Intermolecular Interactions of the p85α Regulatory Subunit of Phosphatidylinositol 3-Kinase*

The regulatory subunit of phosphatidylinositol 3-kinase, p85, contains a number of well defined domains involved in protein-protein interactions, including an SH3 domain and two SH2 domains. In order to investigate in detail the nature of the interactions of these domains with each other and with other binding partners, a series of deletion and point mutants was constructed, and their binding characteristics and apparent molecular masses under native conditions were analyzed. The SH3 domain and the first proline-rich motif bound each other, and variants of p85 containing the SH3 and BH domains and the first proline-rich motif were dimeric. Analysis of the apparent molecular mass of the deletion mutants indicated that each of these domains contributed residues to the dimerization interface, and competition experiments revealed that there were intermolecular SH3 domain-proline-rich motif interactions and BH-BH domain interactions mediating dimerization of p85α bothin vitro and in vivo. Binding of SH2 domain ligands did not affect the dimeric state of p85α. Recently, roles for the p85 subunit have been postulated that do not involve the catalytic subunit, and if p85 exists on its own we propose that it would be dimeric.

The Class IA phosphatidylinositol 3-kinases (PI3K) 1 are heterodimeric enzymes with a p110 catalytic subunit and a p85 regulatory subunit (1). The p85 subunit is a multidomain protein comprising an amino-terminal SH3 domain, a BCR homology (BH) domain which has homology to the GTPase activating protein domain of the Break-point Cluster Region protein (2) and Rho subfamily GTPases, and two SH2 domains separated by an inter-SH2 domain through which p85 binds the catalytic subunit (3). The BH domain is flanked by two proline-rich motifs. To date, five isoforms of p85 have been identified. p85␣ has been cloned from bovine (4), human (5), and mouse (6) cDNA libraries, whereas only bovine p85␤ has been identified (4). Two splice variants of p85␣, termed p55 and p50, have been identified in the human (7), the rat (8,9), and the mouse (10). p55␣ lacks the SH3 and BH domains and the first proline-rich motif (PRM1) but retains the second proline-rich motif (PRM2) and has an amino-terminal extension of 34 amino acids. In p50␣, this extension comprises only 6 residues. To date, no splice variants of p85␤ have been identified. A variant known as p55␥ or p55 PIK has been cloned from bovine 2 and human (11) cDNA libraries and is homologous to p55␣, but no higher molecular mass isoforms of this protein have yet been identified.
PI3K has been implicated in a wide range of signaling pathways including those regulating proliferation and cell migration (12). The modular domains in p85␣ possess intrinsic signaling functions, in that they bind numerous intracellular ligands and mediate the formation of multiprotein complexes. The proline-rich motifs bind the SH3 domains of the Src family tyrosine kinases, Lyn and Fyn (13), whereas the small G proteins Rac (14,15) and Cdc42 (16) are potential PI3K regulators or effectors that bind PI3K, presumably via the BH domain. The SH2 domains of p85 bind phosphotyrosine (pY)-containing sequences from a range of receptor tyrosine kinases and docking proteins such as insulin receptor substrate 1 (17)(18)(19).
The roles of the various isoforms of the adaptor subunits of the Class IA PI3Ks are as yet undefined. Some isoforms, especially the truncated isoforms p55␣ and p50␣, have been shown to have restricted tissue distributions (9) compared with the 85-kDa isoforms. In addition, the truncated isoforms are clearly unable to interact with proline-rich motif-or SH3 domain-containing proteins or with small G proteins, and it has been shown that there is some selectivity in the recruitment of p85 isoforms by receptor tyrosine kinases (20). However, it is not yet understood whether the large number of isoforms of both subunits of PI3K represents a functional redundancy or reflects a form of signaling specificity. We have therefore undertaken a detailed study of the potential interactions of the individual domains of the p85 protein in order to elucidate their roles within the adaptor subunit as a whole.

