NEW YORK – A team led by researchers at the University of Copenhagen and the Max Planck Institute of Biochemistry has developed a spatial proteomics workflow that combines laser microdissection and mass spectrometry-based protein analysis.
Described in a study published last week in Nature Biotechnology, the approach allows for the characterization of thousands of proteins in small numbers of cells while retaining spatial information. It could prove useful for a variety of purposes ranging from basic research into disease processes and cellular heterogeneity in conditions like cancer to clinical pathology, said Andreas Mund, an associate professor at the University of Copenhagen's Novo Nordisk Foundation Center for Protein Research and the first author on the study.
Called deep visual proteomics (DVP), the method uses artificial intelligence-based image analysis to identify cells or cellular regions of interest, which are then cut out by automated laser microdissection. It then uses a single-cell proteomic sample prep and mass spectrometry workflow developed in the lab of Max Planck and University of Copenhagen researcher Matthias Mann, the senior author on the study, to analyze these samples.
The spatial proteomics field has exploded in recent years, with companies like Akoya Biosciences, Standard BioTools (formerly Fluidigm), and Ionpath bringing to market platforms capable of measuring dozens of proteins in tissue samples with single-cell resolution. The DVP method extends such spatial analyses by making it possible to look at thousands of proteins, allowing researchers to combine in-depth functional characterizations of cells with spatial information.
Mund said he has worked extensively with existing spatial proteomics platforms, using them to identify cellular phenotypes and looking at their spatial arrangement in tissues of interest.
"Then the question became, how can we now functionally characterize these different populations?" he said. "How do we get the molecular signature of these cells?"
Instead of defining cell phenotypes using proteomic data, DVP uses conventional staining and AI-based image analysis to identify phenotypes of interest and then follows that up with in-depth mass spec profiling of those cells to collect functional information.
"We changed the order," Mund said.
A number of technologies had to come together to enable the workflow, he noted. To develop the AI-based tools for identifying cellular phenotypes, the researchers built on an existing AI approach for isolating cell nuclei and cytoplasm fractions, using synthetic microscopy images to generate the dataset needed to train their tools.
They also had to develop an interface that would enable a laser microdissection microscope to automatically cut out the tissue portions and cellular features of interest identified by their AI-based analysis of a scanning microscopy image.
Following excision, cells of interest are processed using a protocol for extremely small sample volumes previously developed by the Mann lab and analyzed using the PASEF mass spectrometry workflow on Bruker's timsTOF SCP, a version of the company's timsTOF system optimized for high-sensitivity analyses of very small sample sizes, down to single cells. Bruker developed the system in collaboration with the Mann lab and launched it commercially a year ago.
While the system is capable of making measurements at the single-cell level, Mund said the researchers typically use around 100 cells per phenotype "because that is when we start to see robust phenotypes."
"You want to be sure that the phenotype is not just a preparation artifact or something," he said. "You want to see something robust and representative."
Mund said the automated laser dissection system can cut out around 1,200 cells per hour, or handle roughly 12 samples worth of material. Depending on the length of the liquid chromatography gradient used in the mass spec analysis, the researchers are able to analyze around 12 samples per mass spec per day.
In the Nature Biotech study, they used the system to look at cellular changes linked to melanoma progression, profiling the proteomes of several different, spatially resolved, cellular phenotypes and identifying changes in immune signaling and cell metabolism linked to development of the disease.
Mund said that a large proportion of the Mann lab is currently working on the DVP project "because of its great potential."
"We are applying it to almost every disease you can imagine where heterogeneity and spatial context matters," he said, adding that the group has established collaborations with pathologists at different hospitals to explore use of the method.
"We provide an extension of what pathologists are already doing," Mund said. "We look through the microscope, we make images, and then we have the AI part and the digital pathology, which many hospitals are using now. And then the extra step we have is that once we identify the interesting cells, we cut them out and put this additional unbiased [analysis] on top."
"It is still early days, but the people who are involved are very excited," he said. "We are getting flooded with requests."
Lise Mette Rahbek Gjerdrum, a researcher and pathologist at Zealand University Hospital and Copenhagen University Hospital and an author on the Nature Biotechnology paper, said that she and her colleagues are already using the system to help guide treatment in a lymphoma patient who presented a particularly complicated case.
The patient has two subtypes of lymphoma in the same lymph node, and Gjerdrum and her colleagues are using the DVP platform to determine if they are clonally related or not. It's a question with straightforward clinical consequences, Gjerdrum said. If the lymphomas are related, the patient will need a bone marrow transplant, but if they aren't, she can be treated with standard chemotherapy.
As clinical questions go, this is a rare one, Gjerdrum noted. She said that more generally, she envisions the platform as being useful for investigating tumor heterogeneity in cancer patients, with the ability to measure thousands of proteins in different cancer cell populations potentially giving researchers and clinicians a more fine-grained look at the activity of the various clones that make up a patient's cancer.
She is currently using the platform to study interactions between tumor cells and inflammatory cells in the tumor microenvironment in lymphoma as well as the proteomic changes that occur in lymphoma cells during the shift from early-stage to late-stage disease.
"It's fantastic," said Stanford University professor Garry Nolan, a leader in the spatial omics space and cofounder of Akoya Biosciences. "This kind of discovery-based science will help people target the important proteins that are reflective of some sort of disease state that needs to be managed."
"I think especially the open proteomics approach, as with genomics, helps us better understand mechanism," he added.
Nolan, who was not involved in the DVP work, suggested that an ideal next step would be to package the approach into a more streamlined workflow, for instance with the imaging and laser capture combined together in a single instrument.
"That's how progress is made," he said. "You bring together disparate pieces and show that it's useful, and then somebody decides that it's worthwhile putting it into one package."
In addition to the Bruker timsTOF SCP mass spec, the DVP researchers used Zeiss microscopes and slide scanners for tissue imaging and a Leica laser microdissection system for excising the cells of interest.
Mund said he sees room for improvement across the workflow, ranging from the microscopy hardware to the mass spec sample prep and analysis to the data analysis and processing steps.
"We have a functional pipeline, but we don't want to stop there," he said. "We want to improve every step and ultimately have a very robust and stable pipeline that can be used in the clinic."