Novel Mouse Models of Bladder Cancer Identify a Prognostic Signature Associated with Risk of Disease Progression

Bladder cancer is a major health problem with an estimated approximately 18,000 deaths each year in the United States alone (1). Most bladder cancers are urothelial carcinomas that can be broadly classified into non–muscle-invasive bladder cancer (NMIBC) and muscle-invasive bladder cancer (MIBC; refs. 2–5). The vast majority of bladder tumors (∼75%) are NMIBC and, although they frequently recur, their five-year survival is nearly 90%. Although MIBC accounts for only approximately 25% of bladder tumors, patients with muscle-invasive disease have a much worse prognosis with a 5-year survival of approximately 50% for those with regionally invasive disease and approximately 15% for those with metastatic bladder cancer (3, 4).

Until recently, it was believed that NMIBC and MIBC were essentially distinct disease entities (6, 7). However, it is now appreciated that the genetic alterations present in NMIBC and MIBC are more similar than previously thought (8), suggesting that bladder cancer can best be considered as a continuum of disease states. Most notably, approximately 10% to 15% of patients with NMIBC will progress to MIBC (4); however, it is unclear what distinguishes these “at-risk” patients, or what mechanisms underlie their disease progression. Given the profound differences in clinical outcome for patients with NMIBC versus MIBC, improved methods for identifying those with NMIBC who are at risk of progression may facilitate early intervention, thereby leading to improved outcome.

One of the barriers to investigating the biologic differences between NMIBC to MIBC has been the relative paucity of in vivo models that recapitulate progression from non–muscle invasive to muscle-invasive disease (9, 10). A key challenge in developing such models has been difficulties in identifying and targeting for gene recombination the specific cell populations in bladder urothelium that give rise to bladder cancer. The urothelium is comprised of three basic cell types, corresponding to basal cells, which are small cuboidal cells adjacent to the lamina propria that are located immediately above the basement membrane; intermediate cells; and superficial cells, also called umbrella cells, which are highly differentiated cells that line the bladder lumen (11–13). While the precise relationships of these cell types to bladder tumorigenesis have not been fully resolved, studies in both humans and mice have implicated basal cells as potential cells of origin for MIBC (14–20). Furthermore, gene expression profiling analyses of human bladder tumors have revealed basal-like subtypes, which have more aggressive phenotypes, and luminal-like subtypes, which have less aggressive phenotypes (21–26).

Previously, we described a genetically engineered mouse model (GEMM) that progresses from NMIBC to MIBC (27, 28). This GEMM is based on surgical delivery of an adenovirus expressing Cre recombinase under the control of the CMV promoter (AdenoCMV-Cre) into the bladder lumen of mice with conditional floxed alleles for the Pten and Trp53 (hereafter p53) tumor suppressor genes, thereby resulting in coinactivation of Pten and p53 in the bladder. Coclinical analyses of chemotherapy response in this GEMM provided the rationale for an informative clinical trial that is beneficial for patients with high-risk NMIBC (29, 30), demonstrating the relevance of this GEMM as a preclinical model of bladder cancer. However, a significant disadvantage of this first-generation bladder cancer GEMM is that the CMV promoter is expressed in many cell types; therefore, delivery of AdenoCMV-Cre into the bladder lumen can result in gene recombination in multiple cell types within the urothelium and potentially even non–bladder urothelial cells.

In this study, we describe second-generation GEMMs of bladder cancer with improved specificity of Cre-mediated recombination in bladder urothelium. In particular, we use two independent approaches to target gene recombination to specific cell types in the bladder urothelium. Using both approaches, we find that inactivation of Pten and p53 in basal, but not nonbasal, cells is sufficient for progression from NMIBC to MIBC. Cross-species analysis comparing the mouse gene signature of early bladder cancer with a corresponding human bladder cancer signature has identified a conserved prognostic signature for MIBC, supporting the relevance of these GEMMs for studying human bladder cancer and introducing a robust gene signature that may help to stratify patients at risk for developing MIBC.

All experiments using animals were performed according to protocols approved by the Institutional Animal Care and Use Committee (IACUC) at Columbia University Irving Medical Center (New York, NY). The Pten^flox/flox; p53^flox/flox mice were described previously (27); the Pten^flox/flox and p53^flox/flox alleles were originally obtained from the Mouse Models of Human Cancer Consortium Mouse Repository (Pten stock no. 006440; p53 stock no. 008462). For evaluating specificity of gene recombination in the whole organism, mice were crossed with a β-galactosidase reporter allele (ROSA βgeo 26), which was obtained from The Jackson laboratory (stock no. 009427; ref. 31) to generate Pten^+/+; p53^+/+; lacZ/lacZ or Pten^flox/flox; p53^flox/flox; lacZ/lacZ mice. For lineage tracing, Pten^+/+; p53^+/+ or Pten^flox/flox; p53^flox/flox mice were crossed with the Rosa-CAG-LSL-EYFP-WPRE reporter allele (32) to obtain the Pten^+/+; p53^+/+; R26R-CAG^LSL-EYFP/+ (Pten^+/+; p53^+/+; YFP) or Pten^flox/flox; p53^flox/flox; R26R-CAG^LSL-EYFP/+ (Pten^flox/flox; p53^flox/flox; YFP) mice. The Rosa-CAG-LSL-EYFP-WPRE allele was obtained from The Jackson Laboratories on a C57BL/6 background (stock no. 007903). To generate mice having tamoxifen inducible Cre recombinase (Cre^ERT2) under the control of the cytokeratin 5 or cytokeratin 8 promoter, the Pten^flox/flox; p53^flox/flox; YFP mice were crossed with transgenic Cre alleles based on the bovine cytokeratin 5 promoter (33) or the mouse cytokeratin 8 promoter (34) to obtain Ck5-CreER^T2; Pten^flox/flox; p53^flox/flox; YFP (Ck5^CreERT2; DKO) or Ck8-CreER^T2; Pten^flox/flox; p53^flox/flox; YFP (Ck8^CreERT2; DKO) mice, respectively. To generate mice having tamoxifen-inducible Cre recombinase under control of the uroplakin II promoter, we crossed the Pten^flox/flox; p53^flox/flox; YFP mice with the UpkII-iCre^ERT2 allele (35) to obtain UpkII-iCre^ERT2; Pten^flox/flox; p53^flox/flox; YFP mice. Alternatively, we crossed the Rosa-CAG-LSL-EYFP-WPRE reporter allele with the Pten^+/+; p53^+/+ mice to obtain the Ck5-CreER^T2; Pten^+/+; p53^+/+; YFP (Ck5^CreERT2; WT), Ck8-CreER^T2; Pten^+/+; p53^+/+; YFP (Ck8^CreERT2; WT) or UpkII-iCre^ERT2; Pten^+/+; p53^+/+; YFP (UpkII-iCre^ERT2; WT) mice. The following Cre transgenic alleles were obtained from Jackson Laboratories: KRT5-Cre-ER^T2 mice on a FVB background (stock no. 018394); KRT8-Cre-ER mice on a C57BL/6 background (stock no. 017947); and UpkII-iCre^ERT2 on a C57BL/6 x CBA background (stock no. 024768). Mice have been maintained in our laboratory on a predominantly C57BL/6 background. All studies were done using littermates that were genotyped prior to tumor induction; both male and female mice were used.

