Structural and Membrane Binding Analysis of the Phox Homology Domain of Phosphoinositide 3-Kinase-C2α*

Phox homology (PX) domains, which have been identified in a variety of proteins involved in cell signaling and membrane trafficking, have been shown to interact with phosphoinositides (PIs) with different affinities and specificities. To elucidate the structural origin of diverse PI specificities of PX domains, we determined the crystal structure of the PX domain from phosphoinositide 3-kinase C2α (PI3K-C2α), which binds phosphatidylinositol 4,5-bisphosphate (PtdIns(4,5)P2). To delineate the mechanism by which this PX domain interacts with membranes, we measured the membrane binding of the wild type domain and mutants by surface plasmon resonance and monolayer techniques. This PX domain contains a signature PI-binding site that is optimized for PtdIns(4,5)P2 binding. The membrane binding of the PX domain is initiated by nonspecific electrostatic interactions followed by the membrane penetration of hydrophobic residues. Membrane penetration is specifically enhanced by PtdIns(4,5)P2. Furthermore, the PX domain displayed significantly higher PtdIns(4,5)P2 membrane affinity and specificity when compared with the PI3K-C2α C2 domain, demonstrating that high affinity PtdIns(4,5)P2 binding was facilitated by the PX domain in full-length PI3K-C2α. Together, these studies provide new structural insight into the diverse PI specificities of PX domains and elucidate the mechanism by which the PI3K-C2α PX domain interacts with PtdIns(4,5)P2-containing membranes and thereby mediates the membrane recruitment of PI3K-C2α.

Recent structural and modeling studies of a variety of PX domains have provided insight into the stereospecific head group recognition and membrane-binding mechanisms of PX domains. For example, the crystal structure of the p40 phox -PtdIns3P complex revealed that basic residues, Lys 92 and Arg 58 , specifically form hydrogen bonds with the 1-and 3-phosphate of PtdIns3P, respectively, thereby achieving the stereospecific recognition of PtdIns3P (43). The crystal structure of cytokineindependent survival kinase-PX showed that this domain also has all the basic residues necessary for binding the 3-phosphate of PtdIns3P (44). The crystal structure of the PX domain of p47 phox (45) revealed that this PX domain has two phospholipid-binding sites for PtdIns(3,4)P 2 and phosphatidic acid (or phosphatidylserine), respectively. Similarly, a mutational and homology modeling study of the phospholipase D1 PX domain suggested that it also has two binding sites, one specifically binding PtdIns(3,4,5)P 3 and the other interacting nonspecifically with anionic phospholipids (41). The crystal structures of the free and PtdIns3P-bound PX domain of yeast Grd19p protein showed the lipid-induced local conformational changes involving putative membrane-penetrating hydrophobic residues (46).
To gain a better understanding of differential PI specificities and membrane-binding mechanisms of PX domains, we determined the x-ray crystal structure of the PI3K-C2␣ PX domain that was reported to specifically bind PtdIns(4,5)P 2 (28). We also measured the interaction of the PX domain and mutations with model membranes containing various PIs by surface plasmon resonance (SPR) and monolayer penetration analyses. In addition, we measured the PI-dependent membrane binding of the C2 domain and the full-length PI3K-C2␣. Results from these studies provide new insight into how the PI3K-C2␣ PX domain specifically recognizes PtdIns(4,5)P 2 and how the PX and C2 domains mediate the PtdIns(4,5)P 2 -dependent membrane targeting and activation of PI3K-C2␣.
Structure Determination-For the structure determination, DNA encoding the human PI3K-C2␣ PX domain (residues 1405-1541) was amplified by PCR from IMAGE clone 825260 and subsequently cloned with a C-terminal GSH 6 affinity tag in pJL vector. The cloned PX domain was sequenced and showed two differences with respect to the only published human sequence of PI3K-C2␣ (accession code NP_002636). Residue 1450 is Arg in our sequence instead of Trp, and residue 1464 is an Asp instead of Val. Residue 1464 is critical for the lipid specificity of PX domains (see below). All of the PI3K-C2␣ sequences that have been reported have an Asp at the position equivalent to human PI3K-C2␣ residue 1464, and all of the PI3K-C2␤ and PI3K-C2␥ sequences have a Glu at this position. Residue 1450 corresponds to a more variable region among PX domains in general, but it is an Arg in all other mammalian PI3K-C2 sequences. Consequently, it is likely that these differences with respect to the data base reflect errors in the data base entry.
