Interactions of Phosphatase and Tensin Homologue (PTEN) Proteins with Phosphatidylinositol Phosphates: Insights from Molecular Dynamics Simulations of PTEN and Voltage Sensitive Phosphatase

The phosphatase and tensin homologue (PTEN) and the Ciona intestinalis voltage sensitive phosphatase (Ci-VSP) are both phosphatidylinositol phosphate (PIP) phosphatases that contain a C2 domain. PTEN is a tumor suppressor protein that acts as a phosphatase on PIP3 in mammalian cell membranes. It contains two principal domains: a phosphatase domain (PD) and a C2 domain. Despite detailed structural and functional characterization, less is known about its mechanism of interaction with PIP-containing lipid bilayers. Ci-VSP consists of an N-terminal transmembrane voltage sensor domain and a C-terminal PTEN domain, which in turn contains a PD and a C2 domain. The nature of the interaction of the PTEN domain of Ci-VSP with membranes has not been well established. We have used multiscale molecular dynamics simulations to define the interaction mechanisms of PTEN and of the Ci-VSP PTEN domains with PIP-containing lipid bilayers. Our results suggest a novel mechanism of association of the PTEN with such bilayers, in which an initial electrostatics-driven encounter of the protein and bilayer is followed by reorientation of the protein to optimize its interactions with PIP molecules in the membrane. Although a PIP3 molecule binds close to the active site of PTEN, our simulations suggest a further conformational change of the protein may be required for catalytically productive binding to occur. Ci-VSP interacted with membranes in an orientation comparable to that of PTEN but bound directly to PIP-containing membranes without a subsequent reorientation step. Again, PIP3 bound close to the active site of the Ci-VSP PD, but not in a catalytically productive manner. Interactions of Ci-VSP with the bilayer induced clustering of PIP molecules around the protein.

M any cell signaling events are triggered by the association of peripheral membrane proteins with the membrane. 1−5 The cell membrane acts both as a scaffold for the localization of peripheral proteins and as a two-dimensional platform that allows diffusion on the membrane surface, resulting in the formation of protein−lipid complexes. 6 Association of peripheral proteins with specific lipids in the membrane (e.g., phosphatidylinositol phosphates or PIPs) occurs via lipid-binding modules. 7−13 Indeed, it has been shown that the majority of human kinases contain at least one lipidbinding module, 4 demonstrating the importance of the peripheral protein−lipid association in many cellular events. The main binding modules that have been identified in mammals are the C1, C2, FERM, PX, and PH domains. 14 This work will focus on two related proteins that contain a C2 domain and catalyze the dephosphorylation of PIPs: the intensively studied PTEN (phosphatase and tensin homologue) tumor suppressor and the less well characterized voltage sensitive phosphatase from Ciona intestinalis (Ci-VSP).
