StarTrak links somatic mutations to tissue geography in spatial genomics
Spatial genomics has made it possible to see where cells sit inside a tissue and what molecular programs they are running, but lineage has remained harder to pin down. Cells can share a transcriptional state and still arise from different ancestral branches, and mutation signals are often sparse enough to vanish in routine analyses. The preprint introduces StarTrak as a way to recover those hidden histories by combining somatic variant calling with spatial context at single-pixel resolution.
The framework is built around a simple biological premise: migration and clonal expansion leave footprints in DNA even when RNA profiles look similar. By integrating mutational evidence from both nuclear and mitochondrial reads with spatial proximity, histological images, and transcriptomic states, StarTrak attempts to identify patterns of mutations that are informative for lineage reconstruction rather than random noise. The design is meant to address a familiar problem in tissue genomics: dropout can erase the very variants needed to track where a cell came from and where its descendants moved.
One of the central claims is that the method performs across tissue types and across spatial-omics platforms, with the strongest performance in spatial ATAC-seq. In practical terms, that matters because chromatin accessibility data can sometimes preserve signals that are difficult to capture with other readouts, while also giving the model another layer of context when mutation calls are incomplete. The computational approach is not merely searching for isolated variants; it is using the surrounding tissue architecture and molecular state to decide whether a mutation belongs to a spatially coherent lineage.
Tumor sections reveal lineage splits that transcriptomics alone misses
Applied to tumor sections, StarTrak inferred migration of mutated tumor cells into adjacent non-malignant tissues. It also reconstructed distinct lineage trees associated with different spatial regions, showing that the same tissue block can contain ancestry patterns that diverge more sharply than the transcriptome suggests. In one of the most consequential observations, the analysis identified tumor regions with nearly identical transcriptomic states but distinct mutational profiles.
That finding has obvious biological weight. Tumor cells can converge on similar expression programs under shared microenvironmental pressures while still carrying different ancestral mutations, which means a flat readout of RNA can miss clinically relevant clonal structure. A region that looks homogeneous at the transcript level may still harbor multiple evolutionary branches, each with its own capacity for invasion, survival, or adaptation to local stress. In spatial oncology, that distinction can change how investigators interpret the apparent boundaries of disease.
The migration signal extended beyond the tumor compartment itself. StarTrak inferred movement of mutated cells into nearby non-malignant tissue, implying that histologically normal-appearing areas can contain cells with a malignant or premalignant lineage signature. This is the kind of result that makes spatial context indispensable: a mutation found in one neighborhood is not only a marker of identity, but also a clue to how cells travel through tissue over time.
A single transcriptomic profile can hide multiple ancestral routes.
Sunlight-associated epidermal mutations point to deeper skin movement
The skin analysis adds another layer to the picture. In non-malignant epidermis, StarTrak identified sunlight-associated mutations and evidence suggesting migration of these mutant surface cells into deeper skin layers. The biological logic is plausible: ultraviolet damage accumulates in the epidermis, where surface cells are exposed first and most intensely, and clonal descendants can then be tracked if spatial methods are sensitive enough to retain their mutational signatures.
This result moves the method beyond cancer alone. It suggests that somatic evolution in normal tissue, including environmentally induced mutation, can be mapped in situ rather than inferred indirectly. For dermatology and tissue biology, that raises the possibility of studying how mutated clones spread within otherwise non-malignant compartments, how they persist, and how local microenvironments shape their fate.
The preprint repeatedly returns to the same technical bottleneck: sparse data. Spatial genomics is powerful, but dropout makes direct lineage inference difficult, especially when mutations are rare or unevenly represented. StarTrak addresses that by pooling several weak signals into a single decision framework. When nuclear reads, mitochondrial reads, tissue position, microscopic appearance, and transcriptomic state all point in the same direction, the resulting lineage call becomes more credible than any one signal alone.
For now, the work should be read as a computational advance rather than a finished clinical tool. It is a preprint and has not undergone peer review, and the performance claims need independent validation across additional tissues, disease states, and assay conditions. The most important next step is likely to be testing how reliably this kind of integrated model can separate true clonal migration from technical artifacts, especially in samples where tissue structure is distorted or mutation burden is low.
Even so, the conceptual shift is clear: spatial genomics is no longer limited to mapping where cells are and what genes they express. With methods like StarTrak, it can begin to reconstruct who cells are, where they came from, and how far their descendants have moved through tissue architecture.