Bladder Tumor Subtype Commitment Occurs in Carcinoma In Situ Driven by Key Signaling Pathways Including ECM Remodeling

Urothelial bladder cancer (UBC) is one of the most common malignancies worldwide. Muscle-invasive bladder cancer (MIBC) is treated by radical cystectomy and perioperative platin-based chemotherapy (1). Large-scale gene-expression analyses identified three distinct major molecular subtypes of MIBC, basal, luminal, and neuronal-like with prognostic and predictive impact (2–7). Furthermore, both basal and luminal MIBC subtypes significantly associated with an increase or a decrease of inflammation, respectively, as gauged by amount of stromal infiltrating T lymphocytes and secondary lymphoid structures that predicted disease-specific survival (8–10).

Urothelial tumors are considered to arise from different noninvasive urothelial lesions such as hyperplasia (HYP; WHO 2016 nomenclature: urothelial proliferation of uncertain malignant potential, UPUMP) and urothelial dysplasia (DYS) but also carcinoma in situ (CIS). Several groups have characterized these lesions as well as different tumor stages for gene mutations to address urothelial bladder cancer staging utilizing whole bladder histopathologic mapping (WBHM; refs. 11–14). WBHM is a unique and powerful tool, which implements comprehensive and systematic tissue sampling encompassing the entire mucosal bladder lining and tumor mass to determine histologic and molecular progression to non- or invasive UBC. A resulting global picture of the whole bladder typically identifies urothelium, a variety of noninvasive urothelial lesions, CIS, and different tumor stages.

Despite the emerging impact of molecular subtyping, there is only limited data about expression profiles of noninvasive urothelial lesions and CIS (15, 16). Using IHC, Barth and colleagues described that the majority of CIS loses their luminal phenotype and gains basal features during progression, indicating that basal tumors may, therefore, develop via luminal CIS (16). So far, WBHM has not been used to study bladder cancer progression on gene expression level. In this study, we investigated when noninvasive urothelial lesions and/or CIS become committed to MIBC subtypes using gene and protein expression of 22 patient matched noninvasive urothelial lesions, CIS, and tumors and one entire bladder for WBHM. In addition, we investigated whether the tumor environment influences the gene expression of normal urothelium and the role of extracellular matrix (ECM) remodeling and oncogenic signaling including relevant drug targets during tumor progression. A model emphasizing the influence of the tumor environment on normal urothelium, noninvasive urothelial lesions, and the transition of CIS commitment to bladder cancer is presented.

Cystectomy specimens are routinely obtained from the surgery room and then opened with a Y-shaped incision. For a single WBHM specimen, 25 defined regions of the bladder were dissected macroscopically where each could potentially house several noninvasive urothelial lesions, CIS, and carcinomas along with urothelium. Every tissue was then bisected to snap-freeze one half and paraffin embed the counterpart after formalin fixation (FFPE). Tissue samples were reviewed by two uropathologists (A. Hartmann and M. Eckstein) according to the Tumor–Node–Metastasis classification [TNM; Union International Contre le Cancer (UICC), Geneva, Switzerland] and WHO classification (World Health Organization, 2016).

Noninvasive urothelial lesions were classified as follows: (i) UPUMP; synonyms include: papillary or flat urothelial hyperplasia/HYP (currently not recommended by WHO 2016) or flat or papillary-like shape (without true fibrovascular cores and papillary fronds) with a thickened urothelium and increased amount of cell layers (in flat areas usually ≥10 cell layers) and cell density, but only minimal cytologic atypia and preserved urothelial stratification and cell polarity; (ii) DYS is described as a flat urothelial lesion with loss of the regular perpendicular architecture of normal urothelium with minimal-to-moderate nuclear atypia in absence of inflammation and not fulfilling the criteria of CIS.

The separate class of CIS is described as flat urothelial lesions devoid of papillary structures with high-grade nuclear atypia (large hyperchromatic pleomorphic nuclei and presence of irregular nucleoli), loss of cell polarity, and irregular nuclear crowding resembling clear high-grade morphology. To prevent a biased overlap of DYS with reactive urothelial lesions, we exclusively included flat/noninvasive urothelial lesions that clearly matched the WHO 2016 criteria by hematoxylin and eosin (H&E) morphology and were therapy naïve (no prior instillation therapies, no prior history of nonmuscle-invasive bladder cancer, no relevant transurethral resection–associated inflammation in the respective localization). For better legibility and easier interpretability for nonpathologists, UPUMP is abbreviated as HYP in the figures.

A comparative subtyping analysis was performed with 22 pairs of patient-matched MIBCs (tumor stage: pT2-pT4), noninvasive urothelial lesions, and CIS [n = 41: 33 CIS, 7 DYS, 1 UPUMP/HYP; clinical and pathologic data including age distribution (52–84 years) of the 22 MIBCs are summarized in Supplementary Table S1A]. MIBC and CIS single samples were obtained from different locations within the bladder to allow analysis of potential differences of multifocal disease. In addition, from one WBHM specimen [identification number (ID) 4] subtyping analysis was performed with noninvasive urothelial lesions and CIS (n = 18: 5 UPUMP/HYP, 6 DYS, 7 CIS), tumor-associated urothelium (n = 5), and different regions of invasive carcinoma (pT1: n = 9; pT2: n = 5). The WBHM MIBC samples were surrounded by 9 different stromal-invasive carcinoma areas (pT1). The WBHM specimen encompassed the full spectrum of noninvasive urothelial lesions (UPUMP/HYP, DYS), CIS, stroma-invasive (pT1) and muscle-invasive tumor parts (MIBC). In total, 99 tissue samples were investigated for molecular subtypes using mRNA and protein analyses. For analysis of signaling pathways, 37 samples [5 tumor-associated urothelium (U), 5 UPUMP/HYP, 6 DYS, 8 CIS, 9 pT1- and 4 pT2a-carcinoma samples] from the WBHM “ID4” were analyzed. One CIS (mapping position M12 “CIS”) exhibited low RNA amounts and was only included in the signaling analysis. In addition, two independent cohorts for comparison of ECM remodeling genes in normal morphologic urothelium were included: (i) tumor associated urothelium (U; n = 7) isolated from seven independent bladder tumors and (ii) normal urothelium (NU; n = 5) isolated from noncancer control patients (both cohorts not included in the comparative subtyping analysis above). To analyze tumor subtype associations between CIS and papillary high-grade tumors (pTa/pT1, high grade), we analyzed an independent tissue microarray cohort of 199 papillary high-grade tumors and 38 urothelial CIS by IHC (Supplementary Table S1B). All tissue specimens were therapy naïve (no prior instillations or transurethral resections).

