Structure of a Multifunctional Protein

Eukaryotic phosphatidylinositol transfer protein is a ubiquitous multifunctional protein that transports phospholipids between membrane surfaces and participates in cellular phospholipid metabolism during signal transduction and vesicular trafficking. The three-dimensional structure of the α-isoform of rat phosphatidylinositol transfer protein complexed with one molecule of phosphatidylcholine, one of its physiological ligands, has been determined to 2.2 Å resolution by x-ray diffraction techniques. A single β-sheet and several long α-helices define an enclosed internal cavity in which a single molecule of the phospholipid is accommodated with its polar head group in the center of the protein and fatty acyl chains projected toward the surface. Other structural features suggest mechanisms by which cytosolic phosphatidylinositol transfer protein interacts with membranes for lipid exchange and associates with a variety of lipid and protein kinases.

Phosphatidylinositol transfer proteins (PITPs) 1 constitute a highly conserved family of proteins in multicellular animals, with members in species ranging from Homo sapiens to Dictyostelium discoideum (reviewed in Refs. 1-2, see also Refs. [3][4][5]. PITPs catalyze an exchange of phosphatidylinositol (PtdIns) or phosphatidylcholine (PtdCho) molecules between the surfaces of biological or artificial membranes. The protein appears to be multifunctional, with cellular functions ranging from intermembrane phospholipid transport to possible substrate presentation to lipid-modifying enzymes. Exogenous PITP is able to restore hormone-stimulated G protein-coupled activation of PtdIns(4,5)P 2 hydrolysis in human promyelocytic cells (6) and to reconstitute Ca 2ϩ -stimulated, ATP-dependent secretion of catecholamines in rat pheochromocytoma cells (7). A variety of cultured mammalian cells respond to peptide agonists, epidermal growth factor, and antigens for the IgE receptor with a PITP-dependent stimulation of PtdIns 4-kinase and phospholipase C (8 -9). PITP and PtdIns 3-kinase together exhibit a capacity to phosphorylate PtdIns to PtdIns(3)P (10). In cell-free systems the budding of vesicles from the trans-Golgi network and vesicle transport between cis and medial Golgi compartments are both stimulated by PITP (11)(12). Existence of PITPcontaining multiprotein complexes has been proposed, poised to respond to extracellular stimuli and to channel PtdIns efficiently from PITP to kinases and phospholipases for the generation of signal transducing molecules and regulation of vesicular trafficking (13)(14).
Multiple isoforms of cytosolic, 32-kDa PITPs are common and ubiquitous (15)(16). Mammalian PITP␣ and PITP␤ share nearly 80% sequence identity; both isoforms are capable of forming noncovalent and stoichiometric complexes with PtdIns or PtdCho and transporting these lipid substrates through an aqueous environment. Sequence identity is more than 98% among mammalian PITP␣ isoforms (Fig. 1). PITP␣ and PITP␤ represent unique gene products (17p13.3 and 22q12.1, respectively, in humans) and they exhibit differential tissue expression. The N terminus of the 140-kDa membrane-associated retinal degeneration B proteins (rdgBs) contains a region homologous to the soluble PITPs (17)(18)(19)(20). Significantly, the rdgB-PITP region, comprising 266 amino acids, can be expressed as a soluble protein that exhibits both PtdIns and PtdCho transfer activities. We have solved the structure of rat PITP␣ bound to phosphatidylcholine using x-ray diffraction techniques. The results provide a framework for understanding the interactions of PITPs with lipid molecules and membrane surfaces in lipid transport processes and with other cellular proteins during signaling and trafficking events.

EXPERIMENTAL PROCEDURES
Protein Preparation and Crystallization-There were two criteria that guided all crystallization experiments: use of unmodified protein and maintenance of a protein-bound ligand of physiological significance. Considerable biochemical data have accumulated indicating that the C-terminal residues of PITP exhibit a conformational change upon binding to membranes. We therefore avoided the use of truncated or tagged proteins, particularly at the C-terminus. To improve the likelihood that a bound phospholipid would be observed in the x-ray data, no detergents of any kind were employed in any step of purification or crystallization. Cells of Escherichia coli B834(DE3), a methionine auxotroph containing plasmids for expression of groELS and rat PITP␣ (21), were grown in chemically defined media at 20°C with selenomethionine in place of methionine (22). The selenomethionyl-substituted protein was purified by established procedures for the wild-type protein and the bound bacterial phosphatidylglycerol (PtdGro) was replaced with sn-1,2-dioleoyl-PtdCho (21). The protein was crystallized under similar conditions as the wild-type protein when supplemented with seeding (23).
