.4) 0 (0.0) 1 (0.7) NA NA NA 
Pathologic tumor stage (%)b pT0 NA 0 (0.0) 0 (0.0) 80 (35.9) 4 (2.9) 0 (0) 0 (0.0) NA 
 pTa NA 0 (0.0) 1 (0.4) 3 (1.3) 1 (0.7) 0 (0) 13 (4.2) NA 
 pT1 NA 9 (4.0) 0 (0.0) 13 (5.8) 2 (1.5) 0 (0) 44 (14.3) NA 
 pTis NA 0 (0.0) 0 (0.0) 7 (3.1) 1 (0.7) 0 (0) 1 (0.3) NA 
 pT2 NA 68 (30.2) 72 (28.1) 42 (18.8) 6 (4.4) 131 (32.5) 241 (78.5) NA 
 pT3 NA 102 (45.3) 122 (47.7) 50 (22.4) 11 (8.1) 139 (34.2) 1 (0.3) NA 
 pT4 NA 46 (20.4) 61 (23.8) 24 (10.8) 2 (1.5) 133 (32.8) 1 (0.3) NA 
 pTx NA 0 (0.0) 0 (0.0) 4 (1.8) 109 (80.2) 2 (0.5) 6 (2.0) NA 
Pathologic lymph node stage (%) pN0 NA 145 (64.4) 142 (55.5) 157 (70.4) 14 (10.3) 231 (57) NA NA 
 pN1-3 NA 80 (35.6) 105 (41.0) 52 (23.3) 4 (2.9) 128 (31.6) NA NA 
 pNx NA 0 (0.0) 9 (3.5) 14 (6.3) 118 (86.8) 46 (11.4) NA NA 
TrainingTestingValidation
VariablePTCRC1RC2NAC (20)TMTTCGA (11)LUND (12)Imvigor210 (24)
n  175 225 256 223 136 405 307 195 
NE-like n (%) 8 (4.6) 4 (1.8) 8 (3.1) 3 (1.4) 9 (6.6) 9 (2.2%) 11 (3.6) 2 (1) 
Age Median (IQR) 70 (47–90)a 69 (63–75) 70 (63–77) 62 (56–71) 70 (62–77) 69 (60–76) NA NA 
Gender Female 41 (23.4%) 41 (18.2) 48 (18.8) 69 (30.9) 27 (20.0) 106 (26.2) NA 42 (21.5) 
 Male 134 (76.6%) 184 (81.8) 208 (81.2) 154 (69.1) 108 (80.0) 299 (73.8) NA 153 (78.5) 
Clinical lymph node stage (%) cN0 NA 0 (0.0) 231 (90.2) 140 (62.8) 60 (44.2) NA NA NA 
 cN1-3 NA 0 (0.0) 20 (7.8) 83 (37.2) 2 (1.4) NA NA NA 
 cNx NA 225 (100.0) 5 (2.0) 0 (0.0) 74 (53.2) NA NA NA 
Clinical tumor stage (%) cTis NA 0 (0.0) 1 (0.4) 0 (0.0) 0 (0.0) NA NA NA 
 cTa NA 0 (0.0) 7 (2.7) 0 (0.0) 0 (0.0) NA NA NA 
 cT1 NA 0 (0.0) 22 (8.6) 0 (0.0) 0 (0.0) NA NA NA 
 cT2 NA 0 (0.0) 162 (63.3) 90 (40.4) 96 (70.6) NA NA NA 
 cT3 NA 0 (0.0) 42 (16.4) 90 (40.4) 33 (24.6) NA NA NA 
 cT4 NA 0 (0.0) 21 (8.2) 43 (19.3) 6 (4.4) NA NA NA 
 cTx NA 225 (100.0) 1 (0.4) 0 (0.0) 1 (0.7) NA NA NA 
Pathologic tumor stage (%)b pT0 NA 0 (0.0) 0 (0.0) 80 (35.9) 4 (2.9) 0 (0) 0 (0.0) NA 
 pTa NA 0 (0.0) 1 (0.4) 3 (1.3) 1 (0.7) 0 (0) 13 (4.2) NA 
 pT1 NA 9 (4.0) 0 (0.0) 13 (5.8) 2 (1.5) 0 (0) 44 (14.3) NA 
 pTis NA 0 (0.0) 0 (0.0) 7 (3.1) 1 (0.7) 0 (0) 1 (0.3) NA 
 pT2 NA 68 (30.2) 72 (28.1) 42 (18.8) 6 (4.4) 131 (32.5) 241 (78.5) NA 
 pT3 NA 102 (45.3) 122 (47.7) 50 (22.4) 11 (8.1) 139 (34.2) 1 (0.3) NA 
 pT4 NA 46 (20.4) 61 (23.8) 24 (10.8) 2 (1.5) 133 (32.8) 1 (0.3) NA 
 pTx NA 0 (0.0) 0 (0.0) 4 (1.8) 109 (80.2) 2 (0.5) 6 (2.0) NA 
Pathologic lymph node stage (%) pN0 NA 145 (64.4) 142 (55.5) 157 (70.4) 14 (10.3) 231 (57) NA NA 
 pN1-3 NA 80 (35.6) 105 (41.0) 52 (23.3) 4 (2.9) 128 (31.6) NA NA 
 pNx NA 0 (0.0) 9 (3.5) 14 (6.3) 118 (86.8) 46 (11.4) NA NA 

aAge >90 censored to protect patient privacy.

bNAC: ypT.

The RC cohort 2 (RC2) consisted of 256 consecutive patients with pT2-4N0-3M0/pT1N1-3M0 urothelial carcinoma of the bladder who underwent RC with extended pelvic lymphadenectomy at Vancouver General Hospital (Vancouver, BC, Canada; 118 cases from 2005 to 2013), UT Southwestern (Dallas, TX; 94 cases from 2007 to 2013) or Johns Hopkins University (Baltimore, MD; 44 cases from 1994 to 2002). In both cohorts, RC tissue was analyzed. The trimodal therapy (TMT) cohort consisted of 136 patients with cT2-T4aN0M0 urothelial carcinoma who were treated with TMT at the Massachusetts General Hospital (Boston, MA) from 1986 to 2013. Overall 475 patients were treated in this time period, but only 136 had adequate transurethral resection of bladder tumor (TURBT) tissue available for analysis. Patient selection criteria for TMT and the standard treatment regimen at this institution have been reported previously (19). The NAC cohort consisted of 223 patients from 7 institutions who received induction or NAC followed by RC for cT2-4aN0-3M0 urothelial carcinoma (20). In both cohorts, pretreatment TURBT tissue was analyzed. The true SC/NE cohort consisted of 5 tumors collected at the Vancouver Prostate Centre, 2 clinical-use cases and 1 collected at the University of Minnesota (Minneapolis, MN).