Procedures for gene recombination and associated videos were performed under the guidance and approval of the IACUC and a veterinarian (R.L. Shoulson); detailed procedures are provided in Supplementary Videos S1–S4 and Supplementary Procedures. For gene recombination in the bladder urothelium by administration of adenoviruses expressing Cre recombinase (Adeno-Cre), we used viruses in which Cre recombinase is expressed under the control of the bovine cytokeratin 5 promoter (AdenoCk5-Cre), the mouse cytokeratin 8 promoter (AdenoCk8-Cre), or the human cytokeratin 14 promoter (AdenoCk14-Cre), as described previously (36), or a control adenovirus lacking Cre (AdenoCMV-empty; ref. 27). Adeno-Cre viruses were obtained from the University of Iowa Viral Vector Core Facility at a concentration of 7×10¹⁰ to 1×10¹¹ plaque-forming units (bk5-Cre, #VVC-Berns-1547; mk8-nlsCre, #VVC-Li-535; hk14Bgi-Cre, #VVC-Berns-1548; empty virus, #VVC-U of Iowa-272). We used two alternative procedures to deliver Adeno-Cre into the bladder lumen to achieve gene recombination. For the first method, we introduced Adeno-Cre via surgical delivery into the bladder lumen following our published protocol (27). In particular, virus (25 μL) was diluted 1:1 (v/v) with DMEM (Thermo Fisher) containing polybrene (final concentration 10 μg/mL, Sigma-Aldrich) and Evans Blue dye (final concentration 0.05%, Sigma-Aldrich). The diluted virus (5 μL) was injected into the bladder lumen using a 10 μL Hamilton syringe (catalog no. 1701, Hamilton). Quantification of cellular specificity is shown in Supplementary Table S1. A full description of the method and associated video is provided in Supplementary Videos S1 and S3 and Supplementary Procedures.

For the second method, we have developed a new procedure for transurethral delivery into the bladder lumen. In this approach, the bladder was first isolated from the body cavity under anesthesia, which enables confirmation of accurate entry of the catherer into the bladder lumen. Using an Olympus SZ51 microscope for visualization, a polyethene tubing (outer diameter: 0.024 inches, 2.5 inches in length) was inserted into the urethra. After placement of the tubing, 10 μL of virus diluted 1:3 with media as above was introduced into the bladder lumen using a Hamilton syringe attached to the polyethene tubing. After injection, the lower part of the bladder (close to the urethra) was compressed for 10 seconds using forceps as the catheter was being removed. A full description of the method and associated video is provided in Supplementary Videos S2 and S3 and Supplementary Procedures.

For gene recombination in the bladder urothelium using the transgenic Ck5-CreER^T2, Ck8-CreER^T2 alleles or UpkII-iCre^ERT2 alleles, Cre activity was induced systemically or via intravesical delivery. For systemic induction, tamoxifen (100 mg/kg in corn oil; Sigma-Aldrich) was administered via oral gavage once daily for 4 consecutive days. For intravesical delivery, 4-hydroxy-tamoxifen (final concentration 40 μg/mL, 2 μg in 50 μL DMEM containing 1:40 v/v Tween-80; Sigma-Aldrich) was delivered by ultrasound (US)-guided injection into the bladder lumen using a VEVO 3100 Imaging System (Visual Sonics). Quantification of cellular specificity is shown in Supplementary Table S1. A full description of the method and associated video is provided in Supplementary Video S4 and Supplementary Procedures.

Following tumor induction, mice were monitored every 4 weeks by US imaging to detect tumor growth. Mice were euthanized when their tumors reached 20 mm, when their body condition score was <1.5, or when the mice reached 14 months of age. To study disease progression, a cohort of mice were sacrificed at 3 months (early) or 9 months (late) following tumor induction, rather than at their endpoint. At the time of sacrifice, the bladder was dissected and fixed in 10% formalin, followed by embedding in paraffin for histopathologic and IHC analyses. For β-galactosidase staining, tissues were fixed in 4% paraformaldehyde for 1 hour at 4°C, followed by 30% sucrose in PBS overnight, and embedded in Tissue-Tek O.C.T Compound (Sakura Finetek USA). Alternatively, fresh bladder tissue was collected for cell sorting as described below.

Histopathologic and IHC analyses were performed as described previously (37). Histopathologic scoring of GEMM phenotypes was assessed by a pathologist (H.A. Al-Ahmadie) based on blinded analyses of hematoxylin and eosin (H&E)-stained sections, and is reported in Supplementary Table S2. For IHC analyses, 3-μm paraffin sections were deparaffinized and rehydrated, followed by antigen retrieval in citrate-based antigen unmasking solution (Vector Laboratories). Slides were blocked in 10% normal goat serum, then incubated with primary antibody overnight at 4°C, followed by incubation with secondary antibody for 1 hour. For IHC, the signal was enhanced using the VECTASTAIN ABC system and visualized with NovaRED Substrate Kit (both from Vector Laboratories). Slides were counterstained with hematoxylin and mounted with Permount (Thermo Fisher Scientific). Images were captured using an Olympus VS120 whole-slide scanning microscope. For immunofluorescence staining, sections were counterstained with DAPI solution (BD Biosciences) and mounted with VECTASHIELD mounting medium (Vector Laboratories) for fluorescence. Images were captured using a Leica TCS SP5 confocal microscope. For quantification of IHC and immunofluorescence staining, images from a minimum of five sections from a minimum of four independent mice were counted as described previously (37). All antibodies used in this study, as well as antibody dilutions, are described in Supplementary Table S3. For β-galactosidase staining, 8-μm–thick sections were postfixed with 4% paraformaldehyde, followed by incubation in staining solution [1 mg/mL X-gal dissolved in N,N-dimethylformamide, 5 mmol/L K₄Fe(CN)₆ • 3H₂0, 5 mmol/L K₃Fe(CN)₆, 2 mmol/L MgCl₂, 0.02% NP40, 0.01% sodium deoxycholate in PBS] overnight at 37°C. Sections were counterstained in Nuclear Fast Red (Vector Laboratories) and mounted with Permount (Fisher Scientific). Images were captured using an Olympus VS120 whole-slide scanning microscope.

