When it comes to diagnosing, staging, and assessing cancer, pathologists have for more than a century relied on histology, which typically involves examining cells in stained tissue samples under a microscope to identify telltale features. A team led by researchers at Harvard Medical School (HMS) has now co-developed a tool with Seattle-based company RareCyte, called Orion, which comprises a powerful digital imaging platform that integrates information gained through traditional histology, with details revealed by high-plex immunofluorescence (IF) molecular imaging, on the same cells.
The ability to combine histology with molecular information using Orion could improve the way that pathologists see and evaluate a tumor, by providing more detailed clues about the tumor’s type, behavior, and potentially likely response to treatment. In newly reported research, Sandro Santagata MD, PhD, an HMS associate professor of pathology at Brigham and Women’s Hospital, and colleagues described a study in which they used Orion to analyze tumor samples and identify biomarkers prognostic of tumor progression.
For the study, the researchers looked at tumor samples from more than 70 patients with colorectal cancer. The Orion tool provided complementary histological and molecular information about each tumor sample, and identified biomarkers that were more common in patients with serious disease. These biomarkers consisted of specific combinations of tumor features, typically based on numbers and properties of immune cells and other cells, that predicted how patients with colorectal cancer would fare.
The researchers hope that with further refinement, Orion will improve the diagnosis and treatment of cancer and other diseases. Importantly, the more detailed information could reveal features and patterns that help scientists develop biomarkers to better predict how a disease will behave, allowing treatment to be optimized for each patient.
“This work is a critical next step for taking the principles about tumor features that we are developing in the research space and turning them into a tool that is actually useful in the clinic,” Santagama noted. Santagama, together with co-senior author Peter Sorger, PhD, the HMS Otto Krayer professor of systems pharmacology, and colleagues, described their work in Nature Cancer, in a paper titled, “High-plex immunofluorescence imaging and traditional histology of the same tissue section for discovering image-based biomarkers.”
Precision medicine is “critically dependent” on better methods for diagnosing and staging disease, and predicting drug response, the authors wrote. In cancer and other diseases, for example, the primary diagnostic method relies on a histopathology review of hematoxylin and eosin-stained tissue, “complemented by IHC and exome sequencing.” And while machine learning (ML) and artificial intelligence (AI) methods can automatically extract data from such images, “H&E and IHC images generally lack the precision and depth of molecular information needed to optimally predict outcomes, guide the selection of targeted therapies, and enable research into mechanisms of disease,” the team stated.
More recently developed, highly multiplexed tissue imaging methods are showing promise to enhance research studies and clinical practice with “precise, spatially resolved, single-cell data,” the authors further noted. Sorger, Santagata, and their team have spent the past several years developing and improving tools for imaging human tissue samples. In a recent study, the researchers combined a multiplexed imaging technique called cyclic immunofluorescence, or CyCif, with histology to map colorectal cancer. Their maps, which are freely available to other scientists online, provide an unprecedented level of detail about these tumors, including information about how they are arranged spatially, how they form and progress, and how they interact with various immune cells.
The scientists wanted to take their work further, by making the imaging tools designed in a research setting available to clinicians who spend their days looking at tumor samples under a microscope to gather information needed to diagnose and treat patients.
Study lead author Jia-Ren Lin, PhD, platform director in the Laboratory of Systems Pharmacology at HMS, collaborated with RareCyte to develop a digital imaging platform that can quickly collect and analyze both H&E and multiplex immunofluorescence images from the same tissue sample. “… we describe the development of an approach to one-shot, whole-slide, 16 to 18-channel immunofluorescence (IF) imaging, followed by H&E staining and imaging of the same cells,” the scientists explained in their paper.
In contrast with H&E staining, which highlights the key structural features that pathologists have traditionally used to determine how severe and advanced a cancer is, multiplex immunofluorescence uses fluorescent, tagged antibodies to label important molecular features such as types of immune cells and other cells. The result is a single digital image that integrates information from both techniques. “Orion fully merges the two modalities, so that you can go back and forth on a single slide and say, OK, I see that feature in H&E, and now I can figure out what molecular markers are also present. It’s remarkable,” Santagata said.
For their reported study, the team used Orion to analyze tumor samples from two independent cohorts of colorectal cancer patients, totaling 74 individuals. They found that the information provided by the new tool allowed them to identify a biomarker of outcome, or a specific combination of features that predicted which colorectal cancers were likely to progress and which were not. “Using a retrospective cohort of 74 colorectal cancer resections, we show that IF and H&E images provide human experts and machine learning algorithms with complementary information that can be used to generate interpretable, multiplexed image-based models predictive of progression-free survival,” the team stated.