MATERIALS AND METHODS
Peptides-Peptides were synthesized by Zinsser Analytic or Alta Biosciences and were of the following sequences: P1, KISPPTPKPRPPR-PLPVAPGPS; P2, WNERQQPAPALPPKPPKPT; Tyr-740/Tyr-751, GG-pYMDMSKDESVDpYVPML; Tyr-740, GGpYMDMSKDESVDYVPML; Tyr-751, GGYMDMSKDESVDpYVPML (where pY represents a phosphotyrosine residue). * The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
All cDNAs in pGEX-2T were expressed as glutathione S-transferase fusion proteins and purified by glutathione affinity chromatography according to the manufacturer's instructions. cDNAs in the baculovirus transfer plasmid pVL1393 were co-transfected into Sf9 cells with Bacu-loGold-linearized baculovirus DNA (PharMingen), whereas those in the transfer plasmid pBlueBac4 were co-transfected into Sf9 cells with Bac-N-Blue baculovirus DNA (Invitrogen). The recombinant baculovirus was plaque-purified and amplified as described previously (25). Exponentially growing Sf9 cells at a density of 1.5-2 ϫ 10 6 /ml were infected with recombinant baculoviruses at an multiplicity of infection between 10 and 20 and harvested 2-3 days post-infection. SV40-transformed monkey kidney cells (Cos7) were maintained in Dulbecco's modified Eagle's medium supplemented with 10% fetal calf serum, at 37°C and 10% CO 2 . Transient transfections were performed using DEAE-Dextran (Sigma). Cells (50% confluent) were transfected with 15 g of plasmid DNA per 15-cm tissue culture dish and harvested 48 h post-transfection.
Sf9 cells expressing recombinant p85 variants were lysed in Triton X-100 lysis buffer (50 mM Tris, pH 8.0, 150 mM NaCl, 1% Triton X-100, 5 mM dithiothreitol (DTT), and a range of protease inhibitors (50 g/ml 4-(2-aminoethyl)-benzenesulphenylfluoride hydrochloride, 16 g/ml benzamidine, 10 g/ml 1,10-phenanthroline, 10 g/ml aprotinin, 10 g/ml leupeptin, and 5 g/ml pepstatin A)). p85 variants were purified by affinity chromatography using a phosphotyrosine-affinity column that was synthesized by immobilizing 2 mg of phosphotyrosine (Sigma) per ml of Actigel (Sterogene) according to the manufacturer's instructions. Briefly, up to 50 ml of cell lysate was applied to a 14 ϫ 1-cm column of phosphotyrosine-Actigel equilibrated in Buffer A (20 mM Tris, pH 8.0, containing 5 mM DTT) and eluted with 50 ml of Buffer A followed by 50 ml of Buffer A containing 150 mM NaCl and 50 ml of Buffer A containing 2 M NaCl. Fractions of 2.5 ml were collected at a flow rate of 2.5 ml/min and exchanged into 3.5 ml of Buffer A using pre-packed Sephadex G-25 columns (PD10, Amersham Pharmacia Biotech). If a second purification step was required, fractions containing p85 variants were pooled and applied to a 0.46 ϫ 10-cm POROS 20 HQ column (Perspective Biosystems) equilibrated in Buffer A. Elution was carried out using 17 ml of Buffer A followed by a 75-ml linear gradient to 1 M NaCl in the same buffer. Fractions of 2.5 ml were collected at a flow rate of 2.5 ml/min.
Sedimentation Equilibrium-Analytical Ulracentrifugation (SE-AUC)-All SE-AUC experiments were carried out using an Optima XL-A analytical ultracentrifuge (Beckman) equipped with absorbance optics and an An60Ti rotor. Protein samples were buffer exchanged into SE-AUC buffer (50 mM Tris, pH 7.4, 50 mM NaCl, 7 mM ␤-mercaptoethanol, and 0.02% NaN 3 ) using either pre-packed Sephadex G-25 columns (NAP5, Amersham Pharmacia Biotech) or by dialysis overnight. Three samples of 110 l of protein at approximately 0.6, 0.3 and 0.15 mg/ml were analyzed, and the reference cells contained 125 l of SE-AUC buffer.
Experiments were performed at 4°C at an optimized range of speeds between 4000 and 28,000 rpm. Equilibrium data were collected at 280 nm in step scan mode with a radial increment of 0.001 cm between data points. Five readings were averaged each scan, from which a base-line scan taken at 360 nm was subtracted in order to correct for optical imperfections. Readings were taken at 8-h intervals until no difference could be detected between consecutive scans. The equilibrium distributions from three different loading concentrations and up to three rotor speeds were analyzed both individually and simultaneously using the Nonlin curve-fitting algorithm supplied with the ultracentrifuge (Beckman).
Binding Studies Using an Optical Biosensor-The procedure for measuring interactions between domains and peptides using the BIAcore biosensor (Biacore AB) has been previously described (26). Briefly, biotinylated peptides were captured on immobilized avidin, and then sample was injected in running buffer (20 mM Hepes, pH 7.4, 150 mM NaCl, 3.4 mM EDTA, 4 mM DTT, 0.005% Tween 20), and the response at equilibrium was recorded. For competition experiments, the response at equilibrium was compared with that obtained upon preincubation of the protein or isolated domain solution with increasing amounts of free peptide. In order to calculate IC 50 values, the results were plotted as resonance units versus peptide concentration, and the curve obtained was fitted with the equation R ϭ (R max /1 ϩ (C/IC 50 ) P ), where R is the response, R max is the response obtained in the absence of competitor, C is the concentration of competitor, and P is the Hill coefficient.
