Journal of Proteomics & Bioinformatics- Open Access www.omicsonline.com Research Article JPB/Vol.1/December 2008 Novel Therapeutic Approaches to Treat Alzheimer’s Disease and Memory Disorders

Cognitive decline most commonly associated with Alzheimer’s dementia can also result from other conditions including cerebral ischemia or brain trauma. One quarter of people over the age of 65 are estimated to suffer some form of cognitive impairment underscoring the need for effective classes of cognitive-enhancing agents. In this mini-review we highlight recent work from our laboratory on the structural biology of Alzheimers disease and other memory disorders that was presented at the recent PRICPS – AOHUPO 2008 Conference held in Cairns, Australia. Our current work is focused on two proteins, amyloid precursor protein and insulin-regulated aminopeptidase, that are promising targets for the development of anti-Alzheimer’s drugs and as cognitive enhancers. In both cases structures determined by X-ray crystallography are being used to discover promising lead compounds by structure-based drug design.


Introduction
Approximately one quarter of people over the age of sixty five are estimated to suffer some form of cognitive impairment.The incidence of age-related neurological diseases is escalating primarily due to the increased life expectancy of the general population of many developed nations.One of the more prevalent and debilitating neurological disorders is Alzheimer's disease (AD).
AD is the leading cause of dementia among the elderly and is characterized by the presence of amyloid plaques, extensive neuronal death and shrinkage of the brain.It is increasingly accepted that the neurotoxic Ab peptide is responsible for compromising neuronal functions and triggering cell death (Selkoe, 2002;Pereira et al., 2004).The peptide is derived from the cleavage of the amyloid precursor protein (APP) and is the main constituent of the amyloid plaques (Kang et al., 1987).Ab arises through the sequential cleavage by the b-site APP cleaving enzyme (BACE) in the ectodomain near the transmembrane domain and by the g-secretase protein complex within the membrane.This processing is thought to occur in endosomes or in an endosome-trans golgi network recycling pathway.APP is a Type-I transmembrane protein with a large extracellular portion, which has been structurally and functionally subdivided into several domains.We initiated a structural biology program focused on APP about ten years ago with aim of determining its structure as the basis for understanding its function and for structure-guided drug design.The program has been fruitful with numerous structures of the various APP domains now determined (Rossjohn et al., 1999 AD drug development strategies have focused on modulation of the Ab processing enzymes (β and γ secretases) and prevention of Aβ aggregation or oligomerization (Wolfe, 2002).Currently all drugs approved by the US Food and Drug Administration for AD address disease symptoms.Most belong to the class of cholinesterase inhibitors, which are of limited efficacy and only indicated for the treatment of mild-to-moderate forms of the disease (Birks, 2006).In spite of this, many drugs currently being developed to treat cognitive decline in AD are still targeting central cholinergic systems (www.alzforum.org/drg/drc).Development of memory-enhancing drugs is gaining momentum because of their increasingly widespread application in the treatment of other forms of memory disorders, including mild cognitive impairment, as well as that resulting from brain trauma and ischemic damage.Here we describe our recent studies on APP and on a protein receptor involved in memory.

