Phospholipid-binding Sites of Phosphatase and Tensin Homolog (PTEN)

Background: PTEN activity is enhanced by PI(4,5)P2 via either an allosteric or interfacial effect. Results: 31P field cycling NMR with spin-labeled PTEN detects a PI(4,5)P2-binding site near but not in the active site. Conclusion: A model for PTEN identifies Lys-13 and Arg-47 as part of the PI(4,5)P2 site. Significance: Occupation of the PI(4,5)P2-binding site on PTEN stabilizes productive interfacial binding. The lipid phosphatase activity of the tumor suppressor phosphatase and tensin homolog (PTEN) is enhanced by the presence of its biological product, phosphatidylinositol 4,5-bisphosphate (PI(4,5)P2). This enhancement is suggested to occur via the product binding to the N-terminal region of the protein. PTEN effects on short-chain phosphoinositide 31P linewidths and on the full field dependence of the spin-lattice relaxation rate (measured by high resolution field cycling 31P NMR using spin-labeled protein) are combined with enzyme kinetics with the same short-chain phospholipids to characterize where PI(4,5)P2 binds on the protein. The results are used to model a discrete site for a PI(4,5)P2 molecule close to, but distinct from, the active site of PTEN. This PI(4,5)P2 site uses Arg-47 and Lys-13 as phosphate ligands, explaining why PTEN R47G and K13E can no longer be activated by that phosphoinositide. Placing a PI(4,5)P2 near the substrate site allows for proper orientation of the enzyme on interfaces and should facilitate processive catalysis.

The phosphatase and tensin homolog (PTEN) 2 deleted on chromosome 10 is a tumor suppressor that regulates the PI3K/ Akt signaling pathway. In cells where PTEN function is lost, excessive PI(3,4,5)P 3 recruits and activates Akt and other downstream signaling molecules at higher levels, leading to abnormal cell growth and proliferation, which in turn causes many diseases (1). The deletion, insertion, and/or mutation of the PTEN gene are found in a wide variety of tumor cells, including endometrial cancer, glioblastoma, prostate cancer, breast cancer, and lung tumor as well as in many syndromes (1,2). Overexpression of PTEN in tumor cells that have lost PTEN expression suppresses cell growth and tumorigenicity (3). Along with serving as a tumor suppressor, PTEN has roles in other cellular processes, including cell metabolism, cell polarity, stem cell self-renewing activity, and neuronal injury (1,4). Most of the functions of PTEN are linked to its ability to dephosphorylate phosphoinositides.
The structure of a truncated form of PTEN (12) showed it is composed of the N-terminal catalytic phosphatase domain and a C-terminal lipid-binding C2 domain (Fig. 1A). The construct was missing 7 residues from the N terminus, 23 residues from a loop, and 52 residues from the C terminus. Relevant to phosphoinositide binding, residues 7-13 were not visible in the elec-tron density map. Tartrate occupied the active site and was replaced with inositol 1,3,4,5-tetrakisphosphate in silico to have a sense of substrate binding. Molecular dynamics simulations of different dioctanoyl phosphatidylinositols bound to the catalytic domain of PTEN suggested that adjacent to the active site is another region that can play a role in monomer ligand binding (13). Those simulations suggested that there are specific interactions between the acyl chain of these lipids and the ␤2-␣1 loop containing Arg-47. Mutants of Arg-47 showed much lower activity compared with the wild type protein regardless of the identity of the mutated amino acid, indicating the importance of this particular residue for substrate binding and/or catalysis (13). The interactions of the ␤2-␣1 loop could facilitate the binding of several 3-deoxy-PI derivatives that inhibit enzyme activity (14). Detergent molecules, such as Triton X-100, were shown to reduce the kinetic inhibition caused by the deoxy-PI molecules and hence were proposed to bind in this hydrophobic pocket termed the "adjacent hydrophobic site" (AHS).
Thus, there is a possibility that there are three functionally distinct regions in the catalytic domain of PTEN that can bind phospholipids, two quite specifically (the active site and PI(4,5)P 2 site), whereas the AHS can be occupied by a number of amphiphiles. Discrete binding of the different phospholipids to PTEN could contribute to the regulation of this phosphatase. Several methods, including FRET (11), neutron reflectometry (15), and surface plasmon resonance (16), have been used to investigate PTEN or its isolated C2 domain binding to phospholipid vesicles. However, these methods are based on the overall partitioning of the protein onto the phospholipid surface. Location of an individual lipid-binding site on the protein is harder to define and usually relies on mutants that show altered kinetic or binding behavior. Coupled with these binding studies is the suggestion, based on recent MD simulations, that the protein undergoes a rotation of the phosphatase with respect to the C2 domain (17). The rotation in the membranebound conformation of PTEN allows it to access its substrate (Fig. 1B).
In this study, we provide evidence on how PTEN and its N-terminal point mutant (K13E) as well as an active site mutant (C124S) bind to phospholipid micelles. Both fixed high magnetic field and high resolution field cycling 31 P NMR spectroscopy using spin-labeled protein (18 -21), coupled with enzyme kinetics, are used to characterize binding of a series of phosphoinositides to PTEN. The results, which use short-chain phosphoinositides, identify a discrete site for a PI(4,5)P 2 molecule close to, but distinct from, the active site on PTEN. A proposed structure for membrane-bound PTEN is generated by modeling the N-terminal segment as an interfacial helix (Fig.  1B) and using the constraints from the NMR results and kinetic behavior of mutant PTEN enzymes to identify this discrete binding site for a PI(4,5)P 2 molecule. Its location near the active site would aid in keeping the protein at the surface as substrate diffuses in and product out of the active site. Thus, activator binding provides PTEN with an anchor for processive catalysis. We also demonstrate that K13E, which shows no kinetic activation, can still bind diC 8 PI in the active site (although in a slightly different orientation) with diC 8 PI(4,5)P 2 able to bind to this protein as well. However, the NMR results are consistent with the activator molecule having a shorter residence time on the protein. These results suggest a mechanism for how kinetic activation of PI(3)P or PI(3,4)P 2 by PI(4,5)P 2 occurs.
Overexpression of PTEN and Construction of PTEN Mutants-The pET28a-PTEN plasmid construction has been described previously (16). Generation of recombinant PTEN FIGURE 1. Ribbon diagrams of the crystal structure of PTEN in the absence and presence of substrate-containing membrane. A, PTEN crystal structure (Protein Data Bank code 1D5R) with colors from blue to red for N to C termini. Yellow spheres represent sulfur atoms of Cys residues that are spinlabeled in the NMR experiments. The tartrate molecule bound at the postulated active site is in van der Waals sphere representation. The arrow indicates a postulated movement of the two domains upon membrane binding. B, model for membrane-bound PTEN with the missing N-terminal peptide docked as an interfacial helix (pink). The sulfur of Cys-124 is depicted as a yellow sphere and a diC 6 PIP 3 molecule (magenta) is docked in the active site. The postulated conformational change is based on that suggested by Kalli et al. (17). followed a previously published protocol with minor modifications (13,14) that are summarized below. The plasmid was transformed into Escherichia coli BL21-CodonPlus (DE3)-RIL cells from Stratagene, and the cells were grown at 37°C until the A 600 reached 0.8. Protein expression was induced with 0.1 mM isopropyl ␤-D-thiogalactoside, and the cultures were incubated at 16°C for 20 h. Cells were harvested by centrifugation and stored at Ϫ20°C until needed. His 6 -tagged PTEN was purified from the cell extracts using a Ni-NTA-agarose column (Qiagen). The purification process was conducted at 4°C with 10 mM ␤-mercaptoethanol added to all buffers. Pure protein (Ͼ90% as judged by SDS-PAGE) was combined and dialyzed against 100 mM Tris-HCl, 1 mM DTT, pH 8.0 (a sample of the elution from the Ni-NTA affinity column is shown in Fig. 2, top); protein concentration was determined by UV absorbance as well as Lowry assay. The genes for PTEN mutants K13E, R47K, and C124S were constructed by site-specific mutagenesis using a mutagenesis kit from Stratagene. All mutations were confirmed by DNA sequencing (GeneWiz). Protein expression and purification of the mutant PTEN proteins followed the procedure for the wild type protein (13). The typical yield was 1-1.5 mg/liter culture. CD spectra of the mutant proteins showed that they all were well folded and had essentially the same wild type protein secondary structure (spectra for recombinant PTEN and R47K are shown in Fig. 2. bottom).