The protein was expressed in Escherichia coli strain C41(DE3) and purified by Ni 2ϩ -affinity, heparin, and gel-filtration chromatography as described in the Supplemental Methods. The protein in gel-filtration buffer (20 mM Tris-HCl (pH 7.4 at 25°C), 100 mM NaCl, and 5 mM dithiothreitol) was concentrated to 5 mg/ml. Crystals were obtained by hair seeding drops (3 l of protein plus 3 l of reservoir solution) that were incubated at 17°C over a reservoir consisting of 1.7 M Li 2 SO 4 and 0.05 M MES, pH 5.6. For the hair seeding, a hair from a poodle was passed through a drop with crystals and then passed through a series of six freshly set up drops with no crystals so that the number of nuclei would thereby be serially diluted. When passing through each of the 6 drops, a pattern of an "X" was traced through the drop with the hair. In most cases, the 1st drop of the series had a shower of small crystals along the path of the hair, and in the later drops in the series only a few crystals were formed, but these were much larger. Crystals were visible after 24 h and grew to full size within 1 week.
For diffraction data collection, crystals were cryoprotected in reservoir solution supplemented with 15% glycerol and were flash-frozen in a nitrogen stream at 100 K. A heavy metal derivative crystal was prepared by soaking a crystal in cryoprotectant supplemented with 1 mM ethylmercury thiosalicylate for 15 min. An initial, lower resolution, 2.9 Å native data set was also collected and used for initial single isomorphous replacement phasing. Tables 1-3 summarizes the data collection statistics. Images were processed with the program MOSFLM (49) and refined with SCALA (50). Three mercury sites were located with RANTAN and refined with AutoSHARP. After density modification with SOLOMON (51) and DM (52), an initial model was manually built in a 2.9-Å resolution electron density map using COOT (53) and O (54). A subsequent 2.6-Å data set was obtained for a native crystal, and this data set was used for refinement with REFMAC (55). There is one PX domain in the asymmetric unit. The Ramachandran analysis with the program PROCHECK (56) shows 88.8% of residues in the most probable regions and none in disallowed regions. The refinement statistics are given in Tables 1-3 Mutagenesis and Protein Expression-Mutagenesis of the PI3K-C2␣ PX domain was performed using the overlap extension PCR method (57). Constructs were subcloned into the pET21a vector containing a C-terminal hexahistidine tag and transformed into DH5␣ cells for plasmid isolation. After checking each construct for correct sequence, the plasmid was transformed into BL21(DE3) cells for protein expression. The PX domains used for monolayer and SPR measurements were expressed and purified as described previously (58) with minor variation, i.e. the domains purified by nickel-nitrilotriacetic acid column chromatography were further purified by gel filtration chromatography. Briefly, a 1-ml sample of each PX domain was loaded onto a Superdex 200 column (Amersham Biosciences) equilibrated and eluted with 20 mM Tris-HCl buffer, pH 7.4, containing 0.16 M KCl. Fractions corresponding to the PI3K-C2␣ PX peak were pooled and concentrated to 1 ml. The C2 domain (residue 1562-1679) was cloned into pET21a using NdeI and XhoI sites. For C2 domain expression, BL21(DE3) cells were grown at 37°C in LB media containing 100 g/ml ampicillin. Protein expression was induced when the absorbance at 600 nm reached 0.8 and continued for 4 h at 25°C. Cells were pelleted by centrifugation, and the C2 domain was purified as described previously (59). The full-length PI3K-C2␣ was expressed in Sf9 cells, purified, and assayed as described previously (21). Protein concentration was determined by the bicinchoninic acid method (Pierce).
Monolayer Measurements-The penetration of PI3K-C2␣ and isolated domains into the phospholipid monolayers of different lipid compositions was measured in terms of the change in surface pressure () using a 10-ml circular Teflon trough and a Wilhelmy plate connected to a Cahn microbalance as described previously (60).
