Targeting Ligandable Pockets on Plant Homeodomain (PHD) Zinc Finger Domains by a Fragment-Based Approach

Plant homeodomain (PHD) zinc fingers are histone reader domains that are often associated with human diseases. Despite this, they constitute a poorly targeted class of readers, suggesting low ligandability. Here, we describe a successful fragment-based campaign targeting PHD fingers from the proteins BAZ2A and BAZ2B as model systems. We validated a pool of in silico fragments both biophysically and structurally and solved the first crystal structures of PHD zinc fingers in complex with fragments bound to an anchoring pocket at the histone binding site. The best-validated hits were found to displace a histone H3 tail peptide in competition assays. This work identifies new chemical scaffolds that provide suitable starting points for future ligand optimization using structure-guided approaches. The demonstrated ligandability of the PHD reader domains could pave the way for the development of chemical probes to drug this family of epigenetic readers.


NMR SPECTROSCOPY
NMR experiments were performed using a AV-500 MHz Bruker spectrometer equipped with a 5 mm CPQCI-1 H- 19 F / 13 C/ 15 N/D Z-GRD cryoprobe. All spectra were recorded and processed with TopSpin (Bruker) and analysed with CCPNMR software. 15 N( 1 H)-HSQC spectra of BAZ2 PHDs were recorded at ~100 µM for BAZ2A and ~150 µM for BAZ2B in 200 µL NMR buffer containing 25 mM H 2 PO 4 -/HPO 4 2-, 50 mM KCl, 1 mM DTT and 20% D 2 O at pH 6.9 and 6.5, respectively. Labelled proteins were then incubated with 5 mM of each fragment at final concentration of 1.25% of d 6 -DMSO. Spectra of proteins incubated with fragment were overlaid with the spectrum of the apo protein containing 1.25% d 6 -DMSO.

Chemical shift perturbations and K D estimation
Chemical shift perturbation experiments were performed at the same protein concentration and pH as above but in 50 mM H 2 PO 4 -/HPO 4 2-, 1 mM DTT and 20% D 2 O. Each fragment was titrated in each protein at increasing concentration in a range between 0.5 mM and 5 mM. As reference spectrum was used the apo form. The weighted chemical shift difference (Δδ weighted ) was calculated using the equation: !"#$!!"# = | | ! + | | ! * 0.15 , 3 where ΔδH is the chemical shift on the proton and ΔδN is the chemical shift on the nitrogen which is scaled with a factor 0.15 to account for the different range of the amide proton and amide nitrogen. Chemical shifts for each backbone amide group were measured from the peak detected in apo form spectrum to the peak at the end of the titration. K D of binding for each fragment were estimated using the following equation 3 in the CCPNMR software: [P]t and [L]t are the total concentration of protein and ligand; Δδobs is the observed shift in regard to the reference, while Δδmax is the maximum shift obtained upon saturation and is extracted from the fitting. K D for each fragment was extrapolated as an average value of K D ± s.e.m. of those resonances with Δδ > + σ.

Isothermal titration calorimetry
(ITC) experiments were performed with the ITC200 micro-calorimeter (GE Healthcare) at 298 K stirring the sample at 750 rpm. Before titration, BAZ2 PHD zinc fingers were dialyzed overnight against buffer containing 20 mM Tris, 200 mM NaCl, 1 mM TCEP (tris(2-carboxyethyl)phosphine) pH 8.0, using the D-tube dialyzer MWCO 3.5 KDa (Millipore). Titrations were performed in direct mode titrating 3 mM peptide solutions into 120 µM protein solution loaded in the calorimetric cell. First injection was of 0.4 µL (subsequently discarded during data analysis) followed by 19 injections of 2 µL at 120 s time intervals. For each protein, at the end of this first titration, in order to reach saturation of binding, 20 further injections were performed, after removing the excess of solution from the cell. A control experiment of peptide into buffer was performed. Data were fitted keeping the stoichiometry N fixed at 1. Dissociation constant K D and the enthalpy of binding ΔH were obtained using the MicroCal ORIGIN software package.

Thermal shift assay (TSA)
Thermal shift experiments were performed using a 96-well PCR plate in CFX96 Touch Realtime PCR detection system (Biorad). Each well contained 40 µL reaction and the final conditions were: 2.5X Sypro Orange (Invitrogen Molecular Probes) and 10 µM protein in 100 mM MES, 50 mM NaCl, pH 6.0. Fragments were screened against both proteins at three different concentrations: 3 mM, 5 mM and 15 mM in 5% (v/v) DMSO (dimethyl sulfoxide). Each sample was run in triplicate. The assay was conducted by increasing the temperature from 25 °C to 95 °C at a rate of 1 °C per minute and detecting fluorescence at the end of each interval. Fluorescence was plotted versus temperature and the melting temperature (Tm) was extrapolated from the fitting of the Boltzmann equation using the "DSF analysis" excel spreadsheet available at ftp://ftp.sgc.ox.ac.uk/pub/biophysics. The tabulated ΔTm values are the difference of the mean of three independent measurements ± propagated s.d.

