Target-templated de novo design of macrocyclic d-/l-peptides: discovery of drug-like inhibitors of PD-1

Peptides are a rapidly growing class of therapeutics with various advantages over traditional small molecules, especially for targeting difficult protein–protein interactions. However, current structure-based methods are largely limited to natural peptides and are not suitable for designing bioactive cyclic topologies that go beyond natural l-amino acids. Here, we report a generalizable framework that exploits the computational power of Rosetta, in terms of large-scale backbone sampling, side-chain composition and energy scoring, to design heterochiral cyclic peptides that bind to a protein surface of interest. To showcase the applicability of our approach, we developed two new inhibitors (PD-i3 and PD-i6) of programmed cell death 1 (PD-1), a key immune checkpoint in oncology. A comprehensive biophysical evaluation was performed to assess their binding to PD-1 as well as their blocking effect on the endogenous PD-1/PD-L1 interaction. Finally, NMR elucidation of their in-solution structures confirmed our de novo design approach.

Peptides were purified by semi-preparative HPLC on a Waters 2700 sample manager equipped with a Waters 2487 dual-wavelength absorbance detector, a Waters 600 controller, a Waters fraction collector and Masslynx software by using a Sunfire C18 column (150 x 10 mm x 3.5 μm, 100 Å, Waters), flow rate 6.6 mL/min, solvent A=0.1% TFA in water; solvent B=0.1% TFA in acetonitrile.

Synthesis of cyclic peptides PD-i1-7
Synthesis was performed on a Wang polystyrene resin (theoretical substitution of 1.14 mmol/g). Resin was swelled in 9:1 v/v CH 2 Cl 2 /DMF (15 mL/g). In a separate flask, Fmoc-Asp-OAll (1.2 eq, relative to the resin) and HOBt-Cl (1.2 eq) were dissolved in a minimum amount of DMF and added to the resin. Then, 4-dimethylaminopyridine (0.1 eq) and DIC (1.2 eq) were added. The mixture was allowed to react in an orbital shaker overnight at room temperature. Then, peptide chains were elongated by microwave-assisted automatic peptide synthesis. Fmoc deprotection was carried out using 10% (w/v) piperazine and 0.1 M OxymaPure in a 9:1 mixture of NMP and EtOH. The N-Fmoc-protected amino acids (5 equiv, 0.2 M in DMF) were added with OxymaPure (5 equiv, 1 M in DMF) and DIC (5 equiv, 0.5 M in DMF) to the resin. The mixtures were stirred for 3 min at 90ºC, except for cysteines, histidines and arginines, which were coupled at 50ºC for 10 min. When the chain elongation was finished, Alloc groups were deprotected with phenyl silane (20 eq) and Pd(PPh 3 ) 4 (0.1 eq) in DCM (3 x 15 min treatments). On-resin peptide cyclization was performed with OxymaPure (5 eq) and DIC (5 eq) in DCM until the reaction was complete (typically, 16 h). Finally, peptides were cleaved with concomitant removal of side-chain protecting groups, using TFA, H 2 O and TIS (92.5:5:2.5) for 2 h. After evaporating the residual TFA, peptides were ether precipitated and further purified by semi-preparative HPLC. Peptide characterization was performed by UPLC and UPLC-MS (see above).

