This paper seems to skirt many years of autopsy studies finding that the senile plaques have marginal statistical capacity to distinguish normal from AD patients, such as Davis et al., 1999. Also in 1999, Lue et al. concluded that soluble Aβ posed the greatest toxicity, and that Aβ40 was particularly toxic to synapses. But the issue of deposits seems to pale in the face of new MRI studies concerning gross atrophy of the brain, that is, gross death of brain cells and connectivity. See, for example Stroub and colleagues’ 2006 article on MCI. Is the senile plaque issue becoming a tempest in the teapot?

References:
Davis DG, Schmitt FA, Wekstein DR, Markesbery WR, Alzheimer neuropathological alterations in aged cognitively normal subjects, 1999 Apr;58(4):376-88. Abstract

Lue LF, Kuo YM, Roher AE, Brachova L, Shen Y, Sue L, Beach T, Kurth JH, Rydel; RE, Rogers J, Soluble amyuloid beta peptide concentration as a predictor of soluble synaptic change in Alzheimer's disease, Am J Pathol,1999 Sept; 155(3):853-62. Abstract

Stroub TR, deToledo-Morrell L, Stebbins, GT, Leurgans S, Bennett DA, Shah RC, Hippocampal disconnection contributes to memory dysfunction in individuals at risk for Alzheimer's disease, Proc Nat Acad Sci USA, 2006 June; 103(26):10041-45. Abstract

View all comments by Erik Jansson

This report unequivocally demonstrates that Aβ40 and Aβ42 peptides have opposite effects on amyloid deposition in vivo and that Aβ40 inhibits Aβ42-induced amyloidosis. These results nicely complement our data that reducing Aβ40, without increasing Aβ42, leads to accelerated plaque pathology (Deng et al., 2006 and Wang et al., 2006). This is a welcome addition to the Alzforum discussion initiated by Peter Davies and Bart De Strooper last year concerning the pathogenic mechanisms of the PS1 FAD mutations. All data combined support the notion that a partial reduction of Aβ40 (and γ-secretase activity) may be the primary mechanism for the amyloid pathology seen in certain PS1 patients and may indeed apply to sporadic cases, as well.

View all comments by Hui Zheng

Can Some Forms of Aβ Be Good?
The generation of BRI-Aβ40 and BRI-Aβ42 transgenic mice and the crossing of these mice with the Tg2576 APP-transgenic line has enabled Kim and colleagues to determine whether the Aβ40 and Aβ42 peptides could play different roles in plaque deposition. Increasing Aβ40 by crossing BRI-Aβ40 and Tg2576 transgenic mice resulted in decreased plaque deposition, in contrast to the increased deposition previously reported in the BRI-Aβ42/Tg2576 bitransgenic (McGowan et al., 2005). This anti-amyloidogenic “activity” of Aβ40 was confirmed by crossing BRI-Aβ40 with BRI-Aβ42 transgenic mice, which resulted in reduced amyloid deposition relative to BRI-Aβ42 alone. The dramatically decreased plaque number was paralleled by a similar reduction in insoluble, formic acid extractable Aβ, despite an overall increase in total Aβ. The authors then used an in vitro aggregation assay to support their suggestion that decreased amyloid deposition may relate to the ability of Aβ40 to decrease Aβ42 aggregation.

These experiments raise the possibility that Aβ40...  Read more

Can Some Forms of Aβ Be Good?
The generation of BRI-Aβ40 and BRI-Aβ42 transgenic mice and the crossing of these mice with the Tg2576 APP-transgenic line has enabled Kim and colleagues to determine whether the Aβ40 and Aβ42 peptides could play different roles in plaque deposition. Increasing Aβ40 by crossing BRI-Aβ40 and Tg2576 transgenic mice resulted in decreased plaque deposition, in contrast to the increased deposition previously reported in the BRI-Aβ42/Tg2576 bitransgenic (McGowan et al., 2005). This anti-amyloidogenic “activity” of Aβ40 was confirmed by crossing BRI-Aβ40 with BRI-Aβ42 transgenic mice, which resulted in reduced amyloid deposition relative to BRI-Aβ42 alone. The dramatically decreased plaque number was paralleled by a similar reduction in insoluble, formic acid extractable Aβ, despite an overall increase in total Aβ. The authors then used an in vitro aggregation assay to support their suggestion that decreased amyloid deposition may relate to the ability of Aβ40 to decrease Aβ42 aggregation.

