Single-cell RNA transcriptomics allows researchers to broadly profile the gene expression of individual cells in a particular tissue. This technique has allowed researchers to identify new subsets of cells or new gene programs that have advanced the understanding of development, cancer, and normal biology. Standard single-cell RNA transcriptomics requires researchers to dissociate cells from their tissues for profiling. This process can lead to loss of certain cell types, and dissociation completely destroys the spatial organization of a tissue.
Understanding the spatial organization of single cells is important. This dimension lets researchers know which cells could potentially be interacting or communicating with one another in a tissue full of many diverse cell types. To solve this problem, researchers have combined tissue imaging with highly multiplexed quantification of 100s to 1000s of RNA transcript-specific probes to create a technique called spatial transcriptomics. “Tissues are complex, and we want to understand what programs the cells are running in their native environment. Spatial transcriptomics is really exciting because you look at tissues that are fixed in place and identify all the different types of cells…and learn about what transcriptional programs they’re running,” says Dr. Evan Newell of the Vaccine and Infectious Disease Division at Fred Hutch.
Since its inception, gleaning this information from spatial transcriptomic data has been challenging because methods for segmenting, or computationally separating single cells, frequently misidentify cellular borders. Historically, segmentation has been accomplished through antibody staining and microscope imaging of the tissue section. However, this technique is error prone and frequently misidentifies cell boundaries. “With antibody staining, it’s more of an art…there’s a lot of variation and background,” explains Newell.
Assigning expression patterns to single cells in their native environment is the entire point of spatial transcriptomics, so the fact that cells are misidentified using classic methods is a big problem. To tackle the issue of segmentation, Dr. Daniel Jones, a staff scientist in the Newell lab, built a new computational tool called Proseg to segment cells based on their RNA expression. First, he defines the number of cells in a particular tissue by staining and counting their nuclei. This data is then fed into a probabilistic model that defines cell boundaries based on what transcripts are present. His model takes advantage of the fact that cells behave like a “bag of RNA,” meaning that RNA transcripts are typically randomly distributed throughout the cell. From there, Proseg takes inspiration from the Cellular Potts Model to simulate cells that best explain the distribution of the transcripts.
Having the model in hand, the team next sought to test how well it performed against other segmentation tools in the field. In spatial transcriptomics, genes that are biologically implausible to be expressed at the same time in the same cell are frequently identified due to segmentation errors. To test Proseg’s performance, they decided to quantify the frequency of these suspiciously co-expressed genes across several segmentation models. They found that Proseg reduced the frequency of these “suspicious” gene pairs compared to every other model they tested, indicating that their tool improved segmentation relative to existing tools. Other quality control tests they ran also indicated that Proseg outperformed existing tools.
After validating Proseg, the team used the tool to quantify T-cell infiltration in renal cell carcinoma, a type of kidney cancer. Understanding how many and where immune cells are in renal cell carcinoma is important because immune infiltration is closely linked to treatment outcomes in this type of cancer. “After we re-segmented [the samples], we found that there were quite a few more T-cells than we originally thought in these samples. It was the case that their signal was being smothered by the surrounding tumor cells due to the poor segmentation,” explains Jones. These results highlight the potential for Proseg to expand scientists’ existing understanding of tumor biology through spatial transcriptomics.
Jones’s tool has been widely and quickly adopted by other researchers doing spatial transcriptomics. “Everyone at the Fred Hutch is using it,” says Newell of the tool’s reach. Groups outside of the Hutch have been quick to adopt this method, as well. To date, Proseg has more than 29,000 downloads, and Jones receives frequent requests to expand the tool to different spatial transcriptomics platforms. In the future, the team hopes to continue developing the tool to make it broadly applicable to many biological scenarios.
This work was supported by funding from the National Institutes of Health, the Immunotherapy Integrated Research Center, the Cancer Research Institute Irvington Postdoctoral Fellowship (to David Glass), and an American Society of Hematology Fellow Scholar Award.
Fred Hutch/University of Washington/Seattle Children’s Cancer Consortium Member Dr. Evan Newell contributed to this research.
Jones DC, Elz AE, Hadadianpour A, Ryu H, Glass DR, Newell EW. 2025. Cell simulation as cell segmentation. Nature Methods. 22(6):1331-1342. doi: 10.1038/s41592-025-02697-0.
Kelsey Woodruff is a PhD candidate in the Termini Lab at Fred Hutch Cancer Center. She studies how acute myeloid leukemia cells remodel the sugars on their membranes to reprogram cancer cell signaling. Originally from Indiana, she holds a bachelor's degree in Biochemistry from Ball State University. Outside of lab, you can find her crocheting and enjoying the Seattle summers.
For the Media