We invite you to participate in this "offline" Forum discussion led by Vincent Marchesi of Yale University. Coming from a different research field, Marchesi has in recent years followed the AD literature as closely as have few other outside observers. Earlier this summer, Marchesi published a perspective in PNAS (See Full Text [.pdf]), in which he called on investigators to consider the amyloid hypothesis in a new light.

The amyloid hypothesis commands majority support for its central claim that accumulation of the Aβ peptide plays an important role in AD. Yet Marchesi's ideas come at a time when fundamental questions about this hypothesis remain stubbornly unresolved, slowing down progress toward a deeper understanding and therapeutic approaches. Take advantage of this leisurely format to express your thoughts about Marchesi's article. Do you have supporting evidence? Contradictory evidence? What did Marchesi overlook? How could his hypothesis be tested? We invite you to send questions, comments, critiques, or kudos to Managing Editor Gabrielle Strobel. Gabrielle will post your commentaries and forward them to Marchesi or other participants for their responses.

See Full Text (.pdf). Marchesi VT. An alternative interpretation of the amyloid Abeta hypothesis with regard to the pathogenesis of Alzheimer's disease. Proc Natl Acad Sci U S A. 2005 Jun 28;102(26):9093-8. Copyright © 2005 National Academy of Sciences, U.S.A.


Marchesi, VT. An alternative interpretation of the amyloid Abeta hypothesis with regard to the pathogenesis of Alzheimer's disease. Proc Natl Acad Sci U S A. 2005 Jun 28 ; 102(26):9093-8. Abstract

Abstract: Alzheimer disease is a complex neurodegenerative process that is believed to be due to the accumulation of short, hydrophobic peptides derived from amyloid precursor proteins by proteolytic cleavage. It is widely believed that these Aβ peptides are secreted into the extracellular spaces of the CNS, where they assemble into toxic oligomers that kill neurons and eventually form deposits of senile plaques. This essay explores the possibility that a fraction of these Aβ peptides never leave the membrane lipid bilayer after they are generated, but instead exert their toxic effects by competing with and compromising the functions of intramembraneous segments of membrane-bound proteins that serve many critical functions. Based on the presence of shared amino acid sequences containing GxxxG motifs, I speculate that accumulations of intramembraneous Aβ peptides might affect the functions of amyloid precursor protein itself and the assembly of the PS1, Aph1, Pen2, nicastrin complex.

For further background, read this news summary of the PNAS article:

Marchesi sets up his argument by first laying out a brief synopsis of the main points of consensus in the field. For example, the plaques scattered throughout various brain areas of AD patients comprise many components, most prominently the Aβ40 and 42 peptides generated by cleavage of the APP transmembrane protein. The concerted action of the secretases BACE and γ-secretase releases Aβ peptides, and it is their overproduction or underremoval that is thought to lead to accumulation, extracellular aggregation, neuronal dysfunction, and eventually neuronal death. While Aβ's involvement in the disease is beyond serious dispute, how and where it acts remains unclear, Marchesi writes. Recent research has cast doubt on the conventional notion that the deposits are to blame for the early disease processes, and attention has shifted toward small forms of Aβ, often called soluble oligomers. How they act remains mysterious, and it is this question Marchesi's new ideas address.

Next, Marchesi notes that the fewer than 5 percent of early-onset AD cases who have mutant forms of APP or presenilin—widely thought to encode part of the γ-secretase—probably suffer from an accelerated form of the same underlying pathogenic process that operates in the more prevalent sporadic, late-onset forms. Transgenic animals expressing a variety of mutant forms alone or in combination now exist. They all develop massive Aβ deposits that resemble human pathology, but do not show the other pathologic hallmark of AD, that is, neurofibrillary tangles made of the protein tau. The animals' neurologic and neurodegenerative defects are subtle and vary from strain to strain. A triple transgenic mouse strain expressing mutant versions of APP, presenilin, and tau does develop amyloid deposits and neurofibrillary tangles. In this way, it models the human disease more fully, though it's worth noting that no human with such a heavy genetic burden has as yet been described (Oddo et al., 2003).

The triple transgenic mouse strain confirmed an earlier observation by others that the initial manifestations of Aβ accumulation begin inside neurons, not in extracellular spaces. The earliest detectable material resides in membrane compartments, possibly lysosomes or endosomes. Marchesi further cites a separate claim that APP molecules exist as homodimers inside neuronal plasma membranes in the brain (Scheuermann et al., 2001). The APP dimers are sequestered away in specific "raft" domains that are enriched for cholesterol and sphingolipids, and are thought to affect the regulation of protein-protein interactions. Moreover, these neuronal membrane patches also appear to contain Aβ dimmers (Kawarabayashi et al., 2004), and it is this observation from which Marchesi develops his hypothesis. While its discoverers interpret the intramembraneous Aβ dimers as being on their way to secretion and extracellular accumulation, Marchesi proposes that they could just as well stay inside the neuronal membrane for long periods of time.

To support this notion, Marchesi compares the intramembraneous APP sequence to that of another dimerizing transmembrane protein that is better studied. He suggests that both proteins derive their intramembraneous stability in a similar way, whereby a shared GxxxG motif recruits van de Waals forces and hydrogen bonds such that the transmembrane helices can pack closely and remain dimerized.

There already is a hypothesis dealing with Aβ inside neuronal membranes. It holds that secreted Aβ reinserts itself into the membrane in a channel-like structure (see ARF Live Discussion). In his perspective, Marchesi points to structural and biochemical questions that it still needs to address. He further argues that all arguments supporting the claim that Aβ peptides enter membranes from the outside equally well support the notion that they need not exit the membrane in the first place. They do eventually accumulate outside cells as the disease progresses, but perhaps they are not promptly secreted merely as a consequence of APP cleavage. In fact, the simplest interpretation of Kawarabayashi et al. is that secretase cleavage of APP dimers generates Aβ dimers, of which a significant fraction remains in the membrane, Marchesi contends.

Next, Marchesi asks how Aβ peptides accumulating inside membranes could affect the function of neurons. While APP dimers are anchored to particular sites within the membrane, Aβ peptides are not, and presumably would drift through the plane of the membrane. "It is easy to imagine how such peptides could influence the behavior of intramembraneous segments of receptors of channels, or even intramembraneous segments of enzymes like the presenilins and other secretases," Marchesi writes. He speculates that Aβ peptides inside the membrane could compete with normal APP dimer formation, displacing full-length APP monomers and creating chimeric dimers via their common GxxxG domains. These could be preferentially cleaved to generate more intramembraneous Aβ.

Alternatively, Aβ peptides might destabilize the γ-secretase complex, Marchesi speculates. The GxxxG motif is important in the transmembrane segment of Aph1, a protein that stabilizes the γ-secretase complex by linking its components together. Intramembraneous Aβ peptides could associate with Aph1 or compete for its binding partners, forming heterodimers of various sorts. This is not unheard of in membrane biochemistry. This second speculation could shed new light on a related debate within the field, Marchesi notes. The debate swirls around the question of whether presenilin mutations in familial AD cause well-documented increases in Aβ levels by way of a gain of function, or maybe also through a partial loss of regulatory functions of the γ-secretase complex.

There are broader implications of this scenario, Marchesi notes. It has become clear in recent years that many enzyme reactions occur within membranes, some through a process called regulated intramembraneous cleavage (RIP). Signal transduction mechanisms underlie control by RIP, for example, Notch cleavage. And the γ-secretase complex affects neuronal dysfunction in ways other than generating excessive amounts of Aβ. If tested and confirmed, this perspective of Aβ pathogenesis would call for a new therapeutic approach that aims to inhibit interactions between hydrophobic peptides within a bilayer. While that prospect seems daunting, other fields have already begun exploring it with some success, Marchesi writes. For example, two labs have described peptide segments that correspond to the intramembraneous domain of the transmembrane proteins PDGF or the Erb B receptor, respectively, and modulate their dimerization (Freeman-Cook et al., 2004; Bennasroune et al., 2004). Another possibility worth exploring lies in modulating the lipid membrane itself, Marchesi concludes.—Gabrielle Strobel.


  1. As is often the case, an outsider can give fresh perspectives in ways that someone from the inside has difficulty seeing. In this article, Dr. Vincent Marchesi, a former member of the Alzheimer Research Forum Advisory Board, presented a very thoughtful perspective and hypothesis concerning Aβ-mediated toxicity in brain. The premise with which Dr. Marchesi approached this article is that while the evidence linking Aβ peptide to "the pathogenesis of AD is substantial," how these peptides may be toxic in brain is far from resolved. While current investigators favor the model that neurons process APP and release the cleaved Aβ peptides into the extracellular space where they can form cytotoxic oligomers and aggregate into amyloid deposits, this scenario is indeed far from proven. For example, the Golde laboratory recently showed quite definitively that Aβ42 is necessary for parenchymal amyloid deposits when they creatively fused the Aβ domain to the Bri gene and expressed this construct in mice (1). However, these animals, in spite of substantial amyloid load in brain, have a normal lifespan and do not show "obvious behavioral abnormalities." If this is true, then it suggests that lots of extracellular Aβ, whether oligomers or aggregated fibrils, are insufficient to injure neurons or cause synaptic dysfunction.

