Engineering Neprilysin Activity and Specificity to Create a Novel Therapeutic for Alzheimer’s Disease

Neprilysin is a transmembrane zinc metallopeptidase that degrades a wide range of peptide substrates. It has received attention as a potential therapy for Alzheimer’s disease due to its ability to degrade the peptide amyloid beta. However, its broad range of peptide substrates has the potential to limit its therapeutic use due to degradation of additional peptides substrates that tightly regulate many physiological processes. We sought to generate a soluble version of the ectodomain of neprilysin with improved activity and specificity towards amyloid beta as a potential therapeutic for Alzheimer’s disease. Extensive amino acid substitutions were performed at positions surrounding the active site and inner surface of the enzyme and variants screened for activity on amyloid beta 1–40, 1–42 and a variety of other physiologically relevant peptides. We identified several mutations that modulated and improved both enzyme selectivity and intrinsic activity. Neprilysin variant G399V/G714K displayed an approximately 20-fold improved activity on amyloid beta 1–40 and up to a 3,200-fold reduction in activity on other peptides. Along with the altered peptide substrate specificity, the mutant enzyme produced a markedly altered series of amyloid beta cleavage products compared to the wild-type enzyme. Crystallisation of the mutant enzyme revealed that the amino acid substitutions result in alteration of the shape and size of the pocket containing the active site compared to the wild-type enzyme. The mutant enzyme offers the potential for the more efficient degradation of amyloid beta in vivo as a therapeutic for the treatment of Alzheimer’s disease.


Introduction
Neprilysin, or neutral endopeptidase (NEP), is an integral type II membrane-bound zinc-dependent peptidase of approximately 750 amino acid residues [1] that degrades a number of physiological peptides that are involved in processes such as blood pressure regulation and nociception. NEP is composed of an ectodomain, which contains the catalytic site and belongs to the M13 family of peptidases/proteases, a transmembrane domain and a short intracellular domain. The structure of the ectodomain is composed of two largely a-helical domains (domains 1 and 2) that are arranged to form a central spherical water-filled core that contains the active site of the enzyme. The larger N-terminal domain (domain 1), which is structurally related to the bacterial protease thermolysin, contains a single zinc atom that is critical for peptidase activity [2][3][4] and is coordinated by His and Glu residues [5]. Whilst the substrate specificity of the enzyme is quite broad, NEP has a strong preference for peptides over larger proteins. This specificity seems to result from the enclosed catalytic chamber and size-restricted access to that chamber. Within peptide sequences cleavage is usually at the N-terminal side of a hydrophobic residue, with a strong preference for Phe or Leu at the P1' position [6] and corresponds with a deep lipophylic cavity at subsite S1'. While structures of NEP in complex with different inhibitors display similar main-chain conformation, large ligand induced flexibility is observed for the side chains that form the substrate binding pockets (such as Arg 102 , Phe 106 , Arg 110 and Trp 693 ), which may explain the broad substrate specificity of the enzyme [5,7,8].
Amongst the peptides cleaved by NEP are a variety of physiologically relevant peptides such as enkephalins, tachykinins, and natriuretic peptides [9], where NEP may play a role in their degradation to limit activity: In hypertensive rats inhibition of NEP results in a lowering of mean arterial blood pressure and an increased natriuresis suggesting increased activity of natriuretic peptides [10]. Similarly, inhibition of NEP in normal rats results in a potentiation of exogenously applied atrial natriuretic peptide (ANP) activity, increasing natriuresis and lowering of blood pressure [11]. The antinociceptive activity of endogenously administered enkephalins could also be increased through blockade of NEP activity [12]. NEP has also been implicated in lipid and glucose metabolism, and suppression of NEP activity by genetic or pharmacologic means leads to late-onset obesity in mice [13], suggesting a role in regulating orexigenic peptides such as neuropeptide Y, a known NEP substrate [14].
NEP has attracted interest as a potential therapeutic to treat Alzheimer's disease through its ability to degrade the peptide amyloid beta (Ab) [15,16]. In animal models manipulation of the levels of brain NEP has been shown to have a significant effect on Ab levels. Mice deficient in the NEP gene show defects in their ability to degrade both endogenous and exogenously applied Ab [17] and inhibition of NEP using thiorphan or phosphoramidon results in accumulation of Ab in rats [18], rabbits [19] and nonhuman primates [20]. Over-expression of NEP in the brain through transgenic [21] or viral transduction [22][23][24] results in a lowering of Ab levels in the brain. However, because of its broad substrate specificity, and known effects on blood pressure, increasing the levels of NEP through regular dosing to degrade Ab has the theoretical risk of increasing the degradation of other physiological peptides and generating unwanted side effects. It would therefore be preferable to work with an enzyme that has higher specificity and/or activity on the desired substrate. It has previously been demonstrated that NEP is amenable to protein engineering to alter its activity and specificity [25]. Mutation of two key residues located in the substrate binding pockets adjacent to the active site resulted in NEP variants that had altered specificity towards Ab, insulin-b chain and leu-enkephalin. However, these alterations in specificity were also accompanied by an overall reduction in the catalytic activity of the enzyme. To generate a variant of NEP that may have greater therapeutic use it would be preferable to increase its activity at the desired substrate, Ab, and reduce activity at peptides where the potential for unwanted side effects exist.