Protein in fractions from HP-SEC was precipitated using 10% trichloroacetic acid and analyzed by SDS-PAGE and Western blotting as described above. Proteins were precipitated from Sf9 cell lysates using GST fusion proteins captured on glutathione-Sepharose CL 4B (Amer- sham Pharmacia Biotech) and analyzed for lipid kinase assay as described below. Phosphatidylinositol 3-Kinase Assays-PI3K assays of glutathione-Sepharose CL 4B-precipitated proteins were carried out essentially as described previously (28). Lipid kinase assays contained 2 mM MgCl 2 , 1 mM ATP, 20 Ci of [␥-32 P]ATP, and 200 g/ml phosphatidylinositol. Extracted phospholipids were analyzed by thin layer chromatography in 65% 1-propanol, 0.7 M acetic acid, 50 mM phosphoric acid.

RESULTS
Interactions of the SH3 Domain and the Proline-rich Motifs of p85␣-It has previously been reported that both proline-rich motifs in p85␣ conform to the consensus ligand for the SH3 domain of p85␣ (29). In order to determine which of the two proline-rich motifs was the preferred ligand for the isolated p85␣ SH3 domain (p85␣SH3), we compared its ability to bind to peptides derived from the sequences of the first (P1) or second (P2) proline-rich motifs of p85␣, using an optical biosensor (Fig. 1A). Binding of p85␣SH3 to immobilized P1 was observed in this system, producing a response of greater than 400 resonance units (Fig. 1A), whereas little or no p85␣SH3 injected at the same concentration bound to P2 (Fig. 1A), suggesting that P1 is the preferred ligand for p85␣SH3.
In contrast, wild type p85␣, which contains the SH3 domain, did not bind immobilized P1 to a significant extent (Fig. 1B). The small amount of binding observed was equivalent to that observed for binding of the same concentration of p85␣ to a surface on which P1 was not immobilized (data not shown). A mutant of p85␣, in which two proline residues in PRM1 that have been shown to be important for p85␣ SH3 domain binding (30) were changed to alanine (p85␣⌬PRM1), was able to bind P1; however, a variant of p85␣ with mutations at the equivalent residues in PRM2 (p85␣⌬PRM2) was not. A variant of p85␣ in which both proline-rich motifs contained mutations (p85␣⌬PRM1:PRM2) bound to P1 to a similar extent as p85␣⌬PRM1 (Fig. 1B). Mutations in PRM1 therefore allowed p85␣SH3 to bind exogenous P1 peptide, whereas the SH3 domain in the context of wild type p85␣ could not bind to exogenous P1. Mutations in PRM2 did not affect binding to P1 peptide, confirming that the first proline-rich motif was the preferred ligand for the p85␣ SH3 domain. Thus, mutation of key residues in PRM1 of p85␣ not only affected this site but also the binding characteristics of the SH3 domain. These mutations increased the ability of the SH3 domain to bind exogenous ligands, indicating that in wild type p85␣ the SH3 domain interacts with PRM1 and that when this interaction is disrupted the SH3 domain is free to bind exogenous peptide.
Similarly, deletion of the SH3 domain of p85␣ would be expected to free PRM1 and allow it to bind exogenous SH3 domains. When wild type p85␣ was co-expressed with the p110 catalytic subunit in Sf9 cells, a p85␣SH3 GST fusion protein (GST-p85␣SH3) immobilized on glutathione-Sepharose CL4B was poorly able to co-precipitate PI3K lipid kinase activity from Sf9 cell lysates. However, a monoclonal antibody directed against the SH2 domain of p85␣ was able to immunoprecipitate PI3K activity; thus both p85␣ and p110␣ were expressed in these cells. The same antibody was able to immunoprecipitate PI3K activity from lysates of Sf9 cells co-infected with p110␣ and a mutant of p85␣ in which the SH3 domain had been deleted (p85␣⌬SH3). Immobilized GST-p85␣SH3 was able to co-precipitate the p110␣-p85␣⌬SH3 complex to a much greater extent than the p110␣-p85␣ complex (Fig. 1C); thus deletion of the SH3 domain increased the ability of an exogenous SH3 domain to bind PRM1, confirming that the SH3 domain and PRM1 are bound to each other in both p85␣ and the p110␣-p85␣ complex.
