Novel STAT3 Small-Molecule Inhibitors Identified by Structure-Based Virtual Ligand Screening Incorporating SH2 Domain Flexibility

Abstract

Efforts to develop STAT3 inhibitors have focused on its SH2 domain starting with short phosphotyrosylated peptides based on STAT3 binding motifs, e.g. pY₉₀₅LPQTV within gp130. Despite binding to STAT3 with high affinity, issues regarding stability, bioavailability, and membrane permeability of these peptides, as well as peptidomimetics such as CJ-887, have limited their further clinical development and led to interest in small-molecule inhibitors. Some small molecule STAT3 inhibitors, identified using structure-based virtual ligand screening (SB-VLS); while having favorable drug-like properties, suffer from weak binding affinities, possibly due to the high flexibility of the target domain. We conducted molecular dynamic (MD) simulations of the SH2 domain in complex with CJ-887, and used an averaged structure from this MD trajectory as an “induced-active site” receptor model for SB-VLS of 110,000 compounds within the SPEC database. Screening was followed by re-docking and re-scoring of the top 30% of hits, selection for hit compounds that directly interact with pY+0 binding pocket residues R609 and S613, and testing for STAT3 targeting in vitro, which identified two lead hits with good activity and favorable drug-like properties. Unlike most small-molecule STAT3 inhibitors previously identified, which contain negatively-charged moieties that mediate binding to the pY+0 binding pocket, these compounds are uncharged and likely will serve as better candidates for anti-STAT3 drug development.

Graphical Abstract

INTRODUCTION

Signal transducer and activator of transcription 3 (STAT3) is one of a 7-member transcription factor family (STAT1, 2, 3, 4, 5A, 5B, and 6) that is activated in response to extracellular mediators including cytokines, growth factors, and hormones. It modulates a large repertoire of genes involved in a number of critical functions: inflammation, cellular proliferation, survival and apoptosis, angiogenesis, fibrosis, transformation, as well as tumor invasion and metastasis (1–5). STAT3 is found to be constitutively activated in 50–100% of many different types of cancers and has been associated with poor prognosis (3, 6–9). Multiple studies suggest that targeting STAT3 in cancer suppresses cell growth and induces apoptosis in vitro and in vivo (1, 2, 10–16) making it an attractive therapeutic target for cancer treatment (17, 18)

Examination of the monomer subunit of STAT3 extracted from the 1BG1 crystal structure of STAT3 homodimer bound to duplex DNA reveals four distinct structural domains: an N-terminal four-helix bundle (residues 138–320), an eight-stranded β-barrel (residues 321–465), an α-helical “linker” domain (residues 466–585), an SH2 domain (residues 586–690), and a loop domain (residues 691–715) (19). The loops within the β-barrel and linker domains are responsible for DNA-binding sequence specificity. The loop domain is phosphorylated on Tyr-705 and the pY₇₀₅-peptide motif within each monomer binds in trans to the SH2 domain of the other monomer, leading to dimerization. The N-terminal domain (residues 1–130; not shown in the crystal structure) is involved in oligomerization, which facilitates binding of multiple STAT3 dimers to two or more adjacent STAT3 DNA-binding elements within a gene promoter. The C-terminal domain of STAT3α (residues 716 to 770; not included in the crystal structure) contributes to transcriptional activation, while the C-terminal domain of STAT3β (residues 716–722) is responsible for its prolonged nuclear retention compared to STAT3α (20–22). The SH2 domain of STAT3 is required for its recruitment to ligand-activated receptor complexes and for its homodimerization (18, 23, 24). Due to the moderately high affinity and specificity of SH2 binding to its cognate pY-peptide ligand motifs (25), targeting the SH2 domain is among the most viable strategies for inhibit STAT3 signaling.

The first STAT3 inhibitors identified were phosphotyrosylated (pY) peptides derived from peptide sequences shown to be bound by the STAT3 SH2 domain, such as P-pY₇₀₅LKTK within the C-terminal of STAT3 (26) and pY₉₀₅LPQTV within gp130 (24). However, proteolytic cleavage, which results in short plasma half-life, along with poor oral bioavailability and low cell-membrane permeability have limited the clinical development of peptide inhibitors (27). To achieve better pharmacokinetic properties, peptidomimetic inhibitors derived from STAT3 SH2 pY-peptide ligands have been developed (28–30). Among them, the conformationally constrained peptidomimetic, CJ-887 and its derivatives, which are based on pY₉₀₅LPQTV, achieved high binding affinity as reflected in K_i values of 15 nM (30). Similar to pY-peptides, however, lack of cell permeability and poor drug-like properties remain major obstacles for these compounds to be further developed for clinical use. Consequently, great effort has been expended to identify small-molecule STAT3 inhibitors with favorable drug-like properties and high potency (18). Structure-based virtual ligand screening studies (SB-VLS) have been performed with a number of active hits identified and lead compounds developed from them (31–33). However, agents targeting STAT3 have been slow to enter the clinic, in part, because of difficulties inherent in targeting transcription factors, a class of proteins deemed “undruggable” due to the large size of their protein-protein interaction interfaces (34). Besides, serious adverse events (SAE), including lactic acidosis and peripheral neuropathy, were observed with some small-molecule STAT3 inhibitors in clinical-stage development (35, 36). These were attributed to targeting of STAT3’s non-canonical functions, most notably, its contribution to mitochondrial-mediated oxidative phosphorylation (37), which relies on phosphorylation of STAT3 on serine 727, as opposed to tyrosine 705 (38, 39).

One possible reason that SB-VLS has failed to this point in identifying hit compounds that bind STAT3 with high-affinity may be the high mobility of the STAT3 SH2 domain. In the crystal structure of dimers of the core STAT3 protein bound to DNA, the phosphopeptide binding region within the SH2 domain is resolved to only ~20 Å due to conformational flexibility (19). In addition, the crystal structure provides only a static snapshot of the domain’s structure, which may be close to the “real” conformation in solution for a rigid domain, but may differ substantially from the structure in solution of a highly flexible domain, such as the SH2 domain of STAT3. Of note, the conformational flexibility of a protein is closely related to its functional activity with conformational changes occuring commonly in many types of protein (40, 41). Furthermore, in some cases, the induced binding pocket exhibits a more druggable site than its rigid counterpart and is of higher yield in drug design and discovery (42, 43).

To this end, we conducted molecular dynamics (MD) simulations of the STAT3 SH2 domain in a complex with CJ-887 and found that it induced protein conformation changes that favor ligand binding. An averaged structure from MD trajectory was calculated, optimized further and used as a receptor model for SB-VLS; that takes protein flexibility into consideration. This “induced-active site” strategy involved in silico screening followed by re-docking, re-scoring, selection for hit compounds that directly interact with pY+0 binding pocket, and testing for STAT3 targeting in vitro. Using this strategy we identified six compounds that inhibited cytokine-induced pY-STAT3 in cells at low micromolar inhibitors of (2.7 – 34.5 μM). Two of these compounds are of high potency, low molecular weight, and fulfill Lipinski’s rule-of-five, and, thus, may serve as excellent starting points for hit-to-lead optimization.

