Dynamics and energetics of the mammalian phosphatidylinositol transfer protein phospholipid exchange cycle

Phosphatidylinositol-transfer proteins (PITPs) regulate phosphoinositide signaling in eukaryotic cells. The defining feature of PITPs is their ability to exchange phosphatidylinositol (PtdIns) molecules between membranes, and this property is central to PITP-mediated regulation of lipid signaling. However, the details of the PITP-mediated lipid exchange cycle remain entirely obscure. Here, all-atom molecular dynamics simulations of the mammalian StART-like PtdIns/phosphatidylcholine (PtdCho) transfer protein PITPα, both on membrane bilayers and in solvated systems, informed downstream biochemical analyses that tested key aspects of the hypotheses generated by the molecular dynamics simulations. These studies provided five key insights into the PITPα lipid exchange cycle: (i) interaction of PITPα with the membrane is spontaneous and mediated by four specific protein substructures; (ii) the ability of PITPα to initiate closure around the PtdCho ligand is accompanied by loss of flexibility of two helix/loop regions, as well as of the C-terminal helix; (iii) the energy barrier of phospholipid extraction from the membrane is lowered by a network of hydrogen bonds between the lipid molecule and PITPα; (iv) the trajectory of PtdIns or PtdCho into and through the lipid-binding pocket is chaperoned by sets of PITPα residues conserved throughout the StART-like PITP family; and (v) conformational transitions in the C-terminal helix have specific functional involvements in PtdIns transfer activity. Taken together, these findings provide the first mechanistic description of key aspects of the PITPα PtdIns/PtdCho exchange cycle and offer a rationale for the high conservation of particular sets of residues across evolutionarily distant members of the metazoan StART-like PITP family.

Phosphatidylinositol-transfer proteins (PITPs) regulate phosphoinositide signaling in eukaryotic cells. The defining feature of PITPs is their ability to exchange phosphatidylinositol (PtdIns) molecules between membranes, and this property is central to PITP-mediated regulation of lipid signaling. However, the details of the PITP-mediated lipid exchange cycle remain entirely obscure. Here, all-atom molecular dynamics simulations of the mammalian StART-like PtdIns/phosphatidylcholine (PtdCho) transfer protein PITP␣, both on membrane bilayers and in solvated systems, informed downstream biochemical analyses that tested key aspects of the hypotheses generated by the molecular dynamics simulations. These studies provided five key insights into the PITP␣ lipid exchange cycle: (i) interaction of PITP␣ with the membrane is spontaneous and mediated by four specific protein substructures; (ii) the ability of PITP␣ to initiate closure around the PtdCho ligand is accompanied by loss of flexibility of two helix/loop regions, as well as of the C-terminal helix; (iii) the energy barrier of phospholipid extraction from the membrane is lowered by a network of hydrogen bonds between the lipid molecule and PITP␣; (iv) the trajectory of PtdIns or PtdCho into and through the lipidbinding pocket is chaperoned by sets of PITP␣ residues conserved throughout the StART-like PITP family; and (v) conformational transitions in the C-terminal helix have specific functional involvements in PtdIns transfer activity. Taken together, these findings provide the first mechanistic description of key aspects of the PITP␣ PtdIns/PtdCho exchange cycle and offer a rationale for the high conservation of particular sets of residues across evolutionarily distant members of the metazoan StART-like PITP family.
Phosphatidylinositol (PtdIns) 4 is a metabolic precursor of phosphoinositides, and these lipids function as critical intracellular chemical signals in eukaryotes. Although much effort is invested in understanding the enzymes that produce and consume phosphoinositides, fundamental aspects for how phosphoinositide production is regulated and physically organized define unappreciated gaps in our understanding of how inositol lipid signaling is prosecuted in cells. It is in those contexts that PtdIns transfer proteins (PITPs) command increasing interest as activities that functionally channel lipid metabolic information to dedicated and tightly integrated phosphoinositide signaling circuits. It is therefore no surprise that the biological importance of PITPs is on striking display in both uni-and multicellular organisms. Phenotypes associated with defects in individual PITPs include: (i) non-redundant functions of individual PITPs in channeling phosphoinositide signaling to Golgi/endosomal membrane trafficking, lipid droplet metabolism, and endosomal phosphatidylserine metabolism in yeast (1)(2)(3)(4), (ii) collapse of "extreme" polarized plant membrane growth programs required for tissue organogenesis (5-7), (iii) defects in embryogenesis and retinal degeneration disease in both Drosophila and zebrafish (8 -11), and (iv) neurodegenera-tion, chylomicron retention disease, liver steatosis, and hypoglycemia in neonatal mice (12)(13)(14).
The biological importance of these proteins notwithstanding, PITPs remain poorly characterized with respect to how these proteins work as molecules. PITPs are not enzymes but proteins that mobilize energy-independent transfer of PtdIns and PtdCho between membranes in vitro (15,16). On the basis of this operational assay, PITPs have historically been described to function as lipid carriers that transport PtdIns from its site of synthesis in the endoplasmic reticulum to the plasma membrane where agonist-stimulated phospholipase C-dependent pathways consume phosphoinositides, for which PtdIns is a metabolic precursor (16). An alternative mechanistic framework for thinking about how PITPs function as molecules is gaining strong momentum, however, and it conceptualizes a very different mechanism that does not involve intermembrane PtdIns transfer at all. Rather, PITPs are proposed to chaperone tightly channeled synthesis of privileged phosphoinositide pools dedicated to specific biological outcomes (17)(18)(19).
Regardless of whether PITPs function as PtdIns presentation modules or as bona fide PL-transfer proteins, there is no question that the PL-exchange reaction is central to their biological activity. The broad outline of the exchange reaction can be conceptualized by the PITP ejecting bound PL upon docking onto a membrane surface, followed by reloading of the PITP with another PL molecule prior to its disengagement from the surface. Studies in purified systems with defined liposomes and purified recombinant PITP demonstrate that abstraction of PL from a stable membrane bilayer during the exchange reaction occurs in the absence of any other co-factors or ATP (15,16). Thus, PITPs somehow lower the activation energy of PL desorption from the bilayer, presumably by coupling protein conformational transitions to a cycle of PL unloading and loading. However, the fundamental details of the protein and PL dynamics associated with the exchange reaction are proving exceptionally challenging to dissect using conventional wet biochemistry approaches alone.
The first atomistic insights into the conformational dynamics associated with PL exchange were obtained from unrestrained molecular dynamics (MD) simulations and biochemical analyses of the major yeast PITP Sec14 (20). Although the MD time course was short (32 ns), oscillations between "open" and "closed" Sec14p conformers were simulated, and Sec14 structural elements were identified that control the conformational dynamics pathway associated with such oscillations. These included a flexible helical substructure that gates access to the PL binding pocket, a hinge region that supports helical gate motions, and a conformational switch element (the G-module) that couples gate dynamics with membrane/PL binding (20,21). The Sec14 conformational circuitry described by those simulations is likely to be broadly conserved across the Sec14 protein superfamily, and available evidence makes the case that defects in this circuitry underlie inherited diseaseassociated mutations in mammalian Sec14-like proteins (20).
Although informative, those Sec14 simulations are limited. First, the MD simulations were conducted in the absence of a membrane bilayer. Thus, no information was forthcoming regarding PL-PITP interactions during the exchange reaction (20). Fundamentally, it is those interactions that must be understood. Second, the detailed information gleaned from the Sec14 MD simulations does not translate to the vertebrate StARTdomain PITPs, given that the two classes of PITPs are structurally unrelated. This issue takes on added force given the dramatic developmental and neurodegenerative phenotypes associated with genetic deficiencies in StART-domain PITP function in vertebrates and other metazoans (8 -13). Addressing these major gaps in our knowledge for the first time, we report herein the protein and PL dynamics, and the free-energy landscape, of PtdIns and PtdCho exchange by the mammalian StART-domain PITP␣.