SPR Measurements-All SPR measurements were performed at 23°C in 10 mM HEPES, pH 7.4, containing 0.16 M KCl. A detailed protocol for coating the L1 sensor chip with lipid vesicles was reported previously (61,62). Briefly, after washing the sensor chip surface with the buffer, 90 l of vesicles composed of POPC/POPE/PI (78:20:2) (see Fig. 3) were injected at 5 l/min to give a response of 5500 resonance units. Similarly, a control surface was coated with POPC/POPE (80:20) vesicles, to give the same resonance unit response as the active binding surface. Under our experimental conditions, no binding was detected to this control surface beyond the refractive index change for all proteins. Each lipid layer was washed with 10 l of 50 mM NaOH three times at 100 l/min. Typically, no decrease in lipid signal was seen after the first injection. Equilibrium (steady state) SPR measurements were performed with the flow a Data sets were collected at ESRF beamline ID14-2 at ϭ 0.933 Å using a MAR CCD detector. The single isomorphous replacement phasing and initial model building was carried out using native 1. Subsequently, a higher resolution native data set 2 was collected and used for refinement  rate of 5 l/min to allow sufficient time for the R values of the association phase to reach saturating response values (R eq ) (62, 63). After sensorgrams were obtained for five different concentrations of each protein within a 10-fold range of K d (see Table  4), each of the sensorgrams was corrected for refractive index change by subtracting the control surface response from it. R eq values were then plotted versus protein concentrations (C), and the K d value was determined by a nonlinear least squares analysis of the binding isotherm using an equation, R eq ϭ R max /(1 ϩ K d /C). Each data set was repeated three times to calculate average and standard deviation values.
The PI3K-C2␣ PX domain crystallized in space group P423 with a ϭ 115.9 Å and one molecule per asymmetric unit. Residues 1405-1408, 1488 -1497, and 1540 -1541 are disordered. The fold of the PI3K-C2␣ PX domain consists of a three-stranded meander ␤-sheet subdomain followed by a helical subdomain with four ␣-helices and one 3 10 helix (Fig. 1A). The ␤-sheet is twisted by the bulge at residues 1430 and 1431 in strand ␤1. This twist of the ␤-sheet toward the helical subdomain enables the formation of a pocket bordered on one side by the beginning of strand ␤2. A sulfate is bound in this pocket on the surface of the domain and interacts with Arg 1503 of helix ␣3.
The PI3K-C2␣ PX domain superimposes closely on other PX domains such as the PX domain from p40 phox (Fig. 1B). However, a loop immediately following the PXXP motif at residues 1483-1486 is disordered in the PI3K-C2␣ PX domain (shown as dashed line in Fig. 1, A and B). This disorder gives the lipid-binding pocket a much more open appearance compared with the analogous pocket of p40 phox . The core of the ␤-sheet subdomain is nearly identical among PX domains. However, the ␤1-␤2 turn in the PI3K-C2␣ PX domain has a cis-proline (Pro 1438 ), accentuating the curvature of the concave face of the ␤-sheet relative to other PX domains.
The comparison of the lipid-binding sites from p40 phox (43) and PI3K-C2␣ PX domains immediately reveals why it would be impossible for PtdIns3P to bind the PI3K-C2␣ PX domain. Both negative selection against the 3-phosphate and loss of positive selection for the 3-phosphate contribute to the lipid specificity of the PI3K-C2␣ PX domain. Fig. 2, A and B, illustrate that positive selection is lost because the essential, conserved determinant of the 3-phosphate interaction, which is Arg 58 in p40 phox , has been replaced by Thr 1462 in PI3K-C2␣. Arg 60 of the p40 phox PX domain, which is one of the residues responsible for its 1-phosphate coordination, has been replaced in PI3K-C2␣ by Asp 1464 . The orientation of Asp 1464 is such that it imposes a negative selection against 3-phosphate, as it would electrostatically repel a phosphate.   DECEMBER 22, 2006 • VOLUME 281 • NUMBER 51

JOURNAL OF BIOLOGICAL CHEMISTRY 39399
The PP II /␣2 loop is disordered in the PX domain structure, preventing us from suggesting a detailed model of PtdIns(4,5)P 2 binding to the PI3K-C2␣ PX domain. However, the bound sulfate molecule in the PI3K-C2␣ PX domain seems to indicate a plausible position of 4-phosphate binding. The sulfate in PI3K-C2␣ is hydrogen-bonded to Arg 1503 , which superimposes well with the p40 phox residue (Arg 105 ) that interacts with the 4-and 5-OH of PtdIns3P. In the PI3K-C2␣ PX domain structure, there is no clear indication for a 5-phosphate specificity determinant. However, a few candidate residues could be suggested as illustrated in Fig. 2B. In the p40 phox and Grd19 PX domains, the backbone of the variable loop between PP II and ␣2 embraces the 4-and 5-OH of the bound phosphoinositide, and the side chain of a basic residue analogous to Arg 1503 in helix ␣2 of PI3K-C2␣ interacts with these groups (Fig. 2B) (46). The variable PP II /␣2 loops of p40 phox and Grd19 also provide the 1-phosphate ligand (Fig. 2B).