MATERIAL AND METHODS

Cell Lines and antibodies:

Breast cancer cell lines MDA-MB-231 and MDA-MB-468 and the AML line Kasumi-1 were obtained from the cell line core at BCM and ATCC respectively, and cultured in complete DMEM or RPMI respectively, with 10% FBS, antibiotics penicillin, streptomycin and amphotericin. WT-STAT3 was expressed in STAT3−/− MEFs (44) and selected for cells stably expressing AcGFP1-tagged WT-STAT3 (AcGFP1-STAT3) or empty vector (AcGFP1-vector) using neomycin (G418) selection. The following antibodies, pJak1 (Y1034/1035) (D7N4Z) Rabbit mAb (cat #74129), total jak1 (6G4) Rabbit mAb (cat # 3344), pJak1 (Tyr1034/1035)/Jak2 (Tyr1007/1008) (E9Y7V) Mouse mAb (cat #66245), total Jak2 (D2E12) XP(R) Rabbit mAb; were procured from Cell Signaling technology (MA, USA) and the Anti-β-Actin Mab (cat# A5441) was obtained from Sigma-Aldrich/Millipore.

Molecular docking:

The STAT3 SH2 domain was taken from x-ray crystal structure 1BG1 (19) stored in Protein Data Bank (PDB). Only residues 466 to 716 of the monomer, including the linker domain, SH2 domain and the loop bearing Tyr-705, were kept. Residues 1–465, including the N-terminal DNA-binding, 4-helix bundle and β-barrel domains, were removed since they do not have direct interactions with the SH2 domain and the dimerization interface. The missing residues 689–701 in the crystal structure were constructed and minimized by Prime with OPLS force field. The modeled structure was subjected to Protein Preparation Wizard workflow in Maestro 9.2. Bond orders were assigned and all hydrogen atoms were minimized to reach the convergence of RMSD = 0.3 Å with OPLS force field. A grid-enclosing box was centered on the GLU 638 to enclose residues located within 20 Å, where the phosphorylated peptide located. A scaling factor of 1.0 was set to van der Waals (VDW) radii of those receptor atoms with partial atomic charge less than 0.25. The “Builder” module in Schrodinger was used to build the molecular structure of CJ-887 and OPLS force field was applied to obtain the minimized energy structure. LigPrep (45) module was performed to get the diverse conformation of CJ-887 with pH value of 7.0 ± 2.0.

Molecular dynamics simulation:

To relax the docking pose of CJ-887, molecular dynamics simulation was performed for the STAT3-CJ-887 complex. RED was used to do charge derivation in cooperation with Gaussian 09 program at the b3lyp/6–31G* level for the ligand (46, 47). TLEaP was used to solvate the systems in a truncated-octahedron box of TIP3P water molecules (48) with 10 Å of minimum distance from the atoms of protein or ligand to the border of the box. Na+ atoms were added to obtain an electrically neural system. MD simulations were carried out with AMBER11(49) on BlueBiou of IBM supercomputing cluster in Rice University. The standard amber ff99 force field (50) and the general AMBER force field (GAFF) (51) were used as the parameters for protein and ligand, respectively. Calculations employed the particle mesh Ewald method (52) to treat the long-range electrostatic interactions with periodic boundary conditions and a cut-off value of 10Å. All bonds containing hydrogen atoms were constrained using the SHAKE algorithm (53) along with time step of 2 fs. The temperature is controlled by Langevin thermostat (54). Energy minimization was executed by the steepest descent method for the first 2,000 steps and followed by the conjugated gradient method for the second 2,000 steps with 5 kcal/mol*Ų restraints on all the atoms of protein and ligand. The temperature was increased from 0K to 300K gradually over 40 ps with 10 kcal/mol*Ų restraints on the solvent, and then the force constant was decreased to 5, 1 and 0 kcal/mol*Ų, respectively, for the following three 40 ps simulations, using the NTP ensemble to relax the water molecules and Na⁺ ions. 10 ns production simulation was done at 1atm and 300K under the NTP ensemble with a time step of 2 fs.

Virtual ligand screening protocol:

All molecular structures from SPECS screening compounds library (www.specs.net) were prepared using LigPrep (45) in Schrodinger to add hydrogen atoms and predict diverse protonation states within the scale of pH value from 5.4 to 9.4. Enumeration of stereoisomers and tautomers was also taken into consideration during the ligand prepare process. The averaged structure obtained from MD simulation was minimized by the steepest descent method for the first 2,000 steps and followed by the conjugated gradient method for the second 2,000 steps by using AMBER11. The minimized structure was then submitted to the Protein Preparation Workflow in Maestro (55). Bond orders were assigned and all hydrogen atoms were minimized to reach the convergence of RMSD = 0.3 Å with OPLS force field. The grid generation process was the same as the protocol described above for the CJ-887 docking. The high throughput virtual screening (HTVS) mode of Glide (56) was used to explore the binding modes and evaluate the binding affinities for all the compounds in the active site with default settings. We redocked and rescored thirty percent of top-ranking molecules with standard precision (SP) parameter settings. Top 10% poses from the SP docking experiment served as the inputs for pose filter. Criteria that were applied to do the molecule selection were (1) hydrogen bonding with at least one of the pY+0 pocket residues including R609 and S613 and (2) structural diversity in chemical space. The whole scheme of hit discovery, enumerating the docking and simulation strategies as well as the functional screens of potential in-silico hits, is stated in Figure 1.

Figure 1. Summary diagram of the structure-based virtual ligand screening strategy incorporating SH2 domain flexibility.

CJ-887 was docked to a monomer of STAT3 extracted from the crystal structure 1BG1, followed by 10 ns molecular dynamics simulation to relax the complex structure. An averaged structure from MD simulation was minimized and used as receptor, to perform two-layer docking experiments by using Glide HTVS and SP parameter sets, respectively. Structural criteria, including binding at pY+0 pocket, were used to filter the docking poses for compound selection. Select compounds, as indicated, were screened for inhibition of constitutive and G-CSF-stimulated pY-STAT3, inhibition of STAT3 binding to its pY-peptide ligand (SPR), and inhibition of STAT3-driven growth, as described in the text.

Surface plasmon resonance (SPR) assay of STAT3 binding pY-peptide.

STAT3 (aa 127–722) at a concentration of 200 nM in 20 mM Tris buffer (pH 8) was pre-incubated without or with compounds prior to injection onto an SA chip immobilized with phosphorylated and control non-phosphorylated biotinylated EGFR derived dodecapeptides based on sequence surrounding Y1068 (57), on a Biacore 3000 biosensor (Biacore inc., Piscataway NJ) and analyzed as described (58).

Luminex bead-based assay.

