Novel inhibitors of a Grb2 SH3C domain interaction identified by a virtual screen

Abstract

The adaptor protein Grb2 links cell-surface receptors, such as Her2, to the multisite docking proteins Gab1 and 2, leading to cell growth and proliferation in breast and other cancers. Gab2 interacts with the C-terminal SH3 domain (SH3C) of Grb2 through atypical RxxK motifs within polyproline II or 3₁₀ helices. A virtual screen was conducted for putative binders of the Grb2 SH3C domain. Of the top hits, 34 were validated experimentally by surface plasmon resonance spectroscopy and isothermal titration calorimetry. A subset of these molecules was found to inhibit the Grb2–Gab2 interaction in a competition assay, with moderate to low affinities (5: IC₅₀ 320 µM). The most promising binders were based on a dihydro-s-triazine scaffold, and are the first small molecules reported to target the Grb2 SH3C protein-interaction surface.

1. Introduction

Growth factor receptor-bound protein 2 (Grb2) is a small adaptor protein found in humans and other organisms. Structurally, Grb2 consists of three protein recognition domains: a central SH2 domain flanked by two SH3 domains.¹ Although Grb2 lacks enzymatic activity, it is a component of a wide range of cellular signaling pathways, often as an intermediary between receptor tyrosine kinases and intracellular processes.^{2, 3} Mounting evidence points to a strong role for Grb2 in developing malignancies. In particular, Grb2 is overexpressed in a range of human breast cancer cell lines⁴ and tissue samples.⁵ In breast cancer, Grb2 lies downstream of the EGF family protein Her2 (ErbB2, Neu). Overexpression of Her2 occurs in c. 20% of breast cancers, and is associated with poor survival prognosis.⁶ Grb2 docks to specific pTyr residues in Her2 via its SH2 domain,⁷ then recruits Sos through both SH3 domains,⁸ or the Gab family of large multi-site docking proteins via its C-terminal SH3 domain (SH3C).^{9, 10}

Three therapies targeted towards Her2 have been approved for the clinic. Treatment with trastuzumab, a monoclonal antibody targeting the extracellular domain of Her2, improves survival for patients with Her2 positive cancers. While large clinical trials have revealed benefits from trastuzumab therapy in metastatic cancer¹¹ and as an adjuvant therapy,¹² patients rapidly develop resistance.¹³ Lapatinib, a small-molecule inhibitor of the Her2 tyrosine kinase domain, has also shown clinical benefits in patients with metastatic Her2 positive cancers, but most patients did not respond to therapy and disease progression continued after a short time.¹⁴ More recently, the anti-Her2 antibody pertuzumab has shown some success in combination therapy with trastuzumab,¹⁵ increasing the median progression-free survival time for metastatic disease from 12- to 18-months. While this trial strongly suggests a role for combination therapies in treating Her2 positive cancers, there remains a significant need for novel therapies.

One approach is to target the Her2 pathway downstream of the receptor.¹⁶ Cell-based studies have shown that down-regulation of Grb2 inhibits growth in tumors expressing high levels of Her2.¹⁷ Likewise, knockdown of Gab2 reduces proliferation of Her2-overexpressing cells.¹⁸ Further in vivo studies in mice have identified a peptide inhibitor of Grb2 SH3 domains which may slightly reduce cancer progression in a xenograft model.¹⁹ While inhibitors have been developed for the SH2 and SH3N domains of Grb2,²⁰ small-molecule inhibitors of the SH3C domain have not been reported.

The Grb2 SH3C domain binds an unusual RxxK consensus sequence,^{21, 22} which is distinct from the more common PxxP sequence motif required for binding to many other SH3 domains. A structural study has elucidated the precise molecular interactions involved in Grb2–Gab2 binding.²³ Two distinct Grb2 SH3C binding sites were found in Gab2, each with an RxxK core motif within a variable proline-rich sequence. The Gab2a epitope presents the Arg and Lys residues within a polyproline II helix conformation, while in the Gab2b epitope they are found within a 3₁₀ helix. In both cases, essential hydrogen bonds are formed to Glu13 and Glu16 on the Grb2 protein. An attractive feature of Grb2 SH3C as a drug target is thus its requirement for a mixed positively charged and hydrophobic ligand, and so we investigated, in a first approach, whether any existing drug-like molecules possess such properties and thus potential inhibitory activity.

