Membrane Recognition and Targeting by Lipid-Binding Domains

See allHide authors and affiliations

Science's STKE  16 Dec 2003:
Vol. 2003, Issue 213, pp. re16
DOI: 10.1126/stke.2132003re16


Modular domains that recognize and target intracellular membranes play a critical role in the assembly, localization, and function of signaling and trafficking complexes in eukaryotic cells. Large domain families, including PH, FYVE, PX, PHD, and C2 domains, combine specific, nonspecific, and multivalent interactions to achieve selective membrane targeting. Despite structural and functional diversity, general features of lipid recognition are evident in the various membrane-targeting mechanisms.


Lipid-binding domains that target intracellular membranes play a critical role in the assembly of signaling and trafficking complexes and in dynamic membrane remodeling events, such as vesicle budding, phagocytosis, and cell motility. The fundamental biological significance of reversible membrane targeting is underscored by (i) the prevalence of lipid-binding domains, which rank among the most common modular domains in the eukaryotic proteome, and (ii) the discovery of major proto-oncogene proteins and tumor suppressors containing essential lipid-binding domains or lipid metabolic activities that regulate membrane association (1-4). Consequently, the structure and function of lipid-binding domains have been the focus of intensive investigation. Within the past several years, the number of known lipid-binding domains has expanded considerably, providing fresh insight into the mechanisms underlying lipid recognition and membrane targeting.

This review addresses several major classes of lipid-binding domains, with an emphasis on the large modular domain families that have evolved to recognize and target membranes. These include pleckstrin homology (PH), FYVE (acronym of Fab1, YOTB, Vac1, and EEA1), plant homeodomain (PHD), phox homology (PX), and C2 [named for homology with protein kinase C (PKC)] domains, as well as smaller domain families and peptide motifs (Table 1). The variation in physical properties and recognition mechanisms between and within families is striking. A realistic assessment of the large literature on lipid-binding domains leads to the unavoidable conclusion that each domain represents a unique case. Nevertheless, from the collective insights provided by cell biological, biochemical, structural, and biophysical studies, general principles governing lipid recognition and membrane targeting have emerged.

Lipids, Head Groups, and Cellular Membranes

Membranes consisting of phospholipid bilayers are a ubiquitous component of all cells. With a large repertoire of chemically distinct lipids, the composition of biological membranes is highly complex and variable, depending both on the type of cell and the organelle of interest. Lipid composition also varies within organelles, giving rise to microdomains with distinct physiochemical properties that reflect both the stereochemical and electrostatic characteristics of the head group, as well as the length, saturation, and branching of the hydrocarbon chains. As a consequence, cellular membranes, and microdomains in particular, are rather poorly defined, both structurally and chemically, although technological advances in mass spectrometry have created the opportunity for a nascent field of "lipidomics" to grapple with these issues (5). More in-depth discussion of the literature on membrane microdomains, including lipid rafts, can be found in several recent reviews (6-10).

Some of the more abundant phospholipids include phosphatidylcholine (PC), phosphatidylserine (PS), phosphatidylethanolamine (PE), phosphatidylinositol (PI), and phosphatidic acid (PA). Whereas PC and PE are neutral zwitterions, PS, PI, and PA bear a net negative charge (Fig. 1). In certain membrane regions, such as the inner leaflet of the plasma membrane, the levels of PS are sufficient to confer a substantial negative electrostatic potential near the membrane surface. A useful, albeit less realistic, model derived from physical organic studies depicts membranes as a uniform bilayer with smoothly varying dielectric properties ranging from nonpolar in the hydrocarbon core to polar, charged, or a combination of polar and charged at the surface (11). The variation in polarity is greatest in the "interfacial region" corresponding approximately to the head group, the phospho-glycerol-ester moiety, and the first few methylene groups of the hydrocarbon chains. This model, which has the practical advantage of representing the structural, compositional, and dynamic heterogeneity of membranes in terms of average physiochemical properties, provides a convenient framework for discussing the "nonspecific" contributions to membrane targeting in cases where a more detailed description is not required.

Fig. 1.

General properties of lipids and membranes. A few of the more abundant phospholipids are depicted with carbon atoms in white, nitrogen atoms in blue, and oxygen and phosphorous atoms in red. Phosphatidylcholine (PC) and phosphatidylethanolamine (PE) are neutral zwitterions; phosphatidylserine (PS), phosphatidylinositol (PI), and phosphatidic acid (PA) bear a net negative unit charge. Solid lines denote approximate boundaries between the nonpolar hydrocarbon core and the interfacial region (see text). Molecular volumes were calculated with DS ViewerPro (Accelrys). Except where otherwise noted, the figures in this review were rendered with PyMOL (156).

Mono- and polyphosphorylated derivatives of phosphatidylinositol, collectively referred to as phosphoinositides, play a disproportionately critical role as targets for the known lipid-binding domains. Several physiochemical properties that distinguish inositol from other common lipid head groups may well have contributed to the convergent evolution of functionally related yet structurally distinct phosphoinositide binding domains. With five equatorial hydroxyl substituents and a single axial hydroxyl group at the D2 position, the semi-rigid cyclohexane-based ring of D-myoinositol represents a prominent landmark against the backdrop of typically smaller head groups. Reversible phosphorylation of the D3, D4, and D5 hydroxyl groups transforms an otherwise weakly anionic phospholipid, with a net negative charge of –1, into seven distinct derivatives: three monophosphates, three bisphosphates, and a single trisphosphate, with high net negative charges of –3, –5, and –7, respectively. The extensive literature on phosphoinositide metabolism by lipid kinases and phosphatases is covered in two excellent reviews (12, 13). Given a high negative charge density, distributed over two to four phosphates in close proximity, it is perhaps not surprising that a strong positive electrostatic potential should be a common feature of the various domains that recognize phosphoinositides. What seems more remarkable is the high degree of stereochemical selectivity that some lipid-binding domains have evolved to distinguish among structurally similar phosphoinositides, in particular the pseudo-symmetrical monophosphoinositides PI(3)P and PI(5)P or the analogous bisphosphoinositides PI(3,4,)P2 and PI(4,5)P2.