DNA sequencing libraries were prepared for tumor and matched normal samples (500 ng) using the KAPA Hyper Prep Kit and then sequenced on the Illumina HiSeq sequencing platform at NantOmics. DNA-sequencing data were aligned to the Genome Reference Consortium Human Build 37 (GRCh37, aka hg19) by bwa-mem (v0.7.17; https://sourceforge.net/projects/bio-bwa/files/), duplicates marked by samblaster (version number 0.1.26; https://anaconda.org/bioconda/samblaster), and indel realignment and base quality recalibration performed by GATK v2.3 (RRID:SCR_001876). Clinical Laboratory Improvement Amendments–certified variant analysis was performed using the GPS Cancer analytical pipeline (version 1.1; https://nantomics.com/gpscancer/).

The latest GPS Cancer pipeline (v1.1) classifies all somatic variants into five categories (pathogenic, likely pathogenic, variant of unknown significance, likely benign, and benign). Categories were determined using a combination of variant class (e.g., missense), amino acid change, PhastCons (source: UCSC Genome Browser; RRID:SCR_005780) conservation score of the mutated site, gene type (i.e., tumor suppressor, oncogene), driver status, variant allele frequency in the population from dbSNP (RRID:SCR_002338), and location in a mutational hotspot. The disruption of a particular amino acid change was calculated according to a conservation-controlled amino acid substitution matrix score (CASM; ref. 17), with parameters estimated on the basis of variant calls of >5,000 The Cancer Genome Atlas (TCGA) tumor exomes and their matched-normal tissues. Gene type was obtained from COSMIC Cancer GeneCensus (https://cancer.sanger.ac.uk/census; RRID:SCR_002260; ref. 18), while driver status was obtained from a pan-cancer publication across multiple TCGA cancer types (19). Mutation clusters were discovered using OncodriveCLUST (v0.4.1; http://bg.upf.edu/group/projects/oncodrive-clust.php). Variants found in COSMIC database (release v76; RRID:SCR_002260) were annotated with the number of COSMIC samples harboring mutations that cause the same protein change (18).

Copy-number alterations were determined by the latest GPS Cancer pipeline (v1.1) as follows: relative coverage and majority allele fraction of the tumor sample versus its matched normal were estimated using a single-pass segmentation algorithm that merges fixed-width contiguous regions of the genome unless the estimates of the relative coverage and majority allele fraction of the regions differ in a statistically significant manner (i.e., greater than three SDs). The regions outputted by the single pass segmentation algorithm were corrected for estimated guanine/cytosin (GC) bias. Variable regions with the weakest support are merged with the neighboring region that best matches the region's estimates, and then the newly neighboring regions are merged using the same significance criteria as before. This process was iteratively performed until regions could no longer be merged. Copy-number status for a given region was defined as “amplification” if log₂(rc) > 1.0, “moderate amplification” if log₂(rc) > 0.25, “loss” if log₂ (rc) < −0.25, and “normal” otherwise, where rc was the region's estimate of relative coverage normalized by the total read counts of tumor and matched normal tissue. Mutational data of the MIBC of the WBHM specimen are depicted in Supplementary Table S1A.

Microdissection and RNA isolation of FFPE tissue sections was carried out as described previously (8, 9). The purity of microdissected probes represented approximately 80% tumor or epithelial cells. RNA was extracted by an automated magnetic bead-based approach using the Promega Maxwell 16 Instrument (modified Promega LEV DNA blood kit protocol using the Promega RNA incubation and lysis buffer), purified with chloroform, and measured via Qubit 4 Fluorometer (Thermo Fisher Scientific).

mRNA expression was determined via nCounter MAX/FLEX system (NanoString Technologies). To differentiate luminal and basal phenotypes, a customized 21-gene panel containing nCounter PlexSet (NanoString Technologies), according to the MD Anderson Cancer Center (MDA) subtyping approach, was applied, as described previously (8, 9). SDHA and HPRT1 served as reference genes for data normalization (Supplementary Table S2). Comprehensive gene expression analysis of different signaling pathways like angiogenesis, ECM remodeling, epithelial-to-mesenchymal transition (EMT), and metastasis were performed via the nCounter PanCancer Progression Panel (NanoString Technologies; Supplementary Table S2). In total, 13 reference genes with an average mRNA copy number of ≥ 150 and homogenous data distribution were used for normalization. Fifty-nine target genes with an average expression of 0 were excluded from the analysis. For further investigation, the data were processed and log₂ transformed via the nSolver Software 4.0 (NanoString Technologies). mRNA data distributions are depicted in Supplementary Table S2. Sufficient comparability using NanoString methodology of different tissues, including FFPE tissues and patient fluids to conventional gene-expression methods (RNA sequencing; RNA-seq), has been shown in previous publications (20, 21). Supplementary Figure S1 shows a comparison of RNA-seq and NanoString-based subtyping of MIBCs, from which the 22 MIBCs for this study were chosen.