Enzyme Kinetics-The PTEN phosphatase assay for release of inorganic phosphate has been described previously (14). Generally, the assays were carried out at 37°C in 50 mM Tris, pH 8.0, with 2 mM EDTA, 10 mM DTT, and substrates. In a "monomer" assay system, 0.5 mM diC 8 PI(3)P or 0.1 mM diC 8 PI(3,4)P 2 was used (the critical micelle concentration (CMC) for each compound is Ն0.5 mM). The amount of protein added to each sample was chosen to obtain 5-20% substrate hydrolysis in 20 min. diC 8 PI(4,5)P 2 and diC 7 PC concentrations in the assays were one-half, or equal to, that of the substrate. The Triton X-100 concentration was four times the substrate concentration. A mixed micelle system was also used with longchain phosphoinositides DOPI(3,4)P 2 and DOPI(4,5)P 2 dissolved in Triton X-100 with a phospholipid to detergent ratio of 1:4. All these assays were done multiple times (and at least in duplicate) using different enzyme preparations. The specific activities presented are the mean Ϯ S.D.
For assays using vesicles, large unilamellar vesicles (LUVs) containing 5 mM total lipids with the compositions DOPI(3,4)P 2 / POPC (5:95) or DOPI(3,4)P 2 /DOPI(4,5)P 2 /POPC (5:5:90) were made by mixing the designated amount of lipids in chloroform, removing the organic solvent under a stream of nitrogen, followed by lyophilization overnight, resuspending the dry film in 50 mM Tris, pH 8.0, and then extruding the suspension through a 0.1-m polycarbonate membrane to form LUVs with an average diameter of 1000 Å. Enzyme assays used 1 mM (total phospholipid) LUVs as the substrate. These assays were conducted in duplicate, and the reported activity is the average of the two assays.
Fixed High Field 31 P NMR-31 P NMR spectra of short-chain phospholipids (diC 8 PI, various phosphorylated diC 8 PI molecules, and diC 7 PC) were obtained at 242.7 MHz on a Varian 600 MHz VNMRS system at 25°C. The protein, in 100 mM Tris, 2 mM EDTA, pH 8.0, was added to the lipid mixture (0.25 to 1.0 mM phospholipid) to a final concentration of 0.5 mg/ml (10.1 M). The peak width at half-height (⌬1 ⁄ 2 ) was measured, and the linewidth difference ⌬⌬1 ⁄ 2 was calculated by subtraction of the linewidth of phospholipids alone from the linewidths of protein-bound phospholipid resonances. 31 P spectra were acquired for every lipid concentration at least in duplicate and usually in triplicate using different samples and different protein preparations to obtain a mean value and standard deviation for the change in linewidth caused by addition of the protein.
Spin-labeling PTEN-Prior to labeling of PTEN cysteine groups, the protein (4 mg/ml protein solution in 100 mM Tris, pH 8.0) was incubated for 30 min with 10 mM DTT to reduce any disulfide bonds. The protein was then exchanged into 20 mM Tris, pH 7.5, and excess DTT was removed using a micro Bio-Spin 6 column (Bio-Rad). MTSL, from a 20 mg/ml stock in acetone, was added to the protein solution at a ratio of protein/ MTSL ϭ 1:4, based on the concentration of the protein prior to reduction and elution from the spin column. However, because about 50% of the protein is lost in that step, the ratio of spinlabeling reagent to protein was considerably higher (ϳ10:1). The incubation time, 1 h for PTEN and R47G and 1.5 h for K13E, was chosen so that greater than 80% of the PTEN activity was lost. Two spin columns removed excess MTSL and exchanged the protein back into 100 mM Tris, pH 8.0.

Phosphoinositide-binding Sites on PTEN
The labeling ratio was calculated by [Cys] free, PTEN-SL / [Cys] free, PTEN-WT ⅐10 and showed that for the wild type recombinant protein 8 Ϯ 1 cysteines were modified by the spin label.
Mass Spectrometry (MS) Detection of Modification Sites of Spin-labeled PTEN-LC-MS/MS was used to monitor the extent of MTSL modification of the 10 cysteines on PTEN. The protein, 10 g of wild type or spin-labeled in 50 mM Tris, pH 8.5, was incubated in 8 M urea for 30 min. The sample was subsequently diluted eight times with 50 mM Tris, pH 8.5, and LysC endoproteinase (Wako) was added in a ratio of 1:50 (w/w) endoproteinase/PTEN. After overnight digestion at 37°C, the peptide mixture was desalted on a C18 spin column (Pierce) and dried using a Speedvac (Thermo Scientific). The peptide sample was resuspended in 10 l of buffer A (95% water, 5% acetonitrile, 0.1% formic acid). LC-MS/MS analysis was performed on an LTQ-Orbitrap Discovery mass spectrometer (Thermo Fisher) coupled to an Agilent 1200 series HPLC system. The LysC digest was pressure-loaded onto a 100-m fused silica column (100 m fused silica with a 5 m tip) packed with 10 cm of Aqua C 18 reverse phase resin (Phenomenex). The peptides were then eluted from the column using a 120-min gradient of 5-100% buffer B in buffer A (buffer A: 95% water, 5% acetonitrile, 0.1% formic acid; buffer B: 20% water, 80% acetonitrile, 0.1% formic acid) and injected into the mass spectrometer. The flow rate through the column was set to ϳ0.25 l/min, and the spray voltage was set to 2.75 kV. One full MS scan (M r 400 -1,800) was followed by seven data-dependent scans of the nth most intense ions with dynamic exclusion enabled. Tandem MS data were searched using the SEQUEST algorithm (22) using a concatenated target/decoy variant of the human International Protein Index database. A differential modification of ϩ185.0874 on cysteine was specified to account for MTSL modification. SEQUEST output files were filtered with DTASelect 2.0 (23). Discriminant analyses were performed to achieve a peptide false-positive rate below 5%.