AlphaLISA
AlphaLisa competition assays were developed using biotinylated BAZ2 PHD and H3 double mutant peptide (ARTAATARKS) synthetized with the additional Anti-FLAG epitope at the C-terminus plus an intermediate flexible linker in order to avoid steric hindrance (ARTAATARKS-TGGSGGSG-DYKDDDDK). The assay was set up in a 384-well plate (PerkinElmer) in 100mM HEPES buffer pH 7.5, 0.1% BSA and 0.02% CHAPS. Each well contained a final concentration of 10 nM protein, 160 nM AlphaLISA peptide, fragment at desired concentration in 4% (v/v) DMSO and 10 µgmL -1 of each AlphaLISA beads (Anti-FLAG AlphaLisa acceptor beads and Streptavidin donor beads). Beads were cautiously added to the solution under low light conditions. The plate was incubated for one hour at room temperature before to be read on a PHERAstar FS plate reader (BMG Labtech) at laser excitation of 680 nm and filter set on emission light at 615 nm. BAZ2 PHDs were tested against eight different peptide or fragment concentrations in 1:5 serial dilution. Fitting of the data to extrapolate IC 50 values of the dose-response curves was performed using the GraphPad Prism 6 (GraphPad Software).

X-RAY CRYSTALLOGRAPHY Crystallization
Crystals of BAZ2A and BAZ2B PHD domains were obtained by mixing equal volume of protein and crystallisation buffer (2.2-2.4 M Na/K phosphate at pH 8.5). BAZ2A PHD was crystallized at 6.5-7 mg ml -1 and BAZ2B PHD at 5.5-6 mgml -1 . Crystals were left to grow at 18°C for at least two days.

Crystallization H3 3-mer peptide in complex with BAZ2A PHD finger
Crystals of the complex of BAZ2A PHD with H3 3-mer were obtained soaking overnight preformed apo BAZ2A PHD crystals in a solution containing 20 mM H3 3-mer (ART) in crystallization buffer.

Crystallization H3 10-mer AA mutant peptide in complex with BAZ2A PHD finger
Crystals of the complex of BAZ2A PHD with H3 10-mer AA mutant peptide (ARTAATARKS) were obtained soaking overnight preformed apo BAZ2A PHD crystals in a solution containing 2.5 mM H3 10-mer in crystallization buffer.

Crystallization of Fr19 and Fr23 in complex with BAZ2A PHD finger
Crystals of the complex of BAZ2A PHD with Fr19 and Fr23 were obtained soaking for 24 h apo form crystals of BAZ2A in 10 µL of a solution containing crystallization buffer supplemented, respectively, with 50 mM Fr19 and 50 mM Fr23.

Crystallization of Fr21 and Fr23 in complex with BAZ2B PHD finger
Crystals of BAZ2B PHD in complex with Fr21 and Fr23 were obtained soaking for 24h BAZ2B apo form crystals in a solution of crystallization buffer supplemented, with, respectively, 20 mM Fr21 and 50 mM Fr23.
All the crystals, before to be flash-frozen, were cryo-protected in the soaking solution supplemented with 20% glycerol. All the fragments used were soluble in water up to 500 mM.

Data collection and structure determination
Data collections were performed at the beamlines at the Diamond Light Source (Didcot, UK) synchrotrons. Images of BAZ2A PHD crystals were indexed and integrated using XDS. 4,5 Images of BAZ2B PHD were indexed and integrated using Mosflm. 6 Scaling and merging was performed using Aimless 7 from the CCP4i package and copying R free flags from the apo form pdb models used for the refinement (PDB: 4QF2 for BAZ2A and 4QF3 for BAZ2B). Structures were solved using 4QF2 pdb as model for BAZ2A and 4QF3 pdb as model for BAZ2B. Several rounds of refinement were performed using Refmac5 8 with TLS groups generated via TLSMD server. 9 F o -F c map showed clear electron density in the histone pocket to fit peptides and fragments. Waters were manually added to the model at the latest stages of the refinement before to model the ligands. Residues at the N-terminus of the protein chains were not visible, consequently they were not modeled. Modeling of the fragments was performed using the ligand builder in Coot. 10 The topology files of the fragments were generated using PRODRG2 server. 11 The quality of the models was checked by MolProbity, 12 and all structure figures were generated using PyMOL (The PyMOL Molecular Graphics System, Version 1.7.05, Schrödinger, LLC).