Intramolecular interactions between polyproline regions and SH3 domains have been reported previously for several proteins, including p47 phox (45). It appears that for such intramolecular interactions to occur in PX domains, a type II polyproline motif is required. The p40 phox , Grd19, and PI3K-C2␣ PX domains all contain polyproline regions, which feature the PXXPXR(K) consensus, which corresponds to the type II SH3 ligand motif PXXPXϩ (where ϩ refers to a positively charged amino acid and X to any residue). In p40 phox -PX the last residue of the motif, Lys 92 , participates in lipid recognition. The equivalent residue in PI3K-C2␣-PX, Arg 1488 , is disordered, but it is possible that Arg 1488 might become ordered on lipid binding, and it may be that this residue is the 1-phosphate ligand (Fig.  2B). In all three proteins, the prolines of the PXXP loop are buried. PI3K-C2␣ isoforms do not contain SH3 domains, and to date there has been no report of an interaction between PI3K-C2␣ and the SH3 domain from another protein. These observations suggest that perhaps type II SH3-ligand motifs in PX domains are not sufficient on their own to mediate interactions with SH3 domains.
We then measured the binding of PI3K-C2␣ PX mutants to POPC/POPE/PtdIns(4,5)P 2 (78:20:2) vesicles. In agreement with the structural analysis, the mutation of the conserved  2) surface, and R eq values were measured. B, binding isotherm was generated from the R eq (average of triplicate measurements) versus the concentration of the PI3K-C2␣ PX domain plot. A solid line represents a theoretical curve constructed from R max (ϭ170 Ϯ 2) and K d values (ϭ25 Ϯ 4 nM) determined by nonlinear least squares analysis of the isotherm using the following equation: R eq ϭ R max /(1 ϩ K d /C). 10 mM HEPES buffer, pH 7.4, with 0.16 M KCl was used for all measurements.

PI3K-C2␣ PX Domain Structure
Arg 1503 (i.e. R1503A) abolished binding to PtdIns(4,5)P 2 -containing membranes, corroborating the notion that this conserved Arg residue is essential for recognition of PtdIns(4,5)P 2 and may interact with the 4-phosphate. Mutations of two other basic residues, Arg 1488 and Arg 1493 , that are candidates for 5-phosphate recognition in the PtdIns(4,5)P 2 -binding pocket of PI3K-C2␣ PX (R1488A and R1493A) reduced the vesicle affinity of the PX domain by 7-and 23-fold, respectively, whereas mutations of other neighboring residues (K1440A and K1497A) had little effect. This indicates that Arg 1488 and Arg 1493 , in addition to Arg 1503 , may be involved in specific PtdIns(4,5)P 2 recognition. Gross misfolding was precluded for all mutations as they were expressed at the same level as wild type and bound POPC/POPS (5:5) vesicles with the same affinity as wild type (data not shown).
Recently, it has been shown that hydrophobic residues surrounding the PI-binding pockets of FYVE (67,68), PX (46,58), and ENTH (66, 69) domains can penetrate the membrane following PI binding. We thus measured the effects of mutations of hydrophobic residues (Val 1490 and Leu 1491 ) surrounding the PI-binding pocket of the PI3K-C2␣ PX domain to see if they are involved in membrane penetration. V1490A and L1491A exhibited a 7-and 5-fold lower membrane affinity than the wild type, respectively. Interestingly, mutation of Leu 1479 , which is located on the same molecular surface with Val 1490 and Leu 1491 but distant from the PI-binding pocket, had little effect on binding to PtdIns(4,5)P 2 -containing vesicles. This indicates that only hydrophobic residues adjacent to the PI-binding pocket specifically penetrate the membrane (see below).