The retrospective patient cohorts (RC1, RC2, NAC, TMT) were designed to include only cases lacking any component of variant histology other than mixed squamous or glandular differentiation based on the original pathology report at the treating institution. However, the NAC cohort contained 2 patients with a minor component of micropapillary carcinoma who were included in the analysis. A central pathology review was not conducted. All tumors from the PTC cohort were predominantly conventional urothelial carcinoma and all tissue was reviewed to ensure that only conventional urothelial carcinoma was sampled. However, a minor portion of variant histology elsewhere in the tumor cannot be excluded. There were 4 cases in the prospective cohort with a secondary NE variant reported at the source institution.

The human ethics board of each institution approved this study and all patients provided written consent to analysis of their tumor tissues when required by each institution. The studies were conducted in accordance with the ethical guidelines of the Declaration of Helsinki. Array files for profiled bladder tumors are available under Gene Expression Omnibus accession codes, GSE128701 and 128702 (http://www.ncbi.nlm.nih.gov/geo/).

Transcriptome profiling

Whole-transcriptome analysis was performed on formalin-fixed, paraffin-embedded tumor tissue with GeneChip1 Human Exon 1.0 ST Array (Affymetrix) in a Clinical Laboratory Improvement Amendments–certified laboratory (GenomeDx Inc; 21). Microarray data were normalized and genes summarized using single-channel array normalization (22).

Datasets from the public domain

The TCGA RNA sequencing (RNA-seq) data was previously analyzed as part of the TCGA consortium and was available for 405 patients with MIBC treated with RC in the absence of NAC (11). The Lund microarray data was available for 307 patients with non-MIBC and MIBC (12). The Cleveland Clinic SC/NE cohort utilized the HTG EdgeSeq Oncology Biomarker Panel Assay (HTG Molecular Diagnostics) and included 39 pathologic and IHC-confirmed SC/NE cases, 1 metastatic tumor, and 6 matched normal (23). Expression data from primary bladder tumors (n = 195) in IMvigor210, a phase II trial investigating the efficacy of the PD-L1 inhibitor atezolizumab in the second line metastatic setting, was downloaded (24, 25).

Discovery and validation of the NE-like classifier

The prospective cohort was used for initial model training. Genes with high variance (SD > 0.15 and bimodal distribution) were selected by comparing the Bayesian information criterion of a 2-component to a 1-component Gaussian mixture model (26, 27). Hierarchal clustering was used to cluster genes and sequentially remove clusters of genes without clear biological functions in urothelial carcinoma. The selection of genes was based on overlap with TCGA 2017, gene enrichment analysis, literature review, and expert opinion. This resulted in a list of 72 genes corresponding to 4 biological categories: luminal (21 genes), basal (16 genes), immune (17 genes), and stromal (18 genes). As tumors with SC/NE features are typically rare in most cohorts, neuronal genes were not initially selected by our algorithms. As such, we manually added 12 neuronal genes curated from the TCGA (8 features; MS1, PLEKHG4B, GNG4, PEG10, RND2, APLP1, SOX2, TUBB2B; ref. 11) and the literature (4 features; CHGB, ENO2, SV2A, SYP). All neuronal genes, including those used to highlight the neuronal subtype in the TCGA, were carefully reviewed for additional literature support and an association with NE tumors. There was a total of 84 genes in the final gene list (Supplementary Table S1).

The final set of 84 genes and consensus clustering was used to generate a clustering solution with 8 unique clusters (28). One of these 8 clusters had high neuronal gene expression, but was depleted in all other genes, including genes associated with basal and luminal tumors. The clustering labels were further adjusted by Prediction Analysis for Microarrays (PAMR) (R-package) to ensure the tightness of each cluster. A random forest classification model with 2,000 trees was then trained to predict the cluster labels. For this particular study, we focused exclusively on evaluating the NE cluster. The performance of this classifier was first evaluated in the testing cohort (RC1), where it was shown to be better than other classifier choices, such as PAMR or logistic regression (29). The finalized model was then locked and evaluated on a set of 7 true SC/NE cases and in 6 independent validation cohorts of conventional urothelial carcinoma. In addition, to test the performance of the model, a cohort of true SC/NE tumors was used.

Model porting to RNA-seq platforms

The NE model was ported to the RNA-seq platform (TCGA) using quantile normalization and imputation. For the 82 of 84 genes shared by the RNA-seq and microarray platforms, their expression values from RNA-seq were normalized by quantile–quantile matching with the expression values in our training cohort (PTC, n = 175) as implemented in R package preprocess Core (30). For the 2 genes not captured by the RNA-seq platform, linear regression models with lasso penalty were discovered on the training cohort and applied to the quantile normalized RNA-seq data to impute their expression values.

IHC staining and antibodies

IHC was performed on a subset of 10 NE-like tumors from the TMT (n = 7) and RC2 (n = 3) cohorts, and 5 true retrospective NE tumors. A tissue microarray (TMA) was created from 8 tumors (3 NE-like and 5 true NE tumors) and individual sections were used for the other 7 tumors (TMT cohort) in which not enough tissue was available for a TMA. The positive control was NE prostate cancer and the negative control was benign kidney. IHC was performed with a panel of antibodies against 6 classical NE markers (CgA, SYP, CD56, PEG10, SOX2, ENO2; Supplementary Table S2). All stained slides were digitalized with the SL801 autoloader and Leica SCN400 scanning system (Leica Microsystems; Concord, Ontario, Canada) at magnification equivalent to 20×. The images were subsequently stored in the SlidePath digital imaging hub (Leica Microsystems) of the Vancouver Prostate Centre. Values on a 4-point scale were assigned to each image. Descriptively, 0 represents no staining, 1 represents low, but detectable degree of staining, 2 represents clearly positive cases, and 3 represents strong expression. IHC was quantified for staining intensity (0–3).

Drug response score

Using in vitro drug sensitivity and microarray data from the NCI-60 panel, we generated gene signatures predicting tumor sensitivity to 4 drugs used in the treatment of bladder cancer, including cisplatin, etoposide, gemcitabine, and ifosamide (31). For each drug, we used the CellMiner tool (32) to identify drug response–related genes correlated to the IC50 value of the NCI-60 cell lines. The most significantly correlated genes were selected and the expression of the corresponding genes in the prospective cohort were extracted for drug response score (DRS) calculations. A patient-specific DRS was calculated using these correlation coefficients (Cor) as weighting factors of the corresponding gene expression normalized by the sum of Cor.

Statistical analysis

The Kaplan–Meier method was used for estimates of overall survival (OS). OS was defined as the moment of first diagnosis of MIBC till death. Patients who were lost to follow-up were censored at the date of last contact. The differences between the OS curves of the NE-like patients compared with all other patients were tested using the log-rank test. Multivariable Cox proportional hazard models were used to demonstrate the effect of NE-like subtype on OS-adjusting common clinical variables, including age, sex, and stage. All model training and statistical analyses were performed in R (R Core Team, 2018; R: A language and environment for statistical computing; R Foundation for Statistical Computing; https://www.R-project.org/).

