Differential NEUROD1, ASCL1, and POU2F3 Expression Defines Molecular Subsets of Bladder Small Cell/Neuroendocrine Carcinoma with Prognostic Implications

Abstract

Small cell carcinomas (SMC) of the lung are now molecularly classified based on the expression of transcriptional regulators (NEUROD1, ASCL1, POU2F3, YAP1) and DLL3, which has emerged as an investigational therapeutic target. PLCG2 has been shown to identify a distinct subpopulation of lung SMC with stem cell-like and pro-metastasis features and poor prognosis. We analyzed the expression of these novel neuroendocrine markers and their association with traditional neuroendocrine markers and patient outcomes in a cohort of bladder neuroendocrine carcinoma (NEC) consisting of 103 SMC and 19 large cell neuroendocrine carcinomas (LCNEC) assembled in tissue microarrays. Co-expression patterns were assessed and integrated with detailed clinical annotation including overall (OS) and recurrence free survival (RFS) and response to neoadjuvant/adjuvant chemotherapy. We identified five distinct molecular subtypes in bladder SMC based on expression of ASCL1, NEUROD1 and POU2F3: ASCL1+/NEUROD1− (n=33; 34%), ASCL1−/NEUROD1+ (n=21; 21%), ASCL1+/NEUROD1+ (n=17; 17%), POU2F3+ (n=22, 22%), and ASCL1−/NEUROD1−/POU2F3− (n=5, 5%). POU2F3+ tumors were mutually exclusive with those expressing ASCL1 and NEUROD1 and exhibited lower expression of traditional neuroendocrine markers. PLCG2 expression was noted in 33 tumors (32%) and was highly correlated with POU2F3 expression (p < 0.001). DLL3 expression was high in both SMC (n=72, 82%) and LCNEC (n=11, 85%). YAP1 expression was enriched in non- neuroendocrine components and negatively correlated with all neuroendocrine markers. In patients without metastatic disease who underwent radical cystectomy, PLCG2+ or POU2F3+ tumors had shorter RFS and OS (p<0.05), but their expression was not associated with metastasis status or response to neoadjuvant/adjuvant chemotherapy. In conclusion, NEC of the bladder can be divided into distinct molecular subtypes based on the expression of ASCL1, NEUROD1 and POU2F3. POU2F3 expressing tumors represent an ASCL1/NEUROD1-negative subset of bladder NEC characterized by lower expression of traditional neuroendocrine markers. Marker expression patterns were similar in SMC and LCNEC. Expression of PLCG2 and POU2F3 was associated with shorter recurrence-free and overall survival. DLL3 was expressed at high levels in both SMC and LCNEC of the bladder, nominating it as a potential therapeutic target.

Introduction

Small cell carcinoma (SMC) of the bladder is a rare and aggressive histologically defined bladder cancer subtype. While <1% of bladder tumors exhibit neuroendocrine features, ¹ the bladder represents one of the most common sites of extrapulmonary small cell carcinoma.² Bladder small cell carcinoma is typically advanced at the time of diagnosis, but the outcome of this aggressive histologic subtype has improved significantly with the adoption of chemotherapy in the neoadjuvant or adjuvant setting.³ Furthermore, outcomes remain poor for patients with metastatic bladder small cell carcinoma despite the aggressive use of systemic chemotherapy ^3,4, and more recently, immunotherapy.⁵

Traditional immunohistochemical markers of neuroendocrine differentiation such as synaptophysin, chromogranin, CD56, INSM1 (insulinoma-associated protein 1) are routinely used to confirm the morphologic impression of a neuroendocrine phenotype inferred from H&E staining. However, despite their diagnostic value, these traditional neuroendocrine markers do not provide prognostic or predictive information due to their ubiquitous expression in neuroendocrine carcinoma. Although tumor morphology is similar to its small cell lung carcinoma counterpart, recent studies suggest that neuroendocrine carcinoma of the bladder have a distinct molecular pathogenesis and that most arise from a urothelial carcinoma precursor, with >50% exhibiting areas of both urothelial and neuroendocrine differentiation.⁶