High Resolution Field Cycling 31 P NMR-31 P field-cycling spin-lattice relaxation rate (R 1 ) measurements (18 -21) were made at 20°C on a Varian Unity plus 500 spectrometer using a standard 10-mm Varian probe in a custom-built device (24) that moves the sample, from the conventional sample probe location (at 11.7 T), to a higher position within, or just above, the magnet, where the magnetic field can be as low as 0.04 T; lower fields can be programmed to as low as 0.002 T with the use of a permanent magnet mounted at the top of the Dewar. Concentrations of the phosphorus-containing ligands were usually 3 mM for phosphatidylinositols and 5 mM for diC 7 PC; the protein concentration was usually 0.1 mg/ml (2.1 M) for diC 8 PI samples and 0.25 mg/ml (5.3 M) for other ligands. The sample buffer was 100 mM Tris, pH 8.0, with 2 mM EDTA. Controls used the same concentration of phospholipids but with 0.1 or 0.25 mg/ml unlabeled PTEN. The change in spin lattice relaxation rate due to the spin-labeled protein is ⌬R 1 ϭ R 1, spin label Ϫ R 1, control . The data of R 1 over a wide magnetic field range were fit to Equation 1, where R P-e (0) is the low field limit of ⌬R 1 ; p is the angular frequency of 31 P, and P-e is the correlation time of the dipolar relaxation of 31 P by nearby nitroxide unpaired electrons. The "c" term represents a limiting chemical shift anisotropy contribution due to the paramagnetic interaction and is not used in this analysis. The two parameters extracted are R P-e (0), related to proximity to the spin label, and P-e , the correlation time for the protein/lipid interaction. As discussed previously (18,19), if the protein contains a single spin label and the ligand binds to a discrete site on the protein, the average distance r PϪe between a bound phospholipid phosphorus and the unpaired electron of the spin-labeled proteins can be extracted from both R P-e (0) and P-e along with the concentration of the protein and phospholipid, [PTEN-SL] and [PL] as shown in Equation 2, When more than one cysteine bears a spin label, as is the case for PTEN, the relaxation will be the sum of paramagnetic relaxation enhancement from the different label positions, and the distance calculated, r app , reflects contributions from multiple spin labels. However, because R P-e (0) depends on r P-e Ϫ6 , only very close spin labels will have a strong effect on the 31 P relaxation of the ligand (20,21). Because r app 6 is what is extracted, we will primarily use this parameter for comparisons with different samples.
For each combination of phospholipids reported, the field cycling experiments were carried out on two separately prepared samples (using a different batch of PTEN and often slightly different concentrations of protein or phospholipid). However, each plot of R 1 versus magnetic field shown is for a single sample. Likewise, the entries in Table 3 are for a specific sample. In comparing two samples, the extracted r 6 ϫ 10 54 values differed at most by 30%, leading to an error in r app Ϯ 4%.
Modeling the N-terminal Region of PTEN and Locating the PI(4,5)P 2 -binding Site-The crystal structure is missing the 13-amino acid N-terminal portion of PTEN. Combined analysis of the sequence with the secondary structure prediction produced by Jpred (25) as well as SABLE programs suggests strongly that the initial 14-residue fragment, MTAIIKEIVS-RNKR, is organized as an amphipathic helix. Lack of extensive contacts with the body of the phosphatase domain would create a highly mobile element suitable for interfacial membrane binding. The highly cationic character of this predicted helix would be expected to enhance membrane binding of the PTEN phosphatase domain, although it is possible that such a helix is not stable until the protein is proximal to the membrane surface. With this as a starting point, we used the program COOT (26) to create an idealized helix that had several hydrophobic residues on one side of the helix and several charged residues on the opposite site. The connectivity requirement (residue Arg-14 is observed in the crystal structure), shape, and charge compatibility of this helix with the rest of the phosphatase domain allows for a unique docking of the helix with the hydrophobic face directed toward the membrane (Fig. 1B). Such a model, although speculative, leads to the partial electrostatic neutralization of the acidic residues in the loop adjacent to the proposed N-terminal helix and also extends Lys-13 toward the active site of the enzyme. The new surface of the phosphatase domain created by docking the missing N-terminal peptide as an amphipathic helix contains two distinct indentations that are potential candidates for the binding sites of headgroups of phosphorylated inositol species. Structures with diC 6 PI-(3,4,5)P 3 and diC 6 PI(4,5)P 2 molecules docked to these sites in the PTEN model were evaluated in PyMOL (27).

RESULTS
PI(4,5)P 2 Effects on PTEN Enzymatic Activity-Monomer and micellar ligands are amenable to NMR analyses and can be used to provide information of how and where a PI(4,5)P 2 molecule will bind on PTEN with respect to a substrate analog in the active site. However, for the biophysical results to be convincing, kinetic characterization of PTEN with these short-chain phosphoinositides is needed.
Triton X-100 is a nonionic detergent that we have previously shown to be able to reduce the inhibition potency of shortchain deoxy-PI derivatives by binding to a hydrophobic site adjacent to the active site (AHS) and/or forming micelles that alter the on/off rates of the PI derivative in the active site (13). The presence of the nonionic detergent in the diC 8 PI(3)P/ diC 8 PI(4,5)P 2 assay system had little effect on the PIP 2 activation of PTEN (Fig. 3C). However, the lack of activation of PTEN toward DOPI(3,4)P 2 by DOPI(4,5)P 2 in Triton X-100 mixed micelles (as well as the enhanced inhibition in diC 8 PI(3)P/ diC 8 PI(4,5)P 2 assays) could suggest that occupation of the AHS can alter both substrate and activator binding.
PC is often used as the matrix lipid in preparing vesicles. However, in monomer/micelle assay systems, short-chain PC is an inhibitor of PTEN (Fig. 3D). Comparable concentrations of Triton X-100 have little effect on PTEN enzymatic activity (13). For both diC 8 PI(3)P or diC 8 PI(3,4)P 2 as the substrate, the addition of diC 7 PC below its CMC of 1.5 mM significantly reduced the enzyme activity, although somewhat less so for the better substrate. Adding the nonionic detergent Triton X-100 into the mixture reduced the extent of the diC 7 PC inhibition of PTEN.
These results indicate that a short-chain PC molecule may interact with the protein AHS and in so doing alter access to the nearby active site (13). Triton X-100, in contrast, appears to bind almost exclusively to the AHS with little effect on substrate access to the active site, although it can preclude strong activation by PI(4,5)P 2 .
Lys-13 has been suggested as a key residue for PI(4,5)P 2 activation in PTEN (10,11). Conversion of this cationic side chain to an anionic one should alter electrostatic interactions of protein and ligand. The K13E enzyme was less active than wild type PTEN toward diC 8 PI(3)P or diC 8 PI(3,4)P monomers (less than 10% of wild type PTEN activity) and toward DOPI(3,4)P 2 in PC vesicles (25% of the activity of wild type PTEN). K13E could not be activated either by diC 8 PI(4,5)P 2 or long-chain DOPI(4,5)P 2 (Fig. 4). The ability of PI(4,5)P 2 to activate PTEN has been lost.
Arg-47 was previously shown to be a key residue of the AHS (13). Substituting Arg with Gly, Trp, Leu, or Lys reduced PTEN activity in all assay systems examined (Table 1), although the effect was most pronounced for monomer and micelle systems. The relative activity compared with recombinant PTEN was only decreased 3-4-fold for LUV assay systems, likely because vesicle binding is weak even with wild type enzyme (11). The loss of activity is consistent with this key residue of the AHS affecting the interaction of substrate with the active site. The much higher relative activity for R47K with diC 8 PI(3,4)P 2 as substrate compared with diC 8 PI(3)P might suggest that it is substrate binding that is affected with this mutation. Keeping  this residue cationic is also preferable to placing a small and flexible residue at that position (R47G). Mutation of Arg-47 also alters activation by PI(4,5)P 2 . For R47G, there was virtually no activation by PI(4,5)P 2 (Fig. 4). When the Arg was replaced by a Lys, the relative increase in activity provided by PI(4,5)P 2 was comparable with wild type protein in the monomer assay systems but was significantly lower with vesicles. This kinetic result could suggest that the PI(4,5)P 2 activator site is near the AHS, and the positive charge on Arg-47 is important for PI(4,5)P 2 binding and subsequent activation of PTEN.
All these kinetics are consistent with three distinct but perhaps partially overlapping lipid-binding sites as follows: the catalytic site, the PIP 2 -activating site (likely involving Lys-13 and Arg-47), and the adjacent hydrophobic site (which also appears to involve Arg-47 (13)). However, are these "sites" truly discrete?