Membrane penetration properties of PI3K-C2␣ PX domain and mutants were further characterized by lipid monolayer penetration measurements. Recent studies have shown that PIs can specifically induce the membrane penetration of FYVE (68), PX (58), and ENTH (66, 69) domains. To determine whether or not PtdIns(4,5)P 2 can also elicit the membrane penetration of the PI3K-C2␣ PX domain, we first measured the penetration of the PX domain to monolayers with different lipid compositions (see Fig. 5). The PI3K-C2␣ PX domain cannot penetrate the POPC/POPE (80:20) monolayer with the surface pressure above 26 dynes/cm; however, when 2 mol % PtdIns(4,5)P 2 was included in the monolayer (i.e. POPC/POPE/ PtdIns(4,5)P 2 (78:20:2)), the domain penetrated the monolayer with the surface pressure of up to 33 dynes/cm. Because the surface pressure of biological membranes has been estimated to be 30 -35 dynes/cm (70 -72), these data suggest that PtdIns(4,5)P 2 can induce the penetration of the PI3K-C2␣ PX domain into cell membranes under physiological conditions. The specific nature of this penetration was confirmed by two additional measurements. First, neither 2 mol % PtdIns(3,4) 2 nor 2 mol % PtdIns(3,4,5)P 3 nor 20 mol % phosphatidylserine in the monolayer significantly influenced the penetration behaviors of PI3K-C2␣ PX. Second, mutations of residues involved in specific PtdIns(4,5)P 2 recognition (i.e. R1503A and R1488A) greatly reduced the penetration of the PI3K-C2␣ PX domain to the POPC/POPE/PtdIns(4,5)P 2 (78:20:2) monolayer, whereas mutations that had no effect on K d values, K1440A and L1479A, did not cause reduction in monolayer penetration (Fig. 6A). Finally, V1490A and L1491A had low monolayer penetration into the POPC/POPE/PtdIns(4,5)P 2 (78:20:2) monolayer (Fig.  6B), confirming the notion that these hydrophobic residues are directly involved in the membrane penetration of the PI3K-C2␣ PX domain. In contrast, the mutation of Leu 1479 located at the other end of the membrane-binding surface had little effect on the monolayer penetration (Fig. 6B), showing that this part of PI3K-C2␣ PX does not significantly participate in membrane penetration.
PtdIns(4,5)P 2 Binding of C2 Domain Versus PX Domain-The recent study of the C2 domain of PI3K-C2␣ indicated that it had affinity for PIs, particularly PtdIns(4,5)P 2 and PtdIns(3,4)P 2 (30). This finding raises a question as to whether the PX domain or the C2 domain plays a major role in PI-mediated membrane binding of PI3K-C2␣. To compare the PI specificities and affinities of the PX and C2 domains, we measured the binding of the C2 domain to PI-containing vesicles by   DECEMBER 22, 2006 • VOLUME 281 • NUMBER 51 SPR analysis and also monitored their binding to PI-containing lipid monolayers. We first measured the PI specificity of PI3K-C2␣ C2 domain by SPR analysis using the active surface coated with POPC/ POPE/PI (75:20:5) vesicles. A control surface was coated with POPC/POPE (80:20) vesicles. Fig. 7 shows the binding response changes for 500 nM of the PI3K-C2␣ C2 domain injected over various PI-containing membranes. In agreement with the previous report (30), the PI3K-C2␣ C2 domain binds PtdIns(3,4)P 2 and PtdIns(4,5)P 2 ; however, unlike the PX domain, the C2 domain can significantly bind other PIs and consequently shows low PI selectivity. In fact, quantitative analysis of the binding data (see Table 5) showed that the C2 domain could not effectively distinguish among various PIs. Perhaps more importantly, the affinity of the C2 domain for POPC/POPE/ PtdIns(4,5)P 2 (78:20:2) (K d ϳ 490 nM) was about 20-fold lower than that of the PX domain (K d ϳ25 nM) under the same conditions. We also measured the binding of the PX and C2 domains as a function of PtdIns(4,5)P 2 concentration (i.e. using POPC/POPE/PtdIns(4,5)P 2 (80-x:20:x) vesicles). As shown in Fig. 8, the PX domain had higher affinity than the C2 domain throughout the PtdIns(4,5)P 2 concentration range used, and the difference was the most dramatic at the physiologically relevant PtdIns(4,5)P 2 concentration (i.e. Ͻ5 mol %). Also, the PtdIns(4,5)P 2 concentration for half-maximal binding of the PX domain (ϳ2 mol %) was significantly lower than that of the C2 domain (ϳ7 mol %), and the latter is consistent with the reported value (30). Collectively, these results indicate that the PX domain would play a dominant role in PtdIns(4,5)P 2mediated membrane binding of PI3K-C2␣.