Novel neuroendocrine markers, including ASCL1 (Achaete-scute homolog 1), NEUROD1 (neurogenic differentiation 1) and POU2F3 (POU domain, class 2, transcription factor 3), were recently identified in preclinical models and by immunohistochemical analysis of a large cohort of lung cancers.^7,8 The differential expression of these transcription regulators define distinct molecular subtypes of lung small cell carcinoma and highlight the transcriptional heterogeneity in this seemingly morphologically homogenous disease. Similar results were recently reported in a cohort of bladder small cell carcinoma based on RNA expression profiling, but details on immunohistochemical validation and co-expression patterns were limited.⁹ ASCL1 is a neuroendocrine lineage master regulator that was identified in fetal lung neuroendocrine cells and is required for survival of human lung small cell carcinoma cell lines.^10,11 NEUROD1 contributes to regulation of neuroendocrine cell development, maturation and function of several neural/neuroendocrine tissues, fate of neurons in the central and peripheral nervous system and for insulin gene transcription in adult pancreatic beta cells.^11,12 POU2F3 is another transcription factor that is normally expressed in tuft cells, where it acts as chemosensory regulator in normal lung. Notably, POU2F3 expression has been identified in cell lines with low ASCL1 or NEUROD1 expression levels.^13,14 POU2F3 expression was associated with lower expression of traditional neuroendocrine markers such as synaptophysin and chromogranin in both preclinical models and in human lung small cell carcinoma.^8,15,16 YAP1 (Yes-associated protein 1) has been identified in a subset of lung small cell carcinoma, with a possible association with survival.^7,17 While these markers provide important insights into the heterogenous biology of SMC, their potential association with disease progression or response to chemotherapy has not been fully evaluated.

The aim of the current study was to define the prevalence and pattern of expression of novel neuroendocrine markers in small and large cell neuroendocrine carcinomas of the bladder and their coassociation with traditional markers of neuroendocrine differentiation. Our major translational focus was to identify prognostic markers of cancer-specific outcome and predictive markers of chemotherapy response. Moreover, we evaluated the expression of PLCG2 (Phospholipase C Gamma 2), which was recently detected as the top upregulated gene in an unsupervised clustering of lung small cell carcinoma and was found to be associated with pro-metastatic and stem-cell like features.¹⁸, and DLL3 (delta like canonical Notch ligand 3), a notch ligand expressed exclusively on neuroendocrine cells, that is currently being investigated as a target for novel antibody drug conjugates.¹⁹

Materials and Methods

Patient cohort.

This study was approved by the MSKCC institutional review board, and all patients signed informed consent. Tumors from patients with a diagnosis of small cell or large cell neuroendocrine carcinoma of the bladder diagnosed at MSK between 1995-2021 were analyzed. All H&E and available immunohistochemical (IHC) slides were reviewed to confirm the diagnosis and annotate tumor characteristics including the type and amount of neuroendocrine component and the presence of any other non-neuroendocrine histology. Representative tumor blocks of each case were utilized for the construction of tissue microarrays (TMA), using a Manual Tissue Microarray machine (Beecher, Budapest, Hungary). From each tumor, 1-mm cores from two different areas of the NE component and any non-NE areas, when present, were sampled. From these tumors, a total of 326 cores were assembled into three TMA blocks. The TMAs included 103 small cell carcinoma (SMC) and 19 large cell neuroendocrine carcinoma (LCNEC) tumors. Pure neuroendocrine carcinoma was present in 71 tumors. Of the tumors with mixed histology, the neuroendocrine component comprised >90% (n=16), >50% (n=23) or <50% (n=12) of the tumors. For 25 of the mixed histology tumors the accompanying non-neuroendocrine components was included in the tissue microarrays. To further characterize the extent and pattern of expression of the markers within individual tumors, whole slides were stained for 12 small cell carcinoma and 5 LCNEC cases.

The following novel neuroendocrine markers were analyzed: ASCL1 (Mo/24B72D11.1, BD Bioscience, 1/100), NEUROD1 (RbMo/EPR17084, Abcam, 1/100), POU2F3 (MoMo/6D1, Santa Cruz, 1/1000), DLL3 (RM/SP347, Ventana, ready to use), PLCG2 (RbMo/E5U4T, Cell Signaling) and YAP1 (MM/63.7, Santa Cruz, 1/1000). For comparison, the traditional neuroendocrine markers chromogranin, synaptophysin, CD56, and INSM1 were also analyzed,⁸ as were p53 and Rb1 expression. The total number of cases with at least one core yielding interpretable results for each marker and the details of IHC staining are summarized in Supplementary Table 1.

For the novel neuroendocrine markers, both staining percentage (1-100%) and intensity (1: weak, 2: moderate, 3: strong) were recorded and an H-score was generated by multiplying the percentage of cell staining by intensity score. For traditional neuroendocrine markers, only the staining percentage was recorded. NE markers were dichotomized and analyzed as binary variables (−, no expression; +, any expression), as well as continuous values on the log scale. Only nuclear staining was considered positive for ASCL1, NEUROD1 and POU2F3, whereas nuclear and cytoplasmic staining for YAP1, cytoplasmic/membranous staining for DLL3 and membranous staining for PLCG2 was considered positive.