For isolation of yellow fluorescence protein (YFP)-lineage–marked cells, the bladder urothelium was manually dissected from the underlying connective tissue and the isolated urothelial tissue was digested at 37ºC for 3 hours in 1× collagenase/hyaluronidase (10× stock, Stem Cell Technologies, #7912) in DMEM-F12 and 5% FBS. Samples were pelleted at 350 × g for 5 minutes at 4ºC in an Eppendorf 5810R tabletop centrifuge, resuspended in 0.25% trypsin/EDTA (Stem Cell Technologies), and incubated for 30 minutes on ice. Cells were collected by centrifugation as above, and incubated in a cocktail of prewarmed dispase (5 U/mL) plus DNase I (1 mg/mL stock, final concentration 0.1 mg/mL; Stem Cell Technologies) for 1–2 minutes with vigorous pipetting. Following which, cells were filtered using Corning Falcon test tube with cell strainer snap cap (#08–771–23), pelleted and resuspended in Hank Balanced Salt Solution (Stem Cell Technologies) plus 2% FBS (Gibco Laboratories) for sorting using a BD FACS Aria II sorter. The YFP-expressing cells were isolated using PE/FITC (R-phycoerythrin/fluorescein isothiocyanate) channels to gate the YFP-positive (YFP⁺) population.

To generate organoids, the YFP-sorted cells (as above) were collected in hepatocyte media (Corning Hepatocyte Culture Media Kit, #355056) containing 10 μmol/L ROCK inhibitor as described previously (38, 39). Cells (1,000–5,000 in 100 μL media) were plated in 96-well low-attachment plate (catalog no. 3474, Corning) in organoid media containing hepatocyte medium with 10 ng/mL EGF (Corning), 5% charcoal-stripped FBS, 1× Glutamax (Gibco Laboratories), 5% Matrigel (Corning), 10 μmol/L ROCK inhibitor, and 1× antibiotic–antimycotic (Gibco Laboratories). After 14 days in culture, organoid-forming efficiency was estimated by counting the number of YFP⁺ organoids divided by the number of cells initially plated. For histologic analysis, organoids were collected by centrifuging for 5 minutes at 350 × g at 4ºC, fixed in 10% formalin, and embedded in Richard-Allan Scientific HistoGel Specimen Processing Gel (#HG-4000–012, Thermo Fisher). Histologic and IHC analyses were performed as above.

Allograft models were generated by implantation of YFP-lineage–marked urothelial cells obtained by FACS (as above) into the submucosal layer of the bladder of adult male NCr/Nude mice (6–8 weeks old; Taconic) using US guidance as described previously (38). Briefly, the lamina propria was delaminated from the detrusor muscle by US-guided delivery of PBS (about 100 μL) using a 1 mL syringe with a 30G 0.5-inch needle to create a submucosal pocket, followed by US-guided delivery of the cell suspension (1×10⁴ cells suspended in 20 μL organoid media containing 50% Matrigel) into the resulting submucosal pocket. Mice were euthanized when their tumors reached 20 mm, when their body condition score was <1.5, or when the mice reached 14 months of age. At the time of sacrifice, bladders were collected and histologic and IHC analyses were performed as above.

For RNA sequencing (RNA-seq), RNA was prepared from YFP-sorted cells (as above) from four independent mice (two males and two females) per group. RNA was made using the MagMAX-96 total RNA Isolation Kit (Thermo Fisher), and amplified using SMART-Seq v4 Ultra Low Input RNA Kit (Takara #634891). Total RNA was enriched for mRNA using poly-A pull-down; only RNA samples having a quantity of 200–1,000 ng and with an RNA integrity number (RIN) > 8 were used. Libraries were made using an Illumina TruSeq RNA prep-kit and sequenced using an Illumina HiSeq2500 by multiplexing samples in each lane, which yields a targeted number of single-end/100-bp reads for each sample, as a fraction of 180 million reads for the whole lane. Reads were aligned to mouse genome build M4 using the STAR aligner (version STAR_2.4.2a_modified, RRID:SCR_004463), and raw gene counts were determined using the Subread (Bioconductor) package using R-studio 1.3.1073, R v4.0.2 (The R Foundation for Statistical Computing, ISBN 3–900051–07–0). Mouse RNA-seq raw counts experiments were normalized, variance stabilized, and differential gene expression analysis was defined using DESeq2 (Bioconductor, RRID:SCR_006442, RRID:SCR_000154). Fold changes were corrected using the apeglm package. Genes with Benjamini–Hochberg adjusted P values <0.001 were defined as differentially expressed. A list of differentially expressed genes is provided in Supplementary Dataset S1.

Gene-set enrichment analysis (GSEA, RRID:SCR_003199) was performed as described previously (40); normalized enrichment scores (NES) and P values were estimated using 1,000 gene permutations using the fgsea (Bioconductor) package; visualization of GSEA curves was performed using enrichplot (Bioconductor). For pathway enrichment analyses, reference signatures were ranked by their log₂fold change X -log₁₀ (P value). Pathway enrichment was performed using GSEA to query the Molecular Signatures Database (MSigDB), including the C2 [Kyoto Encyclopedia of Genes and Genomes (KEGG), RRID:SCR_012773 and BIOCARTA RRID:SCR_006917] and Hallmark pathway dataset (41). A list of differentially expressed pathways is provided in Supplementary Dataset S2.

For cross-species analyses, mouse genes were first mapped to their corresponding human orthologs based on the HomoloGene database (NCBI, RRID:SCR_002924). The analyses described herein used the following published human bladder cancer cohorts: (i) the Hedegaard cohort (42); (ii) the Kim cohort (43); (iii) the Lund cohort (44); and (iv) The Cancer Genome Atlas (TCGA) bladder cancer cohort (45); available clinical details for these cohorts are provided in Supplementary Table S4. For cross-species analyses using the Kim cohort as a reference signature, a differential gene expression signature was defined using a two-tailed Student t test, and genes were ranked by their t-statistic X -log₁₀ (P value).

For Kaplan–Meier survival analysis, we first performed single-sample GSEA (ssGSEA) using gsva (Bioconductor) to rank patient samples based on the combined expression levels of the specified gene signature from highest to lowest expression. Kaplan–Meier survival analysis was done to compare the top 25% of patients with highest expression levels with the top 25% of patients with lowest expression levels. Statistical significance was estimated using a log-rank test using the survminer (Bioconductor) package.

For in vivo studies, the number of mice needed to achieve statistical significance was determined using standard power analysis. Statistical analyses were performed using nonparametric unpaired t test (Mann–Whitney test). GraphPad Prism software (Version 8.4.3, RRID:SCR_002798) was used for statistical calculations and data visualization. Log-rank (Mantel–Cox) test and Kaplan–Meier survival analysis were done using GraphPad Prism software. The value of n reported within figure legends represent number of samples. Unless otherwise indicated, data are expressed as mean ± SD.

Raw and normalized RNA-sequencing data are publicly available (GSE110694 and GSE89823) through Gene Expression Omnibus (GEO, RRID:SCR_005012) database.