PITP␣ binding to the membrane bilayer
Both open apo-PITP␣ and closed PtdCho-bound holo-PITP␣ forms have been crystallized, and high-resolution structures have been solved (22,23). The protein consists of four domains. The PL-binding cavity includes: (i) an eight-stranded ␤-sheet capped by two long ␣-helices; (ii) a regulatory loop speculated to be involved in mediating protein interactions; (iii) the C-terminal domain suggested to participate in closure of the PL-binding pocket; and (iv) a small lipid exchange loop speculated to form a lid over the PL-binding pocket and thereby gate access to it (Fig. 1A).
To examine the PITP␣ lipid exchange cycle at an atomistic level, extensive all-atom MD of both the open and closed PITP␣ conformers were run in systems reconstituted with bilayers consisting of PtdCho or PtdIns/PtdCho (supplemental Table  S1). Root mean square deviation (RMSD) and root mean square fluctuation (RMSF) plots of PITP␣ C␣ atoms showed that both conformers were stable during the simulations (Fig. 1, B and C). Higher flexibility was observed for the open conformer in the presence of the PtdCho/PtdIns membrane (Fig. 1B, lower right panel), and this was primarily accounted for by the increased flexibility of the PITP␣ extreme C terminus (Fig. 1C, lower right panel); that is, the substructure suggested to help gate access to the PITP␣ hydrophobic pocket.
The substructure bounded by residues 25-38 similarly exhibited enhanced flexibilities in the open conformer relative to the closed holo-PITP␣. Indeed, it is in that former substructure that the most striking conformational change was observed during the MD. In all simulations involving the open PITP␣ conformer, the lipid exchange loop (bounded by residues 65-83) also transitioned from a turn in the starting structure into a 3 10 -helical conformation during the MD simulation as evaluated by the DSSP (define secondary structure of proteins) algorithm that uses H-bond networks to classify secondary structure (24). In one simulation (system 4, open PITP␣ conformer), that conformational change coincided with insertion of the lipid-exchange loop into the bilayer upon PITP␣ association with the membrane.
To further characterize the stable and firm binding of PITP␣ to membrane, we used both theoretical and experimental data to define two criteria for identifying the most likely "on pathway" trajectories. First, trajectories that reported binding of the Trp 203 -Trp 204 motif to the membrane surface were considered Anatomy of the PITP lipid exchange cycle to be on pathway given experimental evidence suggesting those residues to be important for membrane binding and lipid exchange by StART-like PITPs (25)(26)(27). Second, trajectories that were consistent with theoretical predictions from the Orientation of Proteins in Membranes (OPM) database (28), using PDB files for open and closed PITP␣ structures (PDB codes 1KCM and 1UW5, respectively), were also considered to be on pathway. Based on these criteria, we observed a firm and spontaneous PITP␣ binding to membrane in the following simulations: PtdCho/PtdIns membrane with closed PITP␣ repeats 1 and 2, PtdCho membrane with closed PITP␣ repeats 1 and 2, PtdCho/PtdIns membrane with open PITP␣ repeats 1 and 3, and PtdCho membrane with open PITP␣ repeats 1 and 3. Our discussion on the results and presentation of averaged data is based on these systems unless otherwise stated. Rationale for this choice can be viewed in supplemental Movies S1-S12, in which the residues Trp 203 and Trp 204 are depicted and were observed to undergo association with membrane. Moreover, the pattern of binding scored by the number of H-bonds established between protein and PL (supplemental Fig. S1A) was consistent within these systems. Initial (0 ns) and final (1 s) snapshots for all four representative simulations are presented in Fig. 2, and the time evolution of each membrane association event is documented in supplemental Movies S1-S12.
Detailed analyses of the occurrence of H-bond interactions between individual PITP␣ residues and PLs are compiled in supplemental Tables S2 and S3. Although the residues identified for membrane binding differed between the open and closed conformers in the PtdCho-only systems, some residues were consistently highlighted and identified three common regions involved in membrane association. Those regions were defined by: (i) the lipid exchange loop (residues 68 -81), (ii) residues 135-163, and (iii) residues 204 -220. With regard to specific interactions, Tyr 103 of the open conformer engaged the PtdIns/PtdCho membrane in hydrogen-bond interactions, whereas the C-terminal lid domain of the closed conformer (residues 259 -264) interacted specifically with the PtdCho-only membrane. Parallel analyses in the PtdIns/Ptd-Cho systems identified residues 147-153 and 208 -219 in PITP␣-PtdIns interactions (supplemental Table S3).

Anatomy of the PITP lipid exchange cycle
Occurrence of contacts between PITP␣ and PL atoms were similarly monitored throughout the MD course where contact was set as an interatomic distance of Յ0.35 nm (supplemental Fig. S1B). Those analyses indicated that, in addition to the regions identified by H-bond analyses, both conformers showed that residues 28 -41 constituted another membrane association region we term MAR1. Moreover, the open conformer uniquely presented one additional membrane association region we term MAR2 (residues 99 -110), and this particular set involved interactions other than H-bonds (e.g. charge pairs). This unique membrane association area was able to interact with the bilayer surface because of the slightly different dock orientation of the open PITP␣ conformer on the membrane surface relative to that of the closed form, a difference that renders this protein region more accessible to the membrane for binding. For purposes of clarity, these MAR1 and MAR2 motifs are identified on the PITP␣ structure in Fig. 1A.

