Imagine walking into the doctor’s office in search of a diagnosis. Instead of the typical work up done today, the physician orders tests that use the local hospital’s newly acquired spatial transcriptomics platform. The next day, she delivers a diagnosis, along with a finely tuned treatment plan—both made possible because of the ability of this new technology to read and analyze the gene expression profile of any particular tissue. The process was quick, easy, and highly effective, exhibiting an unprecedented level of precision medicine.
Using spatial transcriptomics in the clinic as a diagnostic, or precision medicine, tool is “everybody’s dream,” said Jasmine Plummer, associate director at the Applied Genomics, Computational & Translational (AGCT) Core and a research scientist at Cedars-Sinai in Los Angeles, California. It’s a dream she thinks will eventually become reality. But the jump from basic research to the clinic will be challenging for several reasons. For one, basic researchers can “leave their findings in the land of ‘this is cool and important’,” Plummer said. But making a medical diagnosis requires a higher level of responsibility.
Spatial transcriptomics, and more broadly spatial biology, is one of the hottest new technologies in genomics. The advent of single-cell genomics and single-cell RNA seq (scRNA-seq) enables researchers to parse out gene expression at the single-cell level. But to do the experiments, the cells need to be pulled apart (dissociated) from each other.
Spatial transcriptomics allows for the profiling of single cells while they remain within their tissues. This enables researchers to reveal that a cell is expressing gene X when (and only when) it’s located near another cell. And this spatial information, which researchers have not been able to access before, is rapidly expanding the knowledge surrounding cells, tissues, and diseases.
While this technology is being rapidly adopted in the research lab setting, led by companies such as 10x Genomics, Nanostring Technologies, and others, its ability to move into the clinic remains unknown. Some in the field, including those companies working to commercialize spatial transcriptomics platforms, have confidence that it will. But not everyone is convinced.
Clinical OMICs spoke to three spatial experts sitting at the interface of research and the clinic. Despite their agreement that spatial is not ready for the clinic right now, the consensus is that new technologies must start somewhere—and that spatial is off to a great start.
Great promise
Many scientists think spatial analysis is new because of the emergence of spatial transcriptomics, explains Evan T. Keller, professor of urology and pathology at the University of Michigan School of Medicine. But spatial analysis, he argues, has been done since people have been looking through a microscope.
Keller recognized the explosion of spatial transcriptomics a few years ago. Since then, he has been building a program at the University of Michigan called the Single Cell Spatial Analysis Program (SCSAP), which launched this summer under his direction with six researchers, an annual symposium, and a monthly seminar series. Keller has devoted considerable attention to understanding the capabilities of the different spatial technologies, both from the preclinical or translational side, to figure out how they could be used in the clinic. There is, in his opinion, “great promise there.”
Spatial research to date has focused largely on creating maps or atlases. But, Keller said, that is not where the true power of the technology lies. Rather, it is in analyzing how cells or signaling pathways are interrelated. This is where the technology starts to bridge into translational. For example, having a deeper understanding of how immune cells and cancer cells interact could lead to a better understanding of how well a cancer patient responds to immunotherapy.
Keller, who works on prostate cancer, thinks spatial could play a role in tumor grading, or understanding how aggressive a tumor is. Typically, a pathologist looks at tissue specimens under the microscope to identify cellular changes that would indicate the presence of cancer. In prostate cancer, needle biopsies, which capture a small part of the tumor (less than 1%) are performed. But sometimes—about 20% of the time that a low-grade diagnosis is made—oncologists discover an aggressive tumor when the full prostate is removed. This error, Keller says, could potentially be avoided by adopting a spatial signature of the tumor microenvironment. In that case, spatial may allow for the detection of an aggressive tumor signature even if the needle biopsy misses the tumor. Although the signature is not yet known, Keller’s lab is using spatial to elucidate it.
More than gene lists and maps
At Cedars-Sinai, Plummer suspects that her specialty of neuroscience will be one of the last fields to adopt spatial in the clinic. Brain tissue is hard to come by (human brain biopsies are typically limited to autopsies or, in some cases, glioblastomas.) And sheer complexity of the brain is far beyond other tissues. Although scientists talk about the heterogeneity of a tumor with hundreds of cell types, the brain by comparison has millions.
But this is what makes spatial so potentially important when studying the brain. There are a lot of genetic and genomic data that need to be spatially understood. And understanding the connections between the cells, Plummer asserts, is of paramount importance. Nothing matters more than which cell sits beside another, she explained. Currently, they are “working with broken pieces.”
Neuroscientists are making progress producing maps of the brain. But how do they go from basic maps to clinical applications? Plummer said it’s a “numbers game.” Maps are made from small sample sizes—too few to assemble the knowledge that would be used to diagnose a patient. “We have to throw numbers at it,” she asserted. You need large numbers of patients—and access to the right type of tissue—to convince a clinician to use it for diagnosis.
