Specificity and Promiscuity in Phosphoinositide Binding by Pleckstrin Homology Domains*

Pleckstrin homology (PH) domains are small protein modules involved in recruitment of signaling molecules to cellular membranes, in some cases by binding specific phosphoinositides. We describe use of a convenient “dot-blot” approach to screen 10 different PH domains for those that recognize particular phosphoinositides. Each PH domain bound phosphoinositides in the assay, but only two (from phospholipase C-δ1and Grp1) showed clear specificity for a single species. Using soluble inositol phosphates, we show that the Grp1 PH domain (originally cloned on the basis of its phosphatidylinositol 3,4,5-trisphosphate (PtdIns(3,4,5)P3) binding) binds specifically tod-myo-inositol 1,3,4,5-tetrakisphosphate (Ins(1,3,4,5)P4) (the PtdIns(3,4,5)P3headgroup) with K D = 27.3 nm, but bindsd-myo-inositol 1,3,4-trisphosphate (Ins(1,3,4)P3) or d-myo-inositol 1,4,5-trisphosphate (Ins(1,4,5)P3) over 80-fold more weakly. We show that this specificity allows localization of the Grp1 PH domain to the plasma membrane of mammalian cells only when phosphatidylinositol 3-kinase (PI 3-K) is activated. The presence of three adjacent equatorial phosphate groups was critical for inositol phosphate binding by the Grp1 PH domain. By contrast, another PH domain capable of PI 3-K-dependent membrane recruitment (encoded by EST684797) does not distinguish Ins(1,3,4)P3 from Ins(1,3,4,5)P3 (binding both with very high affinity), despite selecting strongly against Ins(1,4,5)P3. The remaining PH domains tested appear significantly less specific for particular phosphoinositides. Together with data presented in the literature, our results suggest that many PH domains bind similarly to multiple phosphoinositides (and in some cases phosphatidylserine), and are likely to be regulated in vivo by the most abundant species to which they bind. Thus, using the same simple approach to study several PH domains simultaneously, our studies suggest that highly specific phosphoinositide binding is a characteristic of relatively few cases.

Pleckstrin homology (PH) 1 domains are small protein modules of approximately 120 amino acids with sequence similarity to two regions in pleckstrin, the major protein kinase C substrate in platelets (1)(2)(3). PH domains are found in more than 100 different proteins involved in intracellular signaling processes, cytoskeletal organization, regulation of intracellular membrane transport, and modification of membrane phospholipids (4 -6). The majority of proteins that contain PH domains require membrane association for their function, and the PH domain appears to play a role in this membrane targeting (7). Although there have been reports of protein binding by PH domains or by adjacent portions of their host proteins (8 -10), a consensus is emerging that most PH domains interact directly with cellular membranes by binding to phosphoinositides (7,(11)(12)(13)(14)(15)(16)(17)(18)(19)(20).
Expression of GST-PH Fusion Proteins-The coding region for each PH domain was amplified by polymerase chain reaction from the relevant cDNA clone, incorporating a BamHI site at the end corresponding to the N terminus and either an EcoRI or BamHI site at the opposite end. Appropriately digested polymerase chain reaction products were then subcloned into pGEX-2TK (Amersham Pharmacia Biotech) using the BamHI site or BamHI/EcoRI sites. The identity of each insert was confirmed by dideoxynucleotide sequencing.
GST-PH fusions were expressed in Escherichia coli BL-21 and purified using glutathione-agarose as described (54). Proteins were labeled with 32 P while bound to glutathione-agarose. For solution studies of inositol phosphate and phosphoinositide binding, PH domains were subcloned between the NdeI and BamHI site of the T7 expression vector pET11a (Novagen), and produced exactly as described previously for PLC␦-PH (14). Concentrations of purified protein were determined using A 280 , for which extinction coefficients were measured by quantitative amino acid analysis of samples with known absorbance. 32 P Labeling of GST-PH Fusions-pGEX-2TK-encoded GST-PH fusions were labeled with 32 P essentially as described (55,56). Approximately 10 g of purified GST-PH fusion (on beads) was incubated with 0.75 mCi of [␥-32 P]ATP (NEN Life Science Products; NEG-035C) and 10 -20 units of protein kinase A (Sigma) for 30 min at room temperature in 50 mM potassium phosphate buffer, pH 7.15, containing 10 mM MgCl 2 , 5 mM NaF, and 4.5 mM DTT, in a reaction volume of 75 l. Beads were washed extensively with phosphate-buffered saline (PBS), containing 1 mM DTT and 1 mM phenylmethylsulfonyl fluoride, and 32 Plabeled GST-PH fusion protein was eluted with 15 mM reduced glutathione in PBS. Eluted 32 P-labeled protein was filtered (0.22 m) prior to use in the dot-blot assay.
Dot-Blot Assay-Phosphoinositides at 2 mg/ml in 1:1 chloroform: methanol (containing 0.1% HCl) were spotted (2 l) onto nitrocellulose (MSI) in the pattern shown in Fig. 1. After drying, nitrocellulose was blocked overnight at 4°C in Tris-buffered saline (TBS) containing 3% bovine serum albumin but no detergent. 32 P-Labeled GST-PH fusion protein was then added to TBS/3% bovine serum albumin at a final concentration of 0.5 g/ml, and this solution was used to probe the phosphoinositide-containing nitrocellulose for 30 min at room temperature. Filters were washed five times with TBS (without detergent), dried, and bound radioactivity was visualized using a PhosphorImager (Molecular Dynamics).
