Molecular Classification of Neuroendocrine Carcinoma of the Bladder

**Background
**Neuroendocrine carcinoma (NEC) of the bladder is a rare and aggressive cancer with limited treatment options. Recent studies reported molecular subtypes resembling small cell lung cancer (SCLC), specifically those associated with transcription factors (TFs) ASCL1, NEUROD1, and POU2F3. A subset of bladder NEC remains unclassified. ATOH1 was recently identified as a TF associated with a minority of SCLC tumors.

Methods
The Tempus Lens Platform (Tempus AI, Inc., Chicago, IL) was used to query the multimodal de- identified database and establish a cohort of 121 patients with bladder NEC. We analyzed the genomic and transcriptomic profiles using tempusverse and custom code. K-means consensus clustering of gene expression was used to identify molecular subtypes; subsequently, tumor genetic alterations (single-nucleotide variants [SNVs], indels, and copy number alterations [CNAs]) and gene expression profiles were compared across these subtypes. Fisher’s exact test was used to compare the frequency of events across categorical groups. limma-trend and empirical Bayes statistics were used for differential gene expression analysis. The Cox proportional hazards model and the Kaplan-Meier method were used for survival analysis. consensusMIBC was used for the consensus molecular classification.

**Results
**Within the cohort, 103 were classified as neuroendocrine (NE)-like using consensusMIBC and were further analyzed in this study. The median age at diagnosis was 71 years (interquartile range, 63–76), with a male predominance (77%). Using the four SCLC TFs (ASCL1, NEUROD1, POU2F3, and ATOH1; YAP1 was excluded as it was recently found to be expressed in the non-NE components of mixed bladder NECs), we identified a five-type system that classifies most samples: ASCL1+/NEUROD1- (n = 37; 36%), ASCL1-/NEUROD1+ (n = 23; 22%), ASCL1+/NEUROD1+ (n = 16; 16%), POU2F3+ (n = 23; 22%), and ATOH1+ (n = 4; 4%). Each subtype was associated with a distinct gene expression profile with representative markers: ASCL1, SOX2, and DLL3 (ASCL1+/NEUROD1-); NEUROD1, NEUROD4, and SSTR2 (ASCL1-/NEUROD1+); ETV1, NKX2-1, and NKX2-2 (ASCL1+/NEUROD1+); POU2F3, GFI1B, and POU2AF2 (POU2F3+); ATOH1, POU4F3, and USH2A (ATOH1+). POU2F3+ tumors expressed low NE markers and high PLCG2 and ERBB3; ATOH1+ tumors often expressed NE markers and tended to overexpress ALK. No statistically significant tumor SNVs, indels, or CNAs were differentially observed across the subtypes; TP53 (92%), TERT (82%), and RB1 (62%) were the most frequently altered genes in the entire cohort. Univariate analysis showed a worse real-time overall survival of ASCL1+/NEUROD1+ (hazard ratio [HR] = 3.1, p = 0.01), POU2F3+ (HR = 3.4, p = 0.001), and ATOH1+ (HR = 8.4, p = 0.0003), compared with ASCL1+/NEUROD1- tumors.

Conclusions
We report an effective, clinically relevant molecular classification system for bladder NECs, including the newly described ATOH1+ subtype.