We performed extensive mutagenesis on the solvent accessible core of the protein and screened the resulting NEP variants for their ability to degrade Ab1-40, Ab1-42 and a selection of other physiological peptides. We identified a panel of variant NEP molecules with increased activity in cleavage of Ab1-40 and Ab1-42 and reduced activity on a panel of other peptide substrates. We selected one variant for further analysis and demonstrate it has improvement in its ability to degrade Ab1-40 and 1-42 and up to 3,000-fold reduced activity on the other peptides. Further characterisation revealed a change in the cleavage sites within Ab1-40 and an altered sensitivity to inhibitors. We crystallised this variant and showed that long range structural changes combined with changes in the shapes of the substrate binding pockets, are likely to be responsible for the alterations in substrate specificity. The significantly enhanced Ab cleavage and reduced activity on other peptides make this variant an attractive molecule for the potential treatment of Alzheimer's disease.

Cloning, expression and purification of NEP variants
To generate NEP variants for screening for improved activity and specificity on Ab, DNA encoding the extracellular domain (amino acids 52-749) of wild type human NEP was cloned into the yeast expression vector pYES2. The expression construct was assembled to have the mating factor alpha leader, a triple HA-tag and a Gly-Ser dipeptide linker at the N-terminus of NEP. To attempt to alter the activity and specificity of NEP 134 amino acids lining the active site were identified from the crystal structure [5] and a library of variants was generated by site-directed mutagenesis at each of these positions. In subsequent rounds of screening, mutations that resulted in improved activity on Ab and/or improved specificity over other peptides were combined.
To screen for improved specificity towards Ab, wild-type NEP and the variants were expressed in Saccharomyces cerevisiae YMR307w (EUROSCARF) cultured in SC medium (yeast nitrogen base; Becton Dickinson) or CSM-Ura medium (MPBio) supplemented with 0.5% (w/v) casein hydrolysate, 200 mM HEPES (pH 7.0) and 2% (w/v) galactose for induction of expression for 55-70 h at 30uC. CSM-Ura medium lacks uracil to enable selection of strains carrying the expression construct. Purification of HA-tagged NEP variants was performed by immunoaffinity chromatography with a monoclonal antibody specific for the HA-tag [26]. Eluted proteins were bufferexchanged into 50 mM HEPES (pH 7.0) containing 300 mM NaCl.
Selected NEP variants were expressed in mammalian cell culture to allow their kinetic characterisation. DNA encoding amino acids 52-749 of human NEP variants was amplified by PCR. Sequences encoding an N-terminal 10-His tag and the human immunoglobulin heavy chain leader sequence were added by successive rounds of PCR and the resulting constructs were cloned into pENTR (Invitrogen) before being further sub-cloned into pDEST12.2 (Invitrogen) for expression in mammalian cells.
10-His NEP variants were expressed transiently in CHO suspension cells and secreted into the medium. Expression constructs were used to transfect cells using Polyethylenimine Max (Polysciences) and cells were cultured in CD-CHO (Invitrogen) for 10 days at 37uC, 5% CO 2 and 80% relative humidity with shaking at 140 rpm. The supernatant was harvested, concentrated, and exchanged into 26phosphate buffered saline (PBS) (2.2 mM KH 2 PO 4 , 310 mM NaCl, 5.9 mM Na 2 HPO 4 , pH 7.4) using cross-flow dialfiltration with a 30 kDa membrane. 10His-NEP proteins were purified first by nickel affinity chromatography using a 5 ml HisTrap excel column (GE Healthcare), followed by size exclusion chromatography using a Superdex 200 HiLoad 16/ 60 (GE Healthcare) equilibrated with PBS (1.1 mM KH 2 PO 4 , 155 mM NaCl, 3.0 mM Na 2 HPO 4 , pH 7.4) and eluted with an isocratic flow. Protein concentration was measured by absorbance at 280 nm using a UV Vis spectrophotometer and purity was estimated using SDS PAGE. Typically the yield was 2-20 mg/L of pure protein (.95% pure).
To produce protein for crystallisation, N-terminally His-tagged NEP G399V/G714K (NEPv) was cloned into PICZaA and expressed in Pichia pastoris strain GS115 his4 His-, Mut+ (Invitrogen). A clonal line strongly expressing NEPv was selected for fermentation in BMGY medium supplemented with 12 mL/L PMT1 salts and 0.04 g/L L-histidine or basal salt minimum (BSM) medium (1.7 mM CaSO 4 , 26 mM K 2 SO 4 , 15 mM MgSO 4 , 5% (w/v) glycerol, 24 mM (NH 4 ) 2 SO 4 , 12 g/L Nahexametaphosphate, 12 ml/L PMT1 salts, 0.04 gr/L L-histidine, and antifoam). Fermentors were inoculated at OD of 1 and grown at 30uC with the addition of L-histidine at 10 g/L/day. After intitial glycerol depletion, a constant glycerol feed was applied at ,18 mL/h/L until cell density reached approximately 200 g/L. Subsequently, a three hour starvation period was applied, during which the temperature was gradually reduced to 20uC. Protein production was induced by the addition of methanol feed at 2 mL/L/h, which was increased to 3.5 mL/L/h after 3 h. The supernatant containing secreted NEPv protein was harvested after 24 hrs (BMMY medium) or 48 hrs (BSM medium). NEPv was purified by nickel affinity chromatography using a 5 mL His Trap HP column and dialysed into 18 mM KPO 4 (pH 6.0) containing 500 mM NaCl. The protein was deglycosylated using 833 Units EndoH (New England Biolabs) per mg NEPv with overnight stirring at 4uC, followed by size-exclusion chromatography using a Superdex200 prep grade 26/600 column. Three main species of NEPv (82002, 82204 and 82408 Da) were observed by mass spectrometry.