Intermolecular Interactions of p85 Isoforms-Examination of purified, recombinant p85␣ and p85␤ by HP-SEC under native conditions showed that they had apparent molecular masses that were approximately double (162 Ϯ 14 and 151 Ϯ 12 kDa, respectively) that expected from their amino acid sequences (Fig. 2, A and B). In contrast, the apparent molecular masses of the truncated p85␣ isoforms, p55␥ and p49␣ (82 Ϯ 4 kDa and 52 Ϯ 2 kDa), were much closer to their predicted molecular masses (55 and 49 kDa; Fig. 2, A and B).
There are several mechanisms by which a protein under native conditions can have an apparent molecular mass different from that predicted from its amino acid sequence. HP-SEC measures the hydrodynamic volume of a protein, which is de- PRM2 (--) proline-rich motifs to immobilized P1 peptide. C, precipitation of phosphatidylinositol kinase activity associated with p110␣/p85␣ or p110␣/p85␣⌬SH3 from Sf9 cell lysates using GST-p85␣SH3 or antibodies to p85␣. fined as the spherical volume occupied by that protein as it tumbles rapidly in solution. Non-spherical proteins often elute with an abnormally high apparent molecular mass compared with the globular proteins used to calibrate the column. Proteins can also interact nonspecifically with the column matrix, impeding their progress through the column and resulting in a later elution time, leading to an underestimate of the molecular mass of a protein. High apparent molecular masses may also result from dimerization or oligomerization of the protein; thus we investigated whether the higher than expected molecular mass of p85␣ was due to dimerization by determining the molecular masses of p85␣ and p49␣ under different conditions. HP-SEC was carried out in buffers containing either 5 mM DTT or 8 M urea in order to determine whether the putative p85␣ dimeric interaction could be disrupted. The apparent molecular masses of p85␣ and p55␥ were unaltered in the presence of 5 mM DTT (Fig. 2B), suggesting that dimerization of p85␣ was not due to the presence of an intermolecular disulfide bond. In contrast, p85␣ had an apparent molecular mass of 89 Ϯ 5 kDa in the presence of 8 M urea. High concentrations of urea disrupt non-covalent, but not covalent, bonds, indicating that p85␣ forms a dimer via a non-covalent interaction. The apparent molecular mass of p49␣ was relatively unaffected in the presence of 8 M urea however (Fig. 2B), suggesting that it is monomeric under native conditions. Sedimentation Equilibrium-Analytical Ultracentrifugation (SE-AUC) was employed to determine the apparent molecular masses of a number of p85␣ variants, as the equilibrium distribution of a solute in a gravitational field is dependent on the molecular mass of the solute, but independent of its shape, whereas HP-SEC is dependent on both mass and shape. Analysis of the equilibrium distribution of the concentration of protein with respect to radial position in the centrifuge demonstrated the occurrence of self-association in some samples. Self-association was defined as a lack of adherence to the Lamm equation, which describes the distribution of ideal, nonassociating solute particles in a gravitational field. The occurrence of self-association manifests as a non-random distribution of residuals to the fit of the experimental (absorbance versus radius) data to a derivative of the Lamm equation.
The equilibrium distribution of p85␣ did not fit that of an ideal, non-associating solute, as the residuals were non-random (Fig. 3D). The apparent molecular mass of p85␣ was dependent on protein concentration and ranged between 80 and 110 kDa (Fig. 3D), suggesting that p85␣ exists as an equilibrium of monomers and dimers under these conditions. In contrast, p49␣ was monomeric under the conditions examined. The residuals of the fit for an ideal, non-associating solute were distributed randomly around zero, and p49␣ had an apparent molecular mass of approximately 47 kDa (Fig. 3C). Thus, p85␣,  ). B, summary of data from HP-SEC of recombinant p85 isoforms and p85␣ deletion mutants and a diagrammatic representation of the domains in each p85 isoform is indicated. The predicted molecular mass for each isoform (in kDa) was calculated from its amino acid sequence. Molecular masses determined by HP-SEC were calculated from the retention time (min) of the maximum height of the peak for each recombinant protein.
The retention time was converted to an apparent molecular mass by comparison to a plot of log molecular mass (kDa) versus the retention time (min) of the maximum height of the peak for a range of standard globular proteins chromatographed using the same buffer conditions. Apparent molecular masses are denoted as the mean of at least three measurements Ϯ 1 S.D. Where only one measurement was taken, an approximate molecular mass is indicated. For HP-SEC of transfected Cos7 cytosolic extracts, a range of apparent molecular masses is calculated by converting the fraction numbers that are positive for the appropriate band by Western blotting into retention time (min) and therefore into apparent molecular mass (kDa) as described above.
which was apparently dimeric by HP-SEC, fitted a model for self-association, although the equilibrium dissociation constant for p85␣ dimer formation (estimated from the plot of apparent molecular mass versus protein concentration; Fig. 3D) was in the micromolar range, which is a lower affinity than that expected for a protein that was determined to be constitutively dimeric by HP-SEC. Given that p49␣ was shown to be monomeric using both techniques, this suggested that the dimerization of p85 was mediated by domains that are not present in p49␣.

