N-domain of angiotensin-converting enzyme hydrolyzes human and rat amyloid-β(1-16) peptides as arginine specific endopeptidase potentially enhancing risk of Alzheimer’s disease

Alzheimer’s disease (AD) is a multifactorial neurodegenerative disorder. Amyloid-β (Aβ) aggregation is likely to be the major cause of AD. In contrast to humans and other mammals, that share the same Aβ sequence, rats and mice are invulnerable to AD-like neurodegenerative pathologies, and Aβ of these rodents (ratAβ) has three amino acid substitutions in the metal-binding domain 1-16 (MBD). Angiotensin-converting enzyme (ACE) cleaves Aβ-derived peptide substrates, however, there are contradictions concerning the localization of the cleavage sites within Aβ and the roles of each of the two ACE catalytically active domains in the hydrolysis. In the current study by using mass spectrometry and molecular modelling we have tested a set of peptides corresponding to MBDs of Aβ and ratAβ to get insights on the interactions between ACE and these Aβ species. It has been shown that the N-domain of ACE (N-ACE) acts as an arginine specific endopeptidase on the Aβ and ratAβ MBDs with C-amidated termini, thus assuming that full-length Aβ and ratAβ can be hydrolyzed by N-ACE in the same endopeptidase mode. Taken together with the recent data on the molecular mechanism of zinc-dependent oligomerization of Aβ, our results suggest a modulating role of N-ACE in AD pathogenesis.

In the present study, in order to determine the effect of termini protection on the hydrolysis of Aβ  (Table 1). Each peptide (40 μM) was incubated in two different buffer systems (see the section 2.4.) at 37 °C with N-ACE for 10-40 min. Additionally, these reactions were performed in the presence of lisinopril (10 μM) known as a specific inhibitor of ACE enzymatic activity. Samples from all of the reaction mixtures were subjected to direct MALDI-TOF MS analysis in order to identify the reaction products.

Molecular modeling of complexes of Аβ-derived substrates with the active center of N-ACE
supports the role of N-ACE as an arginine endopeptidase towards Аβ species. We have shown that N-АСЕ demonstrates endoproteolytic activity by cleaving the Arg-His bond in C-amidated Аβ and ratАβ MBDs irrelevant of the bond position whether 5-6 (in human) or 13-14 (in rat). To get more insight into the molecular mechanism of N-ACE endoproteolytical activity, the complexes of N-ACE with tetrapeptides corresponding to several fragments of Аβ and ratАβ MBDs have been modelled.
The active site of N-ACE (the structure of the C-domain of ACE is very similar) is a large channel with a constriction in the middle, which divides the channel into two chambers like in a sand-glass with a catalytic Zn 2+ in the center 52 . The active site is quite large and can accommodate several amino acids in both parts. Since it is difficult to correctly model the complexes of N-ACE with long peptides (Аβ(1-16) or ratАβ(1-16)), in this study tetrapeptides 4FRHD7 (h4_7) and 12VHHQ15 (h12_15) of Аβ and 4FGHD7 (r4_7) and 12VRHQ15 (r12_15) of ratАβ have been used to model the behaviour of (Аβ(1-16) and ratАβ(1-16) as substrates for N-ACE. We have implemented molecular dynamic simulation to probe the stability of Michaelis complexes for N-ACE with h4_7, h12_15, r4_7, and r12_15 substrates in the N-ACE active site and figure out the possible reasons for the abolishment of catalytic activity, associated with R5G, Y10F, and H13R substitutions by which Аβ differs from ratАβ 50 .