Statistical Analysis.

Biomarker values between cores were compared in cases with 2 interpretable cores; high correlation was found (0.99 IQR: [0.97, 1.00]) therefore mean of marker values across cores was calculated and used in the analysis. To leverage additional patients who only had whole slide-stained samples available that were not included in the TMAs, distributions of markers were compared between the TMA samples and the 17 whole slide-stained samples. No significant differences were observed; hence 17 unique whole slide-stained samples were included in the final cohort for statistical analysis.

Dichotomized presence/absence of marker expression was compared between clinical and genomic groups using Pearson's Chi-squared test or Fisher's exact test when observed cell counts < 5. H-scores, percent and mean intensity marker values were compared between groups using Wilcoxon rank sum tests. Associations between binary markers were analyzed using Phi coefficients (ϕ), which are equivalent to Pearson’s correlation for dichotomous data. These coefficients quantify the strength of the association, ranging from −1 for a perfect negative relationship to +1 for a perfect positive relationship. Spearman’s rank correlations (ρ; range from −1 to 1) were calculated to analyze associations between continuous marker values. To examine novel NE-specific markers, 25 unique samples with both NE and non-NE components were compared using paired Wilcoxon signed rank tests with continuity correction for continuous variables and McNemar’s test for binary variables.

For analysis of clinical endpoints including overall survival, progression-free survival and recurrence-free survival, the cohort was split into two distinct sub-cohorts: one cohort included patients treated with radical cystectomy (RC, n=72), and the other included those who did not undergo radical cystectomy (non-RC, n=44). In the RC cohort, 6 patients with metastatic disease at the time of cystectomy were omitted (n=6).

In the RC cohort, overall survival (OS) was defined as the time between cystectomy or start of chemotherapy/radiotherapy and death and recurrence-free survival (RFS) was defined as the time between cystectomy and disease recurrence or death.

For the non-RC cohort, OS was calculated from the start of systemic chemotherapy until death, and progression-free survival was defined as time between systemic therapy and clinical evidence of disease progression or death. Ten non-RC patients who did not receive chemotherapy due to comorbidities were excluded from RFS/OS analysis, to keep the group homogenous for treatment modalities. For all endpoints, patients who did not experience an event were censored at last follow-up.

Time to event endpoints were summarized using Kaplan-Meier methods and RFS and OS differences were assessed between marker groups using log rank tests. Associations between presence of markers and clinical endpoints were assessed using cox proportional hazards models adjusting for known prognostic characteristics including stage and treatment status.

Next generation sequencing.

Targeted next generation sequencing of up to 505 cancer-associated genes was performed using the MSK-IMPACT assay to identify somatic mutations, copy number alterations and structural variants. Detailed methods on sample preparation, sequencing and data analysis have been described previously.^20,21 Somatic mutations were classified as oncogenic and likely oncogenic using the OncoKB knowledgebase.²² From prospectively profiled bladder cancer samples, tumors diagnosed as SMC or LCNEC were included in this analysis. For statistical portion of genomics analysis, we only considered genes with prevalence above 10%.

Results

Patient and sample characteristics

Among 122 unique patients, 103 were classified based on histologic review of H&E slides and expression of neuroendocrine immunohistochemical markers as small cell carcinoma (SMC) and 19 as large cell neuroendocrine cancer (LCNEC). Morphologic features of LCNEC were those utilized in lung LCNEC and consist of large high-grade polygonal tumor cells with variable amounts of cytoplasm and prominent nucleoli. They have variable architectural patterns of neuroendocrine morphology such as nests, trabeculae, organoid, and peripheral palisading.^23,24 Detailed patient characteristics are summarized in Supplementary Table 2. The cohort was predominantly male (80%), consistent with greater prevalence of bladder cancer in male versus female patients. Median age was 70 years. 71 (57%) tumors were derived from biopsies/transurethral resection (TUR), 48 (40%) from radical or partial cystectomies, and 3 (3%) from biopsies of metastatic lesions. In 25 (20.5%) tumors with mixed histology, urothelial carcinoma, NOS, squamous, plasmacytoid, micropapillary, glandular and sarcomatoid subtypes co-existed with the neuroendocrine components. Seventy-two of the 122 patients underwent radical cystectomy and a median follow up of 2.1 years from time from resection (IQR: 10 months - 7.7 years). Among 44 patients who did not undergo radical cystectomy, median follow-up from the start of systemic chemotherapy was 1.3 years (IQR: 8 months – 2.7 years).

