The company they keep

In 2020, Everett Moding, MD, PhD, an assistant professor in Stanford Medicine’s radiation oncology department, noticed that some people with a rare cancer called soft-tissue sarcomas were cured with surgery and radiation while others saw their cancers quickly recur. “Two patients could have the same diagnosis and be treated the same way, but their cancers would respond very differently,” he said. “And there was no effective way to predict who would have a poorer prognosis.”

Around the same time, Magdalena Matusiak, PhD, then a postdoctoral student in the laboratory of professor of pathology Matt van de Rijn, MD, PhD, the Sabine Kohler, MD, Professor in Pathology, was growing frustrated with the traditional methods of predicting cancer cells’ growth based primarily on mutations in their DNA.

“In many instances, we can’t explain tumor biology just by looking at mutations or gene expression,” Matusiak said. “Ductal carcinoma in situ, a common early breast cancer, is not usually life threatening. But in about 1 in 4 patients, these cancers will become invasive for reasons we can’t explain with conventional methods.”

Both young researchers turned for answers to a rapidly growing field defined by leaps in technology and machine learning that allow a close-up look at the thousands of interactions between cancer cells and the healthy cells and tissues in which they reside. This three-dimensional neighborhood is broadly defined as the tumor microenvironment, and our growing understanding of its importance relies heavily on studies of what’s been called spatial biology.

It turns out that the company that cancer cells keep — and the way that company reacts to their presence — is critical to determining whether a new cancer grows, thrives and metastasizes to other parts of the body or is pounced upon and eliminated by the immune system.

“Cells don’t exist in isolation,” said assistant professor of biomedical data science Aaron Newman, PhD. “A cell’s identity, its behavior, its characteristics depend on what other cells are around it in three-dimensional space and what those cells are doing. But even five years ago we didn’t have a good way to identify these interactions. Now we can begin to assess aspects of this nuanced, community-specific biology.”

“Cells don’t exist in isolation. A cell’s identity, its behavior, its characteristics depend on what other cells are around it in three-dimensional space and what those cells are doing.”

Aaron Newman, PhD, assistant professor of biomedical data science

Newman, a member of the Stanford Cancer Institute and a Chan Zuckerberg Biohub Investigator, is one of several Stanford Medicine scientists developing tools and techniques to collect and interpret dizzying amounts of data from human tumors to identify, on a cellular communications level, exactly who says what to whom, as well as where, when and why. It’s a daunting task when you consider that a tumor the size of a small grape contains something on the order of 1 billion cells.

Some heavy hitters back this research, among them the National Cancer Institute, which in 2016 named the Human Tumor Atlas Network as one of the key research initiatives of its Cancer Moonshot — a program created to focus on areas of research deemed most likely to benefit cancer patients. The tumor atlas network aims to detail the evolution of the cellular and molecular interactions among healthy and diseased cells as a precancerous growth develops into full-blown cancer.

“It’s really clear that a tumor is not just a collection of cancer cells,” said Sylvia Plevritis, PhD, chair of Stanford Medicine’s Department of Biomedical Data Science, the head of the Stanford Center for Cancer Systems Biology and the Stanford Cancer Institute’s associate director of cancer AI.

“In fact, some of the most difficult tumors to treat, like pancreatic tumors, are mostly noncancer cells. Techniques to study the spatial biology of tumors, like those developed in Aaron’s lab and several others at Stanford including mine, are changing our understanding of cancer. Now, we can not only see what cell types are in the tumor but who their neighbors are and the molecular interactions that allow them to communicate and sustain each other.”

In just a few years, researchers have gone from deciphering flat, stained slices of tumor tissue highlighting the gross anatomy of a tumor to parsing not just the precise cellular composition of small tumor samples but even identifying specific cellular neighborhoods and interactions that can determine health or disease. The insights are providing important clues to medical mysteries, like this one puzzling Moding and Matusiak: Why do some patients with what seem to be very similar cancers have better outcomes than others?

Proving the link between
cancer cells and their surroundings

The idea that the cells and tissue surrounding a cancer cell may be as important as the cancer cell itself for determining whether the cancer cell thrives, divides and — eventually — metastasizes was first floated in 1863 when German physician Rudolf Virchow, MD, noted a connection between inflammation and cancer. In 1889, English surgeon Stephen Paget, FRCS, advanced his “seed and soil” hypothesis that the cellular environment within which a metastasizing cancer cell landed influenced whether it would flourish or die in its new location.

At that time, there were few ways to prove these hypotheses on a cellular level. Aspiring investigators pored over microscope slides holding thin slices of tissue stained a dull purple to delineate individual cells and structures. Researchers could only infer relationships among cells from a snapshot in time frozen on a two-dimensional grid — a bit like trying to predict how occupants of a high-rise spend their time by looking at the building’s blueprints.

Decades later, in the late 1960s, scientists devised a way to attach color-changing proteins to antibodies that recognize and bind to specific cellular structures — vastly increasing the amount of information that could be garnered from a single slide. Now they could see the arrangement of furniture in individual rooms and predict the function of each space. But still, there was no inkling of how the cells communicated, or didn’t, with one another in living tissue.

