Prognostic impact of molecular muscle-invasive bladder cancer subtyping approaches and correlations with variant histology in a population-based mono-institutional cystectomy cohort

Purpose Recently discovered molecular classifications for urothelial bladder cancer appeared to be promising prognostic and predictive biomarkers. The present study was conducted to evaluate the prognostic impact of molecular subtypes assessed by two different methodologies (gene and protein expression), to compare these two approaches and to correlate molecular with histological subtypes in a consecutively collected, mono-institutional muscle-invasive bladder cancer (MIBC) cohort. Methods 193 MIBC were pathologically re-evaluated and molecular subtypes were assessed on mRNA (NanoString technology, modified 21-gene-containing MDACC approach) and protein levels (immuno-histochemical [IHC] analysis of CK5, CK14, CD44, CK20, GATA3 and FOXA1). Descriptive statistical methods and uni-/multi-variable survival models were employed to analyze derived data. Results Neither gene expression nor protein-based subtyping showed significant associations with disease-specific (DSS) or recurrence-free survival (RFS). Agreement between mRNA (reference) and protein-based subtyping amounted 68.6% for basal, 76.1% for luminal and 50.0% for double-negative tumors. Histological subtypes associated with RFS in uni-variable (P = 0.03), but not in multivariable survival analyses. Tumors with variant histology predominantly showed luminal subtypes (gene expression subtyping: 36/55 cases, 65.5%; protein subtyping: 44/55 cases, 80.0%). Squamous differentiation significantly associated with basal subtypes (gene expression subtyping: 44/45 squamous cases, 97.8%; protein subtyping: 36/45 cases, 80.0%). Conclusion In our consecutive cystectomy cohort, neither gene, protein expression-based subtyping, nor histological subtypes associated with DSS or RFS in multi-variably adjusted survival analyses. Application of a limited IHC subtyping marker panel showed high concordance of 83.9% with gene expression-based subtyping, thus underlining the utility for subtyping in pathological routine diagnostics. In addition, histological MIBC subtypes are strong indicators for intrinsic subtypes. Supplementary Information The online version contains supplementary material available at 10.1007/s00345-021-03788-1.


Introduction
Urothelial bladder cancer (UBC) is among the tenth-most common cancers worldwide, and is a prime example of carcinogen-derived cancer with smoking representing the main risk factor [1]. Based on the depth of invasion, UBC is histologically classified as non-(NMIBC) as well as muscleinvasive bladder cancer (MIBC) [2]. Additionally, these histologically defined borders associate with distinct molecular and clinical behaviors: 80% of bladder tumors are NMIBC, frequently recur and often harbor FGFR3 alterations as oncogenic drivers. In contrast, MIBC is a life-threatening disease with five-year survival rates between 40 and 60% and is mainly driven by TP53 mutations [3].
Modern high-throughput genomic technologies rapidly led to new insights in molecular heterogeneity of UBC, e.g., distinct luminal, basal and neuroendocrine molecular subtypes identified by transcriptional analyses. Several widely overlapping subtyping approaches with heterogeneous nomenclatures have been proposed mainly for MIBC [4][5][6][7][8][9], which were summarized into six consensus subtypes in 2020 coordinated by the Bladder Cancer Molecular Taxonomy Group (BCMTG) [10]. The discovery of these three major molecular phenotypes, luminal, basal and neuroendocrine-like tumors, represents a milestone in bladder cancer research. Luminal tumors can be subdivided into less aggressive "papillary"-like tumors exhibiting expression profiles of non-invasive high-grade UBC, along with two further subtypes, which share an EMT gene signature and are highly invasive [8,10]. Basal tumors exhibit gene expression related to squamous differentiation and are highly proliferative while neuroendocrine carcinomas show classical expression of neuroendocrine differentiation markers [8,10]. While earlier studies suggested an improved response to neoadjuvant chemotherapy (NAC) in basal carcinomas [11], more recent findings indicate that NAC response is rather dependent on immune infiltration than intrinsic differentiation [12]. However, until today, validation of the utility of different subtyping approaches (mRNA vs. protein) in consecutive population-based cohorts is still rare and heavily needed to explore its true predictive and prognostic potential. Importantly, smaller case series also suggested that subtypes correlate with histological subtypes/variant histologies of MIBC [2,5,6,8,10,13,14], but systematic analysis in coherent cohorts is lacking.
Here, we studied the prognostic impact of two subtyping approaches, i) a gene expression-based classifier (adapted MDACC approach [5]), and ii) a protein-based classifier (consensus panel by the BCMTG) in a consecutive population-based mono-institutional cohort of 193 patients treated with radical cystectomy and bilateral lymphadenectomy. Furthermore, subtype assignments were correlated with histological MIBC subtypes.