Isothermal Titration Calorimetry (ITC)-ITC experiments employed the MCS ITC instrument (57) from MicroCal (Northampton, MA). PH domain or inositol phosphate was placed in the calorimeter cell (1.39 ml) at the noted concentration, and sequential aliquots (10 -16 l) of the relevant inositol phosphate or protein were injected, under computer control, to cover a range of InsP n :PH molar ratios of 0:1 to 3:1. PH domain was dialyzed exhaustively into 25 mM MOPS, pH 6.8, 100 mM NaCl, 1 mM DTT prior to the experiment, and all inositol phosphate solutions were generated by dilution of a 10 mM stock solution (in water) into this buffer. Control experiments for heats of dilution of inositol phosphate or protein into buffer and for buffer into PH domain or inositol phosphate were performed in separate titrations, and data were corrected accordingly. Data were analyzed and fit using the ORIGIN software (MicroCal) supplied with the instrument, allowing stoichiometry (n), ⌬H, and K B (binding constant) to float in fitting to a single class of sites.
Spin-column Competition Assay-The ability of a series of different inositol phosphates to compete with Ins(1,3,4,5)P 4 for binding to Grp1-PH was assessed using a spin-column binding assay. The specified molar excesses of unlabeled competitor were included in 35-l samples (final volume) that contained a 1:1 Ins(1,3,4,5)P 4 :Grp1-PH complex at 5 M (final concentration), plus 10 nCi (14 nM) of [ 3 H]Ins(1,3,4,5)P 4 (NEN Life Science Products; 21 Ci/mmol). To assess the degree of competition, samples were applied to a 1-ml Bio-Gel-P6 (Bio-Rad) spin-column, and spun at 900 ϫ g to allow Ͼ95% of the protein to pass through. The number of (bound) counts in the flowthrough was determined for each sample by scintillation counting, and this value was plotted as the percentage of maximal binding (in the absence of competitor) against the molar excess of competitor (Fig. 3). The resulting competition curves were fit (with IC 50 as the fitted parameter) in ORIGIN (MicroCal) to an isotope dilution model of the form: % comp ϭ {% 0 ϫ [IC 50 /(C ϩ IC 50 )]} ϩ % ϱ . In this equation, % comp is the number of counts bound in the presence of competitor at a concentration, C (expressed as a molar excess over the Grp1-PH⅐Ins(1,3,4,5)P 4 complex); % 0 is the number of counts bound in the absence of competitor; % ϱ is the number of residual counts remaining at high excess of competitor (unlabeled Ins(1,3,4,5)P 4 ); and IC 50 is the molar excess of competitor required to reduce % comp to half its initial value. IC 50 for unlabeled Ins(1,3,4,5)P 4 was 3-fold in the best fit. Relative affinities of other InsP n tested were then estimated as [(IC 50 InsPn /3) ϫ 0.027 M]. All experiments were performed in 25 mM MOPS, pH 6.8, 100 mM NaCl, 1 mM DTT.
Subcellular Localization Studies-The Grp1 PH domain was subcloned between the BglII and EcoRI sites of the pEGFP-C1 vector (CLONTECH) for expression of an EGFP-PH domain fusion protein in mammalian cells. Vectors directing the expression of a hemagglutinintagged constitutively active (farnesylated using the signal from c-Ha-Ras) form of the PI 3-K catalytic subunit (p110␤) and its kinase-inactive mutant (D919A/D937E), and their effects on phosphoinositide metabolism in HeLa cells, have been described (58). HeLa cells were maintained in Dulbecco's modified Eagle's medium supplemented with 10% fetal bovine serum. Cells were seeded onto 12-mm circular glass coverslips in wells of a six-well plate, and were transfected with 2 g of EGFP (or fusion protein) vector alone, or together with 20 g of the plasmid encoding farnesylated p110␤ (or its kinase-inactive mutant). Lipo-fectAMINE (Life Technologies, Inc.) was used for transfection according to the manufacturer's suggestions. Cells were grown for 2 days following transfection in Dulbecco's modified Eagle's medium/10% fetal bovine serum, washed with phosphate-buffered-saline, fixed in 2% paraformaldehyde/PBS, and mounted for fluorescence microscopy using Fluoromount (Southern Technology Associates, Birmingham, AL). Microscopy was performed using a Zeiss Axioplan fluorescence microscope or the confocal microscopy facility of the University of Pennsylvania Cancer Center.
Centrifugation Assays for PH Domain Binding to Lipid Vesicles-Small unilamellar vesicles (SUVs) of defined lipid composition were generated by co-dissolving the appropriate lipids in 1:1 chloroform: methanol containing 0.1% HCl. Lipid mixtures contained 80% (molar) di(dibromostearoyl)phosphatidylcholine (PtdCho; Avanti) plus 20% (molar) dipalmitoyl phosphatidyl-L-serine (PtdSer; Sigma). PtdIns-(4,5)P 2 (Sigma) was added to 3% (molar) where indicated. Lipid mixtures were dried under a nitrogen stream followed by high vacuum, and the dried mixtures were rehydrated to a final total lipid concentration of 25 mM in 25 mM HEPES, pH 7.2, 100 mM NaCl by bath sonication. The pH was adjusted to 7.2 if necessary, and lipids were subjected to at least 10 cycles of freezing (liquid N 2 ) and thawing (bath sonication at 45°C), until optical clarity was achieved. The brominated PtdCho was employed in order to produce SUVs that can be pelleted efficiently by ultracentrifugation, as described (59).