Expression characteristics of neuroendocrine markers

ASCL1, NEUROD1 and POU2F3.

For the bladder SMC tumors, any expression of ASCL1, NEUROD1 and POU2F3 was detected in 51% (50/98), 38% (38/98) and 22% (n=22/98) of tumors, respectively. Examples of tumors expressing ASCL1, NEUROD1 and POU2F3 are shown in Figure 1A. For LCNEC tumors, ASCL1, NEUROD1 and POU2F3 were expressed in 44% (8/18), 44% (8/18), and 11% (2/18), respectively. Overall, we did not observe a significant difference in the expression of ASCL1, NEUROD1 and POU2F3 between SMC and LCNEC groups (Table 1). In the overall cohort, the mean H-score for POU2F3 was higher compared to ASCL1 and NEUROD1, although the difference was not statistically significant (278 vs 150 and 110, respectively). Paired analyses comparing non-neuroendocrine components to neuroendocrine components in the overall cohort showed significant differences in the expression of all novel markers with higher expression in the neuroendocrine components of mixed histology tumors (p<0.01 for all markers, Supplementary Table 3).

Figure 1.

A. Representative examples of ASCL1+, NEUROD1+ and POU2F3+ immunohistochemistry of neuroendocrine bladder cancers. Note the overall non-overlapping expression pattern of these markers. POU2F3+ tumors typically had low or no expression of traditional neuroendocrine markers such as synaptophysin. DLL3 was expressed in ASCL1+, and to a lesser extent in NEUROD1+ bladder tumors but was typically negative in POU2F3+ tumors. B. Distribution of markers expression sorted by order of ASCL1, NEUROD1 and POU2F3 binary expression. Each column represents one tumor sample and each raw corresponds to an immunohistochemical marker. For p53, positive indicates strong and diffuse expression (mutated pattern) and negative indicates complete loss of expression (null pattern). For Rb, negative indicates loss of expression and positive indicates retained expression. C. Correlation plot of novel and traditional markers. Points are scaled by the magnitude of correlation coefficients and colored by direction of the correlation. This plot shows Spearman Correlation coefficient values as calculated pairwise between continuous percent values each of marker.

Table 1:

Prevalence of Novel Markers Expression

Characteristic	N	Overall, N= 122	LCNEC, N = 19	SMC, N = 103*	p-value**
ASCL1/NEUROD1/POU2F3	116				0.5
ASCL1+		39 (34%)	6 (33%)	33 (34%)
ASCL1+/NEUROD1+		19 (16%)	2 (11%)	17 (17%)
NEUROD1+		27 (23%)	6 (33%)	21 (21%)
POU2F3+		24 (21%)	2 (11%)	22 (22%)
None		7 (6.0%)	2 (11%)	5 (5.1%)
Unknown***		6	1	5
ASCL1	117	59 (50%)	8 (44%)	51 (52%)	0.6
Unknown		5	1	4
H-score Median (IQR) [Range]	59	150 (50, 190) [3, 300]	45 (28, 188) [10, 300]	160 (70, 190) [3, 300]	0.4
NEUROD1	117	48 (41%)	8 (44%)	40 (40%)	0.8
Unknown		5	1	4
H-score Median (IQR) [Range]	48	110 (20, 248) [1, 300]	85 (23, 218) [1, 270]	135 (20, 248) [1, 300]	0.5
POU2F3	117	28 (24%)	2 (11%)	26 (26%)	0.2
Unknown		5	1	4
H-score Positive Median (IQR) [Range]	28	278 (190, 300) [2, 300]	245 (218, 272) [190, 300]	278 (190, 300) [2, 300]	0.8
PLCG2	102	33 (32%)	7 (54%)	26 (29%)	0.11
Unknown		20	6	14
H-score Median (IQR) [Range]	33	100 (50, 170) [5, 255]	82 (65, 92) [30, 180]	130 (50, 178) [5, 255]	0.5
DLL3	101	83 (82%)	11 (85%)	72 (82%)	>0.9
Unknown		21	6	15
H-score Median (IQR) [Range]	83	110 (15, 180) [1, 285]	100 (55, 110) [1, 180]	110 (10, 180) [1, 285]	0.7
YAP1	114	35 (31%)	9 (50%)	26 (27%)	0.092
Unknown		8	1	7
H-score Median (IQR) [Range]	35	180 (100, 255) [10, 300]	180 (170, 200) [20, 285]	175 (92, 270) [10, 300]	>0.9

^*n (%)

^**Fisher's exact test

^***Cores that are not suitable for interpretation were excluded (tumor exhaustion, tissue fell off from section, etc.)