    The central thesis of Dr. Marchesi, then, is that following cleavage of APP by various secretases or other proteases, part or all of the APP transmembrane domain remains within the membrane. The retained membrane segment containing part of the Aβ sequence subsequently interacts with APP and other membrane proteins and, in so doing, perturbs the normal processing of these proteins. Without being specific, Dr. Marchesi speculates that the function of cell surface receptors, ion channels, etc., can be altered by interacting Aβ fragments. The premise behind this novel idea is that both Aβ and APP exist in dimers, presumably within plasma membrane. The APP and Aβ dimers are held together noncovalently through the consensus GxxxG motif that is present in many single transmembrane proteins, including APLP1, Aph-1, and glycophorin A. Indeed, he speculates that APP may be cleaved by secretases as a dimer and the resultant Aβ peptides would remain as dimeric assemblies. The finding that APP dimerization enhances Aβ production is consistent with this view (2). Further, presenilin has been proposed to exist as a dimer in the γ-secretase catalytic core (3). Admittedly, the latter does not mean the APP substrate is also in dimeric form, but then why not? In this model, a portion of Aβ following secretase cleavages remains in, and, in fact, never leaves the membrane. In a postmitotic neuron, Aβ could be stuck in the membrane for a long time, enough to create havoc on neuronal function as it builds up over time.

    One can be nitpicky about details of this model, especially where supportive evidence is wanting. But, as a speculative position paper, there is a lot to like about this model. For one, this hypothesis is reminiscent of cytotoxicity mediated by the APP C-terminal fragments first proposed by Neve and Yanker, as well as by Fukuchi and Martin, and others (4,5). These are membrane-retained APP fragments created after a- and especially β-secretase cleavages. While there is no consensus as to how C-terminal APP fragments are toxic, their levels are increased in cells expressing various familial AD-associated APP and presenilin mutations. Furthermore, work from Fred Van Leuven's and Jie Shen's laboratories has shown that loss of PS1 activity in adult brain did not rescue long-term behavioral deficits in the animals even though amyloid production was virtually shut down (6,7). Whether neuronal dysfunction is due to the accumulation of the C99 APP C-terminal fragment is unclear, but is certainly consistent with the putative neurotoxicity of these peptides and quite possibly with Marchesi's model, as well. Second, we and others have proposed a model of cellular toxicity mediated by APP dimerization (8-10). Thus, APP dimers may be bad for the cell by augmenting Aβ production (both released and membrane-retained) and activating signals that can result in neuronal death. Finally, Dr. Marchesi cautions that experimental therapies aimed at reducing soluble Aβ may not be targeting the most appropriate sites. Though speculative, could this be an explanation for the lack of clinical response in the aborted active Aβ vaccination trial from Elan?

    In sum, perhaps this idea will be most appealing to those of us who find the myriad of proposed pathways to explain Aβ toxicity rather unsatisfying. Secreted Aβ, whether soluble or oligomeric, could be toxic in vivo, but it should not be so hard to demonstrate this. In the end, maybe this is similar to the story of the drunk looking for his car keys under the street lamp at night. When asked why he was looking there, he replied that he was there because there was light for him to look around even though he dropped his keys somewhere over in the darkened field. So as suggested by Dr. Marchesi, could we have been looking at the wrong place, after all? Or, to paraphrase a now famous political statement: "It's the membrane, stupid."


    . Abeta42 is essential for parenchymal and vascular amyloid deposition in mice. Neuron. 2005 Jul 21;47(2):191-199. PubMed.

    . Homodimerization of amyloid precursor protein and its implication in the amyloidogenic pathway of Alzheimer's disease. J Biol Chem. 2001 Sep 7;276(36):33923-9. PubMed.

    . A presenilin dimer at the core of the gamma-secretase enzyme: insights from parallel analysis of Notch 1 and APP proteolysis. Proc Natl Acad Sci U S A. 2003 Oct 28;100(22):13075-80. PubMed.

    . Alzheimer's disease: a dysfunction of the amyloid precursor protein(1). Brain Res. 2000 Dec 15;886(1-2):54-66. PubMed.

    . Expression of a carboxy-terminal region of the beta-amyloid precursor protein in a heterogeneous culture of neuroblastoma cells: evidence for altered processing and selective neurotoxicity. Brain Res Mol Brain Res. 1992 Nov;16(1-2):37-46. PubMed.

    . Neuronal deficiency of presenilin 1 inhibits amyloid plaque formation and corrects hippocampal long-term potentiation but not a cognitive defect of amyloid precursor protein [V717I] transgenic mice. J Neurosci. 2002 May 1;22(9):3445-53. PubMed.

    . Loss of presenilin function causes impairments of memory and synaptic plasticity followed by age-dependent neurodegeneration. Neuron. 2004 Apr 8;42(1):23-36. PubMed.

    . Amyloid beta protein toxicity mediated by the formation of amyloid-beta protein precursor complexes. Ann Neurol. 2003 Dec;54(6):781-9. PubMed.

    . A monoclonal antibody to amyloid precursor protein induces neuronal apoptosis. J Neurochem. 2000 Jun;74(6):2331-42. PubMed.

    . The cytoplasmic domain of Alzheimer's amyloid-beta protein precursor causes sustained apoptosis signal-regulating kinase 1/c-Jun NH2-terminal kinase-mediated neurotoxic signal via dimerization. J Pharmacol Exp Ther. 2003 Sep;306(3):889-902. PubMed.

  2. It is quite surprising that, despite a huge number of experiments done and papers published, the molecular events responsible for the Alzheimer's phenotype remain obscure. This article is a stimulating exercise that gives support to an alternative explanation for the possible toxic effects of Aβ peptides. In a nutshell, Dr. Marchesi proposes that Aβ peptides exert their toxicity within the cell membranes, where they remain entrapped as dimers after the cleavage of APP. Therefore, extracellular oligomers or larger aggregates could have no toxic effects: Plaques could be even a "safe" storage of toxic peptides, which prevent their reassociation with the membranes.

    I'll add three further points to the interesting speculation of Marchesi. First, we should consider that Aβ42 has two extra residues (Ile and Ala) at its C-terminus. These could render more stable, compared to Aβ40, the association of the peptide itself with the hydrophobic environment of the membrane bilayer. This could explain why Aβ42 seems to be so crucial for AD pathogenesis.

    A second point is that the so-called p3 peptide, always considered not to be toxic, could on the contrary be even more toxic than Aβ. It is generated through the concerted cleavage of APP by a- and ?-secretase, and, as predicted by Marchesi's hypothesis, it could remain inserted in the membrane like Aβ. Furthermore, given that it lacks most of the hydrophilic or charged residues present in the N-terminal half of Aβ, its interaction with the membrane could be more stable than that of Aβ.

    Thirdly, we should observe closely the efforts of numerous research groups that are trying to address the possible relationship between lipid rafts, cholesterol content of membranes, ApoE alleles, and AD. Several clinical studies indicate that individuals treated with statins (cholesterol-lowering drugs) show a low incidence of AD (1). These results apparently conflict with the observation that mice treated with statins show a dramatic increase of Aβ production and of amyloid plaques. Marchesi's hypothesis could easily explain this contradiction: The decrease of membrane cholesterol provokes a release of Aβ peptides from the membranes, so that they accumulate in the extracellular space where they are not toxic. In this way, the subjects treated with statins could gain protection from AD while the mice treated with the same drug have more Aβ and more plaques.

    I think that this hypothesis deserves to have the field put it to the test with experiments.


    . Statins and the risk of dementia. Lancet. 2000 Nov 11;356(9242):1627-31. PubMed.

    . Lovastatin enhances Abeta production and senile plaque deposition in female Tg2576 mice. Neurobiol Aging. 2003 Sep;24(5):637-43. PubMed.