Peptide cleavage assays
To screen NEP variants for improved activity on and specificity towards Ab a high-throughput assay was developed where cleavage activity on Ab1-40 or 1-42 and eight off-target peptides (see Table S1) was determined by fluorescence polarisation. Peptide substrates were labelled with DY-505 or DY-647 at the Nterminus and biotin at the C-terminus. The biotin serves for increasing the molecular size of uncleaved peptides after addition of strepavidin, thereby increasing the assay window and measurable signals. Peptide cleavage assays were performed by incubating NEP variant samples in a microtitre plate containing assay solution composed of 60 nM peptide substrate in 50 mM HEPES (pH 7.4) containing 150 mM NaCl and 0.05% Pluronic F-68 (Sigma). Assays were incubated at 37uC for appropriate lengths of time to allow 5 to 90% turnover of substrate, after which reactions were stopped by the addition of an equal volume of 1.2 mM streptavidin (Calbiochem). In the case of ANP and endothelin the stop solution also contained 10 mM dithiothreitol. The extent of peptide cleavage was determined by measuring the fluorescence polaristion of the stopped reactions on a plate reader (Tecan) with an appropriate set up of polarisation filters.
Since the concentrations of each enzyme variant was constant for all measurements and all substrate concentrations were ,, K M values, the relative rate of peptide cleavage, k app , was proportional to k cat /K M . This allowed NEP variants to be ranked based on their relative k cat /K M ratios on Ab and the other peptide substrates.
For detailed kinetic characterisation of peptide cleavage by NEP, a similar assay to that above was used with modifications based on a published method [27]. Peptides were synthesised by Bachem using 9-fluorenylmethyloxycarbonyl solid-phase peptide synthesis and were labelled with a N-terminal 5(6)FAM via a eaminocaproic acid linker and a C-terminal biotin via a lysine linker (Table S1). Purified peptides were supplied as trifluoroacetate salts with a purity $95%. The assay was performed in a 96well microtitre plate and contained 50 mM HEPES (pH 7.4), 100 mM NaCl, 0.05% (w/v) BSA (Buffer A), 1-200 mM peptide substrate and 1-500 nM NEP protein. Reactions were incubated at 37uC before being stopped at various time points between 2 and 360 min by transferring 5 mL aliquots to 245 mL Buffer A containing 2 mM 1,10-phenanethroline (Sigma) and 2 mM avidin (Invitrogen). For peptide substrates containing disulphide bonds, 1 mM tris(2-carboxyethyl)phosphine (Thermo Scientific) was added to the stop solution. The fluorescence polarisation of the resulting solution was measured on a Victor plate reader and the amount of substrate cleaved was determined with reference to substrate-only controls with and without avidin. Initial rates were obtained by linear regression of the linear regions of time courses. Enzyme velocity was plotted as a function of substrate concentration and the Michaelis-Menten equation was used to fit the data, giving the parameters k cat and K M .
Kinetic parameters for cleavage of endothelin-1 by NEP were determined using a stopped assay in which the amount of substrate cleaved was measured using reversed-phase HPLC (RP-HPLC). Endothelin-1 was incubated at 12.5-200 mM in the presence of wild type NEP or NEPv and reactions were stopped after 10 min at 37uC by the addition of an equal volume of 1:1:1 water:acetonitrile:TFA containing 2 mM tris(2-carboxyethyl)phosphine (Thermo Scientific). RP-HPLC spectra were recorded using an Agilent 1260 Infinity system. Injections of 20 mL were made for each sample and analytes were separated by elution on an Agilent Polaris C8-A stationary phase column (4.66100 mm, 3 mm) at 1.5 mL min 21 using a linear binary gradient of 10-90% acetonitrile in water with 0.1% (v/v) TFA over 15 minutes at 40uC. Signal intensity for intact endothelin-1 was monitored at 210 nm and relative peak area was determined by manual integration of the baseline. Kinetic parameters were derived from the activity data as described above.
The effect of inhibitors on NEP activity was determined using the above fluorescence polarisation assay to measure the rate of Ab1-40 cleavage with a substrate concentration of 10 mM, an enzyme concentration of 20 nM and inhibitor concentrations in the range 2 nM to 100 mM. The NEP inhibitors phosphoramidon and thiorphan were purchased from Sigma and were dissolved to 10 mM stock solutions in DMSO before being diluted in Buffer A. Activity values were plotted as percentage of uninhibited control against log 10 [inhibitor] and the log(inhibitor) vs. normalised response in GraphPad Prism was used to fit the data to generate IC 50 values.

Mass spectrometry
Unlabelled Ab1-40 and bradykinin were supplied as TFA salts by Bachem. Cleavage of the peptide substrates was assessed following incubation of wild-type NEP or mutants with 10 mM Ab1-40 or 2 mM bradykinin in 50 mM HEPES (pH 7.4) containing 150 mM NaCl at 37uC. Reactions were stopped at 0, 10, 60 and 360 min by the addition of 0.5% (v/v) trifluoroacetic acid. Stopped cleavage reactions were purified on C18 Zip-tips (Millipore), eluting directly into MALDI matrix solution (10 mg/ ml a-cyanohydroxycinnamic acid dissolved in 50:50:0.1 water:acetonitrile:TFA). Volumes of 1 mL were applied to a MALDI target and air dried. Mass Spectra were acquired on an AB Sciex 4800 ToF/ToF mass spectrometer operating in positive reflector mode with external calibration on a peptide mixture. Peptides appearing in the digests were analysed by MS/MS with collisioninduced dissociation, and identified by manual assignment in comparison with the Ab1-40 peptide sequence.