PITP␣ dynamics upon membrane association
After membrane binding, the closed PITP␣ molecule remained relatively inert throughout the remainder of the simulation. No unloading of bound PtdCho or opening of the lid was observed (supplemental Movies S1-S6). Specific conformational motions were observed upon membrane binding for the open PITP␣ conformer, however, and it is for this reason that focus was trained on this conformer. Interestingly, residues 25-38, implicated in membrane binding, exhibited enhanced mobility in the open form. Moreover, using the interatomic distance between residues Ser 30 and Glu 41 as a ruler for pocket closure, that distance decreased dramatically until it approached that observed in the closed conformer ( Fig. 3 and supplemental Fig. S2).
The lipid exchange loop was also inserted into the cytoplasmic leaflet of the bilayer at this time where it underwent a transition from a turn to a 3 10 -helix, as indicated by DSSP analysis, and adopted a more extended conformation as reported by RMSF calculations (Fig. 1C) and by visual inspection (Fig. 3, A  and B). Most interestingly, bilayer insertion of the lipid exchange loop (and the partial closure of the hydrophobic pocket) coincided with partial loading of a single PtdCho molecule into the open pocket (Fig. 3, C and D). We interpret the simulation data to conceptualize engagement of the lipid exchange loop with one PtdCho molecule with the result that the engaged PtdCho was extracted from the membrane bilayer and moved toward the hydrophobic pocket/water interface. PITP␣ was unable to consolidate those movements into vectorial transfer of PL into the hydrophobic pocket, however. Rather, the partially extracted PLs either slid back into the bilayer or engaged in bobbing movements at the cavity opening/water interface.
The role of PITP␣ residues Trp 203 and Trp 204 in the lipid exchange cycle is unclear. Several studies conclude that those residues are essential for lipid exchange by PITP␣ and closely related PITPs (25,26), whereas other studies argue for minor roles for Trp 203 and Trp 204 in lipid exchange (27). Nonetheless, Trp 203 and Trp 204 were the first residues to interact with the lipid bilayer, and the Trp 203 -Trp 204 motif maintained interactions with membrane lipids throughout the MD time course. MD suggests that those Trp residues, and the residues that surround them, interact primarily with PtdIns (supplemental Fig.  S1B).  Table S1. The lipids are shown in licorice (PtdIns in yellow and PtdCho inside the binding pocket of PITP in red) and transparent brown surface. PITP␣ is rendered in new cartoon representation in colors corresponding to secondary structure (yellow, ␤-sheet; purple, ␣-helix; blue, 3 10helix; cyan, random coil). Water molecules and ions are omitted for purpose of clarity.  Table S1). The loop consisting of residues 25-38 is depicted in red, residues Ser 30 and Glu 41 are in orange licorice representation, and the indicated distance between their C␣ atoms is marked as a dashed black double-headed arrow. The lipid exchange loop is colored in green, and its conformational change between a turn (red highlight) and a 3 10 helix (blue highlight) is indicated. The PtdCho lipid engaged by the lipid exchange loop is in yellow licorice representation. C and D, snapshots from simulation system 4 (as described in supplemental  Table S1) of the open PITP␣ conformer in new cartoon representation (C) or blue surface representation (D), where a PtdCho molecule (in licorice representation) is engaged by the lipid exchange loop in the final (1 s) step. Other lipids are shown as brown sticks, and water molecules and ions are omitted for clarity.

Anatomy of the PITP lipid exchange cycle Functional analyses of Ser 25 in PITP␣ activity
Residues 25-38 define a flexible region predicted to be involved in membrane binding, and the boundary residue Ser 25 is projected to be involved in H-bonding to PtdCho and PtdIns within the PITP␣ hydrophobic pocket. To assess the importance of Ser 25 to PITP␣ function, this residue was altered to Phe, Ala, Cys, Glu, Lys, and Thr. All mutant proteins were confirmed to be stable when expressed in yeast (supplemental Fig.  S3). Mutant proteins were subsequently analyzed in in vivo and in vitro functional assays.
High-level expression of PITP␣ in yeast is sufficient to effect phenotypic rescue of the growth and membrane trafficking defects associated with temperature-sensitive mutations in the yeast SEC14 gene (29), and this system enabled a general assessment of whether a mutant PITP␣ is functional or not. Expression of PITP␣ S25A , PITP␣ S25C , and PITP␣ S25T rescued sec14-1 ts -associated growth defects at the restrictive temperature of 37°C at an efficiency similar to that observed for PITP␣ (Fig.  4A). By contrast, PITP␣ S25E was only partially functional in this assay, whereas PITP␣ S25F and PITP␣ S25K scored as completely defective. Consistent with the in vivo results, purified recombinant PITP␣ S25F and PITP␣ S25K were also completely inactive for PtdIns and PtdCho transfer in vitro in end-point assays using radiolabeled phospholipid substrates and a single clamped protein concentration (Fig. 4B). Thus, although some substitutions were tolerated for Ser 25 , the functionality pattern did not correlate with H-bonding capacity (e.g. S25A scored as active). Rather, activity correlated best with the accessible surface area of the residue at that position. Residues compatible with efficient function (Ser, Ala, Cys, and Thr) have small accessible surface areas (80, 67, 104, and 102 Å 2 , respectively); those that supported partial function have larger values (Glu, 138 Å 2 ); and residues not tolerated had the largest side chains (Phe and Lys; accessible surface areas of 175 and 167 Å 2 , respectively).
Real-time dequenching approaches were used as independent assays to separately monitor PtdCho-and PtdIns-transfer rates. In those assays, transfer was recorded as enhanced fluorescence of pyrene-labeled phospholipid as it was mobilized from quencher-loaded donor vesicles to quencher-free acceptor vesicles (see "Materials and Methods"). The significantly reduced initial rates of PtdCho and PtdIns transfer by PITP␣ S25E and PITP␣ S25K were recapitulated in those experimental systems (Fig. 4C). These assays were able to resolve a small amount of PtdIns-transfer capacity for PITP␣ S25E , which  25 . A, phenotypic rescue assays. Isogenic SEC14 (CTY182; MATa ura3 -52, lys2-801, his3⌬200) and sec14-1 ts strains (CTY1-1A; MATa ura3-52, lys2-801, his3⌬200, sec14-1 ts ) served as positive and negative controls for PITP␣ activity, respectively. The latter strain fails to grow at 37°C because activity of the major PITP is inactivated at that temperature (29). Isogenic sec14-1 ts yeast strains carrying expression cassettes for RnPITPNA, or versions carrying the indicated substitutions for residue Ser 25 , integrated at the LEU2 locus, were spotted in 10-fold dilution series onto YPD agar plates and incubated at either permissive temperature of 30°C or restrictive temperature of 37°C for 48 h. B, end-point PtdIns and PtdCho transfer assays. Purified recombinant PITP␣ and indicated missense mutants (10 g protein/assay) were assayed for [ 3 H]PtdIns and [ 3 H]PtdCho transfer, as indicated. Transfer efficiencies of PITP␣ were set at 100% (n ϭ 3), and transfer of mutant PITP␣ proteins was normalized relative to PITP␣ activity values obtained from the same experiments. Assay efficiencies and backgrounds for these experiments are summarized in supplemental Table S5. C, rates of PyrPtdIns and PyrPtdCho transfer were measured for the indicated Ser 25 substitution mutants. Fluorescence intensities of PyrPtdIns or PyrPtdCho are plotted as a function of time. Black arrows mark addition of a 1-g increment of the indicated protein. Following protein addition, the observed increase in fluorescence intensity was directly proportional to the relative transfer efficiency of the protein being assayed. The initial slope after each addition was calculated to obtain relative transfer efficiencies (supplemental Table S6). Error bars were calculated from two independent determinations.

Anatomy of the PITP lipid exchange cycle
accounts for the modest ability of the mutant to rescue Sec14 deficiency in vivo (Fig. 4, A and C).

