CytoSignal Estimates the Locations and Temporal Changes of Cell-to-Cell Signaling Using Spatial Transcriptomics
Background: Locating where cell-to-cell signals act
CytoSignal is a computational tool that estimates the locations of ligand–receptor signaling at the cellular level from spatial transcriptomics data. Its extension, VeloCytoSignal, also predicts changes in signal strength over time. The researchers applied both methods to mouse tissue data and compared location predictions with protein-proximity experiments. The study was published in Nature Genetics on June 10, 2026.
Understanding cell-to-cell signaling requires knowing not only which cells express ligands and receptors, but also where those cells are located. Existing methods based on single-cell RNA analysis sometimes fail to account for spatial information or remain limited to inferring interactions between cell populations. Even among methods that use spatial transcriptomics, a limitation remained: contact-dependent signaling and signaling by diffusible ligands were treated in the same way. These two modes differ in the spatial conditions that allow signals to reach their targets.
Key findings: Combining expression-based scores with validation in tissue
CytoSignal calculates a ligand–receptor interaction score, or LRscore, by multiplying ligand expression that accounts for spatial neighbors by receptor expression. For contact-dependent interactions, it considers directly adjacent locations; for diffusion-dependent interactions, it applies distance-based weights to surrounding locations. Spatial permutation testing and false discovery rate correction identify significant locations. The tool can also explore spatial gradients of signaling and related genes. LRscore is an estimate based on expression data and is distinct from direct measurements of the amount of protein binding or signaling activity.
For location validation, the researchers used high-resolution spatial transcriptomics data obtained with Visium HD from adjacent sections of mouse embryos, together with proximity ligation assays (PLA). PLA produces a fluorescent signal when two target proteins are within 40 nanometers of each other. The researchers evaluated a total of five ligand–receptor pairs in sections adjacent to the two sections analyzed with Visium HD. CytoSignal showed consistent performance in region-level classification and outperformed existing methods in several regions. However, when the entire embryo was compared at once, correlations between predictions and PLA measurements were low for all methods.
The team also analyzed differences across samples. Comparing brains from four young and five aged gene-knockout mice exhibiting Parkinson’s-like pathology, they found significant differences in 42 diffusion-dependent and three contact-dependent interactions. The false discovery rate threshold was below 5%. Estimated activity of immune-related signals, including Spp1–Cd44, was higher in the aged group.
Implications and outlook: Connecting predicted rates of change to follow-up experiments
VeloCytoSignal combines RNA velocity for ligand and receptor genes to predict whether signaling scores at each location will increase or decrease. RNA velocity estimates the rate of change in gene expression from transcript information before and after splicing. The researchers separately analyzed data from embryonic days 10.5, 11.5, 12.5, and 13.5, then compared predicted trends with data from subsequent time points. Increased Alb–FcRn signaling in the liver and changes in Wnt5a–Antxr1 signaling in the jaw and tooth regions were consistent with trends in signaling scores calculated from successive time points. These findings did not come from directly tracking protein signals in real time within the same tissue.
CytoSignal requires a gene expression matrix and spatial coordinates; VeloCytoSignal additionally requires RNA-processing information. Both methods have been released as R packages. The study provides an analytical basis for narrowing down the signals and locations to validate in follow-up experiments. The tumor microenvironment and repair processes in injured tissues were proposed as future applications. Diagnostic performance in patients and therapeutic effects were not evaluated in this study. The study design and findings are available in the Nature Genetics paper.
Nature Genetics, Published online: 10 June 2026. DOI: 10.1038/s41588-026-02624-9
Summary: Bypassing the low predictive velocities and spatial structure stripping constraints that historically cloud empirical cell-to-cell communication profiling in heterogeneous niches, this multi-omic translation scales a programmable spatial computing infrastructure termed CytoSignal. Synthesizing cellular-resolution spatial transcriptomic matrices with mathematically calculated RNA velocity vector equations, the computing platform establishes continuous tracking of homeostatic ligand–receptor interactions across discrete tissue coordinates concurrently. The model deciphers the precise mathematical covariance linking multi-layer transcriptional splicing kinetics to real-time signal propagation velocities, effectively mapping unexpected dynamic activation hubs inside complex pathological microenvironments. This computational calibration delivers a validated, non-invasive baseline to isolate raw directional variance, optimize programmatic target drug screening, and guide prospective universal single-cell stratification under digital genomic governance.
Researchers analyzing spatial transcriptomics can use CytoSignal to identify tissue locations where particular signals are estimated to be prominent and prioritize follow-up experiments by distinguishing contact-dependent from diffusion-dependent signaling. For example, they could select ligand–receptor pairs that vary by developmental stage or age, then test the predictions at those locations with protein-level experiments such as PLA. When RNA-processing information is available, VeloCytoSignal can also predict the direction of change for comparison with tissues from other time points. However, expression-based scores and region-level location validation alone do not establish causal relationships in signaling or the validity of a drug target. Analysis of the tumor microenvironment is likewise a prospective research application requiring further validation.