To generate enhanced GEMMs of bladder cancer, we sought to develop approaches to achieve gene recombination in specific cell populations of the bladder urothelium with improved efficiency and specificity. Toward this end, we refined our adenovirus approach for gene recombination in the bladder in two key respects (Fig. 1A). First, rather than using the CMV promoter, which is expressed in many cell types, we used adenoviruses expressing Cre under the control of promoters that are expressed selectively in specific cell types, as reported previously, for analyses of cell type specificity in lung cancer models (36). Second, we complemented our original method of surgical delivery of adenoviruses into the bladder lumen (Supplementary Fig. S1A and S1B; Supplementary Videos S1 and S3) by developing an alternative approach using transurethral delivery of AdenoCre (Supplementary Fig. S1A and S1B; Supplementary Videos S2 and S3). Notably, while the surgical delivery method is rapid and efficient, it can yield off-target Cre-mediated recombination and potentially off-target tumors if there is leakage of the virus outside of the bladder lumen (Supplementary Fig. S1A). In contrast, transurethral delivery is technically more challenging and inherently less efficient for tumor induction, but Cre activity is highly specific to bladder (Supplementary Fig. S1A).

As an independent approach to target specific cell populations in the bladder urothelium, we have utilized tamoxifen-inducible Cre transgenic alleles (Fig. 1B). To confine Cre-mediated recombination to bladder, we developed a procedure for intravesical delivery of 4-hydroxy-tamoxifen into the bladder lumen using US-guided imaging (Supplementary Fig. S2A and S2B; Supplementary Video S4). Notably, while systemic delivery of tamoxifen is more efficient for activation of Cre, intravesical delivery is bladder-specific (Supplementary Fig. S2A and S2B); therefore, intravesical delivery enables the use of Cre alleles that are expressed in tissues other than bladder, because few Cre alleles are expressed exclusively in bladder (9).

As a proof of concept, we considered that specific cell populations in bladder urothelium differentially express high molecular-weight cytokeratins, such as cytokeratin 5, and low molecular-weight cytokeratins, such as cytokeratin 8. In particular, basal cells express high levels of cytokeratin 5 and low levels of cytokeratin 8 (Ck5^high; Ck8^low); intermediate cells express low levels of cytokeratin 5 and medium levels of cytokeratin 8 (Ck5^low; Ck8^med); and superficial cells express low levels of cytokeratin 5 and high levels of cytokeratin 8 (Ck5^low; Ck8^high; Fig. 1C; ref. 12). We reasoned that the cytokeratin 5 promoter should preferentially target gene recombination to basal cells, while the cytokeratin 8 promoter should primarily target intermediate and superficial cells.

Therefore, we used adenoviruses in which Cre recombinase is expressed under the control of a cytokeratin 5 promoter (AdenoCk5-Cre) or a cytokeratin 8 promoter (AdenoCk8-Cre) to target basal or intermediate/superficial cells, respectively (Fig. 1D). To evaluate the specificity and efficiency of the AdenoCk5-Cre and AdenoCk8-Cre viruses, we performed lineage marking using a conditionally activatable reporter allele with enhanced YFP fluorescence, R26R-EYFP (32). Delivery of AdenoCk5-Cre into the bladder lumen led to marking predominantly of basal cells, as evident by colabeling of the YFP-marked cells with cytokeratin 5, but not cytokeratin 8 (P = 0.0286; n = 105 Ck5^highCk8^low cells/109 total cells counted; 95.2% ± 3.1%; Fig. 1D; Supplementary Table S1A). Conversely, following delivery of AdenoCk8-Cre, the YFP-marked cells were primarily coexpressed with cytokeratin 8 but not cytokeratin 5, corresponding to intermediate and superficial cells (P = 0.0286; for intermediate: n = 14 Ck5^lowCK8^med cells/29 total cells counted; 46.3% ± 11.5%; for superficial: n = 14 Ck5^lowCK8^high cells/29 total cells counted; 51.5% ± 14.2%; Fig. 1D; Supplementary Table S1A). Notably, AdenoCk8-Cre results in an underrepresentation of superficial cells, which we attribute to inefficient infectivity and gene recombination in superficial cells likely because they are highly differentiated, large, and multinucleated cells (13).

We obtained similar results using transgenic alleles having an inducible Cre recombinase expressed under the control of the cytokeratin 5 promoter (Ck5-Cre^ERT2) or a cytokeratin 8 promoter (Ck8-Cre^ERT2), which were crossed to the R26R-EYFP reporter allele (Fig. 1E). We found that the majority of YFP⁺-lineage marked cells in bladders from the Ck5-Cre^ERT2 mice corresponded to basal cells (P = 0.0006; n = 3887 Ck5^highCk8^low cells/4,289 total cells counted, 89.1% ± 4.6%; Fig. 1E; Supplementary Table S1B), whereas in bladders from the Ck8-Cre^ERT2 mice, most of the YFP⁺-lineage marked cells were intermediate cells (P = 0.0079; n = 1250 Ck5^lowCk8^med cells/1,582 total cells counted; 78.5% ± 4.1%; Fig. 1E; Supplementary Table S1B). In summary, we have developed a toolkit of complementary approaches to achieve Cre-mediated recombination selectively in basal and nonbasal (i.e., intermediate and superficial) cell populations of the bladder urothelium.

To evaluate the relative contribution of basal and nonbasal cells for bladder tumorigenesis, we induced gene recombination using either the AdenoCre or transgenic Cre approach in mice harboring conditional alleles for Pten and p53, because we have previously shown that co-inactivation of Pten and p53, but not other tumor suppressor alleles, leads to bladder tumors following delivery of AdenoCMV-Cre (27). For lineage-marking, mice were further crossed with the conditionally activatable R26R-EYFP reporter allele to generate the experimental Pten^flox/flox; p53^flox/flox; YFP mice. To induce bladder tumors with AdenoCre, we introduced the AdenoCk5-Cre or AdenoCk8-Cre viruses (or a control adenovirus) into the bladder lumen of adult male or female Pten^flox/flox; p53^flox/flox; YFP mice using the surgical delivery approach (Fig. 2A–D; Supplementary Fig. S3A; Table 1; Supplementary Table S2). To evaluate the consequences of conditional inactivation of Pten and p53 using the transgenic Cre drivers, we first crossed the Pten^flox/flox; p53 ^flox/flox; YFP mice with the Ck5-Cre^ERT2 or Ck8-Cre^ERT2 alleles to generate the corresponding Ck5-Cre^ERT2 or Ck8-Cre^ERT2; Pten^flox/flox; p53^flox/flox; YFP mice (or the control mice without the transgene). We then induced Cre-mediated recombination by intravesical delivery of 4-hydroxy-tamoxifen into the bladder lumen of adult male or female mice to restrict gene recombination to the bladder urothelium (Fig. 2E–H; Supplementary Fig. S3B; Table 1; Supplementary Table S2). Cohorts of the AdenoCre or transgene Cre mice were induced to form tumors at 2 months of age and monitored for up to one year using ultrasound imaging to detect bladder tumor growth (Fig. 2A, B, E, and F).