These experiments raise the possibility that Aβ40 may be a protective molecule, and emphasize the importance of the Aβ42:Aβ40 ratio as a determinant of amyloid pathology. This is of particular concern as some γ-secretase inhibitors can actually increase the Aβ42:Aβ40 ratio, which could potentially increase plaque deposition despite reducing total Aβ levels. These findings also complement a recent report showing that some presenilin mutations increase amyloid pathology in mice by selectively decreasing Aβ40 without affecting Aβ42 (Deng et al., 2006, and Wang et al., 2006), implicating a potential pathogenic role for decreasing Aβ40.

This paper raises a number of interesting questions. Most importantly, what effect does decreased plaque load in the presence of overall increased Aβ in the BRI-Aβ40 have on behavior, especially the memory deficits observed in Tg2576 mice? This is a central question that bears on the issue of whether Aβ40 is truly protective. It is conceivable that increased levels of soluble Aβ could impair memory performance despite reduced plaque numbers, possibly by increasing the formation of Aβ oligomers. These mice might thus be a valuable resource for identifying the structural forms of Aβ that contribute to memory impairment. Biochemical and behavioral analysis of the BRI-Aβ40/Tg2576 brains may enable more precise information about the roles of Aβ monomers, trimers, oligomers, Aβ*56, protofibrils, and fibrils.

There were hints that the BRI-Aβ40 may have protective effects in addition to preventing plaque deposition, since the BRI-Aβ40/Tg2576 showed somewhat reduced premature death. However, this was not straightforward since the BRI-Aβ40/BRI-Aβ42 bitransgenic showed an even greater increase in premature death. In conclusion, these new observations provide another level of complexity to the Aβ story, and suggest that Aβ by any other name may not be the same.

References:
Kim J, Onstead L, Randle S, Price R, Smithson L, Zwizinski C, Dickson DW, Golde T, McGowan E. Abeta40 inhibits amyloid deposition in vivo. J Neurosci. 2007 Jan 17;27(3):627-33. Abstract

McGowan E, Pickford F, Kim J, Onstead L, Eriksen J, Yu C, Skipper L, Murphy MP, Beard J, Das P, Jansen K, Delucia M, Lin WL, Dolios G, Wang R, Eckman CB, Dickson DW, Hutton M, Hardy J, Golde T. Abeta42 is essential for parenchymal and vascular amyloid deposition in mice. Neuron. 2005 Jul 21;47(2):191-9. Abstract

Deng Y, Tarassishin L, Kallhoff V, Peethumnongsin E, Wu L, Li YM, Zheng H. Deletion of presenilin 1 hydrophilic loop sequence leads to impaired gamma-secretase activity and exacerbated amyloid pathology. J Neurosci. 2006 Apr 5;26(14):3845-54. Abstract

Wang R, Wang B, He W, Zheng H. Wild-type presenilin 1 protects against Alzheimer disease mutation-induced amyloid pathology. J Biol Chem. 2006 Jun 2;281(22):15330-6. Epub 2006 Mar 29. Abstract

Bentahir M, Nyabi O, Verhamme J, Tolia A, Horre K, Wiltfang J, Esselmann H, De Strooper B. Presenilin clinical mutations can affect gamma-secretase activity by different mechanisms. J Neurochem. 2006 Feb;96(3):732-42. Epub 2006 Jan 9. Abstract

View all comments by Matthew Hass
View all comments by Bruce Yankner

Amyloid: Getting Less Toxic Every Day
The Alternate Amyloid Hypothesis (1,2), whereby amyloid-β (Aβ) serves as a protective response in the pathogenesis of AD, is supported by this recent paper showing that Aβ is not responsible for the cognitive and pathological changes that are pathognomonic for AD (3). Briefly, in this study, Aβ40 dramatically reduces Aβ deposition (60-90 percent compared with Tg2576 mice) and rescues the premature-death phenotypes of Tg2576 mice. The important question is whether pathological changes observed in Tg2576 mice (e.g., gliosis, synapse degeneration, cognitive deficits) are altered in Aβ40/Tg2576 mice. Interestingly and most importantly, the same research group reported no cognitive improvement in Aβ40/Tg2576 mice compared with Tg2576 mice (4). In this regard, other studies have found that the cognitive function is relatively intact in APP transgenic mice despite massive accumulation of Aβ including soluble and insoluble forms in brain (5,6). Therefore, the role of Aβ in the pathogenesis of AD should be reassessed. It really does appear...  Read more

