NEW YORK – As the popularity of spatial proteomics continues to grow, a number of researchers are developing mass spectrometry-based approaches that allow for unbiased spatial analyses.
These methods provide significantly greater depths of coverage than the antibody-based technologies that have dominated spatial proteomics research to date but face technical challenges such as sample preparation and throughput.
Named the 2024 "Method of the Year" by the journal Nature Methods, spatial proteomics has generated substantial commercial and research interest, with a number of companies, including Akoya Biosciences (recently acquired by Quanterix), Standard BioTools, Bruker, Ionpath, Bio-Techne, Thermo Fisher Scientific, and 10x Genomics bringing spatial proteomic platforms to market.
These systems are largely antibody-based, meaning researchers are limited to the reagents available for each platform. Typically, experiments top out at around 50 to 100 protein targets.
Unbiased mass spec-based approaches, on the other hand, can measure thousands of proteins per experiment. Florian Rosenberger, an assistant professor in the department of medical biochemistry and biophysics at Karolinska Institute, said that the deep visual proteomics (DVP) method he helped develop while a postdoctoral fellow in the lab of Max Planck Institute of Biochemistry researcher Matthias Mann yields quantitative data for up to 8,000 proteins per sample.
Ruijun Tian, a professor at China's Southern University of Science and Technology (SUSTech), said that his lab is able to measure around 3,000 to 4,000 proteins using its Spatial and Cell-type Proteomics (SCPro) method, which, like DVP, is an unbiased mass spec-based approach.
In a January paper in Nature Communications, a team led by researchers at the Pacific Northwest National Laboratory demonstrated the ability of their wcSOP spatial proteomics approach to measure up to 4,600 proteins in human spleen tissue.
Scientists at the State University of New York-Buffalo have mapped roughly 5,000 proteins in mouse brain tissue using a spatial proteomics method developed by SUNY-Buffalo professor Jun Qu called micro-scaffold assisted spatial proteomics (MASP).
Generally speaking, unbiased spatial proteomic approaches involve extracting and analyzing proteins from individual cells or small numbers of cells in tissue in such a way that those proteins can be traced back to the specific cells and their locations within the tissue. In recent years, advances in areas like artificial intelligence and laser microdissection have allowed researchers to characterize and select cells for analysis in a more rapid and streamlined manner.
The DVP method, for instance, combines conventional cell staining with artificial intelligence-based image analysis to identify cells or groups of cells of interest. Researchers then cut out those cells using automated laser microdissection (LMD) and analyze their proteomes via mass spec.
SCPro uses multicolor immunohistochemistry (IHC) to stain centimeter-scale tissue sections, which are then analyzed using algorithms for nuclei and cell membrane identification. Single cells are then isolated using automated LMD and analyzed with mass spec.
The depth of coverage enabled by unbiased approaches provides "quite rich biology," said Rosenberger. It comes with the trade-off of limited throughput, however.
"At the moment, we can only reasonably measure 80 samples per day," he said. "If you want to map a full tissue [at single-cell resolution], that is going to take forever."
Rosenberger said his lab is trying to address this issue by developing methods for clustering cells of interest.
"You don't have to measure every single cell [individually], but you can group cells based on morphological features that then stratify the biology in the tissue in a meaningful way," he said.
Rosenberger cited the example of work he and his colleagues recently submitted for publication on alpha-1 antitrypsin deficiency, a genetic condition that can lead to lung and liver disease. The condition causes protein aggregates to form in hepatocytes, eventually killing the cells. In their study, the researchers imaged diseased tissue and used a convolutional neural network to cluster cells by whether or not aggregates were present and what structure those aggregates exhibited. They then cut out those distinct clusters and profiled the proteome of each.
"This gives us a new granularity of data … but it is not a full spatial mapping," Rosenberger said. "It is more a proteomic mapping based on histological features."
In their SCPro approach, Tian and his team used IHC staining, cell morphology, and location data to select cells of interest from the centimeter-scale tissue sections they analyzed. They identified "phenotype-matched" groups of 60 to 100 cells that they cut out and pooled for their mass spec experiments, allowing them to balance throughput, depth of coverage, and cellular and spatial resolution.
"With those three categories [of data], you can capture all your favorite tissue heterogeneity features," Tian said.
The SUNY-Buffalo researchers' MASP method takes a different tack, using 3D-printed microscaffolds to cut tissue slices into separate microsections, each of which is then processed and analyzed via mass spec. In the group's initial paper describing the approach, they used microscaffolds consisting of 900 separate 400-μm microwells, though they are working to develop scaffolds providing higher spatial resolution.