PI3K-C2␣ PX Domain Structure
We then measured the monolayer penetration of the C2 domain to see if the domain shows significant membrane penetration and if PIs could induce the penetration of the C2 domain. Fig. 9A illustrates that the C2 domain has much lower monolayer penetrating power than the PX domain in the presence of PtdIns(4,5)P 2 and the c value for the C2 domain was significantly lower than 30 dynes/cm. Also, unlike the case with the PX domain, PtdIns(4,5)P 2 (or PtdIns(3,4)P 2 ) enhanced the monolayer penetration of the C2 domain only slightly. This again indicates that the PX domain should be a main player in all aspects of membrane interactions of PI3K-C2␣, including membrane penetration.
PtdIns(4,5)P 2 -mediated Membrane Binding of the Full-length PI3K-C2␣-To verify the notion that the PX domain controls the membrane binding of PI3K-C2␣, we prepared wild type and R1503A mutant (PtdIns(4,5)P 2 binding-defective; see Table 4) of the full-length PI3K-C2␣ and compared their membrane binding properties with the isolated PX and C2 domains. R1503A was expressed as effectively as wild type in the Sf9 cells and had the basal activity comparable with that of wild type, precluding the gross misfolding of the mutant (data not shown).

PI3K-C2␣ PX Domain Structure
SPR analysis (see Table 5) shows that the full-length PI3K-C2␣ and the isolated PX domain have essentially the same affinity for PtdIns(4,5)P 2 -containing membranes (ϳ20 nM), underscoring the predominant role of the PX domain in the membrane binding of PI3K-C2␣. Interestingly, the R1503A mutant of the full-length protein had 25-fold lower affinity than the wild type, which was comparable with that of the isolated C2 domain. This suggests that the C2 domain supplements the function of the PX domain in PtdIns(4,5)P 2 -mediated membrane binding and that the two domains work additively rather than synergistically. Finally, we compared the monolayer penetration of the full-length PI3K-C2␣ and the isolated PX domain (see Fig. 9B). PI3K-C2␣ penetrated the monolayer in a PtdIns(4,5)P 2dependent manner similarly to the isolated PX domain, whereas the R1503A mutation abrogated the PtdIns(4,5)P 2 dependent penetration, demonstrating that the PX domain is essential for the penetration of the full-length enzyme.

DISCUSSION
This study addresses two important questions: how does the PX domain of PI3K-C2␣ specifically recognize PtdIns(4,5)P 2 and how does the PX domain mediate the membrane binding and activation of PI3K-C2␣? PX domains are similar to PH domains in that they have highly variable PI specificities and affinities. High resolution structures of several PX domains that bind the 3Ј-phosphate have been determined as either free proteins or complexes with PI (43)(44)(45)(46). However, the structural information for the PX domains that specifically bind non-3Јphosphate-PI has not been available.
The crystal structure of the PI3K-C2␣ PX domain clearly accounts for its inability to bind PtdIns3P that most PX domains prefer. A canonical 3Ј-phosphate ligand (i.e. Arg 58 for p40 phox PX) is substituted for by Thr, and an acidic residue, Asp 1464 , replaces a 1Ј-phosphate ligand (i.e. Arg 60 in p40 phox PX). This negative selection would not allow PtdIns3P to favorably interact with the PI-binding pocket of the PI3K-C2␣ PX domain. Because of the disorder in the PP II /␣2 loop of the PI3K-C2␣ PX domain, there is no direct structural information on how the domain achieves stereospecific recognition of the 4Ј-and 5Ј-phosphates of PtdIns(4,5)P 2 . However, the location of a bound sulfate ion, sequence alignment, and our mutational analysis suggest that three basic residues, Arg 1488 , Arg 1493 , and Arg 1503 , are involved in PtdIns(4,5)P 2 binding. Presumably, Arg 1503 interacts with the 4Ј-phosphate, whereas Arg 1493 contacts the 5Ј-phosphate. Arg 1488 may be involved in binding to either the 5Ј-phosphate or the 1Ј-phosphate.