Structural studies of an amyloid peptide-antibody complex
One promising stream of research to develop treatments or prophylactics for AD is anti-amyloid-β (Aβ immunotherapy) which aims to promote Aβ peptide clearance from the brain.In January 2002, however, the outlook for immunotherapies to AD took a pessimistic turn when the first clinical trials of active vaccination for AD were halted after a small subset of patients developed sterile meningoencepha-litis.The synthetic Aβ peptide antigen from those trials, AN-1792, is thought to have elicited a severe T-cell mediated immune response inducing a higher incidence of meningoencephalitis.Despite this setback, it was subsequently demonstrated that vaccination with AN-1792 elicited a positive antibody response and that antibodies did not cross-react with APP as first feared.It was also shown that cognitive function was stabilised in over half of patients with high antibody titer (Weksler, 2004).
Six years later, Elan Corporation and Wyeth Pharmaceuticals, who initiated the AN-1792 trials, have begun a Phase 3 clinical program of bapineuzumab (AAB-001), for the treatment of patients with mild to moderate Alzheimer's disease.Bapineuzumab is a humanised monoclonal antibody targeting Aβ and is the first antibody based therapy to AD to enter Phase 3 clinical trials.It is hoped that by taking a passive immunotherapy approach, detrimental immune responses like those elicited by the AN-1792 immunogen can be avoided.
Results of the small Phase 2 clinical trial of Bapineuzumab on 240 patients showed significant slowing in the rate of mental decline and brain atrophy in some Alzheimer's patients receiving Bapineuzumab compared with placebo control patients.Unfortunately, in the Phase 2 trials twelve patients developed an accumulation of fluid on the brain known as vasogenic edema.Vasogenic edema was shown to occur at high doses and predominantly in patients carrying the E4 allele of apolipoprotein, a gene known to predispose a carrier to AD.With the incidence of vasogenic edema in mind, the Phase 3 trials initiated by Wyeth and Elan in 2007 were designed to treat APO E4 carriers and nonhttp://www.alzforum.org/new/detail.asp?id=1850).In addition to safety concerns raised in the Phase 2 trials, efficacy trends were weaker than expected.However, the much larger Phase 3 trials on Bapineuzumab will establish the true efficacy of this particular therapeutic in sufferers with mild to moderate forms of the disease.
A long term follow up study of the Phase 1 AN-1792 trial showed that immunisation of patients with pre-fibrillated Aβ (AN-1792) initiated a long term reduction in Aβ load and post mortem evidence of a plaque removal 5 years after the last injection.(Holmes et al., 2008).In addition, the two patients with almost complete elimination of plaques and lowest Aβ loads were shown to have had the highest antibody response during the Phase 1 trial.Unfortunately, there was no evidence that immunisation with AN-1792 had an impact on cognitive decline in the small cohort assessed in the follow up study.It may be that immunotherapies targeting carriers separately ( fibrillar Aβ (such as AN-1792) are not effective at reducing the concentration of soluble oligomeric forms of the peptide, and it has been suggested that disintegration of plaques may increase the concentration of these oligomers implicated in synaptic dysfunction and dementia in Alzheimer's disease (Patton et al., 2006).
Antibodies raised against Aβ that have been shown to reduce plaque burden and improve cognitive deficits in mouse models of AD are those raised against the N-terminal region of Aβ (McLaurin et al., 2002;Frenkel et al., 2003;Chauhan and Siegel, 2005).Following the AN-1792 trials, development of passive and active immunotherapies targeting Aβ has focused on this N-terminal region (~ residues 1 to 14), which contains the immunodominant B-cell epitope of Aβ but lacks the T-cell epitopes arguably responsible for abandonment of the AN-1792 trials.A number of antibodies specific to this region have been studied for their ability bind Aβ, inhibit fibrillogenisis and cytotoxicity, and reduce plaque burden in vivo (Solomon, 2004).In addition, several engineered immunogens based on this B-cell epitope have been proposed as vaccine candidates.Considering the intense interest in antibody based therapies targeting Aβ, it is surprising that only recently has structural data emerged characterising the molecular basis for antibody recognition of this important pathogen.