Recombinant expression of human PD-1
Procedure adapted from ref. 2 Genes encoding the interacting ectodomain of human PD-1 (aa 33-150) was cloned into pET-24d. A Cys to Ser mutation was introduced at position 93 of PD-1 to aid protein stability and folding. The protein was expressed in E. coli BL21(DE3) in the form of inclusion bodies. The cells were grown at 37 °C in LB medium supplemented with 50 μg/mL kanamycin until OD 600 reached 0.6-1.0, and the protein expression was induced with 1 mM IPTG and incubated for 16 h at 30 °C. The cells were harvested by centrifugation, re-suspended in PBS (containing protease inhibitor cocktail and DNAse) and lysed by sonication on ice. Inclusion bodies were recovered by centrifugation (25,000 g for 30 min at 4 °C) and washed 2 times with 50 mM Tris buffer (200 mM NaCl, 0.5% Triton X-100, 10 mM EDTA and 10 mM 2mercaptoethanol, pH 8.0) followed by a final wash with the same buffer without the detergent. The inclusion bodies were re-suspended in guanidinium hydrochloride buffer (50 mM Tris, 6 M GuHCl, 200 mM NaCl, 10 mM 2-mercaptoethanol, pH 8.0) by stirring vigorously for 2 h at 4ºC. After removing undissolved residue by centrifugation (25,000g for 30 min at 4 °C) and refolded by slow drop-wise dilution in folding buffer (0.1 M Tris, 0.4 M L-Arg hydrochloride, 2 mM EDTA, 5 mM cystamine, 0.5 mM cysteamine, pH 8.0) for 3 days at 4ºC. The protein was then dialyzed 2 times against 10 mM Tris pH, 20 mM NaCl, pH 8.0 buffer. The solubilized fraction was diluted 1:2 in water, acidified to pH 6 and applied to HiTrap SP HP cation exchange chromatography column (GE Healthcare Life Sciences). The protein was washed with five column volumes of wash buffer (30 mM NaCl, 0.7 mM KCl, 2.5 mM phosphate buffer, pH 6.0) and eluted using a 0-50% gradient of elution buffer (1 M NaCl, 0.7 mM KCl, 2.5 mM phosphate buffer, pH 6.0). Finally, the protein was purified by gel filtration chromatography using a HiLoad 16/60 Superdex 200 pg column (GE Healthcare Life Sciences) with 10 mM Tris, 20 mM NaCl, pH 8.0 buffer. The protein purity was evaluated by SDS-PAGE, high-resolution mass spectrometry and multiangle light scattering.

Surface plasmon resonance (SPR)
Binding experiments were carried out in a Biacore T200 system (GE Healthcare) at 25ºC in PBS-T buffer (0.05% Tween 20, pH 7.4). hPD-1 (10 µg/mL) at 25 μg/ml in 10 mM sodium acetate (pH 5.5) was directly immobilized on the dextran matrix of a CM5 sensor chip (GE Healthcare) by amine coupling using the manufacturer's kit (GE Healthcare) and an injection time of 80 sec, resulting in immobilization levels of ∼600 RU. A solution of 10 mM glycine-HCl at pH 2.5 was used for chip regeneration between analyte injections. The protein bioactivity was assessed by injecting serial dilutions of Fc-tagged PD-L1 (Peprotech #310-35) at 1-min injections and performing kinetic analysis of the sensorgrams, which yielded a K D value of 4.6 µM for the interaction, in agreement with previous results. 3 To assess the potential interaction of peptides PD-i1-7, analytes were dissolved in PBS and injected at 10, 100 and 1000 µM concentrations. K D values could not be unequivocally calculated due to their fast kinetics and lack of signal saturation. Due to this limitation, those peptides presenting clear concentration-dependent associationdissociation signals were further evaluated by MST.

Microscale thermophoresis (MST)
MST was used to calculate the binding affinities in solution of PD-i3 and PD-i6 for the extracellular domain of human PD-1. To this end, PD-1 was fluorescently labelled by reaction of the Lys side chains with Alexa-647-NHS (6 eq) for 3 h at 4ºC. The labelled protein was purified by sizeexclusion chromatography and was diluted in PBS-T buffer (0.05% Tween 20, pH 7.4) to a final concentration of 10 nM. MST binding assays were performed on a Monolith NT.115 instrument (NanoTemper Technologies) with standard capillaries at 25ºC. The protein bioactivity was confirmed by using Fc-PD-L1 (Peprotech #310-35) and a specific anti-PD1 mAb (BioCell #BE0193) as positive controls ( Figure S). Peptides were diluted 2:1 from a concentrated stock in 10 µL of PBS-T buffer to make a 16-sample dilution series, which were titrated against a constant concentration of PD-1 (10 nM). A reproducible and negative thermophoresis response was observed, which was baseline-corrected and normalized (ΔFnorm [‰]).The curves were analyzed using a standard Langmuir binding model, from which dissociations constants (K D ) were determined. The experiment was performed in triplicate.

