Pharmacophore design, in silico screening, and interactive docking

The high-resolution crystal structure of the isolated PH domain of human AKT1 in a complex with the head group of Ins(1,3,4,5)P₄ (32) was utilized to define a pharmacophore pocket for screening a 2,000 molecule database (National Cancer Institute) using Unity in Sybyl (version 7.2; Tripos Inc., St Louis, MO). The pharmacophore pocket included all the residues of the AKT1 crystal structure within 5Å of the Ins(1,3,4,5)P₄ binding site: Lys14, Arg15, Gly16, Glu17, Tyr18, Ile19, Lys20, Thr21, Arg23, Pro24, Arg25, Lys39, Pro51, Leu52, Asn53, Asn54, Phe55, Gln79, Ile84, Glu85, Arg86 and Phe88. The program operates by assigning attributes to various atoms on the ligand or protein binding site. The defined pharmacophore pocket was used to search virtual chemical databases and candidate compound “hits” were identified. The FlexX docking algorithm in Sybyl V.7.2 was utilized for the docking of these compounds into the AKT1 PH domain active site. FlexX produces 30 different docking orientations (poses) of the ligand within the active site. Various docking orientations were analyzed on the basis of FlexX scores, G-score, and X-score. The scores are similar to interaction energy, and the more negative the value, the better the interaction. The predicted K_D is calculated by pK_D = 10 exp(−Xscore) (38). In order to investigate the possibility of specific binding of the identified small molecules at the AKT1 PH domain using in silico methods, known crystal structures of the IRS1 PH domain (IRS1, PDB:1QQG) (39) and of the PDK1 PH domain (PDK1, PDB:1W1D, 1W1G) (40) were also used for docking studies similar to those described above.

Synthetic Procedures

Details of the syntheses and characterizations of the compounds used herein are located in the Supplemental Data Section.

Expression and purification of recombinant PH domains

Recombinant mouse AKT1 PH domain amino acids 1–111 (UBI/Millipore, Charlottesville, VA), human AKT1 PH domain amino acids 1–111 (Origene NM005163.2), human IRS1 PH domain amino acids 12–133 (Invitrogen, #IOH29016) and human PDK1 PH domain amino acids 407–549 (Origene, NM002613.3) were cloned by PCR into EcoRI/XhoI sites in pGEX-4T1 inducible bacterial expression plasmid (GeneStorm, InVitrogen, Carlsbad CA) transformed into BL21(DE3) E. Coli.

Surface plasmon resonance (SPR) spectroscopy binding assays

All interaction analyses were performed with a Biacore 2000, Biacore 2000 Control Software v3.2, and BIAevaluation v4.1 analysis software (Biacore, Piscataway, NJ). The PH domain GST-fusion proteins (AKT1, IRS1, and PDK1) were immobilized on a CM5 Sensorchip (Biacore BR-1000–12) using Biacore’s Amine Coupling Kit (Biacore BR-1000–50) to a level of 10,000 Response units (RUs). Small molecule analytes at concentrations ranging from one tenth to ten times the predicted K_D were injected at a high flow rate (30μL/min). Dimethylsulfoxide (DMSO) concentrations in all samples and running buffer were 1% (v/v) or less.

ELISA competitive binding assay

A 96-well Maxisorb plate (Nunc, Rochester, NY) was coated with 1μG/100μl L-α-phosphatidylinositol(3,4,5)P₃ (Biomol, Plymouth Meeting, PA). The proteins (purified GST-PH domains) were incubated with the drugs for 4 hours in 0.2 M carbonate buffer pH 9.4. The proteins mixed with increasing concentration of drugs were then added to the 96-well plate and incubated overnight at 4ºC. The plate was washed 4 times with phosphate buffered 0.9% NaCl (PBS), blocked with 3% bovine serum albumin (BSA) in PBS and 0.01% Tween for 1 hour, washed again 4 times with PBS and mouse monoclonal anti-glutathione-S-transferase antibody (Pierce) in 3% BSA (1: 2000) was added for 1 hr at room temperature shaking. The plate was washed 4 times with PBS and an anti-mouse IgG horseradish peroxidase coupled antibody (dilution 1: 2000 in 3 % BSA) was added for 1 hr. Following 4 washes with PBS, 2,2′-azinobis [3-ethylbenzothiazoline-6-sulfonic acid]-diammonium salt (ABST, from Pierce) was added and the reaction was allowed to develop for 30 min. A stop solution of 1% sodium dodecyl sulfate was then added and the plate was read at 405 nm in a plate reader (μQuant Bio Tek Instruments Inc., Winooski, VT).

