Ca2+/Calmodulin Directly Interacts with the Pleckstrin Homology Domain of AKT1*

AKT kinase, also known as protein kinase B, is a key regulator of cell growth, proliferation, and metabolism. The activation of the AKT signaling pathway is one of the most frequent molecular alterations in a wide variety of human cancers. Dickson and coworkers recently observed that Ca2+·calmodulin (Ca2+·CaM) may be a common regulator of AKT1 activation (Deb, T. B., Coticchia, C. M., and Dickson, R. B. (2004) J. Biol. Chem. 279, 38903–38911). In our efforts to scan the mRNA-displayed proteome libraries for Ca2+·CaM-binding proteins, we found that both human and Caenorhabditis elegans AKT1 kinases bound to CaM in a Ca2+-dependent manner (Shen, X., Valencia, C. A., Szostak, J., Dong, B., and Liu, R. (2005) Proc. Natl. Acad. Sci. U. S. A. 102, 5969–5974 and Shen, X., Valencia, C. A., Gao, W., Cotten, S. W., Dong, B., Chen, M., and Liu, R. (2007) submitted for publication). Here we demonstrate that Ca2+·CaM and human AKT1 were efficiently co-immunoprecipitated, and their interaction was direct rather than mediated by other proteins. The binding is in part attributed to the first 42 residues of the pleckstrin homology (PH) domain, a region that is critical for the recognition of its lipid ligands. The PH domain of human AKT1 can disrupt the complex of the full-length AKT1 with Ca2+·CaM. In addition, Ca2+·CaM competes with phosphatidylinositol 3,4,5-trisphophate for interaction with the PH domain of human AKT1. Our findings suggest that Ca2+·CaM is directly involved in regulating the functions of AKT1, presumably by releasing the activated AKT1 from the plasma membrane and/or prohibiting it from re-association with phosphoinositides on plasma membrane.

on the hydrophobic motif (Ser-473, Ser-474, and Ser-472 for AKT1, AKT2, and AKT3, respectively). Such activation permits a conformational change that results in substrate binding and a significant increase of the rate of catalysis.
AKT family members regulate a wide variety of cellular functions, including apoptosis, cellular proliferation, differentiation, intermediary metabolism, genome stability, and neo-vascularization (25,27). The regulation mechanism of AKT is by deactivation or activation of its downstream substrates through phosphorylation. For instance, AKT kinase facilitates growth factor-mediated cell survival and blocks apoptotic cell death by phosphorylating and deactivating pro-apoptotic factors such as Bad, caspase-9, and Forkhead transcription factors (9, 14 -19). AKT also phosphorylates and inactivates glycogen synthase kinase-3, which allows the activation of glycogen synthase (13). On the other hand, AKT phosphorylates and activates IKK␣, which phosphorylates IB followed by its ubiquitination and proteasomal degradation (20 -22, 29). AKT regulates many cellular processes that, upon deregulation, may contribute to the development or progression of cancer. It is now evident that the activation of the AKT signaling pathway is one of the most frequent molecular alterations in a wide variety of human cancers (30 -34). Overexpression or inappropriate activation of AKT1 is noted in many types of cancer.
Despite its importance in numerous signaling pathways, our understanding of the activation and regulation of AKT kinases is still incomplete. The lipid-binding PH domain of AKT plays an important role in promoting the translocation of AKT to the plasma membrane for activation. Recent results suggest that the PH domain is also important for the interaction with a number of other proteins (35). In our recent effort to scan the mRNA-displayed proteome libraries for Ca 2ϩ ⅐CaM-binding proteins, we found that both human and Caenorhabditis elegans AKT1 bound to CaM in a Ca 2ϩ -dependent manner (2). 4 Here we extend Dickson's work and report that AKT1 interacts with Ca 2ϩ ⅐CaM through two binding motifs, one at the first half of the N-terminal PH domain and another at the N terminus of the kinase catalytic domain. Our results indicate that Ca 2ϩ ⅐CaM and AKT1 can be co-immunoprecipitated and the Ca 2ϩ ⅐CaM binding appears to be general in all three AKT isoforms. The Ca 2ϩ ⅐CaM binding to the N-terminal PH domain of AKT1 could regulate its interaction with PtdIns(3,4,5)P 3 .

