SEARCH

SEARCH BY CITATION

Keywords:

  • allosteric;
  • cancer;
  • diabetes;
  • fragment screening;
  • NMR;
  • PDK1

Abstract

  1. Top of page
  2. Abstract
  3. Methods and Materials
  4. Results and Discussion
  5. Conclusions and Future Directions
  6. References
  7. Supporting Information

Aberrant activation of the phosphoinositide 3-kinase pathway because of genetic mutations of essential signalling proteins has been associated with human diseases including cancer and diabetes. The pivotal role of 3-phosphoinositide-dependent kinase-1 in the PI3K signalling cascade has made it an attractive target for therapeutic intervention. The N-terminal lobe of the 3-phosphoinositide-dependent kinase-1 catalytic domain contains a docking site which recognizes the non-catalytic C-terminal hydrophobic motifs of certain substrate kinases. The binding of substrate in this so-called PDK1 Interacting Fragment pocket allows interaction with 3-phosphoinositide-dependent kinase-1 and enhanced phosphorylation of downstream kinases. NMR spectroscopy was used to a screen 3-phosphoinositide-dependent kinase-1 domain construct against a library of chemically diverse fragments in order to identify small, ligand-efficient fragments that might interact at either the ATP site or the allosteric PDK1 Interacting Fragment pocket. While majority of the fragment hits were determined to be ATP-site binders, several fragments appeared to interact with the PDK1 Interacting Fragment pocket. Ligand-induced changes in 1H-15N TROSY spectra acquired using uniformly 15N-enriched PDK1 provided evidence to distinguish ATP-site from PDK1 Interacting Fragment-site binding. Caliper assay data and 19F NMR assay data on the PDK1 Interacting Fragment pocket fragments and structurally related compounds identified them as potential allosteric activators of PDK1 function.

Abbreviations:
AGC

cAMP-dependent protein kinase/protein kinase G/protein kinase C extended family

ATP

adenosine triphosphate

NMR

nuclear magnetic resonance

PDK1

3-phosphoinositide-dependent kinase-1

PDKtide

KTFCGTPEYLAPEVRREPRILSEEEQEMFRDFDYIADWC

PI3K

phosphoinositide 3-kinase

PIF

PDK1 Interacting Fragment

PIFtide

REPRILSEEEQEMFRDFDYIADWC

PIP2

phosphotidylinositol-3,4-bisphosphate

PIP3

phosphotidylinositol-3,4,5-triphosphate

Short PIFtide

EEQEMFRDFDYIADWC

STD

saturation transfer difference

T308tide

KTFCGTPEYLAPEVRR

TROSY

transverse relaxation optimized spectroscopy

TCEP

tris (2-carboxyethyl) phosphine

One of the most important and extensively studied cellular signalling systems is the phosphoinositide 3-kinase (PI3K)/Akt pathway. Activation of PI3K by extracellular growth factors and hormones triggers the transduction of signals from the cytosol to the nucleus resulting in cell growth, proliferation and survival (1,2). Aberrant activation of this pathway because of genetic mutations of essential signalling proteins has been associated with human diseases including cancer and diabetes (3,4). The phosphorylation of phosphotidylinositol-3,4-bisphosphate (PIP2) by activated PI3K generates the second messenger phosphotidylinositol-3,4,5-triphosphate (PIP3). The binding of PIP3 to the pleckstrin homology domains of Akt and 3-phosphoinositide-dependent kinase-1 (PDK1) co-localizes these enzymes at the plasma membrane inducing a conformational shift in Akt (1,5,6). This shift alleviates the autoinhibition of the active site allowing for PDK1 phosphorylation at T308 resulting in partial activation of Akt. PDK1 recognizes other substrate kinases in the pathway through a distinct regulatory mechanism. The N-terminal lobe of the catalytic domain of PDK1 contains a docking site which recognizes the non-catalytic C-terminal hydrophobic motifs of substrate kinases. As shown in Figure 1, this site is distinct from the ATP binding site. The binding of substrate in this so-called PIF (PDK1 Interacting Fragment) pocket allows interaction with PDK1 and enhanced phosphorylation of the downstream kinases (7). This intricate substrate recognition system allows for the specific and selective control of this signal transduction pathway in response to unique extracellular events (8).

image

Figure 1.  Surface representation of the PDK1 kinase domain. Residues 76–359 from the wild-type structure (Protein Data Bank accession number 1H1W) (22) were used to calculate the surface, with residues 233–236 omitted due to disorder in the published structure. ATP is shown in stick format. Locations of the αB and αC helices in the PIF pocket are also highlighted.