Computational druggability assessment of BAZ2A PHD
The crystal structure of apo PHD of BAZ2A (PDB code 4QF2) was subjected to computational druggability assessment using FTMap. 13,14 Water and solvent molecules present in the structure were removed in advance. Then, identified potential binding sites were classified according to their perceived druggability using the "classify_druggability" tool included in FTMap. Physicochemical properties of the pockets were examined using SiteMap v3.8 (Schrödinger Inc., LLC).

Virtual screening cascade of BAZ2A and BAZ2B PHDs
We assembled a diverse virtual library of a thousand commercially-available fragments and low-molecular weight biomimetics, peptidomimetics, and charged compounds. The molecules were processed using LigPrep v3.7 (Schrödinger Inc., LLC), and the protonation states were perceived using Epik 3.5 (Schrödinger Inc., LLC). The crystal structures of apo PHD of BAZ2A and BAZ2B (PDB codes 4QF2 and 4QF3, respectively) were subjected to molecular docking using Glide v7.0 (Schrödinger Inc., LLC). Water and solvent molecules present in the protein crystal structures were removed, and the proteins were prepared for docking using the Protein Preparation Wizard (Schrödinger Inc., LLC). Amino acid protonation states were assigned using PROPKA 3.0. 15,16 Two grids were prepared for each of the prepared proteins: one covering the histone tail pocket, and another one centred in the newly proposed binding site at the back of the proteins. Each grid was used to screen virtually the compound library using the Standard Precision algorithm in Glide. The virtual screening results were normalized for ligand efficiency by dividing the docking score by the number of heavy atoms in the molecule. Compounds with a normalized docking score ≥ 0.25 were redocked using the eXtra Precision (XP) algorithm in Glide. XP docking scores were again normalized by number of heavy atoms, and the top 200 compounds were subsequently processed with the MM-GBSA protocol in Prime 3.0 (Schrödinger Inc., LLC), considering water solvation and a protein shell of 5.0 Å surrounding the docked compounds as flexible with constraints.

Compound selection
Physicochemical properties and aqueous solubility of top-ranked compounds for each grid in terms of normalized docking score or normalized MM-GBSA E bind were predicted using QikProp 4.7 (Schrödinger Inc., LLC) and the Calculator Plugins included in Marvin 15.9.7, 2015, ChemAxon (http://www.chemaxon.com). Surviving compounds (predicted pK s ≤ 3) were visually inspected and a varied set of the most promising compounds was purchased for experimental validation of binding affinity.

Selection of Fr19 derivatives
We assembled a focused virtual library of commercially-available low-molecular weight compounds extracted from the ZINC 15 database. 17 We selected compounds containing the peptidic backbone of Fr19 using the SMARTS pattern "[N;A;+1][C;A]C(=O)[#7]", which was generated informed by the crystal structure of BAZ2A in complex with the fragment and with assistance of SMARTSviewer. 18 Filtering of the compounds by having at least 1 aromatic ring (using Open Babel 19 ) and a predicted pK s ≤ 3 resulted in 4,196 unique molecules. Compound library preparation was carried out as above.
The histone pocket of the crystal structure of BAZ2A in complex with Fr19 was subjected to the previously described molecular docking protocol to screen the focused compound library. Visual inspection of top-ranked hits based on MM-GBSA E binding led to the final selection of most promising compounds for purchase and experimental validation of binding affinity.

Torsion analysis of model compounds
Model compounds thiazolacetamide and phenylacetamide were subjected to fully flexible torsion analysis of their arylacetamide bond using DFT at the PBF (water) MN15-L/aug-cc-pVTZ(-F) level of theory in Jaguar 9.7 (Schrödinger Inc., LLC). The torsion was gradually rotated from 0 to 180º in 10º steps, and no molecular symmetry was considered during the analysis. Natural bond order (NBO) analysis of the global energy minimum in thiazolacetamide was carried out using NBO 6.0 (http://nbo6.chem.wisc.edu), as implemented in Jaguar.    In white all residues reporting intensity of shift < + σ. In red are fragments that reported shifts for the histone pocket, while in blue fragments that showed shifts for the "back-pocket".  Table S2. K D are calculated as mean ± s.e.m. of single K D extrapolated from CSPs as described in the experimental section. In brackets, ligand efficiency in kcal × mol -1 × heavy atom -1 . NT = not tested.       HSQC-validated fragments were tested in the AlphaLISA competition assay with both BAZ2 PHDs. Dose-response curves are reported above. IC 50 were obtained where it was possible to fit the experimental data. NT = not detectable because precipitation of the fragments was observed during sample preparation. It is interesting to note that Fr7 and Fr14 showed displacement of the histone peptide against both proteins, however by NMR CSP analysis they were identified as binder for BAZ2B or BAZ2A only, respectively. Fr7 was found to bind to the "back pocket" by CSPs, so the ability of this fragment to displace the histone peptide might reflect an allosteric effect.