“Many times, if you just look at tumors as a bag of cells, your ability to predict a patient’s prognosis is not great, even if you know how many of each cell type is in the sample. But if you can incorporate where those cells are in the tumor, those predictions become much better.”

Michael Angelo, MD, PhD, associate professor of pathology

The floodgates started to open when genomic sequencing took off in the early 2000s. Soon researchers learned how to infer the cellular composition of a tumor by identifying the relative levels of RNA messages, or transcripts, expressed by the cells — first in bulk and then, almost incomprehensibly, at the level of individual cells. Suddenly, the high-rise blueprint shows not just rooms and furniture but also people and what was on their minds.

That’s because, although most cells share a common vocabulary in the form of the genes encoded by their DNA, RNA messages are the genetic words a cell mutters to itself to accomplish a certain goal at a particular time. Single-cell RNA sequencing allows researchers to eavesdrop on these internal conversations.

Newman and his peers at Stanford Medicine have developed technologies that build on these earlier advances. One, CIBERSORTx, functions like an eerily accurate fortune teller, predicting the various cell types in a bulk tissue sample based on the relative abundance and patterns of RNA messages in the sample. Another, EcoTyper, builds on this prediction to determine what the cell types are up to (a condition called cell state) and which other cells they are interacting with. The information allows researchers to build a picture of complex cellular neighborhoods called ecotypes within tumor tissue that hint at how the tumor is (or isn’t) thriving.

“Spatial transcriptomics is a new technology that gives us information about gene expression and spatial location so we can understand the modular architecture of healthy and cancerous tissue,” said Newman, the Institute for Stem Cell Biology and Regenerative Medicine Faculty Scholar. “In ecology, a species changes its characteristics and behavior in response to its local environment. Cells do this as well.”

Most recently, another tool, CytoSPACE, developed in Newman’s lab, maps these neighborhoods to precise locations in the tumor tissue, while also assessing the activity of all of each cell’s 20,000 genes.

“Techniques to study the spatial biology of tumors … are changing our understanding of cancer. Now, we can not only see what cell types are in the tumor but who their neighbors are and the molecular interactions that allow them to communicate and sustain each other.”

Sylvia Plevritis, PhD, associate director of cancer AI at the Stanford Cancer Institute

“Many times, if you just look at tumors as a bag of cells, your ability to predict a patient’s prognosis is not great, even if you know how many of each cell type is in the sample,” said associate professor of pathology Michael Angelo, MD, PhD. Angelo developed a way to visualize the locations of up to 50 individual proteins in a cell using a technique called MIBI-TOF. “But if you can incorporate where those cells are in the tumor, those predictions become much better. And they don’t seem to have a whole lot to do with the tumor cells themselves,” Angelo said. “The much more important angle is how the nontumor cells are responding to the presence of the cancer.”

Importantly, the machine learning that drives each of these advances has no preconceptions about what it might find. By simply looking for patterns — this type of cell is likely to be found rubbing membranes with this other type of cell, but only when both are in a particular cell state, for example — the computers can identify interactions that defy expectations.

“When my lab started working with single-cell data of tumors, we kept finding fibroblasts coming up as really important,” said Plevritis, the William M. Hume Professor in the School of Medicine. “Fibroblasts are most known for creating part of the skeleton that cells sit in and are one of the most understudied parts of a tumor, so it is very interesting and exciting to study this association.”

Further studies in Plevritis’ lab found that fibroblasts at the leading edge of a lung tumor had properties that stimulated cancer cells to invade surrounding tissue, while the fibroblasts in the interior appear to be more tumor suppressive.

New tools allow for deeper probes
of archived cancer tissues and types

Taken together, these technologies have given researchers, including Matusiak and Moding, valuable insight as to why people with the same type and stage of cancers can have such different outcomes.

Matusiak compared the location and activity of immune cells called macrophages in breast and colon cancers with healthy tissue. Prior to her study, researchers identified macrophages in tumor tissue by the presence of a protein that appears universally on all macrophages. Matusiak used single-cell RNA sequencing data to identify additional proteins that appear on only a subset of macrophages. She then found antibodies to these subset-specific proteins and used them to probe slides of tissue from colorectal and breast tumors.

She learned that macrophages are found in five distinct and very different cellular neighborhoods, or niches, within the tumors and that the macrophages were acting differently in each location.

“This was a big surprise,” Matusiak said. “We were definitely not expecting to see such distinct and separate spatial regions.”

For example, macrophages with a protein called IL4I1 on their surfaces were found in regions of high cellular turnover in both healthy and cancerous tissue — gobbling dead or dying cells. T

he presence of this class of macrophages correlated with a good response to immunotherapy in breast cancer patients and more favorable outcomes in people with colorectal cancers. In contrast, although macrophages with a protein called SPP1 were associated with tumor cell death, their presence in colorectal tumors correlated with poor outcomes.

“Now we have the first tools to really investigate macrophage biology in different tissues and cancer types in archived human tissue, including ductal carcinomas in situ,” Matusiak said.