Molecular determinants of lipid-protein recognition

The National Institute of General Medical Sciences awarded the University of California, San Diego $384,394 on August 1, 2026, under the Biomedical Research and Research Training program (CFDA 93.859) to investigate molecular determinants of lipid-protein recognition through computational and experimental approaches.

The research integrates deep-learning, molecular dynamics simulation, and de novo protein design to characterize biophysical interactions governing binding affinity and specificity of membrane-associated proteins with lipid headgroups and organellar surfaces. The work will employ de novo protein design to generate a library of headgroup binding proteins with distinct folds and binding pockets, then use this library to characterize membrane-protein association energetics both computationally and experimentally. The research will produce predictive models for thermodynamics and kinetics of lipid-protein binding in native proteins. A second approach focuses on GM2, a glycosphingolipid whose dysregulation contributes to Niemann-Pick Disease Type C, with development of novel de novo proteins to advance understanding of this disease mechanism.

Performance occurs in La Jolla, California, with a period of performance extending from August 1, 2026, through May 31, 2031. This is a Project Grant, a form of assistance that funds investigator-initiated biomedical research under NIGMS's mission to advance fundamental scientific discoveries in human health and disease.

Abstract

Lipids form the membranes that are the primary boundary of organelles. The chemistry and physical properties of these membranes are incredibly heterogeneous and dictate their signaling and mechanical properties. The shape, charge, and flexibility of lipid headgroups will often dictate the binding or regulation of protein activity. Changes in the regulation and distribution of these membrane lipids are a key contributor to diseases including cancer and neurodegeneration. While the composition of these membranes is of key biomedical interest, they are also very poorly understood and difficult to dissect with traditional biochemical and biophysical methods. Our laboratory seeks to integrate deep-learning, molecular dynamics simulation and DE novo protein design to investigate critical biophysical interactions that confer binding affinity and specificity of membrane-associated proteins with lipid headgroups, the plasma membrane or organellar surfaces. We utilize these methods to build generalizable models of protein binding to determine how binding terms like shape complementarity, hydrogen bonding, hydrophobic effect and entropic penalty confer specificity at the heterogenous solvent environment of the membrane surface. In our first approach we will employ DE novo protein design to generate a library of headgroup binding proteins with distinct folds and binding pockets. We will use this library to rigorously characterize the energetics of membrane-protein association both computationally and experimentally. Using this library and the resulting dataset, we will generate a model for predicting the thermodynamics and kinetics of lipid-protein binding in native proteins. In our second approach, we will focus on a specific lipid, GM2, which is a glycosphingolipid whose dysregulation is a key, but poorly understood, contributor to niemann-pick disease type c (NPC). We will focus on generating novel DE novo lipid-biosensors for GM2 to measure the trafficking of this lipid in live cells and to better understand the connection between disease mutations in NPC and the MIS trafficking of GM2. Both approaches, if carried out, will enrich our understanding of how a protein's binding surface and biophysical properties influence membrane-protein association and how the dysregulation of these resident lipids contribute to disease. In addition, the data we extract from our simulations and experimental assays will allow US to RE-evaluate our theoretical models of lipid-protein interactions, which will allow US to better generalize machine learning models of protein interactions with metabolites, small molecule drugs, and other biological macromolecules. Long term, we intend to be able to predict a protein's lipid target and affinity from its sequence, and to develop novel biosensors for every lipid type.