Centrifugation assays were performed in an assay volume of 100 l, in 25 mM HEPES, pH 7.2, containing 100 mM NaCl. PLC␦-PH and PlecN-PH were produced from a pET expression vector as described (14). Assay samples contained 5 M PH domain, and lipid at total concentrations of 0 -6 mM (assuming that only 50% of the vesicle lipid is available on the outer surface, values plotted in Fig. 6 have been divided by 2). Vesicle/protein mixtures were centrifuged for 1 h at 25°C at 85,000 rpm in a Beckman Optima TLX benchtop ultracentrifuge, using a TLA-120.1 rotor. 75 l of the supernatant was removed and assayed for protein content. After discarding the remaining supernatant, the vesicle pellet was resuspended by bath-sonication in 100 l of buffer, and 75 l was taken for protein assay. Protein assays employed the Pierce BCA assay, as directed by the manufacturers. For the vesicle pellet assays, SDS was added to 1% after incubation to remove scattering artifacts. A standard curve for each PH domain was generated in parallel, using proteins quantitated by A 280 (14), in order to determine the percentage of added protein pelleted (from a mean of the values suggested by the pellet and supernatant protein concentrations). Data were best-fit to the following equation, using ORIGIN (MicroCal).
K is a molar partition coefficient corresponding to the constant of proportionality between the concentration of protein bound to the outer SUV leaflet and its concentration in bulk solution (60). K values include no assumptions about stoichiometry, (although K D can be estimated as (mole ratio)/K if 1:1 binding of phospholipid were assumed).
[lipid] is the concentration of total available lipid (Ͼ Ͼ[protein] bound ), approximated by one half of the total lipid concentration (assuming that only 50% is available on the outer leaflet of SUVs).

RESULTS
In an effort to identify PH domains that specifically recognize one or a few phospholipid ligands, we developed a convenient qualitative dot-blot assay to assess apparent binding specificity. We used the assay to compare the ability of 10 different examples to bind a series of phosphoinositides and other phospholipids (Fig. 1). The selection of PH domains was largely based on our ability to obtain pure soluble protein in bacterial expression. PH domains from the N termini of rat PLC␦ 1 and PLC␥ 1 (61), from mouse Grp1 (35), mouse PKB (62), human Ras-Gap (63), human diacylglycerol kinase-␦ (DAGK-␦) (64), human pleckstrin (N-terminal) (1), human ␤ARK-1 (65), human ␤-spectrin (66), and human KIAA0053 (67) were each subcloned into pGEX-2TK for expression of a GST fusion protein containing a protein kinase A site between the GST and PH domain moieties. Purified protein was labeled by phosphorylation with protein kinase A in the presence of [␥-32 P]ATP (55,56), and 32 P-labeled GST-PH fusions were used to probe nitrocellulose filters on which defined amounts of pure phospholipids had been spotted (see "Experimental Procedures"). Fig. 1, the 10 PH domains tested in our initial screen all appeared to bind phosphoinositides. 32 P-Labeled GST alone gave no signals above background. While this assay clearly does not allow a quantitative comparison of phosphoinositide-binding specificity (see legend to Fig. 1 and "Discussion"), Fig. 1 does suggest that the different PH domains show quite different degrees of specificity. Only two of the PH domains tested (from PLC-␦ 1 and Grp1) appeared to select a single phosphoinositide over all others. For the PLC-␦ 1 PH domain (PLC␦-PH), we and others have previously shown that the PtdIns(4,5)P 2 headgroup (Ins(1,4,5)P 3 ) binds significantly more strongly than any other inositol phosphate (14,16). Similarly, the Grp1 PH domain (Grp1-PH), which showed clear selectivity for PtdIns(3,4,5)P 3 ( Fig. 1), specifically recognizes the headgroup of this phosphoinositide (Ins(1,3,4,5)P 4 ), as analyzed in detail below (see also Refs. 35 and 68).

An Initial Screen Indicates That PH Domains Differ in Both Preferred Ligand and Degree of Specificity-As shown in
Two other PH domains (from Akt/PKB and PLC-␥ 1 ) gave a significant signal with only two or three of the phosphoinositides tested. The Akt/PKB PH domain recognized only PtdIns(3,4)P 2 and PtdIns(3,4,5)P 3 , consistent with previous reports that it binds preferentially to these phosphoinositides, and binds both with similar affinities (24,25). The PLC-␥ 1 PH domain (PLC␥-PH) appeared to recognize only 3-phosphorylated phosphoinositides, consistent with previous studies (20). However, while Fig. 1 suggests a preference of PLC␥-PH for PtdIns-3-P, Falasca et al. (20) reported that this PH domain binds PtdIns(3,4,5)P 3 more strongly than either PtdIns-3-P or PtdIns(3,4)P 2 . Using a centrifugation assay to study PLC␥-PH binding to phosphoinositides in lipid vesicles (data not shown) rather than immobilized on nitrocellulose, we also found that PLC␥-PH binds to PtdIns(3,4,5)P 3 more strongly than to Pt-dIns-3-P (Յ3-fold), in agreement with Falasca et al. (20). This discrepancy between approaches highlights a likely limitation of our simple qualitative screen (see "Discussion"), although for Grp1-PH, PLC␦-PH, Akt/PKB-PH, and other examples, the dot-blot assay agrees well with other in vitro studies, arguing that it is a useful and rapid first-pass indicator.
The majority of the PH domains tested showed a much less marked preference for a single phosphoinositide or group of phosphoinositides than the examples discussed above (Fig. 1). Several also gave a significant signal with another anionic phospholipid, PtdSer (␤ARK-1, Ras-Gap, pleckstrin-N, DAGK-␦, and KIAA0053). This result raises the possibility that these PH domains may bind phosphoinositides through nonspecific electrostatic interactions with negatively charged surfaces, which is considered below for one case (PlecN-PH). Consistent with the results in Fig. 1, studies of inositol phosphate binding to the pleckstrin-N and DAGK-␦ PH domains (53) have shown that binding affinity correlates most strongly with the number of phosphate groups, regardless of their spatial arrangement. Thus, while some PH domains are clearly capable of recognizing a particular inositol phosphate or phosphoinositide isomer, the majority (at least when expressed as isolated domains in E. coli) appear to be quite promiscuous, and may simply bind to anionic surfaces.