Measuring Protein-Lipid Interactions

Various techniques have been used to measure in vitro interactions with lipids or head groups. These techniques differ not only in the method of detection but also with respect to the nature of the ligand. For example, the ligand may consist of the head group; a soluble lipid below the critical micelle concentration; pure or mixed micelles; pure lipids immobilized on a filter; or bilayer liposomes consisting of multilamellar vesicles (MLVs), small unilamellar vesicles (SUVs), or large unilamellar vesicles (LUVs). The merits and caveats of several commonly used methods are summarized briefly here. More in-depth discussion can be found elsewhere (14, 15).

One popular method assesses binding to pure lipids (typically phosphoinositides) immobilized on nitrocellulose filters (16). Though convenient as a preliminary screen, this method does not allow quantitative determination of affinity and specificity, nor does it approximate physiological lipid composition or membrane structure. Surface plasmon resonance (SPR) measurements using Biosensor chips allow the determination of affinities and, in principle, the kinetics of association with and dissociation from a continuous lipid-covered surface (17). Some caution with respect to the design and interpretation of SPR experiments is warranted, given known limitations, including the potential for mass transport-limited kinetics and off-rates limited by rebinding at high ligand density. Isothermal titration microcalorimetry provides quantitative solution affinities as well as enthalpies and entropies of binding; however, it is limited to interactions with head groups or soluble lipid analogs and consumes relatively large quantities of both protein and ligand (15). Finally, sedimentation of proteins with sucrose-loaded SUVs or LUVs permits a quantitative assessment of the partitioning of lipid-binding domains with bilayer vesicles of defined composition (14).

Although different methods often lead to similar conclusions regarding the relative affinity and specificity of lipid-binding domains, the potential for conflicting results has been documented. In general, methods that employ SUVs or LUVs approximating physiological lipid compositions and solution conditions are more likely to correlate with in vivo observations than are methods that measure binding to pure lipids.

The remainder of this review is concerned primarily with the modular domain families that recognize phospholipids and the often elaborate mechanisms that have evolved to facilitate selective membrane targeting. The mechanisms employed span a broad range, from interactions driven predominately by stereochemically specific recognition of particular head groups to nonspecific modes of interaction that can be correlated with generic physical properties such as net charge or exposed hydrophobic surface area. It may well be that the majority of membrane-targeting domains rely on a combination of specific and nonspecific interactions.

PH Domains

PH domains are one of the largest and most intensively investigated families of lipid-binding domains (18-20). They were first identified in numerous signaling, cytoskeletal, and metabolic proteins as evolutionarily conserved modules of ~120 amino acids with weak homology to pleckstrin, a PKC substrate in platelets (21-24). As estimated by the human genome project, there are over 250 proteins containing one or more PH domains, making them one of the most common domains (25, 26). Of the PH domains that have been characterized, the majority bind weakly to phosphoinositides with little or no selectivity. A subset representing 10 to 20% of PH domains exhibit relatively high affinity [the dissociation constant (Kd) for the head group is in the low μM to nM range] and varying degrees of specificity for polyphosphoinositides (15, 27-30). PH domains that recognize PI(3)P, PI(4)P, or the low-abundance bisphosphoinositide PI(3,5)P2 have also been identified (30).

Despite high sequence variability, nuclear magnetic resonance (NMR) and crystal structures of more than a dozen different PH domains have established a canonical core fold (Fig. 2A) consisting of a seven-stranded partly open β barrel, capped at one end by a C-terminal α helix (18, 31-39). Outside the core regions, the loops connecting the various secondary structural elements are best characterized as hypervariable with respect to composition, length, and structure, although similarities are apparent within subfamilies. As discussed below, the hypervariable loops play a critical role in determining the functional properties, in particular the diverse affinity and specificity for phosphoinositides. Nevertheless, one property characteristic of PH domains that bind phosphoinositides is a strongly dipolar electrostatic potential (Fig. 2B), with the positive lobe typically centered near the open end of the central β barrel (40, 41). This bulk electrostatic property accounts, at least in part, for the weak phosphoinositide affinities and specificities of many PH domains that correlate directly with the net charge of the head group (15). In these cases, preferences for phosphoinositides over other acidic lipids presumably derive from the higher negative charge density of mono- and polyphosphoinositides rather than stereochemical determinants. Even in cases where the affinities and specificities are dominated by stereospecific interactions with the head group, it is likely that the overall electrostatic potential plays an important role in establishing the orientation of the domain with respect to the membrane.

Fig. 2.

Structure and functional properties of PH domains. (A) Structural architecture of the Grp1 PH domain [Protein Data Bank (PDB) access number 1FGY]. The broadly conserved structural core is depicted in blue. Specificity determining regions (SDRs) corresponding to three hypervariable loops flanking the head group binding site are highlighted in green (β1/β2 loop), magenta (β3/β4 loop), and orange (β6/β7 loop). Also shown is the head group of PI(3,4,5)P3, with carbon and phosphorous atoms in yellow and oxygen atoms in red. (B) Electrostatic potential of the Grp1 PH domain contoured at +25 mV (blue) and –25 mV (red). A strongly dipolar electrostatic potential, with a prominent positive lobe, is a characteristic feature of the PH domains that bind phosphoinositides. The molecular surface and electrostatic potential in the absence of the head group were calculated with GRASP (157). (C) Head group recognition by the Grp1 PH domain. Note the extensive network of stereospecific hydrogen-bonding interactions contributed by residues from the signature motif (blue) and all three SDRs [colored as in (A)]. Bidentate interactions between adjacent phosphate groups and lysine residues are a frequently observed feature of polyphosphoinositide recognition.