An Intermolecular Interaction between the SH3 Domain and PRM1 Contributes to the Dimerization of p85␣-
The observation that p85␣ and p85␤ were dimeric under native conditions, but p55␥ and p49␣ were not, suggested that the amino-terminal half of the higher molecular mass regulatory subunits was involved in intermolecular interactions that caused them to dimerize. Given that the SH3 domain and the first proline-rich motif reside in this region of the protein, and interact with each other (Fig. 1), we investigated whether this interaction was involved in the dimerization of p85␣. Indeed, expression of the amino-terminal half of p85␣ on its own, either with or without one of the SH2 domains (p85␣SH3-BH-SH2 and p85␣SH3-BH; Fig. 2B), resulted in proteins with apparent molecular masses by HP-SEC (101 Ϯ 10 and 53 Ϯ 4 kDa, respectively) that were still approximately double those predicted (50 and 38 kDa, respectively). This confirms that the amino-terminal half of p85␣ contains all the amino acid residues required for dimerization.
When the SH3 domain of p85 was deleted, the apparent molecular mass of the resulting protein (p85␣⌬SH3) determined by HP-SEC under native conditions was 139 Ϯ 7 kDa, which was approximately double that predicted from its amino acid sequence (74 kDa) (Fig. 2B). This suggested that removal of the SH3 domain did not disrupt the dimer interface to a significant extent. Similarly, mutation of two proline residues (Pro-96 and Pro-99) in the first proline-rich motif to alanine (p85␣⌬PRM1, Fig. 2B) did not convert p85␣ to a monomeric species. Deletion of the remaining domain within the aminoterminal portion of p85␣, the BH domain (p85␣⌬BH), also did not disrupt dimerization, as its molecular mass by HP-SEC (145 Ϯ 11 kDa) was still approximately double that predicted (64 kDa). In contrast, deletion of both the SH3 and BH domains, as is the case for both p55␥ and p49␣, had already been shown to result in a monomeric protein. Therefore, the dimerization interface resides in the amino-terminal half of the molecule, either in the SH3 or BH domains or the intervening regions.
A construct comprising just the SH3 domain and PRM1 (p85␣SH3-PRM1, Fig. 2B) had an apparent molecular mass by HP-SEC (23 Ϯ 2 kDa) approximately double that predicted (12 kDa), suggesting it was dimeric. Like p85␣, the equilibrium sedimentation characteristics of p85␣SH3-PRM1 were consistent with that of a protein undergoing self-association (Fig. 3B). The residuals of the fit for an ideal, non-associating solute were non-random, and the apparent molecular mass of p85␣SH3-PRM1 was dependent on protein concentration and ranged between 16 and 22 kDa, with an equilibrium dissociation constant for dimer formation in the micromolar range (Fig. 3B). In contrast, the SH3 domain alone was unable to dimerize, as it had a similar apparent molecular mass by HP-SEC to that predicted from its amino acid sequence (13 Ϯ 1 and 9.7 kDa, respectively).
The apparent molecular mass of the isolated BH domain was determined by HP-SEC (32 Ϯ 2 kDa) and was approximately 1.5 times that predicted (22 kDa). The p85␣BH also appeared to self-associate by SE-AUC, although to a lesser degree than for p85␣SH3-PRM1 (Fig. 3A). The estimated equilibrium dissociation constant for dimer formation was in the millimolar range.
The interaction surface for dimerization therefore does not reside in a single domain of the amino terminus of p85␣, as neither the SH3 nor BH domain alone was observed to dimerize fully. Moreover, deletion of either domain or mutation of PRM1 did not generate monomeric p85␣ (Fig. 2B). The interaction surface involved in dimerization must therefore involve residues that are widely distributed in the primary structure of p85␣, so that no single domain self-associates to a significant degree, but are positioned in three-dimensional space in the folded protein so that they cooperate to form a single binding surface. In contrast, p85␣SH3-PRM1 was able to self-associate, and the estimated equilibrium dissociation constants for dimerization of p85␣SH3-PRM1 and p85␣ were similar, suggesting that SH3-PRM1 binding was intermolecular and contributed to dimerization. However, the contribution of the BH domain to dimerization of full-length p85␣ did not allow us to discriminate between an inter-or intramolecular SH3-PRM1 interactions in the whole protein.