Tissue samples were stained for KRT5/CK5 (clone XM26, mouse monoclonal, Diagnostic BioSystems, dilution 1:50), KRT14/CK14 (clone SP53, rabbit monoclonal, Cell Marque, dilution 1:40), KRT20/CK20 (clone Ks 20.8, mouse monoclonal, Dako; dilution 1:50), GATA3 (clone L50–823, mouse monoclonal, DCS; dilution 1:100), FOXA1 (rabbit polyclonal, Abcam; dilution 1:400), CD44 (clone DF1485, mouse monoclonal, Dako; dilution 1:50), UPK2 (clone bc21, mouse monoclonal, Biocare Medical, LCC; dilution 1:50), and HER2/neu (c-erbB-2 Oncoprotein, rabbit polyclonal, Dako; dilution 1:200) according to a DAkkS (German accreditation society) accredited staining protocol on a VENTANA BenchMark ULTRA autostainer (Ventana).

The IHC panel above was chosen based on the recommendations for IHC-based subtyping provided by the International Bladder Cancer Molecular Taxonomy Working Group (22). Semiquantitative IHC analysis of the subtyping markers was carried out using the immunoreactive score (IRS). The IRS represents a product of multiplication between the score for the percentage of positive tumor cells (0%–100%) and staining intensity (0 = negative, 1+ weak staining intensity, 2+ moderate staining intensity, 3+ strong staining intensity), resulting in a relative protein expression with a range of 0–12.

All samples of the WBHM specimen were hybridized with the ZytoLight SPEC ERBB2/CEN17 Dual Color Chromogenic In-Situ Hybridization Kit (ZytoVision GmbH). Representative spots of tissue samples were identified on H&E slides and analyzed at 400x-magnification on matching spots of hybridized tissue sections. The total amount of CEN17 and ERBB2-gene locus signals was counted in 25 cells and the ERBB2/CEN17 ratio calculated by dividing the cumulative amount of ERBB2-signals by the amount of CEN17 signals.

RSEM RNA-seq data and clinical data were accessed and downloaded via cBioportal (http://www.cbioportal.org/; RRID:SCR_014555; ref. 23). Assignments for TCGA mRNA subtypes for gene expression correlations between basal and luminal MIBCs were adapted according to a modified MDACC subtyping approach as reported previously (8, 9).

Data for signature score calculation were transformed to Z-scores to obtain comparable datasets without metric dimension as described previously (8, 9). Signature scores were calculated by building the median Z-score of all containing variables. Variables included in the different signature scores are depicted in Supplementary Table S3. Gene-expression pathway scores of the nCounter PanCancer Progression Panel were calculated with the nSolver Software 4.0 using the advanced pathway scoring module based on the R programing language (R 3.3.0; https://www.r-project.org; RRID:SCR_001905) and the R-package KEGGprofile.

To characterize the distributions of continuous variables descriptive statistics [mean, SD, quartiles, median, range] and nominal variables (frequency, percentage) were employed. Multiple group comparisons were carried out using nonparametric Kruskal–Wallis tests. Regular group comparisons were statistically tested with nonparametric Mann–Whitney tests. Supplementary Table S4 shows P values of group comparisons where Kruskal–Wallis tests were applied. All P values were two-sided, and a P < 0.05 was considered statistically significant. Cluster analyses were performed by unsupervised hierarchical clustering based on the average link algorithm (weighted pair group method with arithmetic mean) using Euclidean distance as the metric scale. The statistical analyses were carried out by GraphPad Prism 8.1 (GraphPad Software Inc.; RRID:SCR_002798) and JMP 13.3 (SAS Institute; RRID:SCR_008567). Randomization procedures were not performed (not applicable). Blinding was not performed (not applicable). Power analysis was not applicable to the current experimental study because no clinical outcome measure was investigated. Gender distribution (male > female) of analyzed samples ranged in the usual distribution for MIBC.

This study was approved by the ethical review board of the Friedrich-Alexander-University Erlangen-Nürnberg (approval number: 329_16B and 97_18Bc) in accordance with the Declaration of Helsinki. All patients gave written informed consent to allow use of tumor samples.

Unsupervised hierarchical clustering of 21 subtype-associated genes of 99 samples from patient-matched pairs of MIBCs, CIS, noninvasive urothelial lesions, and the WBHM ID4 revealed three distinct hierarchical clusters (Fig. 1A): a “basal” cluster (n = 23) with predominant expression of CDH3, KRT5, SNAI2, ZEB2, VIM and KRT6A/B/C; a “luminal p53-/ECM-like” cluster (n = 45) demonstrating luminal differentiation and high coexpression of fibroblast and ECM-related genes ACTG2, CNN1, CD44, MFAP4, and MYH11; and a “luminal” cluster (n = 31) with high levels of CDH1, CYP2J2, FGFR3, ERBB2, KRT20, GATA3, ERBB3, and KRT7. CD44, which encodes the hyaluronic acid receptor and is associated with ECM remodeling (24), was significantly upregulated in the basal and luminal p53-/ECM-like clusters. Strikingly, 84.4% of samples classified as luminal p53-/ECM-like were noninvasive urothelial lesions and CIS (53.3% CIS, 17.8% DYS, 13.3% UPUMP/HYP), 11.1% tumor-associated urothelium, and 4.5% pT1 carcinomas (belonging to the WBHM specimen). On the other hand, MIBCs clustered exclusively in basal (52.2% CIS, 39.1% MIBCs, 8.7% DYS) and luminal clusters (54.8% MIBCs, 22.6% pT1 carcinomas, 12.9% CIS, 9.7% DYS; Fig. 1A). Subtype classifications according to tissue sample types are depicted in Fig. 1B.

Histologically, five basal MIBCs of the patient matched pairs showed squamous features (5/8), one sarcomatoid (1/8), and two conventional morphologies [not otherwise specified (NOS); 2/8; Supplementary Table S1A]. Luminal MIBCs of the matched pairs included two carcinomas with micropapillary (2/14), one with plasmacytoid (1/14) and eleven with NOS morphology (11/14; Supplementary Table S1A). The invasive carcinoma samples (pT1/MIBC) of the WBHM specimen exhibited NOS-morphology. Nine percent of CIS lesions (n = 3) presented with a pagetoid spread and two with glandular differentiation (6%; all variants of CIS were luminal p53-ECM-like). The remaining CIS showed conventional urothelial morphology.

