In a state-of-the-art lecture delivered last month at the 2014 Annual Meeting of the American Urological Society, Dr. Dan Theodorescu,Director of the University of Colorado Cancer Center, discussed the need for prognostic biomarkers for bladder cancer and the issues associated with nailing down such precision medicine.Calling biomarkers the “pillars of personalized medicine,” Dr. Theodorescu argued that there is no current prognostic biomarker for bladder cancer that is well-validated, prospectively evaluated, and FDA-approved that can add to stage and grade, the current predictive tools—but not for lack of effort. “As we move on from empirical medicine, from the stethoscope to stratified medicine using the Kaplan–Meier curve, to precision personalized medicine using genomic and biomarker technology, we’ll acquire the ability to take patients such as Mr. Smith with a bladder tumor, and, instead of predicting Mr. Smith’s outcome based on 100 tumors from similar patients, we will be able to precisely determine Mr. Smith’s biology based on his tumor and its exact characteristics and at the right time for therapy.”
Dr. Theodorescu established as the background to keep in mind when validating biomarkers in the literature several biomarker types that stem from genomic “regions,” starting at 1) the blueprint of DNA; to 2) RNA and 3) protein, which comprise the machinery of the cell; and then to 4) the output of that machinery, the metabolome. These four biomarker types can be accessed in various materials such as tumor tissue, blood, urine, and normal tissue, if risk is being assessed. The biomarkers can be single or in panels comprising: one type, DNA; multiple types; and multiple materials. Panels are made up not only of one element, but of RNA, DNA, and protein combined.
Dr. Theodorescu emphasized the importance of conceptualizing what happens when the patient is evaluated, with the premise that, since cancer is an interaction between a host and a tumor, both host and tumor are important and both need to be examined at the molecular level. Because of this interaction, the host, for example, can be evaluated for DNA composition as well as other parameters. Body fluids and the tumor itself can be examined using DNA, RNA, and protein with known technologies such as alpha matrix array, mutations and copy number, and protein and RNA/miRNA expression.
When the evaluation is complete, the information can be integrated into what Dr. Theodorescu called a personalized molecular blueprint of the patient and the patient’s tumor. He explained this “blueprint” as providing both prognostic information (will the patient benefit from treatment?) and predictive information (what particular treatment should be used?). And this evaluative process is then repeated as tumors change, as they will with respect to genetics and epigenetics when therapeutic pressures—chemotherapy, radiation, and targeted agents—are applied.
Dr. Theodorescu then focused his discussion on a novel mutation in bladder cancer, citing three articles published last year. In these articles, the protein STAG2 was identified as being mutated in bladder cancer.1–3 STAG2 is very important in cell division and separation of chromosomes; therefore, if the level is low or the protein is missing altogether, an anomaly can occur in how the chromosomes separate in daughter cells, leading to genome instability. From 16% to 32% of urothelial carcinomas have been shown to have STAG2 mutations, and gene mutation leads to a reduction in the protein or its absence. One study showed that mutation is associated with a poor prognosis; therefore, low levels of STAG2 mean poor prognosis for the patients.2 By contrast, another study showed that low levels of STAG2 yielded a good prognosis.3 Therefore, although the investigators in the latter study looked at protein instead of DNA itself, the results were apparently opposite. There are multiple causes for this discrepancy, which Dr. Theodorescu acknowledged “seems surprising and somewhat alarming”; however, multiple explanations can be proposed.
Similar discrepancies have been seen before. For example, in the early 1990s, a period of high excitement in the field of early biomarker work was generated by p53. A paper in the New England Journal of Medicine in 1994 showed in a cystectomy series that higher levels of p53 were a bad prognostic factor.4 Prospective studies were subsequently conducted in 500 patients in a phase III trial evaluating similar endpoints, and the findings were essentially negative. This example was cited by Dr. Theodorescu as highlighting an important issue in biomarker development—studies need to be prospectively validated, either in a truly prospective fashion or in prospectively collected samples, before being used.
Dr. Theodorescu offered another example of a biomarker that hasn’t been prospectively validated, FGFR3. “In bladder cancer, FGFR3 has been associated, counter-intuitively perhaps, with better outcome and lower- rather than higher-grade disease.” He went on to say that the data are fairly convincing. Multiple studies have evaluated the gene, and some of these were prospective. Results indicate that FGFR3 mutation is a good prognostic factor. “More importantly, because it is extremely common in low-grade disease, this marker may be effectively used as an assay in the urine to detect bladder cancer, which may help determine whether cystoscopy is necessary.”
He went on to talk about neither DNA nor a mutation, but RNA, wherein many biomarkers, generally in panels, have been identified. He cited an important study that evaluated over 20 different biomarker signatures. Collections of RNA panels were able to predict stage/grade, and 10 were able to predict for progression and survival. In essence, this study showed that, for stage and grade, the accuracies were 70% to 90%, and the signatures with >150 genes had even more robust prediction. However, some large randomly selected signatures performed even better. “Furthermore, when it came to the 10 prognostic signatures, none performed better than random gene sets when applied to independent validation data.”
He offered as a prototypical model the work of a group of researchers who set out to develop biomarkers for node-positive disease based on expression of 20 genes. They developed the biomarker panel of 20 genes using multiple datasets. They then used the gene set in a completely independent and prospectively collected set of tumors and showed an ability to predict node-positive disease.Dr. Theodorescu wound up by providing an example that brings together both mutations and RNA in the search for biomarkers.5 He and his colleague “sequenced 99 tumors using next-generation sequencing and identified a number of mutations. Both among non–muscle invasive and invasive tumors we can classify these mutations into three large groupings: chromatin regulators, cell cycle regulators, and oncogenes. In parallel studies of RNA expression, we studied samples from almost 2000 patients for tobacco-related cancers—bladder, lung, and head and neck—to find a robust signature of progression with a technology called “gene enrichment.” Surprisingly, the cell cycle also proved to be a significant factor. When we combined the two, we came up with a 15-gene panel that was highly prognostic and able to predict the outcome in multiple datasets. This helps to confirm the principle that combining many datasets and validating them leads to a more robust result.”
- Exciting developments are happening with respect to the molecular biology of bladder cancer. We are hopeful that in 5 to 6 years a validated assay will be available for clinical use.
- Need: standardized assessment protocols (SOPs) and assays
- Need: independent “prospective” validation
- Need: multivariate analysis—to validate new markers with nomograms to determine whether they add additional information
Guo G, Sun X, Chen C, et al. Whole-genome and whole-exome sequencing of bladder cancer identifies frequent alterations in genes involved in sister chromatid cohesion and segregation. Nat Genet. 2013;45(12):1459-1463.
Tony Nimeh MD