4578 Background: Erda, an oral pan-fibroblast growth factor receptor (FGFR) tyrosine kinase inhibitor is approved to treat adult pts with locally advanced or mUC with susceptible FGFR3alt, whose disease progressed on or after at least one line of prior systemic therapy. In superficial and surgically resectable UC, FGFRalt has been shown to be highly associated with luminal subtype characterized by expression of luminal markers, low expression of basal markers and with immune cold / poor immune infiltration. In a randomized open-label phase III THOR study (Cohort 2), erda showed improvement in objective response and progression-free survival (PFS) compared to pembro, with no significant improvement in overall survival (OS). Exploration of subtypes in a large cohort of mUC pts has not been done and so a molecular analysis was undertaken to further understand the clinical findings. Methods: All available tumors from pts enrolled in Cohort 2 ( FGFRalt positive; N = 201) and a subset of FGFR wildtype (N = 116) were used to perform whole transcriptome RNA sequencing, where 152 and 84, respectively, passed quality control. The consensus single-sample classifier was applied to the RNAseq data to determine molecular subtypes. Tumor subtypes was correlated with treatment response to erda or pembro, PFS and OS. Results: Molecular classification of tumors identified a significant proportion of luminal-papillary (LumP) subtype in the tumors harboring FGFRalt compared to FGFR WT (78.3% vs. 36.9%, p < 0.001) vs. other subtypes: basal/squamous (Ba/Sq; 11.2% vs. 31.0%), stroma-rich (6.6% vs. 10.7%), neuroendocrine-like (0% vs. 2.4%), luminal-unstable (3.3% vs. 17.9%) and luminal-unspecified (0.7% vs. 1.2%), respectively. Consistently, FGFRalt type showed differential association with subtypes: 3.2% of fusions and 14.1% of mutations were detected in Ba/Sq subtype while 74.2% and 79.3%, respectively, were detected in LumP. Clinical outcomes evaluated within LumP subset showed a significant improvement in ORR of erda-treated vs. pembro-treated pts (41.7 vs. 19.7%; p = 0.01), which was consistent with the intent to treat (ITT) population (40.0% vs. 21.6%). Numerical improvement were observed in PFS in the LumP subtype between erda vs. pembro (5.5 vs 2.7 months) compared with the ITT population (4.4 vs 2.7). However, the benefit of erda over pembro in the LumP subtype did not translate to OS (10.9 vs. 12.9 months), similar to the ITT population (10.9 vs. 11.1 months). Conclusions: A significant overlap was identified between genomic selection of FGFRalt and the LumP subtype. Clinical outcomes based on transcriptomic analyses were consistent with those observed in the ITT population. These findings warrant further understanding of LumP tumors and association to treatment response. Clinical trial information: NCT03390504 .