By analyzing co-expression patterns of ASCL1, NEUROD1 and POU2F3 in the 116 tumors for which there was available data for all 3 markers, the following five subgroups were identified: ASCL1+ (34% (39/116), NEUROD1+ (23%, 27/116), ASCL1+/NEUROD1+ (16%, 19/116), POU2F3+ (21%, 24/116) and ASCL1−/NEUROD1−/POU2F3− (6%, 7/116) (Figure 1B, Supplementary Figures 1 & 2). Unsupervised consensus clustering based on marker expression identified 4 distinct tumor groups (Supplementary Figure 3A). Cluster 1 is enriched for ASCL1+ alone or ASCL1+ plus NEUROD1+ tumors, cluster 2 is enriched for NEUROD1+ tumors (with no ASCL1 expression), cluster 3 is enriched for POU2F3+ tumors, and cluster 4 is negative for all 3 novel markers. Of note, all the ASCL1−/NEUROD1−/POU2F3− cases were positive for at least one of the traditional neuroendocrine markers as shown in Figure 1B and Supplementary Figures 1 & 2. Similar patterns of co-expression and mutual exclusivity were observed between SMC and LCNEC cohorts when analyzed separately.

Among the 19 ASCL1+/NEUROD1+ cases, 10 expressed NEUROD1 and 9 expressed ASCL1 predominantly. Tumors with POU2F3 expression showed significant mutual exclusivity with tumors that expressed NEUROD1 and ASCL1 (Fisher’s exact test, p<0.001 for both, Figure 1C, Supplementary Figure 3B). We also observed a moderate negative correlation with the extent (%) of expression of NEUROD1 (Spearman’s rank correlation = −0.33, p < 0.001) and ASCL1 (Spearman’s rank correlation = −0.42, p < 0.001).

ASCL1 was positively correlated with all four traditional neuroendocrine markers (Figure 1C, Supplementary Figure 1B), with CD56 and synaptophysin consistently showing higher levels of expression across the tumors, whereas expression of chromogranin, INSM1 and DLL3 was more variable (Supplementary Figure 4). Conversely, POU2F3+ tumors showed weaker and less extensive expression of traditional neuroendocrine markers and a negative correlation between POU2F3 and chromogranin, synaptophysin and INSM1 expression was observed (Figure 1C, Supplementary Figures 3B & 4). The only exception was CD56 which was positive in almost all the neuroendocrine tumors, including those expressing POU2F3 (Figure 1B).

To assess for heterogeneity of ASCL1, NEUROD1 and POU2F3 expression across larger tumor areas (compared to the smaller TMA cores), we performed whole slide staining of 16 tumors. Similar patterns of expression to the TMA sections were observed in 14 of the 16 tumors. The remining two tumors showed distinct, non-overlapping subpopulations of POU2F3+ and ASCL1+ tumor cells (Figure 2A), supporting the mutual exclusive expression patterns of POU2F3 and ASCL1 observed in the TMA, within a subset of individual neuroendocrine bladder cancers. Of note, there were no morphologic differences between the POU2F3 and ASCL1 positive subpopulations on H&E evaluations and moreover, the traditional neuroendocrine markers were stronger and more diffuse in the ASCL1+ compared to the POU2F3+ tumor areas. Finally, there was no significant difference in the expression of the tested markers between SMC and LCNEC except for CD56, which was expressed in higher percentages in SMC (p=0.019).

Figure 2.

A. Whole slide staining of a tumor with ASCL1+ and POU2F3+ subpopulations, showing mutually exclusive expression patterns. Note synaptophysin expression in ASCL1+ tumor region and lack of expression in POU2F3+ areas. B. An example of small cell carcinoma with POU2F3 and PLCG2 co-expression. Note the low chromogranin expression. This tumor is additionally negative for ASCL1, NEUROD1 and DLL3. C. Oncoprint of frequently mutated genes, sorted by expression of novel endocrine markers ASCL1, NEUROD1 and POU2F3. Top panel is number of mutations per tumor among selected genes and right panel is gene alteration prevalence in the entire cohort. Bottom panel shows expression patterns of novel and traditional neuroendocrine markers. Results or p53, RB, PLCG2 and DLL3 expression are also shown. For p53, positive indicates strong and diffuse expression (mutated pattern) and negative indicates complete loss of expression (null pattern). For Rb, negative indicates loss of expression and positive indicates retained expression.

Of note, there was no specific microscopic H&E features that could reliably distinguish between these groups or predict the expression of any of ASCL1, NEUROD1 or POU2F3, suggesting that these groups are currently strictly molecularly defined.