Notably, for both the AdenoCre and the transgenic Cre cohorts, activation of Cre via the cytokeratin 5 promoter resulted in MIBC in Pten^flox/flox; p53^flox/flox; YFP mice, whereas Cre activation via the cytokeratin 8 promoter did not (Fig. 2B and F; Supplementary Fig. S3A and S3B; Table 1; Supplementary Table S2). In particular, in the AdenoCre cohort, delivery of AdenoCk5-Cre into the bladder lumen of Pten^flox/flox; p53^flox/flox; YFP mice resulted in invasive urothelial carcinomas (T3 or greater), some with prominent squamous differentiation or poorly differentiated sarcomatoid neoplasia without metastasis; in contrast, mice that received AdenoCk8-Cre displayed moderate to marked urothelial atypia without overt carcinoma (Fig. 2B; Supplementary Fig. S3A; Table 1; Supplementary Table S2). As we have reported previously (27), inactivation of either Pten or p53 alone did not result in abnormal histopathological or tumor phenotypes (Supplementary Fig. S3C).

In the transgenic Cre cohort, the tamoxifen-induced Ck5-Cre^ERT2; Pten^flox/flox; p53^flox/flox; YFP mice developed invasive and noninvasive high-grade urothelial carcinoma (Tis to T2) with rare squamous differentiation without metastasis, whereas the Ck8-Cre^ERT2; Pten^flox/flox; p53^flox/flox; YFP mice displayed urothelial atypia and dysplasia but no evidence of carcinoma (Fig. 2F; Supplementary Fig. S3B; Table 1; Supplementary Table S2). Consistent with the more aggressive histopathologic phenotype of the AdenoCk5-Cre mice, bladder tumors arising following AdenoCk5-Cre compared with Ck5-Cre^ERT2 had with a shorter latency (median time of 7.6 vs. 10.3 months) and were on average considerably larger, albeit more variable in size (748.0 ± 1,331.1 mg vs. 95.6 ± 72.2 mg; Fig. 2B, C, F, and G; Table 1). In both the AdenoCre and the transgenic Cre cohorts, the bladder cancer phenotypes were similar in the male and female mice, as we have reported previously for tumors induced by delivery of AdenoCre-CMV (27).

In both the AdenoCre and the transgenic Cre cohorts, the bladder phenotype of mice that have activation of Cre via the cytokeratin 5 promoter, but not the cytokeratin 8 promoter, displayed extensive cellular proliferation, as evidenced by Ki67 expression, increased expression of cytokeratin 14, and irregular expression of vimentin, which are characteristic of invasion (Fig. 2D and H). Furthermore, in bladder tumors from both the AdenoCk5-Cre and Ck5-Cre^ERT2; Pten^flox/flox; p53^flox/flox; YFP mice, the YFP-marked cells coexpressed cytokeratin 5, confirming that the tumors consist of basal cells (Fig. 2B and F). Together, these findings suggest that basal, but not nonbasal, cells give rise to tumors in the context of coactivation of Pten and p53.

To further verify these findings, we used alternative Cre drivers to target gene recombination to basal or nonbasal cells using an Adenovirus expressing the cytokeratin 14 promoter (AdenoCk14-Cre; ref. 36) or a Uroplakin II promoter transgenic allele (UpkII-iCre^ERT2; ref. 35), respectively. In particular, targeting basal cells by delivery of AdenoCk14-Cre into the bladder lumen of Pten^flox/flox; p53^flox/flox; YFP mice resulted in invasive urothelial carcinomas (T3 or greater), similar to that of AdenoCk5-Cre although with a shorter latency to tumor development (Supplementary Fig. S4A–S4C; Supplementary Table S2). Conversely, targeting nonbasal cells via induction of gene recombination using the UpkII-iCre^ERT2 transgene leads to noninvasive high-grade (Tis or Tis/a) phenotypes that are more prominent than in the cytokeratin 8 Cre cohort, but did not result in progression to overt tumors (Supplementary Fig. S4D–S4F, Supplementary Table S2). Therefore, using two independent approaches to achieve gene recombination, namely the AdenoCre and transgenic Cre approaches, we have found that coinactivation of Pten and p53 in basal, but not nonbasal, bladder urothelial cells is sufficient to induce MIBC.

We next sought to leverage these GEMMs to investigate progression from noninvasive to muscle-invasive disease. For these studies, we used the transgenic Cre approach to coinactivate Pten and p53, because phenotypes arising in these mice displayed a more consistent time-course of tumorigenesis than those induced by AdenoCre (see Fig. 2C and G; Table 1). In particular, we analyzed the bladder phenotypes of Ck5-Cre^ERT2 or Ck8-Cre^ERT2; Pten^flox/flox; p53^flox/flox; YFP mice (hereafter, Ck5-Cre^ERT2 or Ck8-Cre^ERT2; DKO for “double knockout”) or transgene-negative control mice at 3 months following tumor induction (i.e., in 5-month-old mice), which is prior to the occurrence of overt tumors, and at 9 months following tumor induction (i.e., in 12-month-old mice), which is near the median time-course of tumor development (Fig. 3A).

Histopathologic analysis confirmed that the Ck5-Cre^ERT2; DKO, but not the Ck8-Cre^ERT2; DKO, mice displayed invasive bladder carcinoma at the late time point (i.e., at 12 months; Fig. 3B), consistent with the findings described above (see Fig. 2), whereas at the early time point (i.e., at 5 months), neither the Ck5-Cre^ERT2; DKO nor the Ck8-Cre^ERT2; DKO mice displayed evidence of invasion (Fig. 3B). Correspondingly, only bladder tumors from the Ck5-Cre^ERT2; DKO mice at the late time point were highly proliferative, as evident by robust staining for Ki67 (25%; P = 0.0159; Fig. 3B and C). At the early and late time points, bladders from both the Ck5-Cre^ERT2; DKO and Ck8-Cre^ERT2; DKO mice, but not the control mice, displayed activation of PI3K signaling, as evident by immunostaining with phospho-S6 (Fig. 3B), consistent with our previous observations in bladder tumors from Pten^flox/flox; p53^flox/flox mice that had been induced with AdenoCMV-Cre (27, 28). Interestingly, MAPK signaling was upregulated in the bladders of early but not late Ck5-Cre^ERT2; DKO mice, as well as in both the early and late Ck8-Cre^ERT2; DKO mice, as evident by immunostaining with phospho-Erk (Fig. 3B), also consistent with our previous observations in Pten^flox/flox; p53^flox/flox bladder tumors induced with AdenoCMV-Cre (27). In addition, cytokeratin 14 was upregulated in basal cells of Ck5-Cre^ERT2; DKO bladders at the early time point, and expressed widely in the invasive tumors of the late stage Ck5-Cre^ERT2; DKO mice, but was not expressed in bladders from either the early- or late-stage Ck8-Cre^ERT2; DKO mice (Fig. 3B).

