Lipid-Protein Interactions in Plasma Membrane Organization and Function.

FOR INVESTIGATING SPECIFIC LIPID–PROTEIN INTERACTIONS Investigation of specific lipid–protein interactions and deciphering the homology of lipid-binding domains largely rely on structural biology techniques. There are other techniques such as nuclear magnetic resonance, electron spin resonance, and force spectroscopy; however, due to space restrictions, we focus mostly on structural methods in this review. We briefly address the state-of-the-art technologies that have contributed immensely to our current understanding of specific lipid–protein interactions, as well as of collective membrane behavior. We apologize for leaving out important work in the field performed using other technologies not mentioned in this review. 3.1. X-Ray Crystallography X-ray crystallography allows one to obtain the structure (atom positions and chemical bonds) of crystallized material by studying the diffraction of X-rays by electron clouds of the crystal. Although this approach is widely applicable to deciphering the atomic structure of proteins, it requires protein crystallization, which has historically been challenging for membrane proteins. Membrane proteins contain hydrophobic residues, which require detergents to extract them from the membrane for subsequent purification and crystallization ( 93 ). This requirement is particularly problematic for isolation of proteins together with their lipid ligands because detergents interact strongly with lipids. The selection of detergents is of crucial importance for crystallization, and research on the applicability of different types of detergents has made it possible to identify the best detergents for crystallography ( 102 ). As such, several proteins have been successfully crystallized together with their lipid ligands, which enabled the resolution of structures of lipid-binding pockets. For instance, interactions of the pore-forming protein lysenin with sphingomyelin ( Figure 1a ) or of β2 adrenergic receptor with cholesterol ( Figure 1b ) were confirmed by structural studies. As an alternative, crystallization in lipid mesophases ( 148 ) or nanodiscs ( 14 , 31 ) was used to study the structure of bacterial outer membrane proteins, photosynthetic proteins, and GPCRs ( 23 , 58 , 109 ). Another alternative is electron crystallography ( 107 , 110 ), which enables the solution of the structure of proteins together with their lipid ligands (annular lipids) by the formation of two-dimensional crystal arrays. 3.2.

Cryogenic Electron Microscopy

Over the past decade, cryogenic electron microscopy (cryo-EM) ( 6 , 22 , 42 , 98 ) has emerged as a promising alternative to X-ray crystallography for elucidating protein structure. In cryo-EM, vitrified protein solutions are imaged with electron microscopy to obtain multiple single-molecule images of the same protein in different orientations. Many such images can be processed to reveal protein topography with atomic precision. Unlike X-ray crystallography, cryo-EM does not require protein crystallization; instead, vitrification is achieved by plunge freezing a protein solution in liquid nitrogen. This approach has been extensively employed to solve structures of various membrane proteins, including GPCRs ( 153 ), pore-forming toxins ( 11 ), assembled dynamin polymers ( 66 ), clathrin cages ( 90 ), and Bar domains ( 136 ). Since crystallization is not required, many proteins that could not be studied with X-ray crystallography can now be solved with cryo-EM. One example is the transient receptor potential (TRP) superfamily, the structures of whose proteins could not be solved by X-ray crystallography due to the difficulties in crystallization. Since 2013, developments in cryo-EM enabled researchers to solve at least one protein for each of seven subfamilies of TRPs ( 22 ). Particularly important for this review, the structure of TRPV1 resolved by cryo-EM identified the distinct amino acid residues that bound phospholipid hydrophobic tails in the outer leaflet of the PM, as well as the hydrophilic residue in the extracellular domain that targeted phospholipid headgroups ( Figure 1c ). Unfortunately, isolation of membrane proteins for vitrification still relies on detergents to extract proteins from their native membrane. A potential alternative is protein isolation via nanodiscs, which appears to be advantageous when the aim is to resolve the lipid ligands of the proteins ( 41 , 153 ). Even in these cases, lipid ligands tend to provide low electron microscopy contrast, which introduces difficulties for their resolution with atomic precision. 3.3.