Amyloid: Getting Less Toxic Every Day
The Alternate Amyloid Hypothesis (1,2), whereby amyloid-β (Aβ) serves as a protective response in the pathogenesis of AD, is supported by this recent paper showing that Aβ is not responsible for the cognitive and pathological changes that are pathognomonic for AD (3). Briefly, in this study, Aβ40 dramatically reduces Aβ deposition (60-90 percent compared with Tg2576 mice) and rescues the premature-death phenotypes of Tg2576 mice. The important question is whether pathological changes observed in Tg2576 mice (e.g., gliosis, synapse degeneration, cognitive deficits) are altered in Aβ40/Tg2576 mice. Interestingly and most importantly, the same research group reported no cognitive improvement in Aβ40/Tg2576 mice compared with Tg2576 mice (4). In this regard, other studies have found that the cognitive function is relatively intact in APP transgenic mice despite massive accumulation of Aβ including soluble and insoluble forms in brain (5,6). Therefore, the role of Aβ in the pathogenesis of AD should be reassessed. It really does appear to be becoming less toxic [sic!] every day.

References:
1. Lee HG, Casadesus G, Zhu X, Takeda A, Perry G, Smith MA. Challenging the amyloid cascade hypothesis: senile plaques and amyloid-beta as protective adaptations to Alzheimer disease. Ann N Y Acad Sci. 2004 Jun;1019:1-4. Review. Abstract

2. Lee HG, Zhu X, Nunomura A, Perry G, Smith MA. Amyloid beta: the alternate hypothesis. Curr Alzheimer Res. 2006 Feb;3(1):75-80. Review. Abstract

3. Castellani RJ, Lee HG, Zhu X, Nunomura A, Perry G, Smith MA. Neuropathology of Alzheimer disease: pathognomonic but not pathogenic. Acta Neuropathol (Berl). 2006 Jun;111(6):503-9. Epub 2006 Apr 27. Abstract

4. Janus C, Kim J, Hanna A et al. Dissociation between amyloid pathology and memory impairment Alzheimer's & Dementia, 2(3 (Supplement)), S85-S86 (2006).

5. Bizon J, Prescott S, Nicolle MM. Intact spatial learning in adult Tg2576 mice. Neurobiol Aging. 2007 Mar;28(3):440-6. Epub 2006 Feb 28. Abstract

6. Savonenko AV, Xu GM, Price DL, Borchelt DR, Markowska AL. Normal cognitive behavior in two distinct congenic lines of transgenic mice hyperexpressing mutant APP SWE. Neurobiol Dis. 2003 Apr;12(3):194-211. Abstract

View all comments by Rudy Castellani
View all comments by Hyoung-gon Lee
View all comments by George Perry
View all comments by Mark A. Smith
View all comments by Xiongwei Zhu

This paper shows the generation of a novel model of cerebral (non-Aβ) amyloid deposition. The authors generated transgenic mice expressing a mutant form of the BRI gene, found in patients affected by familial Danish dementia (FDD). FDD is a rare inherited disease that causes progressive dementia that, like AD, is neuropathologically characterized by amyloid deposition (ADan), neurofibrillary tangle formation (identical to that seen in AD), and neuronal cell loss. This model provides an exciting new tool in which to study the abnormal changes in the brain that lead to dementia. Comparing the similarities and differences of these two related neurological diseases may provide important clues to how AD develops.

View all comments by Nikolaos K. Robakis

This is a beautiful paper from Dr. Golde's lab showing for the first time that a peptide derived from the BRI2 gene is able to reduce cerebral Aβ deposition in vivo in an AD mouse model and that the same peptide inhibits Aβ aggregation in vitro. The peptide is a 23 amino acid long (Bri2-23) C-terminal fragment generated by the pro-protein convertases processing (Kim et al., 1999) of BRI2, a 266-amino-acid, type-II single transmembrane domain protein (Vidal et al., 1999). Using recombinant adeno-associated virus 1 (rAAV1)-mediated gene transfer in TgCRND8 mice, the investigators show a dramatic suppressive effect of the BRI2 transgene—and a BRI2-Aβ1–40 fusion protein (Kim et al., 2007)—on parenchymal Aβ accumulation. Importantly, the investigators found no evidence for alterations in the steady-state levels of APP or APP CTFβ in TgCRND8 mice expressing the virally delivered BRI2-Aβ1–40 or BRI2 transgenes, but rather that the Bri2–23 peptide could directly inhibit Aβ1–42 fibrillogenesis in vitro.

