Single-cell protein mapping reveals distinct breast cancer cell states
Multiplexed immunofluorescence is pushing tumor pathology beyond morphology and into the biology of individual cells. In treatment-naive stage II/III breast cancer, investigators measured 27 proteins at single-cell resolution in formalin-fixed, paraffin-embedded tumor specimens and used those data to classify neoplastic cells into discrete protein coexpression states.
The scale of the analysis was substantial: 638,577 tumor cells from 26 breast cancers were interrogated after iterative staining cycles on routinely collected tissue blocks. Unsupervised clustering identified eight distinct protein coexpression clusters. In practical terms, the tumors were not simply more or less abnormal versions of the same disease. They were mixtures of cell populations with different proteomic identities, and those identities were often spatially organized within the same lesion.
One of the clearest findings emerged from tumor dominance patterns. In about one-third of the cancers, more than 95% of neoplastic cells belonged to a single cluster. The remaining tumors were more heterogeneous, with multiple clusters appearing either in separate regions or intermingled across the tumor field. Heterogeneity at that level matters because it implies that a single biopsy core may capture only part of the proteomic biology driving growth, invasion, or treatment resistance.
Proteomic clusters tracked with metabolic and receptor-related features
The clustering strategy was built around protein expression rather than DNA alone, which is clinically important. Genomic profiling can identify driver alterations, but it does not always explain how a tumor is behaving at the tissue level in real time. Protein state can reflect signaling activity, metabolic demand, and lineage plasticity more directly than static DNA sequence.
Among the clusters identified, some were associated with hormone receptor loss, PTEN alteration, and HER2 gene amplification. Those associations are biologically coherent. Loss of hormone receptor expression often signals a shift away from endocrine dependence, while PTEN disruption can release a brake on PI3K-AKT pathway signaling and favor aggressive metabolic adaptation. HER2 amplification can similarly support proliferative signaling and increase energetic demand inside the tumor cell.
The metabolic readout was especially notable. Tumor uptake of 18F-fluorodeoxyglucose was associated with protein expression clusters carrying those adverse biological features. FDG uptake is a surrogate for glucose avidity, and glucose avidity often rises when cells are driven by higher proliferative pressure, altered growth factor signaling, or metabolic rewiring. In this setting, the radiotracer signal did not float free of tumor biology; it aligned with specific proteomic states that are already familiar as aggressive molecular phenotypes.
A single tracer signal can now be connected to a defined cellular state.
Why the tissue architecture of heterogeneity matters
The study did more than catalog protein expression. It linked cellular states to their arrangement inside the tumor. Some tumors contained regional clusters, with one state dominating a particular zone of the cancer. Others showed intermingled cell populations, suggesting fine-grained coexistence of distinct proteomic programs in close proximity.
That pattern is clinically relevant for several reasons. Regional segregation can influence how oxygen, nutrients, and immune cells are distributed across the lesion. Intermingled heterogeneity may allow resistant cells to persist beside more treatment-sensitive cells, creating a substrate for clonal selection under therapy. Either configuration complicates the assumption that one biomarker readout from one tumor region fully represents the entire disease.
The method also makes use of tissue that is already part of routine pathology workflows. A single 3- to 5-μm unstained section from standard FFPE blocks underwent repeated imaging cycles, with background autofluorescence recorded first and then removed computationally after staining. Cell segmentation separated epithelial and stromal compartments using cytokeratin boundaries, while membrane, cytoplasmic, and nuclear regions were defined with Na+K+ATPase, S6, and DAPI signals. The resulting pixel-level information was condensed into cell-level metrics and queried with clustering methods.
A pathology workflow with translational reach
The most important feature of this platform is not simply that it generated large data. It transformed a conventional specimen into a map of proteomic behavior at the level of individual cancer cells. Because the tissue was archived FFPE material, the approach is compatible with real-world pathology samples rather than requiring fresh tissue alone.
The technical workflow also generated an enormous number of measurements, roughly 155 million in total across mean, standard deviation, and hotspot metrics for each marker. That density of data creates a foundation for future studies that may relate protein states to treatment response, recurrence risk, or resistance mechanisms. The current analysis stops short of those clinical endpoints, but it establishes a reproducible framework for measuring heterogeneity in a way that mirrors the biology seen by the pathologist under the microscope and the oncologist in the clinic.
For breast cancer, this matters because subtype boundaries are not always rigid. A tumor can retain hormonal features in one region while adopting HER2-related or metabolically active programs in another. Protein-state mapping offers a way to detect those shifts directly in tissue instead of inferring them only from bulk assays or from limited genomic surrogates.
FDG avidity may reflect a proteomic state rather than a single mutation
The association between FDG uptake and clusters defined by hormone receptor loss, PTEN alteration, and HER2 amplification suggests a useful conceptual bridge between imaging and tissue biology. Nuclear medicine often captures global tumor behavior, while pathology captures local cellular detail. Here, the two domains align.
Such alignment could be useful when a radiotracer signal is unexpectedly high or low relative to standard pathology markers. A metabolically active lesion may contain a proteomic program shaped by growth factor signaling and endocrine escape rather than by one isolated genetic lesion. In turn, a lower-uptake lesion may reflect a different cellular composition, even within the same stage of disease.
The current dataset does not establish causality, and it does not claim that FDG uptake alone can replace histopathology or genomics. It does show that in routine breast cancer specimens, proteomic heterogeneity is measurable, spatially patterned, and biologically meaningful. The remaining challenge is to connect these cellular states to treatment outcomes in larger cohorts, where protein clusters can be tracked alongside recurrence, therapeutic response, and survival rather than only with imaging or baseline tumor composition.