Surface Probing by Fragment-Based Screening and Computational Methods Identifies Ligandable Pockets on the von Hippel–Lindau (VHL) E3 Ubiquitin Ligase

Beyond the targeting of E3 ubiquitin ligases to inhibit protein homeostasis, E3 ligase binders can be repurposed as targeted protein degraders (PROTACs or molecular glues). We sought to identify new binders of the VHL E3 ligase by biophysical fragment-based screening followed by X-ray crystallographic soaking. We identified fragments binding at the ElonginC:Cullin2 interface and a new cryptic pocket in VHL, along with other potential ligandable sites predicted computationally and found to bind solvent molecules in crystal structures. The elucidated interactions provide starting points for future ligand development.


Protein expression, purification, and crystallization
The VCB and VBCH complexes were expressed, purified, and crystallized as previously described [1,2].

X-ray data collection and protein structure determination
VCB and VCBH crystals were soaked overnight with 20 mM fragment in 2 % DMSO, 8 % isopropanol, in the respective crystallization condition. Crystals were cryoprotected by 10 % glycerol in the crystallization condition. X-ray data for the VBC crystals soaked with MB235, MB756, and MB1200 were collected at the Proxima 1 beamline at the Soleil synchrotron facility and at the i03 and i24 beamlines at the Diamond Light Source synchrotron facility (Table S1). X-ray data collected for the VCBH crystals soaked with MB756 were collected at the Proxima 1 beamline at the Soleil synchrotron facility. X-ray data were processed using XDS [3]. The structures were solved using rigid body refinement using the respective VCB (PDB 3ZRF) [1] and VCBH (PDB 4AJY) [2] structures as reference, using BUSTER [4]. Presence of electron density for the fragments was assessed by manual inspection using Coot [5]. Coordinates and restrains for the ligands were obtained using Elbow from the Phenix suite [6]. The ligands were manually fitted in the electron density in Coot. The structure was further refined using Phenix.refine [7]. X-ray data collection parameters, processing, and refinement statistics are listed in Table S1.

Differential scanning fluorimetry (DSF)
DSF experiments [8] were performed using a Roche Lightcycler 480 machine, in a 96-well plate setup, using 100 µl per well. Compounds were assayed at 1 mM concentration in triplicates, using a concentration of 5 µM VCB in 100 mM Tris (pH 8.

Isothermal titration calorimetry
ITC experiments were performed using an ITC200 instrument from Microcal Inc. (GE Healthcare) at 25 ºC. MB1200 and MB756 were dissolved in 100 % DMSO at 200 mM and subsequently diluted to 20 mM in 20 mM Bis-Tris pH 7.0, 150 mM NaCl, 1 mM DTT. The fragments were titrated into buffer and into a solution of 50-100 µM VCB, equilibrated in the same buffer containing the matching DMSO concentration. The titrations comprised 32 × 1.2 µl injections. An initial injection of ligand (0.5 µl) was made and discarded during data analysis. The resulting buffer-subtracted data were fitted to a single binding site model (stoichiometry fixed to n = 1) using the Microcal LLC ITC200 Origin software provided by the manufacturer, to obtain the binding constant (K a ) and the enthalpy of binding (ΔH).

Computational methods
For all the computational work the structure of VHL:EloC:EloB after removing the HIF-1α peptide, water, and other solvents from the VHL:EloC:EloB:HIF-1α crystal structure (PDB 4AJY) [2] was used as model since it has higher atomic resolution than the apo VCB crystal structure PDB 3ZRF (resolution 1.73 and 2.8 Å, respectively).