Download figure to PowerPoint

The pivotal role of PDK1 in the PI3K signalling cascade has made it an attractive target for therapeutic intervention. PDK1 has been demonstrated to phosphorylate numerous downstream protein kinases including other AGC kinase family members such as p70 ribosomal S6 kinase, serum- and glucocorticoid-responsive kinase, protein kinase C, and p90 ribosomal S6 kinase (6,9–12).

Using a virtual screening approach targeting the PIF pocket of PDK1, Engel et al. demonstrated that small molecules could modulate enzyme activity through binding in this pocket (13). The identified small molecules are postulated to allosterically activate enzyme activity by modulating the phosphorylation-dependent conformational transition. The data suggesting that the small molecules bind to the PIF pocket was generated through a series of PDK1-directed mutagenesis experiments, synthesis of compounds to generate rational structure–activity information, displacement studies and isothermal titration calorimetry. In an effort to identify further selective small molecule modulators of PDK1, our laboratory initiated an NMR fragment screening approach to exploit the ATP binding and PIF pocket sites. The goal was to identify both novel ATP scaffolds and allosteric modulators of PDK1. Lead molecules identified from the fragment screening approach would be expected to complement hits from a previous HTS assay that was designed to identify PDK1 inhibitors. Since HTS assays are typically run at much lower compound concentrations than are fragment screens, the resulting hits are usually low micromolar inhibitors but often lack novelty. By contrast, fragment screens are run at higher concentrations and result in weaker hits but with more chemical diversity. Novel ATP competitive scaffolds identified in this manner would potentially enrich the known kinase chemical space. Identified compounds that interact with the PIF pocket, which was blocked by the PDKtide substrate used in the HTS assay, may serve as tools to facilitate the study of the PDK1 pathway, other AGC kinases, or novel pockets within other kinases.

NMR spectroscopy was used to screen a PDK1 kinase domain construct against a library of chemically diverse fragments in order to identify low molecular weight, ligand-efficient binders. The construct was designed such that the ATP site and the allosteric PIF pocket would all be available for potential interactions with ligands. Greater than 300 hits were identified from the ∼10 K fragment collection screened. Binding fragments were subsequently tested for activity using the Kinase-Glo assay measuring ATP depletion or Caliper assay measuring the incorporation of a phosphate into a fluorescent tagged substrate peptide. A large number of fragments possessed some inhibitory activity when assayed. Many of the hits were further characterized by NMR competition studies to determine if they could be displaced by ATP or peptides containing the PIF motif. While the data generated from NMR and biochemical studies suggested that the majority of the fragment hits were ATP-site binders, several fragments appeared to interact with the PIF pocket. More intensive NMR methods were then applied to characterize PIF pocket interactions. Ligand-induced changes in 1H-15N TROSY spectra acquired using uniformly 15N-enriched PDK1 kinase domain provided further evidence to distinguish ATP-site from PIF-site binding. Caliper assay data and 19F NMR assay data on the PIF pocket fragments and structurally related compounds identified them as potential allosteric activators of PDK1 function.

Methods and Materials

  1. Top of page
  2. Abstract
  3. Methods and Materials
  4. Results and Discussion
  5. Conclusions and Future Directions
  6. References
  7. Supporting Information

Expression and purification of PDK1

Human PDK1 (N-terminal His6-tagged, residues 51–359) was cloned into the pFastBacHT A vector (Invitrogen, Carlsbad, CA, USA; 10584-027) and a recombinant baculovirus was generated using the Bac-to-Bac Baculovirus Expression System (Invitrogen, 10359-016). Recombinant P2 viral stocks were used to generate baculovirus-infected insect cell (BIIC) stocks (14).

Expression of unlabelled PDK1 was carried out as follows: Spodoptera frugiperda (Sf21) insect cells were cultured in SF-900II SFM (Invitrogen, 10902-088) media, supplemented with 10 mg/L gentamicin (Invitrogen, 15710-064) and 1X antibiotic-antimycotic (Invitrogen, 15240-062) at 27 °C and 110 rpm in 1 L Erlenmeyer flasks (Corning, NY, USA 431403). For 5 L expression, Sf21 cells were seeded in culture media at 0.7 × 106 viable cells/ml in 10 L Cellbags (GE Healthcare, CELLBAG10L/S-NC) 24 h prior to infection. The bag was placed on a Wave Bioreactor System 20/50EH (GE Healthcare: Chalfont St. Giles, United Kingdom, 20/50EH) rocking at 26 rpm, 27 °C, and rocking angle 7º. When cells reached a density of 1.2 × 106 viable cells/mL, the cells were infected with 1 mL of PDK1 BIIC to the 5 L culture at 27 rpm, 27 °C, rocking angle 7º, and was further incubated for 3 days postinfection. PDK1 expressing cells were harvested by centrifugation at 9000 × g for 10 min. The cell pellets were stored frozen at −80 °C until purification.

