Understanding how stochastic molecular events produce ordered cellular machines that generate force is a key problem in cell biology. Structural methods such as cryo-electron tomography (cryo-ET) can visualize macromolecular assemblies inside cells at near-molecular resolution. However, cryo-ET typically yields static tomograms that are interpreted as single structural snapshots, offering limited direct access to assembly states or temporal progression.
This study uses human macrophage podosomes—physiologically relevant actin-rich force-generating structures—as a model system to test whether information relevant to assembly state can be extracted from cellular tomograms.
The authors combined several complementary approaches applied to cellular cryo-ET data. They performed spatial mapping of filamentous actin (F-actin) and identified Arp2/3-mediated branch junctions within podosome networks. Filament segmentations were generated with deep-learning methods to obtain dense filament representations suitable for quantitative orientation analysis.
To relate local filament orientations and branch geometries to assembly processes, the team applied Markov-chain modeling to orientation-class transitions. This integrated pipeline—spatial mapping, segmentation-based orientation analysis and probabilistic modeling—was designed to infer preferred growth directions and assembly-state distributions from static 3D tomograms.
Analysis of mapped filaments and branch junctions revealed a bias consistent with membrane-directed actin polymerization within podosomes. In addition, Arp2/3-mediated branching showed preferential geometries that align with a local growth axis oriented toward the membrane. Together, these local biases establish a directional cue for network organization at the scale of individual nucleation and branching events.
The source reports that these observed local preferences provide interpretable mechanistic constraints that can be linked to larger-scale network architecture.
At the scale of the entire podosome network, the authors describe a repeating organization they term layered helical order. Filament-orientation classes recur at intervals of approximately 33 nm along the membrane-normal axis. These recurring classes form a three-class cycle that returns to an approximately equivalent, non-polar orientation, implying a repeating, layered arrangement in the direction normal to the membrane.
This network-scale pattern links the local orientation preferences of individual filaments and branch junctions to a higher-order structural motif observable in tomograms.
To formalize how local branching rules translate into the observed architecture, Markov-chain modeling was used to analyze mother-to-daughter filament-orientation transitions. The modeling identified preferred orientation transitions at branch points and produced a stationary composition of branch-orientation states.
Importantly, the stationary distribution predicted by the Markov model closely matched the composition of branch-orientation states observed in structurally more advanced regions of the podosome networks. This congruence supports the idea that the static tomograms contain statistical information reflecting progression along an assembly pathway rather than representing only a single, temporally isolated state.
By integrating spatial filament and branch mapping, deep-learning-based orientation analysis and Markov modeling, the study connects three scales of organization: local actin nucleation and Arp2/3 branching, network-scale layered helical order, and inferred assembly state. The results suggest that static cellular tomograms can retain interpretable traces of assembly dynamics, enabling inference about temporal progression and preferred growth pathways in native macromolecular machines.
The authors propose that this approach advances cryo-ET from purely structural description toward the capacity for temporal inference—moving in the direction of 4D in situ structural biology.
The document is a preprint and has not been peer reviewed. The source does not provide peer-review validation in this version. Specific experimental details, quantitative metrics and full methodological parameters are contained in the full preprint but were not all reported in the abstract. The authors declared no competing interests and acknowledge funding sources listed in the source article.