Our previous studies on the FYVE (68), PX (58), and ENTH (66) domains have indicated that PI binding specifically induces the membrane penetration of surface hydrophobic residues surrounding the PI-binding pocket, presumably by causing local conformational changes of proteins and by neutralizing the positive electrostatic potential surrounding hydrophobic loop residues that interferes with the membrane penetration of domains. Based on these previous studies, we proposed a membrane-binding mechanism for PI-binding domains in which initial membrane adsorption of the domain driven by nonspecific electrostatic interactions between the cationic protein surface and the anionic membrane surface is followed by specific PI-triggered membrane penetration of the domain. This study indicates that PtdIns(4,5)P 2 -mediated membrane binding of the PI3K-C2␣ PX domain follows essentially the same mechanism. Our monolayer data clearly indicate that PtdIns(4,5)P 2 binding by the PI3K-C2␣ PX domain is not a consequence of but a prerequisite for the penetration of the domain into the monolayer, the packing density of which is comparable with that of cellular membranes and large unilamellar vesicles (70 -72). The penetration of hydrophobic residues (Val 1490 and Leu 1491 ) located in the loop adjacent to the PtdIns(4,5)P 2 -binding pocket of PI3K-C2␣ PX is a specific PtdIns(4,5)P 2 -dependent process, because it was abrogated either by removal of PtdIns(4,5)P 2 in the monolayer or by mutation of the PtdIns(4,5)P 2 -coordinating residues, Arg 1488 , Arg 1493 , or Arg 1503 .  This study provides no direct information as to whether or not PtdIns(4,5)P 2 binding causes conformational changes of the PI3K-C2␣ PX domain. Structural studies of other PX domains have indicated, however, that PI binding causes local conformational changes. For example, a recent NMR study on the PX domain of Vam7p revealed that residues in its putative membrane binding loops underwent large changes in chemical shift when dodecylphosphocholine micelles were added to the PX domain-PtdIns3P complex (35). Also, crystal structures of the Grd19p PX domain in the absence and presence of a bound PtdIns3P showed conformational differences in the putative membrane attachment loop (46). This, in conjunction with the PI3K-C2␣ PX domain structure showing high flexibility of the PP II /␣2 loop, suggests that PtdIns(4,5)P 2 binding may also induce the conformational change and thereby facilitate the membrane penetration of hydrophobic side chains.
On the basis of qualitative vesicle pelleting assay, a recent study indicated that the C2 domain of PI3K-C2␣ has high affinity and specificity for PtdIns(4,5)P 2 and PtdIns(3,4)P 2 (30). Our quantitative SPR measurements confirm that the C2 domain of PI3K-C2␣ can indeed bind PtdIns(4,5)P 2 and PtdIns(3,4)P 2 ; however, the C2 domain lacks PI specificity and has about 20-fold lower affinity for PtdIns(4,5)P 2 -containing vesicles than the PX domain. The promiscuity and lower affinity of the C2 domain for PIs has been reported for other C2 domains from synaptotagmin I (73), JFC1 (74), and Rsp5p (75). Also, unlike the PX domain, the C2 domain of PI3K-C2␣ is not able to penetrate the lipid monolayer with the packing density comparable with that of cell membranes with or without PI. Furthermore, the isolated PX domain and the full-length PI3K-C2␣ have comparable affinity for PtdIns(4,5)P 2 -containing monolayers and bilayers. Thus, it is evident that the PX domain plays a more dominant role than the C2 domain in the PtdIns(4,5)P 2 -mediated membrane recruitment of PI3K-C2␣. It should be noted, however, that promiscuous PI specificity of the C2 domain may allow this domain to play a more significant role in the membrane recruitment of PI3K-C2␣ in response to PIs other than PtdIns(4,5)P 2 . It is also possible that the membrane recruitment of PI3K-C2␣ is mediated by both the PX domain and the C2 domain, interacting with PtdIns(4,5)P 2 and another PI, respectively. Interestingly, the C2 domain of PI3K-C2␣ harbors a nuclear localization sequence and associates with nuclear speckles (29). Further studies are needed to understand how the nuclear localization signal and the lipid-binding site of the C2 domain are related.