All four systems were stable along the course of the 100 ns molecular dynamic simulation and the tetrahedral zinc coordination has been retained (Supplementary Table S3). Peptides h4_7, h12_15 and r12_15 have demonstrated similar conformational behavior and interactions with the N-ACE active site (Figs 3 and 4A-C, Supplementary  Table S5). These three tetrapeptides adopted an extended backbone conformation, which has been stabilized by hydrogen bonds with main-chain atoms of β-sheet N-ACE residues A332 and A334 and side chains of H331, H491 and Y501. This behaviour is in line with the experimental and theoretical studies of ACE complexes with known peptide substrates [53][54][55] . Constructs h4_7 and r12_15 demonstrate more than 89% populations of the key contacts, stabilizing the scissile bond (R5 O -Y501 OχHχ, H6 NH -A332 O for h4_7 and R13 O -Y501 OχHχ, H14 NH -A332 O for r12_15, Supplementary Table S5). The h12_15 construct reveals ca. 30% decrease of the H13 O -Y501 OχHχ contact population as compared to h4_7 and r12_15. The side chain of an arginine residue, preceding the scissile bond, interacted with the carboxyl group of D43 and amide group of N494 of N-ACE. Polar groups of C-terminal amino acids of h4_7, h12_15 and r12_15 formed hydrogen bonds with N-ACE residues Q259, K489 and Y498. The hydrogen bond, linking the side chain of the N-ACE catalytic residue E362 and zinc-coordinating water molecule, was stable along the whole simulation. It is interesting to note, that higher flexibility of the arginine side chain, preceding the scissile bond, as compared to the bulky histidine imidazole ring, results in the weaker stabilization of the N-terminus and decreased populations of the H6 O-H331 Nε2 Hε2 and H14 O -H331 Nε2 Hε2 contacts for h4_7 and r12_15 peptides respectively as compared to analogous contacts of h12_15 (Supplementary Table S5). However, the population of H14 O -H491 Nε2 Hε2 hydrogen bond was lower for h12_15 peptide.
The R5G substitution significantly changes the conformational behaviour of the tetrapeptide r4_7 (Fig. 3) and results in the increased backbone motility in the region of the scissile bond (Fig. 5). The characteristic peptide stabilization by hydrogen bonding with main chain atoms of A332 and A334 and side chains of H331, H491 and Y501 of N-ACE breaks down along the coarse of simulation (Fig. 4D, Supplementary Table S5). The peptide adopts a distorted extended conformation, where the position of the peptide bond between G5 and H6 residues moves along the N-ACE tunnel towards the C-terminus. The shift of the peptide position in the catalytic center is reflected by the formation of two new polar contacts: 1) a statistically significant hydrogen bond between the side chain of H331 and the backbone carbonyl oxygen of F4 instead of H6, and 2) a hydrogen bond between the hydroxyl group of Y501 and the backbone carbonyl oxygen of the N-terminal capping group instead of the backbone carbonyl oxygen of G5, which is observed by the end of the trajectory (Fig. 4D). Thus, the R5G substitution destabilizes the Michaelis complex of ratAβ fragment 4-7 with N-ACE. This explains the absence of endopeptidase activity toward the G5-H6 peptide bond of ratАβ(1-16).    The stability of the modelled complex between N-ACE and r12_15 correlates with the observed hydrolysis of bond Arg13-His14 in ratAβ (1)(2)(3)(4)(5)(6)(7)(8)(9)(10)(11)(12)(13)(14)(15)(16). However, the Аβ fragment 12VHHQ15 which is not hydrolyzed by N-ACE also forms a well-stabilized complex in the active site of the enzyme as well as ratАβ fragment 12VRHQ15. The lack of hydrolysis of C-amidated Аβ (1)(2)(3)(4)(5)(6)(7)(8)(9)(10)(11)(12)(13)(14)(15)(16) species by N-ACE can be explained by the influence of the Y10F substitution on the process. Indeed, the scissile bond of angiotensin I links phenylalanine and histidine residues, meaning that the P1 position in the N-ACE active site is well suited for bulky aromatic side chains, like the imidazole ring. The Y10F substitution appears as the fourth residue toward the N-terminus from the scissile bond R13-H14 of 12VRHQ15. The substrate tunnel of N-ACE forces an extended conformation on the ligand peptide, where each residue occupies a distinct pocket 53 . The interactions within these pockets govern substrate specificity of the enzyme 53 . Thus, the replacement of phenylalanine 10 by tyrosine which carries a hydroxyl group on the side-chain benzene ring can result in the destabilization of the position of the 12VHHQ15 substrate in the pocket and can lead to the loss of the catalytic activity toward the H13-H14 peptide bond of Аβ (1)(2)(3)(4)(5)(6)(7)(8)(9)(10)(11)(12)(13)(14)(15)(16). In line with these considerations, several known peptide substrates of N-ACE have residues with significantly different from tyrosine shapes of side chains at fourth position toward N-terminus from the scissile bond (Supplementary Table S4).