C2 domains possess an antiparallel β-sheet architecture with variable loops connecting the eight β-sheets. 15,16 They can be grouped into two types: C2 domains that associate with the membrane in a Ca 2+ -dependent manner and C2 domains that bind to the membrane in a Ca 2+ -independent manner. 17 Both types of C2 domains have been shown to interact with anionic lipids, such as phosphatidylserine (PS) and PIPs, in the plasma membrane. 6,7,11,18 PTEN and related proteins (e.g., Ci-VSP and auxilin) contain Ca 2+ -independent C2 domains, 19,20 the loops of which are thought to form direct interactions with anionic lipids. For example, recent simulation studies of the auxilin PTEN-like domain have shown that the loops of its C2 domain determine its orientation relative to the membrane and promote PIP clustering around the bound protein. 21 PIPs serve as second messengers in many signaling events and are involved in several pathological defects. 22 PIPs have an inositol headgroup that can be phosphorylated at different positions, creating different PIP species. For example, PI(4,5)P 2 and PI(3,4,5)P 3 are the major PIPs in the plasma membrane. 23 The exact percentage of different PIPs in the plasma membrane is difficult to determine because of the reversible turnover of PI(4,5)P 2 to PI(3,4,5)P 3 and other PIP species. It is generally stated that PI(4,5)P 2 comprises ∼5% of all phospholipids in the cytoplasmic leaflet of the plasma membrane. 24,25 For comparison, phosphatidylserine is the most abundant anionic phospholipid in eukaryotic cells and comprises approximately 20% of plasma membrane lipids. 26 PTEN is a cytosolic enzyme that when bound to the inner leaflet of the plasma membrane catalyzes dephosphorylation of PI (3,4,5)P 3 to PI(4,5)P 2 . 27 By reducing the level of PI (3,4,5)P 3 in the inner membrane leaflet, PTEN negatively regulates the phosphatidylinositol 3-kinase (PI3K) signaling pathway, leading to a reduced level of cell proliferation. 28,29 For this reason, PTEN is a tumor suppressor and is one of the most commonly mutated protein in human sporadic tumors. 30 Mutations in PTEN may also lead to Cowden disease, Lhermitte-Duclos disease, and Bannayan-Zonana syndrome. 31 PTEN has four domains: an N-terminal PIP 2 -binding module, a phosphatase domain (PD), a C2 domain, and a Cterminal tail ( Figure 1A). The N-terminal binding module specifically binds PIP 2 molecules, ensuring selective binding of PTEN to the inner leaflet of the plasma membrane. The PD contains the PI(3,4,5)P 3 catalytic binding site, including the Ploop consisting of the signature HCXXGXXR motif, common with many protein tyrosine phosphatases (PTPs). 32 The proposed binding site for PI(3,4,5)P 3 corresponds to the site of a bound tartrate molecule in the crystal structure and is supported by mutational data. 27,33,34 The positively charged face of the C2 domain of PTEN is believed to bind to anionic lipids, e.g., phosphatidylserine and PIPs, in the membrane. PTEN lacks the aspartate residues associated with Ca 2+dependent binding. 19 The C-terminal tail of PTEN is not needed for phosphatase activity but has been shown to have regulatory features. 29,35−38 There are more than 2000 mutations in PTEN associated with cancer and other diseases (http://www.sanger.ac.uk/perl/ genetics/CGP/cosmic). These mutations appear in all four domains of the protein. For example, the R130G mutation (identified in endometrial, ovarian, lung, and central nervous system cancer cells) is in the PD, while the R335L mutation (associated with Cowden syndrome) is located in the C2 domain. Mutations can affect PTEN by reducing or abolishing phosphatase activity or by preventing binding to the bilayer. 39 The productive orientation of PTEN on the membrane has been suggested to involve association of the C2 domain loops with the lipids. 9,19,40 Other proteins (e.g., SUMO1) may form complexes with PTEN and regulate its association with the membrane. 41,42 The C. intestinalis voltage sensitive phosphatase (Ci-VSP) catalyzes the dephosphorylation of PI(3,4,5)P 3 and PI(4,5)P 2 at the 5′-phosphate, 43,44 and PI (3,4)P 2 at the 3′-phosphate. 45 Ci-VSP is expressed in Ciona sperm tail membranes 46 and has been suggested to function in digestive system cells and blood cells. 47 Mammalian orthologs of Ci-VSP (namely, TPTE, 48 TPIP, 49 and PTEN2 50 ) are expressed in the testis. 51 The phosphatase activity of Ci-VSP is regulated by the plasma membrane potential. The membrane potential is sensed by the voltage-sensing (VS) domain, a four-α-helix transmembrane domain at the N-terminus of the protein ( Figure 1B). The VS domain is homologous to the VS domain of potassium-selective voltage-gated channels 52 and is joined to the cytosolic section by a linker domain. The cytosolic catalytic region is comprised of a PD and a C2 domain and is similar in structure to PTEN itself, with a 44% identical sequence and a high degree of conservation of catalytically significant residues 53 ( Figure 1B).