Structure Determination and Analysis-For x-ray data collection, crystals were mounted from the crystallization drop in a fiber loop and frozen directly in a cold nitrogen stream using an ADSC cryostat.
Frozen crystals were sent to NSLS beamline X12C for "Fed-Ex Data Collection" on a Brandeis B4 CCD detector. The crystal structure of the PITP␣⅐PtdCho complex was solved by multiwavelength anomalous diffraction (MAD) methods (24). A three-wavelength MAD x-ray data set was collected, with one wavelength at the selenium absorption edge, one at the absorption peak, and one at a low energy remote (Table I). Five of the eight selenium sites were identified by SOLVE (25) and refined by SHARP (26). The resulting electron density map was modified by solvent flipping as implemented in SOLOMON (27). From the resulting electron density map, the dominant secondary structure elements were traced. The final model was built through several cycles of manual model building with O (28), followed by refinement with CNS (29). The program DALI (30) was used to determine structurally related proteins. The lipid cavity was defined and volumes calculated using the CAST (31) web-server.

RESULTS
The protein features a large concave ␤-sheet and several long ␣-helices (Fig. 2). The eight-stranded ␤-sheet, strands 1 to 8, has a connectivity of 71865432. The ␤-sheet is antiparallel, except for the orientation of strands 1 and 7, which are parallel. There are three long ␣-helices, composed of 5 to 7 turns (A, F, and G), and four shorter ␣-helices, each with less than two complete turns (B, C, D, and E). Helices A and F are oriented on the interior side of the ␤-sheet, with helix A aligned coincident with strand 2 and helix F oriented diagonally across the ␤-sheet. The phospholipid is clearly visible in the electron density map (Fig. 3), with the binding site between the interior side of the ␤-sheet and ␣-helices A and F.
Several regions of the electron density map were difficult to model. These areas consisted of residues 239 -242 connecting helix F to G (containing one of the undetected selenium sites at Met 241 ) and residues 263-271 at the C terminus. The solventexposed side chain density of residues on helix G is somewhat weak, although the interior, buried sidechain density is unambiguous. The solvent-exposed Met 247 has very weak side chain density and is one of the undetected selenium sites from the MAD data. The last undetected selenium site is residue 1, for which no electron density is observed.
Three Functional Regions of PITP-We designate the dominant region of the PITP structure as the lipid-binding core. This region, a cavity in which the phospholipid molecule binds, is formed by ␣-helices A and F and the ␤-sheet. Limited segments of helices A and F and strands 4 and 5 are amphipathic and provide excellent hydrophobic interaction with the phospolipid molecule (Fig. 4). The lipid-binding core comprises residues 1-118 and 191-232. Examination of the PITP structure reveals that the intervening residues 119 -190 form an exten-  sive loop region between strands 7 and 8. Postulating that this loop is likely required for the association of PITP with a wide variety of lipid-and protein-modifying enzymes, we refer to this region as the regulatory loop. The recent identification of Ser 166 as a site for limited protein kinase C-dependent phosphorylation (32) supports this hypothesis. All PITPs and rdgBs studied to date contain a regulatory loop region; only Caenorhabditis elegans PITP lacks the equivalent Ser 166 phosphorylation site. The remaining residues in the structure comprise the C-terminal residues 233-271. Consisting of the long ␣-helix G as the principal secondary structural element and an additional 11 residues that limit access to the lipid-binding cavity, we call this region the C-terminal region and suggest its critical participation in membrane association and dissociation. The C-terminal region is the least conserved region among the cytosolic PITPs and is replaced in the rdgB proteins by up to 1000 additional amino acids, some of which appear to be transmembrane. Our functional region designation correlates reasonably well to the location of conserved amino acid residues in the PITP family (Fig. 1). The majority (nearly 75%) of the absolutely conserved amino acids are located in the lipid-binding core. Most of the others are part of the regulatory loop. Consistent with their proposed functions, both the regulatory loop and the C-terminal region were predicted to exhibit significant polypeptide backbone flexibility (33), a structural feature corroborated by specific B factors from the x-ray data.