Fixed Field 31 P NMR Study of Phospholipids Binding to Recombinant PTEN and Variants-Because the different phospholipids used here have multiple phosphorus atoms and also different chemical shifts, 31 P NMR would seem an obvious choice to explore binding of the different short-chain phospholipids to PTEN. Initially, changes in the 31 P linewidths (⌬⌬1 ⁄ 2 ) of ligands induced by the presence of the protein were used for qualitative characterization of lipid binding to PTEN at 14.02 T. Concentrations of the phosphoinositol compounds were below or close to their CMC in the absence of protein, although the protein was 10 M. The dihexanoylphospholipids exhibited minimal changes in linewidth (Ͻ2.5 Hz). In contrast, the dioctanoylphosphoinositides showed significantly larger linewidths in the presence of PTEN. The D-diC 8 PI and L-3,5-dideoxy-diC 8 PI, which bind to the active site (13,14), exhibited a 25-30 Hz increase in linewidth under the experimental conditions used. diC 7 PC also exhibited a large increase in linewidth (28 Ϯ 4 Hz for 1 mM of the phospholipid), consistent with occupation of the AHS and part of the active site as suggested by the kinetics. In contrast to the active site ligands, diC 8 PI(4,5)P 2 showed much smaller line broadening in the presence of PTEN (ϳ5 Hz increase with 0.25 mM lipid).
Interpreting these linewidth changes is difficult because 31 P at this field strength has contributions from chemical shift anisotropy as well as dipolar interactions that are also coupled with exchange of molecules into a protein-binding site, and perhaps formation of small micelle aggregates with protein. When Triton X-100 was added to the sample with 0.5 mM diC 8 PI, the change in linewidth due to PTEN was significantly reduced from 25 Ϯ 1 to 12 Ϯ 3 Hz. Because that nonionic detergent is not an inhibitor (14), this suggests that either the diC 8 PI-PTEN complex may be large or perhaps that the exchange of ligand on and off the protein is in an intermediate regime and that Triton X-100, by virtue of providing a micelle reservoir, may accelerate such exchange. Adding Triton X-100 to diC 7 PC samples with PTEN reduced line broadening from 30 to 3 Hz. The larger decrease in linewidth (compared with what is observed for Triton X-100 added to diC 8 PI with PTEN) could be explained by competitive binding between diC 7 PC and Triton X-100 in the AHS.
PTEN was also studied with both diC 8 PI(4,5)P 2 and L-3,5dideoxy-diC 8 PI in the same sample. The phosphodiester peaks of these two lipids have slightly different chemical shifts, and therefore linewidth changes that can be measured separately. A large increase in linewidth was detected for the 0.25 mM deoxy-PI bound in the active site (31 Ϯ 1 Hz) with 5.7 Ϯ 0.2 Hz for 0.25 mM diC 8 PI(4,5)P 2 . The values of ⌬⌬1 ⁄ 2 are comparable with what is observed for either separately.
The ⌬⌬1 ⁄ 2 of 31 P resonances in the presence of K13E and Arg-47 mutants was also measured. diC 8 PI in the presence of these three mutant enzymes still showed a large increase in linewidth (30 -40 Hz) with a much smaller linewidth increase (6 Hz) for diC 8 PI(4,5)P 2 . diC 7 PC was also broadened in a similar fashion to diC 8 PI for the R47G or R47K mutants (increases of 42 Ϯ 7, 37 Ϯ 4, and 33 Ϯ 5 Hz, for K13E, R47K and R47G, respectively), and when Triton X-100 (2 mM) was added, the increase in diC 8 PI linewidth caused by PTEN was reduced (to 6.3 Ϯ 0.5, 11.2 Ϯ 3.1, and 8.8 Ϯ 0.9 Hz for K13E, R47K, and R47G), as observed for wild type PTEN (reduction to 11.5 Ϯ 2.8 Hz). diC 8 PI(4,5)P 2 (0.25 mM) linewidths in the presence of K13E, R47K, or R47G were also comparable with the effect of wild type PTEN with linewidth increases of 6.1 Ϯ 0.4, 6.9 Ϯ 0.7, and 6.6 Ϯ 0.9 Hz for K13E, R47K, and R47G, again suggesting this lipid still binds to the protein. At least for the short-chain phospholipid assay system, the loss of activity and PI(4,5)P 2 activation for these mutant enzymes does not appear to be caused by a lack of PI(4,5)P 2 interacting with PTEN. The difficulty is to disentangle somewhat nonspecific interactions of the activating phospholipid with occupation of a discrete site on PTEN that is different from the active site.
Although these fixed field 31 P linewidths provide a fingerprint for phospholipid molecules in the active site (large linewidth change with PTEN added), adjacent hydrophobic site (large change in linewidth that is reduced to less than a few Hz by the addition of Triton X-100), and activator site (small linewidth change unaffected by adding molecules that occupy the active site), there is no information on where such sites might be and whether they are indeed spatially distinct.
Spin-labeled PTEN and Measuring the PR 1 E by Field Cycling-Modifying a protein to introduce a specific spin label, followed by NMR experiments, is useful to estimate proximity of ligands to discrete sites. The large electron dipole can dominate dipolar relaxation of nearby nuclei (in this case 31 P of different amphiphiles) leading to an increase in the relaxation rate (paramagnetic relaxation enhancement (PRE)). A large number of reports of line-broadening effects of this kind have been reviewed recently (28). We use paramagnetic effects on NMR kinetic rates between populations in different spin levels (denoted as R 1 ). Effects of spin labels on 31 P nuclear spins at high field are almost negligible. This hybrid method, named "PR 1 E," is particularly useful in characterizing the location of specific phospholipid-binding sites on proteins (19,20).
Both the more conventional PRE and the PR 1 E are proportional to the inverse sixth power of the distance, but the 31 P R 1 effects that are seen at very low fields (by high resolution field cycling) are completely clear of interference by relaxation from chemical shift anisotropy. Therefore, much lower concentrations of spin-labeled PTEN are required. The method requires that the 31 P species be in moderately fast exchange between free-and enzyme-bound species. The inverse sixth power behavior provides a reasonable degree of specificity as well as a molecular yardstick. Fig. 5 shows how R P-e (0) will vary with distance of the 31 P from the spin label, r P-e , at two different ratios of protein to phospholipid. Also shown is how the overall micelle-protein complex correlation time describing that interaction, P-e , affects R P-e (0). The longer the correlation time, the larger the parameter R P-e (0). The r Ϫ6 dependence of the dipolar interaction means that spin labels more than 15 Å away from discrete phospholipid-binding sites will not contribute significantly to R 1 at the ratios of protein to amphiphile shown in Fig.  4 (these are the same ratios used for PTEN and different ligands). Furthermore, the response can be tuned by adjusting the amount of protein added to a fixed amount of amphiphile. Although two binding sites may be close, for example one 10 Å and the other 15 Å away from the spin label, the PR 1 E will be much larger for the 31 P of the amphiphile in the closer site.
PTEN might not seem an easy target for the PR 1 E method because it has 10 cysteines, 5 on the phosphatase domain, and 5 on the C2 domain. Under the spin-labeling conditions (which used loss of enzyme activity to ensure modification of Cys-124), 9 of the 10 cysteines were modified by the spin-label reagent as monitored by MS, although not all completely ( Table 2). The distribution of modified peptides also agreed well with the overall cysteine labeling determined with Ellman's reagent. Previous work (29) has shown that membrane binding is uncou-pled from enzyme activity, so the use of an inactive enzyme should still provide information on ligand binding.