A significant number of PH domains bind polyphosphoinositides with relatively high affinity and with specificities dependent on the arrangement of phosphate groups attached to the inositol ring (Table 1). These include the PI(4,5)P2-specific PLC-δ PH domain, as well as the PI(3,4,5)P3-specific PH domains of Bruton's tyrosine kinase (Btk) and general receptor for 3-phosphoinositides (Grp1) (15, 27-29, 42-44). The PH domains of dual adaptor for phosphotyrosine and 3-phosphoinositides (Dapp1) and the protein kinase B proto-oncogene (PKB, also known as Akt) are promiscuous for PI(3,4,)P2 and PI(3,4,5)P3, yet discriminate against PI(4,5)P2 (1, 15, 28, 29, 45-47). In a peculiar evolutionary twist, splice variants within the Grp1 family of PH domains, in which a single glycine residue is inserted at the N terminus of the β1/β2 loop, bind promiscuously to either PI(4,5)P2 or PI(3,4,5)P3 (48, 49). Several of these PH domains have the property of binding with higher affinity to the head group than to the corresponding lipid (15). At least for the PH domains of Grp1, Btk, Dapp1, PKB, and PLC-δ, which target proteins to the plasma membrane, the affinity for the head group appears to be sufficient to drive membrane association, although other interactions may well influence the precise localization within membrane microdomains.

Crystal structures of the aforementioned PH domains in complex with inositol polyphosphates provide insight into the determinants of phosphoinositide recognition (33, 34, 39, 50, 51). These PH domains conserve a basic "signature motif," K-Xm-(R/K)-X-R-Xn-(Y/N), with the first lysine located near the C terminus of the β1 strand, the (R/K)-X-R sequence near the N terminus of the β2 stand, and a tyrosine residue in the β3 strand (15, 28, 29, 33, 34). In a variation on the theme, the PKB PH domain substitutes the signature tyrosine with a functionally analogous asparagine residue from the β3/β4 loop (39). The first and third basic residues of the signature motif line the most deeply buried and positively charged region of the binding site and, together with the signature tyrosine or asparagine residue, mediate stereochemically equivalent interactions (Fig. 2C) with either the 3- and 4-phosphates (Grp1, Btk, Dapp1, and PKB) or the 4- and 5-phosphates (PLC-δ). Mutational analyses indicate that the signature residues, in particular the first and third basic residues, are critical, but not sufficient, for head group binding (28, 29, 42, 43, 52). With the exception of the PLC-δ PH domain, the majority of the interactions with the head group are mediated by basic and polar residues from three "specificity-determining regions" (SDRs) corresponding to the hypervariable β1/β2, β3/β4, and β6/β7 loops, which flank the phosphoinositide binding site at the open end of the β barrel. Main-chain NH groups in the β1/β2 loop mediate interactions with either the 5-phosphate (Btk and Grp1) or the 1-phosphate (PKB).

An important lesson from these studies is that similar specificities can be achieved through quite distinct structural mechanisms (Fig. 3). For example, the relatively long (11-residue) β1/β2 loop in the Btk PH domain accounts for all of the interactions with the 5-phosphate and half of the contacts with the 4-phosphate. In the Grp1 PH domain, a 20-residue insertion in the β6/β7 loop adopts a β-hairpin structure, which straddles the 4- and 5-phosphates, thereby compensating for a short (six-residue) β1/β2 loop. Equally significant structural variations are observed between the Dapp1 and PKB PH domains. Beyond an apparent evolutionary penchant for multiple solutions to the same problem, these observations beg an obvious but unresolved question: Do the dramatic structural variations, which preserve phosphoinositide recognition, belie additional functions related to, for example, higher order organization of signaling or trafficking complexes on membranes or the coupling of allosteric conformational changes to membrane targeting? A 3-phosphoinositide-dependent conformational change in PKB has been observed in cells; however, it is not clear whether this is a direct consequence of the interaction with membranes or a secondary consequence of 3-phosphoinositide-dependent phosphorylation (53). Consistent with the former interpretation, a head group-dependent conformational change has been observed in the isolated PKB PH domain (54).

Fig. 3.

Common and variable determinants of phosphoinositide recognition by PH domains. Conserved "signature residues" (beige) play an important role in the recognition of adjacent phosphates but do not distinguish between PI(3,4)P2, PI(4,5)P2, and PI(3,4,5)P3. Variable SDRs (green, magenta, and orange) determine the specificity for phosphoinositides. Although the orientation of these PH domains with respect to the membrane is not well characterized, the 1-phosphate of the head group depicted in each structure is expected to align with the corresponding 1-phosphate groups of the phospholipid bilayer. This would place the β1/β2 loop, and possibly the β3/β4 or β6/β7 loops, in apposition with the membrane. Though roughly similar, the viewpoints have been optimized independently for each PH domain to facilitate visualization of the head group. The structures depicted have the following PDB accession numbers: Btk [1B55], Grp1 [1FGY], PKB [1H10], Dapp1 [1FAO], and PLC-δ [1MAI].

Even though the PH domains cited above have achieved high affinity and specificity for particular phosphoinositide head groups, the necessity of recognizing the head group in the context of a lipid bilayer with a net negative surface potential requires electrostatic compatibility. Thus, altering the electrostatic potential of the membrane-proximal surface indirectly affects the affinity for phosphoinositides (55). For example, structure-function analyses of the PLC-δ PH domain have identified basic residues outside the binding pocket that are essential for membrane binding (52). The requirement for complementary electrostatic potentials probably explains the statistical preference for basic residues in the β1/β2 loop of PH domains that recognize 3-phosphoinositides, given that the majority of these residues do not appear to form specific interactions with the head group (33, 34, 39, 51).

