The EMBO Journal (2004) 23, 3918–3928 www.embojournal.org |& 2004 European Molecular Biology Organization | All Rights Reserved 0261-4189/04 THE EMBO JOURNAL Structural insights into the regulation of PDK1 by phosphoinositides and inositol phosphates David Komander1,3, Alison Fairservice2, Maria Deak3, Gursant S Kular3, Alan R Prescott4, C Peter Downes2, Stephen T Safrany2, Dario R Alessi3 and Daan MF van Aalten1,* 1 Division of Biological Chemistry & Molecular Microbiology, School of Life Sciences, University of Dundee, Dundee, Scotland, UK, 2 Division of Cell Signalling, School of Life Sciences, University of Dundee, Dundee, Scotland, UK, 3MRC Protein Phosphorylation Unit, School of Life Sciences, University of Dundee, Dundee, Scotland, UK and 4 Division of Cell Biology and Immunology, School of Life Sciences, University of Dundee, Dundee, Scotland, UK 3-phosphoinositide-dependent protein kinase-1 (PDK1) phosphorylates and activates many kinases belonging to the AGC subfamily. PDK1 possesses a C-terminal pleckstrin homology (PH) domain that interacts with PtdIns(3,4,5)P3/ PtdIns(3,4)P2 and with lower affinity to PtdIns(4,5)P2. We describe the crystal structure of the PDK1 PH domain, in the absence and presence of PtdIns(3,4,5)P3 and Ins(1,3,4,5)P4. The structures reveal a ‘budded’ PH domain fold, possessing an N-terminal extension forming an integral part of the overall fold, and display an unusually spacious ligand-binding site. Mutagenesis and lipid-binding studies were used to define the contribution of residues involved in phosphoinositide binding. Using a novel quantitative binding assay, we found that Ins(1,3,4,5,6)P5 and InsP6, which are present at micromolar levels in the cytosol, interact with full-length PDK1 with nanomolar affinities. Utilising the isolated PDK1 PH domain, which has reduced affinity for Ins(1,3,4,5,6)P5/InsP6, we perform localisation studies that suggest that these inositol phosphates serve to anchor a portion of cellular PDK1 in the cytosol, where it could activate its substrates such as p70 S6-kinase and p90 ribosomal S6 kinase that do not interact with phosphoinositides. The EMBO Journal (2004) 23, 3918–3928. doi:10.1038/ sj.emboj.7600379; Published online 30 September 2004 Subject Categories: structural biology; signal transduction Keywords: inositol phosphates; phosphoinositides; PI-3 kinase pathway; protein crystallography; protein structure Introduction 3-phosphoinositide-dependent protein kinase-1 (PDK1) plays an important role in insulin and growth factor signalling cascades (Toker and Newton, 2000; Mora et al, 2004). It is *Corresponding author. Wellcome Trust Biocentre, School of Life Sciences, University of Dundee, MSI/WTB Complex, Dundee DD1 5EH, UK. Tel.: þ 44 1382 344 979; Fax: þ 44 1382 345 764; E-mail: dava@davapc1.bioch.dundee.ac.uk Received: 2 June 2004; accepted: 29 July 2004; published online: 30 September 2004 3918 The EMBO Journal VOL 23 | NO 20 | 2004 known to phosphorylate and activate at least 24 protein kinases that belong to the AGC (cAMP-dependent, cGMPdependent, protein kinase C (PKC)) family of protein kinases. These comprise isoforms of protein kinase B (PKB, also known as Akt), p70 ribosomal S6-kinase (S6K), serum and glucocorticoid responsive kinase (SGK), p90 ribosomal S6 kinase (RSK) and PKC. PDK1 activates these enzymes by phosphorylating a Ser/Thr residue in their activation loop. PDK1 is a 556 amino-acid protein comprising an N-terminal Ser/Thr kinase catalytic domain (residues 71–359) and a C-terminal pleckstrin homology (PH) domain (residues 459–550) (Alessi et al, 1997; Stephens et al, 1998). The latter binds with high affinity to the second messenger phosphatidylinositol-3,4,5-trisphosphate (PtdIns(3,4,5)P3) (Stephens et al, 1998; Currie et al, 1999), generated through insulinand growth factor-induced activation of PI 3-kinase (Vanhaesebroeck et al, 2001). Much research has shown that insulin and growth factors control PDK1 by regulating its interaction with substrates, rather than by directly activating PDK1. PKB, in contrast to other PDK1 substrates, possesses a PtdIns(3,4,5)P3-binding PH domain. The mutual ability of PKB and PDK1 to interact with PtdIns(3,4,5)P3 via their PH domains is likely to play a crucial role in regulating PKB activation (Anderson et al, 1998; Currie et al, 1999; Filippa et al, 2000). Consistent with this notion, PKB was not substantially activated in a knockin ES cell line expressing an active form of PDK1 that cannot interact with PtdIns(3,4,5)P3 (McManus et al, 2004). Recent structural and cellular studies have also indicated that binding of PKB to PtdIns(3,4,5)P3 induces a major conformational change (Mora et al, 2004) that is likely to be required for PKB activation by PDK1 (Alessi et al, 1997; Stokoe et al, 1997). Importantly, binding of PtdIns(3,4,5)P3 to the PDK1 PH domain does not directly affect activity of the PDK1 kinase domain, but stimulates PKB activation by inducing the colocalisation of PKB and PDK1 (Stephens et al, 1998; Currie et al, 1999). The other substrates that PDK1 activates, such as S6K, RSK and SGK, are phosphorylated by PDK1 in the cytosol, rather than at the plasma membrane. Moreover, these substrates do not possess PH domains or interact with phosphoinositides. Extracellular agonists regulate activation of S6K, RSK and SGK through PDK1, by inducing the phosphorylation of these enzymes at a conserved C-terminal (noncatalytic) hydrophobic motif, which creates a high-affinity docking site for interaction with a pocket located on the kinase domain of PDK1 (reviewed in Mora et al, 2004). This has recently been supported by a knockin cell line in which the PDK1 hydrophobic motif docking site is disrupted (Collins et al, 2003). The PDK1 PH domain sequence is unusual compared to other PH domains and several computer algorithms, such as the Simple Modular Architecture Research Tool or PROSITE, fail to classify it as a PH domain. However, lipid-binding studies have firmly established that PDK1 binds PtdIns(3,4,5)P3 through a C-terminal domain, and truncation & 2004 European Molecular Biology Organization Structure and function of the PDK1 PH domain D Komander et al or mutation of several conserved residues found in this domain abolishes PtdIns(3,4,5)P3 binding (Anderson et al, 1998; Currie et al, 1999). In addition to binding PtdIns(3,4,5)P3, PDK1 also binds with similar affinity to PtdIns(3,4)P2, a potential signalling lipid generated through the dephosphorylation of PtdIns(3,4,5)P3 by the SHIP 5-phosphatase (Stephens et al, 1998; Currie et al, 1999). Surface plasmon resonance-based lipid-binding assays revealed that PDK1 also interacts with PtdIns(4,5)P2, the substrate for PI 3-kinase, albeit with a 15-fold lower affinity compared to PtdIns(3,4,5)P3 (Currie et al, 1999). This weaker affinity for PtdIns(4,5)P2, may be significant as the basal cellular levels of this phospholipid are 50–200-fold higher than those of PtdIns(3,4,5)P3/PtdIns(3,4)P2 in insulin-stimulated cells (Batty and Downes, 1996). Consistent with this notion, quantitative electron microscopy localisation studies revealed that in unstimulated cells B10% of the cellular pool of PDK1 is associated with the plasma membrane in a manner