Overlapping Binding Sites in Protein Phosphatase 2A for Association with Regulatory A and (cid:1) -4 (mTap42) Subunits*

Diverse functions of protein Ser/Thr phosphatases depend on the distribution of the catalytic subunits among multiple regulatory subunits. In cells protein phosphatase 2A catalytic subunit (PP2Ac) mostly binds to a scaffold subunit (A subunit or PR65); however, PP2Ac alternatively binds to (cid:1) -4, a subunit related to yeast Tap42 protein, which also associates with phosphatases PP4 or PP6. We mapped (cid:1) -4 binding to PP2Ac to the helical domain, residues 19–165. We mutated selected residues and transiently expressed epitope-tagged PP2Ac to assay for association with A and (cid:1) -4 subunits by co-precipitation. The disabling H118N mutation at the active site or the presence of the active site inhibitor microcystin-LR did not interfere with binding of PP2Ac to either the A subunit or (cid:1) -4, showing that these are allosteric regulators. Positively charged side chains Lys 41 , Arg 49 , and Lys 74 on the back surface of PP2Ac are unique to PP2Ac, compared with phosphatases PP4, PP6, and PP1. Substitution of one, two, or three of these residues with Ala produced a progressive loss of binding to the A subunit, with a corresponding increase in binding to (cid:1) -4. Conversely, mutation of Glu 42 in PP2Ac essentially eliminated PP2Ac binding to (cid:1) -4, with an increase in binding to the A subunit. Recip-rocal changes in binding by phosphorylation with mitogen-activated protein kinase. The assay is linear when (cid:3) 30% of the total substrate is hydro- lyzed, and all results were within this range. Parallel samples were assayed for 2 h at 30 °C with and without 50 n M okadaic acid. Samples were precipitated with 15% trichloroacetic acid and centrifuged, and the supernatants were analyzed by scintillation counting. The differ- ence in acid-soluble 32 P released (cid:4) okadaic acid was defined as the PP2A activity. Results were replicated in two or three independent experiments, and the average activities were reported. PP2Ac PP1c— full-length PP2Ac sequence submitted to the SWISS-MODEL base (www. sequence alignment of PP1 shows 49% overall amino acid sequence identity. program used to view and manipulate the three-dimensional structure is called Weblab Viewerpro (Molecular Simulations, Inc., San Diego, CA). in by immunoprecipitation, and assayed for activity by hydrolysis of 32 P-myelin basic protein 50 n M okadaic acid as under

Protein Ser/Thr phosphatases of the type-1 and type-2A classes are essential enzymes similar in terms of structure and catalytic mechanism but with distinctive and multifaceted roles in eucaryotic cells (for reviews, see Refs. 1 and 2). The catalytic subunits have an extraordinarily conserved central domain of ϳ250 residues that forms a bimetallic active site for phosphoester hydrolysis. This active site of both PP1 1 and PP2A binds various natural toxins such as microcystin, okadaic acid, calyculin A, and cantharidin that act as potent inhibitors (3,4). Differences in potency among inhibitors for PP1 and PP2A have been attributed to sequence differences in the ␤-12-13 loop at the edge of the active site (5,6). The threedimensional structures of the PP1 catalytic domain with the inhibitor microcystin (7) or calyculin A (8) have been determined. The N-and C-terminal segments outside the central catalytic domain were not visualized in the structure, probably because of their flexibility. These segments diverge in sequence between type-1 and type-2A phosphatases and also among the isoforms of each type (9,10). No structure of PP2A is yet available. Both phosphatases are phosphorylated in conserved sites in their C termini, PIpTP in PP1 (11)(12)(13) and DpYFL in PP2A (14 -18), with reduction in phosphatase activity. In addition, the C-terminal Leu 309 in PP2A undergoes reversible methylesterification, which affects association with regulatory subunits (19,20).
Association with multiple regulatory subunits is considered the basis for achieving diversity of biological functions and modulation of enzymatic activity for both PP1 and PP2A. The monomeric catalytic subunits are highly active and exhibit relatively broad and overlapping reactivity with phosphoproteins (21). Regulatory subunits restrict and control activity, target the phosphatase to structures such as glycogen particles or myofibrils, and also serve as scaffolds for substrates and kinases (22). Thus, one key to understanding PP1 and PP2A is to understand their distribution among regulatory subunits. The PP1 catalytic subunit directly binds to many different regulatory subunits. A RVXF motif common to most PP1 regulatory subunits is the primary basis for recognition (23)(24)(25). In addition PP1 is selectively inhibited with nanomolar potency by phosphoproteins, such as inhibitor-1, DARPP-32, CPI-17, and inhibitor-2 (26,27). Recent results show that PP1 can bind both a regulatory subunit and an inhibitor protein simultaneously (28 -32). In contrast to PP1, most PP2A binds directly to one regulatory subunit, called A or PR65. In this AC dimer the A subunit functions to alter the kinetics of C, in particular reducing the catalytic rate over 20-fold (33,34), and also acts as a molecular scaffold to mediate binding of various regulatory B subunits to form different ABC trimeric PP2A. There are more than a dozen B subunits in three different families, providing considerable diversity (35)(36)(37)(38). Tumor antigens such as SV40 small t and polyoma middle T bind to the AC dimer in place of the B subunits (39 -41). Mutations in the A subunit gene ap-pear in lung and colon cancers, underscoring the importance of this subunit in the function of PP2A (42).