If the interaction between the p85␣ SH3 domain and the first proline-rich motif was intermolecular and contributed to dimerization, exogenous P1 peptide should compete for this interaction and result in the formation of monomeric p85␣. The affinities for self-association as determined by SE-AUC were lower than the apparent affinities suggested by the molecular masses of the p85␣ variants as determined by HP-SEC, thus competition experiments were performed using SE-AUC. In the presence of a 20-fold molar excess of P1 peptide, both p85␣SH3-PRM1 and wild type p85␣ behaved as ideal, non-associating solutes, and their apparent molecular masses were not dependent on protein concentration (Fig. 4, A and B), suggesting that P1 peptide competed for binding in a site involved in dimer formation. However, the apparent molecular mass of p85␣ bound to P1 peptide was greater than expected, suggesting that either a small degree of self-association was mediated by the BH domain or that the high concentration of P1 in the analyte solution increased its viscosity, resulting in an artificially high estimation of apparent molecular mass. An intermolecular interaction between the p85␣ SH3 domain and the p85␣ PRM1 therefore seems to contribute to p85␣ dimerization, although the BH domain also contributes residues to this interface.
Dimerization of p85␣ in Vivo-In order to confirm whether p85␣ dimerization observed in vitro also occurred in vivo, p85 isoforms were transfected into Cos7 cells and Western blot analysis of HP-SEC fractions of whole cell lysates revealed immunoreactive bands in fractions covering a relatively broad range of molecular masses. p85␣ was detected in fractions 36 -38, which corresponded to a molecular mass range of 93-176 kDa (Figs. 5A and 2B). Although this was a broad range, the majority of the p85␣ eluted at a time that corresponded to a molecular mass greater than the monomeric molecular mass for p85␣ (83 kDa), and there was a population at a dimeric molecular mass (176 kDa). p85␣ therefore is dimeric both as a homogeneous, purified recombinant protein and within the context of a cellular environment, when it is at a relatively low concentration and in a protein-rich environment, which are conditions that would be expected to reduce nonspecific self-association.
In contrast, p55␥ and p49␣ were detected in fractions that corresponded to molecular mass ranges of 40 -61 and 26 -49 kDa, respectively, in this system (Figs. 5, B and C, and 2B). The highest molecular mass populations for both p55␥ and p49␣ FIG. 5. p85␣ from transfected Cos7 cells behaves as a dimeric complex upon size-exclusion chromatography and co-precipitates with p85␣⌬BH but not with p49␣. A-C, cytosolic extracts prepared from unstimulated Cos7 cells, transfected with p85␣ (A), p55␥ (B), or p49␣ (C), were subjected to size-exclusion chromatography. Aliquots of each fraction were analyzed by Western blotting with an appropriate mouse monoclonal antibody (lower strip of each panel A, B, and C). Absorbance at 220 nm (O) and the corresponding elution volumes of the molecular mass standards (kDa) are indicated, as is the void volume (V o ). The Superose-12 column was equilibrated and run in TBS at 0.3 ml/min, and 0.3-ml fractions were collected and analyzed by SDS-PAGE followed by Western blotting with the appropriate antibody. Unfractionated cytosolic extracts were analyzed in parallel (Lys) with the relevant antibody. D and E, Nonidet P-40 lysates were prepared from untreated Cos7 cells transfected with Myc epitope-tagged p85␣ and/or p85␣⌬BH (D) or Myc epitope-tagged p85␣ and/or p49␣ (E). Myc epitope-tagged p85␣ was immunoprecipitated using an anti-Myc tag monoclonal antibody (9E10) and immunoblotted for co-precipitated p85␣⌬BH or p49␣ using U14, a mouse monoclonal antibody directed against the SH2 domain of p85␣ which recognizes p85␣, p85␣⌬BH, or p49␣. The relative migration positions of the various isoforms on SDS-PAGE are indicated.
were not greater than their predicted monomeric molecular masses; therefore they are apparently monomeric within a cellular context. Interestingly, p55␥ and p49␣ seemed to bind to other intracellular proteins from Cos7 cells and form high molecular mass (Ͼ300 kDa) complexes. In addition, two variants of p55␥ were observed, but the origin of these was not investigated further.
cDNAs encoding an amino-terminally Myc epitope-tagged version of p85␣, untagged p85␣⌬BH, or hexa-histidine-tagged p49␣ were transiently transfected into Cos7 cells, alone or in combination. Precipitation with appropriate antibodies immobilized on protein G-Sepharose or affinity matrices showed that each protein was expressed in Cos7 cells transfected with a single cDNA (Fig. 5D, lanes 2 and 3 and Fig. 5E, lane 2). When Myc-tagged p85␣ and p85␣⌬BH, a deletion mutant of p85 that still had the ability to dimerize in vitro (Fig. 2B), were cotransfected, immunoprecipitation with a monoclonal antibody directed against the Myc epitope was able to co-immunoprecipitate untagged p85␣⌬BH (Fig. 5D, lane 1). The difference in molecular mass between these two forms of p85␣ allowed them to be resolved by SDS-PAGE. In contrast, p49␣, a deletion mutant of p85 that did not have the ability to dimerize in vitro (Fig. 2B), was unable to co-immunoprecipitate with Myc-tagged p85␣ (Fig. 5E, lane 1); therefore, forms of p85␣ that have the ability to dimerize in vitro can form mixed heterodimers in vivo, whereas monomeric forms cannot.