dependent on an intact PH domain, which could be mediated through the ability of PDK1 to bind PtdIns(4,5)P2 (Currie et al, 1999). Whether insulin or growth factor stimulation induces further localisation of PDK1 at the plasma membrane is controversial (Anderson et al, 1998; Currie et al, 1999). Nevertheless, a significant pool of PDK1 is required in the cytoplasm to activate its other non-lipid-binding substrates such as S6K, RSK and SGK. The mechanism by which PDK1 is anchored in the cytoplasm is not known. Here we define the molecular basis by which PDK1 interacts with its phosphoinositide ligands and show that PDK1 may be anchored in the cell cytosol through its interaction with D6-phosphorylated inositol phosphates. Results and discussion Structure determination of PDK1 PH domain Constructs containing the predicted PH domain of PDK1 designed on the basis of sequence alignments with previously crystallised PH domains produced insoluble protein. To identify the boundaries of a soluble fragment of PDK1 containing the PH domain, a limited proteolysis approach was performed (described in Komander et al, 2004), resulting in a proteolytically stable fragment of PDK1 comprising residues 408–556. We cloned and expressed in Escherichia coli a fragment of PDK1 encompassing residues 409–556, which showed good stability after purification and was readily crystallisable (Komander et al, 2004). The protein crystals, grown in the presence or absence of phosphoinositides, diffracted to below 1.50 Å on a rotating anode generator. The single cysteine in the sequence (Cys536) was exploited to obtain experimental phases, by derivatisation with AuCN. The structure of the PDK1 PH domain in complex with Ins(1,3,4,5)P4 was solved by a two-wavelength MAD experiment (Table I). The resulting map allowed building of residues 409–549, and showed well-defined electron density for Ins(1,3,4,5)P4 (Figure 1A). Overall structure of the ‘budded’ PDK1 PH domain The structure of the PDK1 PH domain reveals a structural variation of a standard PH domain fold (Lemmon and Ferguson, 2000), with an additional ‘bud’ at the N-terminus (Figures 1B–D). Residues 456–530 fold into two, almost orthogonal, b sheets, one consisting of four (b1–b4) and & 2004 European Molecular Biology Organization one of three strands (b5–b7), constituting a barrel-like structure (Figures 1B–D). On one side, the barrel is blocked by a C-terminal a helix (residues 531–549, Figures 1B and C). The other side of the barrel consists of three loops (termed variable loop (VL) 1–3 (Ferguson et al, 1994)), forming a bowl-like structure which is lined by positively charged residues, constituting the phosphoinositide-binding site. N-terminal to the observed standard PH domain fold, the PDK1 PH domain possesses an extension, here referred to as the ‘bud’. This consists of two additional b strands, which extend the b1–b4 sheet in an antiparallel fashion, and an additional a helix, which packs against the newly formed sixstranded b sheet (Figures 1B and D). The bud packs against the b1–b4 sheets of the PH domain, through several hydrophobic contacts (Figure 1D), but also through b-sheet hydrogen bonds between strands b20 and b1. Overall, the bud of the PDK1 PH domain buries more than 30% of the surface of the standard PH domain fold (567 Å2 of 1749 Å2), and a hydrophobic core is created between the a10 helix and residues on one face of the six-stranded b sheet (Figure 1D). Residues Ile416, Phe424, Leu426, Leu456, Ile457, Met460, Leu477, Tyr485, Val487 and Leu494 from the six-stranded b sheet interact with residues Ile412, Phe430, Leu438, Leu439, Trp448, Phe451, Val452 and Leu456 from the a10 helix and surrounding loops (Figure 1D). The presence of this additional hydrophobic core, which lies outside the standard PH domain fold, suggests that the bud is an integral part of the overall fold. This could explain the insolubility of protein obtained from initial constructs which covered the standard PH domain only (residue 458–556) (Komander et al, 2004). A bud-like extension of a PH domain fold as observed in PDK1 has not been described to date in any other crystallised PH domain containing protein. Although interactions of PH domains with other proteins or protein domains have been described in structures of Gbg binding to the GRK2 PH domain (Lodowski et al, 2003), myotubularin-related protein phosphatase MTMR2 (Begley et al, 2003), or the Rac1-Tiam1like complexes (Worthylake et al, 2000), none of these show the antiparallel b-sheet extension observed in the PDK1 PH domain and, therefore, do not resemble the bud on PDK1. Interactions with the PtdIns(3,4,5)P3 head group, Ins(1,3,4,5)P4 The structure of the PDK1 PH domain was determined from a crystal that was grown in the presence of Ins(1,3,4,5)P4, and well-defined ligand density could be observed in the experimental maps (Figure 1A, Table I). The PDK1 PH domain binds Ins(1,3,4,5)P4 in a location equivalent to that described for other PH domains (Baraldi et al, 1999; Ferguson et al, 2000; Lietzke et al, 2000; Thomas et al, 2002), that is, within a shallow but positively charged pocket formed by VL1-3, at the open end of the b barrel (Figure 1). The phosphoinositidebinding site is lined by positively charged residues that contact the phosphates of Ins(1,3,4,5)P4, and a total of 11 hydrogen bonds are possible between the PDK1 PH domain and Ins(1,3,4,5)P4 (Figure 2A). Arg472 contacts the phosphate at the D1 position (D1-phosphate) with two hydrogen bonds. The D3-phosphate is contacted by Arg474 (two hydrogen bonds) and Lys465. Lys495 is in close proximity (4.4 Å) to the D3-phosphate. The D4-phosphate forms up to four hydrogen bonds to Lys465, Tyr486, and Arg521. Finally, the D5-phosphate makes one hydrogen bond to Lys467 (Figures The EMBO Journal VOL 23 | NO 20 | 2004 3919 Structure and function of the PDK1 PH domain D Komander et al Table I Data collection and refinement statistics PDK1 PH domain –Ins(1,3,4,5,)P4 Beamline Wavelength (Å) Space group Unit cell (Å) Resolution (Å) Observed reflections Unique reflections Redundancy Completeness (%) Rmerge I/sI Reflections in test set Rcryst Rfree Phasing statistics /FOMS Phasing power PDK1 PH domain uncomplexed PDK1 PH domain– PtdIns(3,4,5)P3 BM-14 l1 0.861 P21 a ¼ 35.40 b ¼ 58.92 c ¼ 36.58 b ¼ 101.481 BM-14 l2 1.040 P21 a ¼ 35.42 b ¼ 58.94 c ¼ 36.58 b ¼ 101.471 Rotating anode Rotating anode 1.54 P21 a ¼ 42.69 b ¼ 65.28 c ¼ 101.12 b ¼ 97.481 1.54 P21 a ¼ 36.26 b ¼ 58.94 c ¼ 36.44 b ¼ 102.341 25.0–1.50 (1.58–1.50) 85433 23341 3.7 (3.3) 98.6 (99.5) 0.076 (0.439) 17.7 (3.2) 617 0.150 0.204 25.0–1.80 (1.86–1.80) 33500 13556 2.5 (2.1) 95.5 (95.6) 0.043 (0.181) 17.3 (6.4) 30.0–1.45 (1.50–1.45) 422700 95063 4.5 (3.0) 97.5 (80.1) 0.048 (0.397) 13.2 (2.7) 934 0.171 0.211 30.0–1.45 (1.50–1.45) 73746 25810 2.9 (2.7) 99.7 (100) 0.054 (0.372) 8.8 (2.5) 545 0.185 0.237 0.44 (25.0–2.5 Å) 1.11 (25.0–2.5 Å) Number of groups Protein residues Water Ligand atoms SO2 4 atoms Glycerol atoms Wilson B (Å2) /BS protein (Å2) /BS water (Å2) /BS ligand (Å2) 143 217 28 — 6 14.5 14.6 27.6 14.6 573 715 — 50 24 19.5 18.6 35.8 — 143 163 33 — — 23.0 30.3 45.4 36.2 R.m.s.d. from ideal geometry Bond length (Å) Bond angles (deg) Main chain B (Å2) 0.020 1.8 1.6 0.015 1.6 1.5 0.016 1.7 1.7 Values between brackets are for the highest resolution shell. All measured data were included in structure refinement. 