Mutations in the BRI2 gene cause neurodegenerative diseases characterized by...  Read more

This is a beautiful paper from Dr. Golde's lab showing for the first time that a peptide derived from the BRI2 gene is able to reduce cerebral Aβ deposition in vivo in an AD mouse model and that the same peptide inhibits Aβ aggregation in vitro. The peptide is a 23 amino acid long (Bri2-23) C-terminal fragment generated by the pro-protein convertases processing (Kim et al., 1999) of BRI2, a 266-amino-acid, type-II single transmembrane domain protein (Vidal et al., 1999). Using recombinant adeno-associated virus 1 (rAAV1)-mediated gene transfer in TgCRND8 mice, the investigators show a dramatic suppressive effect of the BRI2 transgene—and a BRI2-Aβ1–40 fusion protein (Kim et al., 2007)—on parenchymal Aβ accumulation. Importantly, the investigators found no evidence for alterations in the steady-state levels of APP or APP CTFβ in TgCRND8 mice expressing the virally delivered BRI2-Aβ1–40 or BRI2 transgenes, but rather that the Bri2–23 peptide could directly inhibit Aβ1–42 fibrillogenesis in vitro.

Mutations in the BRI2 gene cause neurodegenerative diseases characterized by cerebral amyloid deposition (Vidal et al., 1999, 2000), and transgenic mice overexpressing a mutant form of BRI2 show cerebral amyloid (ADan) deposition (Vidal et al., 2008). Interestingly, the amino-termini of the amyloid peptides (ABri and ADan) contain the amino acid sequence of the anti-amyloidogenic peptide Bri2-23. The unexpected findings of Kim et al. generate even more questions regarding the normal role of the still poorly characterized BRI2 gene and how mutations in BRI2 lead to neurodegeneration. More work is needed to determine whether the Bri2-23 peptide is able to depolymerize mature Aβ fibrils and if the anti-amyloidogenic properties of Bri2-23 are also shared by the C-terminal peptides generated by other members of the BRI gene family (Vidal et al., 2001). The use of increasing levels of BRI2 in the brain for the treatment of AD as proposed by Kim and collaborators (Kim et al., 2008) is an interesting idea; however, we believe that since the normal function of BRI2 (and the Bri2-23 peptide) is not known, caution should be taken in attempting therapies based on the overexpression of BRI2 alone.

References:
Kim SH, Wang R, Gordon DJ, Bass J, Steiner DF, Lynn DG, Thinakaran G, Meredith SC, Sisodia SS. Furin mediates enhanced production of fibrillogenic ABri peptides in familial British dementia. Nat Neurosci. 1999 Nov;2(11):984-8. Abstract

Kim J, Onstead L, Randle S, Price R, Smithson L, Zwizinski C, Dickson DW, Golde T, McGowan E. Abeta40 inhibits amyloid deposition in vivo. J Neurosci. 2007 Jan 17;27(3):627-33. Abstract

Vidal R, Frangione B, Rostagno A, Mead S, Révész T, Plant G, Ghiso J. A stop-codon mutation in the BRI gene associated with familial British dementia. Nature. 1999 Jun 24;399(6738):776-81. Abstract

Vidal R, Revesz T, Rostagno A, Kim E, Holton JL, Bek T, Bojsen-Møller M, Braendgaard H, Plant G, Ghiso J, Frangione B. A decamer duplication in the 3' region of the BRI gene originates an amyloid peptide that is associated with dementia in a Danish kindred. Proc Natl Acad Sci U S A. 2000 Apr 25;97(9):4920-5. Abstract

Vidal R, Calero M, Révész T, Plant G, Ghiso J, Frangione B. Sequence, genomic structure and tissue expression of Human BRI3, a member of the BRI gene family. Gene. 2001 Mar 21;266(1-2):95-102. Abstract