Semiquantitative IHC protein subtyping for all tissue samples regarding CK5, CK14, CD44, CK20, GATA3, FOXA1 and UPKII revealed different levels of expression for 20 basal and 79 tissues from luminal tumors/noninvasive urothelial lesions/tumor associated urothelium (Supplementary Fig. S2A). Overall, concordance between protein and mRNA subtyping for all sample types was 96.9%. In matched pairs (n = 22), we further found a strong concordance between the protein and mRNA subtypes of CIS and patient matched MIBCs. For example, in 75% (6/8 pairs) of basal MIBCs, the corresponding CIS were identified as basal by protein and gene expression and for luminal MIBCs concordance was 78.6% (11/14 pairs; Fig. 1C), supporting the hypothesis of early tumor subtype commitment at both the gene and protein expression level.

More rarely noted, only two basal and one luminal MIBC pairs were accompanied by CIS of the opposite subtype (basal: case ID6 and 19; luminal: case ID22; Fig. 1C). CIS and DYS of case 16 differed as both molecularly grouped into the basal cluster, while IHC demonstrated luminal phenotypes. In one of eleven cases with multifocal CIS (1/12, 9.1%; ID12), two different CIS were detected where a corresponding CIS associated with a luminal p53-/ECM-like/luminal phenotype and the other CIS with a basal differentiation. In addition, the basal CIS was accompanied by a second basal MIBC. This indicates that molecularly different MIBCs can occur within the same bladder supporting heterogeneous tumor evolution. Concordance of protein and mRNA subtyping of the WBHM tissues was 97.2% (35/36; Fig. 2A).

We support the idea that bladder cancer progresses from noninvasive urothelial lesions. Thus, we focused on the analyses of one WBHM specimen (ID4). Characterization showed a luminal tumor with partial papillary NOS-morphology, 37 noninvasive urothelial lesions, and CIS supporting a full spectrum of tumorigenesis. The pT1/MIBC samples harbored a pathogenic hotspot TP53-mutation (p.T125M) as well as an ERBB2-locus amplification (fold change, 3.8; Supplementary Table S1A; Supplementary Fig. S2B). mRNA and protein subtyping classified all samples as “luminal” or “luminal p53-/ECM-like” with the exception of a single MIBC, which was “basal” by gene expression but “luminal” by protein expression (35/36; Fig. 1A and B; Fig. 2A). Tumor samples were predominantly “luminal” and surrounded by CIS, noninvasive urothelial lesions, and tumor-associated urothelium of “luminal p53-/ECM-like” phenotypes (Fig. 2A). Because of the absence of an epithelial lining at position M21 and an insufficient amount of RNA for the CIS at position M12, no data were obtained for these two spots (Fig. 2A).

Unsupervised hierarchical clustering of luminal (GATA3, ERBB2, ERBB3, KRT20, KRT7, FGFR3) and basal differentiation markers [KRT5, SNAI2, CD44, (KRT6A-C excluded due to a median expression of 0 in these samples)] demonstrated a striking opposite expression pattern between tumor-associated urothelium, noninvasive urothelial lesions, CIS, and luminal carcinomas. Tumor-associated urothelium (n = 4), UPUMP/HYP (n = 4), and DYS (n = 3) represented a “normal-like” cluster with preserved urothelial expression of basal (KRT5/SNAI2/CD44) and luminal markers (GATA3/KRT20) as well as a normal morphologic distribution of these markers determined by IHC (Fig. 2B). Noninvasive urothelial lesions and CIS (1 HYP, 3 DYS, 4 CIS) and one tumor-associated urothelium surrounding the pT1/MIBC tissues, supports a stepwise oncogenic transformation (“transforming-like” cluster). Progression occurred by the loss of some basal markers and the gain of luminal markers with the exception of one CIS allocated to the “normal-like” cluster, which was most likely caused due to its pagetoid spread within the normal urothelial lining and consecutive “contamination” with normal urothelial cells. Two CIS and all pT1/MIBC samples represented an “invasive-like” cluster characterized by overexpression of GATA3, ERBB2, ERBB3, KRT20, and KRT7 and downregulation/absence of basal markers and FGFR3 expression indicating complete transformation to a luminal phenotype of invasive carcinomas (Fig. 2B). Distribution of the classifications within the WBHM specimen is depicted in Supplementary Fig. S2C.

To further validate and compare luminal with basal tumor progression, we built combined differentiation scores consisting of basal (genes: CD44, KRT5, KRT6A-C, SNAI2; proteins: CD44, CK5, CK14) or luminal (genes: CYP2J2, ERBB2, ERBB3, FGFR3, GATA3, KRT7, KRT20; proteins: CK20, FOXA1, GATA3, UPKII) markers for single MIBC and CIS pairs (n = 22) and the WBHM samples. Applying these scores (Fig. 2B–D), as stratified according to histology, we found a striking increase of “luminal-ness” and in parallel a decrease of “basal-ness,” which occurred mainly during the transition of “transforming” CIS lesions to luminal MIBC. Compared with pT1/MIBC samples the “basal-ness” of luminal CIS was significantly higher in the WBHM (P < 0.0001) and single samples (P = 0.016) while the “luminal-ness” further increased during transition to pT1 carcinomas/MIBCs by a parallel decrease of “basal-ness” (WBHM samples: P = 0.023; single samples: P = 0.0089; Fig. 2C and D). Regarding basal tumor progression, we also found an opposing relation during the transition from CIS to MIBC, where expression of luminal markers decreased along with a parallel upregulation of basal differentiation markers (Fig. 2E). These processes were morphologically detectable by IHC as demonstrated by two transitioning cases of CIS to luminal and basal tumors (Fig. 3).