  3. In his PNAS Perspective, Vincent Marchesi, in characteristic fashion, provides us with substantial food for thought. Dr. Marchesi presents a cogently argued hypothesis that provides an alternative—or perhaps an addition—to the concept that Aβ oligomers and/or amyloid fibrils injure neurons from without. Based on reported evidence that APP and Aβ can each occur as dimers in cholesterol-rich "lipid raft" domains of neuronal membranes, Dr. Marchesi proposes that APP dimers may be processed as such by the β- and γ-secretase s to yield Aβ dimers which, at least in part, remain in the membrane bilayer. He then suggests specific ways in which intramembranous dimers of Aβ could compromise the function of numerous other transmembrane proteins that share with Aβ the hydrophobic GxxxG motif, including the very proteins that help generate Aβ in the first place (presenilin and Aph-1). If membrane-retained rather than secreted Aβ is principally responsible for compromising neuronal function, Dr. Marchesi notes, perhaps experimental treatments designed to lower extracellular Aβ may miss their mark.

    That such a biological scenario could occur seems entirely plausible, and Dr. Marchesi marshals supportive evidence from the literature on the properties of transmembrane segments of proteins, including his own pioneering studies of glycophorin A. But as he also emphasizes, there is as yet no direct experimental evidence that APP dimers are converted per se to Aβ dimers, that Aβ occurs preponderantly as dimers in neuronal membranes, and that these dimers are retained within the membrane for periods sufficiently long to be responsible for triggering the neuronal alterations one observes in Alzheimer disease. This is as it should be for a provocative new hypothesis. I join Drs. Koo and Russo in thanking Dr. Marchesi for creatively pondering the unsolved issue of Aβ's pathogenic mechanism. But I would also like to provide responses to Dr. Marchesi's article that address both certain specific statements he makes and the broader therapeutic implications of his model.

    First, Dr. Marchesi states that the findings of LaFerla and colleagues in the triple transgenic mouse model "confirm what many previous investigators have noted, that the initial manifestations of Aβ accumulation begin inside neurons, with the earliest detectable material located inside membranous compartments," including endosomes and lysosomes. In this regard, one needs to distinguish data about the sites of normal Aβ generation throughout life, which include both secretory and endosomal compartments, from data about where Aβ can first be shown to accumulate abnormally during the prodromal stages of AD. As regards the latter, while early intracellular accumulation can be observed in overexpressing disease models such as the triple transgenic mouse (1) and conventional APP transgenic mice (2), it is far more difficult to ascertain the temporal pattern in humans, as we rarely examine the brain tissue before the end of the disease. In the case of human trisomy 21, however, one can detect many diffuse extracellular Aβ deposits in Down subjects as early as 10-12 years of age, and sensitive immunohistochemistry has generally not revealed clear-cut intraneuronal Aβ? aggregates at this early juncture [see, for example, (3)]. Whereas intracellular Aβ can be detected by immunoelectron microscopy in postmortem AD cortex [e.g., (2)], innumerable extracellular deposits are also present. A study of primary neurons cultured from Down syndrome brains observed robust intracellular Aβ staining (4), but this could be due to the increased expression of APP in this particular situation; in conventional AD, wild-type APP is expressed at normal levels. Whereas intracellular tau aggregates (i.e., neurofibrillary tangles as well as "pre-tangle" tau accumulation) are readily detectable by immunohistochemistry in AD and Down postmortem brains, it has been difficult to similarly detect clear-cut intraneuronal Aβ deposits. In our laboratory, Dominic Walsh first detected soluble Aβ dimers and trimers inside certain APP-expressing cultured cells, apparently before they were exported into the medium, leading to our proposal that Aβ dimerization may begin within the vesicles in which it is generated (5). But this does not tell us where Aβ oligomers first accumulate to levels that might actually be injurious to neurons. Therefore, I don't believe that existing evidence supports Dr. Marchesi's rather definitive statement that I quote above. In this sense, I don't concur with Dr. Marchesi's inference that "Aβ peptides do eventually accumulate outside cells as the disease progresses." Rather, they can accumulate outside cells early on (long before clinical symptoms), and longitudinal studies of Down syndrome brains of increasing age as well as APP transgenic mouse brains support this conclusion.

    Dr. Marchesi states that "it is still a mystery how PS1 mutations cause the disease" and wonders how some 150 different mutations scattered throughout the molecule can achieve the same result (AD). I would respond that missense mutations are known to subtly (or sometimes markedly) alter the native conformation of proteins. Because many of the PS1 mutations have been shown to increase Aβ42 generation, sometimes at the expense of Aβ40 generation, I think a highly plausible model is that their conformational effects increase the likelihood that the Aβ42-43 peptide bond within the APP transmembrane domain interacts efficiently with the two catalytic aspartates of presenilin. Indeed, the elegant FRET/FLIM analyses of Oksana Berezovska and colleagues support just such a PS1-APP conformational interaction model (6).

    Dr. Marchesi's proposed feedback mechanism in which membrane-retained Aβ dimers might interact with PS1 and Aph-1 to destabilize the ?-complex itself seems to me to lack molecular specificity; that is, many other hydrophobic transmembrane fragments generated from γ-secretase substrates which have GxxxG motifs could do this, as well. Indeed, the more hydrophobic APP derivative, p3 (arising from a- and γ-secretase cleavages) would presumably be an even better candidate than Aβ for such an effect. While one can certainly not exclude such a feedback mechanism, I think it is biologically more plausible that Aβ- or p3-like substrate products of intramembranous cleavage are efficiently released into the luminal space of vesicles, where they might begin to oligomerize (due to very high local concentrations) and then be retained in part but also be secreted in part.

    This brings me to the issue of the likelihood of retention of Aβ, whether in dimeric or monomeric form, within the membrane. Again, present information cannot exclude this occurring, at least in part. But the processing of a large and growing number of substrates by γ-secretase has led to the hypothesis that γ-secretase is conserved and ubiquitous because it efficiently removes single-transmembrane proteins from the bilayer. So, beyond its ability to release potential signaling domains from some of its many substrates (e.g., Notch), γ-secretase is viewed increasingly as a special type of "housekeeping" protease that serves to rid the membrane of otherwise long-lived transmembrane domains. If so, do the two resulting cleavage products get released quantitatively from the membrane? In healthy cultured cells expressing APP, one can detect abundant monomers of Aβ in the medium (as one can in plasma and CSF), and size-exclusion chromatography under non-denaturing conditions reveals that by far the major form of Aβ found in the medium is monomer [see, for example, Figure 2 in (7)]. Robust release of Aβ monomer following γ-secretase processing of endogenous APP has also been shown in primary human neurons (8). The CSF levels of secreted Aβ42 monomers (as measured by sandwich ELISAs that don't efficiently detect oligomers) are often significantly decreased in AD patients, even early in their clinical course, suggesting that extracellular monomer levels reflect the progressive accumulation of Aβ in myriad insoluble deposits in the cortex. These and other published data suggest to me that much of Aβ in the human brain occurs as soluble monomers (and small oligomers) that are released from cells. But again, such arguments cannot exclude a role for some Aβ peptides that remain in the membrane to contribute to the neuronal insult, whether as monomers or oligomers.

    Regardless of one's perspective, Dr. Marchesi's intriguing hypothesis that membrane-retained Aβ dimers contribute to neuronal injury should be actively pursued experimentally. One way to do this might be to extend FRET/FLIM approaches to learn whether APP can be observed to undergo dimerization in living cells. For example, one might express two differentially tagged fluorescent APP molecules that can be temporally induced (e.g., via Tet-regulated constructs) in neuronal cells and then observe whether—and how quickly—they show FRET. Perhaps the same could be done to detect Aβ interactions within vesicular compartments of the cell or at the plasma membrane.

    Finally, I conclude with a comment about the therapeutic implications of Dr. Marchesi's model. I don't concur with his interpretation that "many new therapies designed to reduce Aβ levels in AD patients are being proposed, but efforts to reduce peptide levels in blood and tissue spaces may not target the most toxic factors." Inhibitors of β- or γ-secretase currently under development are intended to lower both intracellular and cell-surface Aβ production (i.e., they are cell-penetrant), and they should be able to decrease the levels of intramembranous Aβ dimers. And immunotherapeutic approaches, which have shown salutary behavioral effects in AD mouse models and early evidence of clinical and neuropathological benefits in AD patients (9-11), could lower cerebral Aβ burden in a way that may secondarily affect intraneuronal levels, as well. Although specific anti-oligomer approaches are theoretically attractive, these may be more difficult to achieve than the foregoing examples now entering the clinic.


    . Intraneuronal Abeta causes the onset of early Alzheimer's disease-related cognitive deficits in transgenic mice. Neuron. 2005 Mar 3;45(5):675-88. PubMed.

    . Intraneuronal Alzheimer abeta42 accumulates in multivesicular bodies and is associated with synaptic pathology. Am J Pathol. 2002 Nov;161(5):1869-79. PubMed.

    . Sequence of deposition of heterogeneous amyloid beta-peptides and APO E in Down syndrome: implications for initial events in amyloid plaque formation. Neurobiol Dis. 1996 Feb;3(1):16-32. PubMed.