2A and 4A). Neither the D2- nor the D6-hydroxy groups show any direct hydrogen bonds to the protein (Figure 2A). Figure 2 shows a comparison of the structures of the PDK1, DAPP1 (Ferguson et al, 2000) and PKBa (Thomas et al, 2002) PH domains in complex with Ins(1,3,4,5)P4. DAPP1 binds Ins(1,3,4,5)P4 in a similar fashion to GRP1 (Ferguson et al, 2000; Lietzke et al, 2000) and Btk (Baraldi et al, 1999). The orientation of Ins(1,3,4,5)P4 in PDK1 is identical to that found in DAPP1/Btk/GRP1 (Figure 2A and B), while in PKBa the Ins(1,3,4,5)P4 molecule is rotated by 601 (Figure 2C) (Thomas et al, 2001). However, in all PH domain Ins(1,3,4,5)P4 complex structures described to date, Ins(1,3,4,5)P4 interacts with both protein side chains as well as the backbone, whereas in PDK1 Ins(1,3,4,5)P4 interacts with protein side chains only. This results in a significantly reduced number of protein–ligand hydrogen bonds in the PDK1 PH domain Ins(1,3,4,5)P4 complex (a total of 11), compared to the PKBa (15 hydrogen bonds), DAPP1 (15 hydrogen bonds), Grp1 (16 hydrogen bonds) and Btk (16 hydrogen bonds) PH domain–Ins(1,3,4,5)P4 complexes (Figure 2). A major difference between the PH domain– Ins(1,3,4,5)P4 complexes is the position of VL1 (Figure 2D). In the structures of the DAPP1/GRP1/Btk PH domains, VL1 envelopes the D5-phosphate group, which in the case of 3920 The EMBO Journal VOL 23 | NO 20 | 2004 DAPP1 forms four backbone hydrogen bonds with VL1. In the case of the PDK1 PH domain, the D5-phosphate forms only a single hydrogen bond to Lys467, the residue preceding VL1, and no further direct interactions between VL1 and the inositol headgroup are observed (Figure 2A). In the case of PKBa, due to different orientation of Ins(1,3,4,5)P4, VL1 interacts with the D1-phosphate, leaving the D5-phosphate solvent exposed (Figure 2C). For DAPP1/GRP1/Btk and PKBa, backbone hydrogen bonds between the D2-hydroxyl groups are observed. In the case of PDK1, there is no direct interaction of the D2-hydroxyl group with the protein; instead, we observe it binding to the protein via an ordered water molecule. Strikingly, in the PDK1 Ins(1,3,4,5)P4-binding pocket, a layer of five-ordered water molecules (B factors 10.7–23.4 Å2) separate Ins(1,3,4,5)P4 from the protein. The water molecules mediate a number of hydrogen bonds from Ins(1,3,4,5)P4 to protein side chains and backbone (Figure 2A). One of the water molecules is consistently found in the other PH domain–Ins(1,3,4,5)P4 complexes, in a similar position and contacting the D3-phosphate (coloured yellow, Figure 2). The presence of the four additional ordered water molecules is surprising, as this would appear to be entropically unfavourable. Overall, the PDK1 PH domain phosphoinositide pocket is significantly more spacious than & 2004 European Molecular Biology Organization Structure and function of the PDK1 PH domain D Komander et al A B Arg521 Lys467 β1' α1' β2' Tyr486 α1 β1 Lys465 β2 β5 β3 Lys495 Arg474 β6 β4 Arg472 β7 VL2 VL3 C D α1 C N VL1 β5 β1' α1' β6 β1' α1 VL3 β2' β7 β1 VL1 α1' β2' β2 β3 α1' β1 β2 β2 β4 β3 VL2 β4 β3 β4 Figure 1 (A) Electron density maps calculated from experimental phases after solvent flattening, contoured at 0.6s. Blue electron density covers the protein, the density for the phosphoinositide ligand is shown in green, for clarity. The refined, final model is shown as stick representation. (B) Ribbon representation of the PDK1 PH domain (residues 409–549). The standard PH domain fold is coloured green and red (b strands and the C-terminal a helix, respectively). The extension on the N-terminus is coloured blue. The Ins(1,3,4,5)P4 ligand is shown as a stick representation. A coloured, semitransparent surface covers the protein molecule. (C) Topology diagram of the PH domain of PDK1, coloured according to B. (D) Hydrophobic interactions of the six-stranded b sheet with the N-terminal a10 helix. Hydrophobic residues on the six-stranded b sheet and on the a10 helix are shown as a stick representation, as is the Ins(1,3,4,5)P4 ligand. that of other Ins(1,3,4,5)P4-binding PH domains. Perhaps, this more spacious binding pocket explains the observed ability of the PDK1 PH domain to bind different stereoisomers of phosphoinositides (Stephens et al, 1998). Interactions of the PDK1 PH domain with diC4-PtdIns(3,4,5)P3 It is possible that some of the water molecules close to the D1-phosphate are replaced by the glycerol backbone upon PtdIns(3,4,5)P3 binding. To investigate this, the PDK1 PH domain was co-crystallised with a PtdIns(3,4,5)P3 analogue, containing two C4 acyl chains (diC4-PtdIns(3,4,5)P3). & 2004 European Molecular Biology Organization Crystals were obtained under similar conditions as for the Ins(1,3,4,5)P4 complex, and were of the same space group with similar unit cell dimensions (Table I). After refinement of the protein model, well defined |Fo|–|Fc| density was visible for the inositol ring, the phosphates and the glycerol backbone (Figure 3A). A partial model for diC4PtdIns(3,4,5)P3 was then included in the refinement (Table I). Analysis of the resulting model shows that while Arg472 coordinates the free oxygen atoms on the D1-phosphate, the oxygen atom involved in the ester bond to the glycerol does not make any significant contact with the protein. Of the lipid moiety, only the glycerol backbone is The EMBO Journal VOL 23 | NO 20 | 2004 3921 Structure and function of the PDK1 PH domain D Komander et al A Arg474 Arg474 Lys465 Arg472 D3 Lys465 Arg472 D4 D3 D4 D5 D5 D1 D1 Lys467 Lys467 VL1 VL1 PDK1 B Arg184 Arg184 Lys173 D4 D3 D3 Lys173 D4 D5 D5 D1 D1 VL1 VL1 DAPP1 C Arg25 Arg25 Lys14 D4 D3 Lys14 D4 D5 D3 D1 D1 VL1 VL1 D5 PKBα D DAPP1 DAPP1 PKBα PDK1 PKBα PDK1 Figure 2 Comparison of Ins(1,3,4,5)P4 binding to the PH domains of PDK1, DAPP1 and PKBa. Interactions of Ins(1,3,4,5)P4 (marine) with protein residues (green in PDK1, orange in DAPP1/PKBa) in the phosphoinositide-binding site are shown in a stereo representation. Hydrogen bonds are indicated as black dotted lines, and conserved water molecules are shown as yellow spheres. (A) Protein–ligand interactions of PDK1 PH domain with Ins(1,3,4,5)P4. A layer of ordered water molecules (coloured in magenta) separates the ligand from VL1, only one of which is conserved in other PH domain structures (coloured in gold). (B) Ins(1,3,4,5)P4 binding to DAPP1 (pdb-id 1fao; Ferguson et al, 2000). The D5-phosphate is closely enveloped by VL1. (C) Ins(1,3,4,5)P4 binding to PKBa (pdb-id 1h10; Thomas et al, 2001). The ligand is rotated and VL1 contacts the D1-phosphate, while the D5-phosphate is solvent exposed. (D) Overlay of the structures of the PDK1 (green), DAPP1 (orange) and PKBa (blue) PH domains. The Ins(1,3,4,5)P4 ligand and the layer of ordered water molecules of PDK1 are shown, and coloured according to (A). ordered in the crystal structure (Figure 3A). The glycerol backbone projects away from the surface of the protein and does not display any other interactions. Currently, the only other example of a protein domain in complex with an ordered phosphoinositide is