Stable isotope labelling and expression of PDK1 was performed as follows: Sf9 insect cells were cultured as described above. Sf9 cells were seeded at 1 × 106 viable cells/mL in 3 L Fernbach culture flask (Corning, 431253) 24 h prior to infection at 27 °C and 110 rpm. When cells reached a density of 2 × 106 viable cells/mL, cells were harvested by centrifugation at 400 × g for 10 min at room temperature. The supernatants were discarded. Cells were resuspended in 1 L BioExpress-2000 U-15N labelling media (Cambridge Isotope Laboratories, Inc.: Andover, MA, USA, CGM-2000-N) and transferred into a 3 L Fernbach culture flask. Baculovirus infection was performed by adding 1 mL of PDK1 BIIC to the 1 L culture and was further incubated for 3 days postinfection at 27 °C and 110 rpm. Isotope labelled, PDK1 expressing cells were harvested by centrifugation at 9000 × g for 10 min. The cell pellets were stored frozen at −80 °C until purification.

For isolation of unlabelled PDK1, cell pellets were lysed by sonication in 25 mm Tris–HCl, pH 7.5, 500 mm NaCl, 5 mm TCEP, and complete protease inhibitor cocktail (Roche: Basel, Switzerland, 11 873 580 001). The cell lysates were centrifuged overnight at 40 000 × g and 4 °C. The soluble lysates containing PDK1 were loaded onto a HisTrap HP (GE Healthcare, 17-3248-02) Ni Sepharose affinity column in 25 mm Tris–HCl, pH 7.5, 250 mm NaCl, 5mm TCEP, and 20 mm imidazole. The recombinant His6-tagged PDK1 was eluted using a linear gradient of imidazole from 20–500 mm in 20 column volumes. Fractions containing the PDK1 were pooled and loaded onto a HiPrep 26/10 (GE Healthcare, 17-5087-01) desalting column in 25 mm Tris–HCl, pH 7.5, 250 mm NaCl, and 5 mm TCEP to remove the imidazole. The buffer exchanged PDK1 was digested with AcTEV protease (Invitrogen, 12575-015) overnight at 4 °C to remove the His6 tag. The unlabelled PDK1 was further purified by passing the cleavage reaction over a HisTrap HP column in 25 mm Tris–HCl, pH 7.5, 250 mm NaCl, and 5mm TCEP to remove the uncleaved protein and cleaved His6 tag. The flow through containing purified PDK1 was passed over a Superdex S200 16/60 column (GE Healthcare, 17-1069-01) to exchange into the NMR buffer of 25 mm deuterated Tris–HCl, pH 7.5, 100 mm NaCl, and 2 mm TCEP.

U-15N labelled PDK1 was purified as described above, except the final gel filtration step was substituted by a desalting column equilibrated in 25 mm deuterated Tris–HCl, pH 7.5, 100 mm NaCl, and 2 mm TCEP. For NMR analysis, the U-15N labelled PDK1 was concentrated to 3.5 mg/mL. The amount of label incorporation was determined by LC-MS analysis to be 94%.

Peptides

PIFtide (REPRILSEEEQEMFRDFDYIADWC), short PIFtide (EEQEMFRDFDYIADWC), and PDKtide (KTFCGTPEYLAPEVRREPRILSEEEQEMFRDFDYIADWC) were obtained from the American Peptide Company (Sunnyvale, CA, USA). T308tide (KTFCGTPEYLAPEVRR) was obtained from Sigma (St. Louis, MO, USA).