Construction of Recombinant Genes and Expression of AKT1
Fragments-The cDNA corresponding to the open reading frame of human AKT1 was reversibly transcribed from a human poly(A) ϩ mRNA library (Ambion, Austin, TX). The full-length AKT1 was amplified using a high fidelity Platinum Pfx DNA polymerase (Invitrogen) with the following primers: forward primer, 5Ј-CACCATGAGCGACGTGGCTATT-GTGA-3Ј; reverse primer, 5Ј-GGCCGTGCCGCTGGC-3Ј. The PCR product of full-length AKT1 was gel-purified and cloned into a pcDNA3.1/V5-His TOPO vector (Invitrogen) according to the manufacturer's instructions. The resulting AKT1-containing plasmid was confirmed by PCR and sequence analysis. This construct was used as the template to PCR-amplify various AKT1 fragments.
Cell Culture and Lysate Preparation-HeLa S3 cells were grown to stationary phase in Ham's F-12 medium supple-mented with 10% fetal bovine serum. Cells were harvested and lysed as reported (36). After centrifugation to remove intact cells, the supernatants were cleared at 14,000 rpm for 15 min at 4°C. Lysates prepared were used for pulldown assays and coimmunoprecipitation experiments.
CaM Binding and Pulldown Assays-To investigate the interaction between Ca 2ϩ ⅐CaM and fragment/full length AKT, CaM binding assays were performed using both radiolabeled AKT protein fragments generated by TNT and GST-tagged fragments overexpressed in E. coli. For the TNT approach, radiolabeled AKT fragments were synthesized in the presence of 10 Ci of [ 35 S]methionine (PerkinElmer Life Sciences) in a total volume of 25 l for 90 min at 30°C as previously described (2). To perform the CaM binding assays, an aliquot of the TNT product was mixed with an appropriate amount of biotinylated CaM for 90 min at 4°C in buffer A (25 mM Tris-HCl, pH 8.0, 150 mM NaCl, 1 mg/ml bovine serum albumin, 5 mM 2-mercaptoethanol, 1 mM CaCl 2 ) or buffer B (same as buffer A except 1 mM CaCl 2 was replaced with 2 mM EGTA). After binding, 30 l of the 50% slurry of streptavidin beads (Pierce Biotech, Rockford, IL) was added, and the mixture was incubated for 45 min at room temperature with gentle mixing. The supernatant was removed by centrifugation in an Ultrafree-MC centrifugal filter tube (Millipore, Billerica, MA), and the beads were washed three times with 150 l of buffer A. Proteins captured were then released from the column by chelating Ca 2ϩ with 150 l of buffer B. Each fraction was separated by SDS-PAGE and images were obtained through a Storm 860 PhosphorImager.
When GST-tagged, overexpressed AKT fragments or cell lysates were used for the binding reaction, CaM-Sepharose pulldown assays were performed. To do that, 100 l of preequilibrated CaM-Sepharose 4B beads (Amersham Biosciences) was added to an appropriate amount of purified AKT protein (1-10 g), or HeLa S3 cell lysates (ϳ1 mg of total protein), followed by incubation at 4°C for 2-3 h in a buffer containing 50 mM Tris-HCl, pH 7.4, 150 mM NaCl, 1ϫ EDTA-free protease inhibitor mix, with 1 mM Ca 2ϩ (buffer C) or 2 mM EGTA (buffer D). The beads were washed three times with buffer C or buffer D. Proteins bound to CaM were eluted from the beads with buffer D. Proteins from each fraction (L, purified AKT protein or lysate; FT, flow-through; W, last wash before elution; and E, elution) were resolved on SDS-PAGE gels, transferred to a nitrocellulose membrane (Amersham Biosciences), and probed with either anti-AKT1 (Santa Cruz Biotechnology, Santa Cruz, CA), anti-His tag (R&D Systems, Minneapolis, MN), anti-AKT (for all AKT isoforms, Cell Signaling Technology, Danvers, MA), or anti-Calcineurin A2 (Santa Cruz Biotechnology).
Competitive Binding and Elution Assay-For competitive binding experiments, 2 M GST-tagged PH domain of AKT1 (Upstate) was added to a binding mixture with the full-length AKT1 pre-complexed with Ca 2ϩ ⅐CaM immobilized on Sepharose beads. The elution was performed with buffer B. For competitive elution experiments, 2-50 M PH domain of AKT1 or a CaM-binding peptide from MLCK (Ac-RRKWQKT-GHAVRAIGRL-NH 2 ) (Calbiochem) in buffer A was used to elute the full-length AKT1 pre-complexed with Ca 2ϩ ⅐CaM on agarose beads.