    . Altered metabolism of the amyloid beta precursor protein is associated with mitochondrial dysfunction in Down's syndrome. Neuron. 2002 Feb 28;33(5):677-88. PubMed.

    . The oligomerization of amyloid beta-protein begins intracellularly in cells derived from human brain. Biochemistry. 2000 Sep 5;39(35):10831-9. PubMed.

    . Familial Alzheimer's disease presenilin 1 mutations cause alterations in the conformation of presenilin and interactions with amyloid precursor protein. J Neurosci. 2005 Mar 16;25(11):3009-17. PubMed.

    . Natural oligomers of the amyloid-beta protein specifically disrupt cognitive function. Nat Neurosci. 2005 Jan;8(1):79-84. PubMed.

    . Amyloid beta-peptide is produced by cultured cells during normal metabolism. Nature. 1992 Sep 24;359(6393):322-5. PubMed.

    . Antibodies against beta-amyloid slow cognitive decline in Alzheimer's disease. Neuron. 2003 May 22;38(4):547-54. PubMed.

    . Clinical effects of Abeta immunization (AN1792) in patients with AD in an interrupted trial. Neurology. 2005 May 10;64(9):1553-62. PubMed.

    . Neuropathology of human Alzheimer disease after immunization with amyloid-beta peptide: a case report. Nat Med. 2003 Apr;9(4):448-52. PubMed.

  4. The very interesting review of Dr. Vincent Marchesi deals with the notion that Aβ, once generated by diverse secretases, never leaves the membrane and exerts its toxicity by yet unknown mechanisms. Supporting this view, we and others have seen evidence that intraneuronal accumulation of Aβ42 is the main trigger for the pathological events leading to neuron loss and brain atrophy. Increased Aβ42 has been observed in postmortem brains of patients with beginning Alzheimer disease and Down syndrome, and several APP transgenic mouse models.

    Increasing intraneuronal Aβ42 leads to an age-dependent neuron loss in hippocampus in two different APP/PS1 transgenic mouse models with no correlation to plaque load, that is, extracellular Aβ. Intraneuronal Aβ42 is primarily found in multivesicular bodies in these mice. Intraneuronal Aβ42 accumulation in the somatodendritic compartment may have an influence on axonal trafficking and integrity, an issue which is currently also debated. However, whether Aβ42 is closely associated with intracellular membranes or plasma membrane in these mice is presently unclear.

  5. Response by Vincent Marchesi to comments through 28 September 2005

    Dr. Russo points out that if Aβ peptides or fragments of them remain within the lipid bilayer after their generation, Aβ42, with two additional hydrophobic residues, should be more stable within the bilayer than Aβ40, a factor that could account for its greater toxicity. He also raises the interesting possibility that the P-3 peptide, widely assumed to be nontoxic, whose sequence appears below, might be even more likely to partition within the bilayer, since it has the same hydrophobic core but lacks many of the polar residues of either Aβ fragment. If the β-secretase pathway is favored, as many suspect, there would be much less P-3 than either of the Aβs, but P-3 might be doing more than we realize. For one thing, its primary sequence exactly matches the homologous segments of the Aβs, as shown below, and it should, in principle, dimerize with either monomer. If the P-3 peptide is less toxic, for whatever reasons, could it function to neutralize the Aβs, acting as some kind of intramembranous chaperone? The effects of cholesterol on Aβ production are obviously complicated, as Russo points out, but the properties of lipid rafts and sterols must surely influence intramembranous transactions of the type I postulate.


    Amino acid sequences of the Aβ42 peptide generated by the combined actions of β-secretase and γ-secretase, and the P-3 peptide generated by the combined actions of a-secretase and γ-secretase. P-3 is thought to be nontoxic.

    Dr. Koo reminds us that the potentially toxic role of APP C-terminal fragments has a long history and is not consistent with the prevalent view that soluble Aβ aggregates are the major toxic principles. He wonders whether they represent a source of hydrophobic peptides of varying composition that might have unsuspected functions within the membrane interior. The idea that accumulations of C-terminal fragments might represent a reservoir of potentially toxic peptides is an intriguing speculation.

    Dr. Selkoe raises interesting questions regarding sites where Aβ peptides accumulate inside neurons, and he proposes that a distinction should be made between Aβ production that might be "physiological" as opposed to that which is "pathological." He suggests that conditions in which Aβ is overexpressed (various transgenic models) might not reflect the sequence of events that takes place in the human condition that leads to AD, about which we have only fragmentary information. He believes that results from the study of brains of patients with trisomy 21 may be more revealing than the animal models, in that abundant amounts of extracellular Aβ are present at early ages in such patients. Very little Aβ is seen intracellularly, he contends, and that which is observed is believed to be due to the presence of Aβ dimers in secretory vesicles. There may indeed be important distinctions between the causes of dementia in Down syndrome and those which cause AD in non-trisomy individuals, but I don't think they contradict in a fundamental way the widely held view that Aβ peptides accumulate inside cells, both neuronal and non-neuronal, in the early stages of AD. Dr. Bayer's experience also supports this position.

    Regarding the "mystery" as to why so many mutations of PS1 cause disease (150 as of now), I don't think it's wise, in this case, to fall back on the overused speculation that conformational changes are responsible. In the study Dr. Selkoe cites to support this idea (1), labeled antibodies to different proteins were used as donor and acceptor ligands. Because the fluorophores are attached to bulky, flexible macromolecules, and the analysis is done on fixed, antibody-stained cells, one can fairly question whether this approach has the resolving power to reliably detect conformational changes within individual protein molecules. FRET/FLIM analysis is used routinely to detect larger-scale interactions between neighboring proteins, but conclusions drawn from its use to study intramolecular conformations must be confirmed in other ways. So I believe the question of conformational influences is still open.

    It is entirely reasonable to question how the interactions between hydrophobic peptides of similar sequences can be specific enough to account for the interactions I propose. There are now many examples where such interactions do indeed take place, so the ability to form stable associations is not in question. The GxxxG or GxxxA motifs that I propose to account, in part, for the ability of Aβ peptides to interact with other transmembrane proteins, if they are retained within the bilayer, are thought to be due to their ability to allow closer packing of neighboring polypeptides because they lack bulky side chains. I suspect that their ability to interact with specific proteins may be largely due to the fact that they are generated in close proximity to these potential partners. This is why I suggested that accumulations of such peptides, or fragments of them, might first interact with intact APP molecules or APH1, both of which are by definition at sites of Aβ generation. I have already noted, in response to the suggestion by Russo, that P-3 should also be one of these retained peptides. As suggested above, we should look into its fate more carefully than we have so far.

    Another potential target/partner for Aβ is the neuronal sorting receptor sorLA, which appears to be involved in APP trafficking. A recent report (2) indicates that sorLA, a type-1 transmembrane protein of unknown function, colocalizes with APP in endosomal, Golgi, and plasma membranes, and may in some way block or reduce Aβ production. It is interesting that this receptor also has a transmembrane segment similar to the C-terminal end of Aβ42, as depicted below.


    The C-terminal sequences of GVGFA (sorLA) and the GVVIA of Aβ overlap.

    Furthermore, Dr. Selkoe cites an abundance of experimental evidence that Aβ peptides of various states of aggregation are found in the media of cultured cells, CSF, and other body fluids. I do not dispute these findings. However, they do not bear on whether or not a fraction of such peptides, or fragments of them, also exists in the interior of membranes for periods long enough to affect membrane function. Dr. Selkoe's proposal that the γ-secretase complex is a special type of "housekeeping" protease is itself an interesting speculation, and others in the field share a similar view (Kopan and Ilagan, 2004). I think it likely that such elaborate molecular machinery has evolved to generate peptides with specific biological functions, and we should figure out what they might be doing. It also seems reasonable to assume that since these peptides are derived from the intramembranous domains of transmembrane proteins, and are generated within the lipid bilayer, chances are good that they are also doing something biologically important within the membrane itself.


    . Gamma-secretase: proteasome of the membrane?. Nat Rev Mol Cell Biol. 2004 Jun;5(6):499-504. PubMed.

    . Familial Alzheimer's disease presenilin 1 mutations cause alterations in the conformation of presenilin and interactions with amyloid precursor protein. J Neurosci. 2005 Mar 16;25(11):3009-17. PubMed.

  6. Dr. Marchesi provides refreshing insights into how accumulating membrane-associated Aβ may be involved in AD pathogenesis. For example, it is intriguing to consider that Aβ, which we thought accumulates in the inner aspect of outer limiting membranes of endosomes, may actually be embedded within the membrane bilayer. This could also illuminate why Aβ42 prefers to be retained in cells upon exocytosis of more soluble Aβ40 peptides.

