Single-cell immune ecotypes shape microenvironment-modulated evolutionary dynamics in pediatric leukemia

Tumor evolution is shaped not only by intrinsic genetic alterations but also by selective pressures imposed by the tumor microenvironment (TME). Although single-cell profiling increasingly resolves ecological heterogeneity, quantitative evolutionary models that explicitly incorporate microenvironmental structure remain limited. Here, we introduce a TME-modulated Ornstein–Uhlenbeck (OU) framework with ecotype-informed hierarchical structure to identify microenvironment-associated evolutionary regimes in pediatric leukemia. Using single-cell transcriptomes from the Single-cell Pediatric Cancer Atlas (ScPCA), we quantify patient-level TME composition and define reproducible immune ecotypes—recurring microenvironmental states that represent discrete ecological contexts. We fit an ecotype-informed OU model in which patient-specific trajectories along a pseudotime-derived transcriptional differentiation score evolve under stabilizing selection and stochastic diffusion, while ecotype membership modulates evolutionary parameters through hierarchical priors enabling partial pooling across patients with similar TMEs. Bayesian inference yields uncertainty-aware estimates of evolutionary optima and stabilizing selection strength at both patient and ecotype levels, and posterior predictive checks support model adequacy. Comparative analyses uncover distinct ecotype-specific regimes: immune-enriched microenvironments exhibit stronger stabilizing selection and more constrained dynamics, whereas stromal/other-enriched contexts show weaker stabilizing selection and greater stochastic drift. Together, these findings link single-cell–defined microenvironmental states to evolutionary dynamics and provide an interpretable, uncertainty-aware framework for comparing tumor evolution across patients and cohorts.