Bioluminescent ATP depletion assay

The kinase reaction was carried out in 100 mm Hepes, pH 8.25, 10 mm MgCl2, 1 mm dithiothreitol, and 0.015% Brij. PDKtide substrate and ATP (Roche) concentrations were 2 μm and 8 μm, respectively. The PDK1 enzyme was 15 nm and compounds were tested at 100–300 μm. To each well in a white 384 well non-binding plate (Corning), 4 μL of test compound and 8 μL of PDK1 enzyme were added and incubated at room temperature for 60 min. The reaction was then initiated by the addition of 8 μL of ATP/PDKtide substrate. Plates were shaken for 5 min and then incubated for an additional 55 min at room temperature. 20 μL of Kinase-Glo reagent (Promega, Madison, WI, USA) was then added to stop the reaction. After a 20-min incubation at room temperature the luminescence was read on an LJL Analyst (Molecular Devices: Toronto, Canada). All assays contained positive control wells with buffer only and negative control wells where 100% inhibition of PDK1 was achieved with 1 μm staurosporine (Sigma).

Microfluidic caliper assay

Buffer conditions were as described for the ATP depletion assay. However, the substrate was 1.5 μm T308tide (since this substrate does not bind to the PIF pocket and would thus potentially allow identification of PIF pocket binders). ATP was used at a concentration of 28 μM and PDK1 enzyme at 15 nm. The assay was performed in the ‘off-chip’ mode on a Caliper LabChip 3000 (Caliper Technologies Corp, Mountain View, CA, USA). As with the ATP depletion assay, compounds were incubated with enzyme for 60 min prior to the addition of T308tide and ATP. The plate was then incubated at room temperature for 60 min before stopping the reaction with 20 μL stop buffer (100 mm Hepes, 0.015% Brij, 5% dimethylsulfoxide, 0.227% Caliper coating reagent 3 and 35 mm EDTA). The reaction mixture was sipped onto the Caliper chip 20 min after the reaction was stopped. Substrate and product were separated based on charge applying an upstream voltage of −500 V and a downstream voltage of −2600 V, with a screening pressure of −1.6 psi. Results were quantified by fluorescence intensity. Control wells included buffer only (positive control), 1 μm staurosporine (negative control) or 10 μm PIFtide (determined from a dose–response curve to give maximum T308tide phosphorylation). Data were expressed as a ratio of product to the sum of the product and substrate.

NMR samples

PDK1 solutions were prepared in 90%1H2O/10%2H2O. For fragment screening, 550 μL solutions containing 4 μm unlabelled PDK1 and a screening library mixture of usually 10 compounds, each at a concentration of 300 μm, were prepared using Norell 502 NMR tubes. Similar concentrations were used to make samples for competition binding experiments. For ligand-binding characterization experiments, 550 μL solutions containing 50 μm [U-15N]PDK1, 75–250 μm of a given ligand, and 2 mm MgCl2 were prepared using Wilmad 528 NMR tubes. Compounds and peptides were added to the solutions last from concentrated stock solutions, with the presence of PDK1 improving the solubility of marginally soluble compounds. Samples for activity assays contained 1.2 μm PDK1, 100 μm T308tide, 100 μm 2-fluoro-ATP (Chemilia, Huddinge, Sweden), and 2 mm MgCl2. Reactions were allowed to run for 18 h at 21 °C and were then quenched by adding EDTA to a final concentration of 10 mm. The reaction proceeds very slowly because of the >10 mmKm value of the T308tide substrate (7).

NMR spectroscopy

1H NMR data were collected on a Bruker Avance 600 MHz spectrometer (Billerica, MA, USA) equipped with a 5 mm TXI cryoprobe and an autosampler. Fragment screening and competition binding experiments were carried out at 20 °C utilizing saturation transfer difference (15) with on-resonance irradiation at 0.73 ppm and off-resonance irradiation at −20 ppm (16). The 3 s irradiation period consisted of 50 ms Gauss pulses separated by a 1 ms delay. A 35 ms T2 filter was used to suppress residual protein signals. Excitation sculpting was used for water suppression (17). Data sets were the average of 512 scans. 1H chemical shifts were referenced to the 1H2O signal at 4.70 ppm. For ligand-binding characterization, 1H-15N TROSY experiments (18) were collected at 25 °C using a sensitivity-enhanced, water flip-back sequence (19). Spectra were acquired using 96 scans for each of 256 complex t1 increments, and with spectral widths of 4000 Hz and 9259 Hz in F1 and F2, respectively. 15N chemical shifts were referenced indirectly using the 15N:1H gyromagnetic ratio of 0.101329118. For the activity assays, the 471 MHz 19F{1H-decoupled} NMR spectra were collected at 25 °C on a Bruker DRX 500 MHz spectrometer (Billerica, MA, USA) equipped with a conventional 5 mm SEF probe optimized for 19F detection. Data sets were the average of 256 scans. 19F chemical shifts were referenced to internal 50 μm trifluoroethanol.