Crystallization and structure determination of NEP G399V/G714K
Purified NEP G399V/G714K was concentrated to 10 mg/ml in 25 mM Tris-HCl (pH 7.0), 150 mM NaCl and 2 mM MgCl 2 . Phosphoramidon was added to 2 mM from a 50 mM DMSO stock and the protein-ligand complex was incubated on ice for 30 min. Diffraction-quality crystals were obtained by hangingdrop vapor diffusion at 20uC by mixing 1 ml of protein-ligand solution with 1 ml of reservoir solution containing 100 mM HEPES (pH 7.0), 22% (w/v) PEG3350 and 200 mM NaCl. Rod shaped crystals appeared after 2 weeks. Crystal size and appearance was further improved by streak seeding. The crystals were cryo protected in reservoir solution supplemented with 20% glycerol and flash frozen in liquid nitrogen.
Crystallographic data were collected at 100 K to 2.15 Å resolution at the I04-1 beamline at the Diamond Light Source on a ADSC Q315 CCD detector and processed with MOSFLM [28]) and SCALA [29]. The space group was P3 2 21 with one NEPphosphoramidon complex in the asymmetric unit, and 5% of the reflections were used to calculate R free . The structure was solved by molecular replacement using MOLREP [30] with coordinates from PDB ID 1dmt as a search model. The phosphoramidon ligand was well defined in the difference electron density. Crystallographic refinement was performed with a combination of Refmac5 [31] and autoBUSTER [32], Coot [33] was used for model building. X-ray data collection and refinement statistics are listed in Table 1. The crystal structure has been deposited in pdb under the accession code 4CTH. Figures showing structural representations were prepared using PyMOL [34].

Engineering NEP for increased activity and specificity at Ab
Screening of the variants generated from the first round library resulted in the identification of NEP mutants with altered activity on all of the peptide substrates. To rank these variants the change in activity on the Ab1-40 peptide was divided by the average change in specificity on the other peptides and then by the specificity of the wild-type enzyme. The mutants were then ranked according to these scores and compared to the change in activity on Ab1-40 ( Figure 1A). Improvements in activity on Ab of up to approximately 9-fold and specificity changes of up to approximately 12-fold were observed. The DNA encoding each of the variants with altered specificity was sequenced and the mutations resulting in these changes identified. A wide range of mutations at numerous positions within the central core of the enzyme affected the activity on Ab and the other peptides. In particular, mutations at positions 590 and 377 where replacement of the aspartate at these positions with one of a number of amino acids with larger, hydrophobic side chains resulted in improvements in activity and specificity. At many other positions only a single amino acid substitution was detected as having improved specificity, and some of these provided the largest gains observed. The greatest change in specificity was observed with a change at position 714 from glycine to lysine, a change that introduces both a positive charge and a larger side chain. In the high throughput screening this change improved specificity by ,12-fold and activity on Ab by ,7-fold. There were several other positions where introduction of a lysine or arginine resulted in improvements in activity and specificity, notably positions 705 and 709 suggesting the introduction of a positive charge in this area was beneficial for the binding or cleavage of Ab and inhibitory for the binding and/or cleavage of the other peptides. From this analysis mutations that improved activity on Ab1-40 and reduced activity on the other peptides were identified and used to design a second library. Because of its strong specificity change G714K was fixed and randomly combined with other beneficial mutations to generate the second round library of variants. Following screening on the peptide panel, the variants were again ranked and improvements in Table 1. X-ray data collection and refinement statistics of NEPv in complex with phosphoramidon. Engineering Neprilysin Activity and Specificity PLOS ONE | www.plosone.org specificity of around 40-fold over wild-type were now observed. However, the variants did not demonstrate improvements in the cleavage of Ab over those seen on round 1 and a smaller number of hits were detected than at round 1, ( Figure 1B). While introduction of further mutations resulted in additional improvements in specificity, this was accompanied by a marked decrease in the stability of the enzyme with the majority of the expressed material being aggregated (results not shown).

Increased specificity of NEP mutants towards Ab peptide
The lead variant at round 2, NEP G399V/G714K (NEPv), was selected for further characterisation as it had increased activity on Ab1-40 and Ab1-42 and reduced activity on a panel of other natural substrates of the enzyme. In order to determine the magnitude of the increase in specificity towards Ab, kinetic parameters for cleavage of Ab1-40 and 16 off-target peptides were measured and compared to those of the wild-type enzyme ( Figure 2 and Figure 3). On Ab1-40 wild-type NEP had a k cat of 0.8360.1 s 21 and a K M of 104620 mM ( Table 2). These values gave a k cat /K M of 8.0610 3 M 21 s 21 . The value of k cat measured for human wild-type NEP was consistent with previous reports for the rabbit enzyme, although the K M value was ,8-fold higher (Leisring et al. 2003). However, the difference in K M values may be due to inherent differences between the human and rabbit enzyme or may result from difference in the preparation of the proteins. The k cat of NEPv was increased ,2-fold relative to that of wild-type NEP to 2.160.1 s 21 while K M was reduced to 1461 mM. These changes led to an increase in k cat /K M to 1.5610 5 M 21 s 1 , an increase of 18.5-fold over wild-type The cleavage of Ab1-40 by wild-type NEP and NEPv followed normal Michaelis-Menten kinetics at substrate concentrations between 0.4 and 50 mM (Figure 2). Due to the limited solubility of FAM-Ab1-40-biotin it was not possible to use substrate concentrations above 50 mM in this assay. Additionally, it was not possible to generate accurate kinetic data for the degradation of Ab1-42 due to its propensity to aggregate, however, we were able to demonstrate a very similar improvement in NEPv's activity on Ab1-40 and Ab1-42 using an endpoint assay ( Figure S1).