A Luminex bead-based assay was used to determine levels of pY-STAT1, pY-STAT2, pY-STAT3, pY-STAT5, pY-STAT6, total STAT3, GAPDH and β-Tubulin in protein lysates as described (59). Briefly, Kasumi-1 cells, were serum-starved, pre-treated with compound (0.1/0.3/1/3/10/30/100 μM) or DMSO and then stimulated with 20 μl of G-CSF (10 ng/ml) for 15 minutes at 37°C. The assay was stopped by ice-cold PBS and total protein extracted from cell pellets. Total protein was plated in a 96-well filter plate pre-loaded with beads (Millipore, Danvers, MA) coupled to antibody against pY-STAT1, pY-STAT2, pY-STAT3, pY-STAT5, pY-STAT6, total STAT3, GAPDH and β-Tubulin. Bead-bound analytes were measured using biotinylated detection antibody specific for a different epitope and streptavidin-phycoerythrin (streptavidin-PE) (60, 61). Constitutive pY-STAT1, pY-STAT2, pY-STAT3, pY-STAT5, pY-STAT6 activities were also measured in lysates from non-synchronous cultures of MDA-MB-468 cells treated with either 30 μM of the compound for increasing durations of time (0/15’/30’/1hr/2hrs/4hrs/8hr/12hrs/24hrs/36hrs/48hrs) or increasing doses of compounds for 16hrs (for IC50 estimations) in a similar manner. Data were collected and analyzed using the Bio-Plex suspension array system (Luminex 100 system, Bio-Rad Laboratories, Hercules, CA). GAPDH-normalized pY-STAT3 values from each treatment were corrected for untreated cells, expressed as percentage untreated, and used to determine the IC₅₀ using GraphPad.

Anchorage-independent and dependent cell growth.

Breast cancer cells MDA-MB-468 and MDA-MB-231 previously demonstrated to be STAT3 dependent for growth (53) were (5,000 cells/well) cultured in triplicates in complete DMEM ± drug, in ultra-low attachment 96 well plates for 72hrs or cell-culture treated plates for 48 hrs and viable cells were quantitated using MTT. Optical density (OD) was measured at 590 nm using a 96-well multi-scanner (Synergy H1 microplate reader, BioTek Inc, VT, USA). Relative % viability (viability after any treatment ÷ viability of untreated cells x 100) was plotted along Y-axis. At least 2 replicates experiments were performed and were used for IC₅₀ calculation using GraphPad software. AcGFP1-Vector and AcGFP1-STAT3 cells were plated in triplicates (5,000 cells/well) in complete DMEM. After allowing the cells to adhere for overnight, growth medium was removed and cells starved using serum-free DMEM for 24 hrs and then allowed to grow in low-serum (0.5% FBS) for 72 hrs and viable cells quantified by MTT as stated above.

Western Blotting.

Whole protein from MDA-MB-468 cells treated with 30 μM of SPEC-29 for increasing durations of time (0/15’/30’/1hr/2hrs/4hrs/8hr/12hrs/24hrs/36hrs/48hrs) were quantitated and equal amount of protein (50 μg) were subjected to polyacrylamide gel electrophoresis (PAGE), the gel transferred to nylon membranes, membranes blocked with 5% non-fat milk solution in 1x-Tris Buffered Saline with 0.05% Tween-20 (TBS-T) and probed with antibodies and detected using appropriate HRP-conjugated secondary antibodies and chemiluminescent HRP substrates.

GSH Chemical Stability assay.

10ul of 10mM stock in DMSO, SPEC8, SPEC29 and Stattic were spiked into reaction buffer [50mM HEPES pH7.5 or 50mM Ammonium Bicarbonate, containing 1 mM reduced glutathione (GSH)]. The reactions were monitored by repeated sampling of the same vial at intervals of 5mins for the duration of 55 mins using an Exion LC Sciex equipped with an auto sampler and a UV detector. The stationary phase used was a C18 Synergi™ 4 μm Fusion-RP 80 Å LC column (50 × 2 mm); the mobile phase was water (A) and acetonitrile (B). Reaction rate was monitored at 260 - 295nm and quantified by calculating areas under the curves (AUCs) of each inhibitor using MultiQuant software (sciex).

Alkylation and LC–MS/MS analysis of STAT3 peptides.

10uM of the core fragment of STAT3β (127–722) protein in ammonium bicarbonate buffer was incubated with 500uM STAT3 inhibitors at 37 °C overnight the samples were reduced with TCEP and treated with iodoacetamide (15 mM) for 30 minutes at RT in the dark. Alkylated samples were then digested with trypsin gold in the presence of 1mM CaCl2 for 2hrs 37 °C. Formic acid was added to a final concentration of 5% to stop the digestion prior to LC/MS analysis. Tryptic digests of STAT3 protein samples treated with STAT3 SPEC8, SPEC29 and Stattic were analyzed LC-MS using a QTRAP 5500 Sciex hybrid quadrupole-linear ion trap system with a Turbo Ion Spray ion source in positive mode and equipped with a Sciex LC Exion liquid chromatography system. MRM transitions were generated in Skyline software and exported to the QTRAP 5500 to create the LC-MS methods. Fractionation of the samples was done using a Waters Symmetry C18 column (100Å, 3.5 μm, 4.6 mm x 150 mm,) and a 30 min linear gradient of acetonitrile with 0.1% formic acid at a flow rate of 300 uL/min. Data were analyzed using Skyline, and analyst software.

RESULTS

Binding mode of CJ-887 with STAT3.

CJ-887 is a potent peptidomimetic inhibitor of STAT3 with a K_i value of 15 nM (30). It was designed by modifying the phosphorylated hexapeptide (pY₉₀₅LPQTV) derived from gp130 protein residues 905–910 (62). CJ-887 competitively binds to the SH2 domain of STAT3 and inhibits STAT3 homodimerization, a pre-requisite for its high-affinity binding to duplex DNA (62). The binding mode of CJ-887 revealed by our docking experiments showed that it is similar to the binding orientation of the STAT3 pY705 peptide (AAPpY₇₀₅LKTKFICVTPF) as assessed from the crystal structure of the STAT3-dimer (19). The native pY705 peptide-binding site includes an “U” shaped area of interface that surrounds the projection formed by the side chain of E638 (Figure 2A and B). Three sub-pockets are defined within this area of interface: 1) the pY+0 pocket that binds pY705, 2) the pY+1 pocket that binds L706, and 3) a hydrophobic side pocket that binds pY-X [Figure 2; (63)]. The pY705 residue forms hydrogen bonds with side chains of R609 and S613 in the pY+0 pocket. The phosphorylated phenol in CJ-887, which serves as a pY705 mimic, (2D structure shown in Figure 2D) binds to the pY+0 pocket (Figure 2C) by forming hydrogen bonds with the side chain of R609 within the pocket and K591 within αA, one of the four alpha helices described in the crystal structure of STAT3β (19). The amide side chain in CJ-887 is located on the opposite side of E638, in relation to the phospho-phenol group and forms a hydrogen bond with the carbonyl group of the peptide backbone of E638. Interestingly, the amide side chain of Gln residues within peptide-based STAT3 inhibitors (64) also bind the SH2 domain at this position, which is of critical importance for their binding affinity.

Figure 2. STAT3 pY705-peptide ligand (A, B) and CJ-887 (C, D) binding to STAT3 SH2 domain and averaged structure of CJ-887 bound to STAT3 (E.F).