The switch from the usual for ACE carboxypeptidase activity to the endoproteolytic one may be due to the specificity of the Аβ structure. As it was shown in this study, the endoproteolytic activity was observed only for peptides with a blocked carboxylic group at the C-end, i.e. for those without a negative charge in this crucial for ACE recognition region, and at the same time, the first amino acid in the Аβ peptide is an aspartic acid which carries a free carboxylic group (moreover, another negatively charged carboxylic amino acid, glutamate, is found in the third position). Thus in case of Аβ, the negative charge at its C-end is absent, but instead a negative charge is present on the N-end of Aβ(1-16)-[Amide]. This leads to an error in the recognition mechanism of ACE, and the enzyme instead of the C-terminal carboxylic group binds to the side chain group of Asp1 at the N-terminus. This error probably occurs at the entrance to the channel of the active site of the N-domain, where notable differences in hydrophobicity and charge are observed in the lid-like structure comprising of helices α1, α2 and α3 56 .
The probability of this assumption is also confirmed by the structure of the unique natural substrate of ACE toward which endoproteolitic activity of ACE was demonstrated, a regulatory peptide, luliberin, (gonadotropin releasing hormone, GnRH or LHRH) 57 , from which the N-domain of ACE cleaves an N-terminal tripeptide. This hormone is synthesized in the organism with a modified C-terminal amino-acid residue (Pyr-HWSYGLRPG-[Amide]) and a negatively charged pyroglutamate residue at the N-terminus. This unusual structure of luliberin with a blocked C-terminal carboxylate and a negatively charged N-terminus is similar to that of Aβ(1-16)-[Amide]. Thus, a common binding mechanism for both of these substrates by N-ACE, in which the N-terminus of the peptide imitates the C-end of a typical ACE substrate, can be assumed. In regard with this result, it is interesting to search for new ACE substrates, towards which the enzyme could also demonstrate its endoproteolytic action, among peptides and proteins whose sequence begins with negatively charged amino acid residues like aspartate and glutamate.

Hypothesis: N-ACE aggravates the course of AD through generating isoAβ(6-x) species.
Observational studies indicate that increased activity of ACE [21][22][23]58 , as well as inhibition of interactions between ACE and Аβ 33-37 , appear to be important for modulating AD, but the molecular mechanism of action of ACE on the development of the disease remains unknown. In the current study, we have shown that C-amidated peptides corresponding to the metal binding domains of human and rat Аβs are efficiently cleaved at the Arg-His bonds (Arg5-His6 and Arg13-His14, respectively) by the N-domain of ACE, which acts as an arginine specific endopeptidase. Our data also shows that C-terminal amidation is necessary and sufficient for such N-ACE action on these Aβ species. Molecular modelling has demonstrated that these Aβ substrates enter the active site of N-ACE with their N-termini. Since the N-terminal residues 1-16 form an independent folding unit in the full-length Aβ 45,48,[59][60][61]  Rats and mice are invulnerable to AD-like pathologies 49,50 , but for human beings and all other mammalians which suffer of AD, limited hydrolysis of Aβ by N-ACE resulting in the formation of Аβ(6-x) species may have dangerous consequences. Structurally modified Aβ molecules initiate AD-linked amyloidogenesis of endogenous Aβ in animal models 6 probably through the aggregation seed mechanism 62 . One of such potential seeding agents is supposed to be Aβ carrying the isomerized Asp7 residue (isoAβ) 63,64 . IsoAβ appears to be involved in the AD pathogenesis by means of its zinc-dependent interactions with endogenous Aβ resulting in the formation of zinc-bound heterodimeric seeds causing Aβ aggregation 65 .
