Tumor Dynamics: Predicting Cancer’s Trajectory Using Tumor Atlases
Cancer is a dynamic disease that is unique to each patient. Tumor development and progression are complex, involving factors in the cancer cells themselves as well as multidimensional interactions between other cells and tissues in the body, which are shaped by a person’s genetics and cumulative exposures. As cancer progresses, tumors typically become more heterogeneous, composed of diverse cancer and noncancer cells with different molecular characteristics and behaviors. This staggering complexity makes it difficult to predict how a person’s cancer will progress and respond to treatment.
NCI is addressing this challenge by supporting research that is accelerating our understanding of the dynamics of tumor progression. For example, as part of the Cancer Moonshot℠, scientists are creating 3-D reconstructions of tumors and their evolution through the Human Tumor Atlas Network (HTAN), which consists of 10 interdisciplinary research centers across the country. HTAN brings together information about the cellular and structural makeup of a variety of adult and pediatric tumor types and premalignant abnormalities down to the single-cell and molecular level. In addition, NCI supports research to build and test new computational approaches to predict cancer development, progression, and response to treatment.
Technological advances, such as genetic sequencing of tumors at the single-cell level and spatial analysis of gene expression patterns across cells within a tumor sample, are enabling scientists to study tumors at unprecedented resolution. With these advances, researchers are exploring each tumor’s set of genetic and molecular characteristics, including in the cells and molecules surrounding cancer cells, to create maps of a tumor and its microenvironment.
While each tumor is unique, scientists can begin to understand which genetic and molecular characteristics are linked to tumor behavior by comparing tumor data maps to each other and to patient outcomes. However, this requires processing and interpreting a lot of data. New mathematical approaches and computer programs—including artificial intelligence (AI) and deep learning—can be used to analyze tumor data and enable researchers to better define key transitions during cancer progression, such as premalignancy to cancer and primary cancer to metastasis.
Ending cancer as we know it includes a future where we can predict a tumor’s trajectory based on a detailed profile of each patient’s disease. Additional research investments coupling tumor atlases with advances in computer science and molecular techniques will ultimately enable precision prevention, treatment, and care strategies for patients.
Using Computer Science and Artificial Intelligence to Study Tumor Progression
Recent advances in data access, computing power, and AI methods are enabling a better understanding of tumor progression, from premalignancy to early cancer and from less to more aggressive cancer.
AI is no longer just fodder for science fiction stories. Instead, cancer scientists are using AI and machine learning to make sense of the large amounts of data being generated on the dynamics of tumor evolution. As noted above, these types of data include detailed molecular characteristics and physical features of single cancer cells, tumor images and other spatial data, and clinical information over the course of diagnosis and treatment.
As one example, an NCI-funded collaboration between researchers at Indiana University and the NCI Intramural Research Program applied innovative and faster AI techniques to reconstruct key features of tumor progression, a computational process that can be prohibitively slow. The researchers used deep learning methods to understand the evolutionary history of a tumor using measurements from single cells. Notably, the researchers were able to construct large “family trees” of tumor cells to infer the history of how tumors evolve over time. In the future, this approach should enable scientists to better understand how cancer forms, progresses, metastasizes, and responds to treatment, which can lead to new therapeutic targets and approaches.
There are still major hurdles to address before AI will reach its full potential in cancer research. One major hurdle, called the “black box” problem, is AI’s lack of transparency. We don’t fully understand how the computer uses patterns in the data to make decisions. With additional investments in AI design and implementation, researchers hope to establish transparent and reliable computer programs that can use tumor data input to help predict clinical outcomes for patients when making treatment decisions.
Compiling Data on Tumor Heterogeneity with Cutting-Edge Technology
Large data sets that represent the cellular and molecular diversity of tumor cells are the ingredients for AI tumor analysis. Tumors consist of cancer cells as well as immune cells, blood vessels, fibroblasts, other cells, and components that interact with the cancer cells.
The molecular characteristics of cells in a tumor differ from patient to patient and even among tumors in a single patient. And these characteristics can change over time, including in response to cancer treatments. Understanding tumor cell heterogeneity, and collecting data that capture this heterogeneity, is critical because the extent of molecular differences among tumor cells might affect whether it grows or not (its stability), its ability to evade the immune system, whether and where it spreads (metastasizes), and its susceptibility to treatments.
The heterogenous mixture of cells in a tumor has multiple molecular and physical features that scientists can observe. Collecting and analyzing this combination of data that describes cells is called multi-omics.