(A) and (B) are the binding of pY705-peptide with STAT3 SH2 domain extracted from the crystal structure of STAT3-DNA complex (PDB code 1BG1, residues 466 to 716); (C) is the predicted docking pose of CJ-887 with STAT3 SH2 domain; (D) is the 2D structure of CJ-887; (E) is the MD simulation structure of CJ-887 in complex with STAT3; and (F) is the comparison of the docking pose and averaged MD structure. All the proteins are shown in cartoon model except (A) in solvate-accessible surface model. Residues within 5 Å of ligands are shown in line whereas the ligands are drawn in stick model. Hydrogen bonds are shown in yellow dashed lines. Oxygen atoms are colored in red and nitrogen atoms in blue. In (A-C), carbon atoms are colored in green and orange in protein and ligands, respectively. In (E-F), carbon atoms of protein in the docking pose and the averaged structure are colored in green and cyan, respectively; and the carbon atoms of CJ-887 in the docking pose and averaged MD structure are colored in green and orange for clarity.

Peptide immunoblot affinity assays and mirror resonance affinity analysis of phosphopeptides derived from growth factor receptors, e.g. EGFR peptide pY₁₀₆₈XXQ, also demonstrated that only pY-peptides containing Gln at the +3 position (not Leu, Met, Glu, or Arg) bound to STAT3, through H-bonds between the oxygen within the +3 Gln side chain and the backbone amide of Glu-638 (57). As expected, removal of the amide side chain from CJ-887 resulted in a significant loss of potency, indicating that hydrogen bonds formed here also are key interactions for its binding (64). The bicyclic lactam ring located inside the pY+1 pocket made contact with surrounding residues, including L706. Moreover, the carbonyl moiety from the bicyclic lactam formed an additional hydrogen bond with the amide group from the backbone of E638 (Figure 2C). Thus, three groups of hydrogen bonds formed by different parts of CJ-887, anchored it within the “U” conformation of STAT3 SH2 domain pY-peptide binding interface (Figure 2C), mimicking the interactions of different fragments of the phosphorylated peptide (Figure 2B).

Binding site flexibility.

To relax the docking pose, the structure of CJ-887 and STAT3 SH2 domain complex was subjected to 10 ns molecular dynamic (MD) simulations to allow for conformational adjustments of both ligand (CJ-887) and protein (STAT3). The RMSD analysis showed that the system converged after 4 ns MD simulation and remained stable during the simulation (Supplemental Figure 1A). The averaged structure from the last 2 ns MD simulation was extracted and is shown in Figure 2E, F. CJ-887 in the averaged structure still anchors within the “U” shape interface around the projection of the E638 sidechain. The hydrogen bond formed by the bicyclic lactam and the backbone carbonyl of E638 remained unchanged after the simulation. Additionally, the side chain of Q644 flips toward the amide group of CJ-887 and forms a new hydrogen bond replacing the hydrogen bond formed with the backbone of E638 in the docking pose. The hydrogen bond network in pY+0 pocket also changed after the simulation. CJ-887 is tightly hydrogen bonded with the side chain hydroxyl group and backbone amide group from S613 while the hydrogen bonds between the phosphorylated phenol and R609 and K591 disappear (Figure 2F). By comparing the protein structures before and after MD simulation, it was found that αA helix, where K591 is located, conducts movements outward from the central β-sheet strand, causing a change in the spatial distance between the phosphate group and amino side chain of K591 (Figure 2F, Supplemental Figure 1B). The movements of the αA and K591 are critical in that, they introduce a larger space in the pY+0 pocket in the averaged structure (Supplemental Figure 1C, D). In contrast, the positions of conserved residues R609 and S613, as well as V637 and P639, remained unchanged.

In addition, the orientation of phospho-phenol moiety of CJ-887 was adjusted to position it parallel with the flat “wall” formed by side chain and backbone atoms of E638 and P639, allowing hydrophobic contacts in this area (Figure 2E). In addition to hydrogen bonding, the bicyclic ring also makes contact with the hydrophobic side chains of V637 and T714 (Figure 2E). The relative position of the main chain of CJ-887 did not change significantly before or after MD simulation due to the aforementioned hydrogen networks and hydrophobic interactions. Distinct from the main chain of CJ-887, the side chains including the phenyl ring and the acetamide moiety flip away from the original location. No strong polar or hydrophobic interactions are identified between the acetamide group and STAT3, indicating it contributes little to protein binding (Figure 2E, Supplemental Figure 1B). Indeed, the acetamide site was chosen to introduce long lipid chains to increase the cellular permeability for derivatives of CJ-887 without significant influence to binding affinity (30). The simulation results are consistent with this experimental data.

In silico screening and STAT3 inhibitory properties of the hits.

The averaged structure derived from MD simulation was minimized and then used as the receptor for ligand docking and high throughput virtual screening (HTVS) to evaluate binding affinities of 110,000 compounds in the SPEC database. After re-docking and re-scoring thirty percent of the top-ranking molecules with standard precision (SP) parameter settings, the top 10% were selected as inputs into the pose filter along with further restrictions. Importantly, the pose filter was defined to select only the poses that formed hydrogen bonds with residues R609 and S613 in the pY+0 pocket. Many of the top-ranking poses did not form hydrogen bonds with these residues and were filtered out. Subsequently, 110 compounds, fulfilling these criteria, were purchased to test for their ability to inhibit granulocyte colony-stimulating factor (G-CSF)-stimulated tyrosine phosphorylation of STAT3 (pY-STAT3) in Kasumi-1 cells, as described (65). Twenty-four compounds inhibited G-CSF-induced pY-STAT3 by more than 50% at a concentration of 10 μM with 9 compounds at this concentration inhibiting pY-STAT3 by more than 99% (Supplemental Table 1).

We next performed surface plasmon resonance (SPR) on these 24 compounds to determine if their ability to inhibit G-CSF-stimulated pY-STAT3 was due to blocking of binding of STAT3 to its immobilized phosphododecapeptide ligand (EGFR pY1068-peptide), as described (33). Eight compounds that were very active in inhibiting G-CSF-stimulated pY-STAT3 (20–80% inhibition; SPEC-29, 8, 93, 98, 106, 57, 101 and 85; Supplemental Figure 2), inhibited STAT3 binding to EGFR pY-peptide by 29% to 71% at 10 μM and by 67% to 93% at 100 μM (Supplemental Table 2). The binding poses of the 8 compounds (Supplemental Figure 3) demonstrated two common features—occupation of the pY+0 pocket and formation of hydrogen bonds with R609 and S613 per the molecular selection criteria. These 8 compounds were then evaluated to determine their IC₅₀ of inhibition of G-CSF-stimulated pY-STAT3 in Kasumi-1 cells. Five compounds (SPEC-29, 8, 93, 98, and 106) had appreciable inhibitory activity with IC₅₀s ranging from 2.7–19.0 μM (Table 1, Figure 3A). SPEC29 and 8 were identified as the most potent compounds with IC₅₀s of 2.7 and 4.1 μM, respectively.

Table 1.

Summary of inhibitory activities of hits identified within the SPECS compound library.