the PtdIns(3)Pbinding phox homology (PX) domain of the p40phox subunit 3922 The EMBO Journal VOL 23 | NO 20 | 2004 of NADPH oxidase (pdb-id 1h6h; Bravo et al, 2001). In Figure 3, the structures of the PDK1 PH domain–diC4– PtdIns(3,4,5)P3 and the PX domain–diC4–PtdIns(3)P complexes are compared after superposition on the inositol ring. Despite possessing completely different folds, the binding arrangements by which these proteins interact with the D1-phosphodiester are strikingly similar. The two oxygen atoms not involved in the two phosphoester linkages will carry most of the 1e negative charge, and it is these oxygen atoms that interact with basic residues in the underlying binding pocket on the surface of the proteins. In the PDK1 PH domain, the guanidinium group of Arg472 hydrogen bonds both these oxygen atoms, while in the PX domain each oxygen atom interacts separately with a basic residue (Arg60/Lys92). Consequently, the oxygen atom involved in the phosphoester bond to the glycerol projects away from the binding pocket and in neither structure interacts with the protein. However, in contrast to the PDK1 PH domain diC4– PtdIns(3,4,5)P3 complex, the glycerol backbone in the PX domain–diC4–PtdIns(3)P complex displays hydrophobic interactions with an exposed tyrosine side chain (Tyr94), which is not observed in the PDK1 PtdIns(3,4,5)P3 complex (compare Figure 3A and B). To investigate whether the PDK1 PH domain undergoes ligand-induced conformational change when it binds PtdIns(3,4,5)P3, the PDK1 PH domain structure was solved in the absence of any ligand (Table I). However, all four independent molecules in the asymmetric unit show one molecule of sulphate, presumably originating from the 0.2 M ammonium sulphate containing crystallisation buffer, at the position equivalent to the D3-phosphate in the Ins(1,3,4,5)P4 structure (Supplementary Figure 1). Crystal structures of other noncomplexed PH domains, including Grp1 (Ferguson et al, 2000) and PKBa (Milburn et al, 2003), also contain a phosphate/sulphate at equivalent positions. The uncomplexed structures of PDK1 superimpose well onto the Ins(1,3,4,5)P4 and PtdIns(3,4,5)P3 complexes (pairwise with an average RMSD of up to 0.6 Å, Supplementary Figure 1). No significant conformational changes are visible even in the VLs that are involved in ligand binding, with the exception of Lys467, which occupies the position of the inositol ring in the absence of ligand but rotates to a position to contact the D5-phosphate upon ligand binding. This suggests that the PDK1 PH domain, like most other PH domains analysed to date, assumes the same conformation in the presence or absence of phosphoinositides. In contrast, there is evidence that the PKBa PH domain undergoes a conformational change when it binds to PtdIns(3,4,5)P3 (Mora et al, 2004), which is thought to enhance the rate at which PKB can be activated by PDK1 (Vanhaesebroeck and Alessi, 2000). Mutations of the lipid-binding site: functional analysis Figure 4A shows a summary of the ligand–protein hydrogen bonds between PtdIns(3,4,5)P3 and residues of the PDK1 PH domain. To determine the contribution of the interactions to the affinity for phosphoinositides, these residues were mutated in full-length PDK1 and expressed as GST-fusion proteins in HEK-293 cells. The ability of these mutants to bind PtdIns(3,4,5)P3, PtdIns(3,4)P2 and PtdIns(4,5)P2 was first tested with a qualitative protein-phospholipid overlay assay (Dowler et al, 2002) (Figure 4B). As reported previously, & 2004 European Molecular Biology Organization Structure and function of the PDK1 PH domain D Komander et al A C3 C1 O1 C3 C2 O1 Arg472 Arg474 C2 PDK1 Arg472 Arg474 B C1 Tyr94 O1 C1 Tyr94 C3 Acyl-1 O1 C1 C2 C3 Acyl-1 C2 Lys92 L ys92 Acyl-2 Arg58 Acyl-2 Arg58 Arg60 p40 Arg60 phox Figure 3 Analogy of PH and PX domain binding to phosphoinositides. The protein and ligand molecules are coloured as in Figure 2. Under the semitransparent surface, the conserved Arg residues contacting the D1- and the D3-phosphates are drawn as stick representation. (A) Stereo representation of the PH domain of PDK1 interacting with diC4-PtdIns(3,4,5)P3. A |Fo|–|Fc| simulated annealing omit map, contoured at 2.25s, is shown in purple, covering the ligand molecule. (B) PtdIns(3)P binding to the PX-domain of p40phox (pdb-id 1h6h; Bravo et al, 2001). Lys94 and Arg60 contacting the D1-phosphate are drawn as stick representation, as is Tyr94 from the membrane attachment loop. wild-type PDK1 interacts strongly with PtdIns(3,4,5)P3 and PtdIns(3,4)P2, and weakly with PtdIns(4,5)P2 (Currie et al, 1999) (Figure 4B). Only two of the mutants (PDK1[K465E] and PDK1[R472A/R474A]) failed to bind phosphoinositides (Figure 4B). Lys465 and Arg474 of PDK1 are equivalent to Lys14 and Arg25 of PKBa, and Lys173 and Arg184 of DAPP1, where mutation of these residues was previously shown to inhibit the binding to PtdIns(3,4,5)P3 (Dowler et al, 2000; Thomas et al, 2002) (Figure 2). Three mutations (PDK1[R474A], PDK1[R474A/K467A] and PDK1[K465A]) significantly reduced binding (Figure 4B). The individual mutation of Arg472, Tyr486, Lys 495 or Arg521 had little or moderate effects on phosphoinositide binding. Of particular interest was the effect of mutation of Lys467 on the ability of PDK1 to bind D5-phosphorylated phosphoinositides, as Lys467 is the only residue to contact the D5-phosphate (Figures 2 and 4). In order to assess accurately the contribution of Lys467 to ligand binding, we employed a novel time-resolved fluorescence resonance energy transfer (TR-FRET) assay described in Materials and methods, which quantitatively measures binding of PDK1 to phosphoinositides (Figure 5). In agreement with previous surface plasmon resonance (SPR) studies (Currie et al, 1999), in the TR-FRET assays, PtdIns(3,4,5)P3 and PtdIns(3,4)P2 were found to bind PDK1 with significantly higher affinity than PtdIns(4,5)P2 (Figure 5). Affinities obtained are in close agreement for TR-FRET and SPR for both PtdIns(3,4,5)P3 (4.0 and 1.6 nM), and PtdIns(3,4)P2 (23 and 5.2 nM), respectively. In contrast, for PtdIns(4,5)P2, a lower affinity of 700 nM was measured by TR-FRET compared to 24 nM that was calculated from previous SPR studies (Currie et al, 1999). We next used the TRFRET assay to investigate how mutation of Lys467 affected & 2004 European Molecular Biology Organization binding to PtdIns(4,5)P2. Mutation of Lys467 to Ala resulted in an Beight-fold reduction in affinity for PtdIns(4,5)P2, whereas the more conservative replacement of Lys467 to Arg maintained high-affinity PtdIns(4,5)P2 binding. Interestingly, in agreement with the structural data, the PDK1[K467A] mutant interacted with six- to eight-fold lower affinity to all D5-phosphorylated inositides including PtdIns(3,4,5)P3 (Figure 5). In comparison, affinities for PtdIns(3,4)P2 were reduced by two-fold. Binding of PDK1 to inositol phosphates The structures of the PDK1 PH domain show that the phosphoinositide-binding site of PDK1 is unusually spacious, and, in contrast to other PtdIns(3,4,5)P3-binding PH domains, additional space is present around the D2- and D6-hydroxyl groups, which potentially could accommodate further phosphate groups (Figure 3). Physiologically, only