A Systematic Approach to the Discovery of Protein–Protein Interaction Stabilizers

Dysregulation of protein–protein interactions (PPIs) commonly leads to disease. PPI stabilization has only recently been systematically explored for drug discovery despite being a powerful approach to selectively target intrinsically disordered proteins and hub proteins, like 14-3-3, with multiple interaction partners. Disulfide tethering is a site-directed fragment-based drug discovery (FBDD) methodology for identifying reversibly covalent small molecules. We explored the scope of disulfide tethering for the discovery of selective PPI stabilizers (molecular glues) using the hub protein 14-3-3σ. We screened complexes of 14-3-3 with 5 biologically and structurally diverse phosphopeptides derived from the 14-3-3 client proteins ERα, FOXO1, C-RAF, USP8, and SOS1. Stabilizing fragments were found for 4/5 client complexes. Structural elucidation of these complexes revealed the ability of some peptides to conformationally adapt to make productive interactions with the tethered fragments. We validated eight fragment stabilizers, six of which showed selectivity for one phosphopeptide client, and structurally characterized two nonselective hits and four fragments that selectively stabilized C-RAF or FOXO1. The most efficacious fragment increased 14-3-3σ/C-RAF phosphopeptide affinity by 430-fold. Disulfide tethering to the wildtype C38 in 14-3-3σ provided diverse structures for future optimization of 14-3-3/client stabilizers and highlighted a systematic method to discover molecular glues.


Disulfide Tethering Screen and Data Processing
The primary disulfide tethering screen was performed by incubating target 14-3-3σ protein/phospho-peptide complex with small molecule in a 384-well plate format. The UCSF Small Molecule Discovery Center (SMDC) custom 1600 disulfide-containing fragment library was available as 50 mM stock solutions in DMSO. The screen was performed using 100 nM 14-3-3σ ΔC protein diluted in buffer (10 mM Tris, 250 µM betamercaptoethanol (βME), pH 8.0) and plated in 384-well plates (25 µL/well). 100 nL of each fragment was pinned from library master plates using non-sterile disposable 384 polypropylene pin tools (V & P Scientific) to give a final concentration of 200 µM. The peptide screens additionally contained a concentration of peptide equivalent to twice the KD as established by fluorescence anisotropy and competition experiments: (2 µM ERα-pp, 10 µM SOS1-pp, 10 µM USP8-pp, 18 µM CRAF-pp, and 750 nM FOXO1-pp). The reactions were incubated at room temperature for 3 hours before being measured by LC/MS (I-class Acquity UPLC/ Xevo G2-XS Quadrupole Time of Flight mass spectrometer, Waters). Data collection and automated processing followed a custom workflow, as previously described. 1 Compound resynthesis was done following the published procedures. 2

Dose Response LC/MS Experiments
The initial mass spectrometry dose response follow-up experiments were performed under the same conditions as the primary screen with the exception that the compounds were titrated from 50 mM to 23 μM in a 3-fold dilution series in DMSO. Then 1 µL of the compound was transferred into 24 µL of protein-peptide solution for final concentrations of 2 mM -914 nM and 4% DMSO. A final well of DMSO without compound was used as a control at the end of every dilution series. Compounds which displayed dose dependent activity in the presence of phospho-peptide were taken forward into more extensive dose responses in order to determine DR50 values. Fragments were titrated from 50 mM to 0.129 nM in a 3-fold dilution series in DMSO. Then 1 µL of the compound was transferred into 24 µL of protein-peptide solution for final concentrations of 2 mM -5.16 pM and 4% DMSO. A final well of DMSO without compound was used as a control at the end of every dilution series. Initial validation was done in single replicate. Hits were taken forward to fluorescence anisotropy experiments. DR50 curves were graphed and calculated using log(agonist) vs. response -Variable slope (four parameters) from GraphPad Prism version 9.0 for Mac, GraphPad Software, San Diego, California USA, www.graphpad.com.