Lab ID	SPECS Library ID	G-CSF pY-STAT3 IC50 (μM)	Constitutive pY-STAT3, 468 IC50 (μM)	AI growth IC50 (μM)		AD growth IC50 (μM)
				468	231	468	231	AcGFP1-STAT3
SPEC-29	AN-979/41971071	2.7 ± 1.4	5.0 ± 1.9	1.8 ± 1.6	9.5 ± 4.1	2.4 ± 2.5	3.0 ± 2.2	1.0 ± 0.03
SPEC-8	AP-355/42609662	4.1 ± 2.2	5.4 ± 2.8	6.6 ± 4.1	26.1 ± 1.3	6.3 ± 4.6	5.5 ± 3.0	6.0 ± 3.0
SPEC-93	AG-690/37048015	10.4 ± 0.8	43.3± 0.9	11.1 ± 4.5	15.7 ± 1.3	12.1 ± 5.4	12.2 ± 0.3	21.4 ± 5.9
SPEC-98	AG-690/09291009	14.2 ± 8.1	NA	64.1 ± 3.1	NA	32.5 ± 3.7	25.6 ± 3.9	8.6 ± 1.3
SPEC-106	AF-399/15284578	19.0 ± 12.7	ND	34.9 ± 1.3	38.9	44.3 ± 1.6	39.0 ± 0.0	ND
SPEC-57	AN-023/41981716	34.5 ± 30.4	NA	73.5 ± 52.8	NA	44.0 ± 26.9	65.2 ± 17.3	50.0 ± 18.9
SPEC-101	AG-205/36715027	99.0 ± 43.8	ND	20.4 ± 1.0	48.6 ± 30.8	26.9 ± 0.1	21.0 ± 2.4	NA
SPEC-85	AH-487/41138477	>100	NA	NA	NA	NA	NA	NA

Note: Lab ID provided has been used to refer to a compound all though the manuscript. SPEC Library ID is the ID provided at SPECS database at http://www.specs.net Abbreviations: NA: No Activity; ND: Not Done: AD: Anchorage Dependent, AI: Anchorage Independent. Mean ± s.d (standard deviation) of IC50s calculated from ≥ 2 repeat experiments shown. Kasumi-1 cells were used for G-CSF-stimulated pY-STAT3 assays; MDA-MB-468 (468), MDA-MB-231 (231), and AcGFP1-STAT3 MEF cells (as defined in text) were used for growth inhibition

Figure 3. pY-STAT3 inhibition by SPEC compounds.

(A) Inhibition of G-CSF stimulated pY-STAT3 in Kasumi cells by SPEC compounds. Protein from Kasumi-1 cells, pre-incubated with compounds (0, 0.1, 0.3, 1, 3, 10, and 100 μM, 1hr) and stimulated with G-CSF (10 ng/ml, 15 mins) was assayed for pY-STAT3, 5, and 1 and GAPDH levels by Luminex. GAPDH-normalized pY-STAT3 values (% of control) were plotted as function of Log [M] compound, and IC₅₀ calculated using GraphPad. Data presented are representative of at least two repeats. (B-F) SPEC compound inhibition of constitutive pY-STAT3 in breast cancer cell lines. Levels of pY-STAT3, total STAT3 and GAPDH were measured in protein lysates of asynchronous semi-confluent cultures of MDA-MB-468 cells treated with SPEC-29 or SPEC-8 (30 μM for 0, 1, 2, 4, 7, and 12 hrs) using Luminex. GAPDH-normalized pY-STAT3 (B) and tSTAT3 (C) values (% of control) are plotted. (D) GAPDH-normalized levels (% of control) of pY-STAT3, pY-STAT1, pY-STAT5, pY-STAT6, in lysates from MDA-MB-468 cells treated with SPEC-29 for 0, 15’, 30’, 1h, 2h, 4h, 8h, 12h, 24h, 36h and 48h. (E) GAPDH-normalized levels (% of control) of pY-STAT3, pY-STAT1 in protein lysates from MDA-MB-468 cells treated with increasing doses (DMSO, 0.1, 0.3, 1, 3, 10, 100 μM) of SPEC-29, SPEC-8, SPEC-93, SPEC-98, SPEC-57 and SPEC-85 for 16hrs, were plotted as a function of Log [M] compound, and IC₅₀ calculated using GraphPad. IC50 curves for pY-STAT3 (D) and pY-STAT1 (E) from representative experiments shown.

Ability of SPEC compounds to inhibit constitutive STAT3 activity in breast cancer cell line.

To evaluate compounds for their abilities to inhibit constitutive STAT3 activity, we used the breast cancer cell lines MDA-MB-468 known to express increased levels of pY-STAT3 and to depend on STAT3 for growth (53). In an initial experiment we treated the cells with 30 μM of either SPEC-29 and SPEC-8, the two most promising compounds (Table 1) for 0, 1,2, 4, 7, and 12 hrs; cells were lysed and levels of pY-STAT3, total STAT3 (tSTAT3) and GAPDH measured using Luminex. SPEC-29 was the most potent as it reduced the amount of pY-STAT3 to less than 30% within 1 hr of treatment and to ~5% within 4 hrs (Figure 3B). SPEC-8 was also potent, having reduced pY-STAT3 levels to ~20% of untreated levels by 7hrs (Figure 3B). Total STAT3 protein levels were unaffected by either compound even after 12 hrs of treatment (Figure 3C). In a second experiment, lysates were prepared from MDA-MB-468 cells treated with SPEC-29 for even shorter periods of time as well as longer periods of time up to 48 hrs and examined for relative levels of pY-STAT3, as well as pY-STAT1, pY-STAT5 and pY-STAT6. Similar to the previous results, pY-STAT3 levels were reduced to ~40% by 1hr and to ~15% by 4hrs and the level remained low over 48 hrs (Figure 3D). In contrast, longer incubation in SPEC-29 were required to reduce constitutive levels of pY-STAT1 (to ~40% in 48hrs). In contrast, the levels of pY-STAT5 and pY-STAT6, that were low to start with, were not affected by SPEC-29 treatment.

Since both the potent leads, SPEC-29 and SPEC-8, showed maximal inhibition by 7–12 hrs, we used a 16hr treatment time to treat MDA-MB-468 cells with increasing doses (0, 0.1, 0.3, 1, 3, 10, 30, and 100 μM) of 6 SPEC compounds, SPEC-29, SPEC-8, SPEC-93, SPEC-98, SPEC-57 and SPEC-85, to find the IC₅₀ for inhibition of constitutive pY-STAT3 in these cells. Similar to their effects on G-CSF-stimulated pY-STAT3 in Kasumi-1 cells, SPEC-29 and SPEC-8 showed similar low micromolar IC₅₀s for inhibition of constitutive pY-STAT3, as well (Figure 3E, Table 1). SPEC-93 also showed some activity (IC₅₀ = 44 μM), while the other three did not show any inhibitor activity. As expected, neither SPEC-29 nor SPEC-8 showed any appreciable activity against pY-STAT1 within this time point, i.e. 16hrs (Figure 3F), although, in the previous experiment, the compounds showed similar IC50s for G-CSF-stimulated pYSTAT1 and pYSTAT5 activities (Supplemental Figure 4). Together, these data indicate that both SPEC-29 and SPEC-8 are potent and selective STAT3 inhibitors that are able to inhibit both cytokine-stimulated, as well as constitutive pY-STAT3 activity.