Development of a single-sample classifier to identify NE-like tumors

To develop the NE classifier, we leveraged both prospective and retrospective MIBC genome-wide expression datasets for model training and testing, respectively (Fig. 1). The clinical and pathologic features of the patient cohorts are described in Table 1. The initial model training used a set of 84 genes that were selected from the intersection of literature-curated, MIBC-associated genes (Supplementary Table S1), and the MIBC subtyping model described by the TCGA (11) to identify NE-like tumors. In the PTC of 175 patients with phenotypical urothelial carcinoma, we identified 8 (4.2%) NE-like tumors with gene expression profiles consistent with histologic SC/NE tumors (Supplementary Fig. S1). These patients showed robust, but heterogeneous, expression of various neuronal markers (i.e., NESTIN, TUBB2B, PEG10). Conversely, these patients showed low or no expression of either basal (i.e., KRT5/6, KRT14) or luminal (i.e., PPARG, KRT20) markers. Although the NE-like model includes genes capturing immune and stromal infiltration, here we focused specifically on the characterization of NE-like tumors, using the established basal and luminal axis as a reference. We evaluated the classifier on a testing cohort of 225 patients who underwent RC (RC1), classifying 4 patients as NE-like.

Figure 1.

Summary flowchart of NE classifier development. A cohort of 175 prospective cases was used for initial model training using a set of 84 genes selected from machine-learning approaches and literature review. A clustering solution was used to train a random forest model, which was tested on an additional cohort (RC1) and retrained if necessary. The final model was locked and proceeded to validation on 6 additional cohorts, including the TCGA MIBC cohort and a set of pathologic NE MIBC cases. Candidate NE-like cases were stained with neuronal IHC markers. See main text for additional details.

Figure 1.

Summary flowchart of NE classifier development. A cohort of 175 prospective cases was used for initial model training using a set of 84 genes selected from machine-learning approaches and literature review. A clustering solution was used to train a random forest model, which was tested on an additional cohort (RC1) and retrained if necessary. The final model was locked and proceeded to validation on 6 additional cohorts, including the TCGA MIBC cohort and a set of pathologic NE MIBC cases. Candidate NE-like cases were stained with neuronal IHC markers. See main text for additional details.

Close modal

Histologically SC tumors validate feature selection in NE-like classifier

The model performance was evaluated on a set of 7 MIBC tumors with SC/NE histology (Supplementary Fig. S2) where 100% of the cases were identified as NE-like by the model, supporting the accuracy of the classifier. We also evaluated an independent single-institution cohort of histologically confirmed NE tumors (n = 39) and matched adjacent normal bladder tissue (n = 6) profiled on a targeted RNA-seq platform (23). Despite incomplete feature overlap as a consequence of cross platform comparison, we found that a subset of NE markers was strongly upregulated relative to matched normal tissues, whereas conventional urothelial carcinoma markers were downregulated (Supplementary Fig. S3). Furthermore, 4 patients in the PTC had a histologic component of NE tumor separate from the urothelial component that was sampled for transcriptomic profiling. Three of these 4 cases demonstrated NE-like gene expression. These data provide external supporting evidence of appropriate feature selection for the NE classifier.

NE-like tumors are defined by unique biological characteristics

To further validate the NE-like classifier, we applied our model to 3 additional cohorts of patients with phenotypical urothelial carcinoma who underwent different treatments, including a RC-only cohort (RC2; n = 256), a NAC (plus RC) cohort (NAC; n = 223), and a TMT cohort (n = 136). In all 3 cohorts, we identified NE-like tumors ranging in frequency from 1% to 6.6% (Table 1). Consistent with the observations in the training and testing sets, the NE-like tumors showed weak or no expression of basal or luminal markers, although the non–NE-like tumors in the same cohorts expressed these genes robustly. In contrast, the NE-like tumors robustly expressed a suite of neuronal markers, although there was heterogeneity in the expression of individual markers between tumors (Fig. 2). In general, across all cohorts, the expression of these neuronal markers was much higher in the NE-like tumors, compared with tumors with basal- or luminal-like profiles (Fig. 2).

Figure 2.

Forced-order heatmap for 3 biological categories (luminal, basal, and neuronal) of selected MIBC marker genes in RC2 cohort (A), NAC cohort (B), and TMT cohort (C). Cases identified as NE-like are boxed.

Figure 2.

Forced-order heatmap for 3 biological categories (luminal, basal, and neuronal) of selected MIBC marker genes in RC2 cohort (A), NAC cohort (B), and TMT cohort (C). Cases identified as NE-like are boxed.

Close modal

Deletion of CDKN2A and amplification of CCND1 are commonly found in conventional urothelial carcinoma, whereas the inverse has been reported for phenotypic NE cancers (33–35). The NE-like cases in our study had profiles consistent with these latter reports, including higher expression of CDKN2A and lower expression of CCDN1 relative to basal or luminal cases (Supplementary Fig. S4).

We next evaluated the performance of our NE-like classifier in the TCGA (n = 408) and Lund (n = 307) cohorts (Fig. 3). In the TCGA cohort, we identified 9 of 20 patients of the TCGA neuronal subtype to be NE-like (Fig. 3A). The 11 cases not classified as NE by our model expressed lower levels of neuronal markers compared with the 9 NE-like cases (Supplementary Fig. S5). In addition, 4 of these tumors strongly expressed either basal (n = 3) or luminal markers (n = 1). As the NE-like classifier was trained to exclude non-NE subtypes and enrich for robust neuronal marker expression, with only minimal basal and/or luminal marker expression, this was an expected observation and illustrates the stringency of the NE-like classifier. Despite efforts to exclude nonurothelial carcinoma cases, 3 SC bladder cases were included in the TCGA cohort. All 3 of these cases were identified as NE-like by our classifier.

Figure 3.

Forced-order heatmap for 3 biological categories (luminal, basal, and neuronal) of selected MIBC marker genes in the TCGA cohort (A), Lund cohort (B), and Imvigor210 cohort (C). Cases identified as NE-like are boxed.

Figure 3.

Forced-order heatmap for 3 biological categories (luminal, basal, and neuronal) of selected MIBC marker genes in the TCGA cohort (A), Lund cohort (B), and Imvigor210 cohort (C). Cases identified as NE-like are boxed.

Close modal

In the Lund cohort, we identified 11 of 24 cases flagged as SC/NE or SC/NE-like by the LundTax model, with no overlap with the SC/NE - GU class (Fig. 3B). One patient was identified as NE-like by our model, but not by the LundTax model. As was observed for the TCGA cohort, the nonoverlapping NE-like cases tended to have lower expression of neuronal genes and/or coexpressed basal or luminal markers. When comparing the NE-like calls with the LundTax IHC results, we find that 8 of 11 of the NE-like calls had strong staining for several of the profiled markers, including CD56 (NCAM1), SYP, and TUBB2B (Supplementary Table S3). However, when comparing with the LundTax tumor cell phenotype assessment of SC/NE, our model identified 5 of 11 cases as NE-like (Supplementary Table S3).

In the IMvigor210 cohort, we identified 2 of 195 cases as being NE-like. Consistent with the other validation cohorts, the gene expression profile of these 2 patients (Fig. 3C) showed high neuronal gene expression with low basal and luminal gene expression.

The identification of NE-like tumors with unique gene expression profiles (high neuronal genes, low basal and luminal genes) across 6 independent cohorts strongly supported the identification of a NE-like tumor subtype, independent of SC/NE histology. The clinical characteristics of all NE-like tumors in the testing cohort and 6 validation cohorts are summarized in Supplementary Table S4.