Kinetic parameters for cleavage of a panel of peptides by wildtype NEP and NEPv were determined ( Figure 3 and Table 2). The kinetic parameters determined for wild-type NEP on these substrates agreed reasonably well with those previously reported [4,35]. NEPv displayed significantly reduced catalytic efficiency relative to that of wild-type NEP on all of the other peptides tested. The greatest effect was observed with bradykinin, on which NEPv had 3,200-fold reduction in k cat /K M relative to that of wild-type NEP. The reduction in catalytic efficiency was largely due to a decrease in k cat ; the K M for bradykinin was relatively unaffected suggesting that the mutations introduced into NEPv have not affected binding of the substrate to the active site. This trend was observed across the panel of off-target peptides; the majority of the reductions in k cat /K M could be attributed to decreases in k cat . For example, NEPv showed reductions in k cat of 60-, 67-and 340-fold on angiotensin, GRP and neurotensin, respectively, while the K M values on these substrates were relatively unaffected. The reductions in catalytic efficiency were mediated by increases in K M on some substrates, such as nociceptin and substance P, although the overall changes in k cat /K M were relatively modest with these peptides. The average specificity switch (defined as the ratio of the fold change in k cat /K M on Ab1-40 to that on other peptides) on the panel of 16 peptides was 5,300-fold.
To investigate which of the two mutations in NEPv made a greater contribution to the increase in activity on and specificity towards Ab, G399V and G714K single mutants were generated and characterised. Kinetic parameters for cleavage of Ab1-40, angiotensin, bradykinin, glucagon and neurotensin by these variants were determined (Table 3). Both NEP G399V and NEP G714K showed increases in catalytic efficiency on Ab1-40; k cat / K M was increased by ,6-fold for both variants, relative to that of wild-type NEP, suggesting that the two substitutions make roughly equal contributions to the increase in activity on seen with NEPv.
The G399V mutation significantly reduced catalytic efficiency on angiotensin, bradykinin and neurotensin. For example, on bradykinin NEP G399V showed a 250-fold reduction in k cat /K M relative to wild-type NEP. The G714K mutation had much less effect on the activity towards other peptides; k cat /K M on bradykinin was reduced by only 7.5-fold while catalytic efficiency on angiotensin, glucagon and neurotensin was relatively unaffected compared to that of wild-type NEP. These data suggest that G399 has a greater role than G714 in regulating the substrate specificity of NEP.
We next investigated whether the mutations introduced to NEP had affected its sensitivity to known inhibitors of the enzyme. To determine the potencies of inhibition of wild-type NEP and NEPv by phosphoramidon, Ab1-40 cleavage activity was measured over a range of inhibitor concentrations ( Figure S2). Phosphoramidon inhibited wild-type NEP with an IC 50 of ,10 nM (Table S2). Because of the relatively low activity of wild-type NEP on Ab1-40 it was not possible to use an enzyme concentration lower than 20 nM in the assays and therefore the lowest IC 50 value that could be measured was 10 nM. Phosphoramidon inhibited NEPv with an IC 50 of 21 mM, a potency .2000-fold lower than for wild-type NEP. Although the observed reduction in potency may be partially explained by the increased catalytic efficiency of NEPv on Ab1-40, it nevertheless suggests that the mutant has increased active site binding selectivity. To investigate the relative roles of the G399V and G714K substitutions in this decrease in sensitivity to the inhibitor, inhibition of the NEP single mutants by phosphoramidon was characterised. As was observed with peptide cleavage activity, the G399V mutation had a greater effect than G714K on sensitivity to the inhibitor; NEP G399V was inhibited with an IC 50 of 610 nM compared to 100 nM for G714K. These observations are consistent with the hypothesis that G399 has a greater role than G714 in controlling the selectivity of active site binding.
Similar effects were observed with thiorphan, a highly selective and potent inhibitor of NEP [36]. Wild-type NEP was inhibited with an IC 50 of ,10 nM, whereas the IC 50 for inhibition of NEPv was 77 mM. As was observed with phosphoramidon, the G399V mutation had a greater effect than G714K on sensitivity towards thiorphan; NEP G399V was inhibited with an IC 50 of 4.4 mM whereas the IC 50 for NEP G714K was 75 nM (Table S2).

Altered cleavage site preferences of NEP mutants
Although NEP is relativity promiscuous in its cleavage site sequence specificity, it does exhibit a preference for hydrophobic residues, such as Phe or Leu, in its P1' site (MEROPS database, [37]). Wild-type NEP is able to cleave Ab1-40 at several sites although it appears to initially degrade the substrate preferentially at Lys 16 -Leu 17 and Phe 19 -Phe 20 [25,38]. Given the increased activity on Ab1-40 and reduced cleavage of other peptides, as well as the altered binding of inhibitors to the active site observed with NEPv, it seemed possible that Ab may be cleaved at different sites in the variant compared to wild-type NEP. To investigate this possibility, Ab1-40 was incubated with wild-type NEP or NEPv and cleavage products were analysed after 10, 60 and 360 min by MALDI MS.
Wild-type NEP cleaved Ab1-40 predominantly at Lys 16 -Leu 17 , Leu 17 -Val 18 and Phe 19 -Phe 20 ; the main cleavage fragments observed after 60 min were Asp 1 -Lys 16 , Asp 1 -Leu 17 and Asp 1 -Phe 19 (Figure 4). After 360 min these fragments were degraded further, accompanied by the appearance of Val 12 -Leu 17 and Tyr 10 -Leu 17 fragments. NEPv appeared to be more selective in its cleavage site preference; Ab1-40 was cleaved preferentially at Phe 20 -Ala 21 , as shown by the appearance of the Asp 1 -Phe 20 fragment after 10 min. A small amount of Asp 1 -Leu 17 was detected after 360 min although the Asp 1 -Lys 16 and Asp 1 -Phe 19 fragments were not observed. Cleavage of Ab1-40 at Phe 20 -Ala 21 by NEP has not been previously observed [25,27,38] and occupancy of the P1' site with Ala is relatively uncommon (MEROPS database, [37]). These results indicate that the mutations introduced into NEPv have fundamentally altered the cleavage site preference of the enzyme. It is interesting to note that the increase in catalytic efficiency on Ab1-40 observed for NEPv was largely driven by a decrease in K M , which may result from the substrate being able to adopt an altered binding conformation at the active site.
