A new 50-gene molecular subtype classifier: An evaluation of subtype stability and association with response to neoadjuvant chemotherapy in muscle-invasive bladder cancer.
519 Background: The bladder cancer basal and luminal molecular subtypes have been associated with differential progression patterns and responses to neoadjuvant chemotherapy. However, they are typically identified by whole transcriptome expression profiling, which may be impractical for most academic centers. To make subtype classification more accessible, we developed a 50-gene basal and luminal subtype classifier and examined its performance in matched tumor specimens, including pretreatment biopsies from patients treated with neoadjuvant chemotherapy. Methods: We refined a 50-gene subtype classifier derived from a oneNN classifier containing >~2000 genes that we previously developed using unsupervised methods. We compared its accuracy against calls made using the original oneNN classifier, including 148 tumors from a neoadjuvant chemotherapy meta dataset. To test subtype stability, tumors in 2 different datasets (30 sets of matched primary and lymph node tumors and 43 sets of matched pre- and post-chemotherapy tumors) were assigned using 50-gene subtype classifier with Linear Discriminator Analysis (LDA) prediction algorithms. Results: In the NAC cohort, patients with tumors assigned to the basal subtype by the 50-gene classifier had better survival outcomes compared to the luminal tumors, consistent with the conclusion generated with the parent classifier (p<0.05). Basal subtype assignments were stable in 62.5 % of pairs, whereas luminal tumors displayed 100% stability. In matched pre- and post-chemotherapy tumors, basal tumors displayed 78% stability while luminal tumors showed 94% stability. Conclusions: Based on these preliminary data, it appears that basal tumors display higher plasticity than luminal tumors with these specific contexts. This plasticity may interfere with precise subtype predictions with tumors assigned to the basal subtype at biopsy.