    Marchesi's Perspective article highlights the importance of investigating the biology of Aβ42 accumulating within neurons. What's the evidence for a detrimental role of intraneuronal Aβ, also in human AD? In short, by immuno-EM, accumulating intraneuronal Aβ42 and especially Aβ oligomers are associated with subcellular destruction, which equals neuronal dysfunction. This occurs in the absence of plaques. There is so much that needs to be uncovered, including how extracellular Aβ influences intracellular Aβ, the biology of APP/Aβ, PS1 and other AD-linked proteins, especially within synaptic membrane compartments, and the molecular mechanism by which accumulating intraneuronal Aβ initiates synaptic dysfunction. For a detailed review on intracellular Aβ biology, and a historical perspective, see Gouras et al., 2005.


    . Intraneuronal Abeta accumulation and origin of plaques in Alzheimer's disease. Neurobiol Aging. 2005 Oct;26(9):1235-44. PubMed.

  7. The 12 Faces of Amyloid
    The problem of the amyloid peptides always reminds us of an object in the Gallo-Roman museum in Tongeren, the oldest city of Belgium, located about 30 miles east of Leuven. The pentagonal dodecaeder is made of bronze (the museum shop sells tin replicas for 35 euros), and 90 such artifacts were found north of the Alps, in places ranging from England to the Balkan. They occurred in diverse archeological sites, i.e., military camps, public baths, city houses, theaters, graves, and even buried with a treasure of coins.


    Nobody has the slightest idea what they are for, or what purpose they serve, either as a tool, jewel, symbol, toy, relic, instrument. The hypotheses and guesses are as diverse as they are wild. The parallel with the amyloid peptides is evident, although it's not perfect since the 90 known dodecaeders are all very similar and their 12 faces are identical—as opposed to the amyloid peptides that differ considerably depending on who is looking and through what tool!

    Twenty-odd years after Glenner and Wong published their primary amino acid sequence, the amyloid peptides remain enigmatic, largely because experimental proof of any proposed theory or mode of action is hard to come by. Circumstantial evidence and indirect correlations are the best we've been able to do so far. Dr. Marchesi's present hypothesis is no exception, and it appears to some extent to be based on the correlation of dimeric amyloid peptides in lipid rafts and the behavioral traits of APP transgenic mice (Kawarabayashi et al., 2004). Nevertheless, the hypothesis is interesting and innovative in the twist that the toxic species, be it Aβ-oligomers or β-CTF, need not leave the membrane to exert its evil action—proposing that only, or particularly, that which remains in the membrane is neurotoxic.

    One of the early hypotheses pertains to amyloid peptides that can aggregate and become embedded in the membrane—or vice versa—to form pores that allow rather nonspecifically the exchange of undefined species of chemicals and ions between cellular compartments and with the neuronal environment. A recent revival (Lashuel, 2005) and an eventual refinement to include "synaptic" effects (Selkoe, 2003; Tanzi, 2005) are largely in line with the hypothesis on which we are invited to comment here.

    On the one hand, one wonders how and why the tails of the amyloid peptides leave their natural environment of the membrane. The most logical mode of exit appears during PS1-mediated γ-secretase-cleavage. There, the β-CTF substrate is translocated into the catalytic pore, allowing entry of water to hydrolyze the peptide bond made scissile in and by the γ-secretase-secretase complex. For a brief moment, the C-terminal tail of the nascent amyloid peptide could be bathing in an aqueous environment, although it need not be the 55.6 M of pure water. Release into the lumen of ER or Golgi stacks, of late endosomes, or into extracellular space would then help to dispose of the amyloid peptides. This could happen by simple dilution or aggregation, by diffusion and ApoE-mediated drainage into CSF and blood, by endocytosis and phagocytosis, by specific proteolysis or systematic degradation, or by other means.

    The case for intraneuronal amyloid peptides as culprit in AD continues to pit its defenders and doubters against each other, again mainly due to the lack of "isolated" support. That is, the models are incomplete, and experimental evidence that only intraneuronal Aβ is present and at work remains insufficient. Nevertheless, intracellular Aβ is much more closely located near tau protein and the kinases needed to phosphorylate tau to induce intraneuronal tangles. These then wreck the cell's interior organization, beginning with axonal transport and thereby causing synaptic defects (Billings et al., 2005; Oddo et al., 2004; Terwel et al., 2002).

    Amyloid peptides likely exert pleiotropic neurotoxic effects. This includes effects of the extracellular soluble oligomers on synaptic plasticity, such as LTP, in vitro and in vivo (Klyubin et al., 2005; Walsh et al., 2004; Dodart et al., 2002; Dewachter et al., 2002). This has been demonstrated convincingly, and presented well and extensively by the comment of Dennis Selkoe [a possible role for the same kinases that are also involved in the phosphorylation of tau was demonstrated in oligomer-induced synaptic defects (Wang et al., 2004)], thereby opening the scene for cross-talk between the two defining types of the AD pathology.

    From our view, a most interesting point of Dr. Marchesi's hypothesis is the inclusion of the neurotoxic properties of the CTF derived from APP, particularly β-CTF. We were the first to draw attention to the fact that these CTFs might be as bad or even worse than the peptides in neurons in vivo (Dewachter et al., 2002), and not just when overexpressed in non-neuronal cells, as claimed before. Completely in line with our findings, data published earlier this year (Saura et al., 2005) demonstrate that β-CTFs accumulate in the brain of PS1-deficient mice and can explain the lack of improvement of cognitive and behavioral defects in APPxPS1 (n-/-) mice. That is, these PS-knockout/APP transgenics remain as impaired as the parental APP Tg mice, even though their brain levels of amyloid peptides are decimated. As argued on these pages before, CTFs will inevitably become concentrated in their "neurons of origin" since they cannot leave the membrane, while the amyloid peptides can. Nevertheless, the pleiotropic actions of presenilin 1, both as γ-secretase-secretase with its many substrates and by its action in calcium homeostasis (Herms et al., 2003) and at the synapse (Ris et al., 2003) make these experiments very suggestive, but not conclusive.

    We do declare a point of controversy with Dr. Marchesi's hypothesis, and are at variance also with Dr. Russo's comment. It concerns the innocuous P-3 peptides that are proposed to be toxic, perhaps even more so than amyloid peptides. Experimental evidence in ADAM10-deficient mice convincingly demonstrates that increasing the a-secretase cleavage of APP ameliorates the cognitive, behavioral, and synaptic defects of APP transgenic mice, despite accumulation of the P-3 peptides (Postina et al., 2004).

    The central point of membrane-embedded amyloid peptides acting from within the membrane is interesting and plausible. Even so, it will be difficult to demonstrate experimentally. We believe that the part of the hypothesis Dr. Marchesi calls "speculative," i.e., disruption of APP dimers and of the γ-secretase-secretase complex by amyloid peptides in lipid rafts, will actually be easier to address experimentally.

    And no, we have not yet seen convincing evidence for a physiological role of any of the amyloid peptides.

    The Tongeren dodecaeder remains as enigmatic as ever. The amyloid peptides have gained yet one more face to ponder.


    . Dimeric amyloid beta protein rapidly accumulates in lipid rafts followed by apolipoprotein E and phosphorylated tau accumulation in the Tg2576 mouse model of Alzheimer's disease. J Neurosci. 2004 Apr 14;24(15):3801-9. PubMed.

    . Membrane permeabilization: a common mechanism in protein-misfolding diseases. Sci Aging Knowledge Environ. 2005 Sep 21;2005(38):pe28. PubMed.

    . Aging, amyloid, and Alzheimer's disease: a perspective in honor of Carl Cotman. Neurochem Res. 2003 Nov;28(11):1705-13. PubMed.

    . The synaptic Abeta hypothesis of Alzheimer disease. Nat Neurosci. 2005 Aug;8(8):977-9. PubMed.

    . Intraneuronal Abeta causes the onset of early Alzheimer's disease-related cognitive deficits in transgenic mice. Neuron. 2005 Mar 3;45(5):675-88. PubMed.

    . Abeta immunotherapy leads to clearance of early, but not late, hyperphosphorylated tau aggregates via the proteasome. Neuron. 2004 Aug 5;43(3):321-32. PubMed.

    . Axonal transport, tau protein, and neurodegeneration in Alzheimer's disease. Neuromolecular Med. 2002;2(2):151-65. PubMed.

    . Amyloid beta protein immunotherapy neutralizes Abeta oligomers that disrupt synaptic plasticity in vivo. Nat Med. 2005 May;11(5):556-61. PubMed.

    . Deciphering the molecular basis of memory failure in Alzheimer's disease. Neuron. 2004 Sep 30;44(1):181-93. PubMed.

    . Immunization reverses memory deficits without reducing brain Abeta burden in Alzheimer's disease model. Nat Neurosci. 2002 May;5(5):452-7. PubMed.