As shown in Fig. 2B, of several inositol phosphates tested, only Ins(1,3,4,5)P 4 and Ins(1,3,4,5,6)P 5 gave interpretable titrations by ITC at these Grp1-PH concentrations. Ins-(1,3,4,5,6)P 5 formed a 1:1 complex with Grp1-PH, yielding a K D value of 590 nM ( Table I). Addition of a phosphate group to the 6-position of Ins(1,3,4,5)P 4 therefore compromises its ability to bind Grp1-PH by a factor of 20, or nearly 2 kcal/mol. Interestingly, ⌬H for Ins(1,3,4,5,6)P 5 binding to Grp1-PH is actually slightly more favorable (Ϫ22.8 kcal/mol) than ⌬H for Ins(1,3,4,5)P 4 binding (Ϫ18.9 kcal/mol), despite the less favorable ⌬G. Therefore, the inhibitory effect of the 6-phosphate upon binding must be entropic in origin (⌬S is Ϸ20 cal/mol/K lower than for Ins(1,3,4,5)P 4 binding). By analogy with the crystal structure of the PLC␦-PH/Ins(1,4,5)P 3 complex (44), we expect hydration of the 1,3,4, and 5-phosphates of Ins-(1,3,4,5)P 4 to be largely displaced by hydrogen bonding interactions with functional groups in the Grp1-PH binding site. Without specific groups in Grp1-PH to form hydrogen bonds with a phosphate in the 6-position, the 6-phosphate of Ins-FIG. 1. Comparison of phosphoinositide-binding specificity of PH domains using a dot-blot screen. 32 P-Labeled GST-PH domain fusion proteins were used to probe nitrocellulose filters onto which specific phosphoinositides had been spotted as described under "Experimental Procedures." The pattern of spots is depicted at the bottom of the figure, with PI 3-K products in the top row, PI 3-K substrates in the second row, and common membrane phospholipids in the third row. GST alone did not give any signal above background. Since exposure times differed for individual experiments, and the specific activity of labeled proteins is not well controlled, comparison of absolute spot intensities from blot to blot is not valid. The PH domains are grouped for discussion purposes according to the suggestions of Rameh et al. (30). Group 1 PH domains are those that specifically recognize PtdIns(3,4,5)P 3 , group 2 PH domains bind PtdIns(4,5)P 2 as well as other phosphoinositides, and group 3 PH domains recognize PtdIns(3,4)P 2 (and PtdIns(3,4,5)P 3 ) (see "Discussion").
Grp1-PH Associates with the Plasma Membrane when PI 3-K Is Activated-Given its specificity and high affinity for the PtdIns(3,4,5)P 3 headgroup, we anticipated that Grp1-PH should localize to cellular membranes in a PI 3-kinase-dependent manner. To test this in vivo, we generated a fusion protein of Grp1-PH with the enhanced green fluorescent protein (EGFP), as described under "Experimental Procedures." The EGFP/Grp1-PH fusion construct was transiently cotransfected  (Table I)  into HeLa cells with a plasmid directing expression of a farnesylated, constitutively active, form of the PI 3-K p110␤ catalytic subunit (58). The effects of overexpressing this enzyme in HeLa cells have been analyzed, and include substantial increases in PtdIns(3,4,5)P 3 , PtdIns(3,4)P 2 , and PtdIns-3-P production (58). The EGFP/Grp1-PH fusion showed substantial plasma membrane localization when co-transfected with the activated p110␤ subunit (Fig. 4A). In two separate transfections, the number of EGFP-transfected cells showing such plasma membrane fluorescence was 42 of 53 (79%) in one case, and 50 of 64 (78%) cells in the other. By contrast, when the EGFP/Grp1-PH fusion was similarly co-transfected into HeLa cells with a mutated, kinase-inactive, form of farnesylated p110␤ (58), few cells (Ͻ20%) showed plasma membrane localization (Fig. 4B). No membrane localization of EGFP itself was seen in control experiments upon cotransfection with either p110␤ variant (data not shown), and expression of both the EGFP fusion and hemagglutinin-tagged p110␤ variants were confirmed by Western blot (data not shown).
Thus, significant plasma membrane localization of Grp1-PH occurs in a manner that is strongly enhanced by introduction of a constitutively active PI 3-K catalytic subunit (but not a mutated form) into mammalian cells. Binding of Grp1-PH to PtdIns(3,4,5)P 3 (35,68), through specific, high affinity, headgroup recognition is therefore likely to be sufficient for signaldependent membrane recruitment mediated by PI 3-K products. Similar results have been reported for the PLC-␥ 1 PH domain (20), and, very recently, for the PH domains from the ARF exchange factors ARNO and cytohesin-1, which are very similar in sequence to Grp1-PH (69,70).

FIG. 3. Analysis of Grp1-PH inositol phosphate-binding specificity using a [ 3 H]Ins(1,3,4,5)P 4 competition assay.
The ability of various inositol phosphates to compete with Ins(1,3,4,5)P 4 for binding to Grp1-PH was compared using a spin-column competition assay (see "Experimental Procedures"). Different inositol phosphates were added at a range of molar excesses over a 1:1 Grp1-PH:Ins(1,3,4,5)P 4 complex (at 5 M), which contained 10 nCi of 3 H-labeled Ins(1,3,4,5)P 4 . The number of Ins(1,3,4,5)P 4 counts remaining bound to Grp1-PH was determined after passage through the spin column by scintillation counting, and is plotted against the molar excess of added competitor InsP n as a percentage of that measured without competitor. Competition With such low ⌬H values and high affinities, the calorimetric signal-to-noise ratio is poor (since the heat-per-injection is low) when working at protein concentrations required for accurate determination of K D .