Binding of Phosphopeptides to the SH2 Domains Does Not Affect the Oligomerization States of p85␣ Variants-Binding of a diphosphotyrosine-containing peptide that mimics an activated platelet-derived growth factor ␤-R (Tyr-740/Tyr-751) to the SH2 domains of p85␣ affects the oligomerization state of the p110␣-p85␣ complex 3 ; therefore, we investigated the effect of Tyr-740/Tyr-751 binding on dimerization of p85␣ isoforms. The IC 50 for competition of either Tyr-740/Tyr-751 or peptides with the same amino acid sequence but with only one phosphotyrosine residue (Tyr-740 or Tyr-751) with p85 variants for binding to immobilized Tyr-751 was compared. The IC 50 for the competition of Tyr-740, Tyr-751, or Tyr-740/Tyr-751 with the isolated carboxyl-terminal SH2 domain of p85␣ (p85␣ cSH2) for binding to immobilized Tyr-751 was similar, although Tyr-740 had a slightly lower affinity for p85␣ cSH2 compared with the other peptides (Fig. 6A). The binding characteristics of p85␣ and p49␣ in this system were very similar (Fig. 6, B and C). Both p85␣ and p49␣ had lower IC 50 values for the competition of Tyr-740/Tyr-751 for binding to Tyr-751 compared with either mono-phosphopeptide, suggesting that both SH2 domains engage both phosphotyrosine residues within this peptide (Fig. 6,  B and C).
The apparent molecular masses of p85␣ and p49␣ were determined at a range of concentrations by HP-SEC (Fig. 6D) and agreed with those previously described (Fig. 2B). There appeared to be a slight increase in apparent molecular mass of p85␣ with increasing protein concentration (Fig. 6D), supporting the idea that p85␣ can undergo self-association, whereas p49␣ cannot. Although Tyr-740/751 was shown to bind p85␣ or p49␣ with high affinity (Fig. 6, B and C), addition of Tyr-740/ 751 did not affect the apparent molecular mass of either p85␣ or p49␣, regardless of protein concentration (Fig. 6D); therefore

DISCUSSION
The p85 subunit of PI3K may represent yet another intracellular signal transduction protein which utilizes dimerization as a regulatory mechanism, in a similar manner to cell-surface receptors, which are often activated by ligand-induced dimerization or oligomerization (31). Several other intracellular signaling proteins, such as the STAT transcription factors (32) and c-Raf (33,34), have also been shown to be regulated by dimerization.
In this study, a number of techniques have been used to demonstrate that the amino-terminal portion of the p85␣ subunit of PI3K binds to itself in an intermolecular manner (Fig.  7), which leads to dimerization of p85␣ both in vitro and in vivo. Additionally, we determined that the isolated SH3 domain of p85␣ binds only one of its endogenous proline-rich motifs, PRM1, and the same interaction occurs within the whole protein. In order to determine whether these two interactions were related, we further investigated the binding properties of each of the domains of p85␣. However, it became clear that dimerization was not mediated by a single domain, and the involvement of the BH domain in dimerization made interpretation of this data more complicated. However, the ability of the P1 peptide to disrupt both p85␣SH3-PRM1 and p85␣ dimers (Fig.  4) indicated that the SH3 domain of one p85␣ molecule binds the PRM1 of a second p85␣ (Fig. 7).
Previous studies have also demonstrated that various fragments of p85␣ are dimeric. A fragment of p85␣ encompassing residues 1-101 (similar to p85␣SH3-PRM1 in this study) (35) and a fragment similar to p85␣SH3-BH (36) have been shown to be dimers. Crystals of the BH domain of p85␣ have previously been shown to contain two monomers per asymmetric unit (36); however, the hydrophobic dimerization interface was very small, involving only four residues of each monomer. In comparison, the three-dimensional structure of the p85␣SH3 domain has been determined by both x-ray crystallography and NMR, and no evidence for self-association of the isolated SH3 domain of p85␣ has been described (30,37).
