Pediatric Tumors Made Personal
A mixed collection of relatively rare but often deadly pediatric tumors are collectively known as small round blue cell tumors (SRBCT) for precisely the reason one might imagine. Examined under a microscope after routine processing, bone marrow biopsies from cancers including neuroblastoma, Ewing sarcoma, rhabdomyosarcoma, and lymphoma appear as small, blue, and round cells. Despite some distinguishing molecular markers to guide them, oncologists can, on occasion, find it hard to diagnose these tumors specifically. Javed Khan, M.D., Head of the Oncogenomics Section of CCR’s Pediatric Oncology Branch, has been using genomic approaches to study pediatric cancers for several years. He is now poised to launch an ambitious multicenter project to use comprehensive genomic data to guide the individualized treatment of children with advanced solid tumors.
Tapping Gene Expression
Khan is a strong believer in the power of genomic information to guide solutions to the riddles of cancer. A pediatric oncologist who trained in Cambridge, England, Khan first came to the NIH on a hematology/ oncology fellowship that involved translational research at the National Human Genome Research Institute (NHGRI). Jun Wei, Ph.D., was also at the NHGRI and moved with Khan to CCR when he became Head of the Oncogenomics Section in 2001. At the time, the NHGRI was heavily involved in developing microarray technology to analyze gene expression. "Those were very heady, exciting days," remembered Khan. "Working with pediatric solid tumors, we were one of the first to use microarrays to find a cancer diagnostic."
In 2001, Khan, Wei, and their colleagues published a paper in Nature Medicine in which they demonstrated that relatively small numbers of genes could be used to distinguish four different SRBCTs. In the paper, they used artificial neural networks, a computational technique in which the correct method for finding a solution evolves through a training process. A set of microarray data from identified tumors is used to train the network to recognize patterns in the data that uniquely correspond to each tumor type. Once the network is trained in this way, it can use the rules it learns to predict new cases.
Javed Khan, M.D. (Photo: R. Baer)
"The advantage of our method," explained Khan, "is that it allows you to analyze multiple cancers and generate a score that reflects confidence in any particular diagnosis." It is, for example, easily adaptable to a Web site format so that physicians could load microarray or other gene expression data from their own patients to obtain diagnostic information. In fact, Khan and his colleagues have a patent on their method, which a San Diego-based diagnostic company, AltheaDx, is developing into just such a product for pediatric cancers.
Reading the Whole Genome
"The end game for me is personalized therapy," said Khan, "in other words, being able to use genomics to diagnose cancers and to distinguish those who will survive on existing therapies (prognostication). And in the midst of studying all those genetic alterations, find ones that are the key targets for therapeutic intervention in advanced disease." To search for genetic changes that might be driving these cancers, Khan and his team rely on multiple strategies.
Microarrays measure the expression of genes that are being actively transcribed from only a subset of the entire genome—the transcriptome. These data give you important information about changes that occur during RNA transcription and processing. Although he has firsthand experience with the diagnostic value of gene expression data, when it comes to stratifying disease progression, defining targets, and predicting outcomes, Khan’s first bet is on looking at the DNA directly. DNA sequence information does not tell you which genes are expressed at a given time, but it does tell you directly which genes have been mutated.
"The end game for me is personalized therapy"
"To distinguish one cancer from another, the differences [in gene expression] are quite large," explained Khan. "But to distinguish survival outcomes for one type of cancer, the differences are often much smaller. So it becomes much more of a challenge to distinguish prognostic signatures using gene expression data." The problem with RNA is largely a practical one. The molecules themselves are simply much more dynamic. "If you take a tumor sample out and you don’t freeze it immediately and then wait an hour, the expression profile can be profoundly altered. Also, tumor cells that are hypoxic at the center of a tumor may have a very different profile from cells in the periphery of the mass. DNA doesn’t change. RNA does." Khan noted that although there are several published prognostic gene expression signatures for breast cancer or neuroblastoma, for example, there is very little overlap between each of the gene sets for a given cancer. Thus, to validate these signatures for prognostic purposes requires prospective clinical trials in which sample handling and analysis are stringently controlled with standard operating procedures.
As a result of incredible advances in DNA sequencing technology over the last decade, it is no longer impossible to think about sequencing the whole cancer genome of an individual cancer. "Where it’s going is next generation sequencing," said Khan. "The human genome project sequenced the first human genome in 15 years. Now you can do a whole genome in about a month, which is still too long in terms of using it to make therapy decisions. But, you can sequence all the protein-coding genes—the exome—within a week."
With exome sequences in hand, it is still a long and laborious process to identify the mutations that might be critical to tumor growth and survival. The sequence from the tumor must be compared to the patient’s germline DNA and also to published sequence data to find mutations that are specific to the cancer. From one tumor, a hundred functional mutations might emerge and many of these are probably passenger mutations resulting from an unstable genome that are not critical to cancer progression. Comparing mutations across tumors can help to narrow the field, as can analyzing the pathways that might be compromised by individual mutations.