It is generally thought that specific subcellular localization of PI3K-C2␣ is critical for spatiotemporal regulation of the formation of 3Ј-phospho-PIs. Although this study clearly shows that the PX domain drives the binding of PI3K-C2␣ to PtdIns(4,5)P 2 -containing membranes in vitro, the role of the PX domain (and the C2 domain) in the subcellular localization of PI3K-C2␣ remains less clear. In particular, two recent reports indicate that protein-protein interactions play a primary role in subcellular localization of PI3K-C2␣. Domin et al. (76) reported that PI3K-C2␣ was enriched in clathrin-coated vesicles and trans-Golgi network of various mammalian cells but that neither the PX domain nor the C2 domain was necessary for this observed subcellular localization. Also, Gaidarov et al. (25) reported that clathrin bound the N-terminal region of PI3K-C2␣ and thereby activated the enzyme. This apparent discrepancy between the in vitro membrane binding properties and subcellular localization of PI3K-C2␣ is not totally surprising, however, as similar observations have been made for other peripheral proteins. For example, a study of all PH domains from the yeast genome showed that PI affinity, while contributing to membrane binding, does not specify subcellular localization (77). As a matter of fact, accumulating evidence shows that the subcellular localization of many peripheral proteins, including those proteins that were known to be recruited to the membrane primarily by protein-protein interactions, is orchestrated by a combination of protein-protein and lipid-protein interactions, and that the relative contribution of two types of interactions varies among proteins (1,78). It should also be noted that an interaction that appears to make a minor contribution in terms of the overall energetics of subcellular targeting of a protein may play a pivotal role in regulating its subcellular localization and activities (1,78).
For PI3K-C2␣, protein-protein interactions involving its N-terminal region may play a major role in subcellular localization, but the membrane interactions by the PX domain (and the C2 domain) should also contribute significantly. PtdIns(4,5)P 2 is present in significant concentrations in the plasma membrane (ϳ1 mol %) and internal membranes and can be locally enriched (79). In particular, clathrin-coated pits in the plasma membrane are known to have a high local PtdIns(4,5)P 2 concentration (80 -83). The affinity of the PI3K-C2␣ PX domain for PtdIns(4,5)P 2 is similar to that of other PtdIns(4,5)P 2 -binding domains, including the ENTH domain of epsin1 (66), which were shown to recruit their respective host proteins to the PtdIns(4,5)P 2 -rich region in the plasma membrane. Thus, it is reasonable to postulate that the PX domain of PI3K-C2␣ should make a contribution to the binding of PI3K-C2␣ to PtdIns(4,5)P 2 -rich cell membranes, including clathrin-coated pits, clathrin-coated vesicles, and possibly the trans-Golgi network. Another salient feature of the PX domain to be considered in terms of enzyme activation is the PtdIns(4,5)P 2 -mediated membrane penetration. It has been shown that triggering and sustaining the catalytic activity of many lipid-dependent enzymes, such as protein kinases C (84,85) and phospholipases (60), involves partial membrane penetration of their lipid-binding domains. Therefore, it is possible that the PtdIns(4,5)P 2mediated membrane penetration of the PX domain is a part of the activation process for PI3K-C2␣ and/or is necessary for the processive enzymatic action at the membrane. Additionally, the promiscuous C2 domain could also play a role of prolonging membrane residence of PI3K-C2␣, whereas the local concentration of 3Ј-phospho-PIs increases as a result of PI3K-C2␣ action. Undoubtedly, further studies are necessary to fully understand the relative importance of protein-protein versus lipid-protein interactions in the subcellular localization and activation of PI3K-C2␣.
In summary, this study elucidates the structural basis of specific PtdIns(4,5)P 2 recognition by the PI3K-C2␣ PX domain and the mechanism by which the PX domain interacts with PtdIns(4,5)P 2 -containing membranes. This, in conjunction with our previous work on other PX domains, shows that PX domains achieve unique modes of ligand and membrane binding through the variation of a limited number of nonconserved residues. This work may serve as a framework with which to systematically study the mechanism of specific membrane recruitment and regulation of PI3K-C2␣ as well as to comprehensively study the effects of structural and functional diversity of PX domains on lipid-protein and protein-protein interactions during cell signaling and membrane trafficking.