Brain cancer researchers have a new, panoramic visualization tool to help them navigate the complex disease. Built from several publicly available datasets of gene expression and DNA sequences, the new brain cancer landscape acts like a city-wide map of the disease, carefully assembled from neighborhood maps of different brain tumor subtypes.

“As researchers, we can get so focused on comparing like with like that we lose sight of the proverbial forest for focusing too much on the leaves of a single tree,” wrote Holland Lab computational biologist Sonali Arora, MS, in the paper describing the tool, published in Scientific Reports in March.

Her new map offers a detailed landscape of molecular alterations in brain tumors that will help cancer researchers move between different subtype neighborhoods, as well as between cancerous and healthy brain tissue. It’s the first brain cancer computational visualization tool to combine different molecular datasets from both pediatric and adult brain tumors into one landscape, which will allow researchers to chase down key genes and biological pathways that shape development, tumor growth and treatment response, said Arora, who melded the datasets.

“Now you can compare all kinds of things, like the expression of certain genes or pathways … you can compare across all tumors types and normal brain,” said Fred Hutch brain cancer researcher and Human Biology Division Director Eric Holland, MD, PhD. “And that’s something that hasn’t been done before. … Now you can compare and learn about gene expression in one tumor versus another, or one subset versus another.”

The computational tool reveals biological processes that are ramped up or down in cancer compared to normal brain tissue. And because it draws on datasets from the U.S. and China, Arora was able to show that while tumors from both regions often share molecular alterations, there is a certain subset of glioma in patients from China that is not seen in the U.S.-based data.

And to enable scientists to explore her brain tumor map in 3D detail, Arora uploaded the landscape to the visualization tool Oncoscape, also developed in the Holland Lab.

Arora’s landscape “offers a way to visualize both normal and tumor samples and show tumor relationships not available in standard plotting methods,” Fred Hutch colleague and cancer stem cell biologist Patrick Paddison, PhD, wrote in an email. Paddison and his team used the tool to help identify genes that are critical for brain tumor cells (but not healthy brain tissue), as well as their biomarkers.

Melding Datasets Offers Deeper Insights Into Brain Cancer

Arora got the idea to build the landscape during the initial COVID-19 pandemic shutdown. She saw an opportunity to give brain cancer researchers a wider view of brain cancer by creating a map that combined data from multiple brain tumor subtypes housed in publicly available data repositories.

These datasets include a spectrum of molecular information, including tumors’ DNA sequences and the patterns of genes that are turned on and off in tumors. The Cancer Genome Atlas, or TCGA, a National Cancer Institute–funded program, includes samples from adult tumors that span 33 cancer types. Arora combined information from 702 TCGA patients with glioma and 270 patients with tissue samples in the Chinese Glioma Genome Atlas, or CGGA. To allow scientists to easily compare the differences and similarities between adult and childhood brain tumors, she added brain tumor data from the Children Brain Tumor Network, or CBTN. And to help scientists clarify how tumors differ from normal tissue, Arora included molecular information from 1,409 normal brain tissue samples in the Genotype Tissue Expression Project, or GTEx.

She then used a technique called uniform manifold approximation and projection for dimension reduction, or UMAP, to simplify the large, complex data and make it easier to detect key relationships between molecular changes and tumors. Arora also used a computational technique called batch effect correction, which to her knowledge has not been used at this scale before. This strategy helps scientists clear away variation in data that arises because samples are processed under different conditions — and see true biological variation.

In Arora’s UMAP, brain tumors that share gene expression patterns cluster together in the same subtypes. And like a city map, her landscape shows which subtype neighborhoods are near neighbors, and which are across town. Arora was able to see how biological processes differed between normal tissue and cancer, and between different cancer types, including tumors from adult and pediatric patients.

“There are clearly some pathways that are upregulated,” Arora noted.

Relationships between tumor types, as well as tumors and key genes and biological processes, quickly jumped out, she said. For the most part, tissue from gliomas taken from patients in the U.S. and China clustered into the same neighborhoods, save for two clusters of glioma subtypes seen in the CGGA but not the TCGA data. It’s already known that the same cancer type may manifest differently in different regions around the world, and the new brain tumor landscape could help scientists figure out how and why this occurs for glioma as well, the researchers said.

The UMAP showed that in comparison with normal brain tissue, glioma tumors from adults had higher levels of key cancer-promoting biological processes, including those that promote cell growth and DNA repair. Some pediatric tumors had also ramped up these processes. The UMAP also reveals pathways ramped down in tumors, including some neurotransmitter pathways.

Researchers can use the UMAP to dig deeper into these pathways, and explore individual genes involved at different steps in each biological process, Arora said.

“You can visualize the expression of each of those genes over the brain tumor map,” she said. By using the UMAP, she was able to see that while a specific process — in this case, a type of DNA repair — might be elevated in cancer cells, the expression level of individual genes involved in that process might vary, with some remaining the same as in normal brain tissue.

To help researcher examine the relationship between DNA changes and brain tumors, Arora built a smaller UMAP using the samples from TCGA and the CBTN that included DNA sequence information. This gave a window into the range of DNA mutations that can drive cancer, from changes in single DNA “letters,” to the replication or loss of larger chunks of DNA. Sometimes these changes can cause two genes to be fused together, forming a new “Frankengene” that may drive cancer by acting differently than either of its parent genes. Arora’s UMAP showed that certain gene fusions were more common in specific brain cancer subtypes.

“Now you’re able to compare a group of tumors amongst each other,” Holland said. That’s unusual, he noted: It’s far more common for scientists to study and report on one tumor type. These cross-subtype comparisons could help brain cancer researchers discover treatment targets shared by multiple brain tumor types, or those unique to specific subtypes.

Paddison used the UMAP to uncover vulnerabilities — or genes and biological processes that cancer cells need but healthy cells don’t — in different subsets of both adult and childhood brain tumors.

A Step Toward More Precise Diagnostics and Treatment

Precision oncology is the tailoring of a cancer patient’s treatment plan to their tumor’s unique blend of vulnerabilities. While cancer researchers have made great strides, there remains a lot to discover about the treatment targets lurking within tumors, and how to determine which therapies will provide the greatest benefit to which patients.

Arora’s UMAP, combined with Oncoscape’s data visualization capabilities, will help further those discoveries, Holland said.

It will make it easier for brain cancer researchers like Paddison to uncover what drives brain tumors, whether it’s from DNA mutations or large-scale changes in gene expression patterns. Identifying key genes or pathways in the UMAP could help researchers better choose candidate therapies for clinical trials, Holland said.

“This type of approach could be used to more precisely place patient tumors in continuum of adult or pediatric brain tumors to better predict outcomes and survival,” Paddison said.

Using tumor landscapes, incorporating patient clinical as well as tumor molecular landscapes, to refine diagnosis and treatment is Holland’s hope as well.

“You could imagine a world where a given tumor is sequenced, and then placed on the landscape. Then your nearest neighbors on the landscape could tell you what your diagnosis actually is, but also what your expected outcome would be,” Holland said.

Arora’s multi-dataset brain cancer UMAP is “to my knowledge, the first pseudo pan-cancer approach to a really interactive way of learning about, and predicting, new tumor behaviors,” he said.

This work was supported by the National Institutes of Health, the Jacobs Foundation and the National Science Foundation.

This article was originally published April 10, 2023, by Fred Hutch News Service. It is republished with permission.