By contrast with the 3-phosphorylated inositol phosphates, Ins(1,4,5)P 3 bound only weakly to EST684797-PH (Fig. 5), yielding a K D Ͼ 1,000 nM. Furthermore, as listed in Table III, ITC studies showed that, while Ins(1,3,4,5,6)P 5 , Ins(3,4,5,6)P 4 , and Ins(1,3,4,6)P 4 all bind strongly to EST684797-PH, Ins(1,4,5,6)P 4 does not. These results argue that the primary determinant for inositol phosphate binding to EST684797-PH is the presence of a pair of phosphates at the 3-and 4-positions. Beyond this requirement, the interaction appears to tolerate several different arrangements of phosphate groups. Thus, although Ins(1,3,4,5)P 4 is a high affinity ligand for both EST684797-PH and Grp1-PH, the two PH domains recognize distinct characteristics of this inositol phosphate. For Grp1-PH, 3 adjacent equatorial phosphate groups (e.g. at the 3-, 4-, and 5-positions) are necessary for high affinity binding, and additional phosphate groups can have a negative influence on binding. For EST684797-PH, phosphate groups at only the 3and 4-positions are necessary. This difference may be reflected in the fact that, despite similar affinities, ⌬H for Ins(1,3,4,5)P 4 binding to Grp1-PH is substantially more favorable than for binding to EST684797-PH. It is also of note that ⌬S, which is unfavorable (Ϫ28.8 cal/mol/K) for Ins(1,3,4,5)P 4 binding to Grp1-PH, is favorable for Ins(1,3,4,5)P 4 binding to EST684797-PH under the same conditions (approximately ϩ6 cal/mol/K). Comparison of the structures of the two PH domain/ Ins(1,3,4,5)P 4 complexes will be required for a fuller understanding of the differences in the thermodynamics of these recognition events.
PtdIns(4,5)P 2 Binding by PLC␦-PH, but Not PlecN-PH, Is Efficiently Competed by the Soluble Headgroup-Of the PH domains tested using the dot-blot assay, most appeared quite promiscuous in phospholipid binding (Fig. 1). Half of those tested also gave signals with PtdSer that were as strong as those with phosphoinositides. In two cases (PlecN-PH and Ras-Gap-PH), PtdCho was the only phospholipid that gave no signal. Since PtdCho is a zwitterionic phospholipid with no net charge, and PtdSer (like the phosphoinositides) has net negative charge, we hypothesized that lipid binding by some of the more promiscuous PH domains may be driven by nonspecific electrostatic attraction between the positive face of the domain (49) and negatively charged surfaces formed by acidic phospholipids. Using an ultracentrifugation assay (see "Experimental Procedures"), we compared binding of PlecN-PH and PLC␦-PH (expressed without fusion to GST) to SUVs containing PtdIns(4,5)P 2 and/or PtdSer, and analyzed the ability of Ins(1,4,5)P 3 to compete with PtdIns(4,5)P 2 binding.
Whereas PlecN-PH did not bind to pure PtdCho vesicles (data not shown), it bound significantly to SUVs containing 20% PtdSer in a PtdCho background (Fig. 6A). Nearly 50% of the total protein could be pelleted with these negatively  (1,3,4,5)P 4 competition assay for Grp1-PH K D values were estimated from the IC 50 value for each inositol phosphate compared with that obtained for unlabeled Ins(1,3,4,5)P 4 , as described in the text. IC 50 values were estimated best-fits to the data shown in Fig. 3

FIG. 4. A Grp1-PH/green fluorescent protein fusion localizes to the plasma membrane when co-expressed with constitutively active PI 3-K.
A plasmid encoding Grp1-PH fused to the EGFP was co-transfected into HeLa cells with a 10-fold excess of plasmid encoding a farnesylated, constitutively active, form of the p110␤ PI 3-K catalytic subunit (58) (A) or a farnesylated kinase-inactivated mutant of p110␤ (B). Cells were fixed and mounted for confocal fluorescence microscopy as described under "Experimental Procedures." Similar results were obtained in two independent experiments, as described in the text, the cells shown here being representative for each case. No detectable membrane localization of EGFP alone was seen in control experiments with and without p110␤. The nuclear localization of the EGFP/Grp1-PH fusion was also observed with the ARNO PH domain (but not with whole ARNO fused to EGFP), suggesting that it may reflect the ability of the smaller EGFP fusions to pass through nuclear pores (69). charged vesicles when total available lipid concentration (on the outer leaflet of the vesicles) was 3 mM. Further addition of 3% PtdIns(4,5)P 2 enhanced PlecN-PH vesicle binding by approximately 3-fold, with the best-fit molar partition coefficient (K) increasing from 316 M Ϫ1 to 1,006 M Ϫ1 . This effect may arise from weak recognition of the PtdIns(4,5)P 2 headgroup by PlecN-PH, or could simply reflect increased electrostatic attraction between the basic protein and the negatively charged vesicles as more acidic phospholipid is added (addition of 3% PtdIns(4,5)P 2 should increase effective vesicle charge by 1.5-2-fold). PLC␦-PH also showed substantial binding to PtdCho vesicles containing 20% PtdSer, with K Ϸ 430 M Ϫ1 (Fig. 6B). However, the effect of adding 3% PtdIns(4,5)P 2 to the vesicles was more pronounced for PLC␦-PH than for PlecN-PH (increasing K by more than 6-fold; K ϭ 2, 640 M Ϫ1 ). Thus, while both PH domains have a tendency to bind negatively charged membranes, PLC␦-PH shows a stronger preference for PtdIns(4,5)-P 2 over PtdSer than does PlecN-PH, as indicated in Fig. 1 (see also Ref. 15). To determine the role played by specific headgroup recognition in the enhanced binding of these PH domains