Intracellular signaling and trafficking are regulated by selective protein-membrane interactions. Transfer of cytosolic proteins to the membrane presumably occurs in two steps: an initial approach based on electrostatic attraction followed by lipid-induced protein refolding and/or insertion [110
]. Potential control mechanisms include: (1) modulating the protein's affinity for lipid (e.g., calcium-binding promotes the membrane association of C2 domains by enhancing electrostatic forces), (2) sequestering the lipid at specific locations, and/or (3) restricting access to the lipid in the absence of specific stimuli [10
In-vitro experimental support for the computationally predicted lipid-binding sites of α-Actinin, Arp2, Talin, and Vinculin (site 935–978) was obtained using standard techniques such as hydrophobic labeling, differential scanning calorimetry (DSC), Langmuir Blodgett (film balance), FTIR, T-jump, CD spectroscopy, cryo-electron microscopy (EM), and isothermal titration calorimetry. Similar data are not yet available to gauge the in-vitro binding characteristics of the sites predicted by our algorithm for CapZbeta-1 or the vinculin sites (residues 1020–1040 and 1052–1066).
The three-dimensional structures of the computationally predicted lipid-binding sites described here are, with the exception of Site 1 of CapZbeta-1, predominantly or exclusively alpha-helical. The energy required to insert a polypeptide into a membrane is minimized by the presence of favorable secondary structure [114
]. Membrane-spanning or surface associated amphipathic alpha-helices and beta-strands/sheets are common in biologically active peptides and proteins. Amphipathic alpha-helices may reversibly associate with lipids and function as peptide detergents [115
]. Amphipathic beta-sheets, in contrast, interact with lipids in an essentially irreversible manner, and lack detergent properties. Unfavorable energy costs associated with individual amphipathic beta-strands are likely to drive coalesence into beta-sheets on lipid surfaces. When the axis of an amphipathic helix lies parallel to the membrane surface and partially inserted into the membrane, the polar and non-polar protein surfaces may interact simultaneously with the charged head groups and hydrophobic side chains, respectively.
Four of the five cytoskeletal proteins studied here show a strong preference for acidic phospholipids in vitro
(alpha-actinin, Arp2, talin, and vinculin). The mechanism by which these soluble cytoplasmic proteins become membrane associated is unclear. Only one of the five proteins is known to undergo covalent lipid modification (i.e., vinculin). Although myristoylation and palmitoylation increase hydrophobicity, myristate alone may be insufficient to anchor proteins to the plasma membrane [1
]. The clustering of basic residues adjacent to lipid modification sites found among proteins such as K-ras4B and HIV-1 Gag enhances favorable electrostatic interactions with acidic lipids [19
]. Other peripheral proteins (e.g., type II beta-phosphatidyinositol-3-kinase, AKAP79, myelin basic protein, and a number of proteins containing C2 domains), in the absence of lipophilic modifications, depend solely upon basic groups to bind to membrane surfaces [112
]. The three-dimensional structures of pleckstrin homology domains reveal large positively charged electrostatic patches surrounding the ligand-binding sites, suggesting that the excess charge is useful in improving initial attraction and orientation to the predominantly negatively charged plasma membrane [113
]. Most of the predicted lipid interface sites in this study are either intrinsically electrostatically positive (Table ) or are located in regions that are relatively basic.
Many critical biological pathways are regulated by protein-lipid interactions. Understanding this biology is difficult given the complexity and heterogeneity of the interface. Computational methods, such as our matrix algorithm, provide a potentially powerful means for predicting the region, and orientation, of a protein as it associates with lipid aggregates. Further experimental work will be required to validate and refine this algorithm. However, based on experience with the protein huntingtin, it appears that the methods may be applicable to multiple protein classes [120