2. Results and discussion

To search for small-molecule inhibitors of the Grb2–Gab2 protein-protein interaction, we conducted a virtual screen using the docking program GLIDE (version 5.6).²⁴ Our crystal structure (PDB code: 2VWF),²³ a complex containing the more-tightly binding Gab2b 3₁₀ helix, was used as the docking target. The protein was processed using GLIDE’s default protein preparation workflow. The bound peptide and all water molecules were removed, and the protein structure refined using exhaustive sampling for H-bonding networks. Following a constrained Impref minimization with the OPLS-AA 2005 forcefield,²⁵ optimizing all atoms to an RMSD of 0.3 Å, the receptor grid was generated as a cube 7 Å in radius around Trp35 in the center of the binding pocket. A hydrogen-bonding constraint was applied requiring all docked molecules to form at least one hydrogen bond to the Glu16 sidechain.

A library of c. 6.3 million small molecules (MW < 500 Da) was obtained from the Zinc database and used without further refinement.²⁶ Protonation states of molecules in the database were those predicted to occur in the range of pH 5.75–8.25. The molecules were flexibly docked onto the target protein, and the binding interactions evaluated using the GlideSP scoring function. The resulting report ranked each compound by glidescore, which roughly approximates their free energies of binding (Table S1).

The top 200 virtual hits were screened by eye for binding pose and chemical diversity (Figure 1 and Supporting Information). Of these, 34 molecules were obtained for in vitro testing from the NCI/DTP Open Chemical Repository (Figure 2 and Table S2). Human wild-type Grb2 SH3C protein was prepared as previously described.²³ The in vitro binding of these molecules to the Grb2 SH3C domain was evaluated at 50 and 100 µM concentrations using surface plasmon resonance (SPR).

Figure 1.

Docked poses of the molecules ranked 5 (top) and 164 (bottom) in the virtual screen bound to the Grb2 SH3C domain (purple). Both molecules are predicted to make extensive hydrogen-bond and hydrophobic contacts to the protein.

Figure 2.

Structures of the 34 compounds obtained from the NCI/DTP Open Chemical Repository.

To distinguish binders from non-binders, we defined a practical minimum binding affinity to have a K_D of ≤ 1 mM. The average molecular weight of the molecules being screened was 250, and from the relationship:

• Rmax_ligand = (MW_ligand) / (MW_analyte)×Rmax_analyte

where ‘analyte’ refers to the small molecule and ‘ligand’ the SH3C domain, the maximum theoretical response, assuming 1:1 binding, is 56 response units (RU). So the minimum desired response level (Req) could be determined using the formula:

• Req = (Rmax_ligand×C_analyte) / (K_D + C_analyte)

where C_analyte represents the concentration of compound, giving a minimum cut-off of 5 RU for the 100 µM concentration.

Using this threshold, 24 molecules were classified as non-binders (17 with a response level around zero) and ten as binders: 1, 3, 5, 9, 41, 46, 72, 147, 156, 164. Varying levels and quality of interactions were observed by the analysis of their sensorgrams, nine of which are shown in Figure 3 (compound 156 was rejected due to poor sensorgram shape). In several cases (molecules 5, 9, 147, 164) the observed binding behavior was superstoichiometric. Compound 147, for instance, displayed greatly superstoichiometric binding, characteristic of undesirable promiscuous interactions,²⁷ with a response level of nearly 600 RU (i.e. over 10 times the maximum theoretical response) and a long dissociation time. This can be explained by the availability of multiple binding sites on the SH3C domain and the presence of small-molecule aggregates that have a propensity to adhere to the protein surface. Injections of DMSO-only controls at regular intervals as well as a positive control Gab2b peptide (24 µM; sequence: IQPPPVNRNLKPDRK-amide) indicated that the integrity of the coupled Grb2 SH3C surface was maintained throughout the experiment (Figure S3).