PLCG2 and DLL3.

PLCG2 was expressed at any level in 29% (26/89) of SMC and 54% (7/13) of LCNEC cases that were evaluable for these markers. Expression of PLCG2 showed a strong positive correlation with POU2F3 expression (spearman correlation: 0.76, p < 0.001) (Figure 2B, Supplementary Figure 4).

DLL3 expression was high in SMC and LCNEC tumors, with 82% (72/88) and 85% (11/13) expressing DLL3, respectively. DLL3 expression was positively correlated with ASCL1, synaptophysin, chromogranin and INSM1, and negatively associated with POU2F3 (p<0.001, p=0.008, p<0.001, p=0.008, p=0.027, respectively). (Figure 1C, Supplementary Figures 3B & 4).

YAP1.

YAP1 expression with any intensity was detected in 27% (26/96) of SMC and 50% (9/18) of LCNEC, respectively. YAP1 expression was detected in 96% (24/25) of the non-neuroendocrine components that were evaluated, and the mean H-score was higher compared to neuroendocrine regions of mixed histology tumors, with the staining stronger and more diffuse (p<0.001). YAP1-expressing SMC tumors were those mixed with a non-neuroendocrine component in 75% of cases and the mean H-score was higher compared to the pure neuroendocrine tumors. YAP1 expression was negatively correlated with both traditional and novel neuroendocrine markers (Figure 1C, Supplementary Figure 3B).

Next generation sequencing and immunohistochemistry for p53 and Rb

Thirty-three tumors were analyzed by next generation sequencing (NGS) using the MSK-IMPACT targeted capture assay. Mutations in TERT promoter, TP53 and RB1 were identified in 27 (82%), 30 (91%) and 26 tumors (79%), respectively. Of the 30 TP53 mutated tumors, IHC results were available for only 27 tumors, 25 of which showed a mutated/abnormal expression pattern. Of the 26 RB1 mutated tumors, 23 showed Rb loss of expression by IHC. Immunohistochemical results on the overall cohort corroborated these findings as 93% (101/108) of neuroendocrine tumors showed evidence of abnormal p53 staining pattern (either diffuse strong expression in 67 tumors or 62%, or complete loss/null in 34 tumors or 31%), and 92% (103/112 tumors) showed loss of Rb expression. In a paired analysis comparing neuroendocrine and non-neuroendocrine components within the same tumor, p53 and Rb staining patterns did not significantly differ.

Oncogenic alteration frequencies were separately compared between ASCL1, NEUROD1 and POU2F3 positive and negative groups and summarized in the Figure 2C. RB1 and TP53 mutation frequencies did not show significant differences between marker groups. ASCL1+ tumors harbored more MCL1 Amplifications (p=0.039), though differences in alteration frequencies were not significantly different after adjusting for multiple testing (p=0.8). Both POU2F3+ and PLCG2+ tumors showed more FBXW7 mutations compared to those that were negative for these markers, although the difference was not significant. Of potential therapeutic importance, oncogenic mutations in ERCC2 were enriched in POU2F3+ tumors, although the difference in prevalence was not significant. Out of 9 POU2F3+ tumors in which MSK-IMPACT analysis was performed, 3 had ERCC2 mutation (33%), compared to only 2 ERCC2 mutations detected in 24 POU2F3− cases (8.3%) (p=0.11; q > 0.9). Similar enrichment was seen in PLCG2+ cases as 2 of 6 POU2F3+ tumors harbored ERCC2 mutations (33%), whereas none of the 17 PLCG2− tumors harbored ERCC2 mutations (p = 0.059; q > 0.9).

Association of novel markers with outcome and response to neoadjuvant chemotherapy in RC patients

In the RC cohort (n=72), both RFS and OS differed significantly between the POU2F3 positive and negative groups (p=0.03; p=0.011, respectively) with a 2-year RFS of 40% (95% CI 22% - 75%) in the positive group compared to 63% (95% CI 50% - 79%) in the negative group. In Cox models examining association between RFS and marker H-scores (analyzed on the log scale), the difference was marginally non-significant (0.055) after adjusting for neoadjuvant treatment but remained significant in OS (p=0.010). Similarly, RFS and OS significantly differed between the PLCG2 positive and negative groups (p=0.0018; p=0.003, respectively) and the association remained significant in Cox models examining PLCG2 H-score and adjusting for neoadjuvant treatment (p=0.005) (Table 2, Figure 3).