Crystal Structures
Concomitant structures of monoclonal antibody fragments (Fabs) with specificity for the immunodominant B-cell epitope of Aβ have been reported for three Fabs in unliganded states and in complex with Aβ peptides.Dealwis and co-workers (Gardberg et al., 2007) reported high resolution 3D structures of Fabs from the monoclonal antibodies PFA1 and PFA2 in complex with the Aβ 1-8 peptide (DAEFRHDS).We reported our own near-atomic resolution structures (Miles et al., 2008) for the antigen binding fragment of the WO2 antibody complexed to Aβ 1-16 and Aβ 1-28 peptides.Seven residues of Aβ (Ala 2 to Ser 8) have been modelled from electron density when complexed to PFA1, PFA2 or WO2.From our own work, extending the length of the peptide from 16 to 28 residues yielded equivalent structures with no electron density observed beyond this range, corresponding to peptide outside the restraints of the antigen binding site of WO2.
PFA1 and PFA2 have identical sequences across their light chains and indeed the WO2 light chain sequence is very similar (95% sequence identity) to the PFA light chains.Across the heavy chains PFA1 and PFA2 sequences share 96% sequence identity, whereas the WO2 heavy chain aligns to the PFA sequences with around 85% identity.The clearest point of difference is in the hypervariable CDR3 of the heavy chains, where the WO2 sequence is unique.A comparison of apo and liganded structures showed that the PFA antibodies change conformation by around 1 Å across the backbone of VH CDR3 upon binding Aβ, whereas WO2 CDRs remain virtually unchanged on binding Aβ.In this respect, Figure 1 shows Aβ in the binding cleft of the PFA1 and WO2 antibodies.A point of distinction between these structures is that in WO2 there is a large water filled cavity in the heavy chain beneath Asp 7 of Aβ.Unlike PFA1, WO2 makes no close contact or water mediated interaction with Asp 7 of Aβ.This may be a useful distinction between the two antibody binding sites considering the work of Zirah and coworkers who conducted an in vitro aging study of aspartates in Aβ 1-16 following a finding that an unusually high content of racemised and isomerised aspartates (common protein damage phenomena associated with ageing) were found in Aβ from amyloid deposits (Zirah et al., 2006).Zirah and coworkers showed that the major isomer generated in Aβ 1-16 that was aged in vitro was L-iso-Asp 7. Asp 7 is the only amino acid in the WO2 recognition epitope of Aβ not to make hydrogen bonds, salt bridges, or significant van der Waals contacts with WO2.Rather, it appears to be rigidified in the antibody by hydrogen bonds between its side chain and that of the adjacent Arg 5 of Aβ.Taken together with the fact that WO2 makes no direct contact with Asp 1 of Aβ, the WO2 Fv domain represents a promising lead for the development of an immunotherapy targeting the key region of Aβ, between the α-and β-secretase cleavage sites, with likely promiscuity for preserved and aged forms of the peptide.
Figure 2 shows schematic representations of the molecular basis for PFA1 and WO2 recognition of the Aβ immunodominant B-cell epitope.In the figure, interactions conserved between Aβ and the antibodies are highlighted in yellow, these occur predominantly at the light chain interface with Aβ.Again, the hydrogen bonding network that restrains Asp 7 of Aβ in the PFA antibody is absent in the WO2 mode of binding, and with Asp 7 unrestrained, the conformation of the C-terminus of this epitope is stabilised in WO2 by hydrophobic interaction between Ser 8 of Aβ and Tyr 100B (H) in CDR3 of WO2.Despite these differences in binding, the conformation of Aβ when liganded to WO2 or PFA antibodies is remarkably similar across the core epitope EFRH (see superposition in Figure 3).This conformation should serve as a reliable template for the design of antigens for active immunotherapies targeting Aβ.