CaM Binding Affinity Measurement-To measure the binding affinity of AKT fragments with Ca 2ϩ ⅐CaM, the [ 35 S]methionine-labeled PH domain F 1-149 , F 150 -408 , and full-length AKT1 were generated by TNT and mixed with biotinylated CaM, whose concentration was varied in the range of 10 nM and 2.5 M. Biotin-CaM⅐protein complex was captured by streptavidin beads, and the extent of binding was determined by scintillation counting of each fraction, followed by fitting the data to a binding curve using SIGMAPLOT 8.0 software (Systat, Richmond, VA).
Immunoprecipitation and Co-immunoprecipitation-Immunoprecipitation was performed by incubating 1 mg of HeLa S3 lysate with anti-CaM, anti-AKT1, or an irrelevant IgG, respectively. Samples were separated by SDS-PAGE, transferred to a nitrocellulose membrane, and probed with different antibodies.
For co-immunoprecipitation experiments, equal amounts of pre-cleared HeLa S3 whole cell lysate (ϳ1 mg of total protein) were incubated with 2-5 g of primary antibody (anti-CaM) for 4 h at 4°C, in the presence of either buffer C or buffer D. 20 l of pre-equilibrated protein A/G-agarose conjugate was added, and the mixture was further incubated at 4°C for 2 h with mixing. The pellets were collected and washed three times with buffer C or D. After the final wash, pellets were eluted with buffer D. Different fractions were loaded onto SDS-PAGE followed by Western blotting and probing using anti-AKT1. Conversely, anti-AKT1 was used for co-immunoprecipitation and blots were probed with anti-CaM. As a positive control, blots were probed with anti-Calcineurin A2, for the detection of calcineurin, a well known CaM-binding protein. As a negative control, blots were also probed with an irrelevant anti-TFII antibody.
Protein Lipid Overlay Assay-PtdIns(3,4,5)P 3 , PtdIns(3,4)P 2 , PtdIns(3,5)P 2 , PtdIns(4,5)P 2 , and phosphatidylcholine were obtained from Echelon Biosciences (Salt Lake City, UT). 100 pmol of each PIP were dotted on a Hybond-C Extra membrane (Amersham Biosciences), and protein lipid overlay assay was performed according to the manufacturer's instructions. Different amounts of PtdIns(3,4,5)P 3 were also used to examine its concentration dependence. The membrane was first blocked for 1 h in a TBS buffer (25 mM Tris-HCl, pH 8.0, 150 mM NaCl) with 0.1% (v/v) Tween 20 and 3% (w/v) bovine serum albumin. It was then incubated at 4°C overnight with 0.5 g/ml Histagged full-length AKT1 (Upstate), GST-tagged PH domain of AKT1 (Upstate), or His-tagged AKT1 without PH domain (Upstate) in the blocking buffer containing 0.24 M CaM and 100 M Ca 2ϩ or EGTA. After washing three times, the membrane was incubated with anti-His tag or anti-GST for 3 h at room temperature. The membrane was washed again and incubated with anti-mouse IgG horseradish peroxidase (Amersham Biosciences), followed by development using ECL plus Western blotting detection system (Amersham Biosciences).

RESULTS
To address the difficulty of identifying the numerous conditional interactions that are dependent on the intracellular Ca 2ϩ concentrations, we developed an mRNA-display protein selection platform that allows us to scan a proteome of interest for Ca 2ϩ -dependent protein-protein interactions (2). 4 By employing this novel approach, we have scanned both the human and the C. elegans proteomes for Ca 2ϩ ⅐CaM-binding proteins (2). 4 A large number of both known and previously uncharacterized Ca 2ϩ ⅐CaM-binding proteins were identified, including many protein kinases. These CaM-binding kinases include CaM-dependent protein kinase, CaM-dependent protein kinase kinase, elongation factor kinase, and protein kinase C family member UNC-13 that have been previously characterized to interact with Ca 2ϩ ⅐CaM (37). Several protein kinases that have not been previously reported to interact with Ca 2ϩ ⅐CaM were also identified, and AKT1 is among them. The discovery that both human and C. elegans AKT1 kinases interact with CaM in a Ca 2ϩ -dependent manner strongly suggests that Ca 2ϩ ⅐CaM play an important role in regulating the functions of AKT.
