Skip to main content


Pixelgen Customers Begin Collecting Data on Spatial Layout of Cell Surface Proteins

May 17, 2024 | Andrew P. Han

NEW YORK – Customers of Swedish proteomics startup Pixelgen are starting to collect data using the firm's kits to create spatial maps of surface proteins of single cells.


"Saying [our pilot project] went really well, that's an understatement," said Ankur Sharma, a cancer immunotherapy researcher at Australia's Harry Perkins Institute of Medical Research. Sharma is in the process of starting a new lab and turned to Pixelgen about a year ago after hearing about the technology at a conference.


His team has run 16 samples of patients with liver cancer who received immunotherapy. "The big question I work on is why some patients do not respond to therapy," he said, and Pixelgen's data on protein interactions are beginning to help them understand why that might be happening.


Sharma is one of several customers who are getting ready to put Pixelgen data into the literature, joined by Swedish personalized cancer therapy firm Neogap and others.


Earlier this month, the firm published its own proof-of-concept paper in Nature Methods. Its approach uses antibody-oligonucleotide conjugates to detect proteins, then links the barcoded oligos — the so-called pixels — so that a sequencer can read out which proteins were next to each other. The data depict the cell surface at resolution greater than 100 nm, the firm said. In the paper, Pixelgen researchers showed how the proteins formed more than 1,000 "spatially connected zones" per cell in 3D. Their study created protein networks for 76 targets in around 500 cells.


Pixelgen was founded in 2020 by Simon Fredriksson, a veteran of Olink, another Swedish proteomics firm that Thermo Fisher Scientific is in the process of acquiring for $3.1 billion. It started with $6 million in seed funding and closed a $7.3 million Series A funding round in October 2023.


Fredriksson, who serves as Pixelgen's CEO, said the firm is not actively seeking more funding at the moment. It now has customers "across all geographies and segments of the market," he said, and has moved into new facilities with twice as much space as before.


Last year, the firm told GenomeWeb that its kits would initially target 76 proteins in about 1,000 cells per reaction, but it has not yet disclosed pricing for the kit, which contains eight reactions.


Sharma said the kit worked very well in his team's hands, adding that this is not always the case with a new technology. "That's important because if something is working well, you don't have to worry from one run to another. It means technical errors are at a minimum, and what we're seeing is more biological." He noted that his team captured around 900 cells per sample, for a 90 percent recovery rate.


In the Nature Methods study, the Pixelgen researchers were able to estimate cell populations, with replicate samples showing concordant percentages of T cells, NK cells, B cells, and monocytes. Pixelgen's technology generated an average of 1,737 spatial zones and 9,580 unique molecular identifiers per cell. "Furthermore, cell population frequencies from MPX were consistent with those observed with flow cytometry, with similar but varied signal to noise for each marker," the authors wrote.


Using their graph representations of the data, they were able to identify both known and unknown patterns of spatial organization on chemokine-stimulated T cells, showing how this could help define cell states through cell surface protein arrangement.


One of the applications of this data type is stark visualizations of proteins in 3D. In their supplemental material, the authors posted a video of a cell exhibiting a structure called a uropod, an important feature of migrating immune cells. Uropods draw their name from the appendages that help make up the fan in a lobster's tail.


Another data type is a protein colocalization score, which "reflects the degree of spatial co-occurrence of two proteins by quantifying the deviation from what would be expected by random chance (minimizing the influence of bias from experimental perturbations), as well as protein abundance," the authors wrote, noting that a negative colocalization score may provide insight as "a reflection of segregation of protein markers."


Sharma said the data has been helpful as an additional layer that goes a level deeper than single-cell gene expression. While methods like CITE-seq can also pair cell surface proteins with transcriptomics, they can't say that a given protein was not only expressed but also interacting with other proteins.


"Now, once we know what it means functionally, maybe that's how we figure out what to do with the non-responder population," he said, "so we can make them into responders."

Reprinted with permission GenomeWeb, a business unit of Crain Communications © 2024. All Rights Reserved.  #GW24003