    . Neuronal deficiency of presenilin 1 inhibits amyloid plaque formation and corrects hippocampal long-term potentiation but not a cognitive defect of amyloid precursor protein [V717I] transgenic mice. J Neurosci. 2002 May 1;22(9):3445-53. PubMed.

    . Conditional inactivation of presenilin 1 prevents amyloid accumulation and temporarily rescues contextual and spatial working memory impairments in amyloid precursor protein transgenic mice. J Neurosci. 2005 Jul 20;25(29):6755-64. PubMed.

    . Capacitive calcium entry is directly attenuated by mutant presenilin-1, independent of the expression of the amyloid precursor protein. J Biol Chem. 2003 Jan 24;278(4):2484-9. PubMed.

    . Capacitative calcium entry induces hippocampal long term potentiation in the absence of presenilin-1. J Biol Chem. 2003 Nov 7;278(45):44393-9. PubMed.

    . A disintegrin-metalloproteinase prevents amyloid plaque formation and hippocampal defects in an Alzheimer disease mouse model. J Clin Invest. 2004 May;113(10):1456-64. PubMed.

  8. Vincent Marchesi puts forward a potential alternative mechanism as to how the amyloid-β (Aβ) peptide may be involved in the pathogenesis of Alzheimer disease (AD). Dr. Marchesi proposes the possibility that Aβ peptides might be exerting their toxic effects by never leaving the membrane lipid bilayer after they are generated, and that in the membrane, they might exert their toxic effects by competing with and compromising the functions of intramembranous segments of membrane-bound proteins that serve many critical functions.

    This is an interesting hypothesis. Not being the first person to comment on the hypothesis, I note the thoughtful comments of the other scientists who have done so already. I try to add below some observations not noted by the other commentators. Dr. Marchesi first describes a series of observations described in the literature. While I think several of these observations are correct, I think there are a few points worth noting that I am not convinced occur in AD. For example, in the triple transgenic mouse strain produced by the LaFerla lab, some intraneuronal Aβ staining is noted prior to plaques. It is stated that this confirms an earlier observation by others that the initial manifestations of Aβ accumulation begin inside neurons, not in extracellular spaces. While some groups report this, it is not a universal finding by many other labs. Many groups do not find that Aβ accumulation begins intracellularly in either humans or APP transgenic mice, but that it begins in the extracellular space.

    The main aspect of the Marchesi hypothesis is based on the fact that not only APP but also Aβ may exist within the plasma membrane as a dimer in lipid rafts. That this occurs is based on a paper by Kawarabayashi and colleagues in 2004 in the Journal of Neuroscience. These investigators showed that a fraction of Aβ was found in lipid rafts in transgenic mice that overproduce human Aβ, and that coincident with Aβ accumulation, Aβ dimers began to be seen and increased in the lipid raft fraction. The data in the Kawarabayashi paper appear very solid. However, what is not clear is whether Aβ and Aβ dimers that are seen in lipid rafts after homogenizing the brain are actually present in rafts in vivo or get into rafts after homogenizing the tissue in the presence of detergents. One indication that the latter may be the case is that there was a large increase in Aβ and Aβ dimers seen in rafts coincident with the huge increase in Aβ in the brain that occurs with Aβ aggregation and plaque deposition (in the extracellular space). If some Aβ present in aggregates is relatively loosely associated with plaques, it is quite possible that in the presence of detergents, Aβ forms dimers in rafts only after tissue homogenization (i.e., an artifact of tissue homogenization).

    It is of note that a variety of experiments suggests that under normal conditions, the Aβ peptide is generated constantly by cells in the body, with highest production levels in the brain by neurons. When Aβ is generated in the CNS, it is secreted in a soluble form that is probably monomeric, which is then rapidly cleared. This clearance may occur by cells in the brain, by enzymes, but also in large part by being transported out of the brain interstitial space into the CSF and the blood. Interestingly, and despite the fact that Aβ can build up to very high levels in the AD brain and in the brain of APP transgenic mice with age (up to a few hundredfold higher than in a normal brain), studies in animals suggest that soluble Aβ is generated and cleared very rapidly with a half-life of about 2 hours. Thus, while the level in the CSF and probably the brain ISF under normal conditions is about 10 ng/ml, it is being produced, secreted, and cleared throughout life at very high rates.

    Dr. Marchesi proposes the possibility that Aβ peptides might be exerting their toxic effects by never leaving the membrane lipid bilayer after they are generated, and that in the membrane, they might exert their toxic effects by competing with and compromising the functions of intramembranous segments of membrane-bound proteins that serve many critical functions. It is certainly conceivable that Aβ being so lipophilic could enter the bilayer or remain in the bilayer. Alternatively, Aβ that is soluble in the extracellular space could be in equilibrium with Aβ in the bilayer and in rafts. Given the very high rates of Aβ metabolism, if there is a large amount of Aβ that never exits the bilayer and is present in rafts, turnover in this pool could similarly be quite rapid or it could be much slower. Experiments to directly test this should be possible, particularly in cultured cells that might enable one to test whether a membrane pool differs in its turnover from the extracellular, secreted pool. If so, this could give some insights into what might happen or lead to changes that are present in disease. Given the fact that we still do not understand exactly how Aβ leads to brain injury, it remains important to develop a better understanding of this phenomenon. Perhaps Dr. Marchesi's hypothesis will lead to some new suggestions for experiments that will get at this still-unresolved issue.

  9. I am very glad to see hypotheses launching new ideas on the pathogenesis of AD. My content is even greater that this interesting hypothesis includes the role of cellular membranes in this process, which I find crucial, as well.

    Having said that, I also share many of Dr. Selkoe's reservations towards Dr. Marchesi's concept. I wouldn't rule out the possibility that some fraction of Aβ may remain in the cellular membrane, followed by processes of dimerization and ensuing contributions to toxicity. Nevertheless, the fact is that low soluble products of APP cleavage accumulate mainly in the extracellular space. The long-lasting process of such accumulation may finally lead to formation of senile plaques. If presumed intramembrane Aβ accumulation had a big neurotoxic effect, one can expect that it would result in neuronal death before neurons would be able to produce great amounts of extracellular deposits, that is, unless intramembrane accumulation intensifies in later stages of pathological Aβ overproduction and/or low clearance. I can even give "a prompt" to Dr. Marchesi that it may be a consequence of advanced deterioration of neuronal membranes resulting from a longer disease process and Aβ fragments having more affinity for deteriorated neuronal membranes.

    What is particularly missing in this concept is a lack of explanation of the presumed role of APP enzymatic cleavage. I introduced the concept of cholesterol in cholinergic rescue (Bohr, 2004; Bohr, 2005), which is partially in line with Dr. Koudinov's ideas (see, for example, Koudinov and Koudinova, 2005). According to this concept, Aβ plays an important role in neuroprotection by taking part in supplying neuronal membranes in cholesterol that is delivered by ApoE-containing lipoproteins. This idea has recently received support in findings showing different efficacy of Aβ in ApoE incorporation depending on the isoform of this protein. The authors namely have found that ApoE2 facilitates this process, whereas ApoE4 slows it down (Hone et al., 2005).

    If this thesis is true, then radical "fighting" with Aβ might not be a very good approach. However, if Dr Marchesi's concept finds more evidence in favor, selective targeting of intramembrane Aβ may be a better treatment. In light of the above cholesterol-in-cholinergic rescue concept, progressive deterioration of neuronal membranes may be caused by a worsening supply of cholesterol in these membranes. This might fit with progressive intramembrane Aβ accumulation as being a result of the deterioration of the state of neuronal membranes.

    Even if one agrees that external accumulation of Aβ is the prevalent process (in contrast to presumed intramembrane accumulation) with initial physiological relevance, the question about mechanisms of neurotoxicity of extracellular deposits remains unanswered. As a matter of fact, intraneuronal pathological inclusions have been proven to be more detrimental for neurons than extracellular deposits. Senile plaques observed during postmortem do not correlate with the degree of dementia as strongly as neurofibrillary tangles in retrospective studies. It is not uncommon to find nondemented elderly individuals with numerous neocortical senile plaques, whereas the co-occurrence of neocortical neurofibrillary tangles significantly increases the likelihood of dementia (Polvikoski et al., 2001). Even if accumulation of this peptide is a primary process taking place in AD, with formation of neurofibrillary tangles formation a secondary one, a causative link between those two processes has not been found. Even worse, as Dr. Marchesi states, is that in animal models of AD, accumulation of APP products caused by genetic interventions is not followed by the occurrence of the second neuropathological hallmark. One possible explanation is that in animal models the time span of amyloid overproduction is not sufficiently long to result in this process. An alternative explanation is that brains of humanized mice do not share other important pathological conditions of aging human brains. These missing factors are most probably ischemic processes and/or neuroinflammatory conditions frequently occurring in aging human brains.