Characterization of the Lipid-binding Cavity-The residues in close proximity to the PtdCho head group are shown in Figs. 3 and 4. The majority of these interacting residues are highly conserved in all PITP and rdgB proteins. Nine water molecules are within 3.2 Å of the phospholipid or an amino acid residue lining the phospholipid cavity. The phosphate oxygen O1P forms hydrogen bonds with the NZ atom of the conserved Lys 195 and a water molecule, and oxygen O 2 P hydrogen bonds with the OG1 atom of the conserved Thr 97 and another water molecule. The solvent-accessible volume of the phospholipidbinding cavity (calculated with a 1.4 Å probe) is 670 Å 3 , whereas the molecular-surface volume is 2725 Å 3 . The van der Waals volume of the PtdCho molecule is 671 Å 3 . Of the 271 residues in the protein, 72 residues (27%) have atoms contributing to the lipid cavity. DISCUSSION Characteristics of the interactions between PITP and the hydrophobic region of PtdIns and PtdCho molecules have been described (34 -35). Using lipids with defined fatty acyl species at their sn-1 and sn-2 positions, combined with one or more fluorescent acyl derivatives (pyrenylacyl, parinaroyl), both binding to PITP and protein-catalyzed transfer from a vesicle surface were measured. For the sn-1 group a broad range of structures could be accommodated, with preference for 14-and 16-carbon saturated or monounsaturated fatty acids; nevertheless, 18-carbon polyunsaturated fatty acids were still bound and transferred. In contrast, the sn-2 group exhibits a more restricted range, preferring 16-and 18-carbon mono-and polyunsaturated fatty acids whose double bonds were toward the methyl-end of the lipids. PtdCho molecules containing arachidonate (20:4n-6) or docosahexaenoate (22:6n-3) in the sn-2 position bind weakly to PITP. It was suggested that the sn-2 binding site in PITP was "bottle-shaped," with the neck toward the ester linkage. Little difference was observed for acyl chain preferences between the PtdIns and PtdCho class of phospholipids.
In the PITP structure reported here, the PtdCho ligand contains two oleate (18:1nϪ9) acyl groups. The conformation of the sn-2 acyl chain is more extended than the sn-1 chain, such that the sn-2 group extends almost to the surface of the protein. The lipid cavity follows the contour of the sn-2 acyl chain closely, but the cavity around the sn-1 chain is less confining. These features may explain the poor binding of phospholipid molecules with sn-2 acyl moieties longer than 18 carbons and the less restrictive range of groups accommodated in the sn-1 acyl region of the cavity.
PITP preferentially binds PtdIns and PtdCho. Whereas PtdCho was used in the experimental protocol, inositol can be readily modeled into the electron density for the choline moiety (except for the sugar hydroxyls). There are numerous charged or polar amino acid residues in the region that may be potential hydrogen bonding partners to the inositol hydroxyls (Fig. 5). Nevertheless, the tertiary structure of the model does not completely explain polar head group specificity of PITP, for which further studies are needed.
Conformation, Stability, and Function of the C-terminal Region-To transfer a phospholipid, PITP must first bind to the membrane, exchange its bound lipid for a membrane lipid, and then dissociate from the membrane. Previous studies have shown that there are significant conformational changes in the C-terminal residues when PITP binds to lipid membranes (21). Although cytosolic PITP is highly resistant to trypsin digestion, protein bound to membranes is rapidly cleaved at Arg 253 and Arg 259 . In the crystal structure these two sites are part of helix G in the C-terminal region. Studies have shown that generally sites that are subject to limited proteolytic cleavage by trypsin assume a loop-like structure with a conformation similar to that of trypsin inhibitors (36). Thus, this helix must "melt" prior to cleavage by trypsin, a process facilitated by interaction with a membrane surface. The relative instability of portions of this helix is supported by the high B factors calculated for residues 243-250.
Residues 260 -271 also appear to be important for maintaining a tight compact structure. PITPs truncated at residues 253 or 259 have more relaxed conformations than full-length PITP, defined by increased quenching of Trp fluorescence by acrylamide, increased Tyr exposure, increased reactivity of Cys, and increased anilino-naphthalene sulfonate binding (37). Inspection of the native structure shows that such changes cannot be caused by removal and exposure of residues without a confor-mational change in the remainder of the protein. Several studies have shown that removal of one or a few amino acids from the N-or C-terminus of a protein may have a significant effect on protein stability and/or dynamics without producing major changes in conformation (38 -39). Our data on the conformation and biological activity of truncated PITPs are consistent with a critical role of the C-terminus in the complete packing of the hydrophobic core and folding and stabilization of the native structure.