Free energies of PL loading/unloading by PITP␣
The 1000-ns atomistic MD runs were of insufficient length to describe a complete PL loading/unloading half-cycle. Thus, suitable free-energy calculation methods were employed to study the energetics of PL loading/unloading by PITP␣. A steered MD simulation approach was used to obtain a pathway for positioning of PtdCho and PtdIns within the PITP␣-binding pocket through a short pulling simulation (see "Materials and Methods"). Once this regime yielded a PL-loaded conformer, the pulling constraints imposed on PL position were released, and subsequent trajectories were used to extract the windows for umbrella sampling simulations. Those umbrella sampling simulations were then analyzed with the weighted histogram analysis method (see "Materials and Methods"). Our selection of the force constant, together with the window spacing, provided sufficient sampling of the configuration space along the reaction coordinate as indicated by the satisfactory overlap of the histograms in supplemental Fig. S4A. Moreover, although each window was simulated for 40 ns, the obtained PMF stabilized within 10 ns of simulation, indicating that the free-energy profile converged to a stable value. We interpret this rapid convergence to reflect the fact that the starting structures for the umbrella sampling windows were extracted from unbiased trajectories. This interpretation follows previous demonstrations that the qualities of the starting structures for umbrella sampling have a significant impact on subsequent free-energy profile quality (30).
Because the exchange cycle occurs efficiently with purified PITP␣ and pure liposomes and without any ATP requirement, the activation energy for PL loading/unloading must be sufficiently low to be overcome by thermal fluctuations. Indeed, during the ensuing MD simulations (150 and 240 ns for the systems with PtdCho and PtdIns, respectively), bound PL was unloaded from the binding pocket and released into the bilayer (Fig. 5A). The PL unloading took place without any externally applied pulling forces. Taking advantage of these well equilibrated starting configurations, umbrella sampling simulations were extracted from the unloading trajectories, and the free energies related to PITP␣ interaction with PtdCho and PtdIns were calculated. The free-energy profiles for both PLs as a function of the distance along the bilayer normal are presented in Fig. 5B. Those profiles confirmed that PL loading into PITP␣ occurs within the realm of thermal fluctuation. The free energy of PtdCho uptake by PITP␣ was ϳ20 kJ mol Ϫ1 (8 k B T), as calculated from the bottom of the global minimum to the maximum peak at 3 nm (Fig. 5B). The values for PtdCho and PtdIns desorption from the membrane bilayer were 90 and 120 kJ mol Ϫ1 , respectively (supplemental Table S4). The dramatic differences in the free energies of PL desorption in the presence versus absence of PITP␣ emphasize the remarkable role the protein plays in lowering the activation energy for that critical step.
For both PtdCho and PtdIns, the free-energy profiles identified two distinct minima. The first, at 1.5 nm, corresponded to the state where the PL was at equilibrium in the bilayer. The second was located in the membrane-distal region of the PLbinding cavity where the PL headgroups are coordinated by appropriate PITP␣ residues and the fatty acyl chains are incorporated into the most hydrophobic regions of the cavity. Unexpectedly, the free energies of loading for both PLs were similar, suggesting that the affinity of PITP␣ for PtdIns in the initial loading stage(s) was not appreciably higher than that for PtdCho. This was an unexpected outcome given crude estimates that PITP␣ has a 16-fold higher affinity for PtdIns than for PtdCho (15,16). Another remarkable feature of the calculated free-energy profiles was that the minima at ϳ3.5 nm were very shallow. This property neatly accounts for why PL molecules that had partially loaded or were otherwise steered into the PITP␣ pocket were spontaneously and efficiently released into the bilayer via a reverse unloading reaction during simulation; that is, the energy barrier for unloading of bound PL is low.

PITP␣-PL interactions within the hydrophobic pocket
Lipid-protein H-bond interactions within the PITP␣ hydrophobic pocket are likely required for serial consolidation of the loading reaction so that vectoriality is imposed on this process, which culminates in a complete loading reaction (i.e. rather than the reversible and futile partial loading reaction observed in the 1000-ns MD). Indeed, both PtdCho and PtdIns engaged in H-bonding within the PITP␣-binding pocket, and the highest peaks in average H-bond interaction coincided with minima in the free-energy profiles (Fig. 5, B and C). Those results argued that H-bond interactions make critical contributions to the free-energy landscapes of the PtdCho and PtdIns loading/ unloading reactions.
All pocket residues involved in H-bonding with PtdCho and PtdIns are highlighted in Fig. 5C. The unloading MD runs forecast that Lys 68 interacted primarily with PtdCho, whereas Thr 59 , Lys 61 , Glu 86 , Asn 90 , Thr 114 , and Glu 218 interacted primarily with PtdIns. Satisfyingly, the former four residues were known from previous studies to engage PtdIns specifically (25,31), whereas the latter two interactions represent new insights. Tyr 18 , Gln 22 , Ser 25 , Tyr 63 , Asn 101 , Lys 195 , and Gln 217 formed H-bonds with both PtdCho and PtdIns, and those residues had not previously been recognized to play any role in the PL-exchange cycle. The strong single H-bond interaction of Lys 68 with PtdCho was particularly noteworthy because it was observed early in the loading reaction sequence (supplemental Fig. S4B), suggesting that this previously unsuspected interaction plays a role in the PtdCho extraction process. PtdIns did not interact strongly with PITP␣ residue Lys 68 . Rather, it formed two H-bonds with Glu 218 , another residue previously not suspected to play any role in lipid exchange (supplemental Fig. S4B). The MD projected that interactions of PtdCho and PtdIns with Lys 68 and Glu 218 , respectively, defined the major contribution to peaks in the average H-bond plot at ϳ2.3 nm (Fig. 5C).

Functional analyses of Glu 218 and Lys 68 in PL transfer and binding
Lys 68 is a candidate residue for engaging PtdCho early in the PITP␣-loading reaction. Structure-function experiments demonstrated PITP␣ K68E or PITP␣ K68R expression was less efficient

Anatomy of the PITP lipid exchange cycle
in phenotypic rescue of sec14-1 ts -associated growth defects relative to wild-type protein (Fig. 6A). Other substitutions were reasonably well tolerated as evidenced by PITP␣ K68A , PITP␣ K68C , and PITP␣ K68Q scoring as functional proteins in that assay. The biochemical properties of purified PITP␣ K68E were further examined in the end-point PtdCho and PtdIns transfer assays. Consistent with the predictions of the molecular dynamics simulation (MDS), PITP␣ K68E showed a ϳ70% reduction in PtdCho-transfer activity (Fig. 6B). However, the mutant PITP was also strongly defective in PtdIns-transfer, even though the MDS did not reveal strong interactions of Lys 68 residue with this PL. Real-time dequenching analyses confirmed the significantly reduced initial rates of PtdCho and PtdIns transfer and binding by PITP␣ K68E (Fig. 6, C and D).
Unlike Lys 68 , residues Thr 114 and Glu 218 were scored to interact predominantly with PtdIns. Analyses of Thr 114 substitution mutants proved fruitless because the mutants were unstable proteins when expressed in yeast and, in all cases, the mutant proteins quantitatively partitioned into inclusion bodies when expressed as recombinant proteins in Escherichia coli. Thus, our analyses focused on Glu 218 , and this residue was of interest given that it was projected to form two H-bonds with PtdIns during the MD. Nonetheless, PITP␣ E218A , PITP␣ E218D , and PITP␣ E218K all scored as functional in the phenotypic rescue assay (Fig. 7A). Both end-point and real-time PtdIns and PtdCho transfer dequenching assays also showed that PITP␣ E218A was not significantly compromised for either PtdIns-or PtdCho-transfer activity (Fig. 7, B and C). However, the mutant protein was compromised for both PtdIns-and PtdCho-binding (Fig. 7D).