Structure Prediction and Simulations

The progress of computational technologies has enabled structure prediction of proteins based on their amino acid sequences. Clearly, the prediction of protein secondary structure and the geometry of protein folding domains ab initio (i.e., physics-based) is challenging, so methods of protein structure prediction often rely on existing knowledge of 3D structures obtained for similar proteins by experimental techniques. Several approaches are employed. Homology modeling ( 138 ) compares amino acid sequences of unknown structure to those of well-resolved proteins and maps unknown residues onto the known structural template of the related homologous protein. This approach often provides an efficient starting point for sequences with at least 30% similarity ( 144 ). An alternative fold recognition approach ( 100 ) relies mainly on the notion that the number of folds is much lower than the number of sequences, and it attempts to identify the most likely fold for a given sequence. Similar fragment-based methods ( 126 ) allow for assembly of protein structure based on structures of sequence fragments similar to various proteins available in the protein data bank. Finally, the most difficult protein structure prediction approaches rely only on physical potentials of atomic interactions ( 64 , 105 , 129 ). The modern methods of structural predictions often combine several of the above-mentioned approaches, and they became especially powerful with the development of neural networks and machine learning algorithms. The most relevant example is the recent Alphafold project ( 117 ), which uses a convolutional neural network trained on Protein Data Bank structures to estimate the distances between the atoms of residues of a protein. Therefore, it can accurately predict the structure of a protein given its sequence by minimizing the potential by gradient descent. Structure predictions can be highly useful for determining lipid binding sites from the structures of membrane proteins. Lipid binding pockets can be recognized from known amino acid sequences ( 97 ), characteristic protein folds ( 86 ), hydrophobicity ( 89 ), or charge ( 8 ). The orientation and affinity of lipid ligands can then be further assessed by molecular docking ( 149 ). Molecular dynamics (MD) simulations enable the elucidation of possible lipid binding pockets reconstituted together with membrane lipids in silico ( 16 , 25 , 116 ) (e.g., a p24 transmembrane domain interacting with sphingomyelin, as in Figure 1d ). Such molecular- or atomic-scale dynamic simulations provide the unique opportunity to explore molecular interactions at the Angstrom- and nanosecond-scale. Transmembrane proteins can be simulated as embedded in the dynamic ensemble of the multicomponent PM ( 20 ) to resolve specific lipid–protein interactions involved in membrane transport ( 35 ), viral infection ( 65 ), or antimicrobial activity ( 128 ). However, simulation of thousands of molecules for multiple microseconds can be computationally demanding, necessitating alternative methods such as standard residue interaction networks ( 147 ) to interpret protein structures and resolve molecular interactions. An alternative to computationally demanding full-atom simulations is the use of coarse-grained simulations where molecules are simplified ( 131 ). Using coarse-grained simulations, the duration of the simulation, as well as the number of simulated molecules, can be increased ( 39 ). The combination of structural studies and MD simulations of the membrane proteins in the lipid bilayer have led to the establishment of the MemProt database of simulations of transmembrane proteins embedded in the lipid bilayer ( 94 ) (as an example, β2 adrenergic receptor in the lipid bilayer is shown in Figure 1e ). Compared to the other techniques discussed above, MD simulations can provide unique information on the energetics and kinetics of lipid–protein interactions. Since certain time windows can be simulated dynamically, free energy calculations can be performed to test lipid binding affinities (and thus relative probabilities of binding) to certain proteins or protein binding affinities to certain lipid environments (for reviews, see 26 , 27 ).

4. METHODS TO STUDY COLLECTIVE LIPID–PROTEIN INTERACTIONS In addition to their roles as individual ligands for protein interactions, membrane lipids collectively determine the biophysical properties of membranes. Therefore, it is crucial to consider membranes as more than the sum of their individual components. Technologies that enable us to see nanoscale biophysical properties of membranes help us understand how these collective properties take part in protein functionality. We discuss various aspects of collective membrane properties in the following sections. 4.1.