The ability of PITP to bind phospholipid substrates is necessary but not sufficient for subsequent phospholipid transfer. Removal of up to 20 C-terminal residues greatly reduces phospholipid transfer, yet these truncated proteins produced in E. coli contain PtdGro, or less commonly, phosphatidylethanolamine (PtdEtn) with a near normal 1:1 stoichiometry (37,40). Hence, residues 254 -271 are dispensable for lipid binding, although they are important in conferring selectivity for PtdIns and for efficient transfer of phospholipids. It is not immediately obvious from the PITP structure why phospholipids such as PtdEtn and phosphatidic acid (PtdOH) are bound and/or transferred so poorly.
We had earlier reported that residues 260 -271 could be eliminated with variable effects on transfer activity that reflected the composition of the membranes employed in the assay system (41). With membranes containing only PtdCho, the truncated PITP (1-259) has transfer activity identical to the full-length protein. However, with as little as 2% PtdOH in the donor and/or acceptor membranes, transfer activity decreases to 0 -20% relative to that observed with PtdCho membranes. Such concentrations have minimal effect on transfer by native PITP. We demonstrated that this and the more extensively truncated PITP (1-253) have greatly enhanced affinities for membrane binding, which would result in inefficient lipid transfer activity. We suggest that the C-terminal region of PITP acts to maintain a low affinity of the full-length protein for membranes, especially those containing acidic phospholipids. Structural perturbations within the C-terminal region lead to enhanced electrostatic interactions and increased membrane affinity (41).
Structure-Function Correlation of Point Mutations-Knowledge of the tertiary structure of PITP provides an opportunity to evaluate studies describing and characterizing the random mutagenesis of the cDNA encoding rat PITP␣ (42). Mutations of amino acids Ser 25 , Thr 59 , Pro 78 , and Glu 248 were, in general, observed to abolish PtdIns transfer, whereas PtdCho transfer was not affected. With the exception of Ser 25 , all are absolutely conserved among PITPs and rdgBs. Our data support a role of the Thr 59 hydroxyl group in a putative hydrogen bonding interaction with the inositol moiety (Fig. 5) and place Ser 25  choline and inositol polar head group moieties. Pro 78 is near the end of the portion of the cavity that accommodates the sn-2 acyl chain. Why mutations of these residues impact selectively and variably on PtdIns binding and transfer remains unresolved.
Mutations to Glu 6 and Pro 12 , two absolutely conserved residues, yield poor expression of recombinant mammalian PITP in a yeast host (42). Based on our structure, these residues likely stabilize the association between the lipid-binding core and the regulatory loop. Glu 6 participates in salt bridges with the conserved Arg 8 and with His 127 ; similarly, Pro 12 packs tightly against the conserved Leu 174 .
Membrane Binding and Mechanism of Phospholipid Transfer-Analysis of the exterior features of PITP, such as charge distribution and hydrophobicity, provided little insight into what region might constitute the membrane-binding surface. We presently favor a model in which the C-terminal region is a part of this surface. In the interaction between protein and membrane, most if not all of the C-terminal region must be displaced from helix A in the lipid-binding core and acquire increased backbone flexibility sufficient to melt a portion of helix G and become sensitive to proteolytic digestion. We further envision that the refolding and reassociation of the Cterminal region with the bulk protein provides a driving force for eventual dissociation of PITP from the membrane surface.
The detailed mechanism by which PITP releases its bound lipid, exchanges it for another present in the membrane surface and dissociates from the membrane remains to be elucidated. A plausible model is one in which the C-terminal region of the protein binds to the membrane and the lipid-binding core opens, somewhat like a clam shell, exposing its bound lipid to the membrane surface. The protein-bound lipid is absorbed into the membrane and replaced by a membrane lipid. Subsequently, the C-terminal region refolds and reassociates with the body of the protein, i.e. the clam shell closes, and the protein dissociates from the membrane surface. The affinity for the membrane surface and the affinity for a specific phospholipid must not be too great, otherwise dissociation and transfer would be too slow and inefficient. As suggested above, the decreased rate of membrane dissociation, at least in part, makes C-terminal-truncated PITP derivatives weak catalysts of lipid exchange with membranes containing acidic phospholipids.