To determine whether the change in cleavage site preference observed for NEPv could be attributed to either the G399V or G714K mutations, cleavage fragments were also analysed for the NEP single mutants. NEP G399V cleaved Ab1-40 preferentially at the Phe 19 -Phe 20 site, as shown by the appearance of the Asp 1 -Phe 19 fragment after 10 min. Asp 1 -Lys 16 and Asp 1 -Leu 17 were also present after further incubation. NEP G399V also cleaved at Phe 20 -Ala 21 , in common with NEPv, although there was a much lower preference towards this cleavage site with the single mutant. NEP G714K cleaved Ab1-40 preferentially at Leu 17 -Val 18 . This mutant also showed some cleavage at Phe 20 -Ala 21 , although this did not represent the major cleavage site. These results suggest that while the G399V and G714K single substitutions do alter the cleavage specificity of NEP, both substitutions are necessary for the full change in cleavage site preference observed for NEPv.
The products of bradykinin digestion by wild-type NEP and NEPv were also determined by MALDI-TOF MS. Bradykinin was cleaved at Pro 7 -Phe 8 by both wild-type NEP and NEPv, as indicated by the appearance of the Arg 1 -Pro 7 peptide fragment. This cleavage site was consistent with that previously reported for porcine NEP [4].

Crystal structure of the NEP variant
To investigate how the two mutations had altered the structure of the enzyme the crystal structure of the extracellular domain of NEP G399V/G714K in complex with phosphoramidon was Figure 1. Relative activities of the mutants identified after screening. For each variant the relative activity at Ab divided by the relative activity on the off target peptides is shown in the black bar and the relative activity at Ab is shown in the open bar. Mutants are ranked by relative activity at Ab divided by the relative activity on the off target peptides after the first (A) and second (B) rounds of screening. Each bar represents and average of four readings for each variant on Ab alone, or Ab and up to 9 other peptides. Error bars are not shown due to the complex nature of the propagation of errors calculation for the relative specificity change where activity on up to 11 different peptides was measured for both wild-type and variant enzymes. For the first round of screening only the top 50 mutants are shown. Despite the intention of introducing only single amino acid substitutions into the enzyme in the first round a number of the variants with improved specificity had acquired two mutations. Activities are relative to the wild-type enzyme. doi:10.1371/journal.pone.0104001.g001 Engineering Neprilysin Activity and Specificity PLOS ONE | www.plosone.org refined to 2.15 Å resolution. The double mutant crystallized in the same space group and with similar cell dimensions as the wild-type structure with phosphoramidon as described by Oefner et al., 2000 [5]. The overall structure of the double mutant NEP is similar to wild-type NEP, with two domains which are connected by three interdomain linker segments, together enclosing a large water filled cavity. Both mutated residues (G399V/G714K) have defined electron density as shown in Figure S3. Residue Lys 714 is located in the larger domain (domain 1), close to the catalytic site. The second mutated residue Val 399 is at a distance of 20 Å from Lys 714 , located in the smaller domain (domain 2) at the surface of the protein at the interface between the two domains. Despite the Engineering Neprilysin Activity and Specificity PLOS ONE | www.plosone.org overall similarity a superposition shows substantial changes in the relative orientation of secondary structure elements predominantly in domain 1, which are illustrated in figure 5. The interdomain distance, as measured by Ca-Ca distance from residue Val 399 in domain 2 to e. g. residue Arg 698 in domain 1, increases by more than 2 Å (from 5.9 Å to 8.4 Å in wild-type NEP and NEPv, respectively). In domain 2, substantial main chain differences are only observed in the helix preceding the mutated Val 399 and the loop (residues 400-404) following this residue.
While the combined effects of the two mutations on structural changes cannot be fully deconvoluted, it is apparent that Lys 714 protrudes in to the S2' pocket of the enzyme and forces Arg102 away from the positions it has in wild-type NEP ( Figure 6). Furthermore Lys 714 limits the conformational flexibility of Trp 693 and thereby reduces the plasticity and size of the S1' pocket. Further changes in the vicinity of Lys 714 involve Arg 110 and Phe 106 .
The overall binding mode of phosphoramidon in the NEP double mutant is similar to that observed in wild-type (Figure 7). The rhamnose moiety, the tetrahedral N-phosphoryl interaction with the catalytic zinc and the L-leucyl moiety in the S1' pocket are nearly identical in both NEP forms. The C-terminal extension

Discussion
Proteases represent an attractive class of therapeutics because of their potential to irreversibly inactivate protein or peptide targets using lower amounts of the drug than would be required for stociometric binding and inactivation by molecules such as antibodies. However, because of the large substrate repertoires of many proteases, one of the key challenges to exploiting them therapeutically is the lack of specificity towards the target molecule, which may result in undesirable off-target activities. To address this challenge, protein engineering has been used to generate protease variants with increased activity on the target substrate relative to that on off-target substrates. For example, the substrate specificity of rat neurolysin has been altered by a single amino acid substitution, giving a 10-fold specificity switch [39].