Confirmation of neuronal gene expression in NE-like tumors by IHC staining

To confirm the neuronal character of the NE-like cases at the protein level, we selected tumors classified as NE-like by our model in the retrospective cohorts that had tissue available for IHC analysis (RC2, 3; TMT, 7). As these tumors were selected for retrospective studies exploring conventional urothelial carcinoma, we did not expect to observe variant histology. Consistent with this expectation, these tumors were confirmed as conventional urothelial carcinoma by histologic criteria, with no obvious SC or NE features identified. IHC staining was performed with a panel of antibodies against 6 classical NE markers [chromogranin A (CgA), synaptophsin (SYP), CD56, PEG10, SOX2, and ENO2; Supplementary Fig. S2). In Fig. 4, we show a representative subset of tumors, including 1 true NE tumor and 2 NE-like tumors. The NE tumor (Fig. 4A) had nests of small round malignant cells distributed evenly with “salt and pepper chromatin,” as well as the typical crush artefact often observed in these stroma-poor tumors. These histologic characteristics were absent in the NE-like tumors, which instead show irregular clusters of larger tumor cells with nuclear pleomorphism invading deep into the wall. The true NE tumor had strong staining for most markers, whereas the NE-like cases were positive for a subset of neuronal markers, with reduced staining intensity. In a broader panel of tumors (Supplementary Fig. S2 B), some NE-like tumors had IHC profiles that were similar to true NE tumors, whereas others show more modest staining intensity for various NE markers. Like the transcript-level data, the neuronal marker staining was heterogeneous across the 10 NE-like patients, and no specific pattern was observed.

Figure 4.

H&E and IHC images (SYP, CgA, CD56, ENO2, SOX2, PEG10) of NE bladder cancer (A) and 2 cases of NE-like bladder cancer (B and C). Dotted line boxes represent low magnified images of corresponding adjacent high magnification view. Scale bar, 100 μm. The IHC staining scores and gene expression values are indicated above each image.

Figure 4.

H&E and IHC images (SYP, CgA, CD56, ENO2, SOX2, PEG10) of NE bladder cancer (A) and 2 cases of NE-like bladder cancer (B and C). Dotted line boxes represent low magnified images of corresponding adjacent high magnification view. Scale bar, 100 μm. The IHC staining scores and gene expression values are indicated above each image.

Close modal

NE-like tumors have poor prognosis

In general, patients with histologic SC/NE tumors have aggressive disease and poor outcomes (2, 3, 5, 13). To determine whether NE-like tumors have a similarly poor prognosis, we performed both univariable and multivariable OS analysis stratified by NE-like status in all 4 retrospective cohorts.

In the cohort of patients who underwent cystectomy without preceding chemotherapy (RC2), NE-like patients had significantly worse outcomes, with 1-year progression-free survival of 65% compared with 82% in the rest of the cohort (P = 0.046; Fig. 5B). We did not capture which adjuvant and salvage therapies these patients received. All 8 NE-like patients had died within 30 months. NE-like classification correlated strongly with death on multivariable analysis adjusted for age, sex, T-stage, and lymph node involvement [adjusted HR, 6.03 (2.68–13.58, P < 0.001); Table 2].

Figure 5.

Prognosis of NE-like patients in different treatment settings. KM plots for RC1 cohort (A; n = 225), RC2 cohort (B; n = 255), NAC cohort (C; n = 223), TMT cohort (D; n = 136), TCGA RC cohort (E; n = 402), and Imvigor210 cohort (F,n = 195).

Figure 5.

Prognosis of NE-like patients in different treatment settings. KM plots for RC1 cohort (A; n = 225), RC2 cohort (B; n = 255), NAC cohort (C; n = 223), TMT cohort (D; n = 136), TCGA RC cohort (E; n = 402), and Imvigor210 cohort (F,n = 195).

Close modal
Table 2.

Multivariable analysis of OS on the RC2 surgery-only cohort

VariableUnadjusted HRPAdjusted HRP
Age 1.02 (1.00–1.04) 0.023a 1.02 (1.01–1.04) 0.004a 
Gender: male 0.97 (0.66–1.45) 0.9 0.91 (0.61–1.36) 0.644 
>pT2 1.34 (0.94–1.90) 0.107 1.14 (0.79–1.65) 0.493 
pN1–3 2.35 (1.70–3.26) <0.001a 2.45 (1.76–3.41) <0.001a 
NE-like 3.88 (1.81–8.34) <0.001a 6.03 (2.68–13.58) <0.001a 
VariableUnadjusted HRPAdjusted HRP
Age 1.02 (1.00–1.04) 0.023a 1.02 (1.01–1.04) 0.004a 
Gender: male 0.97 (0.66–1.45) 0.9 0.91 (0.61–1.36) 0.644 
>pT2 1.34 (0.94–1.90) 0.107 1.14 (0.79–1.65) 0.493 
pN1–3 2.35 (1.70–3.26) <0.001a 2.45 (1.76–3.41) <0.001a 
NE-like 3.88 (1.81–8.34) <0.001a 6.03 (2.68–13.58) <0.001a 

aP <0.05.

There were only 3 NE-like patients in the NAC cohort; 2 were treated with gemcitabine/cisplatin (Gem/Cis) and 1 with methotrexate/vinblastine/adriamycin/cisplatin (MVAC). NE-like classification correlated with death [adjusted HR, 58.18 (10.94–309.28, P < 0.001)] in this cohort (Table 3). Two patients died within 2 months of surgery (Fig. 5C), including 1 elderly patient who died from metastatic progression shortly after surgery without any salvage therapy and 1 who died due to postoperative complications. In contrast to the cystectomy-only and NAC cohorts, the NE-like patients in the TMT setting showed no significant difference with respect to 1-year progression-free survival compared with patients with urothelial carcinoma (Fig. 5D). Although in the other cohorts, all the NE-patients had died within 30 months of follow-up, in the TMT cohort (Table 4) 7 of 9 NE-like patients were still alive at 30 months [(adjusted HR of 0.87 (0.35–2.15, P = 0.458)]. The stage of NE-like tumors was comparable with the patients with urothelial carcinoma. In this cohort, 2 of 9 patients with NE-like tumors underwent salvage cystectomy, compared with 30 of 127 with non–NE-like tumors.

Table 3.

Multivariable analysis of OS on the NAC cohort

VariableUnadjusted HRPAdjusted HRP
Age 1.02 (1.00–1.05) 0.026a 1.03 (1.01–1.06) 0.005a 
Gender: male 0.97 (0.60–1.56) 0.899 0.83 (0.50–1.39) 0.485 
>pT2 1.56 (0.98–2.49) 0.062 1.84 (1.12–3.03) 0.017a 
pN1–3 4.06 (2.55–6.47) <0.001a 4.91 (3.05–7.91) <0.001a 
NE-like 58.18 (10.94–309.28) <0.001a 105.32 (18.78–590.71) <0.001a 
VariableUnadjusted HRPAdjusted HRP
Age 1.02 (1.00–1.05) 0.026a 1.03 (1.01–1.06) 0.005a 
Gender: male 0.97 (0.60–1.56) 0.899 0.83 (0.50–1.39) 0.485 
>pT2 1.56 (0.98–2.49) 0.062 1.84 (1.12–3.03) 0.017a 
pN1–3 4.06 (2.55–6.47) <0.001a 4.91 (3.05–7.91) <0.001a 
NE-like 58.18 (10.94–309.28) <0.001a 105.32 (18.78–590.71) <0.001a 

aP <0.05.