Vidal R, Barbeito AG, Miravalle L, Ghetti B. Cerebral Amyloid Angiopathy and Parenchymal Amyloid Deposition in Transgenic Mice Expressing the Danish Mutant Form of Human BRI(2). Brain Pathol. 2008 Apr 10; Abstract

View all comments by Bernardino Ghetti
View all comments by Ruben Vidal

There are between 50 and 100 experimental manipulations that have been shown to alter the pathologic and/or behavioral phenotypes of various transgenic models of human Alzheimer disease. The description in this paper of the effect of the Bri protein, the agent of familial British dementia, by Todd Golde and his colleagues, is the latest example in which overexpressing a transgene encoding a wild-type protein in TgCRND8 model AD mice has an ameliorative effect on the AD phenotype. These observations are quite striking in the context of three other instances in which the expressed protein suppressing the AD phenotype is a precursor of a protein in which the wild-type or a mutant form is the proximal cause of human CNS or systemic amyloidosis. Similar effects have been found for cystatin C in Aβ Tg2576 (Mi et al., 2007) or APP23 (Kaeser et al., 2007) double transgenics; animals in which gelsolin, the precursor in the Finnish form of familial amyloidotic polyneuropathy (  Read more

There are between 50 and 100 experimental manipulations that have been shown to alter the pathologic and/or behavioral phenotypes of various transgenic models of human Alzheimer disease. The description in this paper of the effect of the Bri protein, the agent of familial British dementia, by Todd Golde and his colleagues, is the latest example in which overexpressing a transgene encoding a wild-type protein in TgCRND8 model AD mice has an ameliorative effect on the AD phenotype. These observations are quite striking in the context of three other instances in which the expressed protein suppressing the AD phenotype is a precursor of a protein in which the wild-type or a mutant form is the proximal cause of human CNS or systemic amyloidosis. Similar effects have been found for cystatin C in Aβ Tg2576 (Mi et al., 2007) or APP23 (Kaeser et al., 2007) double transgenics; animals in which gelsolin, the precursor in the Finnish form of familial amyloidotic polyneuropathy (Hirko et al., 2007), has been expressed in Tg2576 and APP695/mutantPS1 mice transgenic for Aβ, and our own work describing the profound effect of overexpressing a transgene encoding wild-type human transthyretin in the APP23 model of AD (Buxbaum et al., 2008).

Why should these proteins in particular have such an effect? If we assume that the excessive generation of Aβ1-42, its misfolding and subsequent aggregation into toxic oligomers and fibrils, is intrinsic to AD (as represented by these models), there are a variety of possible mechanisms that could explain the results. The overexpressed amyloid precursors may have a direct interaction with the Aβ fragment or its oligomers in the brain to either disaggregate them or accelerate their aggregation into larger non-toxic multimers that can be more rapidly engulfed and degraded by glia. They may bind to some factor that is critical for the generation of Aβ or its aggregation, reducing the concentration of fibrillogenic precursor. They may interfere with a downstream process responsible for neurotoxicity, having no impact on aggregation per se but a strong effect on the behavioral phenotype.

In the gelsolin instance, the gene was introduced by hydrodynamic gene delivery and appeared to only be expressed in the periphery, not in the brain. Hence, its effect is hypothesized to be based on its action as a “plasma sink” for Aβ, increasing its transport from the brain to the systemic circulation, thereby decreasing the effective intracerebral Aβ concentration. A similar notion involving the CSF compartment has previously been proposed for the transthyretin effect. We think this unlikely (see below).

The observations could be trivial since it is also possible that the effects may be mouse specific and have no relationship to human disease. Equally unlikely is the possibility that the apparent proclivity of this set of proteins to have the observed effect may represent a strong ascertainment bias in which the proteins in question are only a small sample of the universe of proteins that can do this, and the molecules that have been assayed for this property have been chosen precisely because they are amyloid precursors. For the purposes of the rest of my discussion I will ignore the last two possibilities and assume that the observations in the double transgenics and the gelsolin animals have some biologic relevance.

