A single tumor biopsy reveals how cell populations change over time

Tumors and inflamed tissues evolve through interactions between cell types over days to weeks, but a human biopsy provides only one snapshot in time. Repeated biopsies are impractical, and mouse, organoid and co-culture models lack native human tissue context. A way to infer dynamics from a single sample would let researchers and clinicians study tissue circuitry and possibly forecast treatment response.

The authors introduce one-shot tissue dynamics reconstruction (OSDR). Using spatial proteomics data (imaging mass cytometry), they tabulate for each cell its type and the number and types of cells within an 80 µm radius. Division is defined by Ki67 thresholds, and logistic regression learns division probability as a function of neighborhood composition. Removal is assumed constant per cell type and approximated by mean division rate. These rules drive stochastic spatial simulations and yield phase portraits showing fixed points and basins of attraction. Simulations of all four two-cell topologies showed that a few thousand cells per type suffice to recover known dynamics.

On 715 breast cancer biopsies (859,710 cells), neighborhood counts predicted Ki67 significantly (P < 10^-13). The inferred fibroblast-macrophage portrait contained a hot fibrosis fixed point (both cell types), a cold fibrosis fixed point (fibroblasts only), and a semi-stable macrophage-rich point, matching in vitro co-culture measurements. Higher tumor density shifted the system toward the hot state. Hot fibrosis was linked to worse survival (median 132 vs 192 months; log-rank P = 0.0046), and remained significant after controlling for tumor cell number. Results held in two further cohorts (1,012 patients, 2.1 million cells).

For T and B cells, OSDR inferred an excitable system: a stable zero state, but a T cell density above threshold triggers a pulse of T cells, then B cells, then collapse, followed by a refractory period. A 3D model indicated CD4 T cells initiate the pulse and B cells provide negative feedback. In triple-negative disease, the pulse ended with stable B cells rather than collapse, consistent with high lymphocyte infiltration. In the NeoTRIP trial (279 patients; chemotherapy vs chemotherapy plus anti-PD-L1), OSDR applied to early-treatment biopsies predicted tumor cell collapse in responders but not non-responders.

Caveats: the full text provided was truncated before the detailed trial results. The approach assumes division depends on current neighborhood composition, uses a constant death rate because no adequate death marker exists, and ignores massive cell influx. Predictions are model-based inferences validated mainly against one co-culture dataset and retrospective data, not prospective testing.