Figure 3.

SPR sensorgrams of the 9 compounds accepted as positive Grb2 SH3C binders (see Figure S3). Shown are two traces at different compound concentrations: 50 µM (grey line) and 100 µM (black line).

To determine which compounds could competitively target the Gab binding groove in Grb2 SH3C, we performed a biochemical competitive binding assay on the full set of 34 molecules. GST–Grb2 SH3C was incubated with each molecule before contact with streptavidin beads coupled to a biotinylated Gab2b peptide. GST alone did not couple to the peptide-bound beads. Neither did GST nor GST–Grb2 SH3C bind to beads lacking peptide, nor to beads coupled to a Gab2 double alanine mutant peptide (RxxK → AxxA) that cannot bind to the Grb2 SH3C domain²³ (Figure S4). Thus, interaction with the beads was a direct result of the Gab2 peptide – Grb2 SH3C coupling.

Given this assay’s lower sensitivity, a stringent threshold of 20 % minimum inhibition was applied to eliminate ambiguous-, weak-, or non-binders. This represents a margin greater than the error on any individual sample, and five times the average error on the data points (Figure 4A). Consequently, 24 molecules showed no binding (Figure 4B, grey bars), and ten molecules: 3, 5, 41, 46, 50, 55, 72, 139, 147, 164, exhibited various levels of competitive inhibition > 20 % (Figure 4B, black bars). Importantly, seven of these competitive inhibitors overlapped with those determined to bind well by SPR: 3, 5, 41, 46, 72, 147, 164. However, three of these molecules: 50, 72, 164, displaying nearly complete inhibition precipitated during the assay. This made them problematic to handle and may have contributed to their apparent superior inhibitory behavior (compound 72 was the least soluble); molecule 46 also displayed poor aqueous solubility, and so these four compounds were rejected. The superstoichiometric binder from SPR analysis – compound 147 – was excluded too, leaving a subset of three compounds (3, 5, 41) for further assessment. We thus sought to measure the affinity of these compounds for the Grb2 SH3C surface. Above concentrations of 200 µM, these compounds displayed superstoichiometric binding using the SPR experiment, preventing the collection of meaningful affinity data. Thus, the biochemical competition assay was employed using increasing compound concentrations, with the aim of assessing only the competitive-binding component (Figure 5).

Figure 4.

Grb2 SH3C competitive binding screen with 34 hit molecules using a biochemical competition assay. Purified GST–Grb2 SH3C was incubated with each compound (5 mM) in duplicate before contact with streptavidin-sepharose beads previously coupled to biotinylated Gab2b peptide (Bio-EAHK-IQPPPVNRNLKPDRK-amide). Washed beads were run on a 12% SDS-polyacrylamide gel and the extent of binding visualized by SDS-PAGE and Coomassie blue dye staining (A), and the band intensities quantified by densitometry (B). In (A), ‘no’ refers to no compound i.e. DMSO-only control. Numbering indicates the compound used, and relates to the rank of each in the virtual screen. Unbiotinylated Gab2b peptide was used as a positive binding control for inhibition (+ve) at two concentrations. In (B), chart bars are shaded as follows: the ten positive binders showing ≥ 20% inhibition (black); non-binders (grey); controls, or 10 µg of GST–Grb2 SH3C input (light grey). Error bars indicate the spread of the data.

Figure 5.

Concentration-response curves determined by biochemical competition assay for two competitively binding compounds. Increasing concentrations of (A) compound 3, and (B) compound 5, were incubated with GST–Grb2 SH3C before mixing with streptavidin beads coupled to a biotinylated Gab2b peptide. (A) and (B) lower panels: Coomassie bluedyed bands of the GST–Grb2 SH3C/beads samples run on a 12% SDS polyacrylamide gel; upper panels: densitometric quantification and curve-fit of the protein band intensities. The calculated IC₅₀ values and Hill slope parameters are shown. Error bars show standard deviations; the gels are representative of 4 to 6 repeats.