Table 2:

Cox models for association of novel markers expression and recurrence free survival in radical cystectomy cohort

	Univariate				Adjusted For Neoadjuvant Treatment
Marker expression (H-score, log scale)	N	HR*	95% CI*	p-value	HR*	95% CI*	p-value
PLCG2	62	1.25	1.08, 1.44	0.002	1.24	1.07, 1.44	0.005
POU2F3	69	1.16	1.02, 1.32	0.026	1.14	1.00, 1.30	0.055
ASCL1	69	0.88	0.77, 1.01	0.064	0.87	0.76, 0.99	0.039
DLL3	61	0.92	0.81, 1.05	0.2	0.9	0.78, 1.03	0.11
NEUROD1	69	1.03	0.92, 1.17	0.6	1.08	0.95, 1.22	0.3
YAP1	67	1.02	0.91, 1.14	0.8	1.04	0.93, 1.17	0.5

^*HR = Hazard Ratio, CI = Confidence Interval

Figure 3.

Overall and Progression-free survival in POU2F3+ (A) and PLCG2+ (B) groups.

RFS and OS were higher in ASCL1+ groups though the differences were not significant (2-year RFS for ASCL+ was 50% [34%, 74%], for ASCL− was 36% [23%, 57%]; log rank p = 0.19) (2-year OS for ASCL+ was 67% [51%, 89%], for ASCL− was 50% [35%, 70%]; log rank p = 0.21). In Cox models adjusted for neoadjuvant treatment, higher ASCL1 H-score (analyzed on the log scale) was associated with a better RFS (HR: 0.87; 95% CI: 0.76- 0.99; Table 2). ASCL1 and DLL3 expression was significantly higher in M1 patients, compared to the M0 group (p=0.007, p=0.001, respectively). Finally, among patients treated with RC, there was no significant association between the expression of any of the markers and pathologic response. Of note, there was no significant difference in the expression of any markers between those receiving neoadjuvant treatment and not, suggesting that the expression of these markers was not impacted by chemotherapy treatment. Furthermore, no associations were observed between the expression of the tested markers and outcome in non-RC (n=44) patients, which represent a more heterogeneous group with various treatment histories.

Discussion

In this study, we evaluated the expression of recently identified novel neuroendocrine markers in a large cohort of bladder neuroendocrine carcinomas and correlated their expression profiles with traditional neuroendocrine markers and clinical outcome. We showed that small cell carcinomas of the bladder can be divided into distinct subgroups based on their expression of ASCL1, NEUROD1 and POU2F3, with similar results in large cell neuroendocrine carcinomas. Similar to recently reported findings in small cell lung cancers ^8,25, the largest subset was ASCL1+ (34%), followed by NEUROD1+ (23%), POU2F3+ (21%), and negative for all 3 markers (6%). Similar findings have been recently reported in bladder small cell carcinoma with three molecular subtypes defined by these lineage-specific transcription factors ASCL1, NEUROD1, and POU2F3.⁹ Except for a subset with ASCL1 and NEUROD1 co-expression, which represented 16% of cases, these clusters were largely non-overlapping. This non-overlapping pattern appears to be unique to human neuroendocrine tumors, as it was not reported in preclinical models of single cell clones, which suggested more homogenous subgroups ^{7,11,13,26-28}.

Moreover, our study identified a subset of tumors that expresses POU2F3 that is largely negative for the traditional NE markers synaptophysin, chromogranin, INSM1 and to a lesser extent CD56. This POU2F3 expressing subset of neuroendocrine bladder tumors would have been missed or difficult to diagnose due to absent or reduced expression of traditional neuroendocrine markers if such studies were to be performed to confirm the H&E impression of high-grade neuroendocrine carcinoma. The rate of POU2F3 expression in bladder SMC in our study (21%) is slightly higher than what has been reported in lung SMC at 12%.^8,16

Our results confirm that these markers are specific to the neuroendocrine component, small or large cell, as all novel markers tested were exclusively present in the neuroendocrine components of all of cases of mixed histology tumors for which we were able to analyze a non-neuroendocrine component. These findings further support recent reports implicating lineage plasticity from urothelial to neural phenotypes in bladder small cell carcinoma. Although the exact mechanisms underlying such lineage switch have not been fully elucidated, development of bladder small cell carcinoma involves suppression of urothelial phenotype and activation of neural regulons, contributing to its uniformly double negative phenotype for both luminal and basal markers.^29,30 By correlating the patterns of expression of these novel markers individually and with each other, we also found that POU2F3 expression is mutually exclusive of ASCL1 and NEUROD1, which is consistent with recent reports in lung SMC ^{13,15,16,31,32}. In further support of this observation, we were able to demonstrate that expression of POU2F3 is mutually exclusive of that of ASCL1 within the same tumor when we evaluated whole tissue sections. In two such examples, there were two different tumor subpopulations exclusively expressing either ASCL1/NEUROD1 or POU2F3, without overlap.

