Abstract
<jats:p>How stochastic molecular events are converted into ordered cellular machines that perform mechanical work remains a central question in biology. Cryo-electron tomography enables visualization of macromolecular assemblies inside cells at molecular resolution, but static tomograms are generally interpreted as structural snapshots, with limited access to assembly state and temporal progression. Here, we show that cellular tomograms retain interpretable information about the assembly of a force-generating actin network. Using human macrophage podosomes as a physiologically relevant system, we combine spatial mapping of actin filaments (F-actin) and Arp2/3-mediated branch junctions, orientation analysis of deep-learning-based filament segmentations and Markov-chain modeling to infer actin-network assembly in situ. We identify favored membrane-directed actin polymerization and Arp2/3-mediated branching, establishing a local growth axis for podosome network organization. At the network scale, podosomes display layered helical order: filament-orientation classes recur at intervals of approximately 33 nm along the membrane-normal axis, forming a three-class cycle that returns to an approximately equivalent non-polar orientation. Markov modeling identifies preferred mother-to-daughter filament-orientation transitions and yields a stationary composition of branch-orientation states that closely matches that observed in structurally more advanced network regions. Together, these results link local actin nucleation, network-scale architecture and structural assembly state in a native cellular machine. Our study advances cryo-electron tomography from structural description toward temporal inference and 4D in situ structural biology.</jats:p>