Table 1.
Clinical characteristics by cohort for samples used for classifier training and validation.
Fig 1.
Workflow for analysis of the UNC-Training dataset.
RNA from UNC-Training set samples (n = 52) were profiled by both RNAseq (RNA Access) and NanoString nCounter assay for the BASE47 probeset. The previously published Transcriptome BASE47 classifier was applied to the RNAseq data as well as the NanoString-derived BASE47 gene expression values. A NanoString BASE47 classifier was developed for NanoString nCounter expression values, tested by Monte Carlo cross validation, and independently validated on the UNC-Validation dataset.
Fig 2.
Development of a NanoString BASE47 classifier that performs on NanoString derived RNA expression values.
RNA from UNC-Training set samples (n = 52) were profiled by both RNAseq (RNA Access) and NanoString nCounter assay for the BASE47 probeset to generate RNAseq and NanoString expression matrices respectively. “True” biologic calls were assigned by applying the previously published Transcriptome BASE47 classifier to the RNAseq data. (A) Application of the Transcriptome BASE47 classifier to the NanoString expression data resulted in a 46% misclassification error. (B) Application of the newly derived NanoString BASE47 classifier to NanoString expression data resulted in only 13% misclassification error (relative to the “True” biologic subtype designations).
Fig 3.
Heatmap of BASE47 genes on samples supervised by correlation to basal centroid.
(A) Samples were supervised by correlation to the basal centroid (NanoString BASE47 classifier applied to NanoString expression) and clustered on BASE47 gene expression values derived by NanoString nCounter assay. Annotation of sex, race, smoking status, and T stage are indicated. (B) Log2 normalized RNA expression of canonical basal and luminal genes based on NanoString subtype demonstrate expected basal and luminal expression patterns.
Fig 4.
Validation of NanoString BASE47 classifier on UNC-Validation dataset.
(A) UNC-Validation samples were profiled by both RNAseq (RNA Access) and NanoString nCounter assay for the BASE47 probeset to generate RNAseq and NanoString expression matrices respectively. “True” biologic calls were assigned by applying the previously published Transcriptome BASE47 classifier to the RNAseq data. NanoString BASE47 calls were determined by application of the newly derived NanoString BASE47 classifier to NanoString expression data. (B) Application of the newly derived NanoString BASE47 classifier to NanoString expression data resulted in only 6.7% misclassification error (relative to the “True” biologic subtype designations). (C) The basal centroid correlation values derived from application of the Transcriptome BASE47 classifier to RNAseq data (X axis) and the NanoString BASE47 classifier applied to NanoString expression data (Y axis) were significantly correlated, R = 0.90, p < 0.001. (D) Histogram of frequency of correlation coefficients of BASE47 genes between basal centroid correlation values derived from application of the Transcriptome BASE47 classifier to RNAseq data and the NanoString BASE47 classifier applied to NanoString expression data. (E) Scatter plots of expression values of a representative luminal (UPK2) and basal (AHNAK2) BASE47 genes derived from NanoString and RNAseq.
Fig 5.
Black patients are not enriched in basal-like bladder cancer.
(A) Workflow of analysis of the samples used to examine the molecular subtype designations of bladder cancers in black patients. (B) Percentage of patients with each subtype by ethnicity in the UNC/JHU metadataset. (C) Percentage of patients with each subtype by ethnicity in the TCGA BLCA.