Other more recent results have expanded the distribution of the A and PP2Ac subunits to other partners. The A subunit was found by two-hybrid analysis to associate with HSF2, a heat shock transcription factor (43). Binding of HSF2 to the A subunit only occurred when the PP2Ac was displaced, posing alternative functions for the A subunit protein separate from PP2A. The A subunit also was reported to bind to the TPR repeats in PP5, a protein phosphatase with a sequence related to both PP1 and PP2A (44). Immunoprecipitations indicated that PP2Ac simultaneously bound to both the A subunit and eRF1, a release factor for translation termination (45). Similarly, either PP2Ac or AC dimer associates with the cytoplasmic domains of CD28 and CTLA-4 receptors (46). The PP2Ac also directly binds to the protein ␣-4, which is related in sequence to the yeast protein Tap42 that binds yeast phosphatases Sit4 or Pph21/22 and functions downstream of target of rapamycin (TOR) in a signaling pathway sensitive to nutrient availability (47). ␣-4 was first cloned using a monoclonal antibody against a phosphoprotein associated with the ␣ chain of the B cell receptor complex (48,49). Binding of ␣-4 to PP2Ac occurs with displacement of the A subunit, so the two regulatory subunits are mutually exclusive (50). However, unlike the A subunit that binds only to PP2Ac, the ␣-4 subunit also binds the related phosphatases PP4 and PP6 (51).
We sought to define surface determinants in PP2Ac for binding to the A subunit and ␣-4 that would account for specific recognition but mutually exclusive association. Recombinant forms of PP2Ac (and PP1c) have been notoriously difficult to express, and recombinant catalytic subunits exhibit properties different from the native enzymes (52). Therefore, using binding assays with purified subunits was not a feasible approach. Besides, it may be more informative to assay for subunit association in living cells, with the other partners for each of the subunits present and competing for interaction. Transient expression of PP2Ac in mammalian cells is under stringent regulation at transcriptional and translational levels (53,54), limiting the amount of ectopic protein expressed. This affords conditions where ectopic PP2Ac is expressed at low levels relative to endogenous PP2A. Here we used expression of wild type and mutated epitope-tagged PP2Ac and co-immunoprecipitation with endogenous subunits or co-expressed tagged subunits.

MATERIALS AND METHODS
Yeast Two-hybrid Screen-Full-length murine ␣-4 was cloned into the BamHI/EcoRI site of the pGBT10 vector (donated by Ian Macara, University of Virginia). The pGBT10 is a derivative of pGBT9 containing the GAL4 DNA binding domain. The library was 9-day mouse embryo cDNA size-selected for fragments of 350 -700 nucleotides that were inserted into the NotI site in pVP16, downstream of the GAL4 activation domain (obtained from Stan Hollenberg, Fred Hutchinson Cancer Center, Seattle, WA). The two-hybrid screen was done in a sequential manner with ϳ6 ϫ 10 6 independent clones screened by plating onto synthetic medium (Difco) containing 10 mM 3-aminotriazole (Sigma) and lacking the amino acids histidine, leucine, and tryptophan. Positives were analyzed by growth on selective media and induction of ␤-galactosidase using filter assays. Clones were rescued by electroporation into Escherichia coli HB101 that was grown on plates with a medium containing M9 salts lacking leucine. The positive clones were subjected to DNA sequencing and transformation tests. Chemicals were from Sigma unless otherwise noted.
Plasmids and Mutagenesis-Full-length DNA for the human PP2A wild type catalytic subunit ␣ isoform and the A subunit ␣ isoform was amplified by PCR from HeLa cDNA generated by ThermoScript poly(dT) reverse transcription-PCR (Invitrogen) following the manufacturer's protocol. The PP2Ac insert was ligated into the BamHI/EcoRI sites in the mammalian expression vector pKH3, and the PP2A-A insert was cloned in-frame with the FLAG epitope at the BamHI/EcoRI sites of a pcDNA3-FLAG vector. Preparation of the (HA) 3 -PP2Ac(H118N) and FLAG-␣-4 was described previously (55). All mutants of PP2Ac were made using the QuikChange mutagenesis kit (Stratagene, La Jolla, CA) following the manufacturer's protocol.
Cell Culture and Transfection-COS7 cells were grown in Dulbecco's modified Eagle's medium containing 10% newborn calf serum at 37°C in humidified 5% CO 2 atmosphere. Cells at 60% confluence in either 60or 100-mm dishes were transiently transfected and incubated overnight. Single transfections used 2-5 g of plasmid DNA, and co-transfections used 2-4 g of the DNA of each plasmid with FuGENE 6 (Roche Applied Science) as the delivery method.