Cell culture and compound treatments

Human HT-29 colon cancer cells and NIH3T3 mouse fibroblast cancer cells were obtained from the American Type Culture Collection (Rockville, MD). Cells and maintained in bulk culture in Dulbecco’s modified Eagle medium (DMEM) supplemented with 10% heat-inactivated fetal bovine serum (FBS), 4.5 g/l glucose, 100 U/ml penicillin and 100 μg/ml streptomycin in a 5% CO₂ atmosphere. Cells were passaged using 0.25% trypsin and 0.02% EDTA. Cells were confirmed to be mycoplasma free by testing them with an ELISA kit (Roche-Boehringer Mannheim, Indianapolis, IN). The compounds were dissolved in DMSO at a stock concentration of 10 mM and then added at different concentrations directly into the culture media of the cells.

Inhibition of phospho-Ser⁴⁷³AKT, phospho-Ser²⁴¹PDK1 and phospho-(pan)-PKC by Western analysis

Inhibition of the phosphorylation of AKT,1 PDK1 or PKC was measured by Western blotting as described previously (36) using rabbit polyclonal antibodies to phospho -Ser⁴⁷³-AKT, pan-phospho-PKCs, and phosho-Ser²⁴¹-PDK1 (New England Biolabs/Cell Signaling Technology Inc., Danvers, MA). Bands corresponding to phospho-AKT, phospho-PKC or phospho-PDK1 were quantified using Eagle Eye software (BioRad, Richmond, CA) and Kodak X-Omat^™ Blue XB (NEN^™, Life Science Products). β-actin was used as a loading control in all Western blots.

Cell cytotoxicity assay

Cell growth inhibition was determined using a micro-cytoxicity assay. Cells were plated in 96-well micro-cytoxicity at 5,000–10,000 cells per well (depending on cell doubling time) and grown for 7 days. Compounds dissolved in DMSO were added directly to the media, at various concentrations ranging from 1 to 50 μM. The endpoint was spectrophotometric determination of the protein content of each well using 3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide. A concentration-response relationship at two or more concentration levels was used to obtain an IC₅₀. for the compound.

Cell Apoptosis assay

Apotposis was measured as described previously in reference 41. Briefly, HT-29 cells were grown to 70–75% confluency in 6-well tissue culture plates. Cells were treated with the compounds for 24 hours. To measure apoptosis, 10 μl of cells were mixed with ethidium bromide and acridine orange solution (100 μg/ml each in DMEM) and visualized by immunofluorescence for morphological changes. A minimum of 200 cells was counted and the percentage of apoptotic cells determined.

Anti-tumor activity

Approximately 1×10⁷ HT-29 colon cancer cells in log phase growth were injected subcutaneously (sc) in 0.1 ml PBS into the flanks of female severe combined immunodeficient (scid ) mice. When the tumors reached volumes between 100 and 170 mm³, the mice were stratified into groups of 8 animals having approximately equal mean tumor volumes and administration of compound 1 suspended in 0.2 ml 25% dimethylsulfoxide in 20% pharmaceutical grade hydroxypropyl-β-cyclodextrin (Trappsol^®, Cyclodextrin Technologies Development, High Springs, FL) in water was started at a dose of 250 mg/kg per day per os (po) daily for 5 days. The animals were weighed weekly and tumor diameters measured twice weekly at right angles (d_short and d_long) with electronic calipers and converted to volume by the formula volume = (d_short)² × (d_long)/2 (42). When the tumor volume reached 2,000 mm³ or became necrotic, the animals were euthanized.