Structural studies of insulin-regulated aminopeptidase -a protein receptor involved in memory
A variety of brain functions are regulated by a disparate family of neuropeptides.An interesting example of these is one of the bioactive degradation products in the renin-angiotensin system, Angiotensin IV (Ang IV; Val-Tyr-Ile-His-Pro-Phe), a metabolite that possesses very different biological properties to its precursors.Intra-cranial administration of Ang IV or its analogue Nle1-Ang IV has been shown to enhance memory retrieval and retention in several animal models (Braszko et al., 1988;Wright et al., 1993;Wright et al., 1999;Braszko, 2004), while its precursors Ang I to Ang III have no effect.While able to facilitate learning and memory in normal rodents, Ang IV was also able to ameliorate the deficits in a range of rat models of amnesia, including scopolamine (Pederson et al., 1998) and mecalymine treatment (Olson et al., 2004), ischemic damage (Wright et al., 1996), chronic alcohol exposure (Wisniewski et al., 1993) and bilateral perforant pathway lesions (Wright et al., 1999).Ang IV is known to stimulate DNA synthesis and enhance thymidine incorporation and may be involved in several neurobiological actions including neural development and neuron survival (Hall et al., 1995).
It has been proposed that the AT4 receptor, the only specific binder of Ang IV in the brain, might be responsible for these effects (Wright and Harding, 1994).The AT4 receptor was subsequently identified as the transmembrane enzyme, insulin regulated aminopeptidase (IRAP), which occurs abundantly in vesicles associated with the insulin-sensitive glucose transporter GLUT4 in a wide variety of tissues (Albiston et al., 2003).Along with GLUT4, IRAP translocates to the plasma membrane in response to insulin and is rapidly internalised in the absence of this signal (Kanzaki, 2006;Huang and Czech, 2007).IRAP belongs to the M1 family of zinc metallopeptidases, which includes the aminopeptidases A, N and B (Rogi et al., 1996).This class of enzymes is characterized by two common structural features -a Zn 2+ -binding motif HEXXH(X) 18 -E and an exopeptidase motif GXMEN (Iturrioz et al., 2001).IRAP is a type II membrane-spanning protein with an extracellular catalytic site (Keller et al., 1995;Rogi et al., 1996).The extracellular domain of IRAP contains 18 potential N-linked glycosylation sites as well as a putative cleavage site (F 154 / A 155 ) (Rogi et al., 1996) for release of the extracellular domain.
Following identification of IRAP as the AT4 receptor (Albiston et al., 2003), it was shown that Ang IV was a specific inhibitor of IRAP's aminopeptidase activity (Lew et al., 2003).This observation has led to the hypothesis that modulation of IRAP's aminopeptidase activity may be responsible for the cognitive effects seen with Ang IV, possi-  ) have been shown to facilitate learning and memory in rodents in aversion conditioning and spatial learning tasks.Testing of this hypothesis was carried out through examining the effects of another known IRAP inhibitor, the peptide LVVhemorphin-7, on memory and it was shown that this inhibitor had a very similar effect to Ang IV (Lee et al., 2004).While this does not unambiguously establish the actual mechanism and role of the inhibitors and IRAP in improving memory, it adds further support to the neuropeptide protection model.Other hypotheses for these effects that have been suggested include: (1) the regulation of GLUT4 translocation and therefore glucose uptake and neuronal activity, or (2) the direct activation of a signalling pathway.Evidence for the direct role of IRAP in the facilitation of this memory enhancement remains circumstantial.

Bioinformatics
Computational analysis of the IRAP amino acid sequence suggests that IRAP is comprised of several distinct regions (Fig. 4).Starting at the N-terminus, the cytoplasmic tail is predicted to be primarily disordered with some α-helical structure.Interestingly, IRAP is the only member of the M1 family that has a large cytoplasmic tail and it is believed that this tail may be involved in the trafficking of IRAP and GLUT4.This tail is followed by a typical single pass transmembrane region.Immediately proximal to the other side of the membrane is a β-sheet region, followed by an αhelical region that contains the conserved active site residues HEXXH(X) 18 -E and GXMEN.A short β-sheet region then separates this from the helical final region whose function is unknown, although it has been proposed to act as a molecular chaperone (Rozenfeld et al., 2004).

Crystal Structures
To date no crystal structure of IRAP has been solved.Structures of several similar enzymes, however, have been determined including human leukotriene A4 hydrolase (LTA4H, 1HS6, 10% sequence identity to IRAP, (Thunnissen et  LTA4H was the first M1 aminopeptidase family member for which a structure was solved (Fig. 5A, (Thunnissen et al., 2001)).LTA4H has a relatively short C-terminal region, similar to the more recent structure of cold aminopeptidase and the backbones of LTA4H and ColAP superimpose well with a root mean square deviation (r.m.s.d.) of 1.2 Å for 509 Cα atoms (Bauvois et al., 2008).They are both folded into three distinct domains (equivalent to D1, D2 and D4 of the proposed domain structure of IRAP, (Fig. 4)).D1 is composed of β-sheets which present a large surface to the solvent (Fig. 5).The catalytic domain (D2) is mainly composed of α-helices with the active site residues located on two helices.The active centre is located in a groove that consists of one β-strand and three α-helices in what has been described as a thermolysin-like catalytic domain (Thunnissen et al., 2001;Bauvois et al., 2008).D4 is αhelical and, of the three domains, has the lowest level of sequence conservation between LTA4H and ColAP (18% identity and 1.6 Å r.m.s.d.compared to 38% and 0.8 Å for While APN (Fig. 5C and 5D) and Tricorn interacting factor F3 (Fig: -5E) share the three domains described for ColAP and LTA4H, they contain an additional, small, barrel-like domain of β-strands (equivalent to D3 in the proposed domain structure of IRAP (Fig. 4)) located between the catalytic domain, D2, and the C-terminal domain, D4.This arrangement more closely resembles the predicted secondary structure of IRAP.As with LTA4H and ColAP, the major differences at both the sequence and 3D structure levels are found in D4.One of the most striking differences is in respect to the relative positioning of this domain with respect to D1 and D2.With LTA4H, ColAP and APN this C-terminal domain is folded back upon D2, such that they are situated quite close to each other.By contrast the overall structure of the tricorn protease more closely resembles that of a hook, with D4 positioned further away from D2, opening up the catalytic site.This means that Tricorn protease has a deep and wide cleft with its active site largely exposed to solvent, although it has been proposed that it may undergo a conformation change in D4 that may generate a closed-structure similar to those seen for the other M1 family members (Kyrieleis et al., 2005;Addlagatta et al., 2006).While these closed structures may only result from crystal contacts the ability of crystals of E. coli APN to absorb small molecules in crystal-soaking experiments suggests that the active site is still accessible.Movement between open and closed conformations may be a mechanism of regulating activity or substrate specificity (Ito et al., 2006).