Phosphatase Assay-COS7 cells were transfected with different (HA) 3 -PP2Ac constructs or empty vector as a control. After 24 h lysates were prepared, and 500 g of total protein was used for immunoprecipitation with 3 g of anti-HA antibody. The precipitates were recovered, washed by centrifugation, and suspended in phosphatase assay buffer (20 mM MOPS, pH 7.5, 50 mM NaCl, 1 mM MgCl 2 , 1% 2-mercaptoethanol, 1 mM dithiothreitol, 10% glycerol, and 0.1 mg/ml bovine serum albumin). The suspended beads were split, one portion for assay and one for immunoblotting with ␣-HA to show the amount of (HA) 3 -PP2Ac in the assay. The assay used 1 M 32 P-myelin basic protein as substrate, prepared by phosphorylation with mitogen-activated protein kinase. The assay is linear when Ͻ30% of the total substrate is hydrolyzed, and all results were within this range. Parallel samples were assayed for 2 h at 30°C with and without 50 nM okadaic acid. Samples were precipitated with 15% trichloroacetic acid and centrifuged, and the supernatants were analyzed by scintillation counting. The difference in acid-soluble 32 P released Ϯ okadaic acid was defined as the PP2A activity. Results were replicated in two or three independent experiments, and the average activities were reported.
Protein Threading of PP2Ac Structure onto PP1c-The full-length PP2Ac sequence was submitted to the SWISS-MODEL data base (www. expasy.ch/swissmod/SWISS-MODEL.html). The sequence alignment of PP2A versus PP1 shows 49% overall amino acid sequence identity. The program used to view and manipulate the three-dimensional structure is called Weblab Viewerpro (Molecular Simulations, Inc., San Diego, CA).

Location of Potential Regulatory Subunit Binding Sites in
PP2Ac-Using the ␣-4 (mTap42) protein as bait in yeast twohybrid screening yielded 20 clones of the catalytic subunit of PP2A (PP2Ac) out of 65 total positives from a mouse embryo cDNA library. Both the ␣ and ␤ isoforms of PP2Ac were recovered, and a summary of the PP2Ac clones is depicted in Fig. 1a, showing that residues 19 -165 were common to all of the PP2Ac proteins that interacted with ␣-4. This region of the phosphatase catalytic subunit corresponds to a compactly folded, ␣-helical domain seen in the structure of the closely related phosphatase PP1 (7). We aligned the sequence of residues 19 -165 of PP2Ac with the corresponding regions of PP4c and PP6c and compared them with phosphatases PP1c and PP2B, also known as calcineurin (Fig. 1b). The alignment exposed residues conserved among all the phosphatase catalytic subunits. These included acidic side chains known to bind active site metal ions, several glycines needed for protein folding, and His 118 (marked with an asterisk in Fig. 1b) in the conserved RGNHE motif that is essential for phosphatase catalytic function.
The binding of ␣-4 and A subunits to PP2Ac is known to be mutually exclusive, suggesting that they use overlapping sites on the ␣-helical domain. However, only PP2Ac binds to the A subunit, not PP4c or PP6c; therefore we expected that such a binding site would involve residues that appear uniquely in PP2Ac and have different chemical properties in PP4c or PP6c. We located six such residues (Fig. 1b), and four of these had basic side chains, namely Lys 41 , Arg 49 , Lys 74 , and Arg 108 (marked with a plus sign in Fig. 1b). The Arg 108 in the PP2Ac ␣ isoform is a Pro in the ␤ isoform. Both isoforms bound ␣-4 in the two-hybrid screen, and we also mutated Arg 108 to Pro in PP2Ac ␣, which did not influence binding to either ␣-4 or the A subunit using the assays described below (data not shown). The other three charged residues were individually substituted with Ala and assayed for binding to ␣-4 and the A regulatory subunit (see below) ( Fig. 3).
To find a possible ␣-4 binding site we located clusters of residues that are identical in PP2Ac, PP4c, and PP6c (proteins that all bind ␣-4) but have different charged side chains relative to PP1c (Fig. 1b). We initially focused on three sites (solid bars in Fig. 1b), YY (91-92), RIT (110 -112), and SRQITQ (120 -125), and substituted Ala or Ser for residues in each of these sites. With the Y91A/Y92A mutant there was no discernable effect on binding of either regulatory subunit (not shown). On the other hand, RIT/AIA and SRQ/SAS mutants produced distortion of the catalytic site, with loss of phosphatase activity and reduced binding to microcystin-LR beads (not shown). It was interesting that mutations of surface charged residues could produce such structural perturbations in the catalytic subunit; however, the loss of native conformation prevented any valid conclusions about the role of these side chains in binding to regulatory subunits. Another possible site for ␣-4 interaction with PP2Ac is Glu 42 in the ESNVQ motif, common to PP2A, PP4, and PP6. Mutations of Glu 102 in Pph21 and Glu 38 in Sit4, the yeast homologue of PP2A and PP6, respectively, caused a loss of binding to Tap42, the yeast homologue of ␣-4 (56). Mutation of Glu 42 in PP2Ac altered binding to both ␣-4 and A subunits (see below).