It is noteworthy that the PH domain fold occurs in other domains, such as the phosphotyrosine-binding (PTB), enabled/VASP homology (EVH), and Ran-binding (RanBD) domains (18). Despite similar folds, the functions of these domains are quite distinct. PTB domains bind phosphotyrosine peptides; EVH domains bind polyproline peptides; and RanBD interacts with the conformational switch regions of Ran in a manner reminiscent of inositol polyphosphate interactions in PH domains, involving residues in the β1/β2 and β3/β4 loops. Thus, the PH domain superfold can be regarded as an evolutionarily conserved structural scaffold, which has been adapted for both protein-protein and protein-lipid recognition (41). Within the PH domain family, both lipid and protein interactions, in some cases mediated by the same PH domain, contribute to function. For example, the PH domain of the Dbs proto-oncogene protein binds weakly to phosphoinositides and also cooperates with the Dbl homology (DH) domain to catalyze nucleotide exchange on the monomeric guanosine triphosphatases (GTPases) RhoA and Cdc42; both interactions contribute to recruitment of Dbs to membranes, activation of Rho GTPases, and cellular transformation (38, 56, 57). Consequently, the membrane-targeting and exchange activities of Dbs are inextricably distributed over the DH-PH tandem.

Double Zinc-Finger Motifs

FYVE domains

FYVE domains are compact double Zn2+ finger modules consisting of approximately 70 residues (58-61). Proteins containing FYVE domains have been implicated in endocytic membrane trafficking, cytoskeletal regulation, and signal transduction (62). The amino acid sequences of FYVE domains are readily distinguished from those of other double Zn2+ fingers by a distinctive set of nearly invariant residues that includes three distal motifs: an N-terminal WXXD motif, a central R(R/K)HHCR motif, and a C-terminal RXC motif. The majority of FYVE domains characterized to date bind with relatively high affinity and specificity to membranes containing PI(3)P (63-65). In contrast, FYVE domains bind to the head group of PI(3)P with considerably lower affinity and moderate selectivity (66-68). Many proteins that contain FYVE domains exhibit a PI(3)P-dependent localization to early endosomes that can be partially or completely disrupted by the cell-permeable phosphatidylinositol 3-kinase (PI3K) inhibitor wortmannin (69-71). One of the best characterized of these proteins, early endosome antigen 1 (EEA1), interacts with the small GTPase Rab5 and has an essential function in early endosome fusion, in addition to a widespread practical application as a marker for early endosomes (72, 73).

The crystal structure of the unliganded FYVE domain of Vps27 (Fig. 4A), which functions in vacuolar protein sorting in yeast, revealed a fold distinct from double Zn2+ fingers, with a concave binding site lined with conserved basic and histidine residues (74). A pair of exposed hydrophobic residues at the tip of a short membrane insertion loop directly preceding the R(R/K)HHCR motif were proposed to partition into the membrane bilayer. Direct evidence in support of this hypothesis was provided by NMR studies, which demonstrated that the analogous loop in the FYVE domain of EEA1 partitions into dodecyl phosphatidylcholine micelles in the presence and absence of dibutyl PI(3)P (67, 75). Structural data on PI(3)P recognition by FYVE domains comes from NMR and crystallographic studies of the EEA1 FYVE domain (67, 68). The head group of PI(3)P is recognized entirely by conserved residues from the WXXD, R(R/K)HHCR, and RXC motifs. Residues from these motifs mediate stereochemically specific interactions with the 1- and 3-phosphates as well as the 4-, 5-, and 6-hydroxyl groups. When coupled in a structurally constrained configuration with stereoselective head group contacts, the partitioning of exposed hydrophobic residues is an important mechanism for enhancing both the affinity and specificity of PI(3)P-dependent membrane binding. This mode of PI(3)P recognition is likely preserved in most other FYVE domains, given the strong conservation of residues that contact the head group and compositional similarities in the membrane insertion loop. However, substantial variability in the hydrophobicity and basicity of the membrane insertion loop (compare, for example, the SVTV sequence in the EEA1 FYVE domain with the TFTK sequence in the SARA FYVE domain) suggests that the energetic contribution from membrane partitioning, and consequently the ability to target membranes, is likely to vary substantially within the FYVE domain family.

Fig. 4.

FYVE domain structure and determinants of membrane targeting. (A) Crystal structure of the Vps27 FYVE domain [PDB accession number 1VFY] showing exposed leucine residues in a loop proposed to insert into the interfacial region of the bilayer (74). This hypothesis is supported by NMR studies on the EEA1 FYVE domain (67). The partitioning of exposed hydrophobic residues into the interfacial region is a common feature of many membrane-binding domains. (B) Crystal structure of the homodimeric C-terminal membrane-targeting region of EEA1 in complex with the head group of PI(3)P [PDB accession number 1JOC]. The coiled coil proximal to the FYVE domain provides the driving force for dimerization and imposes an organized quaternary structure that is ideally configured for bivalent PI(3)P binding. The single unique orientation that aligns the 1-phosphate groups in the structure with those of a model membrane places the membrane insertion loop identified in NMR studies of the EEA1 FYVE domain (67) within the interfacial region of the membrane.