The N-terminal peptide of PTEN has been suggested to contribute to PI(4,5)P 2 binding, which localizes it near the phosphatase domain. In the PTEN crystal structure, the distances of the C2 domain cysteine sulfur atoms to tartrate carbons are Ͼ20 Å. Therefore, spin labels at those sites should not contribute much to the relaxation of 31 P nuclei of the bound PI phospholipids. Spin labels attached to Cys-83 and Cys-105, located in the phosphatase domain, are far from the suggested membrane binding interface (12) and protein active site and should be only minor contributors to 31 P dipolar relaxation. The distances of the Cys-83 and Cys-105 sulfur atoms to C(1)/C(4) of tartrate are 23/20 and 18/19 Å, respectively. Thus, the major contributors to relaxation of the phospholipids that dock into the active site should be spin labels on Cys-71 and Cys-124. The distances of the Cys-71 and Cys-124 sulfur atoms to C(1)/C(4) of tartrate are 10/6.7 and 6.0/3.6 Å, respectively. If the site for PI(4,5)P 2 is near the active site, it is likely to be relaxed by these spin labels; if it is further away (but still around the face that is thought to interact with the membrane interface) then relaxation by spin-labeled PTEN will be smaller than for the PI bound in the active site.
The NMR-detectable effect of spin-labeled PTEN on different short-chain phosphoinositides was first explored at 242.7 MHz to ensure that ligands still bound in the same manner. Although the contribution of dipolar relaxation to relaxation of 31 P at high fields is small (30), the large dipole of the unpaired electron on the spin label should enhance transverse relaxation of the 31 P nuclei in a distance-dependent manner. At a ratio of diC 8 PI to spin-labeled PTEN of 25:1, there was an additional 7-10 Hz increase in the linewidth over the increase produced by unlabeled PTEN; for diC 8 PI(4,5)P 2 , the linewidth change due to the spin labels on the protein was not significant under these conditions. The lack of an additional increase in observed linewidth upon addition of the spin-labeled PTEN suggests the following: (i) it is not in the active site, and (ii) it is further away from spin labels attached to the protein. Therefore, although the protein is heavily spin-labeled, it has not lost the capacity to bind either active site ligands or diC 8 PI(4,5)P 2 .

TABLE 2 MS identification of PTEN cysteine modified by the 2,2,5,5-tetramethyl-l-oxyl-3-methyl methane-thiosulfonate spin label
The peptides in the tryptic digest containing spin-labeled Cys (C*) and unmodified Cys (C) are indicated. The number of the peptides detected in the sample is also shown.

Peptide
Cys no.  6 PI has a CMC around 10 -14 mM (31); adding phosphates to the inositol ring will increase the CMC so that at the concentrations used in the field cycling experiment (3 mM), the lipids are monomeric in the absence of protein. In the field cycling NMR experiments, the ratio of phospholipid to PTEN was always greater than 500:1, providing an excess of phospholipid so that observed high field phospholipid linewidths were not dramatically affected. As expected, unlabeled PTEN had no effect on the 31 P relaxation behavior of either compound at low fields at these ratios of protein to phospholipid. The profile for diC 6 PI interacting with PTEN is shown in Fig. 6A with the field dependence of the 31 P PR 1 E (⌬R 1 ) shown in the inset. Clearly, there is a significant enhancement of the diC 6 PI 31 P R 1 by the presence of spin-labeled PTEN at fields below 1 T.
There are several interesting features of the protein-diC 6 PI complex PR 1 E data (Table 3). First, P-e is 172 Ϯ 21 ns, a time significantly larger than expected for rotational diffusion of a single PTEN molecule complexed to a single phospholipid. This indicates that the presence of the diC 6 PI ligand in the active site causes protein/phospholipid aggregates to form, even though by itself the diC 6 PI is monomeric at this concentration. Second, the parameter r app 6 is about 4.0 ϫ 10 Ϫ54 m 6 , suggesting moderate proximity of the lipid to spin labels on the protein. A second sample of spin-labeled PTEN and slightly different PTEN and lipid concentrations led to a value of 4.1 ϫ 10 Ϫ54 m 6 for r app 6 , indicating the reproducibility of the experiments. If a single spin label were responsible for this PR 1 E, r P-e would be 12.5 Å. However, Cys-71 and Cys-124, both fairly well labeled, should be close enough to generate a larger PR 1 E for diC 6 PI. The modest effect on relaxation strongly suggests that the active site of the PTEN is not saturated by 3 mM diC 6 PI.
For diC 6 PI(4,5)P 2 , the magnitude of the relaxation enhancement from the spin-labeled PTEN, R P-e (0), was considerably smaller than for diC 6 PI ( Fig. 6B and Table 3). However, P-e was also noticeably shorter, 35 versus 170 ns for diC 6 PI. It appears from our data that binding of diC 6 PI causes large protein/lipid aggregates to form, whereas the phosphorylated activator molecule has a much smaller effect.
I(1,4,5)P 3 is a soluble molecule that mimics the PI(4,5)P 2 headgroup and might be expected to bind into the active site or perhaps a discrete PI(4,5)P 2 site on the protein. However, there was no low field rise in R 1 indicative of that small molecule (3 mM) binding to the spin-labeled protein (4.9 M) and similar to what was seen with diC 6 PI(4,5)P 2 (Fig. 6B, inset). This indicates that inositol 1,4,5-trisphosphate (I(1,4,5)P 3 ) binds poorly to the spin-labeled protein, consistent with the observation that IP 3 had virtually no inhibitory or activating effect on enzymatic activity. It appears that acyl or alkyl chains are needed for the ligands to bind well to PTEN (at least at low millimolar concentrations of the inositol compound used here).
PTEN-induced PR 1 Es, Micellar Phosphoinositides-The effect of spin-labeled PTEN on micellar diC 8 PI and diC 8 PI(4,5)P 2 was much larger than for the dihexanoyl chain compounds. The field cycling profile for diC 8 PI mixed with unlabeled PTEN (Fig.  7A, open circles) showed a low field component significantly greater than for diC 8 PI micelles alone (14). This indicated that PTEN interacts with the longer chain PI and forms larger micelles than the diC 8 PI by itself. Analysis of this curve yielded two distinct dispersions with different correlations times as follows: 5 ns (postulated to represent fast motions, e.g. phospholipid rotation or wobbling (30,32)) and 190 ns (representing diffusive rotation of the entire protein/phospholipid aggregate (33)). The field dependence of the PR 1 E, obtained by subtracting R 1 for diC 8 PI mixed with unlabeled protein from the rates obtained with spin-labeled protein, is shown in Fig. 7B. These data were also not well fit with a single correlation time. To extract parameters for the low field dispersion, a 200 Ϯ 50 ns correlation time, comparable with the value seen for the same phospholipid interacting with the unlabeled PTEN sample, was used and a value of 5 ns for the faster dispersion (comparable with the value seen for unlabeled protein mixed with the micelles). As shown in Table 3, the value of r app 6 was 0.52 ϫ 10 Ϫ54 m 6 , a value significantly smaller than for diC 6 PI. Analyzing the diC 8 PI dependence of ⌬R 1 on field with a single dispersion yielded r app 6 ϭ 0.49 ϫ 10 Ϫ54 m 6 , indicating that the r app 6 value is fairly robust with large ⌬R 1 at low fields. Clearly, the 31 P atom of diC 8 PI bound to PTEN is close to the spin labels. The dispersion with a 5-ns correlation time can also yield information about amphiphile binding. The P-e and R P-e (0) extracted generated a value of r app 6 as 0.06 ϫ 10 Ϫ54 m 6 . On this short time scale, there are many interactions of the spin labels on the protein with PI molecules in the micelle; the bulk of these reflect nonspecific interactions (i.e. not occupation of a discrete binding site) because they do not persist for the length of the correlation time of the entire complex (ϳ200 ns). By comparison, the large PR 1 E for diC 8 PI at the lower fields means a phospholipid must occupy a discrete binding site on the protein for at least 200 ns. The effect of spin-labeled PTEN on 3 mM of the activator molecule diC 8 PI(4,5)P 2 was also examined (Fig. 7C). For these micelles, more spin-labeled protein was needed to get effective relaxation of the 31 P nuclei. Unlike what was observed for diC 8 PI, there was no measurable effect of unlabeled PTEN on the diC 8 PI(4,5)P 2 relaxation rates at fields below 1 T. This is expected because small micelles (14) tumble more rapidly making P shorter so that the whole relaxation dispersion is pulled out to higher fields. The excess relaxation caused by the spinlabeled protein was fit with a single exponential. The average correlation time from the three 31 P resonances for the phospholipid-protein complex, P-e , was 119 Ϯ 17 ns. The values of r app 6 for the three phosphorus resonances follow the same pattern seen for diC 6 PI(4,5)P 2 (Table 3); r app 6 is smallest for P-1 with P-4 and P-5 significantly larger. A key point is that P-1 has the largest PR 1 E, meaning it is closest to the spin labels on the protein.