YAP1 expression was suggested to represent another subgroup of lung SMC in preclinical models ⁷, but subsequent studies did not confirm such an observation in human tumor samples.^8,28 In the current study, the majority of YAP1 expressing neuroendocrine tumors were of mixed histology and co-existed with non-neuroendocrine components, in which the staining intensity and percentages were lower in the neuroendocrine component. YAP1 did not show a correlation with any of the novel neuroendocrine markers and YAP1-positive neuroendocrine tumors showed an overall low neuroendocrine profile. These observations suggest that YAP1-expressing tumors maybe in transition to a full neuroendocrine phenotype where YAP1 is downregulated and neuroendocrine markers are upregulated. Supporting these observations are recent reports that YAP1 loss promotes neuroendocrine differentiation in lung and prostate cancers.^33-35

In addition to their role in understanding the biology and their potential contribution to the diagnosis of neuroendocrine carcinomas, ASCL1, NEUROD1 and POU2F3 were reported to offer different therapeutic vulnerabilities in lung small cell carcinoma.^15,36 In our study, these novel markers provide additional prognostic values, particularly POU2F3, whose expression is associated with a worse RFS and OS. In contrast, ASCL1 expression was associated with better RFS. Interestingly, PLCG2, which was recently identified as one of the top altered genes in a subgroup of lung small cell carcinoma associated with pro-metastatic and stem cell-like features¹⁸, strongly correlated with POU2F3 expression and showed a similar low neuroendocrine profile. More importantly, PLCG2 expression in our study was associated with worse OS, regardless of the extent of its expression.

Despite the significant improvement in the outcome in bladder small cell carcinoma following administration of cisplatin-based chemotherapy either in the neoadjuvant or adjuvant setting, the disease is still associated with significant morbidity particularly in patients with chemorefractory metastatic disease.^3,4 Furthermore, the use of anti-PD1/PD-L1 immune check point inhibitors did not result in improvement of outcome for patients with locally advanced or metastatic bladder small cell carcinoma in a recent phase II study of non-urothelial carcinomas of the urinary tract treated with durvalumab and tremelimumab.⁵ DLL3 is an inhibitory Notch ligand which has been shown to be expressed specifically on neuroendocrine cells, including bladder small cell carcinomas.^37-45 In our study, DLL3 expression was found at high levels in the vast majority of SMC and LCNEC tumors (82% and 85%, respectively). Preclinical models and recent clinical trials of targeting DLL3 by antibody-drug conjugate or bispecific T-cell engager approaches showed promising results in patients with recurrent lung small cell carcinoma.^19,46-51 The high DLL3 expression of bladder small and large cell neuroendocrine carcinomas in our study suggests that it is also an attractive therapeutic target for this rare but aggressive subset of bladder cancers. It is suggested that DLL3 expression is regulated by ASCL1, which promotes NE differentiation in coordination with Notch signaling.^52,53 In support of this observation, our study showed a strong correlation between DLL3 expression and that of ASCL1 and traditional NE markers. Of potential therapeutic relevance, DLL3 expression was limited to the NE component of mixed histology tumors.

Finally, and consistent with prior reports⁶, results of our mutational profiling analysis showed high rates of mutations in TERT promoter, RB1 and TP53 in all of the ASCL1, NEUROD1 and POU2F3 defined subgroups, including the POU2F3+ tumors. ERCC2 mutations were more frequent in POU2F3+ and PLCG2+ groups when compared with the other groups, which may have therapeutic implications given the association between ERCC2 mutational status and cisplatin response in prior studies.^54,55 Finally, we did not detect increased rates of mutations in NOTCH1-4, MYC amplification, amplifications of the 20q13 locus, or PTEN inactivation as was recently reported in lung SMC with POU2F3 expression.¹⁶

In sum, we report on the expression of novel neuroendocrine markers in a large cohort of bladder neuroendocrine carcinomas and show that ASCL1, NEUROD1 and POU2F3 define distinct molecular subgroups with potential prognostic value. More specifically, POU2F3+ and PLCG2+ tumors define a new molecular subset characterized by low expression of traditional neuroendocrine markers and a worse patient outcome. Finally, DLL3 is highly expressed in both small cell and large cell neuroendocrine carcinomas of the bladder and represents an attractive therapeutic target as novel antibody-drug conjugates or bispecific T-cell engagers are being tested in early-stage clinical trials.

Data Availability

The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request.