Functional analyses of Lys 195 in PtdIns and PtdCho transfer and binding
Lys 195 was highlighted in the MD simulations as a major contributor to the interaction of both PtdIns or PtdCho within the PITP␣ lipid-binding cavity (Fig. 8). Moreover, in indepen- Anatomy of the PITP lipid exchange cycle dent analyses, we serendipitously highlighted this same residue in a Cys-scanning mutagenesis screen on the basis of the 2.3 and 2.7 Å distance between Lys 195 and the phosphate moieties of bound PtdIns and PtdCho, respectively. 5 Thus, PITP␣ K195C was purified and analyzed with respect to PtdCho-and PtdInstransfer rates and efficiencies of PtdCho and PtdIns binding (Fig. 8, A and B, respectively). PITP␣ K195C scored as nonfunctional in the yeast PITP activity assay because its expression was unable to rescue growth of sec14 ts yeast mutants at 37°C (Fig.  8A). PITP␣ K195C did not transfer PtdIns or PtdCho in the radiolabel transfer assay (Fig. 8B) and, as expected, was unable to transfer (Fig. 8C) or bind (Fig. 8D) pyrene-labeled PtdIns or PtdCho.

Conformational transitions of the PITP␣ C-terminal helix contribute to ligand specificity
PtdIns is the preferred exchange substrate for PITP␣ but the 1000-ns all-atom simulations of membrane-bound PITP␣ were not informative regarding what headgroup-specific conformational transitions of the PITP might contribute to such ligand specificity. To understand how large conformational changes contribute to the ability of PITP to bind lipids, PITP␣ dynamics were simulated in a solvated system. In those simulations, the conformational transitions of apo-PITP␣, the open PtdInsbound conformer, and the open PtdCho-bound were compared. The latter two starting structures were generated by assigning PtdIns or PtdCho to their respective positions within the hydrophobic pocket as defined by X-ray crystallography (see "Methods and Materials"). MDS (250 ns) revealed conformational transitions from the open to the closed conformation.
Of relevance to this study, we focus on the C-terminal Ghelix. Even though G-helix lies on the PITP␣ surface, and its side chains have no direct interaction with either PL (distance, Ͼ8 Å), this substructure exhibited differential dynamics in the 250-ns MD simulations depending on whether PtdIns or PtdCho was loaded into the open PITP␣ conformer. In both the open apo-PITP␣ and PITP␣-PtdCho MDS, the G-helix remained stable, and no major uncoiling was observed (helix length average, 22 Å from Thr 240 to Asp 256 ; Fig. 9A). This helix

Anatomy of the PITP lipid exchange cycle
underwent a conspicuous melting in systems modeling the open PITP␣-PtdIns conformer, however (helix length average, 15 Å from Thr 240 to Glu 250 ). Of note was a kinking of the helix at residue Glu 248 , an invariant residue among the StART-like PITPs (Fig. 9B).
In the open PITP␣-PtdCho conformer, the Glu 248 side chain was disposed to solvent on the G-helix surface and pointed away from the PtdCho headgroup. In that orientation, the side chain was in H-bond contact with G-helix residues Arg 245 and Lys 252 (both invariant StART-like PITP residues), thereby stabilizing the helical secondary structure. In the case of PITP␣-PtdIns, however, Glu 248 was initially oriented toward the PtdIns headgroup. As the open PITP␣-PtdIns conformer closed, however, the G-helix distorted, and the Glu 248 side chain twisted away from the PtdIns headgroup toward the protein surface. In that orientation, Glu 248 was primarily engaged in electrostatic interaction with Lys 252 . Residue Met 241 sits at the N-terminal end of the G-helix, and it too is invariant among the StART-like PITPs. Interestingly, Met 241 and Ile 244 (also highly conserved) engaged Leu 23 on the N-terminal A-helix via van der Waals interactions (Fig. 9C), suggesting an involvement for Met 241 and Ile 244 in stabilizing the A-helix. Those results offered a spe-cific hypothesis for why the G-helix showed headgroup-specific dynamics given A-helix residues Tyr 18 , Gln 22 , and Ser 25 were identified as interacting with PtdIns and PtdCho in the 1000-ns MDS of membrane-bound PITP␣.

Glu 248 and Met 241 execute specific functions in PtdIns transfer and binding
To determine whether the G-helix exhibits consequential headgroup-specific involvements in PITP␣-mediated PL exchange, the functional properties of Glu 248 mutants were analyzed in detail. Indeed, missense substitutions at that position levied strong selectivity for PtdIns transfer and PtdIns binding relative to PtdCho binding/transfer. In the phenotypic rescue assay, only PITP␣ E248D showed activity, whereas PITP␣ E248K , PITP␣ E248A , and PITP␣ E248T all scored as strongly defective (Fig. 10A). The defects from the rescue assay were recapitulated in PtdIns-transfer assays, even though these mutant proteins retained wild-type abilities to transfer/bind PtdCho in end-point assays (Fig. 10B). PITP␣ E248A was also analyzed in the dequenching assay system where it again scored as completely defective for PtdIns transfer but showed essentially wild-type PtdCho-transfer activity (Fig. 10C). Those

Anatomy of the PITP lipid exchange cycle
headgroup-selective defects were further confirmed in realtime binding measurements for which PITP␣ E248A showed no PtdIns-binding activity in the time course of the assay while remaining fully competent for PtdCho binding (Fig. 10D).
The biochemical consequences of altering Met 241 were similarly analyzed. This residue was of particular interest not only for structural reasons as bridge between the G and A helices, but the M241T missense substitution is the consequence of a SNP circulating in the human population (32). As observed for the PITP␣ E248A,K,T mutants, PITP␣ M241T was selectively compromised for PtdIns-transfer activity. The mutant protein exhibited a 50% decrease of PtdIns-transfer activity with no diminution in its ability to transfer PtdCho (Fig. 10B).