Molecular modelling

An internally developed program AGDOCK (20) was used for docking and scoring. Compounds were docked into the structure of the S241A mutant (Protein Data Bank accession number 2BIY) (21) where all solvent molecules had been removed. This structure was used instead of that for the wild-type (22) because the latter has several disordered residues in the activation loop. The PIF pockets in both structures are indistinguishable (21). Initially, Compound 2 was manually placed into the PIF pocket with a random orientation. This step is required for binding site specification in AGDOCK but is independent of pose generation and scoring. Flexible ligand docking was then performed in the absence of any restraints on Compounds 2 and 3. The initial 3D conformation was generated using Corina (Molecular Networks) and was optimized using the AMBER force field and a truncated Newton method. The best scoring poses for each compound were then selected for flexible protein–ligand minimization using MacroModel (Schrodinger) with a 6 Å shell. The minimization generally involved backbone and sidechain movement of the residues on the αB and αC helices, as well as the sidechains of the adjacent residues such as Q150 and F157, and resulted in an improvement of the free binding energy by 3–5 kcal/mol based on the AGDOCK scoring function.

Results and Discussion

  1. Top of page
  2. Abstract
  3. Methods and Materials
  4. Results and Discussion
  5. Conclusions and Future Directions
  6. References
  7. Supporting Information

Fragment screening provides a complementary method to high throughput screening for identifying ligands for biomacromolecules (23–25). By definition, fragment screens are designed to identify ligand-efficient inhibitors. Since the fragments are limited to low molecular weight, typically 150–300 Da, an efficient sampling of chemical space can be obtained with ∼10 K fragments or even less. Hits identified are typically much weaker than hits from a high throughput screen, but are often comparable on a ligand-efficiency basis (26,27). Fragment libraries can be designed for downstream success. Thus, in addition to being chemically diverse, the fragments are typically hydrophilic and amenable to chemical elaboration. Fragments identified as hits can serve as starting points for similarity searches or chemical elaboration to arrive at lead series for further medicinal chemistry work.

inline image

In an effort to identify novel fragment ligands for PDK1, an NMR-based fragment screen was performed (28–31). Saturation transfer difference (STD) NMR experiments were collected to identify compounds that bind to PDK1. Identification of Compounds 1 and 2 as fragment hits for PDK1 is illustrated in Figures 2 and 3, respectively. In an STD experiment, signals are only observed for compounds that experience magnetization transfer from the protein. Thus, in a mixture of compounds, only ligand signals will be observed. No signals from non-binding compounds will be present. Since the magnetization transfer is detected in the signals of the free ligand, a ligand that binds too tightly (with affinity in the low μm range or tighter) may be missed by this method. Fragment compounds are expected to have much weaker affinity (∼100 μm or weaker) and thus be detected in an STD-based screen. A comparison of the STD signals observed for the mixture of fragments in well H3 of library plate 3 with the reference 1D 1H NMR spectra for each component fragment identifies binding signals for Compound 1 in spectrum D of Figure 2. Likewise, a comparison of the STD signals observed for the mixture of fragments in well F10 of library plate 6 with the reference 1D 1H NMR spectra for each component fragment identifies binding signals for Compound 2 in spectrum A of Figure 3. Interestingly, in Figure 2, binding signals are also observed in the STD spectrum for the compound in spectrum A, indicating that this mixture contains two PDK1 ligands. Since an STD signal is indicative of weak binding which can often be non-specific, it is not uncommon to observe STD signals for more than one compound in a given mixture. A total of 372 fragment hits were identified from a fragment library containing 10 237 compounds.

image

Figure 2.  Identification of Compound 1 (spectrum D) as a ligand for PDK1 by comparing the STD NMR spectrum recorded for a mixture of 10 fragments (bottom spectrum) with the reference 1D 1H NMR spectra for each fragment in the mixture (spectra A–J). Arrows connect resonances observed in the STD spectrum with their corresponding signals in the 1D 1H reference spectrum of the ligand. The reference spectrum of Compound 1 contains other resonances that do not correspond to its structure. These are either impurities or degradation products. Since they are not observed in the STD spectrum, they do not interact with PDK1.

Download figure to PowerPoint

image

Figure 3.  Identification of Compound 2 (spectrum A) as a ligand for PDK1 by comparing the STD NMR spectrum recorded for a mixture of nine fragments (bottom spectrum) with the reference 1D 1H NMR spectra for each fragment in the mixture (spectra A–I). Arrows connect resonances observed in the STD spectrum with their corresponding signals in the 1D 1H reference spectrum of the ligand.