The effect of the spin-labeled PTEN is much easier to detect and quantify for diC 8 PI(4,5)P 2 because the micelles interacting with the protein form a larger complex with a longer P-e . The higher value of r app 6 for the phosphodiester of diC 8 PI(4,5)P 2 compared with diC 8 PI suggests it is occupying a different site on the protein and is further away from spin-label sites that efficiently relax diC 8 PI in the active site.
The similarity between the observed rates (values of R P-e (0) in Table 2), particularly for the two phosphomonoester 31 P spins, suggested to us the possibility of diffusion of spin magnetization (usually called "spin diffusion" in the NMR literature) that could complicate the analysis. The rate for this transfer between two 31 P spins separated, for example, by 2.5 Å, with tumbling correlation times of 100 ns, is about 40 s Ϫ1 (34). This is fast compared with the observed R P-e (0) rates of 1.5 s Ϫ1 , at most, for the phosphodiester 31 P of diC 8 PI(4,5)P 2 samples.
In this case, the distances from the 31 P nuclei and the spin labels for the monoesters are equal within 10%, and the two 31 P nuclei would be expected to relax at nearly the same rate with or without spin diffusion, so therefore it can be ignored. In cases where the distance from the spin label to the nearest 31 P is less than 80% as large as the distance from the spin label to the most distant 31 P, spin diffusion could distort the rates and confuse the model building process. However, this is not the case here for the phosphodiester. Therefore, the distances we extract should not be severely distorted by spin diffusion.
Enzyme kinetics with short-chain phospholipid monomers below the CMC strongly showed that diC 8 PI(4,5)P 2 at moderate concentrations can act as an inhibitor. If the inhibition arises from that PTEN product binding to the active site, then some of the relaxation of PI(4,5)P 2 31 P resonances could be caused by its binding into the active site. To assess this, we examined a sample of diC 8 PI, diC 8 PI(4,5)P 2 , and spin-labeled PTEN. If the PI(4,5)P 2 relaxation is primarily caused by binding to the active site, then the presence of the diC 8 PI should reduce R 1 substantially. We know from measuring linewidths at high field that in mixtures of the two phosphoinositides the linewidth increases were comparable with what was observed for either alone. In the field cycling experiment, the diC 8 PI reso- The data for diC 8 PI required fitting with two dispersions: for the slower one, the value of P-e is 200 ns, and the value from the low field dispersion measured for the system with unlabeled PI-PLC was used for determining R P-e (0). Error in was estimated as Ϯ50 ns. c PI concentration was 3 mM. d The dispersion for diC 7 PC was fit with a single P-e . e The sample contained 5 mM diC 7 PC and 3 mM diC 8 PI(4,5)P 2 ; overlap of both phosphodiester resonances and similar relaxation profiles makes ascribing r app 6 to a particular phospholipid impossible. nance relaxed too quickly at low fields to obtain P-e and R P-e (0). However, the field dependence profiles of the two phosphomonoesters with this enzyme preparation were basically the same in the absence and presence of diC 8 PI, indicating that PI(4,5)P 2 is binding to a site spatially distinct from the active site but near the labeled cysteine residues of PTEN.
Localizing the diC 8 PI(4,5)P 2 Site with C124S-Spin labels on Cys-71 and Cys-124 are the major contributors to relaxing diC 8 PI. However, they or other spin labels may contribute to the PR 1 E for PI(4,5)P 2 . One approach to obtaining a value of r P-e from a particular spin label is to remove one of the active site spin labels. To this end, we spin-labeled the variant C124S. When modified, this variant will still have a nitroxide on Cys-71 near the active site.
For diC 8 PI binding to spin-labeled C124S, the profile shows a 50% decrease in the PR 1 E (Fig. 8). The diC 8 PI profiles have two components, and the low field component was used for com-paring effects of spin-labeled wild type and C124S proteins. That difference in R P-e (0), 1.7 s Ϫ1 (adjusting for the slightly higher concentration of C124S), represents the contribution of spin-labeled Cys-124 to the relaxation of the diC 8 PI. At the concentrations of PTEN (ϳ0.71 M) and diC 8 PI (3 mM) used, the value for the active site-bound diC 8 PI phosphodiester 31 P distance to the spin label attached to Cys-124 is 10.0 Å.
The R P-e (0) values for the three phosphorus nuclei in diC 8 PI(4,5)P 2 binding to spin-labeled C124S are also reduced compared with spin-labeled wild type protein. The ⌬R P-e (0) contributions from a spin label on Cys-124 (subtracting the value obtained for C124S from the wild type PTEN value of R P-e (0) adjusted to the same protein concentration) were 0.39, 0.24, and 0.35 s Ϫ1 for P-1, P-4, and P-5. From the fitted correlation time of 160 ns and these R P-e (0) values, the distances between each 31 P and the unpaired electron of the nitroxide on Cys-124 are estimated as 13.8, 15.0, and 14.1 Å for P-1, P-4, and P-5 of diC 8 PI(4,5)P 2 . For the phosphomonoesters, R P-e (0) is fairly small, and differences in P-e (which is less well defined) could also add some uncertainty in the ⌬R P-e (0) value needed to estimate r P-e for PI(4,5)P 2 with respect to the spin label on Cys-124. However, the ⌬R P-e (0) values are unlikely to be off by as much as a factor of 2, which would translate to Յ10% error in r P-e . Thus, the r P-e values extracted for diC 8 PI(4,5)P 2 indicate that the activator site is close to the active site.
Occupation of the AHS-Our kinetics with the synthetic short-chain PIs showed that diC 7 PC is inhibitory, whereas Triton X-100 has little effect on the observed hydrolysis rate by  itself but is able to counteract the inhibition by diC 7 PC. With this in mind, we explored the effect of spin-labeled PTEN on zwitterionic diC 7 PC micelles in the absence and presence of Triton X-100 and diC 8 PI(4,5)P 2 . For 5 mM diC 7 PC micelles interacting with 5.3 M spin-labeled PTEN, there was a large effect of the protein on the 31 P resonance; r app 6 was 0.40 ϫ 10 Ϫ54 m 6 for this sample (Table 3), a value comparable with the value for diC 8 PI and consistent with the diC 7 PC phosphocholine moiety binding in the PTEN active site, as suggested by the kinetics. diC 7 PC also interacts with the AHS because inclusion of Triton X-100 in the micelles dramatically altered its relaxation by the spin-labeled protein. When 5 mM diC 7 PC was mixed with 10 mM Triton X-100, the effectiveness of the spin-labeled PTEN in relaxing the diC 7 PC was reduced about 2-fold ( Table 3). The kinetics, carried out with substrate concentration below its CMC, suggest that a 4-fold excess of Triton X-100 rescues about half the activity lost by diC 7 PC binding in/near the active site, so not all of the PC is likely to be removed from near the active site. The PR 1 E reduction by Triton X-100 is consistent with competition between the nonionic detergent alkyl chain and one or both chains of diC 7 PC binding to the AHS.