Although isolated FYVE domains bind selectively to liposomes containing PI(3)P in vitro, they are not necessarily sufficient for membrane targeting in cultured cells. For example, the FYVE domains of EEA1 and Hrs are largely cytosolic, whereas the full-length proteins localize to early endosomes in a manner dependent on PI3K activity (76-79). The minimal region of EEA1 required for endosomal targeting consists of the FYVE domain and approximately 60 residues of heptad repeat proximal to the FYVE domain (58, 76). Whereas the FYVE domain of EEA1 is monomeric, the membrane-targeting region forms a stable homodimer, as does the full-length protein (68, 80). The crystal structure of the membrane-targeting region in complex with the head group of PI(3)P (Fig. 4B) reveals an organized quaternary structure with a dyad symmetric FYVE domain dimer optimally configured for bivalent PI(3)P binding (68). Simultaneous interaction with two PI(3)P head groups requires a unique orientation that places the membrane insertion loops within the interfacial region of the bilayer, in agreement with the NMR studies on micelles and mutational data on the interaction of the EEA1, Vps27, and Hrs FYVE domains with liposomes and cellular membranes (75, 81). Electrostatic calculations predict a preferred orientation for most FYVE domains consistent with that inferred from the structure of the EEA1 homodimer (82). Increased avidity afforded by bivalent PI(3)P recognition provides a straightforward explanation for the strict requirement for the proximal coiled coil. Finally, two lines of evidence favor the bivalent binding model: (i) Mutations in the proximal coiled coil that disrupt a weak interaction with the Rab5 GTPase have no apparent effect on membrane targeting (78), and (ii) the requirement for the coiled coil can be bypassed in a tandem construct that fuses the N terminus of one EEA1 (or Hrs) FYVE domain to the C terminus of another (79). A tandem construct of the Hrs FYVE domain has been used to map the intracellular distribution of PI(3)P (79). A naturally occurring FYVE domain tandem has also been implicated in endosomal localization (83).

PHD domains

The PHD motif is a C4HC3-type zinc finger found in a large number of chromatin regulatory factors and is widely distributed throughout the eukaryotic proteome. An NMR structure of the PHD domain from the Williams-Beuren syndrome transcription factor (WSTF) revealed a fold with striking similarity to that of the FYVE domain (84). Recently, the PHD domain of the candidate tumor-suppressor protein ING2 was shown to bind specifically to PI(5)P (85). Basic residues implicated in PI(5)P recognition are also required for the function of ING2 in the regulation of p53-dependent apoptotic pathways. Several other PHD domains show selective binding in dot blot assays to PI(5)P or other mono- or polyphosphoinositides. Intriguingly, the chromatin regulatory function and localization of PHD domain proteins suggest that the relevant phosphoinositide pools are likely to be nuclear.

PX Domains

PX domains were originally identified in the p40phox and p47phox subunits of the phagocyte NADPH oxidase (phox) complex and have since been implicated in diverse cellular processes, from the regulation of cell polarity to yeast vacuolar morphology (86-89). With 65 human sequences in the protein families (Pfam) database, PX domains constitute a sizable family of phosphoinositide-binding domains (90). As with PH domains, variability at the amino acid sequence level is reflected in the phosphoinositide-binding properties. A number of PX domains bind PI(3)P with sufficient affinity and specificity to direct localization to early endosomes or nascent phagosomes (91-96). Other PX domains bind preferentially to polyphosphoinositides (92, 95, 97). For example, the PX domains of p47phox and TCGAP (TC10/cdc42 GTPase activating protein) bind PI(3,4)P2 and PI(4,5)P2, respectively.

Although all PX domains in Saccharomyces cerevisiae bind selectively to PI(3)P, two distinct groups are evident: a high-affinity class (apparent Kd ~ 2 to 3 μM) and a low-affinity class (apparent Kd ≥ 100 μM) (98). It is unlikely that the low-affinity PX domains would be capable of driving membrane localization as isolated modules. However, it appears that the weak intrinsic avidity for PI(3)P can be amplified through oligomerization mediated by external regions. Consistent with this hypothesis, sorting nexins (Snx's) with low-affinity PX domains are known to form homo- or heterodimers, whereas those with high-affinity PX domains are typically monomeric (99-102). Snx's have been implicated in membrane trafficking and protein sorting (103). With respect to the lipid specificity of PX domains, it is noteworthy that Snx1 appears to bind PI(3,4,5)P3 in a filter-binding assay, whereas it binds specifically to PI(3)P and PI(3,5)P2 in a liposome-binding assay. The endosomal localization and lysosomal sorting functions of Snx1 correlate with the latter specificity (104).

PX domains have a general architecture consisting of a helical subdomain nestled against a three-stranded antiparallel β sheet (105-108). Despite a similar overall fold, crystal structures of the p40phox PX domain in complex with PI(3)P and the unliganded form of the p47phox PX domain (Fig. 5) reveal structural variations in the phosphoinositide-binding sites that help to explain the observed specificities (106, 107). In the case of p40phox, the PI(3)P head group is recognized by basic residues that contact the 1- and 3-phosphates as well as the 4- and 5-hydroxyl groups. Unfavorable steric clashes would prevent the binding of other phosphoinositides in the same orientation. Although the details differ, FYVE domains also combine PI(3)P recognition with steric selection against other phosphoinositides (68). A more complete picture of phosphoinositide recognition in PX domains awaits structural data on the appropriate complexes with polyphosphoinositides.

Fig. 5.

PX domain structure and determinants of membrane targeting. (A) Crystal structure of the p40phox PX domain in complex with dibutyl PI(3)P. The head group recognition pocket is located between a three-stranded antiparallel β sheet (orange) and a helical subdomain (blue). (B) Electrostatic potential at the surface of the p40phox PX domain in the range from –25 to +25 kbT, where kb is Boltzmann's constant and T is temperature. Labels denote exposed nonpolar residues implicated in membrane association (107, 109). These residues are predicted to partition into the interfacial region of the bilayer. (C) Electrostatic potential at the surface of the unliganded p47phox PX domain in the range from –25 to +25 kbT. Sulfate ions (green sulfur and red oxygen atoms) from the crystallization medium occupy the putative PI(3,4)P2 head group binding site (right) and secondary anion-binding site (left). Exposed nonpolar residues flanking the primary head group binding site have been implicated in membrane binding and are thought to partition within the interfacial region. The PDB accession numbers for the structures depicted in this figure are p40phox [1H6H] and p47phox [1O7K]. Molecular surfaces and electrostatic potentials were calculated with GRASP (157).