We first characterized the interaction of Ca 2ϩ ⅐CaM with AKT1 by sequence analysis of the selected fragments. Only one fragment originating from human AKT1 was isolated in the Ca 2ϩ ⅐CaM-binding selection using an mRNA-displayed human proteome library. Sequence analysis showed that this was a fragment that covered the entire PH domain of human AKT1. However, four fragments originating from C. elegans AKT1 were isolated, namely, F 1-144 , F 54 -138 , F 105-138 , and F 103-172 (Fig. 1). All these fragments are located within or close to the PH domain of C. elegans AKT1. F 1-144 contains the whole PH domain (the arrowed region), whereas F 54 -138 contains its second half. The shortest overlapping region is between residues 105 and 138 ( 105 VRQRWIHAIESISKKYKGTNAN-PQEELMETNQQP 138 ). This region covers the C-terminal part of the PH domain and part of the LINK region. Using a webbased motif analysis program (37), it is found that this fragment contains a positively charged amphiphilic ␣-helix (underlined), which is a characteristic of many Ca 2ϩ ⅐CaM binding motifs.
To investigate the interaction of these fragments with CaM, we generated radiolabeled AKT fragments by TNT and used them in the CaM binding assay in the presence or absence of Ca 2ϩ . We first validated the assay using both positive and negative controls. For positive controls, more than 20 well known human Ca 2ϩ ⅐CaM-binding proteins or their fragments con-taining the Ca 2ϩ ⅐CaM binding motifs were synthesized by TNT and used to test their CaM binding with this assay. It was found that all these positive controls bound to CaM specifically in a Ca 2ϩ -dependent manner (data not shown). For negative controls, a number of proteins or their fragments that have never been reported to interact with CaM were used. In addition, several known Ca 2ϩ ⅐CaM-binding proteins with the Ca 2ϩ ⅐CaM binding motifs deleted were also used as negative controls. No binding was detected for these negative control proteins under the conditions used in such in vitro assay, as shown in Fig. 2 using SNAP25 as an example. The results of both positive and negative controls demonstrate this CaM binding assay is a very robust method to elucidate protein interactions with Ca 2ϩ ⅐CaM. Using this CaM binding assay, it was found that all the four selected C. elegans AKT fragments bound to CaM in a Ca 2ϩ -dependent manner as shown in Fig. 2, and the binding was completely abolished when Ca 2ϩ was chelated with EGTA. When irrelevant proteins such as SNAP25 were used as negative controls, no CaM binding was observed. Interestingly, F 54 -138 , a fragment that covers both the second half of the PH domain and the LINK region, bound to Ca 2ϩ ⅐CaM very tightly, implying it contains at least one of the major binding motifs. It appears that the interaction of Ca 2ϩ ⅐CaM with the longer fragment F 1-144 was not as strong as that with F 54 -138 or F 105-138 . This phenomenon was also observed in the interaction of CaM with a number of other proteins (2), presumably because the CaM binding motifs are more accessible when present as short fragments.
Because the antibodies for C. elegans AKTs are not readily available, we focused our investigation on human AKTs. We hypothesized that human AKTs should use different region(s) at or close to the PH domain to interact with Ca 2ϩ ⅐CaM, presumably because the LINK region of C. elegans AKT1 has very low sequence homology with human AKT1 (supplemental Fig.  S1). To map the location of the CaM binding motif(s) on human AKT1, we generated by TNT [ 35 S]methionine-labeled protein fragments that cover all the possible regions of AKT1 (Fig. 3).  ]Methionine-labeled protein fragments were generated by TNT. An aliquot of the TNT reaction mixture was incubated with an appropriate amount of biotinylated CaM in the presence or absence of Ca 2ϩ , followed by mixing with streptavidin-agarose beads. The beads were washed, and the bound proteins were eluted using a buffer containing EGTA. Each fraction was separated by SDS-PAGE, and images were obtained by autoradiography. Numerous positive and negative controls were performed but only SNAP25 was shown here as a negative control. L, TNT product; FT, flow through; W, last wash before elution; and E, elution.