For comparison, the effect of Triton X-100 on diC 8 PI binding to spin-labeled PTEN was also examined. Protein binding to mixed micelles of diC 8 PI/Triton X-100 (3:6 mM) exhibited a smaller average size as indicated by a shorter correlation time for diC 8 PI in the presence of unlabeled protein. Once the control was subtracted, the dependence of the diC 8 PI PR 1 E on field was best fit with a correlation time of 68 Ϯ 10 ns; r app 6 was 0.52 ϫ 10 Ϫ54 m 6 ( Table 3), comparable with what is observed with diC 8 PI alone. Thus, the polar headgroup interactions of the PI are more than adequate to compensate for disruption by Triton X-100 of any hydrophobic interactions of acyl chain moieties with the AHS.
If diC 7 PC partially occupies the active site and the AHS, does it have an effect on the PI(4,5)P 2 interaction with PTEN? To answer this, we obtained a field cycling profile for 3 mM diC 8 PI(4,5)P 2 , 5 mM diC 7 PC, and 5.3 M spin-labeled protein.
Although the PC phosphodiester overlapped with the PI(4,5)P 2 phosphodiester resonance, the behavior of the phosphomonoesters in the absence or presence of the micellar PC can act as sensors of the added PC. The two PI(4,5)P 2 phosphomonoester peaks exhibited little change in P-e or R P-e (0)with the excess of diC 7 PC added, again consistent with a discrete PI(4,5)P 2 site on PTEN different from the active site (Table 3).
Modeling PIP 2 Binding to PTEN-The field cycling NMR studies provide distances for placement of the activating PI(4,5)P 2 molecule with respect to spin-labeled Cys-124. These distances, the mutagenesis data implicating Lys-13 and Arg-47 in the PI(4,5)P 2 activation process, and the model for PTEN bound to a substrate-containing membrane (Fig. 1B) were used to identify a discrete site on the protein that is close to but does not compete with substrate occupying the active. As shown in Fig. 9A, diC 6 PI(4,5)P 2 can be docked so that its 31 P nuclei are 10.2, 12.8, and 10.4 Å, respectively, from the spin-label on Cys-124. Lys-13 and Arg-47 are critical for the orientation with P-4. These distances are all shorter than the values estimated by the NMR experiment, although the model tracks which 31 P is clos-est to the spin label (P-1, with P-5 only slightly different) and which is decidedly further away (P-4). Movement of the spin label and faster dynamics of the short-chain molecules could easily reduce the effectiveness of the PR 1 E, but trends would be preserved.
The model also provides an explanation for the difference in kinetic versus NMR effects of what happens with higher concentrations of PI(4,5)P 2 and PI(3)P. Enzyme kinetics show that increasing PI(4,5)P 2 can inhibit PTEN acting on PI(3)P as the substrate. However, in the NMR both PI(4,5)P 2 and PI (the surrogate substrate) clearly bind to PTEN at discrete sites at the same time. The spin label in the active site would make it difficult for PI(3)P to bind well in the active site cleft. However, PI, lacking that phosphorus, can still fit in the cleft.
Both substrate and activator phospholipids are shown docked to the phosphatase domain (without the spin label) in Fig. 9B. Here, the surface electrostatic potential of PTEN is FIGURE 9. Ribbon diagrams of the model of diC 6 PI(4,5)P 2 binding to PTEN at a discrete site near the active site are compatible with the 31 P electron distances to the Cys-124 spin label. A, view of the phosphatase domain as seen from the membrane. The sulfur of Cys-124 is the yellow sphere to which the 2,2,5,5-tetramethyl-1-oxyl-3-methyl methanethiosulfonate is attached. The PI(4,5)P 2 molecule, placed according to the surface features and to the relative distances determined by NMR, forms hydrogen-bonded contacts with Arg-47 and Lys-13, both critical residues for the PIP 2 activation to be observed. B, surface representation of the phosphatase domain, in a similar view to that in A, colored by the electrostatic potential (red, negative, and blue, positive) with both substrate and activator docked to their sites. The PI(3,4,5)P 3 molecule (magenta) is docked at the active site marked by the dark loop and the yellow sphere depicting Cys-124. The PI(4,5)P 2 molecule (in cyan) occupies the same site as in A. The N-terminal helix is docked into the phosphatase domain, and its positive electrostatic potential complements the charges of the domain to create a positive membrane binding profile.
shown. Given the choice of PI(3,4,5)P 3 and PI(4,5)P 2 , the higher electropositive character of the active site will preferentially attract PI(3,4,5)P 3 . Modeling the missing N-terminal peptide as a helix provides a ligand for P-4 of PI(4,5)P 2 but also creates a surface with a higher cationic character to enhance membrane binding. The two phospholipids are in close van der Waals contact with each other as well as the active site loop. Thus, occupation of the PI(4,5)P 2 site will enhance anchoring of PTEN to membranes allowing substrate and product to enter and exit the active site cleft.
Perturbation of the Activator Site in K13E-The model presented for PI(4,5)P 2 binding to PTEN shows that Lys-13 as well as Arg-47 interact with the phosphates of the activating ligand. Our kinetic studies show that the variant K13E is not activated by PI(4,5)P 2 , indicating that the activator site has been altered, i.e. either it has been disrupted or its communication with the active site has been affected. The fixed field studies showed that the increased linewidth for each phospholipid was the same whether wild type or K13E spin-labeled protein was used. However, does PI(4,5)P 2 still bind in the same way? The field cycling experiments provide insights for the mechanism behind this kinetic effect (Fig. 10). diC 8 PI binding to spin-labeled K13E exhibited a similar PR 1 E to that for binding to native PTEN as reflected in R P-e (0). There is a statistically real decrease in r app 6 for diC 8 PI in the mutant that could represent a very small shift in the position of the phosphodiester with respect to the spin labels. Changes in extracted parameters were also observed for diC 8 PI(4,5)P 2 . In particular, the r app 6 for P-5 is now larger than that for P-4, meaning its orientation with respect to the active site spin labels has changed (Table 4). This suggests that the diC 8 PI(4,5)P 2 inositol ring is adopting a different orientation when bound to spin-labeled PTEN K13E. In the model presented for diC 6 PI(4,5)P 2 binding to PTEN (Fig. 9), P-5 is closer to the spin label in the active site than P-4, but it is P-4 that interacts with Lys-13. If Lys-13 is mutated, a major constraint on orienting the activating PI(4,5)P 2 inositol ring has been removed. In the micelle system, the phosphoinositide headgroup can still bind in this site, but it must be reoriented.
Another intriguing difference between the effects of spinlabeled native PTEN and K13E is that the correlation times for the diC 8 PI(4,5)P 2 bound to K13E are shorter than for normal PTEN, particularly with diC 8 PI present (Table 4). With both phospholipids present, the average P-e for the diC 8 PI(4,5)P 2 resonances was 43 Ϯ 8 ns (Table 4) compared with 142 Ϯ 34 ns when the same two phospholipids are mixed with spin-labeled wild type PTEN ( Table 2). The bound diC 8 PI resonance in the same sample with K13E had a P-e twice that of the activator molecule. If one assumes the longer rotational correlation time (for the diC 8 PI) is for the overall protein-micelle complex, then the diC 8 PI(4,5)P 2 molecule has a shorter residence time on the enzyme. These changes in relaxation parameters for both diC 8 PI and diC 8 PI(4,5)P 2 are consistent with a change in the protein conformation of K13E that weakens diC 8 PI(4,5)P 2 binding (and likely alters the orientation of the inositol ring) and slightly alters the placement of diC 8 PI in the active site. This indicates that the orientation of PI(4,5)P 2 when bound to PTEN must be very specific for kinetic activation to occur. Because the two lipids are within van der Waals contact in the model, the loss of activation with K13E is consistent with a requirement of close contact of the two phospholipids for optimal membrane binding and catalysis.