Dose Response Fluorescence Anisotropy Experiments
The fluorescence anisotropy compound dose responses were performed using 100 nM fluorescein-labeled peptides (with the exception of FOXO1-pp which was 10 nM) and 14-3-3σ protein diluted in buffer (10 mM HEPES pH 8.0, 150 mM NaCl, 0.01% TWEEN-20). 14-3-3σ protein concentration was equivalent to approximately half of the KD determined by fluorescence anisotropy and competition experiments (C-RAF: 1 μM 14-3-3σ; ERα: 250 nM 14-3-3σ; FOXO1: 25 nM 14-3-3σ; USP8: 2.5 μM 14-3-3σ). Fragments were titrated from 50 mM to 11.9 nM in a 2-fold dilution series in DMSO. 1 µL of compound was transferred in triplicates into 24 µL of protein-peptide solution for final concentrations of 2 mM -0.477 nM and 4% DMSO. A final well of DMSO without compound was used as a control at the end of the dilution series. A row of 5-FAM-labeled peptide alone was used as a lower limit and a row of 5-FAM-labeled peptide with 14-3-3 excess of the KD was used as an upper limit when calculating the EC50 of the fragments. EC50 measurements were graphed and calculated using log(agonist) vs. response -Variable slope (four parameters) from GraphPad Prism version 9.0 for Mac, GraphPad Software, San Diego, California USA, www.graphpad.com. 14-3-3σ protein titrations were performed using 2-fold dilution of 14-3-3σ in buffer from starting concentration of 250 μM to 59.6 pM for C-RAF and USP8 and 25 μM starting concentration to 5.96 pM for FOXO1 and ERα. Final well contained only 5-FAM-labeled peptide and compound as a control. 5-FAM-peptides were used at 100 nM except FOXO1 which was at 10 nM. Fragments were added at a single concentration of 1 mM in order to maximize 14-3-3σ engagement. Apparent KD values of 14-3-3σ/phosphopeptide in presence of compounds were compared to KD values of DMSO control. Fold stabilization (ΔKD) was determined as the quotient of the apparent KD of the DMSO control over the apparent KD with compound. EC50 and apparent KD measurements were graphed and calculated using log(agonist) vs. response -Variable slope (four parameters) from GraphPad Prism version 9.0 for Mac, GraphPad Software, San Diego, California USA, www.graphpad.com.

X-ray Crystallography
The 14-3-3σΔC protein, acetylated ERα, C-RAF and FOXO1 peptides and fragments (stock solution of 50mM in DMSO) were dissolved in complexation buffer (25 mM HEPES pH 7.5, 2 mM MgCl2 and 2 mM βME) and mixed in a 1:2:2 molar stoichiometry (protein:peptide:fragment) at a final protein concentration of 12 mg/mL. The complex was set up for sitting-drop crystallization after overnight incubation at 4 °C, in a custom crystallization liquor (0.095 M HEPES (pH 7.1, 7.3, 7.5, 7.7), 0.19 M CaCl2, 24-29 % PEG 400 and 5% (v/v) glycerol). Crystals grew within 10 -14 days at 4 °C. Crystals were fished and flash-cooled in liquid nitrogen. X-ray diffraction (XRD) data were either collected at the Deutsche Elektronen-Synchrotron (DESY) PETRA III beamline P11, Hamburg, Germany (all datasets except for 14-3-3σ/C-RafpS259/Compound 1) or at the Diamond Light Source (DLS) beamline Io3, Oxfordshire, United Kingdom (14-3-3σ/C-RafpS259/Compound 1). Automatic processing was done by DESY or DLS using XDS9 3 for data indexing and integration Initial processed data was then scaled using AIMLESS 4,5 and Molrep 6 was used for limited molecular replacement using PDB ID's 4JC3 and 3IQU as template. Presence of co-crystallized ligands was verified by visual inspection of the Fo-Fc and 2Fo-Fc electron density maps in Coot. 7 If electron density corresponding to the co-crystallized ligand was present, its structure and restrains were generated using eLBOW 8 before final model rebuilding and refinement was done using phenix.refine 9,10 and Coot. See Table S6 and S7 for data collection and refinement statistics. The structures were submitted to the PDB with IDs 8AFN, 8AV0, 8ADM, 8A62, 8A65, 8A68, 8A6H, 8A6F.

Tethering and stabilization of 14-3-3/C-RAF primary screen hit compounds
where ! is the number of observations of reflection ℎ e Correlation of experimental intensities with intensities calculated from refined model, as described by Karplus and Diederichs. 11 S10 Table S7. 3.36 a Number in parentheses is for the highest resolution shell used in the refinement b CC1/2 = Pearson's intra-dataset correlation coefficient, as described by Karplus and Diederichs. 11

14-3-3σ ΔC
, where !" is the intensity of the 1th observation of reflection h and < ! > is the average intensity of reflection h where ! is the number of observations of reflection ℎ e Correlation of experimental intensities with intensities calculated from refined model, as described by Karplus