Protein Sorting

(Enrichment) into Certain Lipid Environments Multiple mechanisms have been proposed for protein sorting into particular lipid environments independent of specific lipid–protein interactions. One such mechanism is based on membrane thickness matching. Hydrophobic transmembrane domains of membrane proteins (or their post-translationally acylated parts) can attract lipids of optimally matched chain length and/or saturation degree ( 88 ). Furthermore, for acylated (e.g., palmitoylated, myristoylated) proteins and GPI-anchored proteins (GPI-APs), compartmentalization largely relies on the length and saturation degree of the lipid anchor. Another sorting mechanism is lipid-driven assembly of ordered domains (lipid rafts) that sequester membrane proteins and define their transmembrane positioning and spatial organization. In this context, lipid rafts are usually defined as nanoscale signaling platforms rich in cholesterol, sphingolipids, glycolipids, and raft proteins, i.e., proteins that prefer more ordered membrane environments ( 32 , 81 ). Despite the debatable nature of membrane lateral domains, it is evident that lipid–protein interactions go far beyond specific recognition of lipids by certain proteins and should also be considered in the context of collective behavior. To obtain the comprehensive picture of such complex, collectively driven lipid–protein interactions, the PM has been studied using various biochemical and analytical techniques, as detailed below ( Figure 2 ). 4.1.1. Detergent resistance to study membrane heterogeneity and protein sorting. The content of ordered domains was first studied by using detergents to extract detergent-resistant membrane (DRM) and detergent-soluble membrane (DSM) fractions, with further analysis of these fractions conducted using biochemical methods (e.g., western blotting) or mass spectroscopy ( 13 , 44 ) ( Figure 2a ). This approach showed the association of some proteins [such as GPI-APs ( 15 )] with sphingomyelin- and cholesterol-rich membrane fractions. Moreover, for caveolin ( 103 ) or GPCRs ( 71 ) that were found in DRMs, the association with sphingomyelin and cholesterol is preferential due to specific lipid interactions mediated by consensus structural motifs. Despite being useful in demonstrating the general heterogeneity in PMs, DRM fractions showed major discrepancy in the composition when different detergents were used ( 115 ). This made clear that information on membrane compartmentalization obtained by application of detergents should be addressed with caution, and more sensitive approaches must be employed. Therefore, alternative approaches were developed, including lipid bilayer nanodiscs ( 31 ) using membrane scaffold proteins ( 7 ) or synthetic polymers ( 53 , 99 ). Membrane nanodiscs can resemble some features of the native membrane environment ( 87 ), which allows for characterization of membrane order and protein orientation, in addition to the content of an extract. For example, order parameters of lipid chains are widely assessed by nuclear magnetic resonance ( 57 ). Notably, the developments in nanodisc extraction contributed to the sample preparation methods required for structural studies ( 14 , 153 ), namely crystallization (X-ray crystallography) or plunge freezing (cryo-EM). 4.1.2. Liquid–liquid phase separation as a biophysical basis for lateral protein organization. Extensive investigation of self-assembly and self-organization of lipids in synthetic multicomponent lipid bilayers supports the concept of lipid-driven compartmentalization of the PM. Specifically, lipid bilayers undergo phase transition from a liquid-crystalline phase (low lipid order and high mobility) to a gel phase (also sometimes called the solid phase; low lipid mobility and high order). PMs at physiological conditions adopt a liquid-crystalline phase, which is a fundamental basis of the fluid-mosaic model. Two-component lipid bilayers composed of lipids with distinct phase transition temperatures and saturation exhibit formation of spatially separated domains of the liquid-crystalline phase and gel phase at intermediate temperatures. Interestingly, introduction of cholesterol can fluidize the gel phase and convert it into a liquid-crystalline phase. Because this phase is distinct from the cholesterol-poor liquid phase, lipid bilayers composed of saturated lipids, unsaturated lipids, and cholesterol can exhibit liquid–liquid phase separation (LLPS). Such bilayers contain two liquid-crystalline phases of distinct lipid order and mobility: a liquid-disordered (Ld) phase (low order, high mobility) and a liquid-ordered (Lo) phase (higher order, lower mobility). Classically, the lipid bilayer composition that was used to model biologically relevant LLPS was a combination of phospholipids with low transition temperatures (e.g., DOPC or POPC), phospholipids with high transition temperatures (e.g., DPPC, sphingomyelin), and cholesterol. Such LLPS can be observed in a variety of model membrane settings. In particular, spherical cell-sized vesicles have been applied to investigate the biophysical determinants of LLPS and incorporation of membrane proteins into lipid domains ( 62 , 143 ). These synthetic vesicles provide certain advantages that make them ideal for studying lipid–protein interactions. First, their lipid composition (e.g., amount of charged lipids, cholesterol, saturation) can be tightly controlled to evaluate interactions of isolated proteins with different lipid species ( 113 , 140 ). Furthermore, transmembrane proteins can be incorporated