inositol phosphates but not phosphoinositides are known to be phosphorylated at the D2 and/or D6 position (Michell, 1997; Irvine and Schell, 2001). Ins(1,3,4,5,6)P5 and InsP6 are present at high micromolar levels within the cell, at higher levels than Ins(1,3,4,5)P4 or Ins(1,4,5)P3 (Orchiston et al, 2004). These observations prompted us to analyse the binding of inositol phosphates to full-length PDK1 employing the TR-FRET assay, which can be used for this purpose. Interestingly, Ins(1,3,4,5,6)P5 and InsP6, as well as Ins(1,3,4,5)P4, were found to bind PDK1 with high affinity, with Ki values of 20–60 nM (Figure 6A). In contrast, Ins(1,4,5)P3 binding to PDK1 could not be detected. The hydrogen bonds to the D1-phosphate (Figures 2–4) appear to play a crucial role in mediating binding of inositol phosphates to PDK1, as Ins(3,4,5,6)P4 had a five-fold lower The EMBO Journal VOL 23 | NO 20 | 2004 3923 Structure and function of the PDK1 PH domain D Komander et al 120 A K495A R521A Y486F R474A % of control bound A D4 D3 K465A K465E D5 80 60 40 20 0 0 D1 R472A/R474A double 100 K467A K467R K467A/R474A double 200 300 400 500 [bPIP3] (nM) 125 B % of control bound R472A B wt PDK1 PDK1-PH PDK1 K467R PDK1 K467A 100 250 100 50 25 12.5 6.3 3.1 nmol PtdIns(3,4,5)P3 PtdIns(3,4)P2 PtdIns(3,4,5)P3 PtdIns(3,4)P2 PtdIns(4,5)P2 PtdIns(3,5)P2 PtdIns(5)P 100 75 50 25 0 −3 PtdIns(4,5)P2 −2 −1 0 1 2 3 log[competing ligand] (µM) PtdIns(3,5)P2 Wild-type R521A C K D for bPIP3 (µM) R472A K495A * R474A Y486F * R472A / R474A * R474A / K467A * K465A * K465E Figure 4 Phosphoinositide-binding properties of wild type and mutant PDK1. (A) Representation of the Ins(1,3,4,5)P4-binding pocket on the PDK1 PH domain. The residues that make contacts with Ins(1,3,4,5)P4 and that are mutated in this study are labelled. (B) The ability of the indicated wild type and mutant forms of PDK1 to interact with phosphoinositides was analysed using a proteinlipid overlay assay. Serial dilutions of the indicated phosphoinositides (250, 100, 50, 25, 12.5, 6.3, 3.1 and 1.6 pmol) were spotted onto nitrocellulose membranes, which were then incubated with the purified GST-PDK1 species. The membranes were washed, and the GST-PDK1 bound to the membrane by virtue of their interaction with lipid was detected using a GST antibody (Dowler et al, 2002). A representative of at least two separate experiments is shown. (*) indicates a long exposure of the film to detect weak binding. affinity compared to Ins(1,3,4,5,6)P5 (Figure 6A). This is also consistent with our notion of Arg472 interacting with the delocalised negative charge on the D1-phosphate (Figure 3). Irvine and colleagues have suggested that binding studies using InsP6 should be performed in the presence of Mg2 þ as the counter ion as the physiological form of InsP6 is complexed to Mg2 þ (RF Irvine, unpublished observation). 3924 The EMBO Journal VOL 23 | NO 20 | 2004 Phosphoinositides PtdIns(3,4,5)P3 PtdIns(3,4)P2 PtdIns(4,5)P2 PtdIns(3,5)P2 PtdIns(3)P PtdIns(5)P wt PDK1 0.012 PDK1 [K467A] 0.196 PDK1 [K467R] 0.009 K i (µM) K i (µM) K i (µM) 0.003 0.019 0.59 0.84 2.6 4.0 0.023 0.046 5.9 6.7 21.8 42.4 0.001 0.017 0.41 0.54 1.9 3.4 Figure 5 TR-FRET measurements of phosphoinositide binding to PDK1. Affinities of wild-type and mutant PDK1 for binding to phosphoinositides was measured by the TR-FRET assay, as described in Materials and methods. (A) Saturation-binding isotherms of the indicated forms of PDK1 binding to biotinylatedPtdIns(3,4,5)P3 (bPIP3). (B) The PDK1-bPIP3 complexes formed at 80 nM bPIP3 were incubated with nonlabelled phosphoinositides and the displacement of bound bPIP3 measured. The data in (A) and (B) are presented as an average of three independent experiments7s.d. (C) The KD values were calculated from (A). IC50 values obtained from (B) were used to calculate Ki values using the Cheng and Prusoff equation (Cheng and Prusoff, 1973). The KD value for bPIP3 for the wild type and indicated mutant of PDK1 is also stated. Binding of inositol lipids and inositol phosphates to PDK1 was not affected by the inclusion of 1 mM Mg2 þ (data not shown). Evidence for physiological interaction of PDK1 with D6-phosphorylated inositol phosphates Initially, we attempted to design a mutant of PDK1 that inhibited binding of D6-phosphorylated inositol phosphates without affecting binding of phosphoinositides. Unfortunately, all the point mutants of PDK1 that we generated, which reduced binding of InsP6, also affected binding to PtdIns(3,4,5)P3 and PtdIns(4,5)P2. We also attempted to cocrystallise the PDK1 PH domain with Ins(1,3,4,5,6)P5 or InsP6, without success. This prompted us to compare the affinities of the isolated PH domain and full-length PDK1 for phosphoinositides and D6-phosphorylated inositol phosphates. & 2004 European Molecular Biology Organization Structure and function of the PDK1 PH domain D Komander et al Figure 6 Interaction of PDK1 with inositol phosphates. (A) Affinities of wild-type PDK1 and its isolated PH domain (residues 409–556), for the indicated phosphoinositides and inositol phosphates were measured by the TR-FRET assay using the approach described in Figure 5A and B. The KD value for biotinylated-PtdIns(3,4,5)P3 (bPIP3) is also stated. (B) HEK-293 cells were transiently transfected with DNA constructs encoding the GFP-tagged PKBa PH domain (residues 1–123, panels 1–3), wild-type PDK1 (panels 4–6), the isolated PDK1 PH domain (residues 409–556, panels 7–9) and the non-PtdIns(3,4,5)P3-binding isolated PDK1[K465E] PH domain (residues 409–556, panels 10–12). At 10 h posttransfection, the cells were serum starved for 16 h and either left unstimulated, stimulated with IGF1 for 10 min, or treated with 100 nM wortmannin for 20 min and then stimulated with IGF1 for 10 min. The cells were then fixed and imaged using a Zeiss LSM 510 META confocal microscope with an optical section thickness of 0.5 mm. The cells shown are representative images obtained in three separate experiments. (C) Wild-type GST PKBa was incubated with PDK1, magnesium, g[32P]ATP in the presence ( þ ) or absence of 100 mM phosphatidylserine (PS), 100 mM phosphatidylcholine (PC) vesicles containing 10 mM PtdIns(3,4,5)P3 and/or 300 mM of the indicated inositol phosphates, and the activation of PKBa was assessed as described previously (Alessi et al, 1997). In samples in which PtdIns(3,4,5)P3 was omitted, the PC/PS vesicles were still added. Phosphorylation of the PKB was determined following electrophoresis on a 4–12% gradient polyacrylamide gel, either by immunoblotting with the phosphospecific Thr308 antibody or by autoradiography of the Coomassie Blue-stained PKBa band. PDK1 activity was assayed as above, employing the PDKtide peptide substrate rather than by measuring PKB activation and phosphorylation. Experiments were performed in quadruplicate and similar results obtained in two to four separate experiments. Surprisingly, we observed that the isolated PH domain of PDK1, although it still interacted with PtdIns(3,4,5)P3 and PtdIns(4,5)P2 with high affinity, possessed 40-fold reduced & 2004 European Molecular Biology Organization affinity for InsP6 and five-fold lower affinity for Ins(1,3,4,5,6)P5 (Figure 6A). Inspection of the structure of the