Homology Modelling
For such large proteins with relatively low overall sequence identity, the crystal structures determined to date superimpose quite well, particularly over the catalytic region.This provides confidence in the reliability of homology models for other members of the M1 family developed using these structures as templates.To date the LTA4H structure has been used to generate models of the catalytic sites for APA (Rozenfeld et al., 2002;Goto et al., 2007;Claperon et al., 2008), APB (Pham et al., 2007), pyroglutamyl peptidase II (TRHDE) (Chavez-Gutierrez et al., 2006) and IRAP (Ye, et al., 2008) to try and identify key residues important in the biological activity of these enzymes.One example where this has been quite spectacularly successful is in the identification of residues responsible for the modulation of APA activity by calcium (Claperon et al., 2008).Earlier studies had identified other residues that exerted similar effects (Iturrioz et al., 2000;Goto et al., 2007), but with the model the authors were able to pinpoint the exact residues respon-sible and propose a mechanism by which calcium exerts its effects on APA.Another example of note is where models of TRHDE were used to try and explain the structural differences between standard M1 aminopeptidases and TRHDE, which is an omegapeptidase but still classed in the M1 family (Chavez-Gutierrez et al., 2006).By using their model to identify key differences between the omegapeptidase and aminopeptidase, they were able to mutate TRHDE such that it lost its omegapeptidase activity and displayed instead alanyl-aminopeptidase activity.
The lower sequence identity over D4 reduces the reliability of models built upon this domain.Coupled with this lower reliability is the variable orientation and close interaction of D4 with the active site seen in the published structures, meaning that models built using the structures as templates can only paint a limited picture of the regulation of enzyme activity.With the availability of the new structures of M1 aminopeptidases, however, more detailed models and improved insights will hopefully be gained.

Mechanism of activity and understanding substrate specificity
A detailed mechanism of action for M1 aminopeptidase activity was proposed by Ito et al. (2006) and computational docking of the synthetic substrate Leu-MCA into a molecular model of IRAP generates highly favourable solutions showing interactions consistent with this mechanistic description (Ye et al., 2008).Proton donation by Tyr 529 of IRAP completes hydrolysis of the N-terminal peptide bond of the substrate with interactions with the zinc ion stabilising the transition (Figure 6).
The crystal structures provide structural support for other biochemical data in proposing a mechanism for cleavage.These structures by themselves, however, have not provided significant insight into the substrate specificity of the M1 family, particularly considering the active site region shows high sequence and structural similarity.It is apparent that only small changes are responsible for determining substrate specificity.For example, only two mutations resulted in the change of TRHDE from an omegapeptidase to an aminopeptidase (Chavez-Gutierrez et al., 2006).Models of other members of the M1 family can used to provide insights into this question and to help design mutants to further understand substrate specificity.
With this aim in mind we compared our model of IRAP to LTA4H.Two key residues, Ala 427 and Leu 483, were identified which presented a more open arrangement of the S1 subsite of IRAP compared to LTA4H, where the corre- sponding residues (Tyr 267 and Phe 314) are bulky aromatics (Ye et al., 2008).Alteration of the S1 subsite of IRAP by the corresponding mutations resulted in altered activity and substrate specificity.The IRAP mutant L483F, although capable of efficiently cleaving the N-terminal Cys from vasopressin, was unable to cleave the tyrosine residue from either leu-enkephalin or cyt6-vasopressin.The mutant also showed a dramatic reduction in the affinity of the inhibitor norleucine 1 -angiotensin IV whereas the affinity of Ang IV remained unaltered.Introducing an A427Y mutation to IRAP, by contrast, resulted in increased binding affinity of the peptide substrates without a corresponding increase in the rate of cleavage.This not only suggests a specific role for the two residues, Ala 427 and Leu 483, in defining the catalytic cleft of IRAP but confirms that substrate specificity is likely a result of small differences imparted from the chemical nature of the residues between the family members.