Specifically, F 1-115 covers the PH domain, F 116 -149 the LINK region, F 150 -408 the kinase catalytic domain, and F 409 -480 the C-terminal extension domain, respectively. Consistent with the result from the selection using an mRNA-displayed human proteome library, the PH domain of AKT1 (residues 1-115) bound to CaM well (Fig. 4A). This interaction is highly dependent on the presence of Ca 2ϩ . When EGTA was included, the binding was completely abolished. All the fragments that contain the PH domain, including F 1-149 , F 1-408 , and F 1-480 , also bound to CaM in a Ca 2ϩ -dependent manner (Fig. 4A). We further mapped the binding motif within the PH domain using F 1-42 and F 43-115 that cover two different parts of the PH domain. It appeared that the Ca 2ϩ ⅐CaM binding of the PH domain of AKT1 could be attributed to the first 42 residues, because F 1-42 strongly bound to Ca 2ϩ ⅐CaM, whereas F  showed no binding at all (Fig. 4B).
We then examined whether the PH domain-deleted AKT1 also interacted with Ca 2ϩ ⅐CaM. Surprisingly, F 116 -480 also bound to Ca 2ϩ ⅐CaM (Fig. 4A), suggesting the presence of another CaM binding motif. To map this second binding motif, three protein fragments that cover the LINK region (F 116 -149 ), the kinase catalytic domain (F 150 -408 ), and the C-terminal extension domain (F 409 -480 ), respectively, were generated and used to interact with CaM. Interestingly, it was found that this second CaM-binding site was not in the LINK or the C-terminal extension regions, but within the kinase catalytic domain. We further mapped this CaM binding motif to F 150 -197 (Fig.  4B), a region at the N terminus of the kinase catalytic domain. Indeed, all the fragments that contain F 150 -197 also bound to Ca 2ϩ ⅐CaM, including F 150 -408 , F 150 -480 , and F 116 -480 (Fig. 4A). As already shown in Fig. 2, negative control proteins did not bind to CaM in these assays.
The results using TNT-generated protein fragments were consistent with the CaM-Sepharose pulldown assay using GSTtagged proteins overexpressed in E. coli (supplemental Fig. S2). To rule out the possibility of nonspecific interaction due to the presence of GST tag in such a pulldown assay, overexpressed GST tag alone was used in the binding. No binding was detected between GST tag and CaM, both in the presence and absence of Ca 2ϩ . Taken together, our results suggest that there exist two Ca 2ϩ ⅐CaM binding motifs on human AKT1, one at the first half of the PH domain (F  and the other at the N terminus of the kinase catalytic domain (F 150 -197 ).
To analyze the interaction quantitatively, we determined the binding affinity of these two fragments of AKT1 with Ca 2ϩ ⅐CaM by varying the concentration of biotinylated CaM used in the CaM binding assay. The binding affinity for F  and F 150 -408 was ϳ750 Ϯ 100 nM and 500 Ϯ 80 nM, respectively. The full-length AKT1 has an affinity similar to that of these fragments. This relatively weak interaction suggests that each binding motif might require the elevation of local Ca 2ϩ concentration to effectively compete with many other CaM-binding proteins in vivo.
To investigate the interaction between AKT1 and Ca 2ϩ ⅐CaM under physiological conditions, we first tested whether the  . In vitro CaM binding assays using different human AKT1 fragments. All the fragments were generated by TNT, except F 1-42 , which could not be TNT-synthesized and was obtained by overexpression in E. coli. For TNT-generated fragments, an appropriate amount of [ 35 S]methionine-labeled AKT1 fragment was incubated with biotinylated CaM in a buffer containing either Ca 2ϩ or EGTA. The beads were washed, and the bound molecules were eluted using a buffer containing EGTA. The fragments were separated by SDS-PAGE and detected by autoradiography. For F 1-42 , the pulldown assay was performed with CaM-Sepharose in the presence of either Ca 2ϩ or EGTA as detailed under "Experimental Procedures." F 1-42 was detected by Western blot analysis using anti-GST. L, AKT protein fragment; FT, flow-through; W, last wash before elution; and E, elution. A, list of fragments that are relatively longer; B, list of fragments that are shorter. endogenous AKT1 could be pulled down by CaM-Sepharose. Fig. 5 shows that the AKT1 from human HeLa cells complexed with CaM in a Ca 2ϩ -dependent fashion. When EGTA was present in the pulldown buffer, the binding of AKT1 to CaM-Sepharose was totally disrupted. As a positive control, a well known CaM-binding protein, calcineurin, was shown to bind to CaM through the pulldown assay. We further investigated whether the interaction exists in vivo by co-immunoprecipitation when both CaM and AKT1 are present at their physiological concentrations. Fig. 6A illustrates that AKT1 was co-immunoprecipitated with CaM when anti-CaM antibody was used for the immunoprecipitation. As a positive control, calcineurin was found to be co-immunoprecipitated with CaM as previously reported. As a negative control, we checked an irrelevant TFII protein and found it was not co-immunoprecipitated with CaM. Conversely, CaM was co-immunoprecipitated with AKT1 when anti-AKT1 was used in the immunoprecipitation (Fig. 6B). We also performed immunoprecipitation using an IgG and could not detect the presence of CaM or AKT1 by probing the Western blots using corresponding antibodies (data not shown). These results indicate that the AKT1⅐Ca 2ϩ ⅐CaM complex is present in vivo. Since direct binding of AKT1 with CaM has been detailed in vitro, the observed interaction with CaM is most likely to be direct rather than mediated by other proteins.