PH domain–Ins(1,3,4,5)P4 complex suggests that a The EMBO Journal VOL 23 | NO 20 | 2004 3925 Structure and function of the PDK1 PH domain D Komander et al D6-phosphate would be solvent exposed and unable to interact with the PH domain. However, in the context of the fulllength PDK1, the D6-phosphate may well form interactions with non-PH domain regions of the protein, explaining the observed higher affinity of full-length PDK1 for D6-phosphorylated inositol phosphates. Clearly, further structural studies are required to investigate this specific interaction and to understand the alignment of the PH domain and the kinase domain in context of the full-length PDK1 protein. We next studied the cellular location of green fluorescent protein (GFP)-tagged full-length PDK1 compared to the isolated PH domain of PDK1. As observed previously (Currie et al, 1999), full-length PDK1 was mainly cytosolic, with only a small fraction being at the membrane, and was excluded from the nucleus in unstimulated HEK-293 cells. PDK1 did not markedly translocate to the plasma membrane following stimulation of cells with the PI 3-kinase agonist IGF1. Similarly, localisation of full-length PDK1 was not affected by the PI 3-kinase inhibitor wortmannin (Figure 6B, panels 4–6). In parallel experiments, we observed IGF1-induced plasma membrane translocation of the PtdIns(3,4,5)P3-binding PH domain of PKBa, which was inhibited by wortmannin (Figure 6B, panels 1–3). The isolated PDK1 PH domain was localised in the cytoplasm and nucleus of unstimulated cells, and, in contrast to full-length PDK1, IGF1 induced a marked translocation to the plasma membrane that was inhibited by wortmannin (Figure 6B, panels 7–9). A non-phosphoinositide-binding isolated PDK1[K465E] PH domain mutant was localised throughout the cell and did not translocate to the plasma membrane upon IGF1 stimulation (Figure 6B, panels 10–12). Control Western blotting demonstrated the absence of proteolytic degradation of the expressed constructs (data not shown). Taken together, these findings are consistent with a model in which full-length PDK1 is anchored in the cytosol through high-affinity binding to D6-phosphorylated inositol phosphates. Similar observations have been made in Dictyostelium discoideum, in which Luo et al (2003) followed the localisation of the PH domain of Crac (PHCrac). Crac shows high affinity for diphosphoinositol pentakisphosphate (PP-InsP5, the isomer was not stated), and appears mostly cytosolic, as is PDK1 in our study. Following cAMP stimulation, which causes PtdIns(3,4,5)P3 accumulation in D. discoideum, the movement of PHCrac is potentiated in mutants lacking inositol hexakisphosphate kinase, the enzyme responsible for PP-InsP5 synthesis. It should be noted that levels of inositol diphosphates in D. discoideum are about 100 times those of mammalian tissues, but still only about half those of InsP6 (Luo et al, 2003). PDK1 was found to interact with 5PP-InsP5, with a significantly lower affinity (Ki 400 nM) than Ins(1,3,4,5,6)P5 and InsP6 (Figure 6A). As the cellular concentration of 5PP-InsP5 is much lower than Ins(1,3,4,5,6)P5 and InsP6, it is unlikely that 5PP-InsP5 is a physiological ligand for PDK1. The PKBa PH domain shows rapid translocation to the plasma membrane (Figure 6B, panels 1–3) upon IGF1 stimulation. PKBa PH domain and PDK1 show similar affinities for phosphoinositides, but PKBa PH domain shows significantly lower affinity towards higher inositol phosphates (Frech et al, 1997). Indeed, InsP6 binds PDK1 with 80-fold higher affinity than PKBa PH domain. As such, the competition between cytosol-localised inositol phosphates and membrane-localised phosphoinositides greatly favours phosphoi3926 The EMBO Journal VOL 23 | NO 20 | 2004 nositide binding in the case of the PKBa PH domain, resulting in a marked translocation of PKBa PH domain to the plasma membrane in IGF1 treated cells (Figure 6B, panels 1–3). Consistent with the notion that high-affinity binding to inositol phosphates maintains PDK1 located in the cytoplasm, the isolated PH domain of PDK1, which displays high-affinity interactions with phosphoinositides, but has an affinity (4.1 mM) similar to that of PKBa for InsP6 (8.7 mM, (Frech et al, 1997)), is seen to translocate in a fashion analogous to the PKBa PH domain. Our data do not rule out the possibility that other ligands/factors could interact with full-length PDK1, contributing towards its cytosolic or membrane localisation. In this regard, it has recently been reported that PDK1 can be stabilised at the cell membrane by interaction with Grb14, which binds the insulin receptor through an SH2 domain (King and Newton, 2004). In many studies of PH domain-containing proteins, the lipid-binding and localisation properties of the PH domain are assessed employing the isolated PH domain rather than the full-length protein (Irvine, 2004). Our data emphasise a pitfall of this approach, as the isolated PH domain of PDK1 has quite distinct inositol phosphate-binding properties and localisation pattern, compared to the full-length protein. Recently, it was shown that the isolated PH domain of phospholipase Cd4 (PLCd4) was localised at the plasma membrane, whereas the full-length protein localises at intracellular membranes, and also showed differences in PtdIns(4,5)P2 versus Ins(1,4,5)P3 binding (Lee et al, 2004). Thus, in order to determine the key localisation signals of lipid-binding proteins, we would recommend comparing the lipid-binding and localisation properties of the isolated lipid-binding domain with that of the full-length protein. We next investigated whether Ins(1,3,4,5,6)P5 and InsP6 could compete with PtdIns(3,4,5)P3 in regulating the activation of PKB by PDK1 in an in vitro assay. Addition of 300 mM InsP6 or Ins(1,3,4,5,6)P5 inhibited the activation and phosphorylation of PKB in the presence of PtdIns(3,4,5)P3, by over 80% (Figure 6C). InsP6 or Ins(1,3,4,5,6)P5 did not directly inhibit PDK1 catalytic activity, as the phosphorylation of the PDKtide peptide substrate by PDK1 was not affected by D6-phosphorylated inositol phosphates (Figure 6C). It has also been reported that increasing extracellular Ins(1,3,4,5,6)P5 in cells lowered PKB activation, through an undefined mechanism (Piccolo et al, 2004). The data presented here indicate that these effects could be mediated, at least in part, through binding of PDK1 to Ins(1,3,4,5,6)P5 and anchoring it in the cytosol. It is also possible that Ins(1,3,4,5,6)P5 might bind to the PH domain of PKB in vivo, and thus affect its translocation to the plasma membrane. Several other PtdIns(3,4,5)P3-binding PH domains including Grp1 and Btk have been shown to interact in vitro with InsP6 and Ins(1,3,4,5,6)P5. Btk interacts with both InsP6 and Ins(1,3,4,5,6)P5 with Ki of 90 nM (Kojima et al, 1997), whereas Grp1 binds to InsP6 and Ins(1,3,4,5,6)P5 with KD’s of 6 and 0.3 mM, respectively (Kavran et al, 1998). PtdIns(3,4,5)P3 recruits Grp1 and Btk to the plasma membrane, where they are thought to exert their physiological effects. Thus, if it occurred in vivo, inositol phosphate binding to these PH domain-containing proteins would inhibit their function. In the case of PDK1, cytosolic localisation