via detergent dilution into free-standing model membranes ( 5 ), whereas extramembrane domains can be anchored to lipids via His-Tags or biotin-avidin coupling ( 19 , 60 ). Despite these advantages, the limited complexity of these synthetic systems restricts their use in studying native membrane interactions. As an alternative to fully synthetic membranes, cell-derived vesicles can be exploited for studying protein compartmentalization. These membranes are extracted from living cells by application of vesiculating agents, thus largely conserving PM lipid and protein content. Moreover, phase-separated cell-derived vesicles display LLPS between microscopic ordered and disordered domains. Due to the easy reconstitution of desired membrane proteins in these cell-derived systems (by forcing the cells to express these proteins), they have been used to investigate the structural determinants of ordered-domain association for membrane proteins ( 81 ). Liquid phases in artificial and cell-derived membranes differ in composition and organization: Synthetic systems are simpler in composition with extreme difference in order between ordered and disordered phases. In contrast, cell-derived membranes are more complex in composition with marginal difference in membrane order between ordered and disordered phases ( 137 ). As protein sorting into lipid-driven domains is a function of the packing of the domains, which varies between different model membrane systems ( 119 , 137 ), compartmentalization of proteins can differ in artificial and cell-derived membranes ( 5 , 62 ). As such, a GPI-AP (one of the bona fide ordered domain components), placental alkaline phosphatase (PLAP), partitions into Ld domains when incorporated into synthetic vesicles, presumably due to extreme order difference between the domains. Another GPI-AP, CD59, clearly prefers ordered domains in phase-separated cell-derived vesicles ( Figure 2b ). Therefore, synthetic and cell-derived systems can provide complementary insights into the comprehensive picture of lipid and protein compartmentalization. 4.1.3. Environment-sensitive probes to predict the lipid environment of membrane proteins. Biophysical properties of the PM can be assessed using fluorescence lipophilic probes sensitive to membrane order ( 69 , 112 ), viscosity ( 70 ), and tension ( 24 ). When used in combination with fluorescent proteins, these probes can report on the immediate environment of fluorescently labeled proteins ( 142 ). Notably, such probes can also be modified with ligands of specific proteins to measure the biophysical properties of the protein microenvironment ( 48 , 141 ) ( Figure 2c ). Using this method, Hanser et al. ( 48 ) conjugated an oxytocin receptor (a GPCR) ligand to the environment-sensitive fluorescent dye Nile Red and reported different lipid packing around the oxytocin receptor compared to the bulk membrane environment. With a similar strategy, Umebayashi et al. ( 141 ) developed a Nile Red construct that specifically binds insulin receptor. Similarly, the membrane environment of insulin receptor was distinct compared to the average PM order ( 141 ). Both papers highlight the importance of a linker between the protein binding site and the Nile Red molecule, as this linker changes the probe localization in the membrane. 4.1.4. Protein–lipid proximity sensing by Förster resonance energy transfer. Förster resonance energy transfer (FRET) provides a unique opportunity to measure the molecular distance between two molecules, a donor and an acceptor (a FRET pair), with a sensitivity of units of nanometers. The main requirement for FRET is that molecules of interest are tagged with fluorescent moieties, and the fluorescent spectrum of the FRET donor overlaps with the absorption spectrum of the FRET acceptor. Notably, the intrinsic fluorescence of proteins (e.g., via excitation of Trp) can also be used as a FRET donor. FRET effectively reports on the position and orientation of membrane proteins ( 92 ), as well as on protein–protein interactions and oligomerization in the PM ( 63 , 88 , 151 ). In the context of lipid–protein interactions, such a FRET pair can obviously be constituted of a single protein and lipid species of interest ( Figure 2d ). As FRET acceptors, lipids can be modified by a fluorescent moiety at the headgroup or fatty acyl chain. Lipid–protein FRET can report on specific lipid–proteins interactions ( 25 ), where FRET occurs between a single protein donor and a single lipid acceptor. Moreover, FRET between a single protein donor and multiple lipid acceptors can provide information on the recruitment of lipids to the protein annular region ( 40 ). Finally, FRET is employed to resolve protein incorporation into nanodomains enriched in specific lipids ( 1 , 83 , 111 ). Deciphering the characteristics of lipid–protein interactions (affinity, avidity, presence of lipids in the annular region of a protein, or protein incorporation into the lipid domains) from data obtained by FRET experiments requires the development of appropriate models by analytical solutions or simulations ( 82 , 111 , 135 ). A major advantage of the FRET approach is the possibility to monitor energy transfer in living cells employing live-cell fluorescence microscopy techniques. FRET measurements are highly sensitive to lipid and protein orientation and concentration, as well as to spectral overlap between the donor and the acceptor; therefore, these measurements require very reliable experimental controls. 4.1.5. Lipid environment of proteins by native mass spectrometry.