High-grade papillary urothelial carcinomas and CIS are both high-grade lesions, which are differentiated by either flat configuration (CIS) or formation of true papillae with stromal cores carrying lymphatic and blood vessels. These lesions are often similar based on cytologic criteria and are sometimes difficult to discriminate in transurethral resection specimens. Correct differential diagnosis of high-grade papillary tumors versus CIS is important because treatment strategies can differ greatly (e.g., follow up in papillary high-grade tumors without CIS after resection vs. adjuvant BCG-instillation if CIS present). To analyze whether a subtype analysis might help to distinguish these lesions, we applied IHC subtyping to an independent cohort of 199 high-grade papillary tumors and 78 CIS (38 new cases of CIS, all with conventional morphology; and the previously analysed 40 CIS from single samples and the WBHM specimen). Results showed that 238 tissue samples classified as luminal (CIS: n = 59, 75.6%; pTa: n = 29, 85.3%; pT1: n = 150, 90.9%) or 39 as basal subtypes (CIS: n = 19, 24.4%; pTa: n = 5, 14.7%; pT1: n = 15, 9.1%; Fig. 4A). However, we noted that the luminal cluster separated into two distinct lesion types: (i) a “CK20 ++/+++” cluster with moderate to strong expression of CK20 (71.8% of luminal lesions; 171/238), and (ii) a “CK20 −/+” with absent or low CK20 expression (28.2%, 67/238; Fig. 4A). Distribution of different phenotypes across CIS, pTa, and pT1 high-grade lesions is depicted in Fig. 4B. Remarkably, 96.6% of luminal CIS (57/59) showed a “CK20 ++/+++” phenotype, which could be found in 68.0% of papillary pT1 high-grade carcinomas (102/150) and 41.4% (12/29) of papillary pTa high-grade carcinomas.

Although subtyping does not allow a clear discrimination between CIS, pTa, and pT1 high-grade carcinomas, we further analyzed the utility of the subtyping IHC panel to support the diagnostic verification of CIS. We categorized subtyping marker expression based on protein expression strength and tissue distribution as depicted in Fig. 4C. For example, a regular distribution of subtyping markers resembling normal urothelial architecture with basal stratification of CD44 and/or CK5 and luminal stratification of CK20 (in umbrella cells) was found in all normal urothelium (n = 5), U (n = 12) and 50.0% of UPUMP/HYP (3/6) and 15.4% of DYS (2/13) (Fig. 4C). Aberrant full thickness with a high intensity of CK20 staining was present in all 59 luminal CIS (100%), 2 HYP/UPUMP (33.3%), and 2 DYS (15.4%), but was not identified in all 19 basal CIS (0%; Fig. 4C). On the other hand, 63.2% of basal CIS (12/19), 69.2% of DYS (9/13), and 50.0% of UPUMP/HYP (3/6) did show a weak intensity/patchy and partially full thickness staining of CK20. In addition, all basal CIS exhibited a strong aberrant overexpression of CK5 and/or CK14 (19/19; 100.0%) with mostly an aberrant full thickness of CD44 expression (17/19; 89.5%), which was also present in 8 luminal CIS (13.6%) and 6 DYS (46.2%; Fig. 4C). Thus, subtyping seems to be sufficient to confirm CIS, but not helpful for differential diagnosis against reactive changes, DYS, or UPUMP/HYP.

Based upon our initial findings where noninvasive urothelial lesions and CIS expressed high levels of genes associating with ECM remodeling and cancer-associated fibroblasts (CAF; luminal p53-/ECM-like; Fig. 1A and B), we analyzed specific gene sets and established expression scores for comparing normal urothelium, noninvasive urothelial lesions, CIS, and carcinomas. Five genes were chosen for calculating an ECM remodeling (ACTG2, CNN1, CD44, MFAP4, and MYH11) and three for an EMT expression score (SNAI2, ZEB2, and VIM). Hierarchical clustering of the scores revealed a strong upregulation of ECM remodeling in noninvasive urothelial lesions, CIS, and tumor-associated urothelium (Fig. 4D). In contrast, MIBCs exhibited low expression of ECM remodeling genes but high EMT expression (Fig. 4D).

Because of the observation that tumor-associated urothelium coclustered with noninvasive urothelial lesions and CIS in terms of ECM-like gene expression (Fig. 1A; Fig. 4D), we asked the question whether this could be due to the influence of the tumor environment. An independent cohort of tumor-associated urothelium from MIBCs (n = 7) demonstrated a similar ECM remodeling score comparable with tumor-associated urothelium, noninvasive urothelial lesions, and CIS from the matched sample pairs and the WBHM specimen, thus confirming our initial results (Fig. 4E). However, in strong contrast, normal urothelium from noncancer patients (n = 5) revealed a significant lower ECM score compared with tumor-associated urothelium and noninvasive urothelial lesions (Fig. 4E). These findings support an important early role of ECM remodeling during tumor progression mediated by the tumor environment. Furthermore, the ECM score decreased in invasive tumors (Kruskal–Wallis test for multiple group comparisons: P < 0.0001; Fig. 4F; single group P values in Supplementary Table S4), while the EMT score represented the lowest expression during pT1 progression with a subsequent increase in MIBCs (Kruskal–Wallis test for multiple group comparisons: P = 0.0004; Fig. 4G; single group comparison P values are depicted in Supplementary Table S4). Congruent with prior reports, the sarcomatoid MIBC showed the highest EMT expression score (Fig. 4G; refs. 25, 26).

To further investigate how genetic pathways influence bladder cancer progression, we applied the nCounter PanCancer Progression Panel to all 37 samples of the WBHM specimen. Unsupervised hierarchical clustering of 681 expressed target genes revealed four major gene clusters (A*–D*; Supplementary Table S5A), and lesions were stratified into three major lesion clusters (Supplementary Fig. S3A). Lesion cluster “A” consisted of tumor-associated urothelium (n = 5), UPUMP/HYP (n = 4), DYS (n = 4), and the single CIS with a pagetoid spread pattern, and was enriched with genes associating with ECM-remodeling, cell motility/polarity, cell differentiation, angiogenesis, and wound-healing (Supplementary Fig. S3A and S3B; Supplementary Table S3). Lesion cluster “C” exclusively was represented by pT1 carcinomas (n = 7) and MIBCs (n = 3) and associated with genes related to cellular growth factors, HIF1α signaling, ERBB2/ERBB3 signaling, and EMT (Supplementary Fig. S3A and S3B). Lesion cluster “B” contained one UPUMP/HYP, two DYS and CIS (n = 7), as well as two pT1 carcinomas, which surrounded the MIBC tissues. This cluster showed variable expression of all genes (A*–D*), thus further supporting the importance of the above mentioned pathways in tumor progression (Supplementary Fig. S3C).