Constitutive pYSTAT3 activity in MDA-MB-468 cells had been previously ascribed to aberrant Jak1/2 activation (66). We wanted to test whether the lead SPEC compounds inhibit constitutive pYSTAT3 in these cells by inhibiting the Jaks. In a repeat experiment (as in Figure 3D), lysates were prepared from MDA-MB-468 cells treated with SPEC-29 (30 μM) for increasing periods of time (from 15’ to 48hrs) and tested for pSTAT3 and tSTAT3 levels by Luminex (Supplemental Figure 5A). As before it showed marked reduction of pYSTAT3 levels to around 10% of the untreated levels by 1 hr and remained low over 48hrs. We used the same lysates to run western for pYJak1, total Jak1, pYJak1/2, total Jak2 (Supplemental Figure 5B) to enquire the probable upstream effects of the inhibitors. As indicated in Supplemental Figure 5B, we could not detect constitutive levels of pYJak1 and pyJak2 in these lysates, although an IL6-treated AcGFP-STAT3 control lysate showed specific bands for pYJak1. But did see a delayed decrease in levels of total Jak1 and Jak2 over time, that couldn’t possibly be responsible for a much rapid decrease in pYSTAT3 levels, as shown in Supplemental Figure 5A.

Evaluation of STAT-addicted cancer cell growth inhibition by SPEC compounds.

To evaluate compounds for their anti-cancer properties, we measured their ability to inhibit growth of two breast cancer lines, MDA-MB-468 and MDA-MB-231, known to express increased levels of pY-STAT3 (53) and to depend on STAT3 for their survival (53), drug resistance (54), and metastatic ability (55). SPEC-29, 8, and 93 (Table 1) potently inhibit MDA-MB-468 cell growth under anchorage dependent conditions (AD IC₅₀ = 2.4–12.1 μM; Figure 4A), as well as anchorage independent conditions (AI IC₅₀s = 1.8–11.1 μM; Figure 4B). Similar results were obtained for MDA-MB-231 (AD IC₅₀s = 3.0–12.2 μM; Figure 4C). Inhibition of anchorage independent growth of MDA-MB-231 was similar to that for MDA-MB-468, except that SPEC-8 showed an unexpectedly high IC₅₀ (26.1 μM; Figure 4D). The remaining five compounds showed less potency in inhibiting growth of these cell lines (AD and AI, IC₅₀ ~20–80 μM) with SPEC-85 showing no activity against either cell line. The ability of the eight compounds to inhibit growth of MDA-MB-468 cells, correlated positively with their abilities to inhibit G-CSF-stimulated pY-STAT3 levels (AD: Spearman r = 0.8333, p = 0.015, Figure 4E; AI: Spearman r = 0.8333, p = 0.015, Figure 4F), as well as their ability to inhibit growth of MDA-MB-231 cells (AD: Spearman r = 0.8571, p = 0.0107, Figure 4G; AI: Spearman r = 0.7807, p = 0.0315, Figure 4H). When we compared the constitutive pY-STAT3-inhibitory abilities of 6 out of 8 compounds in MDA-MB-468 cells, these corelated even more strongly with their abilities to inhibit anchorage independent growth of MDA-MB-468 cells (AI: Pearson r: 0.9395, p = 0.0054; Spearman r = 0.9411, p = 0.0167, Figure 4I). These results strongly suggest that the ability of these SPEC compounds to inhibit growth of pY-STAT3-dependent cells, depends on their ability to reduce levels of pY-STAT3 in these cells.

Figure 4. Inhibition of growth of pY-STAT3-high breast cancer cell lines cells by SPEC compounds.

MDA-MB-468 (A, B) or MDA-MB-231 (C, D) cells, were treated with compound (0, 0.1, 0.3, 1, 3, 10, and100 μM) in cell-culture-treated (48 hrs, A, C) or in ultra-low attachment 96-well plates (72hrs, B, D) and viable cells quantitated using MTT. Relative % viability (viability after any treatment ÷ viability of untreated cells x100) was plotted as a function of Log [M] compound, and IC₅₀ calculated using GraphPad. Data show representative experiments from ≥ 2 replicates. Mean IC₅₀ values are shown in Table 1. (E-I) Ability of SPEC compounds to inhibit growth of pY-STAT3-dependent breast cancer cell lines correlates to their ability to inhibit pY-STAT3. Linear regression lines are plotted for [IC50 of Growth inhibition of breast cancer cell lines MDA-MB-468 (E,F) or MDA-MB-231 (G,H)] as a function of [IC50 for inhibition of G-CSF-stimulated pY-STAT3 (E-H) or constitutive pY-STAT3 (I, 468 only)] and the Spearman (rank) correlation co-efficients and p values for each calculated using GraphPad Prism.

Evaluating STAT3-dependent growth inhibition by SPEC compounds.

In order to further assess the relationship between the STAT3-inhibitory properties of the compounds and their abilities to block cell-growth, we expressed WT-STAT3 in STAT3−/− MEFs and selected for cells stably expressing AcGFP1-tagged WT-STAT3 (AcGFP1-STAT3) or empty vector (AcGFP1-vector, Figure 5A). The two cells plated in cell-culture treated 96 well plates were serum starved for 24 hrs and then allowed to grow under reduced serum (0.5% FBS) DMEM medium for 72 hrs. Under these conditions, only the AcGFP1-STAT3 cells grew (Figure 5B), whereas the AcGFP1-vector-expressing cells failed to grow or grew very slowly. Treatment of AcGFP1-STAT3 cells with DMSO or increasing doses (0, 0.1, 0.3, 1, 3, 10, 30, and100 μM) with SPEC compounds (SPEC-29, SPEC-8, SPEC-93, SPEC-98, SPEC-57, SPEC-101 and SPEC-85) inhibited the STAT3-driven growth of these cells (Figure 5C, Table 1) with their growth inhibition IC₅₀ values correlating well with their ability to inhibit G-CSF-stimulated pY-STAT3 (AD: Pearson r = 0.9848, p < 0.0001; Spearman r = 0.9190, p = 0.0071, Figure 5D). We further showed that the AcGFP1-STAT3 cells could form more colonies than the AcGFP1-vector cells (Figure 5E) when plated in very low numbers (200 cell in 6 well plates) and that SPEC-29 with potent anti-STAT3 activity inhibited the ability of AcGFP1-STAT3 cells to form colonies more efficiently (no colonies at 0.3 μM) than SPEC-101 that did not have anti-STAT3 activity. This data provides further compelling evidence that the ability of SPEC compounds to inhibit STAT3-driven growth depends on their ability to inhibit STAT3 activity.

Figure 5. SPEC compounds inhibit STAT3-dependent growth of STAT3^−/− MEF cells expressing ACGFP1-STAT3.