Discussion
The PL-exchange cycles of PITPs are essential to their critical functions in stimulating the activities of PtdIns 4-OH kinases for biologically sufficient production of phosphoinositides (18). The basic form of this cycle, however, is not at all understood for any PITP. Herein, we describe a series of unrestrained MD simulations of mammalian PITP␣ that most likely modeled interfacial processes associated with the initial steps of PITP␣ reloading with PL after ejection of a previously bound PL molecule. Moreover, the simulations supported first calculations of the free-energy landscape of PtdIns and PtdCho exchange by mammalian PITP␣. Additional simulations of PITP␣ in membrane-free systems described conformational transitions that modulate the lipid exchange cycle in surprisingly long-range and lipid headgroup-specific ways. These collective studies report new insights into the initial stages of PL exchange by PITP␣, describe H-bonding interactions between PL and protein within the PITP␣ hydrophobic pocket, and provide the first clues as to how PITPs lower the activation energy of PL desorption from the membrane bilayer.

PITP␣ binding to membrane surfaces
Both the closed and the open PITP␣ conformers spontaneously bound membrane surfaces in equilibrium MD simulations, and the binding events were stable and occurred in the course of relatively short time scales (ϳ200 ns). Membrane binding consisted of three components. The first was direct interactions of the PITP with lipid at the membrane surface as counted by the number of H-bonds established between protein and PL. This component involved two separate regions

Anatomy of the PITP lipid exchange cycle
bound by residues Lys 135 -Lys 163 and Trp 204 -Arg 220 as membrane-association domains. The second component was unique to the open PITP␣ conformer and involved a membrane insertion unit (the lipid exchange loop defined by residues Lys 68 -Ala 81 ) that penetrated into the bilayer and initialized the first steps of the PL-loading reaction of the PITP␣ lipid exchange cycle. The third component described PITP␣ conformational transitions that accompanied membrane binding/ insertion events. One of these motions involved lipid exchange loop (Lys 68 -Ala 81 ) penetration into the bilayer. The other involved increased conformational flexibility and membrane association of residues Ser 25 -Glu 41 . The importance of this region to PITP␣ activity was confirmed by mutational analyses that were consistent with Ser 25 playing a key role in the conformational flexibility of this PITP␣ subdomain and this flexibility being required for PL exchange.

Initial steps of PL loading into apo-PITP␣
The MD simulations predicted that a consequence of lipid exchange loop insertion into a mixed bilayer was a partial load-ing of the open PITP␣ conformer with a single PtdCho molecule (at 700 ns in the simulation). This partial extraction of PtdCho from bilayer into the PITP␣ hydrophobic pocket was accompanied by a shielding of the fatty acyl chains from bulk membrane by the lipid exchange loop and partial closure of the unit that regulates accessibility to the lipid-binding pocket. Functional assays indicated amino acid substitutions in this substructure (i.e. for Lys 68 ) compromised both PtdCho and PtdIns binding and exchange activities. It is now a both reasonable and attractive proposition that the inserted lipid exchange loop is the first protein substructure to interact with a PL molecule and that this interaction marks that PL molecule as a PITP␣ exchange ligand.
The progress in describing initial steps of the PL exchange cycle notwithstanding, we were unable to simulate full incorporation of PL into the PITP␣ lipid-binding pocket. Rather, the "chosen" PtdCho molecule entered a futile exchange reaction characterized by a shallow breathing conformational dynamics regime and ultimately fell back into the bilayer. Our interpretation is that the partially loaded PL encounters a free-energy well at an early step in the exchange cycle and that this barrier could not be negotiated during the 1000-ns simulation windows used in these experiments. Complete uptake of PL into the PITP␣ lipid-binding pocket requires extending simulation time scales sufficiently to allow additional conformational transitions to occur that direct the open PITP␣ from a shallow breathing regime to one that consolidates the PL-extraction process by supporting trajectories converging onto a closed conformer. An alternative, and not mutually exclusive, possibility is that the compositionally simple bilayers modeled themselves created a significant energy barrier that could not be negotiated during the 1000-ns MD. Natural membranes exhibit packing irregularities that might be productively exploited by the PITP␣ molecule in initiating a lipid exchange cycle, and such packing irregularities were minimal in the virtual bilayer systems analyzed in this study.

Free-energy landscape of PITP␣ interactions with lipid
Previous electron spin resonance studies rationalized the ATP independence of yeast Sec14-driven exchange cycle as a system where a PL molecule is engaged in a simple partitioning between two chemically similar microenvironments, i.e. the bilayer and the PITP PL-binding cavity (33,34). The PITP␣ studies reported herein extend that concept. First, PITP␣ dramatically lowered the free energy of PL desorption from the bilayer to within the realm of thermal fluctuation. This lowering of an otherwise very significant thermodynamic barrier was driven by specific H-bond interactions between PITP␣ and lipid ligand headgroups.
Second, free-energy profiles for PtdIns and PtdCho trajectories through the lipid-binding pocket indicated that energy minima were shallow, thereby accounting for why PL molecules that had partially loaded, or were otherwise steered, into the PITP␣ pocket were spontaneously and efficiently released into the bilayer via a reverse unloading reaction during simulation. This feature likely reflects the fact that no larger conformational transitions of the PITP from the open to the closed state were observed in those 1000-ns simulations. Such conformational transitions would have the dual effects of dramatically

Anatomy of the PITP lipid exchange cycle
impacting the free-energy landscapes of the exchange cycle and would likely contribute to the purported 16-fold greater affinity of PITP␣ for PtdIns versus PtdCho. Such a marked difference in PITP␣ affinity for PtdIns versus PtdCho was not at all apparent in our calculations because those analyses reported the free energies of loading for both PLs to be similar.
In support of this idea, a solvent box-based MDS system that does model larger conformational motions of the PITP␣ molecule indicated that headgroup-specific conformational transitions within the C-terminal G-helix exerted headgroup-specific control of the PL-exchange cycle. This finding is interesting for two reasons. First, the G-helix does not contact bound PtdIns or PtdCho directly and is far removed from both headgroups. Second, the M241T missense substitution in the G-helix selectively compromises PtdIns-transfer activity and represents a recognized SNP circulating in the human population.

PITP-PL interactions during the exchange cycle
Umbrella sampling data of PL "unloading" provided novel insights into which PITP␣ residues play direct roles in facilitat-ing PtdIns and PtdCho transit through the PITP␣ lipid-binding pocket. Superposition of the structural elements and residues identified in this study to play important roles in the PITP␣ PL-exchange cycle, with primary sequences of other members of the StART-like PITP family, lend further confidence in the veracity of the data. MD simulations projected six residues of the PITP␣ lipid-binding pocket interact with PtdIns (Thr 59 , Lys 61 , Glu 86 , Asn 90 , Thr 114 , and Glu 218 ) and, of those, all but Glu 218 are absolutely conserved in evolutionarily distant members of the larger StART-like PITP family (Fig. 11) Fig. 11). This correspondence is remark-

Anatomy of the PITP lipid exchange cycle
able given that MDSs are agnostic to experiment and to primary sequence alignments.
Of the set of identified specific interactions put to the functional test, the MD simulations were not absolutely accurate in their precise descriptions of headgroup-specific interactions. However, whereas the suggested lipid headgroup specificities were not borne out by experiment, the MD simulations did accurately identify these residues as important for optimal lipid exchange activity. The functional involvements of these residues in the exchange cycle were otherwise invisible to analyses of PITP␣ crystal structures alone.
If there are degrees of freedom for what residues are tolerated at these positions, why would those residues be so conserved across large evolutionary distances? We speculate that although there might be compensating interactions that are permissive for PL exchange in an in vitro system, strong evolutionary pressures are at play in cellular systems. For example, appropriate presentation of PtdIns to a PtdIns 4-OH kinase for biologically sufficient PtdIns-4-P production (i.e. in the face of cellular antagonists of PtdIns-4-P signaling) might demand stringent geometric requirements for PtdIns/PtdCho trajectories during the exchange cycle that are not apparent in in vitro lipid transfer assays.