Co-precipitation Assays for PP2Ac Binding to Regulatory Subunits-Binding to either ␣-4 or the A subunit was assayed by transient expression of (HA) 3 -PP2Ac in COS7 cells and immunoprecipitation by the triple hemagglutinin epitope tag. We also tested an inactive version of (HA) 3 -PP2Ac with a H118N mutation at the active site (55,57). Endogenous ␣-4 co-precipitated with both wild type and the inactive H118N (HA) 3 -PP2Ac (Fig. 2a, upper blot). Recovery of ␣-4 in the precipitates was quantitated consistently as 8 -10% of the input, relative to an internal standard in the same immunoblot (2% of the cell lysate). Immunoblotting for HA (Fig. 2a, lower blot) was the loading control for recovery of the different (HA) 3 -PP2Ac in the immunoprecipitates. Note that the H118N mutant migrated slightly differently from wild type. Addition of microcystin-LR, an active site inhibitor of PP2Ac, did not interfere with the co-precipitation of endogenous ␣-4 with active or inactive (HA) 3 -PP2Ac (Fig. 2a, compare ϩ and Ϫ lanes). Thus, it is unlikely that ␣-4 interacts with the PP2Ac active site or that binding of this inhibitor prevents binding of ␣-4. This observation contradicts a report that microcystin interferes with ␣-4 binding (58). In other experiments cells were co-transfected to express both (HA) 3  were recovered together either by anti-HA precipitation or by anti-FLAG precipitation (not shown). The results showed that wild type and H118N versions of (HA) 3 -PP2Ac both bound to endogenous or epitope-tagged ␣-4.
In contrast, low yields of co-precipitation made it appear as if the endogenous A subunit did not associate with (HA) 3 -PP2Ac (Fig. 2b, upper blot). The A subunit was intensely stained as an internal standard by immunoblotting 2% of the cell lysates, showing the low yields recovered with (HA) 3 -PP2Ac wild type or H118N. The apparent difference in recovery of the A subunit Ϯ MCLR seen in Fig. 2b was not taken as significant, considering the low yield in both samples. We reasoned that in these co-precipitations the A subunit was probably displaced by the anti-HA antibody because previously we showed by column chromatography that ectopically expressed (HA) 3 -PP2Ac did associate with the endogenous A subunit and eluted as PP2A-AC heterodimer and ABC heterotrimer (55). As an alternative approach to examining PP2A subunit association we co-expressed FLAG-PP2A-A with either wild type or H118N versions of (HA) 3 -PP2Ac, and under these circumstances we recovered the A and C subunits together by anti-FLAG precipitation (Fig. 2c). This co-precipitation was unaffected by the addition of MCLR to the lysates (not shown). We used coexpression and anti-FLAG precipitation to assay for association of the A subunit with different mutated forms of (HA) 3 -PP2Ac.
Basic Residues Necessary for PP2A Binding to the A Subunit-Basic residues at positions 41, 49, and 74 are unique in PP2Ac compared with PP4c, PP6c, PP1c, and PP2B (see above) (Fig.  1b). These residues were mutated to Ala, and the single-, double-, and triple-substituted forms were tested for binding to ␣-4 and the A subunit relative to the wild type PP2Ac. Substitutions of these basic residues caused a progressive reduction in binding of (HA) 3 -PP2Ac to co-expressed FLAG-A subunit (Fig.  3a). Single substitutions of K41A, R49A, or K74A did not significantly decrease co-precipitation of (HA) 3 -PP2Ac relative to wild type (90 versus 100% by densitometry). Results for K41A shown in Fig. 3a are representative of other experiments with FIG. 2. Co-precipitation of (HA) 3 -PP2Ac with endogenous ␣-4 and the A regulatory subunit. a, co-precipitation of endogenous ␣-4 with transiently expressed wild type (HA) 3 -PP2Ac (wt) and inactive (HA) 3 -PP2Ac (H118N). Transfection with empty pKH3 vector was used as a negative control. Proteins were immunoprecipitated (IP) using anti-HA. Immunoblotting used ␣-4 antipeptide antibody (upper blot) and anti-HA (lower blot). Cell lysate (2% of total) was included as an internal standard for the quantitation of recovery. Endogenous ␣-4 was recovered in similar yields with both versions of PP2Ac. Microcystin-LR was added prior to precipitation where indicated (ϩ) at a final concentration of 1 M. b, cell lysates with transiently expressed wild type (HA) 3 -PP2Ac or inactive (HA) 3 -PP2Ac or with empty vector as a negative control were used in anti-HA immunoprecipitation. Immunoblotting was done with anti-HA (lower blot) to show equal loading and with anti-PP2A-A (upper blot) to detect endogenous A subunit. The cell lysate (2% of total) was analyzed as an internal standard. c, cells transiently co-expressed FLAG-PP2A-A plus wild type (HA) 3 -PP2Ac or inactive (HA) 3 -PP2Ac or empty vector as a negative control. Anti-FLAG precipitates were prepared and immunoblotted for recovery of (HA) 3 -PP2Ac and for FLAG-PP2A-A as a loading control.
