时间序列单细胞RNA-seq的状态转换模型识别慢性髓系白血病（CML）疾病微状态稳定性的基因层面起源

CML is defined by evolution from chronic phase (CP) to increased disease burden during blastic phase, but the cellular mechanisms that create these disease states and produce the transition between states is not understood. We previously used state-transition models to show that CML evolution is not encoded in single-cell transcriptional microstates but instead emerges only when gene expression is aggregated into population-level macrostates where distinct phenotypic disease states emerge. Here, we extend this framework to ask how antagonistic teams of genes and their regulatory network defined steady states give rise to these macrostates. Using weekly time-series single-cell RNA sequencing from both CP and blast crisis (BC) inducible CML mouse models, we assessed the origin of phenotypic disease macrostates in each cell type by identifying antagonistic teams of genes. We identified these teams for each cell type lineage by selecting the genes where their eigenvalue in the state-space construction and their observed expression change combine to indicate that the gene either strongly promoted (pro-CML) or strongly opposed (anti-CML) leukemia. To coarse grain the large number of resulting of genes per lineage, we applied weighted gene coexpression network analysis (WGCNA) to define gene modules and module eigengenes that define coordinated transcriptional programs. Each module produced by this process were strongly enriched for either pro- or anti-CML which suggests that they define functional units in leukemia development. We then inferred gene regulatory networks for these modules using Bayesian network inference constrained by prior knowledge from curated interaction and regulatory databases. This produced module-level networks that were unique for each of the B, T, myeloid, and stem cell compartments. For each inferred network, we computed steady states (attractors) and projected the stable transcriptional configurations into the state-space to determine whether the gene derived attractors align with lineage-specific macrostates in the state-space. Preliminary analyses reveal that module networks can reproduce the early, transitional, and late CML macrostates observed from our previous study. Further, we performed in silico perturbations of the networks to predict shifts in attractor occupancy and recapitulate our previous findings that the dominant contributions of B and myeloid compartments to disease progression observed previously. These results support a mechanistic view of leukemia where CML macrostates arise from cell type-specific teams of genes organized into low-dimensional regulatory networks. These network level attractor states could provide a new approach to identify therapeutic targets that are directly related to disease phenotypes and, therefore, new approaches for preventing CML disease evolution.