It has been suggested that membrane depolarization activates Ci-VSP by moving the PD and C2 domain toward the membrane into the optimal orientation for PIP dephosphorylation. 52 However, more recently, it has also been proposed that conformational changes in a "gating loop" (residues 398− 413) result in voltage-triggered activity due to the movement of E411, which competes with the substrate for the active site. 54 Indeed, a crystal structure of Ci-VSP with E411 in a hydrophobic cavity and not in the active site [Protein Data Bank (PDB) entry 3V0H] contains a bound IP 3 molecule at a site on Ci-VSP equivalent to that for binding of tartrate on PTEN. 54 The C2 domain of Ci-VSP is similar in structure to that of PTEN and thus is thought to contribute to membrane binding in a Ca 2+ -independent manner. However, despite similar C2 domains, there are notable differences between the PTEN CBR3 loop and the analogous Ci-VSP loop (P519−P526). While the PTEN loop contains four positively charged residues and points toward the membrane, 19 the Ci-VSP CBR3 loop contains only one positive arginine residue and points toward the PD, forming hydrogen bonds with a loop in the PD. 55 This indicates a possible regulatory role for the C2 domain. The Ci-VSP PD also shares similarities with the PTEN PD: both contain a P-loop. Ci-VSP, however, has a tyrosine residue (Y522) that fills a shallow pocket present in the PTEN active site and a glycine at residue 365, rather than the equivalent alanine in PTEN.
Molecular dynamics simulations may be used to probe the interactions of both integral and peripheral membrane proteins with lipid bilayers. 56 Such simulations have yielded initial models of the likely mode of interaction of PTEN with anionic lipid bilayers that are in good agreement with experimental biophysical data. 9,38,40 Furthermore, the resultant models of PTEN−model bilayer interactions are in good agreement with, e.g., recent mutational studies of PTEN that used exogenous expression in Dictyostelium to probe membrane association and PTEN function in living cells. 57 There have also been simulations of a homology model of the complete Ci-VSP (including the PTEN domain) bound to the surface of a PIPcontaining lipid bilayer. These have focused in particular on the role of the linker region between the voltage sensor and PTEN domains. 58 However, a number of aspects of the detailed molecular interactions of PTEN and of the PTEN domain of Ci-VSP with PIP 3 molecules in lipid bilayers remain unclear. In this work, we use multiscale molecular dynamics simulations 59 to improve our understanding of the similarities and differences in the mechanism(s) of recognition of the PIP-containing lipid bilayer by PTEN and by the PTEN domain from Ci-VSP.

■ METHODS
Structures Used in the Simulations. The PTEN crystal structure (PDB entry 1D5R, resolution of 2.10 Å) consisted of the PD and the C2 domain, 19 as did the Ci-VSP crystal structure (PDB entry 3V0H, resolution of 1.85 Å) 54 (see Figure  1). The Ci-VSP structure was crystallized with inositol 1,4,5triphosphate (IP 3 ) bound.
Lipid Bilayers. All lipid bilayers were generated by selfassembly coarse-grained MD simulations. In these simulations, POPC lipids (without any protein) were randomly placed in a simulation box and solvated with coarse-grained water molecules and ions to neutralize the system. Subsequently, a production simulation was performed for 100 ns. After the first ∼15 ns of the simulation, the bilayer was formed with an equal distribution of lipids in the two leaflets. The bilayers for the pten_away, vsp_away, and pten_bound simulations contained 466, 476, and 680 lipids, respectively (see Tables 1 and 2).
Prior to the simulations described in this study, using a locally written code the POPC lipids were randomly replaced with POPS or PIP lipids in both leaflets to reach the final concentration of lipids required for each system (see Tables  1 and 2 for the final lipid concentration). In the pten_bound simulations, PIP lipids were placed in the desired position, as shown in Figure 3A, using the GROMACS genbox command.