NEP has been suggested as a potential therapeutic because it cleaves a wide range of physiological peptide substrates including Ab, which is thought to be a key pathological component of Alzheimer's disease. However, because NEP is promiscuous in its substrate preference and has relatively modest cleavage activity on Ab, improvement in the specificity of the enzyme is required to maximize its therapeutic potential. By introducing mutations at Phe 563 and Ser 546 , Sexton et al. [25] were able to alter the cleavage preference of NEP towards Ab, insulin-b chain and leuenkephalin, although the specificity changes observed were relatively small and accompanied by reductions in overall activity. By using a directed evolution approach, we have generated a variant of NEP that shows 20-fold increased catalytic efficiency on Ab1-40 and up to 3.2610 3 -fold reduced cleavage of a range of offtarget peptides.
Significant increases in activity on Ab could be achieved from single amino acid substitutions at a number of residues. For example, the G399V and G714K variants showed ,6-fold increase in k cat /K M on Ab1-40. NEPv, which contains both of these mutations, had an even greater improvement in activity, with a 20-fold increase in k cat /K M on Ab1-40 relative to that of wildtype NEP. As well as improved activity on Ab, NEPv displayed significantly reduced activity on a range of other peptide substrates. For example, on bradykinin k cat /K M was reduced by 3.2610 3 -fold giving a specificity switch of 6.4610 4 in favor of cleavage of Ab. While this specificity change is impressive, switches of greater magnitude have been reported for proteases in the literature; the E. coli endopeptidase OmpT has been engineered to preferentially cleave Ala-Arg rather than Arg-Arg by introduction of a single amino acid substitution, giving a specificity switch in excess of 10 6 [40]. However, it is likely that further engineering of NEP could lead to even greater improvements in specificity towards Ab. Indeed, we were able to isolate NEP variants with greater activity than NEPv on the target substrate, although these variants had significantly lower stability and were therefore less attractive for biopharmaceutical development. Nevertheless, the enhancements in specificity towards Ab are expected to improve the efficacy and safety profile of NEPv as a potential therapeutic for Alzheimer's disease.
The changes in kinetic properties observed for NEPv raised the question of which of the two mutations made the greatest contribution to the increase in activity and specificity towards Ab. Analysis of Ab1-40 cleavage by the NEP G399V and G714K single mutants revealed that the mutations caused similar increases in activity on this substrate. However, the G399V substitution caused a much greater reduction in activity on the other peptide substrates tested, although both mutations appear to be required to achieve the full specificity switch observed for NEPv. Consistent with this observation, NEP G399V also showed a much greater reduction in sensitivity to competitive inhibitors than NEP G714K suggesting that despite its distance from the active site, Gly 399 plays a greater role in regulating substrate binding.
As well as displaying altered substrate preference and sensitivity to inhibitors, NEPv also showed significant changes in its cleavage site preference on Ab. NEP generally exhibits a preference for cleavage of peptide substrates on the amino terminal side of hydrophobic residues such as Phe or Leu [6] and Ab is cleaved predominantly at Lys 16 -Leu 17 , Leu 17 -Val 18 and Phe 19 -Phe 20 by wild-type NEP [25,38]. These cleavage sites were also observed for wild-type NEP in the present study. In contrast, NEPv preferentially cleaved Ab at Phe 20 -Ala 21 and showed little cleavage at the Table 3. Steady-state kinetic parameters for cleavage of Ab1-40 and off-target peptides by NEP G399V and NEP G714K. sites favoured by the WT enzyme. Degradation of Ab by NEPv does not appear to be processive; the Asp 1 -Phe 20 fragment generated by cleavage at Phe 20 -Ala 21 was not cleaved significantly further over the course of 360 min. In contrast wild-type NEP is processive in its action, first cleaving Ab at Leu 17 -Val 18 and degrading the resulting fragment further to generate Tyr 210 -Leu 17 . The lack of processivity displayed by NEPv suggests that the specificity has changed to such an extent that initial cleavage products may be released from the active site before any further degradation occurs. Thus release occurs at a greater rate, thereby accelerating k 1 , which would be consistent with the reduced K M observed for the variant. Analysis of the cleavage events reported in the literature reveals that while NEP is able to cleave on the amino terminal side of Ala, this is relatively uncommon. For example, cleavage has been observed at Trp 16 -Ala 17 of proadre-  Engineering Neprilysin Activity and Specificity nomedulin although this appears to be secondary to cleavage at Glu 8 -Phe 9 [41]. The structural background to the substrate promiscuity in wildtype NEP is partially illustrated by crystal structures of various inhibitors bound to NEP. Large ligand induced flexibility of amino acid side chains is observed in the substrate binding site, with particularly Trp 693 functioning as a switch between the S1' and S2' pockets in wild-type NEP ( Figure 8); (P1' and P2' dependency shown in Oefner et al., [7], Glossop et al., [42]). In the crystal structure of the NEPv presented here, the two mutations have reduced the substrate binding sites S1' and S2' as well as decreased the binding site plasticity. While we cannot fully deconvolute the contributions of the individual mutants to these changes, it is likely that introduction of Lys at position 714 predominantly contributes to these local changes. This is supported by our observations that the largest change in specificity for a single mutant after the first round of screening was observed with G714K and the second largest effect for D709K.
One of the Ab1-40 cleavage fragments produced by wild-type NEP is Asp 1 -Phe 19 . This fragment is not observed as a NEPv cleavage product, which may partly be explained by the altered and reduced S1' pocket. The affinity of the P1' phenylalanine for the S1' pocket is most likely decreased and cleavage at Phe 19 -Phe 20 is less likely to occur. Furthermore, the electrostatic interactions on the prime side of the cleavage site are changed due to the introduction of the charged Lys 714 in the S2' pocket, and conformational changes of e.g. Arg 102 , Arg 110 and Phe 106 . It could be envisaged that these changes favour cleavage at Phe 20 -Ala 21 with a Glu at the P2' position. This is supported by the recent modelling of Ab interactions with wild-type NEP [43]. The model suggests a strong interaction between Glu 22 of Ab with the side chains of Arg 102 and Arg 110 of NEP. Our data indicating large conformational changes in the orientation of these side chains would be consistent with a change in the preferences for amino acids in the P2' position.