Compound 41 demonstrated unquantifiably weak inhibition in the limited dose range used (up to 20 mM; data not shown). Compound 3 yielded a concentration-response (inhibition) curve with an apparent IC₅₀ of 5.7 mM, indicating low affinity (Figure 5A). Despite its structural similarity to compound 3, compound 5 bound competitively to Grb2 SH3C with an apparent IC₅₀ of 320 µM, representing an 18-fold increase (Figure 5B). Notwithstanding this positive competitive inhibitory effect, an important feature of the inhibition curves for both compounds 3 and 5 is their steepness (high Hill slope values of −3.1 and −5.0 respectively; Figure 5). This indicates marked non-stoichiometric binding, as near complete inhibition takes place over one log unit rather than two log units as expected from simple 1:1 binding governed by the law of mass action. Molecules displaying such binding characteristics are less ideal but by no means purely artifactual.²⁸

To verify further the different behaviors observed between compounds 3 and 5, isothermal titration calorimetry (ITC) was performed. Consistent with the low affinity calculated using the biochemical assay, no measurable response above background was detected from compound 3 (Figure S5A). In contrast, compound 5 revealed significant reaction heats, resulting in a more complex behavior characterized by a multiphasic isotherm (Figure S5B). No suitable fitting function could be applied to describe this complex binding, though it might be explained by multiple binding sites or a conformational change of the protein.

While a proportion of the binding observed may be a result of multiple interaction sites on Grb2 SH3C, the biochemical competition assay confirms two molecules (compounds 3 and 5) that interact competitively with the Gab2b peptide-binding surface of Grb2 SH3C. Their related structures provide some structure-activity relationship. In fact, among the 34 molecules, seven, 1, 3, 5, 9, 11, 41, 47, featured a common dihydro-s-triazine motif which has previously been investigated for the inhibition of dihydrofolic reductase²⁹ and voltage-gated sodium channels.³⁰ These seven molecules span several orders of magnitude in binding affinity, ranging from 320 µM to undetectably low levels. Consistent with previous peptide array studies,²³ only mixed cationic and hydrophobic molecules bound to Grb2 SH3C whereas molecules 11 and 47, which lack significant hydrophobic motifs, showed no activity in either assay. Compound 9, which differs from 3 in having a hydrophilic polyethylene glycol linker in place of a hydrophobic alkyl chain, similarly shows no activity. The most strongly binding compounds 3 and 5 have linkers of equal length, though the more rigid alkene linker for compound 5 may contribute to its improved binding affinity by lessening the entropic penalty of binding in an extended conformation.

To confirm the nature of the active inhibitory species, the purity and identities of the two most promising compounds (3, 5) were verified using high-resolution mass spectrometry and ¹H and ¹³C NMR spectroscopy (Supporting Information). Their protonation states were further assessed using the composite CBS-QB3 quantum chemical method.³¹ Based on the computed heats of formation (Figure S2), the compounds, which exist as the mono-hydrochloride salt, are protonated at position-5 of the ring.

3. Conclusions

This is the first computational screening reported for the identification of small molecules able to dock to the Grb2 SH3C domain surface. Given the challenging nature of targeting protein-protein interaction surfaces, we were encouraged that a virtual screen allowed identification of small molecule in vitro binders of Grb2 SH3C. Despite the modest number of compounds acquired for in vitro testing, it is remarkable that over 20 % of these displayed inhibitory activity in two orthogonal screens. Of these active compounds, all but two are cationic, and the apparent activity of the other two is likely an artifact due to insolubility. A cationic molecule, while apparently necessary for binding, is not sufficient, as many non-binders were also cations.