FIG. 3. Basic residues are required for association of the A subunit with PP2Ac. a, association of FLAG-PP2A-A with wild-type (wt) and mutated versions of (HA) 3 -PP2Ac. Mutated versions of PP2A were co-expressed with FLAG-PP2A-A that was immunoprecipitated (IP) with anti-FLAG. The recovered proteins were immunoblotted with anti-HA (upper blot) to detect co-precipitated PP2Ac and with anti-FLAG (lower blot) as a loading control. A trace of (HA) 3 -PP2Ac was recovered in the blank control with anti-FLAG beads, even in the absence of FLAG-PP2A-A (left lane). Results are representative of five separate experiments. b, mutants of (HA) 3 -PP2Ac were expressed individually by transient transfection and immunoprecipitated using anti-HA beads. The proteins recovered were analyzed by immunoblotting to detect endogenous ␣-4 (upper blot) relative to cell lysate (2%) included as an internal standard. Immunoblotting with anti-HA was done as a loading control (lower blot) singly substituted R49A or K74A. However, double-substituted PP2Ac-K41A/K74A consistently showed in multiple independent experiments a 50% reduction in binding to the FLAG-A subunit relative to wild type. Triple-substituted PP2Ac-K41A/ R49A/K74A lost essentially all (Ͼ95%) specific binding to the A subunit (Fig. 3a, upper blot). The amount of K41A/R49A/K74A triple-mutated HA-tagged protein recovered on anti-FLAG beads was at background level, comparable with control experiments with cells not expressing any FLAG-A (Fig. 3a, upper  blot, compare far left and far right lanes). In addition to the loading control showing equal recovery of the FLAG-A subunit (Fig. 3a, center blot) we demonstrated that these mutated (HA) 3 -PP2Ac all bound to immobilized microcystin-LR in a pull-down assay (Fig. 3a, bottom blot). Binding to MCLR provided one line of evidence for the overall conformational integrity of the mutated PP2Ac. We concluded that the basic side chains Lys 41 , Arg 49 , and Lys 74 are involved in cooperative binding to the A subunit.
On the other hand, binding to endogenous ␣-4 by single K41A or double K41A/K74A mutants was the same (appearing even slightly increased) compared with wild type PP2Ac (Fig.  3b, upper blot). Binding of ␣-4 to the triple K41A/R49A/K74A mutant was increased substantially relative to wild type (Fig.  3b). Equivalent amounts of the different (HA) 3 -PP2Ac were recovered by precipitation, shown in the loading control (Fig.  3b, lower blot). The results indicate that these basic residues are not required for binding of PP2Ac to ␣-4. Furthermore, the results reinforce the conclusion that these mutations did not alter the overall conformation of the phosphatase catalytic subunit. Diminished binding of the A subunit by the triple mutant PP2Ac-K41A/R49A/K74A apparently favored its binding to endogenous ␣-4 by reducing competition between the regulatory subunits.
Binding of PP2Ac to the ␣-4 (mTap42) Subunit-The single residue substitution of E42A in (HA) 3 -PP2Ac significantly impaired (by ϳ80%) binding to endogenous ␣-4 protein (Fig. 4a). Dual substitution of K41A/E42A showed about the same severe reduction in co-precipitation of endogenous ␣-4. A loading control (Fig. 4a, lower blot) showed equal recovery of the wild type and mutated (HA) 3 -PP2Ac in the immunoprecipitates. The results demonstrated that Glu 42 in PP2Ac is critical for recognition of the ␣-4 subunit. On the other hand, the E42A mutation enhanced binding of PP2Ac to co-expressed FLAG-A subunit (Fig. 4b). The results mirror the effects seen in Fig. 3, where reduced binding to one regulatory subunit yielded increased binding to the alternate partner. Interestingly, the dual mutation K41A/E42A did not enhance binding to the FLAG-A subunit like the single E42A mutation but yielded about the same co-precipitation as wild type. The loading controls (Fig. 4b, lower blot) revealed equal recovery of the FLAG-A subunit in the precipitates. We imagine that binding to the A subunit could have been influenced by local ionic effects at this site. The net zero charge in wild type (from positive and negative charges at residues 41 and 42, respectively) became a positive charge in the E42A mutant and was returned to zero net charge in the double K41A/E42A mutant.