GAPDH-normalized tSTAT3 levels in lysates from STAT3^−/− MEF cells stably transduced with either WT-STAT3 (AcGFP1-STAT3) or empty vector (AcGFP1-vector) are plotted (A). AcGFP1-Vector and AcGFP1-STAT3 cells were plated overnight in cell culture treated 96 well plates, serum starved for 24 hrs, and cultured for 72 hrs in DMEM with 0.5% FBS. Viable cells were quantitated using MTT. Relative cell growth, calculated by dividing the OD values after 72 hrs by the OD values after 24 hrs starvation (0 day) are plotted (B). AcGFP1-Vector and AcGFP1-STAT3 cells, after 24hrs serum-starvation, were cultured for 72 hrs in DMEM with 0.5% FBS ± increasing doses (0, 0.1, 0.3, 1, 3, 10, and 100 μM) of SPEC compounds (SPEC-29, 8, 93, 98, 57, 101 and 85); viable cells quantitated using MTT, relative % viability (viability after any treatment ÷ viability of untreated cells x100) was plotted as a function of Log [M] compound, and IC₅₀ values calculated using GraphPad. Data show representative IC₅₀ curves.

To test the specificity of action of the two lead compounds, further, we cultured the two cells AcGFP1-vector and AcGFP1-STAT3 under conditions (increased number of initially plated cells) when even the vector cells grew (albeit slower than the STAT3-overexpressing clones) and tested the ability of SPEC8 and SPEC-29 to inhibit growth of the two cells. We also included STAT3−/− MEFs overexpressing three known gain-of-function STAT3 mutants, Y640F, D661Y and K658Y (AcGFP1-Y640F, AcGFP1-K658Y, AcGFP1-D661Y, Supplemental Figure 6A). As is evident (Supplemental Figure 6B, C, Supplemental Table 3) that both SPEC-8 and SPEC-29 inhibited the cells overexpressing WT or GOF mutant STAT3 with lower IC50s (~3 fold) as compared to the Vector-expressing cells devoid of STAT3, once again reiterating the preference of these compounds to target STAT3-driven cell survival pathways.

DISCUSSION

We report the first successful use of the structure of the SH2 domain of STAT3 derived from molecular dynamic simulation for SB-VLS of a chemical library to identify small-molecule STAT3 inhibitors. The structure used was the averaged structure from the last 2 ns MD simulation of the STAT3 SH2 domain bound to a high-affinity peptidomimetic ligand (CJ-887); the library screened was the SPECS screening compounds library. The virtual ligand docking portion of the screen yielded 110 compounds as potential STAT3 inhibitors. Five of these 110 compounds (SPEC-29, SPEC-8, SPEC-93, SPEC-98, and SPEC-106) potently inhibited G-CSF-stimulated pY-STAT3 activation (IC₅₀ ranging from 2.7–19.0 μM, Table 1). Three of these five compounds (SPEC-29, SPEC-8, and SPEC-93) also were efficient inhibitors of constitutive pY-STAT3 and were the most potent in inhibiting pY-STAT3-driven, anchorage-dependent growth of breast cancer lines (IC₅₀ ranging from 2.4–12.2 μM); inhibition of breast cancer cell growth by these compounds correlated with their ability to inhibit G-CSF-stimulated pY-STAT3 activity and their ability to inhibit constitutive pY-STAT3. The latter also correlated with their ability to block STAT3-dependent growth of STAT3^−/− MEF cells expressing AcGFP1-STAT3. These results, combined with their drug-like properties, indicate that these hit compounds, especially SPEC-29 and 8, deserve consideration for further development as small-molecule STAT3 inhibitor drugs to treat cancer, chronic inflammation, and fibrosis.

The drug-like features of the hits (2D chemical structures shown in Supplemental Figure 2) was assessed using Lipinski’s four “rules-of-five” (67). Most hit compounds comply with three or more of the four rules (Supplemental Table 4), with the four compounds (SPEC-29, 8, 93, and 98) fulfilling all 4 rules.

The ranking of compounds based on docking results using crystal structure as receptor and standard precision parameter set (Ranking SP1), as well as docking using the averaged structure from MD simulation as the receptor (Ranking SP2), are shown in Supplemental Table 5. It is clearly evident that the binding site flexibility remarkably affects the ranking of compounds in virtual screening. Most of the hits were poorly scored or ranked in the docking experiment based on crystal structure, e.g. SPEC-85 was ranked among the top 30% (33,000) from crystal structure HTVS screening (HTVS1) and ordered as 5201 in the SP1 screening. The two others that were ranked high in SP1 were SPEC-93 (954) and SPEC-106 (1007). The other compounds were not ranked within the top 10,000, in the crystal structure SP screening. However, in the two-step averaged structure screening (SP2), six compounds (SPEC-85, 57, 8, 93, 98 and 29) were ranked within top 1,000 while two (SPEC-106 and 101) ranked within top 5,000.

The binding pattern analysis of the docking experiments using the averaged structure derived from MD simulation revealed that, although the 2D chemical structures (Supplemental Figure 2) are diverse for the hits, they occupied similar binding sites on SH2 domain (Supplemental Figure 3). Screening criterion (described in methods and Figure 1) ensured that all hits exhibited hydrogen-bonding interactions with S611 and S613 in the pY+0 pocket area (Supplemental Figure 3, Supplemental Table 6) although each had different chemical moieties. Phosphate or phosphorylated phenol group occupied the pY+0 pocket in the poses of the crystal structure of SH2 dimer (Figure 2, Supplemental Table 6) and the docking model of CJ-887 (Figure 2C, 3, Supplemental Table 6), respectively. Most of the known STAT3 inhibitors also harbor negatively charged moieties to mimic pY-peptide interactions in this area (18, 33). Interestingly, the negatively charged carboxyl group in SPEC-85 forms hydrogen-bonding interactions in pY+0 pocket; while for the other hits, neutral charge groups are located in this area (Supplemental Table 6). Thus, most hits (except SPEC-85) are electrically neutral at physiological pH suggesting that charged groups are not necessary to bind the pY+0 pocket of STAT3 in our model. This is a very novel finding and prompts one to speculate whether many potentially strong inhibitors might have been overlooked in earlier studies, based on the classical modeling using static STAT3 crystal structures (2). Due to their inherent bias for a negative charge at the pY+0 pocket, these studies might have, in essence, looked for only charged compounds, unlike the hits we found, which have polar atoms, instead of negatively charged groups, forming hydrogen-bonding interactions. This is especially important, as charged molecules such as CJ-887 suffer from poor cell membrane permeability and, thus, are far less suitable for clinical development. In fact, this has been a main reason that inhibitors targeting the SH2 domains of many other targets e.g. Src kinase, the Src-family kinase Lck, p85-the regulatory subunit of PI3K and Grb2 have also been generally unsuccessful (68). The phosphotyrosine (pY) residue was estimated to provide one half of the binding energy of binding of phosphopeptides to the SH2 domain (69–71) and hence considered an absolute necessity. At one point in time, therefore, the idea of targeting a SH2 domain was virtually abandoned (68). The neutral compounds identified in this study (e.g. SPEC-29, SPEC-8), are thus, good candidates for STAT3 hit-to-lead drug development and a similar strategy might be successful in designing inhibitors targeting SH2 domains within other oncogenic targets as well.