Results from our recent study suggest that removal of the N-terminal region 1-5 from Aβ and isoAβ enhances the ability of respective N-truncated Aβ(6-x) and isoAβ(6-x) species to form zinc-mediated oligomers 66 . It is worth noting that isoAβ is cleaved by N-ACE much more efficiently than native Aβ 32 , and at the same time isoAβ(6-x) is immensely more susceptible to zinc-driven oligomerization 66 . Thus, inhibitors of ACE should mainly suppress the formation of isoAβ(6-x) species, what could explain the positive effect of these inhibitors on patients with AD [33][34][35][36][37] and the slowing of neurodegeneration in animal AD models [38][39][40] .
Translating the role of isoAβ as a trigger of amyloidogenesis in AD animal models 63,64 for human patients and taking into account above mentioned considerations, we have assumed the following scenario of N-ACE linkage to AD: (i) in a healthy organism endoproteolytical cleavage of native Aβ at the Arg5-His6 bond is quite rare and a rather normal processing event; (ii) when isoAβ species are formed (for example, due to Aβ ageing, neurotrauma, etc), a rapid limited hydrolysis of these species by N-ACE results in the formation of isoAβ(6-x) molecules which are extremely susceptible to zinc-induced oligomerization and by this reason should significantly enhance the SCIEnTIFIC REPORTS | (2018) 8:298 | DOI:10.1038/s41598-017-18567-5 pathological aggregation of endogenous Aβ. This scenario, on one hand, supports the amyloid cascade hypothesis of AD, and, on the other hand, for the first time links together several molecular agents such as Aβ, isoAβ, zinc ions, and ACE, in a potentially pathogenic network.
In summary, the presented study showed that N-ACE specifically cleaves synthetic C-amidated peptide analogs of the metal-binding domains of Aβ and ratAβ at Arg-His bonds 5-6 and 13-14, respectively. Computer modeling provided evidence that these peptides enter the active site of N-ACE with their N-termini, thus assuming that full-length Aβ and ratAβ molecules should be hydrolyzed by ACE in the same way as the C-amidated peptides under the study. Concerning the possible clinical applications, our results indicate that N-ACE seems to play an aggravating role in AD pathogenesis by generating extremely susceptible to zinc-induced oligomerization isoAβ(6-x) species, and thus N-ACE inhibitors should slow down AD progression.

Mass spectrometry (MS).
Due to the low complexity of the studied system -only one highly purified peptide-substrate and enzyme per sample -high mass-accuracy and MS/MS confirmation were not necessary for reliable identification of the reaction products, while for quantitative measurements fast sample analysis procedure and low sample and H 2 O 18 consumption were required, thus it was decided to use Bruker Microflex MALDI TOF instrument (Bruker Daltonics, Germany) for the study. Mass spectra were acquired in a positive-ion reflector mode, 200-500 laser shots were summed per spectrum. To prepare the matrix solution, HCCA was dissolved to a concentration of 10 mg/mL in acetonitrile /0.1% TFA (70:30 v/v). Usually, for MALDI probe preparation, the dried-droplet method was used: 0.5 μL of 2% TFA was mixed with 0.5 μL of the sample (0.5-2 pmol per target) and 0.5 μL of the matrix solution, then loaded onto a MALDI sample plate and measured by MS.
Quantitative determination of ACE digestion products using 18 O-labeled internal standards. A method for quantitating the products of enzyme degradation has been based on the use of MALDI-TOF MS with internal 18 O-labeled standards. A simple procedure allows to produce such internal standards for the tested sample by enzymatic hydrolysis of the same sample (of a known concentration) in 18 O-water as described earlier 68 . Briefly, to prepare the 18  25 µL of 18 O-water solution containing 20 µM of an appropriate peptide, 50 mM of ammonium bicarbonate (pH 7.8), and 1 µg of trypsin. In order to completely hydrolyze the substrate the reaction was incubated for 48 h, and then the sample was kept at −20 °C until analysis. To obtain the final standard solution, 5 µL of the terminated reaction mixture were added to 45 µL of the matrix solution (see previous section). For quantitation assay, 5 µL of the final standard solution were mixed with an equal volume of an ACE digestion mixture pre-incubated for 10, 20, 40 and 60 min, then, 1 μL of the resulting mixture was applied directly onto the MALDI target plate and subjected to MALDI-TOF MS analysis to obtain the isotopic pattern of the corresponding analyte/internal standard mixture. The previously described algorithm 68 was used to calculate the absolute concentration of the peptide of interest on the basis of experimentally determined isotopic patterns of the analyte and the 18 O-labeled standard (of a known concentration) and of the analyte/internal standard mixture. The method error was estimated to be less than 10% 51 .