Table 4.

Multivariable analysis of OS on the TMT cohort

VariableUnadjusted HRPAdjusted HRP
Age 1.02 (1.00–1.04) 0.058 1.02 (1.00–1.04) 0.015a 
Gender: male 1.19 (0.66–2.18) 0.562 1.40 (0.75–2.60) 0.289 
>cT2 2.23 (1.42–3.50) <0.001a 2.24 (1.41–3.55) <0.001a 
Hydrophr. 1.62 (0.92–2.87) 0.096 1.55 (0.85–2.81) 0.15 
NE-like 0.87 (0.35–2.15) 0.759 0.70 (0.28–1.78) 0.458 
VariableUnadjusted HRPAdjusted HRP
Age 1.02 (1.00–1.04) 0.058 1.02 (1.00–1.04) 0.015a 
Gender: male 1.19 (0.66–2.18) 0.562 1.40 (0.75–2.60) 0.289 
>cT2 2.23 (1.42–3.50) <0.001a 2.24 (1.41–3.55) <0.001a 
Hydrophr. 1.62 (0.92–2.87) 0.096 1.55 (0.85–2.81) 0.15 
NE-like 0.87 (0.35–2.15) 0.759 0.70 (0.28–1.78) 0.458 

aP <0.05.

The outcomes for NE-like patients in the TCGA cohort, who also underwent cystectomy without NAC, were equally poor, with 1-year progression-free survival rates of 57%, compared with 80% for the rest of the cohort (Fig. 5E). None of the 9 NE-like patients in the TCGA cohort survived beyond 30 months of follow-up [adjusted HR 3.80 (1.76–8.23, P < 0.001); Fig. 5E; Table 5]. In the IMvigor210 cohort (Fig. 5F), the outcomes of the 2 NE-like patients was similar to the non-NE patients (P = 0.931).

Table 5.

Multivariable analysis of OS on the TCGA RC cohort

VariableUnadjusted HRPAdjusted HRP
Age 1.04 (1.02–1.05) <0.001a 1.04 (1.02–1.05) <0.001a 
Gender: male 0.89 (0.64–1.23) 0.48 0.88 (0.63–1.22) 0.44 
>pT2 2.14 (1.48–3.09) <0.001a 1.61 (1.07–2.43) 0.023a 
pN1–3 2.10 (1.56–2.83) <0.001a 1.70 (1.22–2.38) 0.002a 
NE-like 2.67 (1.25–5.71) 0.011a 3.80 (1.76–8.23) <0.001a 
VariableUnadjusted HRPAdjusted HRP
Age 1.04 (1.02–1.05) <0.001a 1.04 (1.02–1.05) <0.001a 
Gender: male 0.89 (0.64–1.23) 0.48 0.88 (0.63–1.22) 0.44 
>pT2 2.14 (1.48–3.09) <0.001a 1.61 (1.07–2.43) 0.023a 
pN1–3 2.10 (1.56–2.83) <0.001a 1.70 (1.22–2.38) 0.002a 
NE-like 2.67 (1.25–5.71) 0.011a 3.80 (1.76–8.23) <0.001a 

aP <0.05.

Overall, the NE-like patients showed significantly worse outcomes in the surgical cohorts, which is consistent with the poor prognostic features that have been described for pathologically diagnosed NE tumors.

Predicting treatment response to chemotherapy for NE-like tumors

Since conventional urothelial carcinoma is usually treated with Gem/Cis or MVAC, but SC/NE bladder cancer is typically treated with cisplatin/etoposide (EP/PE; ref. 18), we sought to explore the potential drug sensitivities of the NE-like tumors. Using the CellMiner analysis tool that contains gene expression and IC50 data for the NCI-60 cancer cell lines treated with a panel of oncology drugs, we generated DRS to establish a numerical metric for predicting sensitivity for each tumor. Applying this to our training (PTC) and retrospective cohorts (RC2, NAC, TMT), we observed that NE-like patients had higher predicted sensitivity scores to drugs relevant to MIBC treatment, including cisplatin, gemcitabine, and etoposide when compared with basal and luminal patients (Supplementary Fig. S6). Consistent with reports of basal and luminal sensitivity to cisplatin, basal tumors were predicted to have higher sensitivity to cisplatin compared with luminal tumors. To confirm these results, we applied the DRS model to the TCGA and Lund datasets and observed similar patterns (Supplementary Figs. S7 and S8).

Transcriptome profiling of MIBC has revealed several unique molecular subtypes with distinct biological and clinical features. In 2 recent studies, a neuronal or SC/NE subtype has been defined, having robust neuronal marker expression and poor outcomes (11, 12). Notably, the majority of the tumors classified as NE-like did not have histologic features consistent with this tumor type, but were phenotypically similar to conventional urothelial carcinoma. This suggests NE-like gene expression profiles, in conjunction with pathologic examination, can provide additional information for identifying tumors with NE biology.

Here, we present a single-sample NE-like classifier capable of identifying patients with NE-like biology regardless of their underlying pathologic presentation. We applied this classifier to multiple unique patient cohorts and identified NE-like tumors in similar proportions in all cohorts. Our classifier proved to be more stringent than the TCGA and Lund subtyping models, categorizing 9 of 20 of the TCGA neuronal cases and 11 of 24 of the Lund SC/NE cases as NE-like (Supplementary Table S2). The NE classifier was constructed with a strict definition of NE, particularly in having robust expression of neuronal markers with minimal expression of basal and luminal markers. This was done to maximize high-confidence calls to identify the tumors most likely to harbor NE-like biology. Moreover, porting a model from 1 platform (microarray) to another very different platform (RNA-seq) can introduce confounders, which may impact the model performance and, therefore, these results should be applied judiciously.

The NE-like tumors identified by our classifier demonstrated poor OS across cohorts, similar to the clinical outcomes of patients with pathologically confirmed NE tumors. The NE-like tumors in the TMT cohort were a noteworthy exception, as they had similar OS to conventional urothelial carcinoma. Although the clinical and pathologic parameters of these patients were comparable with the NE-like patients in the surgical cohorts, selection of more favorable patients for TMT cannot be excluded. The NE-like patients in the TMT cohort received radiosensitizing chemotherapy, but not at a dose that would have been expected to affect disease outside the field of radiation. The NAC cohort fared poorly despite receiving NAC, but this cohort contained only 3 NE-like patients, and 1 died from postoperative complications, so it is unclear what impact the addition of chemotherapy had to the better outcome in the TMT cohort. The NE-like patients in the NAC cohort also presented with more advanced stage. Such comparisons across different cohorts are problematic and are limited by the very small number of cases in each cohort. Because the primary risk of recurrence after loco-regional therapy for SC/NE carcinoma of the bladder is systemic, one could postulate that the same will be true for NE-like tumors. This would make systemic chemotherapy a particularly important component of the treatment algorithm, but loco-regional treatment with either TMT or RC is likely to lead to similar outcomes (36, 37).