The suggestion from the crystal structure of the BH domain that the 4-residue interface represents a small portion of a larger interaction (36) is confirmed by the contribution of the BH domain to the overall dimeric interface of p85␣ (Fig. 2B). The size of the interface shown in the crystal structure suggests that the affinity of dimerization would be low. If monomeric and dimeric BH domain existed in equilibrium, and the interconversion rate was fast, the low affinity for self-association would result in only a small proportion of dimeric BH domain and thus may explain the slightly higher average molecular mass observed by both HP-SEC and SE-AUC (Figs. 2B and 3A).
Interactions between SH3 domains and proline-rich motifs have been reported to regulate several intracellular signaling molecules. In p85␣, this interaction not only participates in the dimerization interface but may also block the binding of exogenous ligands for the p85␣ SH3 domain and PRM1. Stimulation of the B cell receptor has been shown to lead to the binding of the SH3 domains of the Src family tyrosine kinases, Lyn and Fyn, to the p85␣ PRM1 and the up-regulation PI3K activity (13). The endogenous, intermolecular interaction may therefore also regulate the binding of exogenous ligands to the SH3 domain and PRM1. A similar mechanism has been demonstrated in Itk, a member of the Tec family of cytoplasmic tyrosine kinases (38), although the SH3 domain proline-rich motif interaction was intramolecular in this protein. Interaction of the SH3 domain with an adjacent proline-rich motif in Itk prevented the binding of the SH3 domain to proline-rich motifs in Sam-68 and Grb2. Tyrosine phosphorylation within the SH3 domain of another Tec family tyrosine kinase, Btk, was shown to disrupt an intramolecular SH3 domain prolinerich motif interaction and release the Btk SH3 domain to allow it to recruit substrates of the Btk kinase domain (39). Intramolecular SH3 domain binding to proline-rich motifs may therefore be a general mechanism for the regulation of SH3 domain function, although there seems to be a number of ways in which the SH3 domain can be released to bind its exogenous ligands. The interaction of the p85␣ SH3 domain and p85␣ PRM1 seems to be a special case of this general mechanism in which an intermolecular SH3-PRM1 interaction occurs between the two units of a dimer.
Recent evidence has suggested that the p85 subunit may have functions that are independent of the p110 catalytic subunit. A truncated form of p85␣, lacking the carboxyl-terminal SH2 domain and 52 residues of the inter-SH2 region, was identified as a potential oncogene in a number of murine lymphomas (40). Overexpression of this mutant protein (p65) caused transformation of NIH-3T3 fibroblasts, and p65 and v-Raf had synergistic transforming activities. Murine embryonic fibroblasts in which the gene for p85␣ was deleted by homologous recombination had a defect in a p53-mediated apoptotic pathway (41). Apoptosis caused by oxidative stress was reduced in cells lacking p85␣. Apoptosis was not inhibited by the PI3K inhibitor, wortmannin, suggesting that the lipid kinase activity of the p110 subunit is not a requisite component of the p53-and p85␣-mediated pathways.
One of the best defined functions of the regulatory subunit of PI3K is the down-regulation of p110-mediated lipid kinase activity (42,43), suggesting that populations of PI3K comprising a complex of the p110 and p85 subunits or the p110 subunit alone would have different activities. A 100-kDa PI3K activity has been detected in lysates of bovine thymus by size-exclusion chromatography (SEC) (44) suggesting that the p110 subunit may exist on its own and implying that the p85 regulatory subunit may also exist as a separate entity with a biological function distinct from its role in the regulation of p110 activity. Another potential role of a pool of dimeric p85 in vivo could be to bind and stabilize the p110 catalytic subunit as soon as it is translated, as p110␣ has been shown to be prone to inactivation and degradation at 37°C without the bound regulatory subunit (42).
The existence of forms of the regulatory subunit of PI3K, such as p49␣ or p55␥, that do not have the potential for selfassociation may represent a mechanism for divergent signaling. PI3K-mediated signaling through the insulin receptor and insulin receptor substrate 1 has been shown to preferentially utilize forms of the PI3K regulatory subunit that do not have the potential for self-association (8,20). As yet it is unclear what implication a lack of dimerization of some regulatory subunit isoforms has for PI3K signaling. The use of proteinprotein interactions of SH3 and BH domains and proline-rich motifs to regulate the nature and localization of PI3K activity would not be an option for the lower molecular mass isoforms. The regulatory subunit of PI3K may therefore represent both an adaptor for the catalytic subunit and a mediator of other protein-protein interactions that utilize these domains. This study has highlighted that these same domains are also implicated in self-association of the p85 protein, which adds a further layer to the potential complexity of the regulation of PI3K activity in cell signaling events.