It remains to be explored whether the potency of these molecules can be further developed, potentially allowing cell-based studies. The modest affinities achieved from these commercially available molecules (cf. 15-mer Gab2 peptide, K_d 3.2 µM towards Grb2 SH3C) and their likely multi-point interaction on the SH3 surface suggests that a more bottom-up approach could be beneficial for producing effective inhibitors against Grb2 SH3C. Wider screening may probe chemical space that is more favorable for protein-protein interaction inhibitors, though delineating that space is an ongoing challenge. Alternatively, the rational design of peptidomimetics that match the native biological partners Gab1 and 2 might provide a better strategy to generate more potent and specific inhibitors to this important new protein target in breast and other cancers.

4. Experimental Details

4.1. Surface plasmon resonance

SPR was performed using a Biacore T200 optical biosensor. A fresh CM5 chip was docked to the instrument and hydrated with duplicate 12 s injections of 50 mM NaOH, 10 mM Glycine•HCl pH 1.5, and 0.1 % SDS at 50 µL.min⁻¹ flow rate. The flow cells were then normalized with 70 % glycerol using the default normalization wizard. The chip was primed into HBS-N running buffer (10 mM Hepes pH 7.4, 150 mM NaCl), and the flow cells activated with a 1:1 mixture of 0.4 M EDC and 0.1 M NHS for 10 minutes at 10 µL.min⁻¹ and 25 °C. Untagged Grb2 SH3C (100 µM in pH 4.0 acetate buffer) was coupled to the surface for 8 minutes to an immobilization level of 1500 RU. After coupling, all flow cell surfaces were deactivated with 4×30 s injections of a 1:1 mixture of 1.0 M ethanolamine and running buffer. The chip was subsequently primed twice into the experimental running buffer (HBS-N, 5.0 % DMSO, 0.01 % Tween 20).

Each molecule was dissolved in running buffer (HBS-N, 5.0 % DMSO, 0.01 % Tween 20) at 50 µM and 100 µM concentrations and then screened using 120 s contact time at 40 µL.min⁻¹ flow rate and 25 °C, with a data collection rate of 10 Hz. The data were analyzed using standard double referencing procedures by subtracting the response from the protein-free cell and the response from a blank injection of running buffer from the response from the SH3C flow cell.³²

4.2. Biochemical competition assay

GST–Grb2 SH3C (10 µg per sample) was pre-incubated with each inhibitor (5 mM, initially) in duplicate for 2 hours at 4 °C in assay buffer (25 mM Tris, pH 7.5, 150 mM NaCl, 1 mM DTT, 0.05 % Tween 20, 5 % DMSO). Biotinylated Gab2b peptide (binding affinity 3.2 µM for Grb2) was synthesized as previously described.²³ 0.3 nmol of peptide per µL of packed streptavidin-sepharose beads (GE Healthcare) were incubated together for 30 minutes in assay buffer before extensive washing of the beads to remove any unbound peptide. Next, 8 µL of peptide-coated beads were transferred to each assay tube containing GST–Grb2 SH3C and compound for 30 minutes, the solution clarified by centrifugation (2000×g for 1 minute) and the beads washed three times with PBS-T (phosphate-buffered saline, 0.05 % Tween 20). The samples were analyzed by SDS-PAGE and visualized after staining the protein bands with Coomassie Blue dye (Figure 3A). As controls, parallel samples were run in the absence of inhibitor, and with preincubation of 10 or 100×molar excess of free, unbiotinylated Gab2b peptide. The band intensities, indicating the levels of GST–Grb2 SH3C binding to the beads, were further quantified by densitometry and scaled relative to the control sample without inhibitor (100 % level; Figure 3B). This normalization allowed the cross-comparison of multiple gels.

For concentration-response experiments, increasing concentrations of compound were used. The data were fitted to a sigmoidal inhibition model using nonlinear regression analysis in GraphPad Prism software.