To account for other biological activities of PITP, we suggest that the protein-bound phospholipid becomes accessible to phosphorylation following interaction of a lipid kinase with a complementary site on PITP, perhaps on the regulatory loop. One-third of the residues within this 72-amino acid segment are identical or Ͼ90% conserved among all PITP and rdgB proteins (Fig. 1). Protein-protein interactions are proposed to alter the conformation of PITP at the head group region of the lipid-binding core (Fig. 6). This could lead to either the presentation of PtdIns to the active site of a membrane-bound PtdIns 4-kinase or the exposure of the inositol moiety of a PITP-bound PtdIns to the action of a soluble PtdIns 3-kinase or 4-kinase (43)(44). Evidence supporting such a function of PITP in substrate presentation to a PtdIns 3-kinase has been reported (10). Following modification of the inositol 3-or 4-hydroxyl function, the affinity of the product for PITP would likely decrease so significantly that it would remain with the kinase or return to the lipid surface. Consistent with this mechanism is the inability of PITP to bind and transfer PtdIns (4)P (45).
Comparison with other Lipid-binding Proteins-Using the DALI algorithm (30) to search for proteins structurally similar to PITP, the recently described START (StAR-related lipidtransfer) domain (46) was identified as sharing significant tertiary structure. MLN64 (PDB:1EM2), a protein associated with some human breast carcinomas, has served as a prototype for structure determination and mechanistic studies of the START domain (47). The MLN64 START domain contains a hydrophobic tunnel that extends the length of the protein with small openings on both ends. A nine-stranded antiparallel ␤-sheet and two ␣-helices form an interior hydrophobic chamber that is believed to be just large enough to accommodate a single molecule of cholesterol. Cholesterol was not included in the crystallization solutions, but docked into the finished protein model. START domain proteins are proposed to bind lipids and interact with membranes, properties also common to PITPs. Steroidogenic acute regulatory (StAR) protein, whose function is to deliver cholesterol to the inner mitochondrial membrane and cytochrome P450scc, contains a START domain near its C-terminus. Nearly the entire sequence of PtdCho transfer protein (PCTP), another cytosolic lipid transfer protein, consti- tutes a single START domain. In contrast to PITP, PCTP has a marked specificity for PtdCho and highly restricted tissue expression in mammals (2). Although not identified by the DALI search, the recently reported structure of sterol carrier protein-2 (PDB 1QND, Ref. 48) is another example of a protein with a lipid-binding core similar to that observed in MLN64 and PITP.
Despite reasonable similarity at the secondary and tertiary levels of structure (Fig. 7), there is no primary structure homology between members of the PITP family and proteins with a START domain. It should be noted that the lipid cavity in PITP is larger than the cavity in MLN64. Upon superposition, one of the ␣-helices comprising the putative lipid-binding cavity of MLN64 clashes with the sn-1 acyl chain of the bound PtdCho in the PITP structure. This is not surprising in that PITP binds diacyl phospholipids, whereas MLN64 binds the more compact cholesterol molecule. The START and PITP protein families represent an excellent example of functional evolutionary convergence to generate comparable domain architecture and related biological function from apparently unrelated genetic information.
Fungi and plants contain Sec14p, a cytosolic protein that transports PtdIns and PtdCho (13, 49 -52) and is required for protein export and cell viability. Recently, the three-dimensional structure of Saccharomyces cerevisiae Sec14p, one member of this conserved family, was reported (53). PITP and Sec14p are similar in size, exhibit similar in vitro lipid-binding specificity and transport abilities and display significant and highly complementary interspecies activity. However, the two proteins have little structural similarity. In contrast to the PITP␣⅐PtdCho structure, Sec14p structure has no bound phospholipid molecule.
Conclusions-Elucidation of the PITP␣⅐PtdCho structure reveals a complex amphipathic lipid completely enclosed within a folded polypeptide cavity. The three functional regions of the protein provides a molecular framework for understanding the details of protein-lipid interactions on several levels. Within its lipid-binding core, PITP has been shown to sequester phospholipids from the aqueous environment. Lipid ligand specificity, whereas not completely explained by the tertiary structure, is dictated by extensive noncovalent interactions. The C-terminal region of PITP is proposed to promote the transient association to and dissociation from lipid surfaces. Through its regulatory loop, PITP is thought to present lipid substrates to kinases for subsequent phosphorylation. Further mutagenic and structural studies of PITP and other members of the PITP-rdgB family are required to test these and other hypotheses. Detailed analyses of PITP interactions with membranes and kinases are needed to gain a more meaningful understanding of the role of PITP in cellular phospholipid metabolism.