In addition to stereochemically specific interactions with the head group, the partitioning of exposed nonpolar residues into the membrane bilayer plays a critical role in orienting and targeting PX domains to membranes (107, 109). Mutation of hydrophobic residues in a putative membrane insertion loop of the p40phox and p47phox PX domains results in significantly decreased affinity for phosphoinositides that correlates with reduced monolayer penetration (109). As with FYVE domains, both stereospecific interactions with the head group and the partitioning of nonpolar residues contribute to high-affinity membrane targeting.

In addition to the primary phosphoinositide-binding site, the unliganded structure of the p47phox PX domain (Fig. 5C) reveals a second anion-binding site adjacent to the membrane insertion loop (107). The second anion-binding site, which is not observed in the p40phox or Vam7 PX domains, binds preferentially to either PS or PA. Simultaneous occupation of the PI(3,4)P2-binding pocket and the secondary anion-binding site is supported by the observation of synergistically enhanced affinity for liposomes containing both phospholipids. Thus, the PX domain of p47phox couples stereoselective and nonstereoselective modes of head group recognition with the partitioning of hydrophobic residues to achieve high-affinity and high-specificity membrane targeting. Lastly, it is interesting to note that phosphorylation of the PX domain of phospholipase D1 (PLD1) by PKC significantly reduces phospholipase activity (110). Furthermore, some PX domains, such as the p47phox PX domain, contain a polyproline sequence consistent with the consensus for binding to Src homology 3 (SH3) domains (105). These observations raise the possibility that the association of PX domains with membranes might be regulated by posttranslational modification or protein-protein interactions.

C2 Domains

Initially identified in PKC, the C2 domain represents a relatively large family with some 225 human sequences in the Pfam database (90, 111). C2 domains share a common fold in which an eight-stranded antiparallel β sandwich scaffolds several adjacent loops that define a membrane-binding surface (55, 112). These loops exhibit considerable variability and often contain key acidic residues that reversibly coordinate two or three Ca2+ ions. In contrast to the other domains discussed above, C2 domains exhibit relatively weak selectivity for particular lipid head groups, so that the extent of membrane partitioning is determined primarily by the bulk electrostatic properties of the membranes.

The interaction of C2 domains with membranes can be largely explained in terms of the Ca2+-dependent or -independent electrostatic properties of the variable loops (113). Consider, for example, the dramatic "electrostatic switch" (114), from a negative to a positive potential, displayed by the C2A domain of synaptotagmin (Syt1-C2A), a regulator of synaptic vesicle fusion, in response to Ca2+ binding (Fig. 6). In the absence of Ca2+, the Syt1-C2A domain does not partition favorably with membranes of any composition. In the presence of Ca2+, however, the electropositive surface of the Syt1-C2A domain drives nonspecific association with anionic but not neutral membranes (115). The Syt1-C2B domain also exhibits Ca2+-dependent phospholipid binding, suggesting that both C2 domains in Syt1 contribute to a common function in membrane targeting and neurotransmitter release (116, 117). Electrostatic calculations indicate that several C2 domains of known structure have Ca2+-dependent potentials consistent with an analogous electrostatic switch (113). In a variation on this theme, the C2 domain of cytoplasmic phospholipase A2 (cPLA2) undergoes a Ca2+-dependent switch from a negative to a neutral potential, allowing nonspecific association with neutral zwitterionic membranes, the driving force for which is provided by the partitioning of exposed nonpolar residues into the bilayer (118-120). Although other C2 domains exhibit Ca2+-independent membrane binding, the general modes of interaction appear to be analogous. As a case in point, the C2 domain of the PTEN tumor suppressor (a protein and lipid phosphatase), which preferentially associates with anionic membranes, has a positive potential similar to that of the Ca2+-bound form of the Syt1-C2A domain (121). In addition, the membrane association of PTEN may be negatively regulated by phosphorylation of the C-terminal tail region (122). Finally, Ca2+ coordination by the carboxylate group of PS in the crystal structure of the PKCα C2 domain accounts for the weak specificity for PS observed in some C2 domains (123, 124). Further discussion of membrane targeting by C2 domains, as well as the C1 domain, can be found in two excellent reviews (55, 125).

Fig. 6.

Electrostatic properties and membrane-binding mechanisms of C2 domains. The membrane-proximal (top) surface of the Syt1 C2 domain [PDB accession number1BYN] switches from a negative electrostatic potential in the absence of Ca2+ to a positive electrostatic potential in the presence of Ca2+. The analogous surface of the cPLA2 C2 domain [PDB accession number 1RLW] undergoes a Ca2+-dependent switch from a negative to a neutral potential. The Ca2+-bound forms of the Syt1 and cPLA2 C2 domains associate nonspecifically with negatively charged and neutral membranes, respectively. The driving force for the latter reflects the partitioning of exposed nonpolar residues within the interfacial region. Several non-Ca2+ binding C2 domains have electrostatic potentials similar to those of the Ca2+-bound forms of the Syt1 or cPLA2 C2 domains. The orientation of the Ca2+-bound C2 domains with respect to the interfacial region (IR) of the phospholipid bilayer (solid lines) corresponds to that predicted by electrostatic calculations (113). Molecular surfaces and electrostatic potentials were calculated with GRASP (157).

Other Lipid-Binding Domains and Motifs

FERM domains

ERM family proteins, which contain a FERM (acronym of band four.1, ezrin, radixin, moesin) domain, serve as regulated cross-linkers connecting actin filaments to membranes (126). The interaction with membranes is mediated by the FERM domain, which binds PI(4,5)P2 as well as the polybasic tails of integral membrane proteins (127, 128). Crystallographic studies of FERM domains reveal a composite cloverleaf fold with clearly discernable subdomains (127, 129-133). The three subdomains, labeled A, B, and C, have folds similar to ubiquitin, acyl-coenzyme A-binding protein, and the PH domain, respectively. A co-crystal structure of the radixin FERM domain bound to the head group of PI(4,5)P2 (Fig. 7A) revealed an unexpected location for the phosphoinositide-binding site in a basic cleft between subdomains A and C (129). An extended flat surface flanking either side of the binding cleft contains several clusters of basic residues, giving rise to a strongly dipolar electostatic potential reminiscent of other lipid-binding domains. Mutation of basic residues in the binding cleft or adjacent electropositive surface impairs PI(4,5)P2 binding, membrane targeting, or both (134, 135).