Phosphatase Activity of PP2Ac Mutants-Another stringent test of conformational integrity of mutated PP2Ac is enzymatic specific activity. We immunoprecipitated wild type, the E42A mutant, and the K41A/R49A/K74A mutant of (HA) 3 -PP2Ac and assayed for okadaic acid-inhibited phosphatase activity by hydrolysis of 32 P-labeled myelin basic protein (Fig. 5). There was no significant difference in the PP2A activity of these proteins in independent experiments. We note that this may depend on the substrate used in the assay (see Ref. 50). The amount of precipitated (HA) 3 -PP2Ac was shown to be identical by anti-HA immunoblotting (Fig. 5, inset). Thus, mutation of the surface charged residues Glu 42 or K41/R49/K74 did not affect the specific activity of the phosphatase. Temperature-dependent Binding to PP2Ac-Temperaturesensitive phenotypes arise from mutations in yeast type-2 phosphatases Sit4 and Pph21/22, in particular mutations of the acidic residue that corresponds to Glu 42 in human PP2Ac. This raised the possibility that the E42A mutation might reduce conformational integrity of the PP2Ac at elevated temperatures, although it had normal specific activity at 30°C (see Fig.  5). We used a pull-down assay with MCLR-conjugated beads to test for binding of PP2Ac from lysates of transfected cells with equal expression levels of the wild type, E42A, and K41A/E42A forms of (HA) 3 -PP2Ac. The same amount of these proteins bound to MCLR beads at 4°C (Fig. 6a), and densitometry of independent experiments showed no significant differences. As a negative control Sepharose CL-4B beads did not pull down (HA) 3 -PP2Ac (not shown). The same pull-down assay with the same lysates performed at 37°C (or 40°C) yielded much more of these forms of (HA) 3 -PP2Ac bound to the MCLR beads, and on the same immunoblot the bands in samples from higher temperatures were overexposed. We tested reduced volumes of the 37°C pull-down samples until we could approximate the same staining intensity on immunoblots as with samples from pull-downs done at 4°C (Fig. 6a). Based on these results (5 versus 15 l) we estimated about 3-fold more (HA) 3 -PP2Ac bound microcystin-LR at 37°C compared with 4°C. The wild type and mutant (HA) 3 -PP2Ac both bound in higher yields at the elevated temperatures, arguing against a temperature-dependent loss of conformation because of the E42A mutation. A separate experiment compared the binding of endogenous PP2Ac in untransfected cells with MCLR beads at 4 and 37°C and found a ϳ3-fold increase in binding at the higher temperature (not shown). We concluded that the temperature-dependent increase in binding of PP2Ac to microcystin-LR was not an artifact resulting from ectopic overexpression or the epitope tag in the protein. Furthermore, MCLR beads bound endogenous PP1c from these cell lysates with a Ͼ3-fold increase in recovery at 37°C relative to 4°C based on staining different volumes of eluted proteins on the same immunoblot (Fig. 6b). These results showed that binding of PP2Ac and PP1c to MCLR beads was enhanced significantly at 37°C relative to 4°C.
The binding of (HA) 3 -PP2Ac to the ␣-4 subunit also appeared to increase with temperature. COS7 cells were transfected to express (HA) 3 -PP2Ac, and the endogenous ␣-4 was immunoprecipitated from lysates with or without added microcystin-LR (Fig. 7). As a loading control the immunoprecipitates were immunoblotted for recovery of ␣-4, and as expected the amount recovered was the same in each sample (Fig. 7, lower  blots). However, more (HA) 3 -PP2Ac was co-precipitated with ␣-4 at 37°C compared with 4°C. There was no effect of adding MCLR to the lysates on co-precipitation of (HA) 3 -PP2Ac at either temperature (Fig. 7). We concluded that binding of (HA) 3 -PP2Ac to ␣-4 increased with temperature, indicative of an activation energy barrier. Co-precipitation of (HA) 3 -PP2Ac with co-expressed FLAG-A subunit as done in Figs. 2c, 3a, and 4b was not different at 37°C compared with 4°C (not shown). This posed a curious difference of temperature-dependent binding of (HA) 3 -PP2Ac to either MCLR or ␣-4 but not to the FLAG-A subunit.

DISCUSSION
Association with various regulatory subunits allows protein Ser/Thr phosphatases such as PP2A to fulfill multiple, yet distinctive, biological functions. Most PP2A catalytic subunit in cells is bound to the A regulatory scaffold subunit, but it can alternatively bind to the ␣-4 (mTap42) subunit. Because the binding of these regulatory subunits is mutually exclusive, one can imagine that there are overlapping areas of surface contacts on PP2Ac, or at least steric exclusion of one by the other. If there is overlap of the binding sites, there still must be determinants for differential recognition of these subunits by PP2Ac because ␣-4 but not the A subunit binds to related phosphatases PP4 and PP6. In this study some mutations in the helical domain of PP2Ac affected neither A subunit nor ␣-4 binding (e.g. Y91A/Y92A), whereas other mutations produced conformational deformations with loss of phosphatase activity and binding to microcystin and regulatory subunits (e.g. R110A/T112S). Other mutations identified positively charged surface residues that were required for binding the A subunit (Lys 41 , Arg 49 , Lys 74 ) and an acidic residue required for binding FIG. 6. Temperature-dependent binding of protein phosphatases PP2A and PP1 to microcystin-LR. a, wild type (wt) and mutated PP2Ac were expressed in COS7 cells, and cell lysates were adsorbed with MCLR-agarose or Sepharose CL-4B as a blank control. The binding of (HA) 3 -PP2Ac was analyzed by immunoblotting. Binding at 4°C is shown in the top blot of a in comparison with 37°C in the bottom blot, where recovery of both phosphatases was 3-fold higher. Endogenous PP2A and (HA) 3 -PP2Ac showed identical temperature dependence, as analyzed in parallel experiments (data not shown). The results are representative of four different experiments with similar temperaturedependent increases in binding. b, PP1c immunoblots were performed on the same MCLR-agarose pull-down assays. Binding at 4°C in comparison with 37°C showed the same temperature dependence as PP2Ac with Ͼ3-fold higher binding to MCLR.