Transthyretin, cystatin C, and gelsolin have been found in Aβ deposits in human AD brains. It has also been shown that in vitro the proteins directly interact with some form of Aβ, in the case of transthyretin most likely a subfibrillar aggregate. These proteins are apparently protective. We believe that their intrinsic amyloidogenicity indicates that they are predisposed to transiently expose their internal hydrophobic sequences to the external (with respect to the protein’s structure) aqueous milieu. If this occurs for a prolonged period or in a substantial portion of their conformational ensemble—conditions more likely for mutant forms of the proteins—the molecules will self-aggregate. However, if the molecule interacts with the hydrophobic portion of another similarly predisposed protein, the interaction can create a hydrophobic micro-environment for that protein domain. If the time frame is short enough, the remaining portions of the two interacting molecules re-fold to re-submerge the hydrophobic region into the internal portion of the native folded molecule. This process most resembles domain swapping but involves regions smaller than full domains and is temporally much more transient. Thus, there could be a series of proteins that are capable of protectively interacting with Aβ or its pre-toxic aggregates serving as “amateur” or “non-professional” chaperones for this particular cargo molecule.

Why should such a mechanism be necessary? The relative frequency of neurodegenerative disorders related to gain of toxic function by misfolded proteins suggests that the usual proteostatic mechanisms operating in neurons are limited. The relative hypersensitivity of neurons to hyperthermia is consistent with this view. It is apparent that during the evolution of the central nervous system, selection has favored the production of limited amounts of functional small peptides. These, because of their size, are less likely to misfold, and are secreted in vesicles that are at neuronal termini, thus not exposing the rest of the cellular milieu to high concentrations of potentially misfolded molecules.

These mechanisms serve the neuron well under most circumstances, unless there are destabilizing mutations in intrinsic neuronal proteins (e.g., α-synuclein, Huntingtin, SOD1). They may also fail when there is an interaction with an infectious agent capable of re-templating the folding of an endogenous protein. The system itself may become less effective (as in aging) for as yet unknown reasons. Under such circumstances, other mechanisms, such as those employing the “amateurs,” are recruited to cope. It is noteworthy that the transcription of transthyretin in the brain has been seen to increase in transgenic AD models. Interestingly, the AD models all require some degree of overexpression of the mutant Aβ construct, suggesting that the intrinsic murine neuronal proteostatic system functions well until it is overloaded. Old mice do not have an AD equivalent in the absence of overexpression of a human AD gene.

It is also possible that the amyloidogenic proteins are not truly “non-professionals” but represent previously unrecognized elements of the neuronal chaperome. Richard Morimoto’s work in C. elegans is consistent with such a hypothesis in that mutations in known elements of the proteostatic machinery reduce the number of glutamines required to produce a neuropathologic phenotype in a poly-Q model of Huntington’s disease, but the effects of such mutations are not seen until the system is stressed, for example, by a misfolded protein challenge [see Bar Harbor Report 2007]. More broadly, cellular proteostasis networks and their role in health and disease are elegantly reviewed in Balch et al., 2008.

Can these notions be experimentally tested for the proteins discussed here? Each observation should be validated by silencing the gene in question. Thus far, only deletion of the transthyretin gene has been tested for its effect on the development of a model of human Aβ transgene-induced murine AD. It accelerated the development of Aβ deposits in two different transgenic models, displaying a gene dose effect strongly supporting the notion that the observations were biologically relevant. If homozygous silencing of the gene in question is lethal, the effect of hemizygous silencing or siRNA knockdown of the gene on amplifying the Aβ phenotype should be reproduced as independent validation of the effect of the particular protein in question.

The protein should be tested for its ability to bind to Aβ in vitro by some standard assay of protein interaction, and the nature of the molecular species of both the “chaperone” protein and Aβ involved in the binding should be defined. The protein should quantitatively inhibit the cytotoxicity of Aβ to neuronally derived targets at concentrations consistent with those attainable in vivo. Most difficult, but certainly most definitive, would be the demonstration of complexes between the protein and Aβ isolated from the target tissue of animals expressing both transgenes and controls.

It would be desirable to determine whether introduction of the gene encoding the protein of interest somewhere in the course of the disease, rather than from conception, would have an impact on the development of the AD phenotype, suggesting that there might be some elements of these interactions that could be therapeutically exploitable. While it is conceivable that the observations made with respect to these four amyloid precursors are the result of ascertainment bias, until such bias is demonstrated the limits of the phenomena should be precisely defined and the underlying chemistry and biology thoroughly explored to determine if there is any “there” there.

View all comments by Joel Buxbaum