4.3. Isothermal titration calorimetry

ITC was performed with a VP-ITC microcalorimeter (MicroCal). Purified Grb2 SH3C (20 µM) was diluted into ITC buffer (50 mM Hepes, pH 7.5, 50 mM NaCl, 0.5 % DMSO) at 25 °C, and then clarified and degassed (ThermoVac; MicroCal), before being placed into the sample chamber holding 1.43 mL. Compound (1 mM in ITC buffer) was titrated from a syringe (300 µL total volume) into the sample chamber with one injection of 4 µL followed by 18 injections of 15 µL. The resulting peaks of measured heat change from the equilibrium temperature were integrated to yield the quantity of heat generated per mole of compound (kcal.mol⁻¹). This was then plotted against the molar ratio of compound to protein using the manufacturer’s Origin 7 software.

4.4. Spectroscopic analysis of compounds 3 and 5

NMR spectra were recorded on a Bruker AVC500 (¹H: 500 MHz; ¹³C: 125 MHz). Chemical shifts are quoted relative to the residual solvent peak. ¹H spectra are reported as follows: chemical shift δ/ppm (number of protons, multiplicity, coupling constant J / Hz [where appropriate]). Multiplicity is abbreviated as follows: s, singlet; br, broad; d, doublet; t, triplet; quint, quintet; m, multiplet. Coupling constants J are reported to the nearest 0.1 Hz as observed. ¹³C spectra are reported in δ/ppm. High-resolution mass spectra were recorded on a Bruker MicroTof mass spectrometer under conditions of electrospray ionization (ESI) by the internal service at the Department of Chemistry, University of Oxford. Values reported are the ratio of mass to charge in Daltons.

4.4.1. 1-(3-(4-(3-Aminophenyl)butyl)phenyl)-6,6-dime thyl-1,6-dihydro-1,3,5-triazine-2,4-diamine•HCl (3)

δ_H (500 MHz, (CD₃)₂SO) 9.00 (1H, br s), 7.64 (1H, br s), 7.44 (1H, t, J 8.0), 7.35 (2H, br s), 7.34 (1H, d, J 7.8), 7.19-7.15 (2H, m), 6.89 (1H, t, J 7.6), 6.38-6.34 (2H, m), 6.31 (1H, d, J 7.4), 6.25 (1H, br s), 4.91 (2H, s), 2.66 (2H, t, J 7.3), 2.44 (2H, t, J 8.0), 1.66-1.51 (4H, m), 1.10 (6H, d, J 2.1); δ_C (125 MHz, (CD₃)₂SO) 157.6, 157.2, 148.5, 144.5, 144.5, 142.6, 134.7, 129.9, 129.7, 128.6, 127.1, 115.9, 113.9, 111.5, 69.6, 35.2, 34.5, 30.3, 30.3, 27.3, 27.2; HRMS (ESI) found 365.2442, C₂₁H₂₉N₆⁺ [M+H]⁺ requires 365.2448.

4.4.2. 6,6-Dimethyl-1-(3-((1E,3E)-4-(3-nitrophenyl)buta-1,3-dien-1-yl)phenyl)-1,6-dihydro-1,3,5-triazine-2,4-diamine•HCl (5)

δ_H (500 MHz, (CD₃)₂SO) 9.04 (1H, s), 8.35 (1H, t, J 1.8), 8.10 (1H, dd, J 8.2, 0.6), 8.01 (1H, d, J 7.9), 7.71-7.64 (2H, m), 7.57 (1H, s), 7.53 (1H, t, J 7.8), 7.39-7.32 (1H, m), 7.35 (4H, br s), 7.31-7.24 (2H, m), 6.94 (1H, d, J 5.4), 6.90 (1H, d, J 5.4), 6.42 (1H, br s), 1.39 (6H, d, J 7.5); δ_C (125 MHz, (CD₃)₂SO) 157.6, 157.2, 148.4, 138.9, 138.8, 135.4, 133.4, 132.5, 132.1, 131.0, 130.5, 130.3, 130.2, 129.2, 127.8, 127.8, 122.1, 120.6, 69.6, 27.3, 27.3; HRMS (ESI) found 391.1872, C₂₁H₂₃N₆O₂⁺[M+H]⁺ requires 391.1877.