FIG. 7. Temperature-dependent binding of protein phosphatase PP2A to the ␣-4 subunit. COS7 cells were transfected to express wild type (HA) 3 -PP2Ac, and cell lysates were prepared. Samples were incubated at either 4°C or 37°C with and without added MCLR, and the endogenous ␣-4 was immunoprecipitated (IP) with an affinitypurified rabbit antibody. Immunoblotting with anti-␣-4 showed the amount of ␣-4 recovered as a loading control (lower blots). The amount of co-precipitated (HA) 3 -PP2Ac was determined by immunoblotting with anti-HA antibody.
␣-4 (Glu 42 ). Interestingly, mutations that essentially eliminated binding to one subunit enhanced the binding to the other, supporting the idea that there is a competition for binding to PP2Ac within the cell. Binding of these mutant forms of PP2Ac to one regulatory subunit and not the other makes it unlikely that the loss of binding was because of perturbation of overall conformation, which would preclude binding to either of the regulatory subunits. Instead, we think these surface charged side chains are involved in specific protein-protein contacts between PP2Ac and its different regulatory subunits. These subunit-subunit interfaces apparently mediate allosteric regulation of PP2Ac by the A subunit and ␣-4 to restrict catalytic activity, which has been demonstrated previously (50,59).
Binding of the A Subunit to PP2A-The portion of PP2A that binds the regulatory subunits is within the ␣-helical domain (colored gold in Fig. 8) that fits together with a C-terminal mostly ␤ sheet domain (gray in Fig. 8) containing the active site bimetallic center. We identified four basic residues in this domain that are non-conservatively substituted even in the closest PP2A relatives, PP4 and PP6, leading to the hypothesis that these residues served to specifically recognize the A subunit. Indeed, substitution of these basic residues essentially eliminated A subunit binding. These basic residues (blue in Fig. 8) appear on the rear surface of the PP2A catalytic subunit along the subdomain interface, positioning the regulatory subunit on the back of the PP2Ac while leaving the face of the catalytic subunit exposed for interaction with substrates or inhibitors such as microcystin-LR. We expect multiple acidic side chains in the A subunit to interact with these basic residues on PP2A. Our examination of the A subunit structure (60) and knowledge that HEAT repeats 11-15 are required for binding to PP2Ac (61) lead us to predict that Glu 375 , Glu 409 , Glu 412 , and Asp 414 might interact with Lys 41 , Arg 49 , and Lys 74 on the back of PP2A. Mutation of Asp 414 is reported to diminish binding of the A subunit to PP2A (62). On the other hand, mutation K416E in the A subunit severely reduces binding to PP2A (33,62), and this would predict a charge interaction of a positive residue on the A subunit and a negative residue on PP2Ac. Such a negative residue has not been assigned, but at least it is unlikely to be Glu 42 based on our results. It would be interesting to create models that accommodate all of these putative charged pairs between the A subunit and PP2Ac.
Both the N and C termini of the PP2Ac influence association with the A subunit without similar effects on association with ␣-4. Mutations to the phosphorylation and carboxyl methylation sites in the C terminus (Y307F/L309Q) reduced association of (HA) 3 -PP2Ac with the A subunit (see Ref. 55). Binding of anti-HA antibodies to transiently expressed (HA) 3 -PP2Ac prevented recovery of the A subunit ( Fig. 2) but not ␣-4, presumably because of selective steric interference. It is curious that a single HA tag reportedly does not prevent co-precipitation of the A subunit (63,64). It is possible that the tandem arrangement of multiple HA epitopes allows the binding of antibody to interfere with the A subunit. The results are consistent with the N and C termini of PP2Ac being in proximity to one another and with the site of association with the A subunit. In contrast, binding of endogenous ␣-4 was unaffected either by mutations in the C-terminal YDFL motif or by antibody binding to the N-terminal triple HA epitope of (HA) 3 -PP2Ac. These differences give some additional spatial information about the differences between the A subunit and ␣-4 docking onto PP2Ac.
One intriguing result from our two-hybrid screen was the recovery of many more clones of the ␤ isoform of the PP2A catalytic subunit compared with clones of the ␣ isoform. Based on Northern hybridization, expression of the ␣ isoform mRNA predominates over expression of the ␤ isoform mRNA (65-67).