Coarse-Grained Molecular Dynamics Simulations (CG-MD). Coarse-grained molecular dynamics (CG-MD) simulations were performed using GROMACS version 4.5.4 60 with the Martini 2.1 force field. 61,62 In these coarse-grained models, groups of approximately four atoms are represented by a single particle. An elastic network model (ENM) 63 was applied to the backbone particles of the protein. Within the network, a harmonic potential was applied to all pairs of backbone particles falling within a cutoff distance of 7 Å to model the secondary and tertiary structure of the protein. All CG-MD simulations were equilibrated for 5 ns and then run using a time step of 20 fs. During the equilibration stage, the protein backbone particles were restrained (force constant of 10 kJ mol −1 Å −2 ). A Berendsen thermostat 64 and a Berendsen barostat were used for temperature and pressure. The LINCS algorithm was used to constrain bond lengths. 65 Note that the protein and the bilayer were fully solvated in all simulations using the standard (nonpolarizable) Martini CG water particles. 61 For the simulations with the PTEN and the Ci-VSP displaced from the bilayer, the dimensions of the simulation box were 12.5 nm × 12.5 nm × 28 nm, containing ∼30000 CG waters. In the simulations with the PTEN bound to one leaflet of the bilayer (Figure 3), the dimensions of the simulation box were 14.5 nm × 14.5 nm × 13 nm, containing ∼14500 CG waters.
Atomistic Molecular Dynamics (AT-MD) Simulations. AT-MD simulations were conducted as described in Tables 1 and 2. The final snapshots of selected CG-MD simulations were converted to AT representations using a CG-to-AT fragment-based protocol. 66 For the AT-MD simulations, the PME (particle mesh Ewald) method 67 was used to model longrange electrostatics. A V-rescale thermostat was used for temperature coupling (temperature of 323 K), and the Parrinello−Rahman barostat 68 was used for semiisotropic pressure coupling (pressure of 1 bar). The LINCS algorithm was used to constrain bond lengths. 65 Prior to the production simulations, the system was equilibrated for 2.5 ns. During the equilibration stage, the protein backbone atoms were restrained (force constant of 10 kJ mol −1 Å −2 ). Note that no restraints, other than the LINCS algorithm that constrains bond lengths, were imposed during the production AT-MD simulations. Analysis was conducted using GROMACS, PyMol (http:// www.pymol.org), VMD, 69 and locally written codes. For these simulations, the SPC water model 70    implicated in membrane binding in previous biophysical, modeling, and simulation studies. 9,38,40,57 However, a dynamic model of the mechanism of PTEN−membrane association has not been developed.
To study the association of PTEN with PIP 3 -containing bilayers, a series of three simulations [pten_away-1−3 (see Table 1)] was performed, in which the PTEN molecule was placed ∼12 nm (i.e., 120 Å) from the center of mass of a lipid bilayer containing increasing numbers of PIP 3 molecules. In these simulations, PTEN initially diffused in the aqueous environment before interacting with the bilayer (Figure 2A).
Overall, increasing the concentration of PIP 3 lipids in the bilayer enhanced the formation of the PTEN−bilayer complex. In particular, when only one PIP 3 lipid was added to each bilayer leaflet (pten_away-1), just two of six repeat simulations resulted in binding of PTEN to the bilayer ( Figure S3A of the Supporting Information), and of these two simulations that yielded a PTEN−lipid complex, only one demonstrated movement of PTEN into a productive orientation ( Figure  S3D of the Supporting Information) (a productive PTEN− bilayer complex is considered to be one that agrees with the previous experimental and computational data 9,38,40 ).
In the systems where the number of PIP 3 lipids in each leaflet was increased to two (pten_away-2) and four (pten_away-3), a productive PTEN−bilayer complex was reached in four of six simulations in each case ( Figure S3 of the Supporting Information). Note that in all cases PTEN did not initially bind to the bilayer in the productive orientation but the initial encounter was followed by rotational reorientation (see Figure  2) to form more specific interactions via the residues implicated by mutational and computational studies. 9,38,40 The resultant productive orientation was retained for the remainder of the simulation. To further determine the stability of the final orientation, all six repeat simulations of the pten_away-2 system were extended from 3 to 4 μs, resulting in PTEN adopting the productive orientation in five of six simulations. Analysis of the radial distribution functions of lipids ( Figure S4 of the Supporting Information) indicated preferential interactions of PTEN with PIP 3 and, to a lesser extent, POPS.