Inhibitors
The development of new, specific M1 aminopeptidase inhibitors is a difficult and complicated task due to the high sequence identity and structural homology of the catalytic sites of these proteins.The multiple biological roles of members of the M1 aminopeptidase family, however, require highly specific inhibitors for use both therapeutically and for physiological investigations.The existing crystal structures of family members, however, do provide a basis for attempting the rational development of selective inhibitors.An example of this is the use of the structure of LTA4H computationally docked to a known inhibitor to develop a pharmacophore model then used to identify a subset of compounds that inhibited LTA4H activity (Grice et al., 2008).One of these compounds was a high affinity inhibitor in vitro, orally bioavailable and showed in vivo activity.
Most physiological studies of IRAP activity have utilised the two known peptide inhibitors, Ang IV and LVV-H7, and their modified forms (Albiston et al., 2007).These have some key limitations, including their short half-life and promiscuous inhibitory activity.One way of overcoming some of these limitations has been the development of cyclic peptide inhibitors that have higher IRAP affinity and longer halflives (Axen et al., 2006;Axen et al., 2007).Unfortunately these cyclic peptides are still not selective enough for therapeutic use or detailed physiological investigation; for example the best cyclic inhibitor also inhibits APN.
We have taken a slightly different approach to our attempts to identify new inhibitors.Using a model of the active site of IRAP based upon the crystal structure of LTA4H, we used in silico screening and in vitro assays, some of the latter performed in the lab of our collaborators, Drs Anthony Albiston and Siew Yeen Chai, to identify a family of highly specific, competitive, non-peptide inhibitors (Albiston et al., 2008).Since these compounds are not peptidic but instead small molecule inhibitors they are amenable to medicinal chemistry modification to tune their activities and pharmacokinetic properties.Members of the family of compounds show negligible activity against some of IRAP's closest relatives, including ERAP1, ERAP2, APN and LTA4H.When administered intracranially they showed the same memory enhancement effects seen for classic IRAP inhibitors Ang IV and LVV-H7, providing further support for the effects being mediated directly through IRAP.These inhibitors are providing the basis for ongoing research into the development of a new class of therapeutics for the treatment of memory amelioration associated with Alzheimer's and more generally, other dementias.
Paper Accepted from the Joint 2nd Pacific Rim International Conference on Protein Science and 4th Asian-Oceania Human Proteome Organization, Cairns-Australia, 22-26 June 2008

Figure 1 :
Figure 1: Electrostatic surface electrostatic representation of the WO2Fab antibody (A) and the PFA1 antibody (B) in complex with the Aβ peptide.In both representations the solid white is the heavy chain of the antibody and the translucentdark region is the light chain.The surfaces are coloured by electrostatic potential with negative charge shown in red and, positive charge in blue.The Aβ peptide is shown as ball-and-stick and water molecules as small spheres.

Figure 2 :
Figure 2: Schematic representations of the molecular basis for PFA1 and WO2 recognition of the Aβ immunodominant Bcell epitope.This figure was produced using LIGPLOT (Wallace et al., 1995).

Figure 4 :
Figure 4: Diagrammatic representation of the domain arrangement of IRAP based on analysis of the IRAP amino acid sequence.

Figure 6 :
Figure 6: Schematic showing the mechanism for M1 aminopeptidase activity proposed by Ito and colleagues (2006).