Systemic chemotherapy in the form of cisplatin and etoposide is usually favored for patients with histologic NE tumors (14). Therefore, the question must now be posed whether NE-like tumors based on gene expression should receive the same treatment. Preliminary analysis using DRS metrics suggested that NE-like tumors are more likely to respond to cisplatin, gemcitabine, and/or etoposide than conventional urothelial carcinoma. However, predicted response does not necessarily translate into a durable response with survival benefit, and the DRS model did not suggest that EP/PE would be superior to cisplatin/gemcitabine. Regardless, the relative rarity of this tumor entity will make it difficult to study this clinical question.

The detection of tumors that resemble a specific histologic subtype by gene expression, but are classified as conventional urothelial carcinoma by morphologic criteria, suggests that morphology alone may be inadequate in identifying variant biology. It is well described in the pathology literature that identification of variant histologic subtypes is subjective and highly susceptible to interobserver variability (38, 39). Although the NE-like gene expression pattern is particularly striking and therefore straightforward to identify in a larger patient cohort, it remains to be determined how many other variant histologic subtypes can also be readily identified by gene expression profiling. Early evidence suggests that the clinically aggressive micropapillary variant of bladder cancer has a unique gene expression profile driven by the activation of miR-296 and RUVBL1 target genes (18). Moreover, sarcomatoid transformation may be driven by the initiation of an epithelial-to-mesenchymal transition transcriptional program, potentially offering a signature to capture this variant as well (18, 40).

The different histologic subtypes of bladder cancer frequently occur together with conventional urothelial carcinoma, and the usual dictum is that these subtypes share a common origin with the concurrent urothelial carcinoma (33, 34). It is, therefore, conceivable that the NE-like tumors with conventional urothelial morphology but a NE transcriptional program represent an intermediate stage in the evolution towards a histologic NE phenotype. An analogous transdifferentiation pathway has been described for the sarcomatoid variant of bladder cancer (41). In this study, rather than expressing classic “endpoint” markers of terminal NE differentiation (i.e., CgA, SYP), the NE-like tumors expressed intermediate neuronal markers (i.e., SOX2, PEG10) while demonstrating urothelial differentiation. Taken together, these data suggest the NE-like tumors are differentiating towards a NE phenotype. With early detection, it is conceivable that timely intervention may interrupt this transition towards histologic NE carcinoma. The concept that the NE-like tumors represent a transition state is supported by the finding that the conventional urothelial component of 3 of 4 PTC cases with a secondary NE variant region was classified as NE-like by our classifier. We know from a recent molecular characterization of SC bladder cancer (42) that urothelial carcinoma and NE are genomically (i.e., mutational and copy number changes) similar but transcriptomically different, as has also been observed in prostate and lung cancer (43). The transition towards NE is supported by the loss of RB1 and TP53 in NE tumors (44), but the transcriptional regulators that lead to the transcriptomic and ultimately the phenotypic shift are poorly understood. The plasticity of this transition appears to go in both directions, because local recurrences in the bladder after chemoradiation for muscle-invasive SC carcinoma are frequently conventional urothelial carcinoma (45).

This study is limited by the lack of central pathology review of all cases, although the majority of cases were assessed by dedicated genitourinary pathologists at the respective tertiary care centers and the NE-like cases were centrally reviewed. The prospectively collected clinical-use cases underwent stringent sample processing with only the muscle-invasive urothelial component of the tumor being selected and profiled. Four of these PTC cases had secondary NE variant histology in other regions of the tumor that were not sampled. An additional limitation is the retrospective nature of the study. Prospective validation will be necessary before this classification can be used clinically.

In summary, we report the development and validation of a robust single-sample genomic classifier trained to identify NE-like patients, regardless of histologic presentation.

E. Davicioni has ownership interests (including patents) at GenomeDx. J.L. Wright reports receiving speakers bureau honoraria from Onc Live and Seattle Cancer Care Alliance, reports receiving commercial research grants from Merck, Altor Biosciences, Nucleix, Movember, and receives royalties from UpToDate. J.L. Boormans is a consultant/advisory board member for Janssen Pharmaceuticals, Roche, and Merck, and reports receiving commercial research grants from GenomeDx. B.W.G. van Rhijn is a consultant/advisory board member for Ferring and Astra Zeneca. P. Grivas reports receiving speakers bureau honoraria from Genentech and Bristol-Myers Squibb, is a consultant/advisory board member for Merck, Genentech, Dendreon, Bristol-Myers Squibb, Astra Zeneca, Pfizer, EMD Serono, Clovis Oncology, Biocept, Jansssen, QED Therapeutics, Heron Therapeutics, Driver Inc., Seattle Genetics, Foundation Medicine, Exelixis, and reports receiving commercial research support from Pfizer, Clovis Oncology, Bavarian Nordic and Immunomedics. K.W. Mouw is a consultant/advisory board member for Pfizer, EMD Serono, and reports receiving commercial research support from Pfizer. R. Seiler has ownership interests (including patents) at Classifier. J.A. Efstathiou reports receiving speakers bureau honoraria from Genentech, Janssen, EMD Serono, and Bayer Healthcare, and is a consultant/advisory board member for Blue Earth Diagnostics and Taris Biomedical. P.C. Black has ownership interests (including patents) at GenomeDx. No potential conflicts of interest were disclosed by the other authors.

Conception and design: E.A. Gibb, Y. Liu, M.A. Dall’Era, C.-L. Wu, B.R. Konety, Y. Lotan, P.C. Black

Development of methodology: E.A. Gibb, Y. Liu, M. Alshalalfa, P.C. Black

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): J. Batista da Costa, T.J. Bivalacqua, Y. Liu, H.Z. Oo, D.T. Miyamoto, E. Davicioni, J.L. Wright, M.A. Dall’Era, J. Douglas, M.S. van der Heijden, C.-L. Wu, B.W.G. van Rhijn, P. Grivas, P. Murugan, B.R. Konety, R. Seiler, O.Y. Mian, J.A. Efstathiou, Y. Lotan, P.C. Black

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): J. Batista da Costa, E.A. Gibb, T.J. Bivalacqua, Y. Liu, H.Z. Oo, M. Alshalalfa, E. Davicioni, J.L. Wright, M.A. Dall’Era, B.W.G. van Rhijn, B.R. Konety, R. Seiler, S. Daneshmand, O.Y. Mian, J.A. Efstathiou, Y. Lotan, P.C. Black

Writing, review, and/or revision of the manuscript: J. Batista da Costa, E.A. Gibb, T.J. Bivalacqua, Y. Liu, H.Z. Oo, D.T. Miyamoto, E. Davicioni, M.A. Dall’Era, J. Douglas, J.L. Boormans, M.S. van der Heijden, B.W.G. van Rhijn, S. Gupta, P. Grivas, K.W. Mouw, P. Murugan, S. Ra, B.R. Konety, R. Seiler, S. Daneshmand, O.Y. Mian, J.A. Efstathiou, Y. Lotan, P.C. Black

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): Y. Liu, E. Davicioni, J.L. Wright, O.Y. Mian, P.C. Black

Study supervision: E.A. Gibb, T.J. Bivalacqua, J.A. Efstathiou, P.C. Black

Other (contributed to data discussion and context): P. Grivas

Other (pathology): L. Fazli

J. Batista da Costa's salary was partially funded by the Association Française d'Urologie. We are grateful for the insightful comments provided by A. Gordon Robertson on this manuscript.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.