When similar simulations were run with the PTEN displaced from a POPC/POPS/PIP 2 (75:20:5) bilayer, the same mechanism for the formation of the PTEN−bilayer complex was observed. In particular, PTEN associates with the bilayer in a nonproductive orientation, but the initial encounter was followed by rotational reorientation to move PTEN in a productive orientation. Productive association of PTEN with the bilayer resulted in clustering of PIP 2 lipids around the protein in the adjacent bilayer leaflet. See Figure S9 and the text of the Supporting Information for more details.
Interactions of a PTEN−Bilayer Complex with PIP 3 Molecules. Following definition of the association mechanism to form a "productive" PTEN−bilayer complex, we investigated how PIP 3 might interact with a preformed PTEN−bilayer complex. To this end, simulations [pten-bound-1−3 (Table 1)] were initiated with PTEN bound to a POPC/POPS bilayer in a productive orientation (as derived from the simulations described above), but with PIP 3 molecules placed in the bilayer away (∼60 Å) from the bound PTEN molecule ( Figure 3A). In  Table 1) with the PTEN molecule displaced from a bilayer that contained two PIP 3 molecules (green) in each leaflet. Simulation snapshots are shown at 0, 1.2, 1.7, and 3 μs. The blue arrows indicate the process of initial encounter of the protein and bilayer followed by reorientation of the protein at the bilayer surface. All the systems shown in this figure were fully solvated with CG water particles. However, for the sake of clarity, these are not shown. (B and C) Progress of selected CG-MD simulations of PTEN and Ci-VSP with PIP 3 -containing bilayers [simulations pten_away-2 and vsp_away-4 (Tables 1 and 2)]. Panel B shows the distance between the center of mass of the protein and the center of mass of the bilayer as a function of time. Panel C shows the cosine of the angle between the protein plane (as defined by the protein's principal z axis) and the bilayer plane. This angle is equal to 0°(and hence the cosine is equal to 1) if the protein is in the "correct" binding orientation (see the text for more information).  Table 1 the first case, a single PIP 3 lipid was placed in seven different positions in the bilayer. Four of six simulations in which the PIP 3 was displaced from PTEN revealed PIP 3 moving into a cleft between the PD and the C2 domain ( Figure 3 and Figure  S5 of the Supporting Information). The cleft was comprised mainly of PTEN residues K128 (from the active site P-loop), K164, V166, and T167 (from the T1 loop of the PD), and K327 and K330 from the C2 domain Cα2 loop. Interestingly, immediately after the initiation of the simulations, the PD partly dissociated from the bilayer and rebound to the bilayer only after the formation of the PTEN−PIP 3 complex ( Figure  3C).
Two different routes of the PIP 3 to the PD−C2 cleft were observed in the four simulations, two via an initial interaction with the PD ( Figure 3B) and two in which the initial interaction occurs via the C2 domain ( Figure 3B). Both routes lead to PIP 3 binding to the PD−C2 cleft described above and were exploited an equal number of times. Detailed analysis of the routes, however, shows that on its way to the cleft, the PIP 3 did not interact with exactly the same residues on the PD or on the C2 domain. Simulations of in silico mutations that reversed the charge of the positive residues in the cleft and surrounding regions (i.e., K164E, K269E, K327E, and K330E; system pten_bound-3 in Table 1) exhibited no binding of PIP 3 to the cleft site (see the Supporting Information).