One of the reasons for large polar groups not being able to bind to the pY+0 pocket might be the relatively narrow size of the pY+0 pocket in the crystal structure, which is incapable of accommodating large groups with potential to form hydrogen-bonds with S611 and S613 (Supplemental Figure 1D). For instance, SPEC-8, and SPEC-98 are not able to dock into the crystal structure due to the limited space, whereas both of the compounds were reordered within the top 1,000 list based on the average structure (Supplemental Table 5). The relatively large chemical moieties (the trienone group within SPEC-8 and the thiazolidine group within SPEC-98) are positioned at the bottom of pY+0 pocket according to the docking poses derived from the STAT3 averaged structure (Supplemental Figure 3C & D, Supplemental Table 6). It seems that the movement of αA helix and side chain of K591 in the induced-fit model resulted in a larger pY+0 pocket, and hence better ranking for compounds harboring larger chemical moieties (Supplemental Figure 1). The strategy of incorporating the MD simulation to accommodate the SH2 domain flexibility, thus, tends to uncover new classes of compounds not identified previously.

In addition to the hydrogen bonding at the pY+0 pocket, another common feature of hit compounds is the hydrophobic interactions between the aromatic or hydrophobic groups in the compounds and the flat “wall” formed by side chain and backbone atoms of V637, E638 and P639, which are also observed in the docking pose of CJ-887 (Figure 2). For SPEC-85 and SPEC-57, a hydrogen bond was predicted to form between the carbonyl group within compounds and the amide group from backbone of E638 (Supplemental Figure 3A, B). This hydrogen bond is also observed in the docking pose of CJ-887 and maintained during the MD simulation. The high IC₅₀ for SPEC-85 and SPEC-57 in assays examining inhibition of G-CSF-stimulated pY-STAT3 supports the conclusion that the contribution of this interaction might not be as important as the pY-interaction.

SPEC-8 is the smallest inhibitor and has the best calculated ligand-efficiency-of-binding-energy per atom (72). It therefore would appear to contain the minimum set of interactions necessary for potency—the hydrogen-bonding network in the pY+0 pocket and the hydrophobic interactions with the flat wall (Supplemental Figure 3C). The IC₅₀ for inhibition of G-CSF-stimulated pY-STAT3 of SPEC-8 also is one of the lowest (4.0 μM). SPEC-29 is the hit with second-lowest molecular weight. The binding pose of SPEC-29 is typical of the group (Supplemental Figure 3F), with its acetamide group forming hydrogen bonds in the pY+0 pocket, its phenyl ring forming hydrophobic interactions with E638 within the flat wall, and its 1-bromo-4-methoxybenzene group locating near T714 and making contact with the hydrophobic pY+1 pocket. The IC₅₀ for inhibition of G-CSF-stimulated pY-STAT3 of SPEC-29 is the lowest of the hits identified (2.7 μM); its ability to form hydrogen bonds with S611 and S613, to form hydrophobic contacts with the pY+0 and pY+1 pockets, and its relatively low rotatable bonds may contribute to this high potency.

Since both of the two lead compounds SPEC 8 and SPEC 29 are α,β-unsaturated carbonyl containing compounds (Supplemental Figure 2) they may act as Michael acceptors towards thiol groups of cysteine containing proteins. Thus, we sought to rule out alkylation as the main mechanism of action of SPEC-29 and SPEC-8. To this end, we compared the reactivity of SPEC-8 and SPEC-29 towards reduced glutathione (GSH) to a well-studied STAT3 inhibitor; Stattic, whose mechanism of action has been attributed to its ability to alkylate STAT3 at key residues (73–75). 100 μM of Stattic, SPEC-8 and SPEC-29 were incubated with a 10-fold excess of GSH, the reaction was then monitored by HPLC UV to determine area under the curve (AUC) at five minutes intervals. GSH reaction times with Stattic were observed to be >10-fold faster, compared to SPEC-8, and SPEC29. (Supplemental Figure 7A). Further, even after 50 minutes of GSH-exposure, SPEC-8 was depleted to ~60% of the original amount, while SPEC-29 was at ~30%, while the loss of Stattic was both rapid and complete (~4% of original by 20’).

In a more definitive experiment, recombinant, core STAT3 protein was mixed with 50-fold molar excess of the three STAT3 inhibitors in the presence of the reducing agent TCEP (tris(2-carboxyethyl)phosphine) and alkylated with excess iodoacetamide 15 mM, prior to digestion with trypsin and analyzed by LC-MS/MS, using a quadrupole-linear ion trap MS (Sciex QTrap 5500), for the presence of STAT3 cysteine adducts of Stattic, SPEC-8 and SPEC-29. As expected, based on the presence of predicted MRM (multiple reaction monitoring) signal for possible STAT3 peptides containing cysteine residues, we detected Stattic adducts, on eight of the 11-cysteine residues (data not shown). SPEC-29 showed a very weak signal for six of the 11 cysteine (data not shown) however only cysteine 712 could be identified and confirmed unambiguously by performing a full MS/MS spectra on the detected transitions. SPEC-8 did not show any adducts at similar conditions.

Because of inherent physiochemical properties, different adductions could have varied responses in a mass spectrometer. As such, direct comparison of multiple reaction monitoring (MRM) signal may not be valid. To address this issue, we sought to evaluate the extent to which SPEC-8 and SPEC-29 react with sulfhydryl groups on STAT3 proteins compared to Stattic, using a different approach. After incubation of STAT3 protein with inhibitors for 12hrs at 37°C at a molar ratio protein: inhibitor (1:50) followed by iodoacetamide treatment for 1hr. We used LC-MS to access the efficiency of alkylation of a STAT3 peptide most amenable to alkylation, viz FICVTPFIDAVWK (C712), by quantifying the amount of peptide labeled with Cysteine Carbamidomethylation (CAM) by Iodoacetamide. We reasoned that efficient alkylation would block all the sites available for reaction with excess Iodoacetamide. Thus, reduced MRM signal for CAM peptide would denote efficient prior alkylation by the compound. After normalization to an internal cysteine free peptide (GLSIEQLTTLAEK), we show (Supplemental Figure B-E and F) that Stattic markedly reduces MRM of C721 whereas SPEC-8 and SPEC-29 do not. Thus, the above data proves that though the reactive olefinic double bonds in SPEC-8 and SPEC-29 do provide to these compounds, some ability to form adducts with STAT3 cysteines, the extent and speed of these reactions are extremely low and unlikely to be the major mechanism through which they bind and inhibit STAT3.

CONCLUSION

SB-VLS, using an averaged structure from molecular dynamics (MD) simulations of STAT3 SH2 domain in a complex with CJ-887, a known peptidomimetic binder, identify two highly potent, neutral, low-molecular weight STAT3-inhibitors with favorable drug-like properties.

IMPLICATIONS.

SB-VLS, using an averaged structure from molecular dynamics (MD) simulations of STAT3 SH2 domain in a complex with CJ-887, a known peptidomimetic binder, identify two highly potent, neutral, low-molecular weight STAT3-inhibitors with favorable drug-like properties.