to PtdIns(4,5)P 2 -containing vesicles, we investigated the effectiveness of Ins(1,4,5)P 3 as a competitor for this interaction (Fig.  6C). This experiment was performed with a fixed (available) lipid concentration of 0.5 mM, corresponding to an available PtdIns(4,5)P 2 concentration of 15 M. Addition of Ins(1,4,5)P 3 to around 12 M reduced PLC␦-PH vesicle binding by 50%, indicating that Ins(1,4,5)P 3 and PtdIns(4,5)P 2 bind PLC␦-PH with similar K D values, as reported previously (14,15). By contrast, PlecN-PH binding was affected very little, even upon addition of 80 M Ins(1,4,5)P 3 . At face value, these results suggest that PLC␦-PH binds PtdIns(4,5)P 2 and Ins(1,4,5)P 3 with similarly high affinities, while PlecN-PH binds much more strongly to PtdIns(4,5)P 2 than to Ins(1,4,5)P 3 . However, PlecN-PH appeared to bind PtdSer and PtdIns(4,5)P 2 similarly in Fig. 1, and enhancement of its SUV binding by PtdIns(4,5)P 2 is not greatly different from that expected from charge effects alone. Therefore, it is possible that PlecN-PH binding to these vesicles is driven largely by nonspecific electrostatic interactions with the negatively charged vesicle surface. The positively charged face of the PH domain would interact simultaneously with multiple acidic lipid molecules in the vesicle. Since these interactions will be cooperative, Ins(1,4,5)P 3 present in the 100 M range would not compete efficiently through ionic strength effects. DISCUSSION In a simple qualitative dot-blot assay for first-pass analysis of PH domain binding to phosphoinositides, clear ligand-binding specificity was seen in only few cases. The majority of the PH domains appeared quite promiscuous in their binding to these anionic phospholipids. For most of the cases in Fig. 1, both the identity of the preferred ligand (when relevant) and the degree of specificity indicated by our dot-blot assay have been confirmed using other methods. For example, studies of headgroup binding presented here confirm the findings for both Grp1-PH (see also Refs. 35 and 68) and EST684797. Similarly, the qualitative interpretation of the dot-blots for PLC␦-PH and Akt/PKB-PH are confirmed by several other studies (14,16,24,25), and the promiscuity in phosphoinositide binding by PlecN-PH and DAGK␦-PH is also reflected in inositol phosphate binding to these PH domains (Ref. 53; see Table III).
Despite the good agreement between the dot-blot assay and other methods in all of these cases, suggesting that our approach is useful as an initial screen, the discrepancy described earlier for the PLC-␥ 1 PH domain (see "Results") highlights an important caveat in interpreting results from the dot-blot assay. PLC␥-PH was previously reported to bind 3-phosphoinositides (20). However, while the dot-blot suggested that Pt-dIns-3-P is the preferred ligand, the results of Falasca et al. (20) and our own centrifugation assays (data not shown) indicate that that PtdIns(3,4,5)P 3 binds most strongly (although PtdIns-3-P binds only about 3-fold less strongly). The observed PI 3-K-dependent localization of PLC␥-PH to the plasma membrane (20) further supports this selectivity. In performing the dot blot assay, it is likely that some of the phospholipid applied to the nitrocellulose filter is removed during the blocking and washing steps. Since water solubility of the phospholipids correlates with their headgroup charge, removal will be most significant for highly phosphorylated phosphoinositides (particularly PtdIns(3,4,5)P 3 ). We therefore suggest that the discrepancy with PLC␥-PH results from a greater degree of removal of PtdIns(3,4,5)P 3 than PtdIns-3-P from the nitrocellulose filters, which leads to a false impression of the selectivity for PtdIns-3-P over PtdIns(3,4,5)P 3 (both of which bind to a significant extent). This differential lipid removal does not cause problems for Grp1-PH since it selects PtdIns(3,4,5)P 3 over PtdIns-3-P much more strongly than does PLC␥-PH.
Similarly, although the dot-blot assay for PLC␦-PH agrees very well with our previous studies of headgroup recognition (14 , Table III), Garcia et al. (15) have reported that PLC␦-PH binds to PtdIns(3,4,5)P 3 slightly more strongly than to PtdIns(4,5)P 2 in vesicles that also contain 33% PtdSer. More extensive removal of PtdIns(3,4,5)P 3 than of PtdIns(4,5)P 2 from the filters could explain why this is not seen in Fig. 1, although the higher negative charge of the lipid vesicles in Ref. 15 could provide an alternative explanation. Differential loss of phospholipids in the dot-blot assay will also tend to enhance apparent PtdSer binding compared with that for phosphoinositides. For example, with PlecN-PH, the spot for PtdSer in Fig. 1 appears more intense that that for PtdIns(4,5)P 2 , yet Fig.  6 indicates that PlecN-PH binding to vesicles containing 20% PtdSer is enhanced (3-fold) upon further addition of 3% . To obtain accurately measurable heats for individual injections in these titrations, the [sites]/K D was close to (or slightly more than) 100, resulting in a larger than usual error in K D determination. PtdIns(4,5)P 2 . This difference may reflect the fact that the PtdIns(4,5)P 2 -containing vesicles are simply more negatively charged than those containing PtdSer alone, while the spots in Fig. 1 contain the same mass of lipid. Alternatively, if the interaction is specific, greater removal of PtdIns(4,5)P 2 than PtdSer from the nitrocellulose would explain the difference.