DISCUSSION
What Does PIP 2 Binding Do?-A number of model membrane binding studies (11,16) as well as cell studies (29,35,36) have shown that PI(4,5)P 2 (but phosphatidylserine as well) enhances membrane anchoring of PTEN. Deletion or mutation of cationic residues in the first 15 residues reduces membrane binding of the protein and can relocalize the protein to the cytosol. These results clearly demonstrate that PI(4,5)P 2 is critical for tight PTEN binding to membranes. However, there is no definitive evidence that an individual PI(4,5)P 2 molecule binds to the N-terminal binding motif of PTEN, evidence for a specific conformational change, or a defined position for the N-terminal peptide with respect to the active site. The NMR studies in this work clearly show that there is a very specific site for a PI(4,5)P 2 molecule binding to the protein near the cleft where substrate binds. Molecular modeling with the NMR and mutagenesis constraints allowed us to identify such a site (Fig.  9A).
One of the intriguing results of a recent MD simulation study was that productive membrane binding of PTEN required a rotation of the two domains. Substrate PI(3,4,5)P 3 promoted correct orientation at the membrane surface, and Kalli et al.

TABLE 4 Comparison of low field relaxation parameters for phosphoinositides interacting with spin-labeled PTEN K13E
The ratio of protein/phospholipid was 0.00071 for diC 8 PI alone; for the sample containing 3 mM of both phospholipids, the protein/phospholipid ratio was 0.00141.

Ligand
P-e R P-eP-e (0) r 6 (17) suggested that perhaps PI(4,5)P 2 , at significantly higher concentrations in cells, could have the same effect. However, the simulation used the crystal structure of PTEN with both C-and N-terminal peptides missing. In our model, the N-terminal peptide is docked as an interfacial helix and is critical in creating the PI(4,5)P 2 site (Lys-13 as a ligand for P-4). This portion of the protein could be inherently flexible and require a phosphoinositide to adopt a stable helix oriented for membrane interactions. As a helix, the electrostatics of this segment would contribute to overall tighter membrane binding. Thus, PI(4,5)P 2 binding at the site would lead to better binding of the protein on membranes. A PI(4,5)P 2 molecule adjacent to the cleft occupied by substrate could also enhance processive catalysis through a van der Waals contact that helps substrate or product navigate in and out of the cleft.
PTEN Aggregates with Monomeric Phosphoinositides-The phosphoinositide monomer and micelle systems used in this study provide an easier way to characterize the PI(4,5)P 2 -binding site than working with vesicles. The micelles are dynamic with small linewidths (for micelles, less than 8 Hz at 14.08 T in the absence of protein) compared with phospholipids in vesicles, allowing experiments with low millimolar concentrations of lipids. Fixed high field studies of concentrations of the shortchain phosphoinositides below their CMC provide a marker for molecules binding in the active site, i.e. large linewidth increases occur for molecules known to bind to the active site and inhibit PTEN activity, while significantly smaller increases in linewidth are observed for diC 8 PI(4,5)P 2 . The large linewidth changes for molecules that bind to the active site could reflect aggregation of the PTEN, but they could also be caused by exchange broadening. The use of high resolution field cycling, under conditions where there is a very large excess of phospholipids compared with protein (typically phospholipid/PTEN is 1500:1 for a PI and 500 -600:1 for PIP 2 ), provides unambiguous information on the size of PTEN⅐short-chain phosphoinositide complexes, as well as an averaged distance between the 31 P of interest and unpaired electrons on nitroxides introduced in the protein.
Although diC 6 PI is monomeric in the absence of protein (31), it is in a large complex characterized by a correlation time of ϳ170 ns when it is bound to PTEN. That large rotational correlation time strongly suggests that the complex consists of more than one or two PTEN molecules, along with many diC 6 PI. In contrast, a much smaller P-e (35 ns) was obtained for the PTEN-diC 6 PI(4,5)P 2 complex. The experimental rotational diffusion constant, P-e , is related to an isotropic rotational diffusion constant by P-e ϭ 1/(6 ϫ D) where D ϭ kT/(8 r 3 ), and r is the hydrodynamic radius. From this, we estimate the average radius of the PTEN⅐diC 6 PI(4,5)P 2 complex as 32 Å. For comparison, a monomer of PTEN, assuming it can be approximated by a sphere, would have a radius r ϭ {3M/(4N o )} 1/3 , where M is the molecular weight; is the average protein density, and N o is Avogadro's number. The extracted value is 24 Å for a monomer PTEN binding a single phospholipid at the activator site. This would be increased 1-2 Å with a hydration layer. Again, considering the protein as globular, a PTEN dimer would have a rotational correlation time of 30 -32 Å. Recently, it has been shown that PTEN can form dimers and that this is the active species of the enzyme (37). The experimental P-e for the diC 6 PI(4,5)P 2 ⅐PTEN complex is consistent with diC 6 PI(4,5)P 2 binding to a PTEN dimer in solution.
If PTEN/diC 6 PI(4,5)P 2 is dimeric in solution, why is the complex with diC 6 PI so much larger? The dimeric PTEN structure is suggested to have a more "open" or relaxed conformation of PTEN when it is bound on the membrane (37). This could resemble the model shown in Fig. 1B. A more extensive membrane-binding surface in the dimer would likely attract a significant number of the short-chain phosphoinositides with it and lead to the observed large correlation time for the complex. That a larger complex is observed for the PTEN with diC 6 PI suggests that a molecule binding to the active site is critical for this more extended conformation.
The r app 6 extracted for diC 6 PI compared with that for diC 8 PI provides another useful piece of information. r app 6 is 7-fold larger for diC 6 PI, indicating that the active site is not saturated with the ligand. This is consistent with productive substrate binding requiring a very specific conformational change of PTEN that is enhanced by the presence of an interface.
In contrast, the r app 6 parameters for diC 6 PI(4,5)P 2 were only 1.4 -2.2-fold larger than the values for diC 8 PI(4,5)P 2 and very different from those for diC 8 PI. The interaction of this monomeric phosphoinositide with PTEN also leads to a complex larger than for a single molecule binding to PTEN. However, the similar values for r app 6 for the two different chain length PI(4,5)P 2 molecules are consistent with a discrete and moderately tight site on the protein for the PI(4,5)P 2 . These results also suggest an explanation for the acyl chain dependence of PI(4,5)P 2 activation kinetics (Fig. 3B). diC 6 PI(4,5)P 2 is poor at activating PTEN toward diC 6 PI(3)P because, by analogy to what we see with diC 6 PI versus diC 8 PI binding, less substrate is bound at the active site. diC 8 PI(4,5)P 2 is also not very effective at enhancing hydrolysis of diC 6 PI(3)P for the same reason.
In summary, we have shown that there are three functionally distinct but spatially close sites for amphiphiles to bind to PTEN, at least for ϳ100 ns: (i) the active site, (ii) the PI(4,5)P 2 "allosteric" site, and (iii) the AHS. A discrete binding site for PI(4,5)P 2 involving the N-terminal peptide of PTEN (specifically Lys-13) and Arg-47 has been proposed and is consistent with the NMR data. Its placement adjacent to the active site provides a pathway to stabilize the conformation of PTEN needed for optimal processive catalysis. The interplay/synergism of the three sites detected by field cycling NMR also provides ways to regulate PTEN activity. Like many other signaling proteins that cycle on/off the membrane surface, PTEN possesses a C2 domain that should transiently anchor it to the membranes via PS interactions. However, the presence of a hydrophobic site and PIP 2 activator site in the vicinity of the active site on the phosphatase domain provides novel ways to direct the PTEN conformational change needed for catalysis.