Interestingly, we observed a striking increase in Cd45-positive cells in invasive tumors from late stage Ck5-Cre^ERT2; DKO mice, but not in early-stage Ck5-Cre^ERT2; DKO mice or either the early or late stage Ck8-Cre^ERT2; DKO mice (Fig. 3B), consistent with an increased proinflammatory phenotype in disease progression. Similarly, higher levels of Cd45-positive cells were also observed in invasive tumors from the AdenoCk5 and AdenoCk14 Pten^flox/flox; p53^flox/flox mice but not the noninvasive phenotypes from the corresponding AdenoCk8 or the UpkII-iCre^ERT2 transgenic mice (Supplementary Fig. S5). Taken together, these findings suggest that Ck5-Cre^ERT2; DKO mice display a preinvasive phenotype at early time-points after tumor induction, which at later stages can progress to invasive carcinoma that is associated with a proinflammatory phenotype.

To investigate the cell-intrinsic features of the preinvasive bladder cells, we isolated YFP lineage-marked cells from the bladder urothelium of Ck5-Cre^ERT2; DKO or Ck8-Cre^ERT2; DKO mice or the corresponding Ck5-Cre^ERT2 or Ck8-Cre^ERT2; Pten^+/+; p53^+/+; YFP mice (hereafter called Ck5-Cre^ERT2; WT or Ck8-Cre^ERT2; WT mice for “wild type” mice) at 3 months postinduction (Fig. 4A). We generated bladder organoids based on a protocol we had established previously for mouse prostate cells and human bladder organoids (Supplementary Fig. S6A and S6B; refs. 38, 39). These analyses revealed that YFP-lineage marked cells from the Ck5-Cre^ERT2 mice, and particularly the Ck5-Cre^ERT2; DKO mice, were significantly more efficient at generating organoids than those from the Ck8-Cre^ERT2 mice (P = 0.0075; Supplementary Fig. S6A and S6B).

We next evaluated the ability of the YFP-marked urothelial cells to grow and/or form tumors when implanted orthotopically into the lamina propria of host mice in vivo (Supplementary Fig. S6C and S6D; ref. 38). We found that the Ck5-Cre^ERT2; DKO urothelial cells developed invasive bladder tumors (P = 0.0205; n = 10), whereas the Ck8-Cre^ERT2; DKO urothelial cells displayed noninvasive growth (n = 8; Supplementary Fig. S6C and S6D). No tumor incidence was detected in the Ck5-Cre^ERT2; WT cells, whereas the Ck8-Cre^ERT2; WT cells did not exhibit any growth in vivo (n = 8/group; Supplementary Fig. S6C and S6D). These findings demonstrate that, by 3 months following tumor induction, basal urothelial cells of Ck5-Cre^ERT2; DKO mice have acquired a cell-intrinsic preinvasive cancer phenotype that is capable of progressing to overt carcinoma.

To identify molecular pathways that are activated in the preinvasive bladder cells, we performed RNA-seq analyses on YFP-marked urothelial cells isolated from Ck5-Cre^ERT2; WT, Ck5-Cre^ERT2; DKO, Ck8-Cre^ERT2; WT, and Ck8-Cre^ERT2; DKO mice (Fig. 4B–D). Principal component analysis showed that expression profiles from the Ck5-Cre^ERT2; DKO and Ck8-Cre^ERT2; DKO urothelial cells clustered more closely together than those of the corresponding Ck5-Cre^ERT2; WT and Ck8-Cre^ERT2; WT samples (Fig. 4B). To identify biological pathways associated with the preinvasive phenotype in the urothelial cells from these GEMMs, we performed pathway-based GSEA using a signature of differential gene expression comparing the Ck5-Cre^ERT2; DKO versus Ck8-Cre^ERT2; DKO expression profiles to query the Hallmark, KEGG and Biocarta datasets (Fig. 4C and D; Supplementary Dataset S2; ref. 41). Among the significantly enriched pathways were those associated with cell adhesion or cell signaling (e.g., KEGG cell adhesion molecules, Biocarta integrin pathway, Hallmark hedgehog signaling), epithelial-to-mesenchymal transition (Hallmark epithelial-mesenchymal transition), tumorigenicity (Hallmark p53 pathway, Hallmark hypoxia), and immune regulation (KEGG cytokine cytokine receptor interaction, Hallmark TNFα signaling via NF-κB; Fig. 4C and D; Supplementary Dataset S2). Pathway activation was further evident by the differential expression of relevant leading-edge genes (Fig. 4D). Overall, these findings showing that preinvasive basal cells are primed to develop organoids in culture and tumors in vivo and that they express pathways associated with tumorigenesis, suggest that these cells are predisposed to progress to invasive bladder cancer.

On the basis of these findings, we sought to identify a gene signature associated with bladder cancer progression in the GEMMs that is conserved with human bladder cancer (Fig. 5). Toward this end, we first identified a mouse gene signature of early bladder cancer by comparing the preinvasive signature from the Ck5-Cre^ERT2; DKO versus Ck8-Cre^ERT2; DKO expression profiles with an invasive gene signature based on differential gene expression analysis of lethal MIBC from AdenoCMV-Cre-induced Pten^flox/flox; p53^flox/flox as described previously (Fig. 5A) (46). We reasoned that genes common to the preinvasive and lethal gene signatures would be enriched for those associated with disease progression.

Indeed, the intersection of these mouse signatures identified 63 genes that are upregulated in both the pre-invasive urothelial cells and the lethal bladder tumors, which we designated as the mouse early bladder cancer signature (Fig. 5A; Supplementary Dataset S3). Notably, this mouse signature was upregulated in a subset of human NMIBC from the Hedegaard cohort, which is comprised mainly of NMIBC (n = 460 NMIBC and n = 16 MIBC; Supplementary Table S4; Supplementary Fig. S7A; ref. 42). Furthermore, expression of the 63-gene mouse signature was significantly associated with adverse outcome in two independent datasets, namely the Lund cohort, which is comprised mostly of NMIBC but has a significant number of MIBC (stages Ta and T1, n = 213; stages >T1, n = 93; Supplementary Table S4; ref. 44), and TCGA bladder cancer cohort, which is comprised primarily of MIBC (n = 412; Supplementary Table S4; ref. 45). In particular, Kaplan–Meier survival analyses revealed that a high level of expression of the 63-gene mouse signature was associated with reduced overall survival in both the Lund cohort (P = 1.76×10⁻⁵; Supplementary Fig. S7B) and the TCGA cohort (P = 3 × 10⁻⁴; Supplementary Fig. S7C).