Despite these problems, and the fact that relative off-rates rather than affinity per se will determine which phospholipids appear to bind most strongly, we have found the dot-blot approach to be a valuable starting point for designing experiments (and choosing examples) for analysis of specific ligand recognition by PH domains. It should also be noted that, when specific recognition of highly phosphorylated phosphoinositides is seen in this assay (as with Grp1-PH), the caveat of differential phospholipid removal from the nitrocellulose actually lends substantial confidence that the specificity is significant, as borne out by the results for Grp1-PH and EST684797-PH.

PH Domains Can Be Grouped According to Their Likely Physiological Ligand(s)
For the purposes of discussion, the PH domains in Fig. 1 6. Ins(1,4,5)P 3 competes for binding of PLC␦-PH, but not PlecN-PH, to lipid vesicles containing PtdIns(4,5)P 2 . Binding of PlecN-PH and PLC␦-PH (expressed without GST fusion) to SUVs was analyzed using a ultracentrifugation assay (see "Experimental Procedures"). Increasing concentrations of SUVs were added to assay samples containing a final PH domain concentration of 5 M. Vesicles (which contained brominated PtdCho to increase density) were pelleted by ultracentrifugation, and the percentage of protein pelleted with the lipid was determined by protein assay. The noted lipid concentrations refer to total available lipid (total lipid, divided by 2 to approximate the concentration available on the SUV outer surface). Binding to PtdCho vesicles containing either 20% (molar) PtdSer (Ⅺ) or 20% PtdSer plus 3% PtdIns(4,5)P 2 (q) was analyzed for PlecN-PH (A) and PLC␦-PH (B). Each experiment was repeated at least three times, and the mean data were fit to Equation 1 (see "Experimental Procedures") for determination of a molar partition coefficient (K). In C, the effect of increasing concentrations of Ins(1,4,5)P 3 on binding of PlecN-PH and PLC␦-PH to the PtdIns(4,5)P 2 -containing vesicles was analyzed. The total available lipid concentration for these competition experiments was fixed at 0.5 mM (corresponding to 15 M available PtdIns(4,5)P 2 ). Error bars represent the standard deviation around the mean of at least two independent experiments. PtdIns(4,5)P 2 , PtdIns(3,4)P 2 , and PtdIns(3,4,5)P 3 .
The third group (30) was for PH domains that specifically bind PtdIns(3,4)P 2 , and included only Akt/PKB-PH. As shown in Fig. 1, and reported elsewhere (24,25), Akt/PKB-PH appears to bind similarly to PtdIns(3,4)P 2 and PtdIns(3,4,5)P 3 . The EST684797 PH domain shows similar specificity, as reported above ( Fig. 5; Table III), and would therefore also be in group 3. Finally, the PLC␥ 1 PH domain has not been placed into any of these groups in Fig. 1, since it is not clear whether group 1 or 3 are most appropriate.
Additional dot-blot experiments that included PtdIns-5-P and PtdIns(3,5)P 2 (data not shown), both of which have recently been found to occur in mammalian and yeast cells (see Ref. 72 and references therein), did not alter any of the groupings suggested above. Binding of group 1 and group 3 PH domains to these phosphoinositides appeared substantially weaker than their binding to PtdIns(3,4,5)P 3 (group 1 and 3) or PtdIns(3,4)P 2 (group 3). PLC␦-PH gave much weaker signals with PtdIns-5-P or PtdIns(3,5)P 2 than with PtdIns(4,5)P 2 . Other group 2 PH domains did not distinguish PtdIns-5-P or PtdIns(3,5)P 2 from the multiple other phospholipids that they appeared to bind in Fig. 1, consistent with their promiscuity in phosphoinositide binding.

Relative Inositol Phosphate-binding Affinities and Specificities for PH Domains from Different Groups
For PH domain-directed membrane recruitment by a particular phosphoinositide, the PH domain must recognize the phosphoinositide with an affinity and specificity that reflects its cellular levels with respect to other phospholipids. For example, based on estimated local PtdIns(4,5)P 2 and PtdIns(3,4,5)P 3 concentrations in activated cells (51), specific membrane recruitment of an isolated PH domain by PtdIns(3,4,5)P 3 would require both high affinity binding and at least 20-fold selectivity for this PI 3-K product over PtdIns(4,5)P 2 .
Group 1 (Specific Binding to PtdIns(3,4,5)P 3 )-As shown in Table III, PH domains placed in group 1 by our studies or from other reports (Grp1, Btk, and Gap1 IP4BP ) all fulfill these requirements. Each binds the PtdIns(3,4,5)P 3 headgroup with K D Ͻ 50 nM, while binding over 100-fold more weakly to the PtdIns(4,5)P 2 headgroup. PI 3-K-dependent localization to the plasma membrane has been demonstrated for each of these isolated PH domains, in addition to those from the close Grp1relatives, cytohesin-1 and ARNO (34, 69, 70) (Fig. 4).
Group 3 (Binding to PtdIns(3,4)P 2 and PtdIns(3,4,5)P 3 )-Like those from group 1, group 3 PH domains also recognize PI 3-K products specifically in vitro and in cellular contexts. This has been shown for Akt/PKB-PH by several studies (22)(23)(24)(25)(26), as well as the dot-blot assay (Fig. 1) and the yeast rescue assay developed by Isakoff et al. (52). However, the quantitative extent of selection for PI 3-K products in vitro differs between reports. The EST684797 PH domain was found to recognize PI 3-K products specifically in vivo by Isakoff et al. (52), and we find that this PH domain binds to Ins(1,3,4,5)P 4 with a K D (43 nM) similar to those seen for group 1 PH domains (Table III). However, unlike the group 1 PH domains, EST684797-PH binds similarly strongly to Ins(1,3,4)P 3 . Both Ins(1,3,4,5)P 4 and Ins(1,3,4)P 3 bind to EST684797-PH Ͼ20-fold more strongly than does Ins(1,4,5)P 3 , consistent with the requirements for membrane recruitment of this PH domain by PI 3-K products in vivo. Recognition of both PtdIns(3,4,5)P 3 and PtdIns(3,4)P 2 by group 3 PH domains, by contrast with the group 1 specificity for PtdIns(3,4,5)P 3 may have physiological relevance. For EST684797-PH, the role is not clear. However, for Akt/PKB, it has been reported that PtdIns(3,4)P 2 can activate the kinase in vitro, while PtdIns(3,4,5)P 3 cannot (23,24,25,73). It is possible that the two phosphoinositides differ in allosteric regulation of kinase activity upon PH domain binding (18), although the mechanism is not at all clear.