Outstanding questions
Although the studies reported here describe the initial steps in a pathway of conformational transition from an apo-PITP␣ to a phospholipid-bound holo-PITP␣ and describe conformational transitions in the C-terminal G-helix that exhibit surprisingly long-range headgroup-specific involvements in the PLexchange cycle, there remain outstanding considerations that will impact the details in important ways. First, the most productive MD simulations involved a membrane-docked PITP␣ conformer. At this point, it remains unknown how PL unloading is spatially or temporally related to PL loading. That is, are these concerted processes, or are these separable reactions that proceed through an apo-PITP␣ intermediate? Although soluble probe accessibility experiments argue against the latter mode, 6 the MD simulations reported here do not and cannot explicitly consider concerted PL-loading/unloading exchange mechanisms. Whether the PL-loading and -unloading reactions are perfectly symmetrical processes also remains to be determined. Such symmetry is unlikely if PL-loading and unloading are concerted reactions. Finally, these simulations do not consider involvement of the PtdIns 4-OH kinase.
Those open questions notwithstanding, the satisfying congruence of the in silico data with the results of functional assays lends confidence that atomistic details of the PITP␣ PLexchange reaction reported herein describe a reasonable first conceptualization of the initial stages of this key biological reaction. Extended MD production runs that capitalize on the membrane-docked apo-and holo-PITP␣ models described in this report now define promising approaches for linking the insights from the atomistic simulations of phospholipid loading/unloading by membrane-bound PITP␣ with insights culled from simulations of solvated PL-bound protein. We expect such extended simulations will extract new sets of fundamental mechanistic details regarding how the lipid-exchange cycle operates in StART-like PITPs.

Anatomy of the PITP lipid exchange cycle
the RCSB Protein Data Bank (22,23). The former structure is lipid-free, whereas the latter describes PITP␣ in complex with a single bound PtdCho molecule. These two starting structures were each introduced into each of the two bilayer systems to generate a total of four distinct systems that were run in 1-s all-atom simulations. The simulations were repeated three times for each system with different starting positions of the protein in water solution above membrane, together resulting in 3 s of simulation time for each system and in overall 12 s of simulation time. In all cases, PITP␣ was initially positioned above the bilayer in the water phase. Simulations were run at physiological salt concentrations (150 mM KCl), and counter ions were introduced to neutralize the total charge of the system. Details of the compositions of each simulated system are provided in supplemental Table S1.
The OPLS all-atom force field was used to describe all molecules (35). For water, the TIP3P model that is compatible with the OPLS parameterization was employed (36). The system structure used in this study is identical to those employed in our previous simulations of lipid bilayers (37,38). To parameterize lipid molecules, parameters specifically derived for lipids were used (39 -41). Periodic boundary conditions with the usual minimum image convention were employed in all three dimensions. The length of each hydrogen atom covalent bond was preserved by the LINCS algorithm (42). The integration time step was set to 2 fs, and the simulations were carried out at constant pressure (1 bar) and temperature (310 K). The temperature and pressure were controlled by the Parrinello-Rahman and velocity-rescale methods, respectively (43,44). The temperatures of the solute and the TIP3P solvent were coupled separately. For pressure, a semi-isotropic scaling was employed. The Lennard-Jones interactions were cut off at 1.0 nm. For the electrostatic interactions, the particle mesh Ewald method (45) was employed with a real space cutoff at 1.0 nm, ␤-spline interpolation (sixth order), and a direct sum tolerance of 10 Ϫ6 . The simulations were performed with the GROMACS 4.5.5 simulation package (46,47).

Molecular dynamics simulations of PITP in a solvated system
Extensive (250 ns) MDS were carried out on PITP␣ structures prepared from the closed conformation bound to PtdCho (PDB code 1T27: 2.20 Å), the open conformation not bound to lipid (PDB code 1KCM: 2.00 Å), and the closed conformation complex with PtdIns (PDB code 1UW5: 2.20 Å). From these available crystal structures, several starting structures were generated in open conformation in combination with PtdCho and PtdIns. A total of 3 MD simulations were run for 250 ns each using the Desmond module of Schrodinger version 2016-3 (D. E. Shaw, Schrodinger Inc.), with the OPLS-AA forcefield as defined in Desmond. System Builder within Desmond/Schrodinger was used to generate a solvated system of protein-ligand complex. An orthorhombic box with a 12 Å buffer around protein-ligand complex was used to generate the periodic box. An all-atom atomistic scale MD simulation was carried out in a periodic box of explicit water molecules. For water, the TIP3P water model compatible with the OPLS-AA parameterization was employed. Simulation was carried out at 300 K at a physi-ological salt concentration of 150 mM NaCl. Counter ions were included to neutralize the total charge of the system.
The OPLS-AA all-atom force field was used to describe all molecules, and the system was relaxed using Maestro's system relaxation protocol before the production run. That regime included a multistage relaxation protocol with two stages of minimization with 5000 steps of restrained minimization and 5000 steps of unrestrained minimization followed by five stages of MD runs with gradually diminishing restraints on soluteheavy atoms. These stages include 500 ps of restrained and 500 ps of unrestrained equilibration under an NVT ensemble. The system was further equilibrated under NPT ensemble with 300 ps of restrained equilibration, a 350 ps of restrained equilibration at 300 K with annealing off, and another 350 ps of unrestrained equilibration at 300 K. A separate MD production run of 250 ns was carried out at constant temperature and pressure.
Long-range electrostatic interactions were estimated using the particle mesh Ewald, whereas bonds involving protons were constrained using the SHAKE algorithm (48). A 2-fs time step was used throughout the simulation. The Thermostat Nose-Hoover chain method was applied with relaxation time of 1 ps (49). Barostat parameters were set according to Martyna et al. (50) with a relaxation time of 2.0 ps with isotropic coupling. A 9 Å cutoff was applied to Lennard-Jones interactions, and the nonbonded list was updated every 1.2 ps. The production run was continued for 250 ns, and snapshots of the coordinates were written out every 1 ps. The results were analyzed, and plots were generated within Maestro's simulation event analysis and simulation interaction diagram modules.