Download figure to PowerPoint

Since all sites on PDK1 were available for interactions during the assay, ligands can potentially bind at the ATP site, other pockets of the active site, the allosteric PIF pocket, or even to other specific or non-specific sites on the protein. The low information content of a first-pass binding assay should be complemented by activity measurements to prioritize fragment hits. Binding fragments were thus tested for activity using a Kinase-Glo bioluminescent assay that measures ATP depletion or a Caliper microfluidic mobility-shift assay that measures incorporation of a phosphate into a fluorescent-tagged substrate peptide (data not shown). A subset of fragments that were either inhibitors with high ligand efficiencies, activators, and/or had very novel chemical structures were then subjected to competition STD NMR experiments in order to distinguish binding at the ATP site from binding at the allosteric PIF-pocket. Competition data are indicative of binding specificity. In the absence of cooperative binding, a compound that binds in the ATP site should only be competed out by a known ATP-site binder, while a compound that binds in the PIF pocket should only be competed out by a peptide targeting this site. In our experiments, STD NMR spectra recorded for a given compound in the presence of PDK1 were compared with similar spectra recorded after the addition of either the known ATP-site binder staurosporine (IC50 = 6.5 nm) (32) or PIF pocket binder PIFtide (Kd = 65 nm) (13). Binding of Compound 1 is completely displaced by staurosporine (Figure 4B) but not at all by PIFtide (Figure 4C). The opposite result is observed for Compound 2. Its binding is completely abolished by PIFtide (Figure 4F) but is not affected by staurosporine (Figure 4E). This data strongly suggests that Compounds 1 and 2 bind specifically at the ATP site and PIF pocket, respectively.

image

Figure 4. 19F{1H-decoupled} NMR spectra following the conversion by PDK1 of 2-fluoro-ATP (23.99 ppm) to 2-fluoro-ADP (24.02 ppm). (A) Control reaction without fragments. (B) 300 μm Compound 1. (C) 300 μm Compound 2. (D) 300 μm Compound 3. (E) Control reaction without fragments and with EDTA added at time zero.

Download figure to PowerPoint

An 19F NMR-based activity assay that monitors the conversion of 2-fluoro-ATP to 2-fluoro-ADP (33,34) was also developed in order to assay fragments under conditions similar to those used in the initial screen (35). NMR-based activity assays provide an important orthogonal assay format since high fragment concentrations can interfere with signal detection in typical activity assays. T308tide was used as the peptide substrate since it is known not to interfere with binding at the PIF site (7). The results for Compounds 13 are shown in Figure 5. Compound 3 is a close analog of Compound 2 identified in a substructure search that has improved aqueous solubility. Compared with control reactions with no added ligand (Figure 5A) and with EDTA added at time zero (Figure 5E), Compounds 1 and 2 are observed to completely or partially inhibit PDK1 at 300 μm (Figure 5B and C), while Compound 3 is observed to accelerate the PDK1 reaction about two-fold at 300 μm (Figure 5D). The 19F NMR assays were run several times with consistent results.

image

Figure 5.  STD NMR spectra for Compounds 1 and 2 in the presence of PDK1 (A and D) and after the addition of staurosporine (B and E) or PIFtide (C and F). Staurosporine and PIFtide were added at 10–15 μM (about three times the concentration of PDK1).

Download figure to PowerPoint

Compounds 13 were also run in the Kinase-Glo and Caliper assays up to concentrations as high as 313 μm. Since the PIF pocket is blocked by the PDKtide substrate in the Kinase-Glo assay, this format would only measure the activity of highly potent PIF pocket-binding compounds that would be capable of displacing bound PDKtide from this pocket. By contrast, since the T308 substrate used in the Caliper assay does not bind into the PIF pocket, this format allows for the measurement of allosteric modulation from interactions at the PIF pocket. Compound 1 had an IC50 of 52 ± 6 μm in the Kinase-Glo assay (Figure S1). This is consistent with the STD competition experiments indicating that the compound is soluble, does not precipitate the protein, and binds at the ATP site. Compounds 2 and 3 were inactive in this assay, and did not show any inhibition even at 313 μm. However, in the Caliper assay, Compounds 2 and 3 showed 33 ± 13% and 30 ± 12% activation, respectively, compared with control reactions, at 313 μm (Figure S2). Unfortunately, neither compound was sufficiently soluble to determine AC50 values. The observed activation for Compound 2 is consistent with the STD competition data indicating that this compound binds at the allosteric PIF pocket. The discrepancy between the 19F NMR-based assay and the Caliper assay for Compound 2 probably results from the tendency of this compound to aggregate in aqueous solution over time. The much shorter duration of the Caliper assay experiment, which because of its higher sensitivity compared with the NMR measurements is quenched at approximately 10% product formation, more likely reflects the true activity of Compound 2.