1.
Ploeg
M
,
Aben
KK
,
Hulsbergen-van de Kaa
CA
,
Schoenberg
MP
,
Witjes
JA
,
Kiemeney
LA
. 
Clinical epidemiology of nonurothelial bladder cancer: analysis of the Netherlands Cancer Registry
.
J Urol
2010
;
183
:
915
20
.
2.
Choong
NWW
,
Quevedo
JF
,
Kaur
JS
. 
Small cell carcinoma of the urinary bladder: the mayo clinic experience
.
Cancer
2005
;
103
:
1172
8
.
3.
Quek
ML
,
Nichols
PW
,
Yamzon
J
,
Daneshmand
S
,
Miranda
G
,
Cai
J
, et al
Radical cystectomy for primary neuroendocrine tumors of the bladder: the University of Southern California experience
.
J Urol
2005
;
174
:
93
6
.
4.
Moschini
M
,
D’Andrea
D
,
Korn
S
,
Irmak
Y
,
Soria
F
,
Compérat
E
, et al
Characteristics and clinical significance of histological variants of bladder cancer
.
Nat Rev Urol
2017
;
14
:
651
68
.
5.
Cheng
L
,
Pan
C-X
,
Yang
XJ
,
Lopez-Beltran
A
,
MacLennan
GT
,
Lin
H
, et al
Small cell carcinoma of the urinary bladder
.
Cancer
2004
;
101
:
957
62
.
6.
Pasquier
D
,
Barney
B
,
Sundar
S
,
Poortmans
P
,
Villa
S
,
Nasrallah
H
, et al
Small cell carcinoma of the urinary bladder: a retrospective, multicenter rare cancer network study of 107 patients
.
Int J Radiat Oncol Biol Phys
2015
;
92
:
904
10
.
7.
Fischer-Valuck
BW
,
Rao
YJ
,
Henke
LE
,
Rudra
S
,
Hui
C
,
Baumann
BC
, et al
Treatment patterns and survival outcomes for patients with small cell carcinoma of the bladder
.
Eur Urol Focus
2018
;
4
:
900
6
.
8.
Choi
W
,
Porten
S
,
Kim
S
,
Willis
D
,
Plimack
ER
,
Hoffman-Censits
J
, et al
Identification of distinct basal and luminal subtypes of muscle-invasive bladder cancer with different sensitivities to frontline chemotherapy
.
Cancer Cell
2014
;
25
:
152
65
.
9.
Sjödahl
G
,
Lauss
M
,
Lövgren
K
,
Chebil
G
,
Gudjonsson
S
,
Veerla
S
, et al
A molecular taxonomy for urothelial carcinoma
.
Clin Cancer Res
2012
;
18
:
3377
86
.
10.
Cancer Genome Atlas Research Network
. 
Comprehensive molecular characterization of urothelial bladder carcinoma
.
Nature
. 
2014
;
507
:
315
22
.
11.
Robertson
AG
,
Kim
J
,
Al-Ahmadie
H
,
Bellmunt
J
,
Guo
G
,
Cherniack
AD
, et al
Comprehensive molecular characterization of muscle-invasive bladder cancer
.
Cell
2017
;
171
:
540
56
.
12.
Sjödahl
G
,
Eriksson
P
,
Liedberg
F
,
Höglund
M
. 
Molecular classification of urothelial carcinoma: global mRNA classification versus tumour-cell phenotype classification
.
J Pathol
2017
;
242
:
113
25
.
13.
Jung
K
,
Ghatalia
P
,
Litwin
S
,
Horwitz
EM
,
Uzzo
RG
,
Greenberg
RE
, et al
Small-cell carcinoma of the bladder: 20-year single-institution retrospective review
.
Clin Genitourin Cancer
2017
;
15
:
e337
43
.
14.
Lynch
SP
,
Shen
Y
,
Kamat
A
,
Grossman
HB
,
Shah
JB
,
Millikan
RE
, et al
Neoadjuvant chemotherapy in small cell urothelial cancer improves pathologic downstaging and long-term outcomes: results from a retrospective study at the md anderson cancer center
.
Eur Urol
2013
;
64
:
307
13
.
15.
Siefker-Radtke
AO
,
Kamat
AM
,
Grossman
HB
,
Williams
DL
,
Qiao
W
,
Thall
PF
, et al
Phase II clinical trial of neoadjuvant alternating doublet chemotherapy with ifosfamide/doxorubicin and etoposide/cisplatin in small-cell urothelial cancer
.
J Clin Oncol
2009
;
27
:
2592
7
.
16.
Meijer
RP
,
Meinhardt
W
,
van der Poel
HG
,
van Rhijn
BW
,
Kerst
JM
,
Pos
FJ
, et al
Local control rate and prognosis after sequential chemoradiation for small cell carcinoma of the bladder
.
Int J Urol
2013
;
20
:
778
84
.
17.
Sorensen
M
,
Pijls-Johannesma
M
,
Felip
E
. 
Small-cell lung cancer: ESMO clinical practice guidelines for diagnosis, treatment and follow-up
.
Ann Oncol
. 
2010
;
21
:
v120
5
.
18.
Guo
CC
,
Dadhania
V
,
Zhang
L
,
Majewski
T
,
Bondaruk
J
,
Sykulski
M
, et al
Platinum priority-bladder cancer gene expression profile of the clinically aggressive micropapillary variant of bladder cancer
.
Eur Urol
2016
;
70
:
611
20
.
19.
Efstathiou
JA
,
Spiegel
DY
,
Shipley
WU
,
Heney
NM
,
Kaufman
DS
,
Niemierko
A
, et al
Long-term oputcomes of selective bladder preservation by combined-modality therapy for invasive bladder cancer: the MGH experience
.
Eur Urol
2012
;
61
:
705
11
.
20.
Seiler
R
,
Ashab
HAD
,
Erho
N
,
van Rhijn
BWG
,
Winters
B
,
Douglas
J
, et al
Impact of molecular subtypes in muscle-invasive bladder cancer on predicting response and survival after neoadjuvant chemotherapy
.
Eur Urol
2017
;
72
:
544
54
.
21.
Erho
N
,
Crisan
A
,
Vergara
IA
,
Mitra
AP
,
Ghadessi
M
,
Buerki
C
, et al
Discovery and validation of a prostate cancer genomic classifier that predicts early metastasis following radical prostatectomy
.
PLoS One
2013
;
8
:
e66855
.
22.
Piccolo
SR
,
Sun
Y
,
Campbell
JD
,
Lenburg
ME
,
Bild
AH
,
Johnson
WE
. 
A single-sample microarray normalization method to facilitate personalized-medicine workflows
.
Genomics
2012
;
100
:
337
44
.
23.
Koshkin
VS
,
Garcia
JA
,
Reynolds
J
,
Elson
P
,
Magi-Galluzzi