Pharmacokinetic Studies

Male C57Bl/6 mice were administered compound 1 intraperitonealy (ip) or po at 250 mg/kg suspended in 0.2 ml 25% DMSO in 20% Trappsol®. The mice were sacrificed at various times, blood was collected into heparinized tubes and plasma was prepared. Plasma (0.2 ml) was immediately mixed with an equal volume of 0.25 M sodium phosphate buffer at pH 4.0 and extracted for 1 hr by inversion with 4 ml ethyl acetate. After centrifugation, 3.8 ml of the organic layer was removed and evaporated under N₂ and the residue dried on a lyophilizer. Chromatographic separation was achieved with a Waters Symmetry C-18 3.9 × 150 mm column (Waters, Milford, PA) with a mobile phase of 0.1% trifluoroactetic acid in 60% methanol, at a flow rate of l/min with detection at 254 nm. For the assay, the sample residue was dissolved in 100μl mobile phase and centrifuged at 15,000 g for 5 min at 4°C. The limit of detection of the assay for all the compounds from 0.2 ml mouse plasma was 0.01 μg/ml.

Toxicity Studies

Compound 1 was administered at 250 mg/kg per day by oral gavage (po) daily for 5 days to female scid mice. The mice were sacrificed 24 hr after the last dose and changes in body weight from the start of the study, blood lymphocyte, neutrophil, red blood cell and platelet counts, and aspartate amino transferase (AST) and amino alanine transferase (ALT) were measured.

Pharmacodynamic Studies

1×10⁷ HT-29 colon cancer cells were injected sc into the flanks of female scid mice and allowed to grow to approximately 300 mm³. Mice received a single ip dose of compound 1 of 250 mg/kg in 25% DMSO in 20% Trappsol^® in water. Mice were killed at various times and tumors removed and immediately frozen in liquid N₂. Tissues were homogenized in 50mM HEPES buffer, pH 7.5, 50mM NaCl, 1% Nonidet^® P40 and 0.25 % sodium deoxycholate, and Western blotting performed using anti-phospho-Ser⁴⁷³-AKT and anti-AKT antibodies. AKT activation was expressed as the ratio of phospho-Ser⁴⁷³-AKT to total AKT.

Discovery of lead molecule 1 (NSC 348900) and its binding to the AKT1 PH domain

PtdIns(1,3,4,5)P₄ and its reported interactions with the crystal structure of the AKT1 PH domain (32) provided the starting point for the identification and design of AKT1 PH domain small molecule inhibitors. A pharmacophore query search using the National Cancer Institute database led to the identification of several lead molecules. These lead molecules were docked and then ranked based on their docking scores. One of these molecules - NSC348900 (compound 1) exhibited good FlexX score and G-score values as summarized in Table I and was selected as a lead for future studies. The predicted binding affinity (K_D) of compound 1 to the AKT1 PH domain was 1.2 μM, which was three times better than the lipid-based compound, DPIEL with a predicted K_D of 4.0 μM (Table I). Figure 1A shows the predicted binding of compound 1 to the PH domain binding pocket of AKT1 and Figure 1B represents hydrogen bonding interactions that occur between compound 1 and the amino acid side chains, as well as the backbone of the AKT1 PH domain binding pocket. The sulfonamido group interacts with Arg86 through a hydrogen bond while a similar hydrogen bonding interaction is involved with the diazopyrazolyl group with Arg23. These two arginine residues are involved in the strong interaction with the phosphate head groups of the substrate PtdIns(1,3,4,5)P₄ supporting the discovery of compound 1 as an AKT1 PH domain inhibitor. Other hydrogen bonds are also established between the backbones of Ile19 and Asn53 with the sulfonamide function of the compound.

Table I. Structures and calculated properties of lead compound 1 (NSC 348900) and analogues (2–6).

Compounds		AKT1 FlexX Score	AKT1 G Score	AKT1 cKD (μM)	PDK1 FlexX Score	PDK1 G Score	PDK1 cKD (μM)	IRS1 FlexX Score	IRS1 G Score	IRS1 cKD (μM)
		NS	NS	4.0	NS	NS	NS	NS	NS	NS
1		−31.0	−96.9	1.2	−17.4	−109.0	1.74	−16.0	−128.0	1.99
2		−29.6	−31.9	2.4	−17.0	−40.0	2.60	−17.1	−96.2	2.40
3		−28.2	−99.5	1.2	−17.1	−103.4	1.70	−14.8	−79.7	10.70
4		−29.1	−71.9	3.0	−17.5	−88.6	2.20	−17.9	−145.5	1.80
5		−33.0	−120.6	1.3	−20.1	−137.1	2.40	−14.6	−90.1	10.70
6		−24.3	−132.0	0.85	−21.0	−109.1	1.45	−14.5	−140.6	0.52