To further refine this signature of early bladder cancer, we sought to identify the sub-set of genes in the mouse signature that are conserved with human bladder cancer progression (Fig. 5B). Toward this end, we performed GSEA to compare the 63-gene mouse signature with a human bladder cancer signature based on the Kim cohort, which includes both NMIBC and MIBC (n = 103 NMIBC and n = 62 MIBC; Supplementary Table S4; ref. 43). This analysis identified 28 genes in the positive leading edge that are both conserved and upregulated, which we have designated the conserved gene signature of early bladder cancer (Table 2). These genes include several that are associated with molecular pathways of bladder tumorigenesis, such as cell adhesion (e.g., NRCAM, SERPINE1, TGFB1, SDC3), cellular signaling (e.g., AK1, IMPA2, GAS7), gene regulation (e.g., FOSL1, RALY, SRF), and inflammation (e.g., MFHA1, SRXN1; Table 2).

Heatmap representation based on patients in the Hedegaard cohort revealed that a subset with varying grade and tumor stage have upregulation of the 28-gene signature of early bladder cancer (Fig. 5C). To further validate this 28-gene signature, we first verified that the genes were upregulated during bladder cancer progression in both mouse and human bladder cancer cohorts. In particular, for the mouse homologs, we considered the expression level of each gene in the pre-invasive versus lethal signatures (Supplementary Dataset S3), and for the human homologs, we compared gene expression in low grade (Ta/T1) with higher grade (>T1) tumors in the Lund dataset (Supplementary Table S4). We found that all 28 genes were upregulated in both the human and mouse cohorts, albeit to differing extents (Table 2).

Next, we examined the expression of the top 12 most-upregulated genes in the 28-gene signature and found that the expression levels of each of the genes were significantly up-regulated in the basal versus nonbasal preinvasive mouse bladder (Supplementary Fig. S8), and all but one (COTL1) were upregulated in the noninvasive (Ta or T1) versus invasive (>T1) human bladder cancers from the Lund cohort (Supplementary Fig. S9). Among these, the top-most upregulated gene in the mouse preinvasive signature was NrCAM (Supplementary Fig. S10A), which encodes a member of the immunoglobin super family that has been found to be dysregulated in several cancer types (e.g., refs. 47, 48). NrCAM expression was also elevated at the mRNA level in invasive human bladder cancers (P < 1×10⁻⁶; Supplementary Fig. S10B), and at the protein level in both mouse and human bladder cancer (Supplementary Fig. S10C).

Given these findings, we asked whether the 28-gene signature of early bladder cancer is associated with bladder cancer progression and outcome. Indeed, Kaplan–Meier survival analyses based on the Lund cohort revealed that high level of expression of this gene signature was significantly associated with adverse outcome based on overall survival probability (log-rank P = 2.65×10⁻⁶, Fig. 5D). Furthermore, Kaplan–Meier survival analyses based on TCGA also showed that high level of expression of the 28-gene signature was significantly associated with adverse outcome (log-rank P = 0.0019, Fig. 5E). Even the expression of NrCAM alone had some prognostic value (P = 0.01625; Supplementary Fig. S10D).

Finally, we compared our 28-gene signature of early bladder cancer with six other prognostic signatures of bladder cancer that have been published previously (43, 49–54). We found that none of the 28 genes are found in the other reported signatures that are of comparable size (Supplementary Table S5), although our 28-gene signature shares genes in common with the larger gene signatures reported by Robertson and colleagues (54). Interestingly, our 28-gene signature out-performed five of six of these other gene signatures in Kaplan–Meier survival analyses based on both the Lund and TCGA cohorts (Fig. 5D and E; Supplementary Fig. S11A and S11B; Supplementary Table S5). These findings reveal that the 28-gene signature of early bladder cancer is a robust prognostic indicator of bladder cancer progression.

Relatively few genes are expressed exclusively in bladder urothelium, particularly in the cells that give rise to bladder tumors. Consequently, it has proven challenging to develop GEMMs of bladder cancer, in part, because of inherent difficulties in selectively targeting gene recombination to relevant cell populations in bladder urothelium (9). We have overcome this challenge by developing approaches to target gene recombination to specific populations of bladder cells that allow the use of promoters that are not exclusively bladder-specific. This has enabled us to directly compare the consequences of tumor initiation in basal and nonbasal cells using two independent approaches, which has revealed that gene recombination only in basal cells can give rise to bladder tumors, at least in the context of inactivation of Pten and p53. Furthermore, by integrating data from preinvasive and advanced stages in the GEMMs with data from human patients with bladder cancer, we identified a 28-gene signature of early bladder cancer that is prognostic of adverse disease outcome, thus, highlighting the relevance of these GEMMs for studying the biology of human bladder cancer.

Our study builds upon previous work from our group and others (e.g., (16, 20, 27, 55–59)) to assemble a toolkit of complementary approaches that has enabled us to leverage Cre drivers that are not exclusively expressed in bladder to target gene recombination specifically to the bladder urothelium, as well as to isolate lineage marked cells to study the cell-intrinsic features of specific populations of bladder urothelial cells. Since our study utilizes Cre drivers that are not exclusively expressed in bladder, we reasoned that it would be important to base our conclusions on complementary approaches, namely, the AdenoCre delivery and transgenic Cre approaches, each of which has certain advantages and disadvantages.

Indeed, an important advantage of both the AdenoCre and the transgenic Cre approaches is that they enable bladder-specific gene recombination in both male and female mice, although the transurethral delivery method for delivery of AdenoCre is more challenging in male mice. Notably, we have not observed any differences in disease progression in male and female mice in the context of inactivation of Pten and p53. The AdenoCre and the transgenic Cre approaches are complementary in that AdenoCre delivery provides a means to rapidly screen the phenotypic consequences of conditional loss or gain of function of relevant genes of interest without the need to introduce additional alleles expressing Cre drivers. Furthermore, bladder tumors induced by the AdenoCre delivery, particularly the surgical method, are generally more aggressive but also more variable in terms of time-course of tumor development, whereas bladder tumors induced by using transgenic Cre alleles are generally less aggressive but have a more predicable time-course of development and therefore more suitable for controlled longitudinal analyses.

While it can be technically challenging to perform the AdenoCre delivery into bladder, intravesical delivery of tamoxifen also requires skill as well as access to ultrasound guided imaging. To overcome the technical hurdles and improve access to these approaches, we have included videos and detailed procedures that describe these approaches in detail. We note that although we have demonstrated the utility of these methods mainly based on cytokeratin 5 and cytokeratin 8 Cre drivers, these procedures should be readily adaptable for other Cre drivers that are expressed in bladder, as we have shown by extending our findings to cytokeratin 14 and uroplakin II promoters. However, we do not expect that all Cre drivers will be suitable to generate GEMM models; in particular, we found that transgenic mice expressing CreERT2 under the control of the human keratin 18 promoter did not display the expected cell-type specificity in bladder. Furthermore, a Cre allele expressing CreERT under the control of the cytokeratin 14 promoter did not achieve sufficient levels of gene recombination in the bladder to generate tumors. Therefore, demonstration of specificity and efficiency of Cre recombination will be essential to adapt this methodology to other Cre alleles.