Recovery of more ␤ than ␣ isoform cDNA clones with ␣-4 could be because of the relative expression levels of the two PP2A catalytic subunits in the library from a 9-day mouse embryo. Deletion of the ␣ isoform of PP2Ac produced embryonic lethality (65). It is inviting to speculate that the two isoforms associate with different regulatory subunits. We tested whether a single R108P substitution, one of the few differences between PP2Ac ␣ and ␤ isoforms, changed the association of PP2A with ␣-4 or the A subunit, but there was no marked difference.
Binding of ␣-4 to PP2A-Interaction of PP2Ac with ␣-4 requires Glu 42 , and mutation of this residue resulted in severe loss of binding, consistent with previous results with the yeast FIG. 8. Model of PP2Ac showing sites for binding to ␣-4 and the A subunit. The three-dimensional structure of PP2Ac was threaded onto the backbone of PP1c using the SWISS-MODEL data base (www.expasy.ch/swissmod/SWISS-MODEL.html), a structure prediction program. The model is colored gold in the helical domain that binds regulatory subunits and gray in the ␤ sheet domain comprising the remainder of the catalytic subunit. The N and C termini are labeled to allow for orientation of the molecule; the first residue corresponds to Asn 20 in PP2Ac, and the C terminus in the model structure corresponds to residue Ala 296 . The top structure shows the rear view of the catalytic domain with the basic residues Lys 41 , Arg 49 , and Lys 74 that are necessary for A subunit binding shown in blue. The red residue represents Glu 42 , which is required for ␣-4 binding. The ␤-3-␣-4 surface loop of PP2Ac is shown in green. The bottom model shows the three-dimensional structure of PP2Ac from the front view looking into the catalytic cleft. The green residue represents His 118 , which is in ␤-3. The orange residues represent Tyr 91 and Tyr 92 , which when mutated to Ala had no effect on binding to either the A subunit or ␣-4.
homologues Pph21/22 and Tap42 (56). The yeast Pph21/22 phosphatase differs from mammalian PP2A by having an extended N-terminal region, positioning this residue at Glu 102 . This acidic side chain (red in Fig. 8) is in the ␣-helical domain and appears on the top back of the phosphatase at the junction between two helices distal from the active site. The proximity of the Glu 42 residue relative to the Lys 41 , Arg 49 , and Lys 74 residues required for A subunit binding suggests how binding to ␣-4 or the A subunit would be mutually exclusive, with overlapping catalytic subunit-regulatory subunit contact. This region of the PP2Ac surface is implicated in regulation of the phosphatase because the regulatory subunits that bind here reduce activity at the active site on the opposite face of the protein.
Conformational Flexibility and Modulation of Catalytic Activity-Substitution of residues Arg 110 and Thr 112 on the back surface of PP2Ac caused conformational deformation because the mutated PP2Ac had little catalytic activity and did not bind to microcystin-LR. Expression levels of this mutant protein were the same as other versions of PP2Ac, but it bound neither the A subunit nor ␣-4. The Arg 110 and Thr 112 residues are in the ␣-3-␤-4 loop, connected to the short ␤-4 sheet that threads through the center of PP2Ac and forms part of the active site. We imagine that mutations of Arg 110 and Thr 112 perturbed the positioning of the ␣-3-␤-4 surface loop (Fig. 8) as well as the ␤-4 strand, thereby altering the overall protein conformation and phosphatase activity and the association with both regulatory subunits. We speculate that interaction with residues 110 RIT 112 in the ␣-3-␤-4 loop on the back surface of PP2A, PP4, and PP6 may be how regulatory subunits control phosphatase catalytic activity. ␣-4 has been thought to inhibit PP2Ac based on low activity in assays with pNPP and PHAS-I (4E-BP1) (68). Yeast protein Tap42 (homologue to ␣-4) is considered a phosphatase inhibitor because mutations in Tap42 increase phosphorylation levels of putative substrates (69). However, we have shown that expression of ␣-4 enhances dephosphorylation of the Mid1 protein in living cells (70). Mid1 is a microtubuleassociated protein that directly binds ␣-4 and ␣-4::PP2A complexes. We imagine that ␣-4 and particular substrates engage residues at either end of the ␤-4 sheet in PP2Ac on opposite faces of the phosphatase to yield substrate-specific phosphatase activity. Furthermore, we speculate the temperature-dependent increase in PP2Ac and PP1c binding to microcystin reflects an activation barrier that involves a conformational change in the active sites of these phosphatases. We propose that there is inherent conformational plasticity in the PPP phosphatase catalytic subunits that is exploited to modulate catalytic rate as well as substrate specificity in response to binding of regulatory subunits. In this study we have identified surface residues necessary for subunit-subunit association, which should contribute to an understanding of the basis for allosteric regulation of PP2A and its competitive distribution among the multiple regulatory subunits of these residues.