Structure‐Based Design of a Macrocyclic PROTAC

Abstract Constraining a molecule in its bioactive conformation via macrocyclization represents an attractive strategy to rationally design functional chemical probes. While this approach has been applied to enzyme inhibitors or receptor antagonists, to date it remains unprecedented for bifunctional molecules that bring proteins together, such as PROTAC degraders. Herein, we report the design and synthesis of a macrocyclic PROTAC by adding a cyclizing linker to the BET degrader MZ1. A co‐crystal structure of macroPROTAC‐1 bound in a ternary complex with VHL and the second bromodomain of Brd4 validated the rational design. Biophysical studies revealed enhanced discrimination between the second and the first bromodomains of BET proteins. Despite a 12‐fold loss of binary binding affinity for Brd4, macroPROTAC‐1 exhibited cellular activity comparable to MZ1. Our findings support macrocyclization as an advantageous strategy to enhance PROTAC degradation potency and selectivity between homologous targets.


Supplementary Figures
in the interface of VHL with Brd4 BD2 and root-mean-square deviation (RMSD) of all α-carbon atoms in the proteins, after superposition to the crystal structure of VCB:MZ1:Brd4 BD2 (PDB code 5T35). [1] (C) Minimum distance between selected amino acids in VHL and Brd4 BD2 that participate in protein-protein interactions. [1] For clarity, the plotted distances are the average over 5 ns. In all cases, only the last 50 ns in 100 ps intervals of a 200-ns MD simulation of the VHL:1:Brd4 BD2 ternary complex were considered to generate the data. The mean ± 1 s.d. of each measurement is shown.

Torsion analysis of surrogate N-substituted acetamides
Model compounds N-ethylacetamide and N-isopropylacetamide (Fig. 1C) were subjected to a relaxed torsion scan analysis of their alkylacetamide bond using density functional theory (DFT) at the PBF (water) MN15-L/aug-cc-pVTZ(-F) level of theory in Jaguar 9.7 (Schrödinger, LLC).
The C-N-C-C torsion was gradually rotated from 0 to 180º in 10º steps, considering Abelian molecular symmetry when appropriate.

Molecular modeling of macrocyclic MZ1 derivatives
Derivatives of MZ1 macrocyclized using a 2-PEG and a 3-PEG linker (Fig. 1D) were modeled in situ using the VHL:MZ1:Brd4 BD2 ternary complex crystal structure as template and Glide Ligand Designer (Schrödinger). Water and solvent molecules present in the protein crystal structure were removed, and the VHL:MZ1:Brd4 BD2 system was prepared for energy minimization in Prime (Schrödinger) using the Protein Preparation Wizard (Schrödinger). Amino acid protonation states were assigned using PROPKA 3.0. [2] Molecular dynamics (MD) simulation of the VHL:1:Brd4 BD2 ternary complex The energy-minimized model of VHL:1:Brd4 BD2 was used as starting point for a molecular dynamics (MD) simulation. MD simulations were carried out in an nVIDIA TITAN X GPU using Desmond (Schrödinger) and the OPLS3 force field. [3] System Builder (Schrödinger) was used to solvate the complex in a TIP3P water box with a padding of 10 Å from the edge of the box to any solute atom, and to neutralize the system charges with three chloride ions. The solvated system was minimized for 2,000 steps with all protein and PROTAC atoms restrained to eliminate residual unfavorable interactions between the solute and the solvent, followed by another 2,000 steps with unrestrained PROTAC and solvent atoms, and lastly followed by another 5,000 steps with all atoms free to move. The equilibration phase consisted of an initial 100-ps Brownian dynamics simulation at 10 K with restraints on solute heavy atoms in the NVT ensemble, followed by one run of 120 ps of MD at the same temperature, restraints, and ensemble. The restrained system was then subjected to two runs of 120 ps in the NPT ensemble, first at 10 K and then at 300 K, and a final unrestrained simulation at the final temperature (300 K) for 240 ps. We run 200 ns of MD (time step of 2 fs using the RESPA integrator), starting from the equilibrated system. The temperature was controlled with a Nose-Hoover chain thermostat at 300 K, and the pressure with a Martyna-Tobias-Klein barostat at 1 bar. Short-range nonbonded interactions were cut off at 9 Å.
Only the last 50 ns of production simulation were used for data collection and analysis.

Analysis of MD trajectories
The last 50 ns of the 200-ns MD trajectory were analyzed using the Simulation Quality, Event, and Interactions Analysis tools, as included in Schrödinger. The buried surface area (BSA) of the proteins upon complex formation, i.e. the difference in surface-accessible surface area (SASA) between the formed complex and the unbound partners in each system, was computed using VMD v. 1.9.2 [4] and considering all protein atoms and a spherical probe of radius 1.4 Å.

System
Number of atoms

Protein expression and purification
Wild

Fluorescence polarization assay
FP competitive binding assays were run in triplicate on 384-well plates [Corning 3575] as described previously, [8] with all measurements taken using a PHERAstar FS (BMG LABTECH) with fluorescence excitation and emission wavelengths (λ) of 485 and 520 nm, respectively. The  Diffraction data were collected at Diamond Light Source beamline I04 using a Pilatus 6M-F detector at a wavelength of 0.9750 Å. Reflections were indexed and integrated using XDS [9] , and scaling and merging were performed with AIMLESS [10] in CCP4i. [11] The crystals belonged to space group P32, and there were two copies of the ternary complex in the asymmetric unit. The structure was solved by molecular replacement using PHASER MR [12] and search models derived from the coordinates for the VCB:MZ1:Brd4 BD2 ternary complex (PDB entry 5T35). The initial model underwent iterative rounds of model building and refinement with COOT [13] and REFMAC5, [14] respectively. All riding hydrogens were excluded from the output coordinate files but included for refinement. Compound 1 geometry restraints for refinement were prepared with the PRODRG [15] server and optimized using eLBOW [16] from the PHENIX suite. [17] Model geometry and steric clashes were validated using the MOLPROBITY server. [18] Ramachandran plots indicate that 96.8% of backbone torsion angles are in the favored region and there are no outliers. The structure has been deposited in the protein data bank (PDB) with accession code 6SIS; data collection and refinement statistics are presented in table S2. Interfaces observed in the crystal structure were calculated using PISA, and all figures were generated using PyMOL.

Immunoblotting
Protein on gel was transferred to nitrocellulose membrane using the Bio-

Western Blot Quantification
Image processing and band intensity quantification were performed using Bio-Rad Image Lab