    Regardless of these unresolved issues, the hypothesis presented by Dr Marchesi deserves experimental testing. It is also worth pointing out that it fits to a broad array of other concepts stressing the importance of neuronal membranes in pathogenesis of AD.

    Regarding the comment by Dr. Dewachter and Dr. Van Leuven, while I find their parallel of the pentagonal dodecaeder brilliant as an analogy, I also think it is discouraging for different researchers trying to decipher the riddle. Fortunately for us, we are not dealing with a mysterious object from ancient times, but a real protein in cells of our bodies and we can observe in action, not at excavations. In contrast to these authors, I am more optimistic about coming up with definite explanation for function/functions of APP and its metabolites. We just need to keep our eyes open. Of course, my favorite hypothesis is my own. I like it not only because it is mine, but because it matches many facts coming from observations and is also in line with other people's ideas.


    . Does cholesterol act as a protector of cholinergic projections in Alzheimer's disease?. Lipids Health Dis. 2005;4:13. PubMed.

    . Cholesterol homeostasis failure as a unifying cause of synaptic degeneration. J Neurol Sci. 2005 Mar 15;229-230:233-40. PubMed.

    . Alzheimer's disease amyloid-beta peptide modulates apolipoprotein E isoform specific receptor binding. J Alzheimers Dis. 2005 Aug;7(4):303-14. PubMed.

    . Prevalence of Alzheimer's disease in very elderly people: a prospective neuropathological study. Neurology. 2001 Jun 26;56(12):1690-6. PubMed.

  10. In speculation about the potential roles of Aβ in neural health and disease, it is surprising how rarely discussants invoke the topic of physiologically appropriate synaptic remodeling.

    The best working model that we have for the initiation of memory is long-term potentiation (LTP). However, this largely biochemical sequence of events does not explain memory to a level of complete satisfaction; the gap is made up, partially, by the conversion of the biochemical events of LTP into longer-term structural changes in the synapse. An equally important contribution to the development and consolidation of memory is likely to be made by long-term depression (LTD). And, if there is a structural component to synaptic potentiation, then synaptic or dendritic pruning is most certainly a structural analog in the biochemical depression of a synapse.

    A physiological role for Aβ in these normal, depressive aspects of synaptic plasticity has been posited by others (1). In this model, the development of Alzheimer disease reflects a runaway tipping of the synaptic balance towards those depressive events—an untoward emphasis on what is essentially a normal process, rather than development of an entirely new "abnormal" process.

    Insights might be gained by considering this model in light of Marchesi's hypothesis. Evidence accumulating over recent years emphasizes the role in synaptic plasticity to be played by regulated trafficking and membrane insertion of glutamate receptors (2). These events seem well-suited to modulation by the concentration of a peptide like Aβ in the plasmalemma, especially considering their dependence on lipid rafts (3). But just as LTP has a structural analog, it seems only natural to consider that the contribution Aβ makes to synaptic depression may include a structural one, i.e., loss of the synapse.

    Whether the structural aspect of synaptic depression (ablation, actually) is neuron-autonomous or is assisted by the degradative talents of microglia (4) is open for debate. But both scenarios are consistent with a role for Aβ in this type of "constructive disassembly," this consolidation of what was initially a biochemical attenuation of the synaptic strength. Moreover, it may be this disassembly—this structural "raising of the stakes"—that is contributed by extracellular (perhaps even trans-synaptic?) Aβ.


    . APP processing and synaptic function. Neuron. 2003 Mar 27;37(6):925-37. PubMed.

    . AMPA receptor trafficking and synaptic plasticity. Annu Rev Neurosci. 2002;25:103-26. PubMed.

    . Lipid rafts in the maintenance of synapses, dendritic spines, and surface AMPA receptor stability. J Neurosci. 2003 Apr 15;23(8):3262-71. PubMed.

    . Microglia in Alzheimer's disease and transgenic models. How close the fit?. Am J Pathol. 1999 Jun;154(6):1627-31. PubMed.

  11. Dr. Holtzman is correct to point out that my views on the possible fate of cleaved Aβ peptides were heavily influenced by the finding of Kawarabayashi and coworkers, who showed that significant amounts of the Aβ42 peptides were found associated with lipid rafts isolated from both Tg animals and the brains of AD patients. He also notes, very appropriately, that because lipid raft fragments of membranes can at present only be isolated in the presence of nonionic detergents, the possibility exists that homogenizing cells or brain tissue that contain large amounts of extracellular Aβ in the presence of detergent could conceivably cause Aβ dimers to be entrapped in the rafts during homogenization as opposed to their having been generated there in situ. While I think this is unlikely, I agree that this important caveat has to be ruled out.

    Dr. Barger reminds us that AD is much more than a complicated ensemble of proteolytic fragments that accumulate in damaged brain tissue. I like his suggestion, which I'm sure is shared by many, that a deeper analysis is needed of the role that Aβ peptides, or fragments of them, might play, physiologically, in the many complicated membrane events that accompany synaptic function. One can think of many ways in which glutamate receptor trafficking, or the functioning of other critical transmembrane molecules, could be influenced by collisions with intramembranous components of all sorts. It will be important to find out if a significant fraction of Aβ peptide does indeed reside within membranes of susceptible neurons, regardless of how they get there. The physiological events that Barger describes will surely be influenced by the properties of the membrane milieu, as he points out. These properties are likely to vary in different membranes and be heavily influenced by local intramembranous components, both lipids and polypeptides. Not only might such studies provide important insights into pathogenesis, but they also offer the prospect of new therapeutic approaches that focus on ways to influence events that occur within plasma membranes themselves. As one example, recent studies on the role of NSAIDs in AD may well be modifying the properties of the membrane itself rather than their known enzymatic targets.

  12. Research spanning the last two decades has increasingly implicated the Aβ peptide as a critical early player in the pathogenesis of AD. However, the mechanisms of Aβ-mediated toxicity and neurodegeneration in AD have been intensely debated and remain poorly understood. One of the most controversial issues concerning Aβ toxicity has regarded the form of Aβ that ultimately is responsible for the neuronal degeneration and death that are observed in AD brain. Compelling evidence suggests that the 42-amino-acid species, Aβ42, the overproduction of which strongly associates with familial AD, is the toxic agent in AD. However, beyond this, much mystery remains regarding the mechanism of Aβ42 toxicity. Paramount in this controversy over Aβ42 toxicity is the peptide's exact conformation, assembly state, and spatial localization in relation to the neuron. Multiple forms of Aβ, including fibrillar plaque-associated Aβ, small soluble Aβ oligomers, and intraneuronal Aβ accumulations, among others, have all been proposed to be neurotoxic in AD, and compelling evidence for each has been reported. Yet, in all cases, the molecular mechanism underlying the toxicity of each Aβ form has remained, at best, enigmatic or incomplete. Thus, a comprehensive, detailed mechanism of Aβ-mediated toxicity and neurodegeneration in AD is still lacking and sorely needed in the field.

    Vincent Marchesi in his recent PNAS paper has attempted to fill this mechanistic void by proposing the hypothesis that intramembranous accumulation of Aβ peptides may interfere with the normal functioning of potentially a plethora of membrane proteins, including the presenilin/γ-secretase complex and APP itself. He bases his hypothesis on years of work with Glycophorin A (GPA). This type I membrane protein contains GxxxG motifs in its transmembrane domain that self-associate in the lipid bilayer, causing the formation of SDS-stable dimers. Interestingly, APP also contains similar GxxxG motifs in its transmembrane/Aβ domain, thus providing an explanation for the peculiar observation that APP forms homodimers in cells (Scheuermann et al., 2001). Moreover, Marchesi hypothesizes that APP dimers may be cleaved by the secretases to form intramembranous Aβ dimers, which have been recently observed in neuronal membranes from Tg2576 mice and AD patients (Kawarabayashi et al., 2004). These Aβ dimers are likely to be quite stable in the membrane, and if they accumulate in AD, membrane-bound Aβ dimers could become toxic to long-lived non-dividing neurons. Perhaps most interestingly, Marchesi speculates that if Aβ dimers are able to dissociate into monomers, even at low levels, the membrane-bound Aβ monomers may have the opportunity to heterodimerize with other membrane proteins that contain GxxxG transmembrane motifs, the number of which is apparently quite large. Furthermore, membrane protein function might be disrupted in the heterodimer, and the resulting deficiency might contribute to the AD phenotype. As an example, Marchesi proposes that the presenilin/γ-secretase complex, in which Aph-1 is essential for complex assembly, might be disrupted by intramembranous Aβ. Aph-1 contains transmembrane GxxxG motifs, and it is possible that Aβ-Aph-1 heterodimers could form and prevent Aph-1 from performing its role as a stabilizing element in the γ-secretase complex. The consequences of this might be γ-secretase destabilization and loss of function. In this light, it is interesting to note that neuron-specific conditional presenilin knockout mice show dramatic age-dependent neurodegeneration (Saura et al., 2004), thus lending some support to the idea that loss of presenilin/γ-secretase complex function may be contributing to AD pathophysiology.