Similar CG-MD simulations were then run with four PIP 3 molecules in the PTEN-bound leaflet (pten_bound-2). This resulted in four of six simulations with a PIP 3 bound in the aforementioned cleft, three of which went via the C2 domain route and one via the PD route. The interactions with the cleft were again predominantly those with residues K330 and K164. When no PIPs interacted with the cleft or the PD, PTEN again showed transient interaction of the PD with the bilayer. Once interacting with PTEN, the PIP 3 molecules tended to remain bound to the same region of the protein for extended periods of time.
To determine whether PTEN can differentiate between PIP 2 and PIP 3 molecules, similar simulations (compared to those described above) were run with a bilayer containing both four PIP 2 molecules and four PIP 3 molecules. Two of six simulations showed binding of PIP 3 and three of six binding of PIP 2 (the remaining simulation showed no binding) to the cleft. Therefore, it seems that both PIP types can bind to the PTEN cleft.
Atomistic Simulations. Atomistic systems were derived from the CG systems described above to study in more detail the interactions of PIP 3 with PTEN. AT-MD simulations were set up using the final snapshot from one of the pten_bound-1 simulations and one of the pten_bound-2 simulations. In both simulations, the PIP 3 lipid had bound in the aforementioned cleft. Three repeats (with different initial velocities) for each system were run for 50 ns (Table 1). In both systems, PIP 3 moved up partially out of the bilayer to interact with PTEN ( Figure 4A,B) and then remained there for the remainder of the simulation. The contacts made between PTEN and PIP 3 formed in the corresponding CG-MD simulations were maintained during the (short) atomistic simulations ( Figure  S5 of the Supporting Information). In particular, PIP 3 can be seen to form interactions with the P loop (residues 124−129) and T1 loop (residues 164−171) of the PD active site, although the PIP 3 headgroup is displaced slightly from the active site formed by the H 123 CXXGXXR 130 motif of the P loop ( Figure  5A). Thus, the three major areas of frequent contact between PTEN and PIP 3 were residues K164 and T167 of the T1 loop and residue K330 of the C2 domain. PTEN continued to interact with the bilayer in the productive orientation described for the CG-MD simulations with the C2 domain loops penetrating into the bilayer ( Figure 6A). Comparison of the  PTEN structure from the atomistic simulations with the crystal structure suggests a degree of conformational change corresponding to a rotation of ∼10°of the PD relative to the C2 domain ( Figure 5B), although more extended simulations may be needed to capture the full extent of this transition.
Interaction of Ci-VSP with Membranes. Ci-VSP is a rather specialized example of a PTEN-like enzyme. Unlike PTEN, Ci-VSP is permanently anchored to the bilayer via an N-terminal voltage-sensor (VS) domain, more generally seen in voltage-gated potassium, sodium, and related ion channels. The PTEN domain from Ci-VSP has an electrostatic surface somewhat different from that of PTEN ( Figure S1 of the Supporting Information), with a molecular dipole; its positive pole points directly to the presumed location of the membrane surface ( Figure S2 of the Supporting Information). Furthermore, despite the overall conservation of fold, the lipid-binding loops of the C2 domain differ between the two structures. Therefore, we performed a series of simulations of Ci-VSP PTEN (Table 2) comparable to those described above for PTEN itself.