In the context of a tumor, a biomarker often factors in how many and what types of immune cells and other cells are present in different parts of a sample. In this case, Orion identified a biomarker based on the presence of T cells marked with the proteins CD4 or CD45 and tumor cells that contained the proteins PD-L1 or α-SMA. The team showed that this and other Orion-based biomarkers performed as well as or better than an established clinical test, called Immunoscore, which is based on CD8 and CD3 T cells and is used by oncologists to assess colorectal cancers. “Combining models of immune infiltration and tumor-intrinsic features achieves a nearly 20-fold discrimination between rapid and slow (or no) progression, demonstrating the ability of multi-modal tissue imaging to generate high-performance biomarkers,” the investigators commented.
While Orion is still in the early stages of development, the researchers say that their results so far provide proof of principle that the platform can be useful in the clinic. “We anticipate many opportunities for joint learning from H&E and IF data using adversarial reinforcement, and other types of ML/AI modeling for research purposes, development of biomarkers, and analysis of clinical H&E data at scale.” And noting limitations of their reported study, they nevertheless concluded, “… we anticipate that it will be feasible to progress in a few years to validated clinical tests that can be added to colorectal cancer treatment guidelines, substantially improving opportunities for personalized therapy.”
The team plans to refine Orion by testing it on larger numbers of patients, and investigating the most useful combinations of antibodies. Scaling the platform to make it faster and cheaper will be key to moving it into the clinic, they said.
The team also plans to test Orion in other cancers, such as lung cancer and melanoma, and eventually move beyond cancer to conditions, such as kidney disease and neurodegenerative diseases, that could also benefit from dual histologic and molecular profiling to determine their stage and severity.
The researchers are especially interested in exploring how Orion can be used to identify new prognostic, and predicted drug response biomarkers, based on the molecular and structural characteristics of a sample. With Orion, Santagata explained, researchers can easily assess many biomarkers at the same time.
“This platform gives you the ability to search across the biomarker space for a wide range of combinations of markers that could be useful and select the ones with the best performance for a particular measure,” Santagata said. “We feel like we have a tool kit now that will allow us to find these in a relatively rapid manner across cancer types.”
If Orion can indeed find more useful prognostic and predictive clues for cancer, these biomarkers could help oncologists more accurately profile a patient’s tumor and develop a more tailored, individualized treatment plan. Additionally, the researchers expect that insights gleaned from Orion in the clinic would inform new directions for the basic science research they are pursuing in the lab.
The researchers envision Orion as something that pathologists can ultimately incorporate into their existing workflow, allowing them to add molecular details to their histological expertise to gain a more complete understanding of a sample. Notably, the digital nature of the tool means that pathologists can look at the images on any computer, rather than being confined to a microscope in a lab or clinic. “There are multiple ways to exploit the complementary strengths of H&E and IF imaging using ML approaches,” the authors suggested. As an example, they noted, “ML models trained on H&E data can increase the number of identifiable cells in multimodal images relative to multiplexed IF data alone. Conversely, IF images can be used to automatically label structures in H&E images (e.g., immune cell types) to assist in supervised learning on these images.”
Santagata continued, “H&E is like soul food; it’s comfort for pathologists, it tells us where we are. But now we can layer on molecular information, which is a very powerful capability … We’re trying to bring new tools to the group of people that are working hard every day to diagnose tumors and other diseases.”
Lin added, “Pathologists already do a huge amount of work with histology to diagnose a patient and understand their disease, but with this tool to augment their knowledge, they will basically have a ‘super view’ of the sample.”
The researchers hope that the tool will eventually help pathologists identify and quantify new features relevant to a disease. They think that Orion may also provide information that can be integrated into artificial intelligence tools being developed to aid pathologists and oncologists.“The future of diagnostic pathology is digital, and this work could not only transform how clinicians diagnose cancer, but also change how we train the diagnosticians of tomorrow,” said Jon Aster, MD, PhD, the HMS Ramzi S. Cotran professor of pathology and deputy chair of pathology at Brigham and Women’s.
And as for the questions that pathologists might have after being introduced to Orion, Santagata said, “Oh, it’s endless. It’s absolutely endless. This is a whole new world, and each person has their own long list of things they’d like to see. Every pathology field has a set of molecular questions that have remained unanswered, and now we have a tool that can start to provide some answers.”