is important in enabling PDK1 to activate substrates that do not & 2004 European Molecular Biology Organization Structure and function of the PDK1 PH domain D Komander et al bind lipids, such as SGK or S6K. Our data indicate that the micromolar levels of Ins(1,3,4,5,6)P5 and InsP6 found in cells would be sufficient to act as physiological ligands of PDK1, and could compete with PtdIns(3,4,5)P3, PtdIns(3,4)P2 and PtdIns(4,5)P2 for binding to PDK1. Such a mechanism could ensure that sufficient PDK1 was distributed in both the cytosol and plasma membrane to enable PDK1 to fulfil its cellular functions. Thus, where the binding of Btk and Grp1 to inositol phosphates would act as a brake in signalling, for PDK1 this binding would accelerate activation of cytoplasmic substrates. In this respect, PDK1 is unique among PH domain-containing proteins described so far. Conclusions Here we describe the unusual structure of the PDK1 PH domain. It possesses several unique features not previously observed in other PH domains, including the N-terminal bud which forms an integral part of the overall PH domain structure. The PDK1 PH domain in complex with a PtdIns(3,4,5)P3 analogue shows a surprising analogy in the phosphoinositide-binding mode of the PDK1 PH domain and the PX domain of p40phox. We define the relative importance of the residues in the ligand-binding site for enabling PDK1 to bind phosphoinositides. In particular, we identify Lys467 as the residue in PDK1 responsible for binding the D5-phosphorylated ligands and thus permitting interaction with PtdIns(4,5)P2. Moreover, the PDK1 PH domain possesses a much more spacious ligand-binding site, which we demonstrate enables PDK1 to interact with Ins(1,3,4,5,6)P5 and InsP6 with high affinity. We perform studies that support the notion that these ligands could function to anchor PDK1 in the cytosol, where it is needed to activate its nonlipidbinding substrates. Overall, our results provide a molecular explanation for how PDK1 could be localised both at the membrane and the cytosol of cells. Materials and methods Protein expression, purification, crystallisation and data collection PDK1 PH domain (residues 409–556) was expressed in Escherichia coli, purified and crystallised as described previously (Komander et al, 2004). Further details on crystallisation conditions, data collection, as well as structure determination and analysis can be found in the Supplementary data. The refinement statistics can be found in Table I. Analysis of PDK1 phosphoinositide and inositol phosphate interactions For qualitative protein lipid overlay assays, wild type and mutants of full-length PDK1 were expressed as GST fusion proteins (pEBG vector) in human embryonic kidney HEK-293 cells, as described previously (Alessi et al, 1997). The phospholipid overlay assays were performed as described previously (Dowler et al, 2002), using di-palmitoyl-phosphoinositides from Cell Signals Inc. (Columbus, USA). The quantitative TR-FRET analysis of binding was performed as described by Gray et al (2003). To determine KD values for biotinylated diC8-PtdIns(3,4,5)P3, increasing amounts (0–500 nM) of biotinylated diC8-PtdIns(3,4,5)P3 were incubated with GSTtagged wild type and mutant full-length PDK1 or wild-type PH domain of PDK1 (20 nM) in the presence of excess europiumlabeled goat anti-GST antibody and streptavidin-conjugated allophycocyanin (APC). The binding of the biotinylated lipid to the protein allows FRET to occur between europium (donor) and APC (acceptor). Fluorescence is monitored at 665 and 505 nm; the ratio of these signals allows determination of relative amount of binding. Saturation-binding isotherms are presented in Figure 5A. Competition assays were performed with a fixed amount (80 nM) of biotinylated lipid and increasing amounts of competing agent (diC8 phosphoinositides or inositol phosphates). IC50 values were determined, from which Ki values were calculated, according to the equation below (Cheng and Prusoff, 1973). All curve fitting was performed using Prism (GraphPad). Ki ¼ IC50 1 þ ½L=KD Localisation studies Cells were fixed in 4% (w/v) paraformaldehyde in PHEM buffer (60 mM Pipes, 25 mM Hepes, 10 mM EGTA and 2 mM MgSO4, pH 7.0). The cells were washed three times with PBS and mounted on slides with Hydromount. The cells were imaged to monitor GFP fluorescence and confocal sections collected, using a Zeiss LSM 510 META confocal microscope. A detailed description of all other methods employed in this study can be found in the supplementary data of this paper. Supplementary data Supplementary data are available at The EMBO Journal Online. Acknowledgements We thank the European Synchrotron Radiation Facility for beam time at BM14, and Martin Walsh for assistance in data collection, Mike Hodgkinson for assistance with in-house data collection, the School of Life Sciences DNA Sequencing Service and Robin Irvine for helpful discussion and sharing unpublished data. 5PP-InsP5 was kindly provided by JR Falck, University of Texas Southwestern Medical Center, USA. An MRC Predoctoral Fellowship funds DK, and AF is supported by a BBSRC Committee Studentship. STS is a Royal Society University Research Fellow, and thanks The Royal Society and Tenovus-Scotland for financial support. DMFvA is supported by a Wellcome Trust Senior Research Fellowship and an EMBO Young Investigator Award. CPD is supported by a Programme Grant from the MRC. DRA thanks the Association for International Cancer Research, Diabetes UK, the Medical Research Council, the Moffat Charitable Trust, as well as the pharmaceutical companies that support the Division of Signal Transduction Therapy (AstraZeneca, Boehringer-Ingelheim, GlaxoSmithKline, Merck & Co. Inc., Merck KGaA and Pfizer). The coordinates and structure factors have been deposited with the PDB (entries 1w1d (PDK1 PH domain–Ins(1,3,4,5)P4 complex), 1w1g (PDK1 PH domain– PtdIns(3,4,5)P3 complex) and 1w1h (uncomplexed PDK1 PH domain)). All Figures were created with PyMol (www.pymol.org). References Alessi DR, James SR, Downes CP, Holmes AB, Gaffney PR, Reese CB, Cohen P (1997) Characterization of a 3-phosphoinositidedependent protein kinase which phosphorylates and activates protein kinase Balpha. Curr Biol 7: 261–269 Anderson KE, Coadwell J, Stephens LR, Hawkins PT (1998) Translocation of PDK-1 to the plasma membrane is important in allowing PDK-1 to activate protein kinase B. Curr Biol 8: 684–691 & 2004 European Molecular Biology Organization Baraldi E, Carugo KD, Hyvonen M, Surdo PL, Riley AM, Potter BV, O’Brien R, Ladbury JE, Saraste M (1999) Structure of the PH domain from Bruton’s tyrosine kinase in complex with inositol 1,3,4,5-tetrakisphosphate. Struct Fold Des 7: 449–460 Batty IH, Downes CP (1996) Thrombin receptors modulate insulin-stimulated phosphatidylinositol 3,4,5-trisphosphate accumulation in 1321N1 astrocytoma cells. Biochem J 317 (Part 2): 347–351 The EMBO Journal VOL 23 | NO 20 | 2004 3927 Structure and function of the PDK1 PH domain D Komander et al Begley MJ, Taylor GS, Kim SA, Veine DM, Dixon JE, Stuckey JA (2003) Crystal structure of a phosphoinositide phosphatase, MTMR2: insights into myotubular myopathy and Charcot– Marie–Tooth syndrome. Mol Cell 12: 