To investigate the possible role of CaM-binding on AKT1, we first examined whether the CaM binding motif near the N terminus of the kinase catalytic domain affects its kinase activity. To address this question, the phosphorylation of biotinylated cross-tide catalyzed by either active or inactive recombinant human AKT1 was performed in the presence of [␥-32 P]ATP. The kinase was preincubated with Ca 2ϩ , Ca 2ϩ ⅐CaM, or EGTA/ CaM, respectively, before being used in the phosphorylation reaction. Recombinant AKT1 proteins with PH domain deleted were also used in the assay. No change of phosphorylation was detected when Ca 2ϩ ⅐CaM was present (data not shown), indicating the catalytic activity of AKT1 is not regulated by CaM.
We then focused on examining the CaM binding motif at the PH domain of AKT1. To address this question, we investigated whether the interaction between Ca 2ϩ ⅐CaM and the full-length AKT1 could be disrupted by the free PH domain of AKT1.
Although the three-dimensional structure of the full-length AKT has not yet been reported, there exist several structures for the separate AKT and the PH domain (8, 38 -41). It was modeled that the C-terminal helix of the PH domain extends about three turns beyond the consensus core, implying a well separated PH domain from the AKT catalytic region (8). We hypothesized that if the PH domain of AKT directly interacts with Ca 2ϩ ⅐CaM, the formation of the AKT⅐Ca 2ϩ ⅐CaM complex should be disrupted when a free PH domain of AKT is included in the binding mixture. Indeed, compared with that when no PH domain was added (Fig. 7A, panel 1), the amount of fulllength AKT1 that could be pulled-down by CaM-Sepharose was significantly decreased when free PH domain was included in the binding buffer (Fig. 7A, panel 3), similar to that when the interaction was completely disrupted by chelating Ca 2ϩ with EGTA (Fig. 7A, panel 2). This was presumably due to competitive binding from the externally added PH domain with CaM. Furthermore, it was found that the full-length AKT1 pre-complexed with Ca 2ϩ ⅐CaM could be eluted when an excess amount of free PH domain was used to replace EGTA in the elution  buffer (Fig. 7B, panel 1). Significantly, a high concentration of free PH domain (ϳ20 M) was required to elute the full-length AKT1, implying that the AKT1⅐Ca 2ϩ ⅐CaM complex is very stable once formed or that the second CaM binding site is contributing to the stabilization of the complex.
Although the amino acid sequences of the Ca 2ϩ ⅐CaM binding motifs in numerous CaM-binding target proteins are extremely diverse, many Ca 2ϩ ⅐CaM-binding proteins have one or more CaM-binding regions that are characterized by a basic amphipathic helix with ϳ20 amino acids in length, as that structurally detailed in myosin light chain kinase (MLCK) (37,42). To determine if the PH domain of AKT1 competes with other CaM-binding proteins for the same binding region(s) on CaM, a short peptide from the CaM binding motif of MLCK was used to compete with full-length AKT1 in Ca 2ϩ ⅐CaM binding assay. This peptide was known to bind to Ca 2ϩ ⅐CaM with very high affinity (K d ϳ 6 pM) (43). Surprisingly, unlike the PH domain, this Ca 2ϩ ⅐CaM-binding peptide could not competitively elute AKT1 from its complex pre-formed with Ca 2ϩ ⅐CaM (Fig. 7B, panel 2), indicating that the PH domain of AKT1 bind to Ca 2ϩ ⅐CaM at regions that are different from the conventional sites.