Study population
We studied 193 MIBC cases of the Comprehensive Cancer Center Erlangen Metropole Region Nuremberg (CCC-EMN) cohort collected from 2004 until 2016. As previously, described, included patients were treated with radical cystectomy and bilateral lymph node dissection in curative intent [15][16][17]. 56 patients received adjuvant platinumbased chemotherapy within 3 months after cystectomy when patients recovered from surgery. No patient was treated with neoadjuvant chemotherapy. All cases were routinely mapped for approximately 23 standardized positions as described previously [18,19]. All cases were systematically reviewed by two experienced uropathologists (A.H., M.E.) according to the latest TNM staging manual of the UICC 8th edition, 2017 and the WHO 2016 classification for tumors of the genitourinary tract [2]. Histological variants were reported if at least 10% of tumor showed variant histology including pure and mixed tumors [20,21]. Clinico-pathologic characteristics of the analyzed cohort are shown in Supplementary  Table S1. Ethical approval for this study was obtained by the ethical review board of the Friedrich-Alexander-University Erlangen-Nürnberg (Erlangen, Germany; approval number: no. 3755 and 329_16B). All patients gave informed consent, and all analyses were carried out in accordance with the Declaration of Helsinki.

RNA isolation and gene expression-based molecular classification
FFPE sections with at least 50% tumor content were used for isolation. As previously described, RNA was isolated with Maxwell Promega RNA purification kits (Promega). The mRNA expression of target genes was determined via nCounter ® MAX/FLEX system (NanoString Technologies ® , USA). To differentiate luminal, basal and double-negative phenotypes, a customized 21-gene-containing nCounter ® PlexSet ™ (NanoString Technologies ® ) according to the MD Anderson Cancer Center (MDA) subtyping approach was applied as described previously [5,17]. Selected genes are summarized in Supplementary Table 2. Gene counts were normalized using two reference genes (SDHA, HPRT1) and log2-transformed for further analysis using the nSolver 4.0 software (NanoString).

Statistical analysis
Cluster analysis was performed by unsupervised hierarchical clustering based on Ward clustering method using Euclidean distance as metric scale. Disease-specific (DSS) and

IHC-based molecular subtypes in MIBC and their associations with classification approaches
Protein-based subtyping was carried out by applying a sixmarker subtyping panel proposed by the BCMTG. Representative images of IHC stainings are shown in Fig. 2A.