Ligand-induced changes in the 1H-15N TROSY spectrum of PDK1 were characterized in order to provide further evidence for the binding sites of Compounds 13. The 1H-15N TROSY experiment provides a two-dimensional correlation of a protein’s amide proton and nitrogen resonances. It can be thought of as a structural fingerprint. Binding of a ligand to the protein will induce changes in the chemical shifts of the protein’s 1H and 15N nuclei which will be manifested by changes in the pattern of 1H-15N correlations. For the most part, the largest changes will be observed for the 1H-15N correlations that correspond to amide backbone groups of the protein that are closest to the ligand-binding site. Chemical shift changes induced by staurosporine and short PIFtide were used as reference points to identify patterns of 1H-15N correlation changes that occur when the ATP site and PIF pocket are occupied, respectively. Short PIFtide was used instead of full-length PIFtide in order to minimize the overall peptide-induced 1H-15N correlation changes and to concentrate the existing changes as close to the PIF pocket as possible. As shown in Figure 6, these patterns are characteristic for the two binding sites even in the absence of sequential resonance assignments. Staurosporine binding induces large chemical shift changes for several well-resolved 1H-15N correlations as indicated by the circles in Figure 6A. By contrast, these same 1H-15N correlations are only very slightly affected by the addition of short PIFtide as shown in Figure 6B. Short PIFtide does, however, induce large chemical shift changes in a separate set of 1H-15N correlations illustrated by the circles in Figure 6B. By contrast, this set of 1H-15N correlations experiences only slight changes in the presence of staurosporine. Thus, binding to the ATP site and the PIF pocket induce characteristic 1H-15N correlation changes that can be used to distinguish between ligand binding at these two sites.

image

Figure 6.  Overlays of the 1H-15N TROSY spectra of 50 μm PDK1 with no ligand added (red resonances) and in the presence of added ligand (green resonances). Ligands added were 75 μm staurosporine (A), 100 μm short PIFtide (B), 250 μm Compound 1 (C), 250 μm Compound 2 (D) and 250 μm Compound 3 (E). Circles identify 1H-15N correlations that experience large changes in the presence of ligand, while boxes identify key 1H-15N correlations that experience only minimal changes in the presence of ligand.

Download figure to PowerPoint

Comparison of the 1H-15N TROSY spectrum of PDK1 in the presence of a ligand with the corresponding spectrum collected with added staurosporine or short PIFtide can identify the binding site of the ligand. Figure 6C shows the 1H-15N TROSY spectrum of PDK1 in the presence of Compound 1. The pattern of induced 1H-15N correlation changes is very similar to what is observed in the presence of staurosporine shown in Figure 6A. This provides very strong evidence that Compound 1, like staurosporine, binds in the ATP site. By contrast, Figure 6D shows the 1H-15N TROSY spectrum of PDK1 in the presence of Compound 2. The pattern of induced 1H-15N correlation changes is very similar to what is observed in the presence of short PIFtide shown in Figure 6B. This provides very strong evidence that Compound 2, like short PIFtide, binds in the PIF pocket. Likewise, the pattern of 1H-15N correlation changes induced by Compound 3 shown in Figure 6E indicates that this compound also binds in the PIF pocket.

Models suggesting how Compounds 2 and 3 bind in the PIF pocket are illustrated in Figure 7. Compounds were docked into the binding site and then optimized using flexible protein-ligand minimization. The carboxyl groups in both compounds most likely mimic the phosphate group on the substrate’s hydrophobic motif. The modelling results indicate that the carboxyl groups form strong electrostatic interactions with R131 and Q150. Residues that experienced the most movement during the minimization, such as residues in the αB and αC helices and adjacent residues, are thought to be involved in regulating the activation of PDK1 and interact with the hydrophobic motif. More specifically, movements in L113, I118, V124, V127, L155, and F157 that occur during the constrained protein–ligand minimization result in more optimal hydrophobic interactions with the aromatic rings of both compounds. The movement of these residues affects the conformation of R129, which may play a role in PDK1 activation since its side chain is in contact with the activation loop in the 2BIY crystal structure.

image

Figure 7.  Modelling of Compound 2 (A) and Compound 3 (B) binding to the PDK1 PIF pocket. Protein backbone and side chains from the crystal structure are represented in green, whereas the modeled protein backbone and side chains are represented in gold. Docked compounds are shown in purple.