Unsupervised hierarchical clustering of signaling pathways yielded striking differences between the lesion cohorts (Fig. 5A). As predicted lesion cluster “A” showed an increased upregulation of gene pathways related to ECM remodeling, angiogenesis, hypoxia, and TGFβ signaling (Fig. 5A; Supplementary Fig. S3A). In contrast, CIS was enriched within cluster “B” and demonstrated an upregulation of oncogenic ERBB2/3 signaling due to ERBB2 amplification along with other genes involved in plasma membrane component remodeling, cell proliferation, cellular growth factors, cell-cycle dysregulation, and EMT (Fig. 5A; Supplementary Fig. S3A). These pathways were further highly upregulated in invasive carcinoma samples in lesion cluster “C” (Fig. 5A). Figure 5B–F and Supplementary Table S5B and S5C illustrate the most differentially expressed genes and signaling pathways between the different histologic entities and lesion cohorts A/C, respectively (Fig. 5B–F; P values derived by Kruskal–Wallis tests for multiple group comparisons across the different stages are depicted below the graphs; single-group comparison P values are summarized in Supplementary Table S4). The striking gene pathway differences between the cohort clusters points to tumor progression from noninvasive urothelial lesions. In detail, candidate genes like COL4A1 (P < 0.0001) and COL4A2 (P = 0.0006) were upregulated during the transition to invasive carcinoma (Fig. 5B), while matrix metalloproteinases like MMP2 (P = 0.0009) and MMP14 (P = 0.0002) were higher expressed in noninvasive urothelial lesions and CIS (Fig. 5B). “Disintegrin genes,” ADAM15 (P = 0.0002) and ADAM17 (P = 0.0025), showed higher levels in MIBCs (Fig. 5B). Additional upregulated genes associating with tumor progression were CHI3L1, HMOX1, PDGFRB, and SPP1 (Fig. 5C), while TGFβ signaling including SMAD3 decreased during the transition to a carcinoma (Fig. 5A and C). ERBB2/3 levels otherwise increased during progression to an invasive carcinoma due to an ERBB2-gene locus amplification noted within the CIS [fold change 3.8× by next-generation sequencing (NGS); mean ERBB2/CEN7-Ratio = 2.66; Supplementary Table S1A; Supplementary Fig. S2C]. Other important signaling genes like EGFR, FGFR3, and STAT3 decreased during tumor progression (Fig. 5D). In parallel, CADM1, a STAT3 regulator, increased supporting pathway regulation. As demonstrated, EGFR and STAT3 signaling associate with basal tumors while ERBB2/ERBB3 signaling is linked with luminal tumors (Supplementary Fig. S3D). Proto-oncogenes like KRAS, MYCL, PIK3CA (gene locus amplification: 5.7-fold change), PTTG1, and RAF1 were upregulated in carcinomas, while JUN and PTEN were downregulated (Fig. 5E and F). Cell-cycle regulators like CDKN1A were 25-fold decreased in carcinomas (compared with precursors), or in the case of the tumor suppressor CDKN2A, no expression was detected due to a homozygous deletion on chromosome 9p.

The normal urothelial lining comprises a luminal cell layer with expression of uroplakins and CK20, while basal cell layers express CK5 and CD44 (27). Aberrant staining patterns of these markers with loss of physiologic stratification have been proposed to be of diagnostic importance of CIS (28). GATA3 and FOXA1 transcription factors belong to ESR1 signaling and are highly expressed in all cell layers of normal urothelium, thus representing markers for regular urothelial differentiation (27). Urothelial carcinoma is considered to originate from malignant transformed urothelial progenitor cells driven by specific genetic alterations (e.g., loss of chromosome 9p), gene mutations (e.g., TP53 or FGFR3), and aberrant hyper- and hypomethylation of gene families (12, 13, 27, 29, 30). However, little is known about the plasticity and evolution of UBC subtypes during tumorigenesis. Multifocal carcinomas are the result of clonal evolution of malignant precursor cells, which spread throughout the urothelial lining via intraluminal seeding or intraepithelial/pagetoid migration (12, 13, 27, 29, 30). In line with the above findings, we demonstrate that particularly CIS and corresponding patient matched invasive tumors share the same gene and protein expression phenotypes, which define MIBC subtypes. This indicates that commitment to a particular tumor subtype occurs at earlier stages of tumor progression in the majority of MIBCs. In contrast to prior studies, which classified CIS exclusively as luminal lesions, we clearly demonstrate that basal subtype commitment already occurs at the stage of CIS (16, 31). From the diagnostic point of view, the addition of CK5 and CK14 to the infrequently used diagnostic CIS panel including p53, CK20, and CD44 IHC might help to classify the so called “CK20-negative CIS” as urothelial CIS with aberrant overexpression of CK5, CK14, and CD44 resembling basal CIS (28, 32). However, this aberrant expression pattern is also found in DYS and UPUMP/HYP and high-grade papillary carcinomas. Especially DYS and UPUMP/HYP are urothelial lesions difficult to define due to the lack of clear definitions, frequent overlaps with reactive atypia in presence of inflammation and other reactive urothelial lesions as well as high interobserver variability (33). Thus, occurrence of these aberrant expression patterns should be carefully used to verify CIS in cases with strong suspicion, but not to discriminate between CIS, DYS, reactive urothelial changes (e.g., after instillation treatments) or other lesions showing overlaps with papillary urothelial carcinomas like CIS with papillary formations and atypical UPUMP/HYP (34).