In the absence of PIP molecules, the Ci-VSP PTEN domain did not exhibit any interactions with the bilayer ( Figure S6 of the Supporting Information). The presence of PIP 2 or PIP 3 resulted in interactions with the bilayer such that the Ci-VSP domain adopted an orientation comparable to the productive mode of PTEN discussed above ( Figure S6 of the Supporting Information). Interestingly, in the simulations that resulted in a Ci-VSP−bilayer complex, the final orientation of Ci-VSP relative to the bilayer was similar to that of PTEN ( Figure  2). However, the Ci-VSP PD made strong interactions with the bilayer unlike the PTEN PD. This is consistent with the difference in the electrostatic potential surface between the two proteins. Calculation of the orientation of Ci-VSP relative to the bilayer ( Figure 2B,C and Figure S6 of the Supporting Information) revealed binding without subsequent reorientation of the domain relative to the bilayer surface. Of course, in vivo Ci-VSP PTEN is more likely to be bound to the membrane as it is tethered by the nearby N-terminal VS domain. Overall, these simulations provide support for a model in which PTEN of Ci-VSP interacts all of the time with the bilayer and is voltage-activated via a conformational change rather than a voltage-driven on and off movement of the PTEN domain to and from the membrane. 54 Atomistic simulations were performed, using the end points from the vsp_away-3 and vsp_away-4 simulations as starting configurations ( Figure 4C,D). Calculation of the contacts between Ci-VSP and the lipids demonstrated that the three regions of Ci-VSP PTEN forming the principal contacts were the same for both PIP 2 and PIP 3 ( Figure S7 of the Supporting Information). These regions included contributions from both the PD and the C2 domain. The most conserved site of protein−PIP interaction between the PIP 2 and PIP 3 simulations was between residues 362 and 369 (particularly K364 and G365), i.e., corresponding to the H 362 SxxGxxR 369 catalytic loop. As for PTEN (see above) in the atomistic simulations, the PIP 2 or PIP 3 molecule is close to the active site but is not found directly between the two catalytic residues (S363 and A369), unlike IP 3 in the X-ray structure. It is interesting to note that unlike PTEN, the corresponding domain from Ci-VSP does not penetrate beyond the surface of the bilayer. In particular, whereas for PTEN the CBR3 loop penetrates deep into the headgroup phosphates, for Ci-VSP the CBR3 loop sits closer to the phosphate−aqueous interface ( Figure 6). This is suggestive of the possibility of a further conformational change linked to tighter binding of the PIP 3 substrate, although more extended simulations might be needed to test this.

■ DISCUSSION
Our simulations have allowed us to model the nature of the PTEN−membrane association process. The predicted final orientation of PTEN relative to the membrane is in good agreement with previous experimental data and computational studies. 9,19,38,40,71,72 An increased concentration of PIP 3 resulted in more frequent association of PTEN with the model bilayer, and PTEN could achieve its optimal orientation without SUMOylation. 41 To reach the preferred orientation, PTEN was observed to be able to reorient at the membrane surface following an initial encounter with the lipid bilayer. It should be noted that in our simulations only a low concentration of PIP 3 lipids in the bilayer was required for the formation of a productive PTEN−membrane complex. Recent lipidomics data suggest that the global PIP 3 concentration is expected to be lower than, for example, the PIP 2 concentration in the plasma membrane, although it should be noted that it is difficult to be certain of the exact concentration of the PIP 3 and PIP 2 molecules because of the reversible turnover of PIP 2 to PIP 3 and other PIP species. However, the local concentration of PIP 3 is not known, and it remains possible that a high local concentration of PIP 3 may drive the formation of the PTEN−membrane complex. Simulation with bilayers containing the more abundant PIP 2 at physiological concentrations also resulted in the same mechanism for the formation of the PTEN−membrane complex. A similar encounter-plus-reorientation mechanism has been observed for, for example, the auxilin PTEN-like domain. 21 Interestingly, the interaction of the PTEN domain from Ci-VSP did not generally require a reorientation step. In simulations starting with prebound PTEN at the bilayer surface, PIP 3 molecules were able to enter the PD−C2 cleft via lateral diffusion along with some degree of "loosening" followed by "tightening" of the PTEN−bilayer interaction.