Multi-omics, the process of collecting and analyzing a combination of data that describes cells, provides a way to study tumor cell heterogeneity, and can be performed at even the single-cell level. Some types of data that scientists collect include:
- genomics (DNA sequences in the cell)
- epigenomics (DNA modifications that affect whether a gene is turned on or off)
- transcriptomics (RNA sequences that cells use to make proteins)
- proteomics (the set of proteins in a cell)
- metabolomics (the products of cells breaking down proteins and nutrients)
Single-cell sequencing is a rapidly growing technique that allows scientists to compare the genomes and DNA products from each tumor cell. This effort builds on previous NCI-funded cancer genomic studies that relied on sequencing a whole tumor sample containing multiple cell types at a single point in time. Using single-cell technology, a recent NCI-funded study led by investigators at Memorial Sloan Kettering Cancer Center identified a subpopulation of small cell lung cancer cells—with unique genetic features and an immunosuppressed microenvironment—that is more likely to metastasize to other parts of the body. These findings have implications for developing future targeted immunotherapies for small cell lung cancer.
In the next frontier of cancer research, scientists are leveraging cutting-edge technologies to learn about the spatial anatomy of cancer and its molecular features. One such imaging technique uses repeated fluorescent labeling to identify a large set of cellular markers throughout a single biopsy. Imagine cutting a tumor into small sections and then using multiplex imaging to record many cellular and molecular features of each section at the single-cell level. Scientists are doing just that, and then assembling that information virtually to build a 3-D map of the tumor and its microenvironment. By mapping out heterogeneous tumor cells and their microenvironments, scientists are poised to identify cellular interactions and molecular characteristics that predict tumor behavior.
For example, NCI-funded researchers at Stanford University used spatial proteomics to characterize HER2-positive breast cancer collected before, during, and after treatment with neoadjuvant (presurgical) therapy. Typical tumor samples are an aggregate of tumor, immune, and connective or structural cells. By using spatial proteomics, they found that the number and kind of immune cells in the tumor changed during treatment for a subset of patients, and that increased levels of a single immune marker, called CD45, following initial treatment could predict response to the full course of therapy.
These results add to the groundwork for personalized medicine. Imagine tailoring a patient’s treatment plan based on their tumor’s molecular response to initial therapy, increasing the likelihood that treatment will lead to a favorable outcome.
In another study, NCI-funded researchers at the Oregon Health & Science University generated a comprehensive multi-omic analysis of a human cancer response to treatment. They analyzed blood and tumor biopsies from a patient with metastatic hormone receptor–positive breast cancer during multiple rounds of treatment. Samples included the primary tumor and three subsequent metastatic tumors over the course of 42 months. The researchers used the data to generate a spatial image atlas to identify how the patient’s tumors responded to treatment and to identify patterns of tumor progression and mechanisms of drug resistance.
These studies offer a peek into the possibilities for precision medicine in the future, in which multi-omics data will help doctors optimize treatment decisions for their patients. More investments are needed to develop the next generation of technology to perform spatial analysis dynamically, at a larger scale, and more rapidly.
Studying the Role of the Tumor Microenvironment
Just as plants and animals respond to the demands of their environments, tumors also adapt to environmental pressures. Cancer scientists are increasingly thinking about tumors from an evolutionary and ecological perspective. New research aims to reveal how a tumor’s microenvironment affects tumor growth, metastasis, and drug resistance.
Immune system function is an important factor in the tumor microenvironment. The immune system patrols the body to watch out for abnormal cells and can recognize and kill cancer cells—a capacity that has been harnessed by cancer immunotherapies. Scientists want to better understand how tumor cells interact with cells of the immune system so they can develop more effective immunotherapy and immunoprevention approaches. For example, researchers want to learn how and when precancerous tumors that progress to cancer, or cancerous tumors that recur after therapy, hide from the immune system.
In an NCI-funded clinical trial testing different drugs for early-stage HER2-positive breast cancer, doctors collected information about the tumor’s immune microenvironment with each therapy. They found significant changes in the composition of surrounding immune cells 14 to 21 days after treatment. They found that an initial increase in cancer-killing T cells decreased by the time of surgery. In the future, this type of information could help determine the best timing for immune system–based interventions during breast cancer therapy.
Now, tumor atlases are being compiled to study the tumor microenvironment’s role in the transition from premalignancy to invasive cancer—with an eventual goal to incorporate patient exposure data over time. This information could inform prevention and intervention strategies in the future.
For example, NCI-supported researchers at Stanford University and their collaborators observed a correlation between the tumor microenvironment and breast cancer development. Using surgical samples from patients, their study revealed that changes in the location and function of immune cells surrounding the tumor are associated with a shift from preinvasive abnormalities to invasive breast cancer.
Building on NCI-funded research on the tumor microenvironment, scientists have suggested tumor classification systems that may help doctors select the best course of treatment for individual tumors. These classifications are often based on the genetic makeup of the tumor. However, researchers have developed a classification system that is based on evolutionary and ecological principles, incorporating inherent traits of the tumor (such as tumor cell heterogeneity and rate of tumor change) and the ecology of the tumor’s surroundings (such as blood flow and immune system response). A classification system that includes tumor and tumor microenvironment traits could aid the goal of precision medicine.