To date 661 of the 874 samples accrued have been processed using U133 2 Plus mRNA expression chips (Affymetrix, Inc.,Santa Clara, CA),which contains over 54,000 probesets analyzing the expression level of over 47,000 transcripts and variants,including 38,500 well-characterized human genes.From the wealth mRNA expression data generated we were able to develop a novel,comprehensive classification of gliomas based on their mRNA expression profiles,which were analyzed using a variety of class discovery algorithms w/out using any pre-conceived notion of their biological or clinical/histopathological grouping, thus finding the classes according to pure mathematical analyses of the expression characteristics of each sample.The results were validated in separate datasets and classified into six subclasses having a nested hierarchy correlating with the tumors characteristics from both biological and clinical standpoints.This novel classification schema, along with the wealth of genomic data produced by the GMDI, may allow for the development of a more rational,objective,diagnostic classification system for gliomas and serve as an important starting point in the search for new molecular therapeutic targets. Genomic DNA from 72% of samples received (634/874) has been hybridized to the 100K SNP chips,which covers 116,204 single-nucleotide polymorphism (SNP) loci in the human genome with a mean intermarker distance of 23.6 Kb. These arrays give two simultaneous data types:Allelic Calls and Signal intensity,allowing for the determination of both Copy Number alterations (CNAs) and regions of Allelic Imbalances (Loss of heterozygosity, LOH).Calls were determined by the GTYPE software version 3.0 with 25% level of confidence.Only samples with call rates more than 90% were accepted for any analysis.To identify areas of alterations where putative new target genes could be located;we analyzed the results from the genomic survey in our retrospective sample of 180 gliomas.We thus devised an integrative method for analysis of genomic data by combining the CNA/LOH data with mRNA expression obtained from the same samples.This approach yielded a substantially reduced list of candidates than CNA/LOH or mRNA expression studies would have produced alone.Moreover, these candidates were more likely to be biologically relevant due to the fact that the mRNA over/underexpression could be traced directly to a genomic alteration, thus indicating that the genes may be contributing to the pathogenesis of the tumor.However, this analysis approach did exclude genes (particularly putative Tumor Suppressor Genes) that may be epigenetically regulated, lacking correlation with the genomic changes detected. To solve this shortcoming, we then focused on all probe sets mapping to areas where at least 10% of samples showed LOH, and selected those for which a substantial fraction of LOH samples had an mRNA expression lower than the median expression shown by our non-tumor reference samples.In this way, we identified about 400 genes most of which had a clear pattern of bimodal mRNA expression distribution, with some samples having little or no expression, while others had an expression level consistent with their copy number status.Methylation and mutation analysis of these gene/sample pairs confirmed the epigenetic nature of their mRNA modulation, suggesting that the genes may indeed be novel TSGs. In the early phases of GMDI, we recognized that although the use of fresh, flash-frozen tumor samples was vital for generating all of the data types we wanted (i.e. mRNA expression) ultimately the need for fresh tissue would be limiting and impractical for routine clinical application of a mRNA-based,GMDI-generated classification system since most tumor tissue from surgeries performed at outside hospitals is paraffin-embedded and not frozen. Furthermore, we have a wealth of pts with gliomas being treated with molecularly targeted agents on investigational trials.Being able to correlate the pts response to these molecularly targeted agents with the specific genomic alterations of their individual tumors could be profoundly important for beginning to identify subgroups of pts likely to respond to individual treatments.Such correlations would be the basis for the design of future clinical trials enriched for patient populations most likely to benefit from any given treatment and represent a significant movement toward the age of personalized medicine. Thus,we have expended a significant effort to develop methodologies that would allow us to generate high quality,consistently reproducible genetic data from FFPE samples. We have optimized the DNA extraction system from FFPE tissues by means of repairing single-strand breaks before extraction.This has allowed us to increase the median size of the DNA fragments recovered from the paraffin from the standard 300bp to our enhanced 700bp.This quality increase has allowed us to use FFPE-derived DNA in the Affymetrix Sty 250K SNP chips with call rates that range from 80-90% of those obtained with high quality DNA from frozen tissue.This methodological advancement allows us to significantly expand our ability to analyze genomic changes in a variety of rare glioma types as well as well as perform detailed mapping of disease progression in pts from past surgical resections for whom fresh frozen tissue is not available.We are now in a position to map genomic changes in the tumors of all pts who are enrolled on our clinical trials giving us an unprecedented opportunity to begin to make genomic correlations with clinical outcome data from pts being treated with molecularly target agents, making the idea of personalized therapy a realistic goal. REMBRANDT (https://caintegrator.nci.nih.gov/rembrandt/) contains not only the data generated from this prospective study but also a suite of querying and displaying tools allowing researchers around the world to profit from the GMDI data. The importance of REMBRANDT for the scientific community cannot be overemphasized as proven by the award of the Service to America Medal and congressional commendations given to it, and its characterization as a prototype for the caBIG effort. Beyond the provision of the molecular data stored in the database, NOB was essential in the creation of the system as we provided the informaticians and software engineers with more than 50 use-cases that were at the core of the basic infrastructure of the system. The NOB has served as the reference for the type of analysis that should be performed, and acted as the main scientific advisor to the programmers.The web-based interface was designed to allow unfettered access to GMDI data to a wide variety of users allowing to save the results in a personal folder, combine queries without having to type them again and exporting the data to commonly used platforms (BRBTools and GenePattern) to allow for customized sample analysis. In fact, Rembrandt allows for full raw data download for the use of more sophisticated users not only for analysis but also as a testing ground for bioinformaticians developing new applications. We are enhancing the system by adding new data types that are being generated, creating tools for their querying, extending and improving the visualization tools available as w [summary truncated at 7800 characters]
Showing the most recent 10 out of 14 publications