Free-energy calculations
Free energies of PtdCho and PtdIns uptake by PITP␣ were obtained by umbrella sampling simulations of two model systems. The starting structure of the first system was obtained from the final structure after a 1-s production run of simulation 1 of the system composed of the open PITP␣ and PtdCho bilayer (supplemental Table S1, system 3). The second system was generated from the first by replacing the PtdCho molecule positioned closest to the binding site with a PtdIns molecule.
Free-energy profiles for PtdCho and PtdIns uptake by PITP␣ were calculated using the umbrella sampling method (51). The starting structures were obtained as follows. First, a single PL (PtdCho or PtdIns) was pulled from the bilayer toward the PITP␣ lipid-binding pocket along the z axis (membrane normal) in 20 ns of simulation with a constant pulling force constant of 1000 kJ ϫ mol Ϫ1 ϫ nm Ϫ2 and pulling rate of 0.00075 nm ϫ ns Ϫ1 exerted on the lipid headgroup. Next, the position of the lipid headgroup was adjusted by lateral pulling such that it was positioned in the near vicinity of PITP␣ residues Cys 95 and Phe 225 reported to be important for PtdCho binding (22,52) or residues Thr 59 , Lys 61 , Glu 86 , and Asn 90 known to coordinate the inositol headgroup (25,31). After the PLs were optimally positioned within the PITP␣ binding pocket, the constraints imposed on the respective headgroups were released. The PtdCho and PtdIns systems were then simulated for an additional 150 and 240 ns, respectively.
In umbrella sampling simulations, 28 and 26 windows with a spacing of 0.1 nm were employed for PtdIns and PtdCho sys-Anatomy of the PITP lipid exchange cycle tems, respectively. The reaction coordinate () was chosen as the z coordinate of the lipid headgroup with boundaries at the plane of the membrane-water interface and at the upper regions of the lipid-binding pocket. In each window, a harmonic restraining potential of 1000 kJ ϫ mol Ϫ1 ϫ nm 2 was applied in the z direction on the distance between the center of mass of the given headgroup and the PtdCho bilayer bound with PITP␣. Finally, the free-energy profile for both PLs was constructed with the weighted histogram analysis method (53) based on the biased distribution of the given PL and magnitude of the umbrella biasing potential along the reaction coordinate. All free-energy calculations were performed with an identical set of simulation parameters as described above for the equilibrium simulations.

Visualization
All the snapshots and movies presented in this work were prepared using the VMD package (54). Molecular graphics and analyses of Pitpna single-nucleotide polymorphisms were performed with the University of California, San Francisco Chimera package developed by the Resource for Biocomputing, Visualization, and Informatics at the University of California, San Francisco.

Yeast strains and media
Yeast were grown in YPD liquid medium (1% yeast extract, 2% bactopeptone, and 3% glucose), or SD liquid medium (0.67% nitrogen base, 3% glucose with appropriate amino acids supplemented) (61). The respective plate media included 2% agar. Yeast strains CTY182 (MATa ura3-52 lys2-801 ⌬his3-200), and CTY1-1A (MATa ura3-52 lys2-801 ⌬his3-200 sec14-1 ts ) were previously described (1,2,62). PITPNA-expressing yeast strains were constructed by genomic integration of a single copy cassette where the PITPNA open reading frame was expressed under the control of the powerful and constitutive PGK1 promoter (P PGK ) and configured in tandem with a functional HIS3 selection marker. The integration cassette was recombined into the LEU2 locus of the appropriate recipient yeast strain by standard lithium acetate transformation methods (62). Transformants were selected on media lacking histidine, and correct integration of the cassette was confirmed by screening for leucine auxotrophy as unselected phenotype. PITPNA overexpression strains were generated by transformation with the PITPNA expression cassette (driven by the powerful constitutive PMA1 promoter) subcloned into a yeast URA3 episomal plasmid pDR195 (63,64). Transformants were selected on uracil-free medium.

Site-directed mutagenesis
The Rattus norvegicus PITPNA gene was mutated by sitedirected mutagenesis in the P PGK1 -PITPNA HIS3 yeast LEU2 integration cassette, in the pDR195 (P PMA1 -PITPNA) vector, and in the pET28b-His8-PITPNA protein expression plasmid. Mutations were confirmed by DNA sequencing. Nucleotide sequences for primers the corresponding primers are available from the authors by request.

Protein purification
Recombinant PITP␣ and mutant versions were expressed from pET28b-His8-PITPNA vectors propagated in E. coli BL21 (RIL/DE3; New England BioLabs Inc., Ipswich, MA). Bacteria were cultured in 4 liters of TB ϩ kanamycin/chloramphenicol medium for 4 h at 37°C, shifted to 16°C, and induced for recombinant protein expression with isopropyl ␤-D-thiogalactopyranoside (final concentration, 100 M). The cultures were harvested after 18 h, and the cells were disrupted by passage through a French Press in low-phosphate lysis buffer (300 mM NaCl, 25 mM sodium phosphate, 5 mM 2-mercaptoethanol, pH 7.5). Clarified lysates were incubated with TALON metal affinity beads (Clontech), and bound proteins were step-eluted with imidazole (25-200 mM). Proteins were concentrated using the Amicon Ultra filter system (EMD Millipore), and concentration as determined by SDS-PAGE and comparison against a BSA standard mass series.
PyrPtdCho and PyrPtdIns binding and transfer measurements were performed using a real-time dequenching assay as described by Somerharju et al. (65) with modification. For the transfer measurements, two vesicle populations were used. The donor (with TNP-PtdEtn quencher) and acceptor (without TNP-PtdEtn quencher) vesicles were prepared as follows. For the acceptor vesicles stock solutions of egg PtdCho and egg phosphatidic acid were mixed in a 291/9 nmol (97/3 mol %) Anatomy of the PITP lipid exchange cycle ratio and dried under a stream of N 2 . The resulting lipid film was then hydrated with 2 ml of low phosphate buffer (25 mM Na 2 HPO 4 , 300 mM NaCl, pH 7.5) and sonicated on ice for 10 min. For the donor vesicles stock solutions of egg PtdCho, PyrPtdCho/PyrPtdIns, and TNP-PtdEtn were mixed in a 4/0.5/ 0.5 nmol ratio, solvent was evaporated, and the lipid film was resuspended in 10 l of EtOH. The solution was injected into the buffer containing the acceptor vesicles. After a 5-10-min equilibrium period, the fluorescence intensity (excitation, 346 nm; emission, 378 nm) was measured as a function of time at 37°C using a ISS K2 fluorimeter (ISS Inc., Champaign, IL).
To measure transfer from donor to acceptor, 1 g of protein was injected every 250 s for a total of four injections. To obtain relative transfer efficiency, the initial slope after each protein injection was calculated, and the mean slope values for each protein (n ϭ 4) was normalized to that calculated for purified PITP␣. For binding measurements, donor vesicles were injected into 2 ml of low phosphate buffer. After a 5-10-min equilibrium period, the solution was titrated 10ϫ with 0.1 nmol of recombinant protein, and fluorescence intensity was measured as a function of wavelength (excitation, 343 nm; emission, 360 -450 nm) at 37°C (K2 fluorimeter). The intensity at 348 nm was then plotted against protein amount, and the slope value for each protein was calculated and normalized against PITP␣ to obtain the relative binding efficiency.