Sample prep is also a major challenge for unbiased spatial proteomics approaches given the extremely small amounts of material typically involved in these experiments. Some researchers have leveraged sample prep methods commonly used in single-cell proteomics like the nanoPOTS (Nanodroplet Processing in One pot for Trace Samples) approach developed by Brigham Young University professor Ryan Kelly.
Simply collecting single cells after they have been isolated via LMD is also difficult, said Tujin Shi, a senior scientist at Pacific Northwest National Laboratory (PNNL). Shi and his colleagues developed their wcSOP (wet collection of single microscale tissue voxels and Surfactant-assisted One-Pot) sample collection and processing method to help address these challenges.
PCR tube caps are commonly used for collecting small tissue sections, or voxels, cut out via LMD. Typically, the voxel is captured in the top of the cap and then transferred to the bottom of the tube for sample processing. Shi and his colleagues found in their study, though, that in many cases voxels are not effectively transferred from the tube cap to the bottom of the tube. Using a microscope to follow the path of voxels after capture, they found that they got stuck at many points around the PCR tube, leading to low reproducibility.
To address this issue, they developed an optimized voxel collection process in which the buffers required for sample processing are placed in the tube cap prior to voxel collection. The sample is then processed in the tube cap, and the resulting digested peptides are moved to the bottom of the tube using centrifugation.
The approach allows for simple and robust sample collection and processing "even for mass spec labs without much experience" in spatial proteomics, Shi said.
Mass spec sensitivity remains a limitation, Rosenberger said, particularly for measurements at the single-cell level. He noted that while his lab can measure around 4,000 proteins in large cells like hepatocytes, smaller cells like neutrophils are still a struggle.
"We can isolate [these cells] very well, but getting meaningful data out of them is tricky because there is so little material," he said.
Rosenberger said that recent mass spec releases like Thermo Fisher Scientific's Orbitrap Astral are helping in this regard.
"This is making a major difference," he said. "We are getting richer biological data from single cells nowadays. Inevitably there will be more technological development, and with better mass specs, we are going to get deeper [coverage]."
Tian noted that the formalin-fixed paraffin-embedded tissue commonly available for spatial proteomics studies is a challenging sample type for mass spectrometry. The SCPro method uses a solid-phase ion exchange-based protein aggregation capture (iPAC) workflow to improve extraction of proteins from these samples.
Tian's lab is now working to automate the SCPro workflow, using well-annotated cancer tissue samples to train AI agents to recognize regions and cells of interest and then cut them out using LMD.
"We hope it will become a very efficient approach to do this type of analysis," he said. "It's a very aggressive goal, but we're hopeful that in the next five years we can achieve this kind of technology."
Tian is also working on a proximity labeling-based approach for spatial proteomics that could allow researchers to label proteins in particular cell types and then extract them for mass spec analysis.
Proximity labeling typically uses a target protein to tag other nearby proteins with a molecule, often biotin, that allows them to be extracted from a cell and analyzed. Tian suggested the approach could be used to, for instance, tag proteins in all the cells in a sample with a particular cellular marker and then pull them out to be measured by mass spec.
"We could label, for instance, all the CD45-positive T cells on a whole centimeter-scale sample all at once," he said.
The approach is similar to the Microscoop spatial proteomics platform from Taiwanese firm Syncell, which allows users to biotinylate all the proteins in cellular regions of interest and then pull them down for analysis by mass spec or another method.
With the Microscoop system, researchers identify a cellular region of interest, either by tagging a molecular marker of that region via immunofluorescence or by identifying an anatomical marker located in that region. The system then images the tagged sample and creates an image mask defining the specific region to be targeted for protein extraction.
Users then apply to the sample what Syncell calls its Synlight Rich kit, which consists of a photobiotinylation reagent that binds to proteins when exposed to light. Following application of this reagent, a laser exposes the region of interest defined by the image mask to light, binding the photobiotinylation reagent to the proteins in this portion of the sample. Those proteins can then be extracted using a streptavidin-biotin pull-down workflow.
The company launched sales of the platform in 2023 and closed a $15 million Series A funding round in December of last year.
While it is still early days for these spatial proteomics workflows, they have begun to have clinical impact. Last year, Max Planck researchers including Mann used the CVP approach to identify a linkage between increased activation of the inflammatory JAK/STAT pathway and the sometimes-fatal skin condition toxic epidermal necrolysis (TEN). Rosenberger, who was a coauthor on the study, said the findings have been used to successfully guide treatment in several patients suffering from the condition.
While the spread of unbiased spatial proteomics approaches remains limited by factors like the technical complexity of the workflows and the high cost of the equipment required, Rosenberger said he is seeing growing interest in the space. He said that with colleagues he is organizing a workshop on DVP and other methods that will take place this autumn in Vienna.
"The community is growing, and it is now a good time to work on this more together," he said. "I think that for the spatial proteomics field this is a future avenue that will grow more and more."