A pathologist still uses markers developed 30 years ago—cheap, basic stains—to diagnose cancer. Moving spatial into this area is a “dollars and cents” game. How much weight should be put onto a technology that is, she says, all about discovery at the moment? To make the big leap from discovery to clinical diagnostics, Plummer said scientists must do more than simply develop gene lists. Rather, they need to start talking to their clinical partners about what is needed to move spatial into the clinic.
Somewhere in the middle
While many researchers are optimistic about spatial becoming clinically useful, others are more skeptical. Prakash Ramachandran, a clinician scientist in the Centre for Inflammation Research at the University of Edinburgh in Scotland, told Clinical OMICs that he is “somewhere in the middle”.
Ramachandran, a hepatologist, sees how spatial could be used, but he’s not sure that the technology, at least as it exists today, is ready. In liver disease, he noted, “we’re still a bit away from that.” If a spatial profile can be associated with a patient’s prognosis, then spatial could be used as a much-improved method for stratification compared to the current approach. But that data set does not yet exist.
The Ramachandran lab performs single-cell sequencing to understand cell heterogeneity in diseased livers and to better characterize which cell types and pathways are present in the diseased areas of tissues with a view to understanding pathogenic mechanisms and therapeutic targets. One of their challenges is the loss of topographical information held in scRNA-seq data. Spatial helps them better determine ligand-receptor interactions in the diseased areas of tissue.
One of the challenges of single-cell experiments is that the tissue frequently comes from patients with advanced disease, thereby obscuring information about early stages and the evolution of the disease. This is one place where spatial could play a role. Archival formalin-fixed paraffin-embedded (FFPE) tissue samples include liver disease at a range of stages. Being able to access that information might uncover targets that could be used to identify patients before they end up with advanced disease.
Currently, stratification and prognostication of patients with liver disease is based on pathologists’ scoring of a section stained with H&E and a fibrosis stain. The only metric that correlates with patient outcome is the amount of fibrosis. But liver disease can be tricky. While two patients may have similar pathological scores on their biopsies, one might progress to advanced disease quickly while the other could have stable disease for years. Spatial may be able to differentiate between these two outcomes.
Right now, the list of genes needed to distinguish any of these possibilities does not exist. But it may not be too far away either. Researchers have examined bulk gene expression patterns on biopsies of the liver in NASH (Non-Alcoholic Steatohepatitis)—the liver manifestation of obesity and metabolic syndrome. But even if the gene expression profile could be correlated to what the pathologist sees on the slide, there is still a gap in linking gene expression to patient outcome.
Cancer, Ramachandran predicts, will likely be the first field to bring spatial into the clinical setting. Keller agrees, noting that immunotherapy is most likely to be first frontier. In cancer, omics is not an alien concept as it may be viewed in other specialties; omics-type approaches are already used to stratify cancer patients. Spatial would be a natural progression from where the field is now. Keller predicts we’ll see clinical trials within two years, and first use within five. Plummer also thinks that spatial will get there, but the development work needed to facilitate that move is “not a two-year process.” To make it happen, Ramachandran believes that clinicians must come along for the journey. Ultimately that power of persuasion will require a significant amount of hard data.
The hurdles
Ramachandran notes three major hurdles that need to be navigated for spatial to reach the clinic. First, spatial platforms need to be well interfaced with health providers. There is a big gap between clinicians in healthcare, such as pathologists, and researchers in discovery biology. Clinicians not only have to buy into spatial, they also need the infrastructure to be able to perform spatial on their slides.
Second is cost. There must be data showing that spending a few thousand dollars on a spatial profile would change patient management and improve clinical decision making. Third, the bioinformatics portion of the workflow, not to mention managing the sequencing data, must be streamlined.
For clinical adoption, Keller says, the technology needs to be plug and play. Ideally, this would include an AI component that can correlate visual slides and omics and a bioinformatics platform to perform analysis of the data. Plummer agreed that cost must be considered when cancer can currently be diagnosed using an H&E-stained slide—a $10 analysis.
However, Keller does not think that the cost is a big barrier. When talking about chronic diseases, he said, the cost of spatial is “nothing compared to someone spending a $500,000 for cancer treatments.” Although expensive right now, Keller says that if spatial is shown to impact clinical therapies, helping the patient get the right therapy and/or saving patients from unnecessary therapies, it will ultimately prove to be very cost effective.
An even bigger issue than cost is what clinical pathologists—the people running the assays, looking at the slides, and making the diagnoses—need in order to trust the assay. It is only in bringing the clinicians and researchers together, so that each side can understand the other, that spatial will be able to develop escape velocity to make the leap into the clinical space.
