As previously discussed, LG non-invasive, and HG invasive, tumors have distinct genomic alterations. In addition to genomic alterations, numerous group have reported on distinct gene expression patterns (54-60). In 2005, Blaveri et. al. reported that, using hierarchical clustering, muscle invasive and superficial tumors could be distinguished based on gene expression pattern. They then went on to identify a gene signature that can discriminate between superficial and muscle invasive disease, as well as signatures for prognosis and the classification of tumors as transitional cell carcinoma or squamous cell carcinoma (58). Similar approaches used by Sanchez-Carbayo et. al. resulted in gene profiles for the same two groups, superficial and invasive; however, they
additionally reported generation of a signature of progression and metastasis by using patients with lymph node metastases and poor survival outcomes (56).
The Höglund group over the past 5 years has published a series of papers in which they have performed DNA and RNA analysis on a cohort of muscle invasive and superficial tumors. These papers initially identified two subtypes that they termed MS1 and MS2 which correlated highly with LG and HG histology respectively. The MS1 group contained increased FGFR3 mutations and was relatively genomically stable, whereas the MS2 group was enriched for TP53 mutations and high numbers of focal genomic amplifications (61). The group then went on to integrate the copy number, mutation and RNA data to identify five molecular clusters, of which three were primarily composed of invasive tumors (62). This was followed up using microarray data to
Unstable, Infiltrate, Urobasal B, and SSC-like) . One of these subtypes encompassed the MS1 group previously described in Lindgren et. al., however the MS2 group was divided into 4 additional subtypes (63). Overall, Urobasal B and SCC-like have similar gene expression patterns, differing in the Urobasal B subtype having an increase in an FGFR3 signature. The increase in FGFR3 signature in the Urobasal B subtype may be the result of the Urobasal B subtype having a higher number of non-invasive tumors, whereas the SCC-like subtype has a majority invasive tumors (63). While the subtypes appear to be prognostic, it is important to note that the poor performing subtypes, SCC- like and Urobasal B, have the highest percentage of muscle invasive tumors. In 2013, Sjödahl developed an immunohistological staining strategy using 20 markers to identify the aforementioned subtypes. This IHC based strategy was only able to accurately separate the Urobasal A, SCC-like, and genomically unstable tumors, the authors noted that the Urobasal B tumors were unable to be reproducibly identified.
The inability to distinguish these two subtypes may be a reflection of the fact that the Urobasal B and SSC-like groups overall gene expression patterns are very similar. One of the common features of the SCC-like and Urobasal B groups is increased expression of keratin 5 (KRT5) and 14 (KRT14). The Chan group has since reported that high expression of KRT14 is correlated to poor prognosis in bladder cancer (64). Both of these studies were performed using cohorts of HG and LG tumors raising the question of whether stratification of only muscle invasive tumors along the lines of keratin or other gene expression signatures still provide prognostic value.
To this point gene expression profiling efforts have focused on the utilization of a combination of LG and HG tumors, however, since pathologists can reliably identify LG and HG tumors, molecular signatures of these two groups are not clinically useful.
Multiple groups have recently addressed the identification of clinically relevant subtypes within HG tumors (43, 65, 66). These studies used cohorts of only HG tumors to identify multiple subtypes within HG bladder tumors. Choi et. al. identified three subtypes, basal, luminal and p53-like. The basal subtype was associated with decreased overall survival and high expression of high molecular weight keratins, similar to the SCC-like and Urobasal B subtypes reported by Sjödahl, whereas the luminal subtype had improved survival and enrichment for FGFR3 mutations. The third subtype, p53-like, was chemoresistant and had expression patterns related to wild type p53 expression (66).
In parallel, the Cancer Genome Atlas identified four subtypes, two subtypes (clusters I and II) had increased expression of markers of urothelial differentiation and decreased expression of keratins compared to cluster III and IV (43) This pattern of keratin expression was also a hallmark the subtypes identified by Sjödahl et. al. and the basal and luminal subtypes identified by Choi et. al. (63, 66). Additionally, TCGA cluster IV had increased expression of genes associated with immune/tumor infiltrating cells. This similar expression pattern was also present in the subtypes discovered by Choi (p53-like) and Sjödahl (infiltrated). Sjödahl noted that the tumors within the infiltrated group stained for increased numbers of T cells and myofibroblasts, raising the question of whether the differences in gene expression stem from intrinsic tumor cell differences or the gene expression patterns of the tumor microenvironment..
When comparing common features between these three studies, a picture emerges of 2-3 unique subtypes; 1) FGFR3 high, UPK high; 2) KRT high with
squamous differentiation; 3) samples with increased tumor infiltrating cells. While these subtypes seem to be reproducible and have been suggested to be prognostic, there still remains a need to derive a clinically useful tool for discriminating the subtypes based on a minimal set of differentially expressed genes.
This dissertation presents work that defines subtypes similar to those previously mentioned, creates a prognostic gene expression based classifier, and more broadly correlates the expression patterns of bladder cancer to those seen in breast cancer.
Table 1: Comparison of muscle invasive subtypes
Urobasal AGenomically Unstable Infiltrated Urobasal B SCC-Like Luminal p53-Like Basal I II III IV
FGFR3 UPK KRT Squamous Diff Infiltrate Cells
Summary
Taken together, evidence supports the notion that distinct genomic and
transcriptional profiles characterize LG and HG disease. LG disease is characterized by alterations of FGFR3, PIK3CA, andRAS as well as decreased levels of markers of differentiation and proliferation as compared to HG disease. HG bladder tumors are genomically unstable as compared to LG tumors and have alterations in DNA-damage response and cell cycle control. Additionally, HG bladder cancer is heterogeneous with varying degrees in expression of ERBB2, FGFR3 and genes involved in urothelial differentiation. While previous studies have identified molecular subtypes with some success, none have created a gene expression based predictor to differentiate molecular subtypes within HG muscle invasive bladder cancer.
Chapter 3: Intrinsic subtypes of high-grade bladder cancer reflect the hallmarks