Presumed LRP1-targeting transport peptide delivers β-secretase inhibitor to neurons in vitro with limited efficiency

Interfering with the activity of β-secretase to reduce the production of Aβ peptides is a conceivable therapeutic strategy for Alzheimer’s disease. However, the development of efficient yet safe inhibitors is hampered by secondary effects, usually linked to the indiscriminate inhibition of other substrates’ processing by the targeted enzyme. Based on the spatial compartmentalization of the cleavage of the amyloid precursor protein by β-secretase, we hypothesized that by exploiting the endocytosis receptor low-density lipoprotein receptor-related protein it would be possible to direct an otherwise cell-impermeable inhibitor to the endosomes of neurons, boosting the drug’s efficacy and importantly, sparing the off-target effects. We used the transport peptide Angiopep to build an endocytosis-competent conjugate and found that although the peptide facilitated the inhibitor’s internalization into neurons and delivered it to the endosomes, the delivery was not efficient enough to potently reduce β-secretase activity at the cellular level. This is likely connected to the finding that in the cell lines we used, Angiopep’s internalization was not mediated by its presumed receptor to a significant extent. Additionally, Angiopep exploited different internalization mechanisms when applied alone or when conjugated to the inhibitor, highlighting the impact that drug conjugation can have on transport peptides.

ANG-SI systems. Convergence was considered to be reached when the free energy value reached a reasonable plateau with respect to the simulation time.

Metadynamics simulations
Metadynamics simulations allow to reconstruct the free energy of the system by keeping track of the sampling of conformations and discouraging the sampling of structures already visited during the simulations, as a function of few conformational degrees of freedom (collective variables). 2,3 Calculations were performed adopting the approach discussed by Deighan et al., who combined metadynamics and parallel tempering (PTMetaD) in the well-tempered ensemble (PTMetaD-WTE) in order to improve simulation efficiency. 4 All simulations were performed with GROMACS 5.0.2 5 patched with PLUMED 2.1.0 plugin 6 . Linear ANG and ANG-SI structures were built using tleap module implemented in AmberTools package. Each peptide was solvated with about 14500 TIP3P water molecules; 7 Na+ and Cl-ions (whose parameters were taken from Joung and Cheatham 8 ) were added in order to assure electroneutrality and to mimic PBS environment. The ff14SB force field parameters were employed for ANG and ANG-SI; since statin is not a standard residue in ff14SB libraries, atomic charges were derived with a proper protocol (vide infra). 9 The starting system for PTMetaD-WTE simulations was obtained according to the following protocol.
First of all, energy minimization was carried out in order to remove bad solvent/solute and solvent/solvent contacts. Then temperature was raised from 0 to 310 K through 200 ps in NVT ensemble, applying a weak harmonic restraint on the solute in order to avoid wild fluctuations. Finally, the system was equilibrated through 1 ns simulation in NPT ensemble at 310 and 1 atm.
Velocity rescale algorithm and Parrinello-Rahman barostat 10,11 were used in order to maintain temperature and pressure at the desired values. Electrostatic interactions were computed by means of Particle Mesh Ewald (PME) method using a cutoff value equal to 1.2 nm; the same cutoff was employed for Van Der Waals interactions. 12 Neighbor list was updated every 5 fs; all covalent bonds involving hydrogen were restrained using LINCS algorithm. 13 PTMetaD-WTE simulations were performed using 12 replicas, whose temperature ranges from 310 to 588 K; before starting the simulations, 1 ns molecular dynamics simulation were carried out for each replica, so that every system could equilibrate according to its temperature and adjacent replicas exhibit different peptide conformations. A two-step protocol was used for PTMetaD-WTE simulations. First of all, a 1 ns PTMetaD simulation was performed biasing only potential energy, chosen as collective variable. 14 Gaussian were deposited every 0.5 ps, with an height value equal to 0.28 kcal mol -1 , a width value equal to 956 kcal mol -1 and a bias factor equal to 24. In the following step, the radius of gyration related to α-carbon atoms in the protein backbone (Rg) and the intrapeptide hydrogen bonds (Hb) were chosen as collective variables, since they proved to be suitable for the sampling of protein structures in water solutions, 4,15-17 and biased using the well-tempered metadynamics algorithm. 3 No more Gaussians were added in order to bias potential energy, since the cumulative bias was employed as a static additional bias. For what regards radius of gyration and intrachain hydrogen bonds, Gaussians were added every 1 ps, using a height value equal to 0.28 kcal mol -1 , width values equal to 0.01 and 0.1 for Rg and Hb respectively and a bias factor equal to 8.
Temperature control and long-range interactions were treated as described above. Simulation times equal to 200 ns and 300 ns were used for each replica for ANG and ANG-SI respectively, leading to a total simulation time of 2.4 μs for ANG and 3.6 μs for ANG-SI. Exchange attempts between replicas were carried out every 0.4 ps by means of PTMetaD algorithm implemented in

Molecular dynamics simulations
Molecular dynamics simulations were performed adopting the same protocol described above for obtaining the starting system for metadynamics simulations. Folded ANG and ANG-SI were solvated with about 14500 TIP3P water molecules; Na+ and Cl-ions were added in order to assure electroneutrality and mimic PBS environment. After energy minimization, temperature was raised from 0 to 300 K in NVT ensemble. The system was then equilibrated for 1 ns in NPT ensemble in order to reach the correct density, and then 50 ns molecular dynamics simulations were performed at 310 K and 1 atm. Temperature and pressure were kept at the desired values by means of velocity rescaling algorithm and Parrinello-Rahman barostat respectively. Long-range interactions were treated as described. The same force fields employed for metadynamics simulations were employed.

Statin atomic charges
Statin atomic charges were derived adopting a calculation protocol consistent with force field design. 9,18 The initial statin structure was taken from crystallographic structure 3LIY in pdb database; the molecule of interest is here included in a small peptide, and thus constitutes a consistent initial guess for charge derivation. Statin was capped with acetyl (ACE) and N-methyl (NME) groups; the obtained structure was firstly minimized through ab initio calculations at MP2/6-31G* level of theory. Electrostatic potentials were then computed from the optimized structure at HF/6-31G* level of theory. Atomic charges were then fitted adopting RESP formalism starting from electrostatic potentials. 19 A two-step protocol was adopted: first, atomic charges were fitted assigning an overall charge for statin equal to 0, while ACE and NME atomic charges were taken from force field library. In the second step, charge equivalence for chemically equivalent atoms was imposed.

Trajectories post-processing
Electrostatic potentials were computed with Adaptive Poisson Boltzmann Solver (APBS). 20 Electrostatic potential is expressed in kBTe -1 units, where kB is Boltzmann constant, T is the absolute temperature and e is the electron charge. Average solvation free energy was computed through MMPBSA as implemented in AmberTools package. 21