The PH domain of AKT is highly conserved among the three AKT isoforms (supplemental Fig. S3). Therefore, we examined whether other human AKT isoforms also interact with CaM in a similar fashion. Fig. 8 illustrates that CaM generally binds to AKT family members, including AKT2 and AKT3 in a Ca 2ϩdependent manner. Significantly, all the binding with CaM was completely abolished when EGTA was used to chelate Ca 2ϩ , indicating that Ca 2ϩ ⅐CaM complex is necessary for the observed interactions. Because the first 42 residues of the PH domain of AKT isoforms are ϳ74% identical (supplemental Fig.  S3), it is likely that all three AKT isoforms interact with Ca 2ϩ ⅐CaM at the same regions through similar mechanism.
The major role of the PH domain of AKT is to recruit AKT from cytosol to plasma membrane through its interaction with PtdIns(3,4,5)P 3 . To address whether the interaction with Ca 2ϩ ⅐CaM affects the phosphoinositide-binding property of the PH domain of AKT, we first examined the complex formation between the PH domain of AKT1 and Ca 2ϩ ⅐CaM in the presence or absence of a phosphoinositide. It was found that the amount of the PH domain that could be pulled down by CaM-Sepharose was decreased Ͼ60% when PtdIns(3,4,5)P 3 was present (data not shown), implying that Ca 2ϩ ⅐CaM and PtdIns(3,4,5)P 3 directly compete for the PH domain of AKT. We further investigated whether the presence of Ca 2ϩ ⅐CaM disrupted the interaction between the PH domain of AKT and various phosphoinositides, using a protein lipid overlay assay. To this end, a strip membrane spotted with ϳ100 pmol of each PIP was used for the protein lipid overlay experiment with various amounts of full-length AKT1, the PH domain of AKT1, and AKT1 with PH domain deleted. Fig. 9A illustrates that the interactions between the PH domain of AKT1 and various phosphoinositides, including PtdIns(3,4,5)P 3 ,PtdIns(3,4)P 2 ,   PtdIns(4,5)P 2 , and PtdIns(3,5)P 2 , were significantly reduced when Ca 2ϩ ⅐CaM was included in the binding buffer. Such interruption of interaction between AKT1 and PIP was dependent on the concentrations of CaM used in the overlay assay (data not shown). The overlay assay was performed several times, and results were reproducible. The inhibition by Ca 2ϩ ⅐CaM to each PIP was quantitatively analyzed by averaging the results from different overlay assays. For PtdIns(3,4,5)P 3 , Ca 2ϩ ⅐CaM treatment lowered the full-length AKT1 binding from 100 to 20% when compared with EGTA/CaM treatment blot. Such interaction was not due to calcium, because there is no significant difference between Ca 2ϩ and EGTA/CaM treatments in terms of the amount of full-length AKT1 binding to PtdIns(3,4,5)P 3 . For PtdIns(3,4)P 2 , the addition of Ca 2ϩ ⅐CaM decreased the fulllength AKT1 binding to 20% of EGTA/CaM and Ca 2ϩ treatments. When the PH domain of AKT1 was used in the lipid overlay assay, its binding to PtdIns(3,4,5)P 3 decreased from 100% to 30% with the addition of Ca 2ϩ ⅐CaM. Only ϳ20% of the PH domain of AKT1 bound to PtdIns(3,4)P 2 in the presence of Ca 2ϩ ⅐CaM, compared with that with EGTA/CaM. Interestingly, PH domain of AKT1 bound to PtdIns(4,5)P 2 , but the binding was reduced to only 10% after the addition of Ca 2ϩ ⅐CaM. No significant difference in terms of the amount of PH domain of AKT1 bound to each of the lipids was observed between EGTA/CaM-and Ca 2ϩ -treated samples. Consistent with the findings reported in the literature, the interaction of AKT1 with phosphoinositides is PH domain-dependent, and was completely abolished when the PH domain was deleted from AKT1 protein (Fig. 9, panel 3). These results demonstrated that the interaction between the PH domain of AKT1 and phosphoinositides could be conditionally disrupted by CaM when the concentration of Ca 2ϩ is elevated. We further examined the inhibition of the binding of the PH domain of AKT1 to a dilution series of PtdIns(3,4,5)P 3 by Ca 2ϩ ⅐CaM. As shown in Fig. 9B, the inhibition is highly dependent on the PtdIns(3,4,5)P 3 concentrations. Although the interaction of the PH domain of AKT1 with PtdIns(3,4,5)P 3 was still clearly detected when PtdIns(3,4,5)P 3 was used at 13 pmol, the interaction was significantly inhibited by Ca 2ϩ ⅐CaM in the whole concentration range.

