Cryo-electron tomography (cryo-ET) provides a unique window into molecular organization in cellular environments (in situ). However, the interpretation of molecular structural information is complicated by several intrinsic properties of cryo-ET data, such as noise, missing wedge, and continuous conformational variability of the molecules. Additionally, in crowded in situ environments, the number of particles extracted is sometimes small and precludes extensive classification into discrete states. These challenges shift the emphasis from high-resolution structure determination toward validation and interpretation of low-resolution density maps, and analysis of conformational flexibility. Molecular Dynamics (MD) simulations are particularly well suited to this task, as they provide a physically grounded way to explore continuous conformation transitions consistent with both experimental data and molecular energetics. This review focuses on the roles of MD simulations in cryo-ET, emphasizing their use in emerging methods for conformational landscape determination and their contribution to gain new biological insight.
Cryo-electron tomography (cryo-ET) provides a unique window into molecular organization in situ [1][2][3][4][5][6]. While operating in low electron dose and cryogenic conditions, it enables three-dimensional (3D) visualization of biomolecular complexes in their native cellular environments. However, the crowded and heterogeneous cellular contexts and low signal-to-noise ratios (SNRs) of cryo-ET data complicate the analysis of structural and conformational variability of individual molecular complexes [7][8][9][10][11][12]. Additionally, 3D reconstructions (tomograms) from collected cryo-ET images (tilt series) are affected by deformations that result from a limited angular range of the collected data, which is known as missing wedge (MW). This further complicates the determination of the structure of individual molecular complexes from the data.
As a consequence, cryo-ET data analysis workflows for structure determination usually rely on methods for sorting data of individual molecular complexes into discrete classes and class averaging [13][14][15][16][17][18][19]. However, discrete classifications are naturally suited to discrete well-separated structural states and hinder revealing subtle structural differences between individual complexes that can be hidden in class averages. Additionally, a small number of particles extracted from in situ cryo-ET data in some cases precludes extensive classification into discrete states.
On the other hand, tomographic analyses increasingly reveal ensembles of related conformations rather than discrete structural states [7,12,[20][21][22]. This challenges traditional structural biology paradigms and motivates the development of computational approaches capable of interpreting cryo-ET data in terms of conformational landscapes rather than static models. Recently, a few such new methods have been proposed [23][24][25][26]. They analyze conformations of individual molecular instances in the data instead of using discrete classifications.
Therefore, a defining feature of current structural studies by cryo-ET is not merely limited structural resolution due to intrinsic properties of cryo-ET data but also due to the presence of gradual conformational transitions with many intermediate states, known as continuous conformational heterogeneity. Molecular dynamics (MD) simulations and related mechanics-based modeling approaches provide a natural framework to address this challenge. MD simulations encode physical atomic interactions and trajectories in conformational space, and offer a natural way to relate experimental density to molecular motion. Over the last decade, MD-based methods have evolved from validating and interpretive tools, applied post hoc to experimentally derived structures, into integrated components of cryo-ET analysis pipelines, capable of sampling conformational landscapes under experimental constraints. This review focuses on the roles of MD simulations in cryo-ET, emphasizing their use in emerging methods for conformational landscape determination and how these approaches enable new biological insight. Particular attention is paid to conceptual and historical links between single-particle cryo-electron microscopy (cryo-EM) and cryo-ET conformational variability methods, and to recent cryo-ET studies that motivate future methodological developments for determination of conformational landscapes.
Recent deep learning approaches for conformational heterogeneity analysis in cryo-ET, such as tomoDRGN [25], cryoDRGN-ET [26], and cryoDRGN-AI [27], highlight the growing interest in conformational landscape reconstruction from tomographic data. These methods infer low-dimensional conformational landscapes from 2D particle tilt images, enabling visualization of continuous conformational variability without requiring prior atomic models. While powerful for uncovering heterogeneity in cryo-ET data, such approaches yield conformational landscapes expressed in terms of particle density-map predictions, rather than explicit atomic models or trajectories. In these methods, the conformational landscape is a low-dimensional latent space into which high-dimensional input data are embedded by deep learning. In contrast, hybrid methods based on dynamics simulations, such as MDTOMO [24] and HEMNMA-3D [23], yield conformational landscapes expressed in terms of atomic or coarse-grained models. Among them, MDTOMO integrates classical MD simulations (based on Newton’s equations of motion) with cryo-ET data, providing a direct link between experimentally observed heterogeneity and physically grounded models of conformational transitions.
The current methods for determination of conformational landscapes in cryo-ET data are summarized in Table 1.
Early cryo-EM MD-based approaches demonstrated how cryo-EM density maps could bias MD trajectories to refine atomic models [28,29]. These methods focused on deriving atomic models from consensus maps and did not address particle-to-part
This content is AI-processed based on open access ArXiv data.