Discussion
Recent discovery of molecular subtypes in MIBC offers a possible and promising route to improvement of personalized treatment strategies for MIBC patients [23]. The present study was conducted to i) explore the prognostic impact of two different subtyping approaches (mRNA-and proteinbased) in a consecutive population-based cystectomy cohort, (ii) to explore the clinical utility of a protein-based subtyping approach, and (iii) to explore associations of histological MIBC subtypes with molecular tumor subtypes. Several studies suggested that molecular MIBC subtypes correlate with prognosis and response to neoadjuvant chemotherapy [10,11]. However, recent meta-analysis of 2411  and 1750 inhomogeneously collected transcriptomic profiles of UBC were only able to show significant prognostic differences between MIBC with luminal-papillary-like expression profiles compared to neuroendocrine-like subtypes, which present with well-known aggressive clinical behavior [10,24]. In addition, it has to be mentioned that stromal invasive carcinomas have frequently been included in these studies, and no systematic pathological re-review has been conducted in the underlying original patient cohorts. Therefore, these data might be biased by early-invasive papillary carcinomas not matching the criteria of MIBC, which have a significantly better baseline prognosis than true MIBC [24,25]. In contrast to these meta-analyses and matching with this hypothesis of a systematic bias by early-invasive carcinomas, a study by Kollberg et al. failed to demonstrate a prognostic impact of molecular subtypes in a populationbased consecutive cystectomy series with 519 patients who underwent radical cystectomy between 2006 and 2011 in Sweden [26]. Neither gene expression consensus subtypes nor protein-based subtyping showed any association with clinical outcome [26]. These observations are in line with our present findings in 193 consecutively collected MIBC patients where no association of mRNA-or protein-based subtyping with DSS or RFS could be found. These huge differences in terms of clinical outcomes between inhomogenously collected large meta-analysis cohorts and homogenous consecutive population-based cystectomy cohorts underlines the urgent need for prospective observational multicenter trials to investigate the true prognostic and predictive potential of molecular MIBC subtypes. Prospective validation of outcome as well as predictive effects by molecular subtypes will be critical to evaluate the utility of genomic subtyping. Another major issue of molecular MIBC subtyping beside inconsistent data on the prognostic value is the mode of subtype assessment. Most subtyping approaches including the recently published consensus subtyping approach base on whole transcriptome mRNA profiling, which is a timeconsuming and expensive method, which requires high RNA quality and huge bio-informatical resources [27]. Thus, the BCMTG defined an IHC marker panel with the intent to provide a reduced consensus panel of routinely applicable IHC markers, which should be sufficient to determine luminal and basal differentiation [22]. By applying this IHC marker panel, we were able to demonstrate a substantial overlap with gene expression-based subtypes (83.9%). However, discordant classifications occurred in 16.1% and were mainly observable in luminal tumors. Similar to our findings, Guo and coworkers' evaluation of GATA3 and CK5/6 correctly identified approximately 80% of gene expression-based molecular subtypes among 74 MIBC cases. Based on these results, Guo et al. suggested to implement this two-marker panel as a simple immuno-histochemical classifier in routine diagnostic settings [27]. Our results indicate that the accuracy of predicting gene expression-based subtypes can be slightly increased by adding other subtype-specific markers; however, a group of discordantly classified tumors remains. Upcoming studies are necessary to explore if discordant classifications are clinically relevant. Identified subtypes among this study are more general and the consensus classification was not applied.

Gene expression based cluster
Conventional histopathological assessment is still the most important prognostic "marker" in MIBC [2]. Previous studies indicated a negative prognostic effect of variant histology in MIBC compared to NOS morphology [28]. In our present study, histological subtypes of MIBC were not independent predictors for DSS or RFS, which can be attributed in part to a smaller study cohort. Interestingly, histological subtypes of MIBC are strong indicators for specific mRNA and protein subtypes, e.g., squamous differentiation as indicator for basal subtypes, which has been indicated in previous studies [5,14]. In contrast, NOS morphology and variant histology (especially micropapillary, nested and plasmacytoid morphology) strongly correlated with luminal phenotypes, which has only been indicated in small previously published case series [20,21,29].
Limitations of our study are the retrospective character and the relatively low number of analyzed patients. Due to the lack of NAC, we were not able to analyze the possible predictive value of these molecular subtypes, which could be of clinical relevance. The prevalence of variant histology in our cohort is higher than in previously reported cohorts. This might be caused by a lack of neoadjuvant treatment (NAC) with well-preserved residual tumors after TURB, by full embedding of the complete tumor mass according to standardized mapping methodologies applied in our institution [18,19], and potentially due to a higher awareness of variant histology based on our primary research interest in bladder pathology.

Conclusion
Matching with observations of the Lund group, we demonstrate that neither gene expression nor protein-based subtyping associates with patient outcomes in a population-based cystectomy cohort. Both subtyping strategies show substantial overlaps, thus indicating that protein-based subtyping could be used for MIBC subtyping in daily practice as a cost-efficient alternative to transcriptome sequencing. In addition, we provide evidence that histological subtypes correlate strongly with mRNA and protein molecular subtypes.