Glioblastoma subtypes
A major characteristic of glioblastoma is its intertumoral and intratumoral heterogeneity. Subtypes of glioblastoma have been defined best on gene expression profiling. Single cell analyses confirmed that multiple subtypes exist within a single tumor. The dominant subtypes influence the surrounding tumor microenvironment. We identified three dominant subtypes of glioblastoma by separating tumor-cell from microenvironmental-cell gene expression based on the overlap between single cell sequencing data, glioma sphere forming cell line data, and bulk tumor sequencing. The resulting subtypes - proneural, mesenchymal, and classical - each have unique biology and each recruits a distinct immunologic infiltrate. Through whole exome CRISPR screening, our laboratory has identified potential vulnerabilities in the most aggressive subtype, mesenchymal, that are potential targets for therapy.
Mesenchymal radiation resistance
The mesenchymal subtype of glioblastoma demonstrates significant treatment resistance. Radiation therapy is the mainstay of treatment for patients. We found that the inflammatory NF-κB pathway promotes radiation resistance. Subsequent reports by our lab and our collaborators demonstrated a distinct localization of mesenchymal subtype cells in the core of the tumor, the area that receives the highest radiation dose yet is the most likely location of tumor recurrence. CRISPR knockout screening of a radiation resistance glioma sphere forming cell model has identified multiple potential targets for radiosensitization in this aggressive subtype.
Unified diagnostic platform
Whole methylome analysis has become a mainstay of brain tumor diagnosis. An approach using unsupervised clustering of methylation data has greatly improved the classification of brain tumors. For gliomas, the unsupervised approach is not without misclassification errors. We developed a complementary approach using the same technical platform, the Infinium methylation arrays using a supervised model approach, to develop predictors of the key defining molecular biomarkers of gliomas: isocitrate dehydrogenase (IDH) mutation, telomerase promoter mutation (TERTp), alpha-thalassemia/mental retardation syndrome X-linked (ATRX) mutation, chromosome 1p/19q co-deletion, and the gene expression subtypes. This unified diagnostic platform (UniD) has been incorporated as part of routine clinical molecular pathology. Of note, this represents the first clinical diagnostic to permit identification of the dominant molecular subtype of the tumor.
The UniD package is available in our GitHub repository.
Lineage tracing and recoverable DNA barcoding (CAPTURE)
Understanding clonal evolution of tumor populations is central to identifying mechanisms of treatment resistance. Recent approaches for lineage tracing have relied on incorporation of DNA barcodes into model systems followed by comparison of barcode distribution before and after treatments. One challenge of this approach is that the association between an emerging or preexisting alteration and barcode enrichment is inferential at best as the interrogation of the barcode distribution by sequencing prevents further characterization of individual clones. We developed a novel barcoding vector that maintains high complexity using a Cas 9D10A and paired-gRNA targetable unique reporter (CAPTURE) single-cell barcoding library. In a proof of principle study, we demonstrated the emergence of a novel mechanism of mutant BRAF drug resistance in a melanoma model.
Using the CAPTURE vector in a unique approach, we have labeled glioma sphere forming lines with one barcode per line. We then pool the lines for high-throughput drug/therapy screening both in vitro and in vivo. The technique, called CARPOOL (patent pending), allows for screening of large number of candidate treatments in a low-cost, rapid approach.
Deconvoluting tumor educated platelets
The use of liquid biopsies is revolutionizing the diagnosis and longitudinal assessment of treatment response in oncology. For brain tumors, the blood brain barrier limits the free flow of tumor cells, DNA, and extracellular vesicles into the bloodstream. Tumor-specific RNA signatures from peripheral blood platelets (so called "tumor educated platelets") were reported as a novel liquid biopsy and, for patients with gliomas, were suggested as a possible means to track treatment response and detect early tumor recurrence. The original discoverers of these signatures suggested that it may involve direct transfer of RNA from tumor cells to platelets. We sought to understand better the components and source of the RNA signatures in these platelets. Using human/mouse xenografts we found that little if any human tumor cell RNA was present in mouse platelets, suggesting that direct RNA transfer is unlikely to occur. We deconvoluted the RNA signatures of the platelets in silico and found that many of RNAs present came from inflammatory cells or red blood cells and not platelets, possibly from contamination during the platelet RNA preparation or alternatively from transfer from these cell types to the platelets. The analytic approach used to deconvolute the signatures is now being tested to optimize the platelet signatures and improve their diagnostic potential.