Machine learning–driven multiscale modeling reveals lipid-dependent dynamics of RAS signaling proteins

Abstract

RAS is a signaling protein associated with the cell membrane that is mutated in up to 30% of human cancers. RAS signaling has been proposed to be regulated by dynamic heterogeneity of the cell membrane. Investigating such a mechanism requires near-atomistic detail at macroscopic temporal and spatial scales, which is not possible with conventional computational or experimental techniques. We demonstrate here a multiscale simulation infrastructure that uses machine learning to create a scale-bridging ensemble of over 100,000 simulations of active wild-type KRAS on a complex, asymmetric membrane. Initialized and validated with experimental data (including a new structure of active wild-type KRAS), these simulations represent a substantial advance in the ability to characterize RAS-membrane biology. We report distinctive patterns of local lipid composition that correlate with interfacially promiscuous RAS multimerization. These lipid fingerprints are coupled to RAS dynamics, predicted to influence effector binding, and therefore may be a mechanism for regulating cell signaling cascades.

Recapitulation of cell-like KRAS4b membrane dynamics on complex biomimetic membranes

“…The campaign primarily focused on understanding RAS behavior on the membrane at microscale length and nanosecond timescales. In a collaboration involving five different national labs, scientists performed an extensive automated multi-scale simulation of KRAS4b nestled on a biologically relevant PM model ( Ingolfsson et al, 2021 ). The in silico PM model was composed of an asymmetric eight-lipid mixture of cholesterol, phosphatidylcholines (PC), phosphatidylethanolamines (PE), phosphatidylserine (PS), phosphatidylinositol bisphosphate (PIP2), and sphingomyelin (SM).…”

“…Each type of lipid has a different head group, acyl chain length, and saturation. A detailed characterization of the eight-lipid bilayer including diffusion properties of individual lipid types can be found in ( Ingolfsson et al, 2021 ). This computational simulation revealed a dynamic rearrangement of lipid composition, especially increased PIP2 and decreased cholesterol, creating a local lipid fingerprint that induced lateral segregation of KRAS4b into multimers and suggested a greater role for lipid patterning in effector recruitment and downstream signal transduction.…”

“…In this work, we captured a cellular PM-like environment on a glass coverslip and investigated how heterogeneous lipid composition regulates the spatiotemporal organization of KRAS4b on model membranes. We implemented a bottom-up approach where we built up the membrane complexity starting from a simple two lipid, 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine (POPC) and 1-palmitoyl-2-oleoyl-sn-glycero-3-phospho-L-serine (POPS) bilayer and compared this system to a complex eight-lipid bilayer, including the most abundantly found lipid species in the inner leaflet of the PM ( Cooper, 2000 ; Ingolfsson et al, 2021 ). We performed total internal reflection fluorescence (TIRF)-SPT experiments to track single molecules of purified full-length KRAS4b, both farnesylated and methylated, as it diffuses along the 2D plane of the supported lipid bilayer, and utilized atomic force microscopy (AFM) to characterize the lipid bilayers.…”

Ras Multimers on the Membrane: Many Ways for a Heart-to-Heart Conversation

“…How these mutations alter Ras dimerization remains to be further elucidated, so do the functional roles of the β2-β3 interface. These dimer interfaces may just be a subset of what may exist, as revealed in a recent collaborative, large-scale modeling effort [116]. With a Multiscale Machine-Learned Modeling Infrastructure (MuMMI), the authors ran an ensemble of over 100,000 simulations of active WT KRas on lipid bilayers with a variety of lipid compositions.…”

“…On the computation side, homology or data-driven modeling [ 111 , 112 ] as well as template-based modeling [ 113 ] are widely used to generate physically fitting dimer structures, while molecular dynamics (MD) simulations extent the knowledge on the stability and interactions of the potential interfaces [ 91 , 114 , 115 ]. Recently machine learning approaches were coupled with the above to generate more data with increased accuracy [ 116 ]. On the experimental side, recent studies using FTIR, NMR, FRET, and EM in combination with point mutations and functional assays have been used to test and refine the predicted dimer interfaces [ 71 , 84 , 85 , 91 , 92 , 97 , 117 ].…”