Isothermal titration calorimetry (ITC)
ITC experiments were performed at 25ºC using a low-volume nano ITC calorimeter (TA instruments). Ligand and protein samples were dissolved in the same buffer (10 mM Tris, 20 mM NaCl, pH 8.0), centrifuged, and degassed prior to the ITC experiments. For each titration, a concentrated peptide solution was injected into a cell containing 190 µL of protein solution at a concentration of 30 µM. A total of 16 injections of 3 µL per titration were performed with a 4-min delay after each injection. Binding isotherms were analysed using the software provided by TA instruments, assuming a single binding site for the independent domains. Baseline controls were acquired with buffer and pure peptide solutions.
NMR spectroscopy 1 H, 15 N HSQC spectra were recorded at 25ºC on a Bruker 600 MHz spectrometer equipped with a cryoprobe. 15 N-labelled PD-1 (60 µM) was prepared in NMR buffer (25 mM potassium phosphate, 100 mM NaCl, 10% D 2 O, pH 6.4). Spectra were acquired with 180x2048 complex points with a total of 48 transients per increment. 1 H chemical shifts were referenced to internal DSS (4,4-dimethyl-4-silapentane-1-sulfonic acid). Data processing was performed using TopSpin v3.0. Assignment of the human PD-1 1 H, 15 N HSQC spectrum has been previously reported. 3 Relative signal intensity changes (I/Io) due to the addition of macrocyclic peptides (1 mM) were plotted for each protein residue, I/Io > 20% were considered significant.
To assess the solution structures of PD-i3 and PD-i6, NMR spectra were recorded at 5ºC on a Bruker B800 MHz spectrometer equipped with a cryoprobe. The NMR sample was prepared by dissolving the peptide in NMR buffer at a final concentration of 1 mM. 1 H chemical shifts were referenced to internal DSS. Data processing was performed using TopSpin v3.0. Complete proton and carbon resonance assignment was obtained by the combined analysis of 2D homo-(TOCSY, NOESY) and natural abundance hetero-nuclear 1 H-13 C HSQC experiments. The TOCSY and NOESY mixing times were 70 and 200 ms, respectively. Suppression of the water signal was achieved by excitation sculpting. C α and H α secondary chemical shifts (Δδ) were calculated as the difference between the measured chemical shift (δ measured ) and reported values for random coil (δ RC ). 4

NMR structural determination
To calculate the structure of PD-i3 and PD-i6 in solution, simulated annealing calculations were performed with the Xplor-NIH software, 5 which can handle L-and D-isomers and cyclic backbones. Distance restraints derived from NOEs observed in the 1 H, 1 H-NOESY spectrum were sorted into strong (2.5 0.7 2.0), medium (3.0 1.2 2.0) and weak (4.0 2.2 2.0), according to their relative intensities. A square-well potential was used to model these restraints. Amide proton temperature coefficients were calculated to infer the presence of backbone hydrogen bonds. 6 Dihedral angles were left unrestrained due to the lack of sufficient information for handling D-amino acids. A total of 1000 structures were calculated and sorted by total energy. The refined coordinates for the ensemble of 10 lowest-energy conformations are available at the Protein Data Bank under accession codes 6TVJ (for PD-i3) and 6TT6 (for PD-i6).

Molecular dynamics (MD)
Unrestrained MD simulations were performed with the Amber14 software using the ff14SB forcefield. The lowest energy structure of the NMR-calculated ensemble was solvated using the Leap module in a pre-equilibrated octahedral box of TIP3P water molecules. Chlorine or sodium ions were added to obtain an electrostatically neutral system. The initial complex structure was first subjected to a minimization protocol consisting of 1000 steps of steepest decent method followed by 500 steps of conjugate gradient method. Thermalization of the system was performed in the NVT ensemble during 200 ps, using a time step of 1 fs and increasing the temperature from 100 to 298 K, where a force constant of 5 kcal mol -1 Å -2 was applied to protein backbone atoms. Prior to the production run, a short MD simulation (100 ps) in the NPT ensemble was done in order to equilibrate the system density to 1 atm and 298 K. From each equilibrated system, 3 simulations of 100 ns were performed at constant pressure (1 atm) and temperature (298 K) using periodic boundary conditions. Low harmonic constraints (2 kcal mol -1 Å -2 ) were used to reduce the protein mobility. Constant temperature was achieved using the Langevin thermostat with a collision frequency of 2 ps -1 . The SHAKE algorithm was used to keep bonds involving hydrogen atoms at their equilibrium length. The particle-mesh Ewald summation method was used to deal with long range electrostatic interactions and a cut-off of 10 Å was applied for non-bonded interactions. Frames collected every 2 ps were analyzed using the CPPtraj module of Amber. The root-mean-square deviation (RMSD) of the position of the Cα of the peptides was calculated for each frame and compared to the target design.