A few cases in our study presented with coexisting luminal and basal CIS and/or MIBC in the same bladder. This could hint to an oligoclonal evolution from independent precursor cells of these tumor clones, which is postulated by the field cancerization theory; however, this has to be addressed in upcoming studies (27, 29, 30). This phenomenon of cooccurring contrary subtypes in clonally related tumor samples has been previously described by Thomsen and colleagues (35) and similar results have been reported for subtype distribution in differentiated tumor areas or tumor areas with variant histology of MIBC revealing the highest heterogeneity for basal carcinomas (31). Because luminal and basal tumors respond differently to systemic treatment regimens like platinum-based chemotherapy, subtype heterogeneity is of major clinical importance and might contribute to chemotherapy resistance (5, 6).

Our findings demonstrate that the majority of noninvasive urothelial lesions and CIS, along with tumor-associated urothelium, corresponds with a luminal p53-/ECM-like phenotype. Importantly, this luminal p53-/ECM-like phenotype has only been identified in MIBC where it was termed “infiltrated” or “p53-like” based on its gene expression signature that is mostly attributed to ECM remodeling and CAFs (2–4, 36). The ECM remodeling signature and the differences between noninvasive urothelial lesions, CIS, pT1 carcinomas, and MIBCs likely stems from the composition of the stromal component: (i) activated resident fibroblasts in the submucosal layer in U and noninvasive urothelial lesions; (ii) CAFs in the desmoplastic stroma of stromal invasive pT1 tumor parts or (iii) myofibroblastic CAFs or smooth muscle cells in MIBC. Although it is not an established fact that UPUMP/HYP and DYS are direct precursor stages in urothelial carcinogenesis, the marked transcriptional differences between normal urothelium (noncancer patients) with lower ECM remodeling versus U, UPUMP/HYP, and DYS support that early initiation of a complex transdifferentiation process is present in noninvasive urothelial lesions contributing to progression and subtype commitment. However, this has to be validated in upcoming studies by integrating gene expression and clonality status of these lesions.

In contrast to MIBC, significant overexpressed pathways like TGFβ signaling, cell motility, MMP/LOX remodeling, or basement membrane rearrangement mainly associated with tumor-associated urothelium, UPUMP/HYP, DYS, and CIS, which supports a critical role of these signaling pathways in noninvasive urothelial lesions and at the transition from CIS to MIBC. The relatively low expression of TGFβ signaling in invasive UBC might be explained by the strong upregulation of PTTG1 during tumorigenesis, which has been shown to suppress SMAD3 signaling independent of TGFβ, thus promoting EMT, tumor growth, invasiveness, and metastatic spread (37). Furthermore, PTTG1 is a target of rapamycin or Rac1-inhibitors, and therefore might represent a therapeutic target in a subset of MIBCs (38, 39).

Downregulation of integrins, laminins, and collagens, important for maintaining cell differentiation, cell polarity, and regulating cell motility, accompanied by ECM remodeling and TGFβ signaling in both noninvasive urothelial lesions and CIS may recapitulate the pathognomonic loss of cell polarity and cellular cohesion during carcinogenesis (24). ECM restructuring also included a continuous upregulation of COL4A1 and COL4A2 reaching its maximum in some MIBCs that both are usually lowly expressed in regular epithelial tissues and associated stroma, and proposed to promote ECM remodeling, invasiveness, and EMT-related processes like tumor budding at the invasion front of invasive carcinomas (40). Other genes, such as ADAM15, ADAM17, CHI3L1, HMOX1, MMP9, PDGFRB, and SPP1, which promote cell proliferation, ECM remodeling, EMT activation, CD44- and integrin-mediated invasion, and metastasis and angiogenesis in several cancers, were overexpressed in invasive UBC (41–50). Remarkably, CIS, but also noninvasive urothelial lesions, which were adjacent to carcinomas, revealed strikingly higher SPP1 levels compared with nonadjacent lesions supporting a crucial role of SPP1 in promoting invasion by regulation of MMP9 (45, 46). Interestingly, a further poorly understood gene, CHI3L1, a chitinase-like glycoprotein without chitinase function, was one of the most prominently upregulated genes in invasive carcinomas, and has been shown to promote ECM remodeling, invasiveness, and angiogenesis, thus supporting a role of progression to MIBC (41, 42).

Overexpression of ERBB2/ERBB3, KRAS, PIK3CA, and RAF1 signaling and downregulation of regulatory counterparts, such as PTEN, emerged as dominant oncogenic driver signatures. Especially, ERBB2/ERBB3 overexpression is a consistently reported molecular feature of luminal MIBCs; however, little is known about the underlying regulation mechanisms (3). ADAM15 and ADAM17 are known to release HER ligands from their precursor forms (47). For example, ADAM15 can transactivate ERBB2 or ERBB3 and promote cancer cell invasion (47, 51–53). These findings are in line with our results demonstrating upregulation of ADAM15 and ADAM17 along with increased ERBB2 signaling in both CIS and invasive carcinoma samples of the WBHM ID4 specimen. Previous data further indicating that ADAM17 promotes trastuzumab and platinum-based chemotherapy resistance might explain the low efficacy of HER2/3-targeting drugs and platinum-based chemotherapy in luminal urothelial carcinomas with ERBB2/ERBB3 amplification/overexpression and parallel upregulation of ADAM15 and ADAM17 (3, 6, 54). Therefore, exploitation of ADAM15 and ADAM17 might be a promising strategy to sensitize these tumors to anti-HER2/3 treatment and chemotherapy. Addressing this potential interplay of ADAM15, ADAM17, and ERBB2 is of special importance because high ERBB2 is mostly present in the luminal MIBC subtype especially in the aggressive micropapillary variant demonstrating frequent ERBB2 gene amplifications (55, 56) shown to respond poorly to neoadjuvant platinum-based chemotherapy (6).