Group 2 (Binding to PtdIns(4,5)P 2 and Other Phosphoinositides)-For membrane localization mediated by PtdIns(4,5)P 2 the requirements for high affinity and specificity are less stringent, since this phosphoinositide is much more abundant in the cell than PI 3-K products. PH domains in group 2 are likely to be regulated by PtdIns(4,5)P 2 (or other relatively abundant phospholipids), and tend to have both lower phosphoinositidebinding affinities and less stringent specificity (Table III). For example, binding of PLC␦-PH to its preferred ligands (PtdIns(4,5)P 2 and Ins(1,4,5)P 3 ) is 4 -8 fold weaker than seen with group 1 PH domains (Table III). PLC␦-PH localization to the plasma membrane has been reported by several studies (20,74,75), and is independent of PI 3-K activity (20,75). Although somewhat specific, PLC␦-PH does not appear to select against other phosphoinositides as strongly as the group 1 PH domains. Since its preferred ligand is relatively abundant in vivo, there is less requirement for stringent selectivity.
Not all group 2 PH domains bind strongly to cellular membranes when studied in isolation. Plasma membrane localization of a GFP-spectrin PH domain fusion protein has been reported (76). However, we were unable to detect significant membrane localization when isolated ␤ARK1-PH or PlecN-PH were injected into IMR-33 fibroblasts as GST fusion proteins. For ␤ARK1-PH, a very slight increase in membrane localization was seen when cells were treated with serum (data not shown). For PlecN-PH, membrane localization was barely detectable in serum-starved cells, and serum-treatment had no effect. ␤ARK1-PH binds very weakly (K D ϭ 200 M) to Ins-(1,4,5)P 3 (46), and PlecN-PH binds relatively weakly to PtdIns-(4,5)P 2 in vesicles (K D Ͼ 30 M), in an interaction that is not substantially inhibited by addition of Ins(1,4,5)P 3 to 100 M (Ref. 13 and Fig. 6). Strong membrane localization would not be expected to be driven by interactions with these affinities. It is therefore possible that these (and other) PH domains participate in membrane localization of their host molecules by cooperating with other regions of the molecule. Indeed, membrane recruitment of ␤ARK-1 requires both PtdIns(4,5)P 2 binding by its PH domain and G ␤␥ binding by a region C-terminal to the PH domain (12). Similarly, Ma et al. (77) found that pleckstrin does not localize to the plasma membrane unless a region to the C terminus of PlecN-PH is phosphorylated. This finding suggests that the N-terminal PH domain alone is not sufficient for membrane targeting of pleckstrin: there is a requirement for additional interactions mediated by other regions of pleckstrin or for modification of PlecN-PH by this phosphorylation event. Finally, the N-terminal PH domain from Tiam-1, which shows a 3-fold difference in binding to PtdIns(3,4,5)P 3 and PtdIns(4,5)P 2 (30), has been found to require an adjacent protein interaction domain for targeting Tiam-1 to the membrane (78).

Structural Basis for Differences in PH Domain Specificity
Nearly all PH domains appear capable of binding at least weakly to inositol phosphates. Nonspecific electrostatic interactions between the characteristic positively charged face of PH domains (49) and the anionic ligand are likely to contribute significantly to this binding. Indeed, all PH domains tested bind to InsP 6 with a K D below 10 M. For some PH domains, notably PlecN-PH and DAGK␦-PH, inositol phosphate-binding affinity appears to correlate most well with the charge (number of phosphates) on the ligand (Table III). In other cases, notably PH domains from group 1, highly specific recognition of one particular spatial arrangement of phosphate groups appears to be superimposed on this nonspecific electrostatic attraction. Selective headgroup recognition presumably requires a specific binding site on the PH domain, with a precise arrangement of charged side-chains. A specific binding site for Ins(1,4,5)P 3 has been seen in the crystal structure of the PLC␦-PH/Ins(1,4,5)P 3 complex (44), and the unoccupied site for Ins(1,3,4,5)P 4 binding has been seen in the crystal structure of the Btk PH domain (45).
As pointed out by Isakoff et al. (52), most of the PH domains that fall into group 1 or group 3 from various studies share significant sequence similarities in the ␤1/␤2 loop region (a key region of PLC␦-PH/Ins(1,4,5)P 3 interactions). Other PH domains, notable for their conservation of 3-dimensional structure in the absence of significant sequence identity, do not contain these characteristic sequences, and may simply present a positively charged face that can bind negatively charged surfaces such as inositol phosphates or anionic phospholipids in membranes. For PH domains that do contain a specific binding site, there are significant differences in affinity (e.g. for PLC␦-PH and Grp1-PH), degree of specificity (e.g. for EST684797-PH and Grp1-PH), and precise modes of ligand recognition (e.g. unlike Grp1-PH, the Btk and Gap1 PH domains do not show a preference for Ins(1,5,6)P 3 and/or Ins(1,2,5,6)P 4 over other non-cognate inositol phosphates). A comparison of high-resolution structures determined for complexes formed by these PH domains will be required to understand these differences fully.