1391–1402 Bravo J, Karathanassis D, Pacold CM, Pacold ME, Ellson CD, Anderson KE, Butler PJ, Lavenir I, Perisic O, Hawkins PT, Stephens L, Williams RL (2001) The crystal structure of the PX domain from p40(phox) bound to phosphatidylinositol 3-phosphate. Mol Cell 8: 829–839 Cheng Y, Prusoff WH (1973) Relationship between the inhibition constant (K1) and the concentration of inhibitor which causes 50 per cent inhibition (I50) of an enzymatic reaction. Biochem Pharmacol 22: 3099–3108 Collins BJ, Deak M, Arthur JS, Armit LJ, Alessi DR (2003) In vivo role of the PIF-binding docking site of PDK1 defined by knock-in mutation. EMBO J 22: 4202–4211 Currie RA, Walker KS, Gray A, Deak M, Casamayor A, Downes CP, Cohen P, Alessi DR, Lucocq J (1999) Role of phosphatidylinositol 3,4,5-trisphosphate in regulating the activity and localization of 3-phosphoinositide-dependent protein kinase-1. Biochem J 337 (Part 3): 575–583 Dowler S, Kular G, Alessi DR (2002) Protein lipid overlay assay. Sci STKE 2002: PL6 Dowler S, Montalvo L, Cantrell D, Morrice N, Alessi DR (2000) Phosphoinositide 3-kinase-dependent phosphorylation of the dual adaptor for phosphotyrosine and 3-phosphoinositides by the Src family of tyrosine kinase. Biochem J 349: 605–610 Ferguson KM, Kavran JM, Sankaran VG, Fournier E, Isakoff SJ, Skolnik EY, Lemmon MA (2000) Structural basis for discrimination of 3-phosphoinositides by pleckstrin homology domains. Mol Cell 6: 373–384 Ferguson KM, Lemmon MA, Schlessinger J, Sigler PB (1994) Crystal structure at 2.2 A resolution of the pleckstrin homology domain from human dynamin. Cell 79: 199–209 Filippa N, Sable CL, Hemmings BA, Van Obberghen E (2000) Effect of phosphoinositide-dependent kinase 1 on protein kinase B translocation and its subsequent activation. Mol Cell Biol 20: 5712–5721 Frech M, Andjelkovic M, Ingley E, Reddy KK, Falck JR, Hemmings BA (1997) High affinity binding of inositol phosphates and phosphoinositides to the pleckstrin homology domain of RAC/ protein kinase B and their influence on kinase activity. J Biol Chem 272: 8474–8481 Gray A, Olsson H, Batty IH, Priganica L, Peter Downes C (2003) Nonradioactive methods for the assay of phosphoinositide 3-kinases and phosphoinositide phosphatases and selective detection of signaling lipids in cell and tissue extracts. Anal Biochem 313: 234–245 Irvine R (2004) Inositol lipids: to PHix or not to PHix? Curr Biol 14: R308–R310 Irvine RF, Schell MJ (2001) Back in the water: the return of the inositol phosphates. Nat Rev Mol Cell Biol 2: 327–338 Kavran JM, Klein DE, Lee A, Falasca M, Isakoff SJ, Skolnik EY, Lemmon MA (1998) Specificity and promiscuity in phosphoinositide binding by pleckstrin homology domains. J Biol Chem 273: 30497–30508 King CC, Newton AC (2004) The adaptor protein Grb14 regulates the localization of 3-phosphoinositide dependent kinase-1 (PDK1). J Biol Chem 279: 37518–37527 Kojima T, Fukuda M, Watanabe Y, Hamazato F, Mikoshiba K (1997) Characterization of the pleckstrin homology domain of Btk as an inositol polyphosphate and phosphoinositide binding domain. Biochem Biophys Res Commun 236: 333–339 Komander D, Deak M, Morrice N, Van Aalten DM (2004) Purification, crystallization and preliminary X-ray diffraction of a proteolytic fragment of PDK1 containing the pleckstrin homology domain. Acta Crystallogr D Biol Crystallogr 60: 314–316 3928 The EMBO Journal VOL 23 | NO 20 | 2004 Lee SB, Varnai P, Balla A, Jalink K, Rhee SG, Balla T (2004) The pleckstrin homology domain of phosphoinositide-specific phospholipase Cdelta4 is not a critical determinant of the membrane localization of the enzyme. J Biol Chem 279: 24362–24371 Lemmon MA, Ferguson KM (2000) Signal-dependent membrane targeting by pleckstrin homology (PH) domains. Biochem J 350 (Part 1): 1–18 Lietzke SE, Bose S, Cronin T, Klarlund J, Chawla A, Czech MP, Lambright DG (2000) Structural basis of 3-phosphoinositide recognition by pleckstrin homology domains. Mol Cell 6: 385–394 Lodowski DT, Pitcher JA, Capel WD, Lefkowitz RJ, Tesmer JJ (2003) Keeping G proteins at bay: a complex between G protein-coupled receptor kinase 2 and Gbetagamma. Science 300: 1256–1262 Luo HR, Huang YE, Chen JC, Saiardi A, Iijima M, Ye K, Huang Y, Nagata E, Devreotes P, Snyder SH (2003) Inositol pyrophosphates mediate chemotaxis in Dictyostelium via pleckstrin homology domain–PtdIns(3,4,5)P3 interactions. Cell 114: 559–572 McManus EJ, Collins BJ, Ashby PR, Prescott AR, Murray-Tait V, Armit LJ, Arthur JS, Alessi DR (2004) The in vivo role of PtdIns(3,4,5)P(3) binding to PDK1 PH domain defined by knockin mutation. EMBO J 23: 2071–2082 Michell RH (1997) The multiplying roles of inositol lipids and phosphates in cell control processes. Essays Biochem 32: 31–47 Milburn CC, Deak M, Kelly SM, Price NC, Alessi DR, Van Aalten DM (2003) Binding of phosphatidylinositol 3,4,5-trisphosphate to the pleckstrin homology domain of protein kinase B induces a conformational change. Biochem J 375: 531–538 Mora A, Komander D, van Aalten DMF, Alessi DR (2004) PDK1, the master regulator of AGC kinase signal transduction. Semin Cell Dev Biol 15: 161–170 Orchiston EA, Bennett D, Leslie NR, Clarke RG, Winward L, Downes CP, Safrany ST (2004) PTEN M-CBR3, a versatile and selective regulator of inositol 1,3,4,5,6-pentakisphosphate (Ins(1,3,4,5,6)P5). Evidence for Ins(1.3,4,5,6)P5 as a proliferative signal. J Biol Chem 279: 1116–1122 Piccolo E, Vignati S, Maffucci T, Innominato PF, Riley AM, Potter BV, Pandolfi PP, Broggini M, Iacobelli S, Innocenti P, Falasca M (2004) Inositol pentakisphosphate promotes apoptosis through the PI 3-K/Akt pathway. Oncogene 23: 1754–1765 Stephens L, Anderson K, Stokoe D, Erdjument-Bromage H, Painter GF, Holmes AB, Gaffney PR, Reese CB, McCormick F, Tempst P, Coadwell J, Hawkins PT (1998) Protein kinase B kinases that mediate phosphatidylinositol 3,4,5-trisphosphate-dependent activation of protein kinase B. Science 279: 710–714 Stokoe D, Stephens LR, Copeland T, Gaffney PR, Reese CB, Painter GF, Holmes AB, McCormick F, Hawkins PT (1997) Dual role of phosphatidylinositol-3,4,5-trisphosphate in the activation of protein kinase B. Science 277: 567–570 Thomas CC, Deak M, Alessi DR, van Aalten DM (2002) Highresolution structure of the pleckstrin homology domain of protein kinase B/Akt bound to phosphatidylinositol (3,4,5)-trisphosphate. Curr Biol 12: 1256–1262 Thomas CC, Dowler S, Deak M, Alessi DR, van Aalten DM (2001) Crystal structure of the phosphatidylinositol 3,4-bisphosphatebinding pleckstrin homology (PH) domain of tandem PH-domaincontaining protein 1 (TAPP1): molecular basis of lipid specificity. Biochem J 358: 287–294 Toker A, Newton AC (2000) Cellular signaling: pivoting around PDK-1. Cell 103: 185–188 Vanhaesebroeck B, Alessi DR (2000) The PI3K–PDK1 connection: more than just a road to PKB. Biochem J 346 (Part 3): 561–576 Vanhaesebroeck B, Leevers SJ, Ahmadi K, Timms J, Katso R, Driscoll PC, Woscholski R, Parker PJ, Waterfield MD (2001) Synthesis and function of 3-phosphorylated inositol lipids. Annu Rev Biochem 70: 535–602 Worthylake DK, Rossman KL, Sondek J (2000) Crystal structure of Rac1 in complex with the guanine nucleotide exchange region of Tiam1. Nature 408: 682–688 & 2004 European Molecular Biology Organization