DISCUSSION
Among a number of Ca 2ϩ sensors in the eukaryotic cell, CaM is the most widespread and the best studied. CaM interacts with many different protein targets and mediates a wide variety of cellular functions. Most CaM-mediated signaling pathways are through the Ca 2ϩ ⅐CaM complex, which is highly modulated by intracellular Ca 2ϩ concentration (44 -49). Two distinct CaM binding motifs on AKT1 were thoroughly mapped using protein fragments that cover all the possible regions of AKT1. In addition, those regions that contain a CaM binding motif were carefully examined using multiple smaller fragments, to further determine the location of the CaM binding motifs. Taken together, the observed CaM binding can be attributed to two well defined CaM-binding regions, one at the first half of the PH domain (F  ) and the other near the N terminus of the kinase catalytic domain (F 150 -197 ). We found that all the fragments that contain one of the mapped CaM binding motifs bound to CaM in a Ca 2ϩ -dependent manner, whereas all the fragments that contain none of the mapped CaM binding motifs could not interact with CaM. Therefore, such CaM binding is less likely due to nonspecific interaction. Indeed, it is not unusual that CaM-binding proteins contain more than one CaM-binding domain, and the interactions of these proteins to CaM are well known to be specific (37). The presence of a second CaM binding motif near the N terminus of the catalytic domain of human AKT1 is intriguing. It appears that C. elegans AKT1 also contains a Ca 2ϩ ⅐CaM binding motif close to this region. However, the exact location is different from that on human AKT1. Therefore, it is likely that AKT proteins from different species are regulated by Ca 2ϩ ⅐CaM at close but not identical regions. It is also possible that this is not a function that has been conserved in C. elegans. The significance of the Ca 2ϩ ⅐CaM binding motif near the N terminus of AKT1 catalytic domain remains to be determined. We found that the catalytic activity of human AKT1 was not affected by Ca 2ϩ ⅐CaM. We postulate that the Ca 2ϩ ⅐CaM regulation affects the activation or subcellular location of AKT1, rather than the kinase activity of the enzyme directly.
The PH domain of AKT has a ␤-barrel shape that is closed at one end by the C-terminal ␣-helix, whereas at the other end lie three loops that are variable, both in sequence and length, in all known PH domains (39). These loops form a highly basic pocket into which the 3-and 4-phosphates of PtdIns(3,4)P 2 and PtdIns(3,4,5)P 3 can bind. Specifically, before the binding of the lipid, the ligand-binding pocket in the PH domain is occupied by a complex hydrogen bonding network. Upon binding of a lipid such as Ins(1,3,4,5)P 4 , this network of hydrogen bonding is disrupted. In the 1-42 region, Tyr-18 moves away from the ligand-binding pocket to create space for the 5-phosphate, Glu-17 moves outwards to enable the contact of Arg-86 with the 4-phosphate, and Arg-23 moves inwards to make contact with the 1-and 3-phosphates. As for the residues close to the 1-42 region, a short acidic patch is formed through the clustering of three negatively charged residues (Asp-44, Asp-46, and Glu-49) upon ligand binding. We found that the binding of the PH domain of AKT1 could be in part attributed to residues 1-42 (Fig. 4). Because this region and its adjacent residues are critical in lipid ligand recognition, it is not surprising that the relatively weak interaction between the PH domain of AKT and the lipid ligands could be disrupted by Ca 2ϩ ⅐CaM.
The PH domain is ϳ80% identical among the AKT isoforms, implying that other AKT isoforms also bind to Ca 2ϩ ⅐CaM similar to AKT1. Indeed, AKT2 and AKT3 appear to interact with CaM in a manner that is highly dependent on Ca 2ϩ . These results indicate that the interaction between Ca 2ϩ ⅐CaM and the PH domain of AKT is not just limited to AKT1, but could be generally applied to all the three isoforms of AKT kinase. PH domains are found in many proteins involved in cellular signaling (50). However, AKT was the only PH domain-containing protein isolated from our Ca 2ϩ ⅐CaM-binding selections using mRNA-displayed proteome libraries.
Our finding that AKT1 is a Ca 2ϩ ⅐CaM-binding protein implies that the activity and functions of AKT1 are directly regulated by the calcium signal. It is well known that the activated AKT1 is released from the plasma membrane to switch