Download figure to PowerPoint

Conclusions and Future Directions

  1. Top of page
  2. Abstract
  3. Methods and Materials
  4. Results and Discussion
  5. Conclusions and Future Directions
  6. References
  7. Supporting Information

Small molecule allosteric activators of PDK1 have been reported previously (13). In that work, PIF site-targeted in silico screening of a 60 000 compound chemical library identified several compounds that increased the intrinsic activity of PDK1. Structure–activity relationships for related compounds indicated that a negatively charged carboxyl group was critical for activation. Biochemical studies correlating the effect of PIF site mutations on PDK1 activation, combined with biophysical studies using surface plasmon resonance that indicated competition with PIFtide, provided convincing evidence that the identified compounds interacted at the PIF site of PDK1. The structure of the lead compound identified by this process is shown here as Compound 4.

The work described here has used a complementary, orthogonal approach and has validated the conclusion by Engel et al. (13) that the PIF pocket of PDK1 is amenable to small molecule modulation. Starting from a fragment library of 10 000 diverse compounds, NMR spectroscopic techniques identified binding compounds, distinguished PIF site binding from ATP site binding, and demonstrated allosteric activation of PDK1 by PIF site binding compounds. It is interesting that Compounds 2 and 3 from our work and Compound 4 from Engel et al. (13) are structurally similar. Each contain a negatively charged carboxyl group sandwiched by two aromatic-ring hydrophobic moieties. Just as the carboxyl group mimics the phosphate group of the phosphorylated substrate, the two aromatic groups most likely mimic the two phenylalanine side chains of the PIFtide’s FRDFD motif. In the context of building structure–activity relationships, it might be interesting to combine the features present in Compounds 24 to determine if the hybrid molecule has improved activation properties compared with the original compounds. The relative positioning of the carboxyl group and the aromatic rings in Compounds 2 and 3 compared with Compound 4 are different enough, however, that the compounds identified here may have a distinct binding mode compared with the compound identified by Engel et al. (13). Further structural characterizations of PIF pocket ligand interactions using X-ray crystallography will be required in order to fully understand the molecular recognition features that modulate allosteric activation. Structure–activity relationships of PIF pocket modulators may then be able to distinguish features that are responsible for binding energy from those that are responsible for enzyme activation.

The weakly active fragments described here can potentially serve as starting points for substructure or similarity searches to identify compounds that might demonstrate increased allosteric activation. If such compounds can be identified, they will provide relatively selective tools targeting the PIF pocket. With these tools in hand, it may be possible to further elucidate the contribution of PDK1 activity in the PI3K/Akt signalling pathway in the etiology of such diseases as diabetes and cancer. Small molecule modulation of the PIF pocket validates PDK1 as a novel therapeutic target for diabetes, where allosteric activators of PDK1 may ultimately be found to modulate the insulin signalling pathway in such a manner that leads to new treatments for this disease. Conversely, allosteric inhibitors may prove to be useful cancer treatments because of PDK1’s role in growth-factor signalling. In this case, PIF pocket binding inhibitors would be expected to have improved specificity profiles compared with ATP competitive inhibitors since the PIF pocket motif occurs far less frequently across protein families than do ATP-binding sites. The recent success of targeting the peptide-binding site rather than the ATP site of the protein kinase JNK with small molecules suggests that this may be possible (36).

References

  1. Top of page
  2. Abstract
  3. Methods and Materials
  4. Results and Discussion
  5. Conclusions and Future Directions
  6. References
  7. Supporting Information

Supporting Information

  1. Top of page
  2. Abstract
  3. Methods and Materials
  4. Results and Discussion
  5. Conclusions and Future Directions
  6. References
  7. Supporting Information

Figure S1. Kinase Glo assay data for Compound 1. Averages and standard deviations are derived from four replicates at each concentration.

Figure S2. Caliper assay data for Compounds 2 and 3. Averages and standard deviations are derived from four replicates at each concentration.

Please note: Wiley-Blackwell is not responsible for the content or functionality of any supporting materials supplied by the authors. Any queries (other than missing material) should be directed to the corresponding author for the article.

FilenameFormatSizeDescription
CBDD_768_sm_Figure S1.ppt38KSupporting info item
CBDD_768_sm_Figure S2.ppt39KSupporting info item

Please note: Wiley Blackwell is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.