One of the main conclusions of this study is that inactivation of Pten and p53 in basal cells is sufficient for bladder tumorigenesis, which is consistent conclusions of previous studies showing that basal cells are a cell of origin of bladder cancer (15, 17). While our findings were not unexpected based on these previous studies, the conclusions of the current study are founded on direct comparison of basal and nonbasal cells using two independent approaches to achieve gene recombination and are based on several different Cre drivers. A second important distinction of the current study is that we have based our findings solely on gene targeting rather than combined with carcinogen treatment, which has been used in most previous studies. Notably, our findings differ from a previous study, which utilized a different transgenic Ck5 Cre allele that did not have a bladder cancer phenotype (55). However, we have found that the transgenic Ck5 Cre allele used in the previous work is not expressed in the bladder, unlike the transgenic Ck5 Cre allele used in this study. Finally, we note that our conclusion that invasive tumors arise in basal cells is based on inactivation of Pten and p53; it is conceivable that conditional loss or gain of function of other genes may give rise to luminal tumors and this can readily be tested using the toolkit that we have established. Along these lines, it will be of particular interest to evaluate the functions of RB1, TSC1, FGFR3, and HRAS using these approaches.

A second important advance of our study is that by modeling bladder cancer progression from preinvasive to invasive disease, we have identified key features of disease progression. First, we find that preinvasive basal cells have the cell-intrinsic capacity to develop tumors when implanted in vivo, and they display up-regulation of pathways associated with bladder tumorigenesis. For instance, the Hedgehog signaling pathway is upregulated in the Ck5-Cre^ERT2; DKO urothelial cells, which is of note because hedgehog signaling has been shown to be dysregulated in bladder cancer (60, 61). Second, we find that invasive bladder tumors have a striking upregulation of immune cells, as evident by expression of Cd45, and also have activation of pathways associated with immune regulation. It is interesting to note that these tumors also display upregulation of pathways associated with hypoxia, given the known relationship of hypoxia and immune checkpoint receptors (e.g., ref. 62). These biological and molecular findings provide opportunities to explore translational approaches to curtail disease progression of patients at risk of progression to invasive disease.

This is of potential clinical relevance because a subset of patients with noninvasive bladder cancer will progress to invasive disease, yet currently there are relatively few ways to detect these patients prior to progression or treatment failure. We have now identified a gene signature of early bladder cancer that is expressed in a subset of patients with NMIBC and associated with adverse disease outcome. Previous studies have identified gene signatures for bladder cancer that are associated basal and luminal subtypes of bladder cancer (e.g., refs. 22–24, 26), or disease prognosis (43, 49–54). This study complements these previous ones because our goal was to define signature to specifically identify patients at risk of developing muscle-invasive disease. Notably, a gene signature identified previously based on comparing expression profiles from laser-capture dissected basal and superficial cells from normal urothelium shares some genes in common with the current signature, and supports a basal origin of invasive bladder cancer (63). Furthermore, a recent study has identified a classifier of low-grade bladder luminal tumors that shares genes in common with the current signature, although the directionality of the genes is opposite in the luminal tumors as one might expect (54). Thus, we propose that the 28-gene signature identified herein can be evaluated prospectively, potentially in conjunction with these other basal and prognostic gene signatures, in biopsies from patients with recurrent or high-risk bladder cancer to identify those at risk of progression.

S. Park reports a patent for Prognostic Gene Signature of Bladder Cancer pending. T.B. Owczarek reports grants from AUA/Dornier Medtech during the conduct of the study and other support from Regeneron outside the submitted work. D.B. Solit reports personal fees from Pfizer, Loxo/Lilly Oncology, Vividion Therapeutics, BridgeBio, Fore Therapeutics, and personal fees from Scorpion Therapeutics outside the submitted work. M.M. Shen reports grants from NIH, Bladder Cancer Advocacy Network, and grants from Mark Foundation for Cancer Research during the conduct of the study. H.A. Al-Ahmadie reports personal fees from AstraZeneca, Janssen Biotech, and personal fees from Bristol Myers Squibb outside the submitted work. C. Abate-Shen reports a patent for CU21285 pending. No disclosures were reported by the other authors.

S. Park: Conceptualization, resources, data curation, formal analysis, validation, investigation, visualization, methodology, writing–original draft, writing–review and editing. L. Rong: Conceptualization, data curation, formal analysis, investigation, visualization, methodology, writing–original draft, writing–review and editing. T.B. Owczarek: Conceptualization, resources, data curation, formal analysis, funding acquisition, investigation, methodology, writing–original draft, writing–review and editing. M. Di Bernardo: Conceptualization, resources, data curation, formal analysis, investigation, visualization, methodology, writing–original draft, writing–review and editing. R.L. Shoulson: Supervision, visualization, methodology, writing–review and editing. C.-W. Chua: Resources, methodology, writing–review and editing. J.Y. Kim: Formal analysis, methodology, writing–review and editing. A. Lankarani: Formal analysis. P. Chakrapani: Formal analysis. T. Syed: Data curation, formal analysis. J.M. McKiernan: Resources, project administration. D.B. Solit: Resources, funding acquisition, project administration, writing–review and editing. M.M. Shen: Resources, funding acquisition, methodology, project administration, writing–review and editing. H.A. Al-Ahmadie: Resources, formal analysis, funding acquisition, project administration, writing–review and editing. C. Abate-Shen: Conceptualization, formal analysis, supervision, funding acquisition, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing.

The authors thank Cathy Mendelsohn for helpful discussion and comments on the manuscript. This work was supported by the Genomics and High-throughput Screening, Oncology Precision Therapeutics and Imaging, and Flow Cytometry core facilities of the Herbert Irving Comprehensive Cancer Center at Columbia University, which are supported in part by NIH/NCI grant #P30 CA013696. Core facilities at Memorial Sloan Kettering Cancer Center are supported in part by NIH/NCI grant #P30 CA008748. This work was also supported in part by NIH grants R01 CA193442 (to C. Abate-Shen), R01 233899 (to H.A. Al-Ahmadie), P01 CA221757 (to C. Abate-Shen), and P50 CA221745 (to D. Solit), and by Innovator Awards from the Bladder Cancer Advocacy Network (to C. Abate-Shen and M. Shen), an Endeavor Award from The Mark Foundation for Cancer Research (to C. Abate-Shen, M. Shen, H.A. Al-Ahmadie, and D. Solit), and by a private gift (to J. McKiernan C. Abate-Shen, H.A. Al-Ahmadie, D. Solit., M. Shen). T. Owczarek was supported by a post-doctoral research fellowship from the Urology Care Foundation Research Scholars Program and Dornier MedTech. C. Abate-Shen is an American Cancer Society Research Professor supported in part by a generous gift from the F.M. Kirby Foundation.