Background

Eukaryotic cells are compartmentalized into membrane-bound organelles, each defined by a unique lipid composition essential for its function. Preserving this lipid balance requires selective transport between organelles, mediated by both vesicular and non-vesicular mechanisms. A major class of mediators of non-vesicular lipid trafficking is lipid transfer proteins (LTPs), which encapsulate lipids in hydrophobic cavities and transfer them between membranes. In recent years, many new LTPs have been identified, highlighting the widespread role of this process in cellular physiology. Some LTPs are thought to operate at membrane contact sites, forming extended hydrophobic tunnels that physically bridge organelles and enable bulk lipid transfer.

Although significant progress has been made, the molecular mechanisms by which LTPs function remain only partially understood. Key challenges include determining how LTPs recognize donor and acceptor membranes with high specificity, and how they overcome energy barriers to extract and release lipids and achieve transport directionality. Structural biology has provided valuable insights, yet many high-resolution LTP structures lack bound lipids, likely reflecting their dynamic behavior within the binding cavity. For bridge-like LTPs (BLTPs) in particular, structural models often rely on AlphaFold predictions, leaving uncertainties about lipid positioning and transport pathways. These proteins are a recently discovered type of LTPs that are large enough to span the whole distance between organelle membranes at membrane contact sites (MCS), consisting of a long hydrophobic cavity that is able to accommodate tens of lipids simultaneously.

Computational approaches offer complementary tools to address these gaps. Traditional methods, such as molecular docking, are limited by the high flexibility of lipid substrates, while atomistic molecular dynamics (MD) simulations, although informative, are computationally demanding and not easily applied to large-scale studies. AI tools are also emerging as a popular tool for protein–ligand predictions, but they are untested in the case of lipids. Recent advances in coarse-grained molecular dynamics (CG-MD) simulations provide a promising alternative, allowing protein–ligand interactions to be captured with reduced computational cost [1]. In this sense, despite the lack of flexibility of the protein structure on CG-MD simulations, the hydrophobic character of the lipid-binding cavity allows reproducing the experimental binding of lipids with no false positives. More importantly, the dynamic view that MD simulations provide allows analyzing aspects of the lipid-binding process that cannot be inferred by docking or AI methods, such as the pathway for lipid entry.

In our recent publication [2], we adapted unbiased CG-MD simulations to investigate lipid–LTP interactions. By placing lipids in the solvent, we observed their spontaneous binding to LTPs, enabling the identification of binding pockets, entry pathways, and lipid density distributions within these proteins. Applying this protocol to bridge-like LTPs such as Vps13 and Atg2 revealed structural features of their lipid-binding activity that have remained unresolved experimentally. Our approach thus provides an efficient, accurate, and scalable framework to dissect the mechanisms of lipid transport and to guide the discovery of new LTP functions.