Fig. 7.

Structure and membrane-targeting mechanisms of FERM and ENTH domains. (A) Crystal structure of the radixin FERM domain in complex with Ins(1,4,5)P3 [PDB accession number 1GC6]. The head group binding site is located between subdomains A and C. Although not shown here, the binding site and flat surface on either side have a strongly positive electrostatic potential. (B) Crystal structure of the epsin1 ENTH domain in complex with the head group of PI(4,5)P2 [PDB accession number 1H0A]. In the unliganded structure, the N-terminal helix (orange) is disordered. In the complex, it adopts an ordered conformation with nonpolar residues (blue) on the outside and polar residues facing the head group binding site. Although not shown, basic residues from the polar surface of the helix mediate interactions with all three phosphates of the head group, explaining the ordered conformation of the helix in the liganded form. The exposed nonpolar residues are thought to induce membrane curvature by insertion into the bilayer (150, 153).


In addition to globular lipid-binding domains, unstructured peptides that target membranes through nonspecific interactions have been identified (136). A well-characterized example is a peptide motif from the MARCKS (myristoylated alanine-rich C kinase substrate) protein, a ubiquitous PKC substrate present at high cellular levels (137). The association of MARCKS with membranes requires N-terminal myristoylation, as well as a region of ~24 amino acids, termed the effector domain (ED), which contains 13 basic residues interspersed with phenylalanine and other nonpolar residues (138, 139). The MARCKS ED binds weakly to neutral membranes but exhibits high affinity for acidic membranes containing physiological concentrations of PI(4,5)P2, PI(3,4)P2, or PS (140). The MARCKS ED binds membranes in an extended conformation in which the hydrophobic residues partition into the bilayer, whereas the basic side chains interact nonspecifically with negatively charged head groups (138). Although the hydrophobic residues also contribute, membrane association of the MARCKS ED is driven primarily by electrostatic interactions with the basic residues (141). Consistent with these observations, 13-residue peptides of polylysine or polyarginine bind to vesicles containing PI(4,5)P2 with affinities comparable to the MARCKS ED. Interestingly, the MARCKS ED binds two to three PI(4,5)P2 molecules, suggesting that it may function as a phosphoinositide buffer by sequestering a significant proportion of the PI(4,5)P2 in the plasma membrane. Finally, the association of MARCKS with membranes can be reversed by PKC phosphorylation or the binding of Ca2+/calmodulin to the MARCKS ED (142).


The Tubby transcription factor contains a 260-amino acid domain that binds nonspecifically to phosphoinositides that are phosphorylated on adjacent positions (143). Consistent with this observation, Tubby associates constitutively with the plasma membrane and is released in response to PI(4,5)P2 hydrolysis after activation of PLC-β (144). The phosphoinositide-binding pocket is located at the end of a basic groove proposed to bind DNA (145). Dual interactions with a lysine residue in the binding pocket appears to be a key determinant underlying the requirement for adjacent phosphates (143). This contribution to bisphosphate recognition parallels that of the signature lysine in PH domains.

ENTH domains and membrane curvature

Beyond targeting and recruitment, protein-lipid interactions play an important role in membrane remodeling. For example, vesicle formation during endocytosis at the plasma membrane, as well as vesiculation and tubulation of the endoplasmic reticulum, involve proteins that bind lipids and influence membrane curvature. Clathrin-mediated endocytosis requires several cytosolic factors, including epsin1 as well as the coat components clathrin, eps15, and the AP2 complex (146-148). Targeting of epsin1 to sites of endocytosis on the plasma membrane is mediated by an epsin N-terminal homology (ENTH) domain, which binds PI(4,5)P2 and PI(3,4,5)P3 (149-151). Crystal structures of the unliganded and Ins(1,4,5)P3-bound forms of the epsin1 ENTH domain provide insight into the mechanisms through which membrane curvature may be coupled to membrane binding (150, 152). The unliganded form consists of a superhelical arrangement of seven helices, with an eighth helix parallel to the superhelix axis. However, in the Ins(1,4,5)P3-bound form, the otherwise disordered N terminus forms an additional helix flanking the phosphoinositide-binding site (Fig. 7B). The N-terminal helix has an inverted polarity reminiscent of integral membrane proteins, with hydrophobic residues on the exterior surface and polar residues facing the phosphoinositide-binding site. Several basic residues from the interior surface mediate interactions with 1-, 4-, and 5-phosphates. Mutational studies support the hypothesis that the hydrophobic surface of the N-terminal helix penetrates into the bilayer. Lipid displacement resulting from insertion of the N-terminal helix likely contributes to the mechanism by which ENTH domains induce membrane tubulation (153).


The interaction of protein domains with lipid ligands differs fundamentally from other classes of macromolecular recognition, a distinction that is readily apparent in the repeating structure and physiochemical characteristics of lipid bilayers. To achieve selective membrane targeting, lipid-binding domains must by definition contend with or take advantage of the bulk properties of phospholipid bilayers. One solution to the problem is the evolution of domains with high affinity and selectivity for particular lipid head groups (Fig. 8A). Representative of this category is the subset of PH domains that recognize polyphosphoinositide head groups through stereospecific networks of electrostatic and hydrogen-bonding interactions. Although higher affinity for the head group in the absence of membranes is a hallmark of these PH domains, the necessity of accessing the head group in the context of a lipid bilayer requires that the electrostatic properties of the surfaces adjacent to the binding site be complementary to those within the interfacial region of the membrane.

Fig. 8.

