Treasure maps for cancer research: Online atlases are pairing tumor anatomy with gene activity to deliver rich troves of information
Bryn Nelson · Cancer Cytopathology · 2015
A growing number of ambitious research projects are combining molecular data with high-resolution anatomical images to yield finely detailed cancer maps. By matching molecular changes with where they are occurring in the body, these online atlases could point researchers toward promising places to unearth clues about how the tumors originate and spread and lead to better diagnostic tools and treatments. The unprecedented level of detail is also highlighting the surprisingly heterogeneous nature of specific cancers, in which nearby niches may harbor startlingly different gene activities. The eagerly anticipated maps, it seems, may take a while to fully decipher. In one major project, a team led by the Allen Institute for Brain Science and the Swedish Neuroscience Institute, both in Seattle, recently mapped the deadliest form of brain cancer. The Ivy Glioblastoma Atlas, created with donated brains from 41 patients, will allow researchers to investigate how glioblastoma tumors, which kill half of all patients within a year of diagnosis, grow and colonize healthy brain tissue. The project, funded chiefly by the Ben and Catherine Ivy Foundation, began after local researchers compared notes on how little was known about the cancer's genetic and cellular basis. Amy Bernard, PhD, director of structured science at the Allen Institute, says that scientists have enthusiastically greeted the bounty of new data embedded in the map. However, the project's participants were surprised by just how much diversity their atlas revealed. “From a molecular perspective, it's quite compelling how diverse these tumors are,” Dr. Bernard says. “I think from the perspective of addressing this clinically, it should give people pause because it's going to be very, very difficult to find a single smoking gun to any of these types of tumors just because of the diversity of them.” In a machine-learning training session, the desktop application Mill (White Marsh Forests, Inc) was used to manually label images of an H & E–stained tissue section with colored boxes (green for the cellular tumor and magenta for the infiltrating tumor). Nevertheless, Dr. Bernard says that she has seen an explosion of interest in creating and using such cancer atlases as valuable new tools: “Researchers need some basic repositories of information that they can turn to that are standardized and reliable.” The Human Protein Atlas, a separate project led by the Royal Institute of Technology in Stockholm and Uppsala University, both in Sweden, includes a similar cancer atlas component that maps gene activity on the basis of patterns observed in multiple human cancer specimens. In a 2014 press conference introducing the 13th iteration of the atlas, the project's leaders described a massive effort that had already produced roughly 13 million annotated images and 300 peer-reviewed articles. The resulting tissue-based map “shows the precise location, to a single-cell level, of all the human proteins,” says collaborator Matthias Mann, PhD, a proteomics researcher at the Max Planck Institute of Biochemistry in Martinsried, Germany. The project leaders also briefly demonstrated their cancer atlas, which now includes data on nearly 530 gene mutations strongly implicated in cancer. Users can look at the relative activity levels of the estrogen receptor 1 gene, ESR1, in the 20 most common cancer types, for example. A summary page shows that the gene is expressed most strongly in breast, ovarian, and endometrial cancer; users can then take a deeper dive into the supporting data, such as antibody-stained histology slides from multiple patients. “I think it's great and amazing,” Dr. Bernard says of the effort. “The approach that they've taken is very similar in concept to the types of approaches that we've taken, as well. It takes a very industrialized and systematic approach to creating bodies of data that other people can draw from to accelerate their own research. So I think it's very much a sister in concept.” As part of its expansion plans, the Human Protein Atlas group hopes to begin working on a cancer-focused pathology version by 2017. Meanwhile, a multicenter research collaboration in the United States has published an early version of its combined anatomical and disease-based MRI atlas of the prostate.1 The Cancer Genome Atlas, run jointly by the National Cancer Institute and the National Human Genome Research Institute, does not contain the same level of matched anatomical information. Even so, the atlas may offer yet another launch pad for additional efforts by cataloging the genomic changes in 33 cancer types on the basis of tissue collected from 11,000 patients. Like its counterparts, the glioblastoma atlas offers multiple entry points for a plunge into its data, such as the framework of stained histology images, implicated genes, and anatomical features, including blood vessels and tissue margins. This fine level of detail was not easy to gather. Nameeta Shah, PhD, a bioinformatics scientist at the Swedish Neuroscience Institute and one of the principal investigators, recalls that she and her colleagues spent the first few years trying to acquire enough intact tissue to construct the atlas. Eventually, the team collected sufficient tissue from 41 tumors, each cut into centimeter-sized cubes and then sliced into 20-μm sections. Some sections were stained, and others were used for gene expression arrays; the team then layered the data into an informatics platform that tied precise anatomical regions to the magnetic resonance imaging–based images, histological staining, and gene activity. The collaborators worked together to ensure that the project included high-resolution, pathology-grade images that captured the local milieu of the tumor tissue. By the time they were done, the researchers had amassed 22,000 annotated images. Past studies of glioblastoma-associated genes, Dr. Shah says, often involved “taking a piece of the tumor tissue and mashing it up,” and this resulted in an inevitable loss of information about the tumor's anatomy and microenvironment. “What this atlas allows you to do is actually study these components separately, and what we found is that location is the most important factor that influences gene expression. So similar locations from 41 different patients will have gene expression close to each other.” In other words, the cancer's varied anatomy is driving the divergent gene activity. “And this is the first time you can study it in so much detail,” Dr. Shah says. “What I'm really hoping is that we will be able to model the disease much better and that will allow us to have much more effective therapeutics for this cancer.” Researchers hoping for straightforward associations of low-, medium-, and high-activity tumor-associated genes, however, may be disappointed. “It's not necessarily what people want to hear, but the truth is that the tumor landscape is extremely, extremely complex,” Dr. Bernard says. “And we don't just mean that to minimize the problem or to restate what's already known.” From investigating the tumor's ultrastructure, for instance, the researchers noticed that several nearby pockets could have completely different gene activity profiles. “You realize that the tumor landscape itself is highly, highly heterogeneous,” Dr. Bernard says. “And as such, you realize that you're working with a little complex ecosystem.” Because of the complexity, she says, the layered information can provide a crucial navigational aid toward a more realistic picture of glioblastoma behavior. “We're providing the information to say, ‘Well, in what way is that more complex, and what specific markers are you talking about?’” she says. One gene marker may be turned up high throughout the tumor tissue, whereas another may be highly active only at the leading edge of the periphery. “The advantage of using the RNA sequencing approach is that you're literally sequencing each and every transcript that gives rise to a gene [product],” she says. For each isolated piece of tumor tissue, the researchers could ask what all of the genes are doing en masse. The atlas also offers more in-depth information for a smaller subset of genes already implicated in glioblastoma tumor pathology. Dr. Bernard says that she hopes the open-access information is ultimately tapped by a wide variety of scientists and clinicians. “We want this to be as open to as many different kinds of users as possible, because we can't possibly predict who is going to be making the next big discovery,” she says. “We're just providing the data there for them to mine.” With such maps to guide them, however, cancer researchers may be considerably closer to hitting a rich new vein. We want this to be as open to as many different kinds of users as possible, because we can't possibly predict who is going to be making the next big discovery.—Amy Bernard, PhD