Open in a new tab

NS: for not shown

Rational Design, Chemical Synthesis, and Docking Studies of Compounds (2–6)

In order to test whether modifications of compound 1 could affect the binding properties and activity against AKT1, compound 1 and five structural analogs (compounds 2 to 6) with varying chemical properties were synthesized (Scheme 1). Diazotization of sulfadiazine (7) with sodium nitrite under acidic conditions, followed by treatment of the diazonium salt with ethyl acetoacetate and sodium acetate, gave β-ketoester 8 in 95% yield. Condensation of 8 with different benzoylhydrazides in glacial acetic acid at 100 °C produced NSC 348900 (1) and analogs 2–5 in yields ranging from 19%–71%. Compound 6 was prepared by treatment of compound 1 with sodium hydride and methyl iodide in THF (reaction not shown). Details of these syntheses and spectroscopic data for compounds 1–6 and 8 are given in the Supplemental Data Section. The structures and docking scores for these compounds are summarized in Table I. Analyses of the docking poses of these compounds in the PH domain of AKT1 revealed different docking orientations between compounds 1, 3 and 5 as compared to compounds 2, 4 and 6. Since there are only small changes in the structures of these compounds, it is expected that these differences in docking orientations are due to limitations of the FlexX docking program (for flexible compounds). It was expected that compounds 1 – 6 would exhibit similar binding to the AKT1 PH domain.

The sequence alignments of critical binding residues in the PH domains of the three human isoforms AKT1, AKT2 and AKT3 are identical (Table II) and thus indicate that compounds 1–6 should bind equally well to the three isoforms of AKT. K_Ds were also calculated for the binding of the compounds 1 to 6 to the PH domain of PDK1 and were found to be very similar to those for AKT1 (Table 1). The calculated K_Ds for the binding of compounds 1 to 6 to the PH domain of IRS1 showed greater variability with compound 6 having the greatest affinity and compounds 3 and 5 more than an order of magnitude lower affinity. Figures 1C and 1D represent the binding mode of compound 1 in the binding pocket of the PH domain of PDK1 (Supplemental Figures 1A and 1B show the binding mode of compound 1 in the binding pocket of the PH domain of IRS1). Compound 1 exhibits the reverse docking mode in IRS1 PH and PDK1 PH domains similar to compound 2 in the PH domain of AKT1.

Table II. Sequence alignments of the PH domain of AKT.

The PH domain binding residues are shown underlined in the sequence alignment below obtained using CLUSTAL W (1.82) multiple sequence alignment.

Open in a new tab

Measured Binding Affinity of Compound 1 and Analog Compounds 2–6

In order to validate the in silico data and the docking studies, binding assays using the Biacore SPR biosensor and an ELISA competitive binding assay were used to measure the binding affinity (K_D) of the compounds to all three PH domains. Table III summarizes the results obtained from the Biacore SPR measurements, which correlated well with the predicted K_D values for the compounds for each PH domain. Representative saturation curves as well dose response curves are shown in Figure 2 for compounds 1 and 2 to the PH domain of AKT1 (panel A) and to the PH domain of IRS-1 (panel B). Compounds 1 and 2, which modeling suggest bind in a reverse binding pose in the PH domain binding pockets of the three different PH domains, also exhibited very different biosensor binding curves. An ELISA competitive binding assay was conducted using the PH domains of AKT1 and IRS1 with compounds 1 (Figure 3 panel A) and 2 (Figure 3, panel B). The ELISA gave binding IC₅₀s with AKT1 for compound 1 and 2 of 0.08 μM, and with IRS1 for compound 1 of 1.0 μM, and compound 2 of > 100 μM. These IC₅₀s were similar to ones obtained using the Biacore.

Finally, consistent with the docking studies, compounds 1, 4 and 5 exhibited low K_D while compounds 2, 3 and 6 do not show any binding to IRS-1 PH domain as measured by Biacore SPR. However, compounds 4 and 6 did not bind the PH domain of PDK1 while the calculated K_D was predicted to be 2.2 and 1.4 respectively. Taken together, these data suggest that the structural modifications in compound 1 have altered the binding positions of the compounds in the AKT1 PH domain as well as modified their specificity against IRS1 or PDK1 PH domains. Overall, the predicted binding affinities correlated well with measured binding constants for AKT1 and PDK1 PH domains but were less accurate for IRS-1 PH domain.