Schematic summarizing general modes of lipid recognition and membrane targeting. (A) Stereospecific head group recognition. Although the affinity for the head group may be sufficient to drive membrane localization, the surfaces adjacent to the binding site must have a polarity complementary to that of the membrane. (B) Nonspecific modes of membrane binding driven by favorable electrostatic interactions and/or the partitioning of nonpolar residues within the interfacial region of the bilayer. (C) Modes of membrane targeting that combine stereoselective head group recognition with nonspecific membrane interactions or multivalent head group binding.

Table 1.

Representative lipid binding specificities.

At the other extreme are domains and peptide motifs that target membranes on the basis of bulk electrostatic properties (Fig. 8B). C2 domains, many PH domains, and various peptide motifs, including the MARCKS ED, rely on nonstereospecific modes of interaction. The partitioning of exposed hydrophobic surfaces into the hydrocarbon core, the favorable interaction between electropositive surfaces and negatively charged membranes, or a combination of these provides the driving force for membrane association. Weak specificities observed in certain cases correlate with either the charge of the head group or a limited number of stereoselective contacts.

Many, if not the majority, of lipid recognition domains employ a combination of stereospecific and nonspecific interactions, including PH, FYVE, PX, FERM, and ENTH domains (Fig. 8C). One might reason that such hybrid mechanisms would imply an intermediate degree of specificity. However, if the regions involved in specific and nonspecific interactions are structurally constrained, rather than connected by flexible tethers, then the net interaction may be optimal for one head group and not others. In this case, nonspecific interactions would have the capacity to amplify the intrinsic specificity for the head group. The requirement for a rigid structural linkage between the specific and nonspecific interaction elements appears to be satisfied in the liganded structures of a number of lipid domains. At least in the case of FYVE domains, the observed specificity is significantly higher for the membrane-bound lipid as compared with the isolated head group. The extent to which this mechanism contributes to the specificity of other lipid-binding domains awaits further characterization.

The examples highlighted in this review, as well as many equally important studies not addressed here, provide useful insights into the principles that govern lipid recognition. As a general rule, domains that selectively target membranes couple stereospecific head group recognition with nonspecific or weakly specific electrostatic interactions, with or without contributions from the partitioning of exposed hydrophobic residues into the interfacial region of the phospholipid bilayer. As with other macromolecular recognition problems, however, the devil lies in the details. In the PH, PX, and C2 families, the lipid affinities and specificities are determined by loop regions with variable length, sequence, and structure. Even in FYVE domains, where the residues that contact the head group are strongly conserved, compositional variability in the membrane insertion loop provides the basis for functional distinction. Moreover, external regions, such as the coiled coils in EEA1 and Snx's, can greatly enhance the avidity for membranes through multivalency arising from either homo- or hetero-oligomerization.

Many signaling and trafficking proteins contain one or more modular domains that are known, or have the potential, to interact with membranes, in addition to domains implicated in interactions with proteins, peptide motifs, or nucleic acid. How the myriad of protein-lipid, protein-protein, and protein-nucleic acid interactions are integrated to achieve the dynamic higher-order function characteristic of cellular signaling and trafficking systems remains one of the most challenging and important questions in cell biology.


  1. 1.
  2. 2.
  3. 3.
  4. 4.
  5. 5.
  6. 6.
  7. 7.
  8. 8.
  9. 9.
  10. 10.
  11. 11.
  12. 12.
  13. 13.
  14. 14.
  15. 15.
  16. 16.
  17. 17.
  18. 18.
  19. 19.
  20. 20.
  21. 21.
  22. 22.
  23. 23.
  24. 24.
  25. 25.
  26. 26.
  27. 27.
  28. 28.
  29. 29.
  30. 30.
  31. 31.
  32. 32.
  33. 33.
  34. 34.
  35. 35.
  36. 36.
  37. 37.
  38. 38.
  39. 39.
  40. 40.
  41. 41.
  42. 42.
  43. 43.
  44. 44.
  45. 45.
  46. 46.
  47. 47.
  48. 48.
  49. 49.
  50. 50.
  51. 51.
  52. 52.
  53. 53.
  54. 54.
  55. 55.
  56. 56.
  57. 57.
  58. 58.
  59. 59.
  60. 60.
  61. 61.
  62. 62.
  63. 63.
  64. 64.
  65. 65.
  66. 66.
  67. 67.
  68. 68.
  69. 69.
  70. 70.
  71. 71.
  72. 72.
  73. 73.
  74. 74.
  75. 75.
  76. 76.
  77. 77.
  78. 78.
  79. 79.
  80. 80.
  81. 81.
  82. 82.
  83. 83.
  84. 84.
  85. 85.
  86. 86.
  87. 87.
  88. 88.
  89. 89.
  90. 90.
  91. 91.
  92. 92.
  93. 93.
  94. 94.
  95. 95.
  96. 96.
  97. 97.
  98. 98.
  99. 99.
  100. 100.
  101. 101.
  102. 102.
  103. 103.
  104. 104.
  105. 105.
  106. 106.
  107. 107.
  108. 108.
  109. 109.
  110. 110.
  111. 111.
  112. 112.
  113. 113.
  114. 114.
  115. 115.
  116. 116.
  117. 117.
  118. 118.
  119. 119.
  120. 120.
  121. 121.
  122. 122.
  123. 123.
  124. 124.
  125. 125.
  126. 126.
  127. 127.
  128. 128.
  129. 129.
  130. 130.
  131. 131.
  132. 132.
  133. 133.
  134. 134.
  135. 135.
  136. 136.
  137. 137.
  138. 138.
  139. 139.
  140. 140.
  141. 141.
  142. 142.
  143. 143.
  144. 144.
  145. 145.
  146. 146.
  147. 147.
  148. 148.
  149. 149.
  150. 150.
  151. 151.
  152. 152.
  153. 153.
  154. 154.
  155. 155.
  156. 156.
  157. 157.
View Abstract

Stay Connected to Science Signaling

Navigate This Article