Protocol for the randomized cosolvation with fragment probes for mixed-solvent MD
The protein complex VCB was initially solvated in a TIP3P water box with a padding of 15 Å from the edge of the box to any protein atom. The system charges were neutralized with sodium or chloride ions as appropriate. Then, a PyMOL script was used to randomize the addition of the cosolvent fragment probes. Iteratively, a water molecule beyond 4.0 Å of any non-water atom or the edges of the cell was selected and converted to a molecule probe. The conversion to the probe was rejected if the probe clashed with any non-water atoms or escaped the cell. This step was repeated until a suitable location for the probe was found. Then, waters within 2.5 Å of the new probe were removed. This new structure was stored and used for the next iteration, and the process was repeated until the desired probe concentration was reached. In the present study, a final probe concentration of 1.3 M was used for all probes. Molecular volumes of probes and VCB were calculated using the 3 V volume assessor web server [13] (Table S2). Topology and parameter files for each fragment probe were generated using the CGenFF server [14,15]. Probes were designed to obtain reliable parameters based on the internal CGenFF reliability score. Physicochemical properties of the probes were calculated using QikProp 5.0 (Schrödinger, LLC) and the Calculator Plugins included in Marvin 15.9.7, 2015, Che-mAxon (http://www.chemaxon.com).

Protocol for the molecular dynamics simulations
Molecular dynamics simulations were carried out using the NAMD program [16] and the CHARMM 36 force field [17]. Initially the mixed-solvated systems were minimized for 3,000 steps with VCB restrained to eliminate residual unfavorable interactions between the protein complex and the mixed solvents, followed by another 5,000 steps with all atoms free to move. Heating of the systems from 0 to 300 K was achieved in 100 ps (time step of 1 fs), with fixed protein backbone atoms to allow relaxation of the mixed solvent. The systems were subsequently equilibrated for 600 ps (time step of 1 fs) with all atoms free to move. The production simulations involved three independent replicates of 15 ns (time step of 2 fs) using the NPT ensemble (total production simulation time of 675 ns). The temperature was controlled with a Langevin thermostat at the final temperature, and the pressure with a Nose-Hoover Langevin piston barostat at 1 bar with a damping coefficient of 5/ps. The damping coefficient was set to 7/ps for phenylpyrimidine due to instability during the simulations. A SHAKE constraint was applied to all bonds containing hydrogen atoms. Short-range nonbonded interactions were switched at 10 Å and cut off at 12 Å, and particle mesh Ewald summation was employed for long-range nonbonded interactions. For each simulation, the last 5 ns of production simulation were considered for analysis using a time step of 10 ps, i.e. amounting 500 frames per replica and 1,500 frames per probe in total. Statistics of the MD runs were obtained using VMD 1.9 [18] and are included in Table S3.

Protocol for the quantification of probe enrichment at the protein complex surface
At every frame the percentage of solvent-accessible surface area (SASA) of each probe was computed by calculating the ratio between the SASA of the individual probe in the presence and in the absence of the protein using PyMOL 1.8 (Schrödinger, Fig. S4D). A probe was considered buried if its percentage of SASA was below 50 % (i.e. the molecule had a percentage of buried surface area equal of above 50 %, Fig. S4D). Then, amino acids around 4.0 Å of buried probes were identified and recorded. For each probe type, amino acids that buried probes in at least 10 % of frames were considered as actively contributing to burying the probe (Fig. S6). Finally, only amino acids that contributed to burying two or more probe types were regarded as interaction hotspots and considered for further analysis ( Fig. 2 and S7).

Protocol for the analysis of crystal contacts
Available X-ray crystal structures of VHL, EloB, and EloC were downloaded from the Protein Data Bank (PDB, accession date: 14/11/2017). Systematic inspection of the electron density maps in the surroundings of the validated and predicted binding sites was carried out using Coot 0.8.7 [19]. Subsequent analysis of noncovalent interactions and preparation of the corresponding images for the manuscript were done in PyMOL (Schrödinger, LLC). Figure S1. Scheme for the fragment-based screening cascade.           [a] R cryst = S||F obs |-|F calc ||/S|F obs |, F obs and F calc are observed and calculated structure factor amplitudes [b] R free as for R cryst using a random subset of the data excluded from the refinement (c) Data in brackets are for the highest resolution shell S18  [13]. d Largest partial atomic charge difference, Δc, extracted from the parameters file generated by the CGenFF server (http://cgenff.paramchem.org) [14,15].  HIF-1α peptide