C
,
McKenney
JK
, et al
Transcriptomic and protein analysis of small cell bladder cancer (SCBC) identifies prognostic biomarkers and DLL3 as a relevant therapeutic target
.
Clin Cancer Res
2018
:
25
:
210
21
.
24.
Rosenberg
JE
,
Hoffman-Censits
J
,
Powles
T
,
van der Heijden
MS
,
Balar
AV
,
Necchi
A
, et al
Atezolizumab in patients with locally advanced and metastatic urothelial carcinoma who have progressed following treatment with platinum-based chemotherapy: a single-arm, multicentre, phase 2 trial
.
Lancet
2016
;
387
:
1909
20
.
25.
Mariathasan
S
,
Turley
SJ
,
Nickles
D
,
Castiglioni
A
,
Yuen
K
,
Wang
Y
, et al
TGFβ attenuates tumour response to PD-L1 blockade by contributing to exclusion of T cells
.
Nature
2018
;
554
:
544
8
.
26.
Fraley
C
,
Raftery
AE
. 
Model-based clustering, discriminant analysis, and density estimation
.
J Am Stat Assoc
2002
;
97
:
611
31
.
27.
McLachlan
GJ
,
Peel
D
.
Finite Mixture Models
.
Hoboken, NJ
:
Wiley
; 
2000
.
28.
Wilkerson
MD
,
Hayes
DN
. 
ConsensusClusterPlus: a class discovery tool with confidence assessments and item tracking
.
Bioinformatics
2010
;
26
:
1572
3
.
29.
Hastie
T
,
Tibshirani
R
,
Narashiman
B
,
Chu
G
. 
PAM: prediction analysis for microarrays
.
R package version 1.55. Available from
https://cran.r-project.org/web/packages/pamr/pamr.pdf.
30.
Bolstad
BM
,
Irizarry
RA
,
Astrand
M
,
Speed
TP
. 
A comparison of normalization methods for high density oligonucleotide array data based on variance and bias
.
Bioinformatics
2003
;
19
:
185
93
.
31.
Abaan
OD
,
Polley
EC
,
Davis
SR
,
Zhu
YJ
,
Bilke
S
,
Walker
RL
, et al
The exomes of the NCI-60 panel: a genomic resource for cancer biology and systems pharmacology
.
Cancer Res
2013
;
73
:
4372
82
.
32.
Reinhold
WC
,
Sunshine
M
,
Liu
H
,
Varma
S
,
Kohn
KW
,
Morris
J
, et al
CellMiner: a web-based suite of genomic and pharmacologic tools to explore transcript and drug patterns in the NCI-60 cell line set
.
Cancer Res
2012
;
72
:
3499
511
.
33.
Oser
MG
,
Janne
PA
. 
Small cell neuroendocrine tumors: cell state trumps the oncogenic driver
.
Clin Cancer Res
2018
;
24
:
1775
6
.
34.
Chang
MT
,
Penson
AV
,
Desai
NB
,
Socci
ND
,
Shen
R
,
Seshan
VE
, et al
Small cell carcinomas of the bladder and lung are characterized by a convergent but distinct pathogenesis
.
Clin Cancer Res
2018
;
24
:
1965
73
.
35.
Rebouissou
S
,
Hérault
A
,
Letouzé
E
,
Neuzillet
Y
,
Laplanche
A
,
Ofualuka
K
, et al
CDKN2A homozygous deletion is associated with muscle invasion in FGFR3-mutated urothelial bladder carcinoma
.
J Pathol
2012
;
227
:
315
24
.
36.
Lohrisch
C
,
Murray
N
,
Pickles
T
,
Sullivan
L
. 
Small cell carcinoma of the bladder: long term outcome with integrated chemoradiation
.
Cancer
1999
;
86
:
2346
52
.
37.
Bastus
R
,
Caballero
JM
,
Gonzalez
G
,
Borrat
P
,
Casalots
J
,
Gomez de Segura
G
, et al
Small cell carcinoma of the urinary bladder treated with chemotherapy and radiotherapy: results in five cases
.
Eur Urol
1999
;
35
:
323
6
.
38.
Sangoi
AR
,
Beck
AH
,
Amin
MB
,
Cheng
L
,
Epstein
JI
,
Hansel
DE
, et al
Interobserver reproducibility in the diagnosis of invasive micropapillary carcinoma of the urinary tract among urologic pathologists
.
Am J Surg Pathol
2010
;
34
:
1367
76
.
39.
Mmeje
CO
,
Guo
CC
,
Shah
JB
,
Navai
N
,
Grossman
HB
,
Dinney
CP
, et al
Papillary recurrence of bladder cancer at first evaluation after induction bacillus Calmette-Guérin therapy: implication for clinical trial design
.
Eur Urol
2016
;
70
:
778
85
.
40.
Mcconkey
DJ
,
Lee
S
,
Choi
W
,
Tran
M
,
Majewski
T
,
Lee
S
, et al
Molecular genetics of bladder cancer: emerging mechanisms of tumor initiation and progression
.
Urol Oncol
2010
;
28
:
429
40
.
41.
Cheng
L
,
Zhang
S
,
Alexander
R
,
Maclennan
GT
,
Hodges
KB
,
Harrison
BT
, et al
Sarcomatoid carcinoma of the urinary bladder: the final common pathway of urothelial carcinoma dedifferentiation
.
Am J Surg Pathol
2011
;
35
:
e34
46
.
42.
Iyer
G
,
Balar
AV
,
Milowsky
MI
,
Bochner
BH
,
Dalbagni
G
,
Donat
SM
, et al
Multicenter prospective phase II trial of neoadjuvant dose-dense gemcitabine plus cisplatin in patients with muscle-invasive bladder cancer
.
J Clin Oncol
2018
;
36
;
1949
56
.
43.
Shen
P
,
Jing
Y
,
Zhang
R
,
Cai
MC
,
Ma
P
,
Chen
H
, et al
Comprehensive genomic profiling of neuroendocrine bladder cancer pinpoints molecular origin and potential therapeutics
.
Oncogene
2018
;
37
:
3039
44
.
44.
Rickman
DS
,
Beltran
H
,
Demichelis
F
,
Rubin
MA
. 
Biology and evolution of poorly differentiated neuroendocrine tumors
.
Nat Med
2017
;
23
:
664
73
.
45.
van de Kamp
M
,
Meijer
R
,
Pos
F
,
Kerst
M
,
van Werkhoven
E
,
van Rhijn
B
, et al
Intravesical recurrence after bladder sparing treatment of small cell carcinoma of the bladder: characteristics, treatment, and outcome
.
Urol Oncol Semin Orig Investig
2018
;
36
:
307
.