Molecular modelling studies. Modelling of Michaelis complexes of Aβ peptides with N-ACE and force-field
parameterization. The models of the Michaelis complex have been constructed for the N-domain of ACE, bound with four tetrapeptide fragments of Aβ (4FRHD7 and 12VHHQ15) and ratAβ (4FGHD7 and 12VRHQ15). All tetrapeptides were acetylated at the N-terminus and amidated at the C-terminus. The models have been build using the crystallographic structure of N-domain of somatic ACE with lisinopril, zinc ion bound in the active site and chlorine ion at Y202/R500 site (PDB code 2C6N). The tetrapeptides have been fitted in the active site tunnel by manual superimposition of the main chain peptide atoms on the corresponding atoms of lisinopril 54 . The lisinopril zinc-coordinating carboxyl group has been replaced by a water molecule. The fitted peptide chains of the obtained models have been minimized using 100 steps of conjugate gradient minimization ( Supplementary  Fig. S6). Modelling has been accomplished using the Chimera software 69 .
The bonded plus electrostatic model has been used to describe zinc chelation 70 . Following the previously published studies of the ACE catalytic mechanism 54,55 , we have assumed a tetrahedral coordination of the zinc ion by the side chains of residues H361, H365, E389 and a water molecule ( Supplementary Fig. S7). The force-field parameters for the zinc-chelating environment have been derived using ab-initio calculations in Gaussian 09w 71 . The local geometry of the zinc-binding interface has been optimized and force constants and atomic partial charges have been derived following the procedure implemented in the Metal Center Parameter Builder (MCPB) package 72 . The quantum mechanical calculations have been performed at the B3LYP level of theory with the 6-31 G* basis set. The force-field constants have been derived from the Cartesian Hessian matrix by the Seminario method 73 and partial charges have been obtained from the Merz-Singh-Kollman charges using Restrained Electrostatic Potential (RESP) fitting 74 . Calculated force-field parameters are summerized in Supplementary Tables S1 and S2.
Molecular dynamics simulations. The molecular dynamics simulations have been performed using the GROMACS 4.6.5 software package 75 and Amber ff99SB-ILDN force field 76 . The model of N-ACE complexed with a tetrapeptide has been placed in a cubic cell with a minimum distance between the protein and the box of 0.8 nm and solvated using TIP3P water molecules 77 . The total charge has been neutralized by Na + ions. The chlorine ion at the Y202/R500 site of N-ACE was retained. The system was minimized using the steepest descent minimization algorithm. Positions for the protein complex atoms were restrained and the system was equilibrated with 100 ps of constant volume molecular dynamics followed by 100 ps of constant pressure molecular dynamic. The production of 0.1 µs molecular dynamics trajectory has been obtained. Calculations have been done with 2 fs integration steps at a constant pressure of 1 atm and temperature of 300 K using the Berendsen barostat and the velocity rescale method for the thermostat. The particle-mesh Ewald method 78 has been implemented to treat long-range electrostatic interactions and the LINCS algorithm controlled the lengths of covalent bonds 79 . The procedure has been repeated for each of the four modelled complexes. Hydrogen bond population analysis has been done using h-bond utility of GROMACS 4.6.5 75 and in-house written scripts.
Molecular dynamics calculations have been performed using the equipment of the shared research facilities of HPC computing resources at Lomonosov Moscow State University. Structure visualization has been done in PyMOL (Schrödinger, LLC).
Data Availability Statement. The datasets generated during and/or analyzed during the current study are available from the corresponding author on reasonable request.