In the same way that the reduced size and plasticity of the substrate binding sites may affect the Ab1-40 cleavage specificity, the mutations may affect specificity towards other natural substrates of NEP. Most notable is the reduced catalytic efficiency on bradykinin compared to wild-type NEP, which in part might originate from steric hindrance of the bradykinin P1' phenylalanine and the NEPv S' site.
However, as an additive effect is seen on specificity and catalytic efficiency with both mutations present, and the NEP single mutant G714K showed some Ab1-40 cleavage at Phe 20 -Aal 21 , but preferentially cleaved at Leu 17 -Val 18 , other additional factors must play a role. The mutational effects on NEP conformation extend from the S1' pocket out to the surface of the protein. To our knowledge, there is no public structural data available with information on substrate binding beyond the S2' site. However, it is likely that the change in structure due to the mutations will further affect substrate binding specificity outside the S1'-S2' region.
The size of the opening to, as well as the size of, the catalytic chamber has been noted [44] to be incompatible with diffusion and fitting of folded Ab into the catalytic cavity. Two hypothetical conformational changes have been proposed to accommodate entry of large substrates into the active site: one where a localized loop shift allows substrate access, the other a hinge-mediated opening of domain 1 and 2. The related neutral endopeptidase thermolysin has been observed as both substrate analogue-bound closed [45] and unoccupied open [46] structures, and transition between these conformations, via hinge bending, is thought to occur during catalysis [47]. Recently, for insulin degrading enzyme, another member of the cryptidase family, both domain opening as well as localized loop-shifts have been observed, that allow substrate access to the active site [48]. The authors speculate that the substrate interaction with the open states contribute to substrate specificity. We have observed an alternative crystal form of NEPv, which showed a domain opening where increased access to the catalytic chamber is observed (manuscript in preparation). It is enticing to assume that the NEP variant has an altered substrate access dynamics; particularly the G399V mutation located at the domain interface could be involved in domain opening dynamics. This, in addition to the local changes in substrate binding sites, could contribute to the observed altered substrate specificity.
These hypotheses are further supported by the observation that NEPv had considerably reduced sensitivity compared to wild-type NEP to inhibition by thiorphan and phosphoramidon. While the crystal structure of phosphoamidon bound to NEPv shows only modest changes in the binding mode of the inhibitor, changed domain opening dynamics and the propensity of ligands to induce domain closure may govern the reduced inhibition observed by these small molecule inhibitors. Whereas these inhibitors effectively block the activity of wild-type NEP with low nM potencies, mM concentrations were required to have any effect on NEPv activity. The observations on the effects of the inhibitors provide further evidence that the mutations introduced into NEPv have substantially altered the active site binding selectivity of the enzyme. Figure 8. Plasticity in S1' and S2' substrate binding pockets. Superposition of 1r1h (cyan), NEPv (yellow) and 2yb9 (violet), illustrating flexibility of size and shape of S1' and S2' subsites. Protein surfaces of each complex are shown: A) 1r1h B) NEPv C) 2yb9. The inner parts of the S1' and S2' sub pockets will probably not be accessible in NEPv. doi:10.1371/journal.pone.0104001.g008 Our engineering of NEP appears to have fundamentally changed the cleavage site preference of the enzyme, and it is possible that this change has increased cleavage activity on other peptides not tested here. The engineered variant of NEP described here has significantly increased activity on and specificity towards Ab, and therefore has potential to be a more efficacious therapeutic for Alzheimer's disease with less activity on other peptides than wild-type NEP. Further development of NEPv as a therapeutic may require analysis of the cleavage of a greater range of peptide substrates, although this was beyond the scope of the present study.
Recent studies have shown that reduction of peripheral Ab through intravenous administration of NEP has no effect on brain Ab levels, suggesting that a peripheral sink of the peptide that can be targeted therapeutically to reduce brain amyloid burden does not exist [49,50]. Therefore, more targeted approaches, using either direct delivery into the brain or via blood brain barrier transport vectors, may be required for amyloid-reducing drugs to be effective. Table S1 Sequences of substrates used peptide cleavage assays. a Peptide substrates used for screening NEP variants. Peptides were labelled at the N-terminus with DY505 or DY647 and at the C-terminus with biotin. b Peptide substrates used for detailed kinetic characterisation of NEP variants. Peptides were labelled at the N-terminus with 5(6)FAM and at the C-terminus with biotin. c Unlabelled endothelin-1 used for determination of kinetic parameters in HPLC assay. d Modified endothelin-1 contained an additional N-terminal Gly residue had 1-15 or 3-11 disulphide bond removed to improve peptide solubility when modified with fluorescent dye and biotin. ANP = atrial natriuretic peptide; BNP = brain natriuretic peptide; GLP-1 = glucagon-like peptide-1; GRP = gastri-releasing peptide. (DOCX) Table S2 IC 50 values for inhibition of wild-type NEP and mutants by phosphoramidon and thiorphan. a IC 50 values were determined by measuring Ab1-40 cleavage activity with a substrate concentration of 10 mM, a NEP concentration of 20 nM and inhibitor concentrations between 2 nM and 100 mM. Values are means from three replicate experiments and are quoted 6 S.E.M. b Due to the relatively low activity of wild-type NEP on Ab1-40, the lowest enzyme concentration that could be used in the assay was 20 nM, and therefore it was not possible to determine IC 50 ,10 nM. (DOCX)