Upregulation of FGFR3 signaling due to amplifications or activating point mutations, which is a frequent key driver in luminal UBC, particularly in nonmuscle-invasive UBC, did not play a crucial role for cancer progression in the WBHM ID4 specimen matching with a FGFR3 wild-type (57). EGFR and STAT3 signaling, which are key features of basal MIBCs, were downregulated (2, 3, 57). Interestingly, CADM1, a tumor suppressor and STAT3 inhibitor, whose expression loss is crucial for STAT3 activation in squamous cell carcinomas/basal-like tumors (58, 59) was overexpressed in the luminal UBCs of the WBHM ID4 specimen. This points to a positive or negative regulation of CADM1, which might contribute to either basal-squamous (i.e., via downregulation of CADM1) or luminal (i.e., via upregulation of CADM1) tumorigenesis via STAT3. This hypothesis is further supported by upregulation of CADM1 and downregulation of STAT3 in luminal and vice versa in basal carcinomas of the TCGA bladder cancer (BLCA) cohort.

Furthermore, the results from the WBHM specimen revealed the dynamics of tumor suppressors like inactivation of PTEN and TP53 (60). PTEN was downregulated, while TP53 expression was upregulated on mRNA and protein level due to a truncating point mutation (p.T125M) known to associate with TP53 overexpression in UBC (61). CDKN1A, a tumor suppressing downstream effector of TP53 (62), revealed significant downregulation during tumor progression while the proto-oncogene MYCL, which belongs to the MYC family, was strongly upregulated (63). CDKN2A located on chromosome 9p was not expressed, which is congruent with our findings of a homozygous CDKN2A deletion of tumor samples (loss of 9p) and previous reports from our group that 9p deletions already occur in normal appearing urothelium, noninvasive urothelial lesions, and CIS (14, 64).

In conclusion, our study provides new insights into the development and commitment of bladder cancer subtypes and signaling pathways associating with progression (Fig. 6; schematic model). The transforming process from a normal urothelial cell to a carcinoma is not only accompanied by genomic mutations and morphologic changes, but also determined by changes in mRNA expression including upregulation of genes regulating cancer stemness, ECM remodeling, proto-oncogenes, and growth factor signaling as well as subsequent downregulation of tumor suppression. The power of analyzing all tissues using the WBHM strategy brings forth new molecular knowledge regarding the entire carcinogenic process from preneoplastic changes to invasive carcinomas. In future studies, it will be important to perform functional studies to understand the molecular basis of how tumor subtypes as well as signal transduction of these key pathways are regulated.

Noninvasive urothelial lesions like DYS and UPUMP/HYP are acknowledged categories by the current 2016 WHO classification, but are poorly defined without clear morphologic characteristics, accompanied by high interobserver variability and show frequent overlaps with reactive urothelial lesions. Although it is known that mutations can occur in these lesions, research is ongoing to establish whether these lesions can progress to a neoplastic state, thus belonging to urothelial carcinogenesis. Implementing future molecular studies will be essential to clarify this important question.

A. Wullweber reports grants from German Cancer Aid during the conduct of the study. B. Wullich reports personal fees and nonfinancial support from Janssen, AstraZeneca, and personal fees and nonfinancial support from MSD outside the submitted work. V. Weyerer reports grants from University Hospital Erlangen-Nürnberg during the conduct of the study. M. Burger reports personal fees and nonfinancial support from BMS, MSD, Bayer, Janssen, Medac, and personal fees and nonfinancial support from Apogepha outside the submitted work. A. Hartmann reports personal fees from BMS, MSD, Roche, Astra Zeneca, Boehringer Ingelheim, Abbvie, Cepheid, Ipsen, Janssen, Pfizer, Diaceutics, and personal fees from Quiagen outside the submitted work. M. Eckstein reports grants from IZKF FAU Erlangen-Nürnberg during the conduct of the study; grants, personal fees, and nonfinancial support from AstraZeneca and Cepheid, Janssen-Cilag; grants from STRATIFYER; personal fees and nonfinancial support from GenomicHealth, Roche, Astellas, Diaceutics; and personal fees and nonfinancial support from MSD outside the submitted work. No other disclosures were reported.

A. Wullweber: Data curation, software, formal analysis, validation, investigation, visualization, methodology, writing–original draft, writing–review and editing. R. Strick: Investigation, visualization, methodology, writing–review and editing. F. Lange: Resources, data curation, investigation, visualization, methodology, writing–review and editing. D. Sikic: Resources, data curation, investigation, methodology, writing–review and editing. H. Taubert: Resources, data curation, investigation, methodology, writing–review and editing. S. Wach: Resources, data curation, investigation, methodology, writing–review and editing. B. Wullich: Resources, data curation, supervision, methodology, writing–review and editing. S. Bertz: Data curation, validation, investigation, methodology, writing–review and editing. V. Weyerer: Resources, software, validation, investigation, methodology, writing–review and editing. R. Stoehr: Resources, data curation, supervision, investigation, methodology, writing–review and editing. J. Breyer: Formal analysis, validation, investigation, writing–review and editing. M. Burger: Formal analysis, validation, investigation, writing–review and editing. A. Hartmann: Conceptualization, resources, data curation, supervision, validation, investigation, visualization, methodology, writing–original draft, writing–review and editing. P.L. Strissel: Resources, formal analysis, supervision, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing. M. Eckstein: Conceptualization, resources, data curation, software, formal analysis, supervision, funding acquisition, validation, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing.

This work was performed in (partial) fulfillment of the requirements for obtaining the degree (MD) of the Friedrich-Alexander-Universität Erlangen-Nürnberg, Medizinische Fakultät. A. Wullweber was supported by the German Cancer Aid with a Mildred-Scheel doctoral scholarship (No. 70112604). M. Eckstein and the current project is supported by the ELAN program (grant number P060) and the advanced module for clinician scientists (IZKF Erlangen) of the medical faculty of the Friedrich-Alexander-Universität Erlangen-Nürnberg.

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.