Mass spectrometry

(MS) has been a key tool to study both specific and collective lipid–protein interactions. Several MS methods have been applied to observe lipid–protein interactions, with native MS being the most common. In native MS, biomolecules are converted from a 3D, condensed-liquid phase to the gas phase via the process of electrospray ionization MS (ESI-MS) ( 78 ). The term native in the name of this method implies that the former liquid phase should be as physiologically relevant (in terms of pH, osmolarity, and ionic strength) and functional as possible to keep the biomolecules in their native state prior to gas phase transition and MS analysis ( 78 ). Being a gas-phase method, native MS has mild experimental conditions that enable the noncovalent interactions and functionality of the biomolecules to be largely preserved. This allows one to obtain information on stoichiometry, binding partners, and biomolecule topology and dynamics. Therefore, it is widely used to study lipid–protein interactions. Advanced sample preparations such as native nanodiscs or styrene maleic acid lipid particles (SMALPs) make it even more attractive for such applications, since native lipid–protein interactions can be preserved in these polymer–lipid combinations ( 52 ). Through the use of native MS, different aspects of lipid–protein interactions have been studied. For example, Laganowsky et al. ( 74 ) showed that different proteins are bound selectively by different lipid classes. Gupta et al. ( 44 ) showed that interfacial lipids are required for protein oligomerization. Landreh et al. ( 76 ) demonstrated the role of annular lipids on protein folding and conformation. Native MS and its applications have been reviewed in detail in Reference 12 , and the elaborate protocols can be found in Reference 45 . 4.2. Protein Geometry Influenced by the Membrane Properties Protein recruitment to certain lipid environments, discussed above, leads to increased local concentrations. Such enrichment might ensure a concentration threshold for triggering biological processes. However, enrichment is not a requirement for lipid environment-selective activation of proteins. Even a small fraction of a given protein can be the active pool if it resides in the lipid environment that ensures a favorable geometry for molecular interactions. The CD2 protein, for instance, shows different geometries, and thus different accessibility, in different lipid environments ( 106 ) ( Figure 3a ). Another example is the bacterial protein lactose permease, which can undergo rapid postassembly TMD flipping in response to changes in the lipid environment ( 145 ) ( Figure 3b ). Finally, Frizzled, a receptor for the Wnt ligand, has been shown to partition preferentially into the DSM fraction, as does its coreceptor, LRP6. However, the Wnt ligand binds exclusively to the minor DRM pool of these receptors ( 118 ) ( Figure 3c ). These examples show that protein function can be modulated by the biophysical properties of the local lipid microenvironment.