In CG simulations linked to subsequent atomistic simulations, the PIP 3 molecule interacted with residues close the PTEN active site [identified via a tartrate bound in the crystal structure ( Figure 5A and Figure S5 of the Supporting Information)]. However, the bound PIP 3 in the simulations was somewhat displaced from the optimal binding mode at the active site suggested by the crystal structure ( Figure S8 of the Supporting Information). Furthermore, molecular dynamics simulations with a PIP 3 molecule initially docked using the tartrate coordinates as a guide did not result in the PIP 3 molecule remaining bound to this site. 9 Interestingly, docking of PTEN inhibitors to the tartrate binding site has been demonstrated after the removal of the C2 domain. 73 Because the C2 domain is more cationic than the PD, it is perhaps not surprising that the negatively charged PIP 3 molecule binds initially to a site formed between the PD and the C2 domain and contains multiple lysine residues. It is also possible that a degree of conformational change is needed, and indeed, some limited change in the relative orientation of the PD and the C2 domain is seen during our (short) atomistic simulations. However, it is likely that more extended atomistic simulation studies, using, for example, metadynamics 74 for better sampling, will be required to explore this aspect in more detail.
In the case of Ci-VSP, the structural similarity to PTEN 75 would suggest a similar orientation of binding to the membrane. The simulations described here are in agreement with this (Figure 4), indicating that the Ci-VSP C2 domain binds to the membrane via a Ca 2+ -independent mechanism as is the case for PTEN 19 and auxilin. 20 Ci-VSP PTEN adopted a preferred orientation in simulations directly after initial membrane binding, without a need for subsequent reorientation, reflecting a difference in the molecular dipole orientation between PTEN and Ci-VSP ( Figure S2 of the Supporting Information). In simulations with PIPs present, the residues that formed a substantive number of contacts with PIP 2 or PIP 3 molecules are in the proximity of the IP 3 substrate in the crystal structure (see Figure S8 of the Supporting Information). However, as for PTEN, PIP 3 did not interact directly with the active site, again suggesting the need for some degree of conformational change and/or more exhaustive sampling of interactions in simulations. In this context, it is of interest that crystal structures of Ci-VSP PTEN have indicated conformational changes may be required for binding of the substrate to the active site of the enzyme. 54 It is important to consider possible limitations of the current simulation studies. In particular, one must consider the approximations implicit in the use of a CG-MD-based approach. This has been shown to correctly predict interactions of PIPs with integral membrane proteins 76,77 and also the interactions of peripheral proteins with anionic lipid bilayers. 78 However, the use of an elastic network model within the coarse-grained simulations means that possible conformational changes of the protein are not accurately represented. The use of a multiscale approach, with subsequent atomistic simulations, in part addresses this limitation. However, as noted above, the AT-MD simulations are relatively short and more extended simulations and/or sampling is likely to be needed to more fully address possible conformational changes linked to, for example, binding of PIP 3 to the PTEN active site.
A more "biological" limitation is the focus on the core PTEN domains (i.e., PD and C2 domain), thus excluding the N-and C-terminal tails of PTEN, and the VS and linker region of Ci-VSP. This is necessitated by limited structural data, although there have been simulations of a homology model of the complete Ci-VSP protein. 58 There are not any structural data for the PTEN N-terminal domain or for the PTEN C-terminal tail. In particular, the C-terminal tail is thought to be a long, unstructured, and flexible region, which makes modeling or simulation rather difficult. It seems that the C-terminal tail may adopt different dynamic conformations in the membranebound and membrane-unbound states of PTEN. 38,57 In particular, recent data suggest that the C-terminal tail may interact with those regions of PTEN that otherwise can interact with lipids in the plasma membrane, thus providing an autoinhibitory mechanism. 57 Consequently, in future studies, it will require rather detailed and careful simulations, possibly via an enhanced sampling approach, to fully capture the conformational energy landscape of the C-terminal tail. This will be a prerequisite for creating a predictive dynamic model of full length PTEN and of its interaction with the membrane. It will also be important to treat more fully some of the lipid compositional complexities of mammalian cell membranes. 79 To summarize, our multiscale simulation approach offers novel insights into the molecular mechanism of association of PTEN proteins with model membranes, providing details of the interactions of these proteins with PIP 3 . As discussed by, for example, Nguyen et al., 57 the majority of PTEN in a cell is not bound to the membrane, so an improved understanding of the mechanism of PTEN−bilayer interaction offers a longer term prospect of developing activators of PTEN function for therapeutic applications.