    Marchesi puts forth his ideas in a well-written, cogent argument that is highly interesting and thought-provoking to read. As with any new hypothesis, much work remains to test its predictions thoroughly. Although intriguing and potentially important, the Marchesi hypothesis may be very difficult to prove. For example, mutating the Gs in the GxxxG motif would be predicted to prevent intramembranous Aβ dimerization and rescue neurodegeneration, but even in case of a positive result, one could still argue that some other toxic property of Aβ had been ablated by the mutation to achieve rescue. As we have seen from the arguments that have raged around the amyloid hypothesis, definitive evidence in favor of any hypothesis of AD has been notoriously elusive. Another prediction of the Marchesi hypothesis would be that increased intramembranous Aβ should reduce presenilin/γ-secretase function through heterodimerization with Aph-1. This is easily tested. However, a positive result would lead to an interesting paradox: If accumulation of intramembranous Aβ inhibits γ-secretase function, then Aβ production should drop and thus alleviate the build-up of membrane-bound Aβ to rescue neurodegeneration. Of course, accumulation of C99 might still progress, causing toxicity, or a threshold of intramembranous Aβ may be reached beyond which no neuronal recovery would be possible.

    An additional paradox concerns p3, the peptide generated by α- and γ-secretase, which the Marchesi hypothesis predicts would form intramembranous dimers at least as readily as Aβ, and perhaps even more so due to p3's smaller proportion of polar residues. Levels of p3 production are thought to be greater than those of Aβ, yet unlike Aβ, the majority of evidence to date suggests that p3 is not associated with neurotoxicity. It is clear that thoughtful, well-designed experiments will be essential for determining the validity of this hypothesis. In spite of the potential difficulties outlined above, this is an interesting alternative hypothesis that deserves attention and may help illuminate some of the great unanswered questions of the field regarding the toxic form and mechanism of Aβ-mediated neurodegeneration in AD. Finally, it is also important to consider that the process of neurodegeneration in AD is extremely complex and likely to involve a multitude of mechanisms. In the end, we may find that amyloid plaques, Aβ oligomers, intraneuronal Aβ, Aβ-mediated neuroinflammation, as well as intramembranous Aβ, may all contribute a degree of neurotoxicity during AD, and that no single form of Aβ may be fully responsible for neurodegeneration in AD.


    . Homodimerization of amyloid precursor protein and its implication in the amyloidogenic pathway of Alzheimer's disease. J Biol Chem. 2001 Sep 7;276(36):33923-9. PubMed.

    . Dimeric amyloid beta protein rapidly accumulates in lipid rafts followed by apolipoprotein E and phosphorylated tau accumulation in the Tg2576 mouse model of Alzheimer's disease. J Neurosci. 2004 Apr 14;24(15):3801-9. PubMed.

    . Loss of presenilin function causes impairments of memory and synaptic plasticity followed by age-dependent neurodegeneration. Neuron. 2004 Apr 8;42(1):23-36. PubMed.

  13. I am "Johnny Come Lately" to this lively discussion. Most of the major points have already been made. Our lab has favored a major role for intracellular events as a basis for β amyloid toxicity since the early 1990s (Fukuchi et al., 1992; 1993). I hasten to add, however, that functional and structural damage from within and from without are not mutually exclusive.

    Having known Vin Marchesi for many years, I have learned to pay very careful attention to his thoughtful papers. As a fellow pathologist, I am wondering how seriously he takes the more general proposition that additional forms of amyloidosis might be classified as "channelopathies" (BL Kagan et al., 2002; ARF Live Discussion ).


    . Overexpression of amyloid precursor protein alters its normal processing and is associated with neurotoxicity. Biochem Biophys Res Commun. 1992 Jan 15;182(1):165-73. PubMed.

    . Expression of a carboxy-terminal region of the beta-amyloid precursor protein in a heterogeneous culture of neuroblastoma cells: evidence for altered processing and selective neurotoxicity. Brain Res Mol Brain Res. 1992 Nov;16(1-2):37-46. PubMed.

    . Neurotoxicity of beta-amyloid. Nature. 1993 Jan 14;361(6408):122-3. PubMed.

    . The channel hypothesis of Alzheimer's disease: current status. Peptides. 2002 Jul;23(7):1311-5. PubMed.

  14. I've read this informative forum with interest. I believe our work may have some relevance to the question: Is secreted Aβ a pathogenic agent in Alzheimer disease?

    We have reported that transient viral overexpression of APP depresses excitatory synaptic transmission (1), and now we have unpublished evidence that this also causes dendritic spine loss. This effect is seen by overexpression of C99, but not by APP-MV, a point mutant APP that is not efficiently cleaved by β-secretase but is cleaved by a-secretase (2). Along with other data (1), our results indicate that Aβ is responsible for the synaptic depression. I would note that since APP-MV does not produce synaptic depression, it is unlikely that P3 (the product of α- and γ-secretase) produces synaptic depression.

    Our evidence supporting the view that secreted Aβ can produce synaptic depression comes from an experiment in which many postsynaptic neurons in a small region of a hippocampal slice are driven to overexpress APP. APP is coexpressed with GFP via an IRES construct so that cells overexpressing or not overexpressing APP can be identified; cells are either bright green or not green, so GFP detection should not be an issue. We record from a single non-overexpressing neuron that is surrounded by APP-overexpressing neurons. We find that transmission onto such a neuron is also depressed. (Note: The presynaptic cells originate in a different part of the slice and are not overexpressing APP.) The simplest interpretation of this result is that the cells overexpressing APP secrete Aβ, which depresses synapses on the nearby non-overexpressing cells. Alternatively, an overexpressing cell may make Aβ that is not secreted, but it induces release of something else that causes synaptic depression on the nearby non-overexpressing cells. In any case, it appears that APP-overexpressing cells can produce synaptic depression in nearby neurons.

    One potentially relevant observation is that while expression of C99 produces synaptic depression, we cannot detect soluble Aβ in the media from such slices. In contrast, we readily detect Aβ in media from slices overexpressing APP. I'd be interested in knowing if other investigators have had this experience when expressing C99 in any cell type. This observation is consistent with the view that Aβ does not leave the membrane once produced, and yet can lead to synaptic depression. I want to add, however, that the experiment described above with a non-overexpressing cell in a group of APP-overexpressing cells has not been done with C99.

    It is, of course, difficult to extrapolate from overexpression studies in hippocampal slices to Alzheimer disease. There are likely to be many systems-level processes at work. However, our studies on synaptic physiology may elucidate the impact of APP-related products on synaptic transmission and plasticity.


    . APP processing and synaptic function. Neuron. 2003 Mar 27;37(6):925-37. PubMed.

    . Mutation of the beta-amyloid precursor protein in familial Alzheimer's disease increases beta-protein production. Nature. 1992 Dec 17;360(6405):672-4. PubMed.

  15. Dr. Marchesi proposes that the accumulation of Aβ dimers in lipid rafts may disturb the normal regulation of APP cleavage, resulting in the overproduction of intramembranous Aβ dimers by creating a positive feedback mechanism for Aβ production. He proposes various effects of the abnormally high levels of intramembranous Aβ dimers including deleterious interactions with intramembranous segments of receptors, channels, and enzymes, like the presenilins and other secretases. He believes that these processes ultimately lead to neuronal death, which would in turn disrupt cognitive function. He also raises the important unanswered question of the physiological function of Aβ peptides in normal neurons, and suggests that certain lipid-associated Aβ peptides may facilitate normal function of ion channels or pumps or critical receptors. Presumably, the normal Aβ peptides are not the same as the pathological Aβ dimers, and the production of Aβ dimers may offset the levels or function of normal Aβ interactions within the membrane.

    A provocative parallel is drawn between the GxxxG segment in glycophorin dimers, originally discovered by Dr. Marchesi, and an identical segment in APP and Aβ. The GxxxG segment protects dimers from denaturation by sodium dodecyl sulfate (SDS), a well-documented and distinctive property of Aβ dimers and higher-order soluble Aβ oligomers.

    In the following paragraphs, I argue that the intramembranous Aβ dimers do not contribute substantially to cause memory and cognitive dysfunction or neuronal loss in mice. Because these data were obtained in mice, their relevance to the human disease must be interpreted with caution.

    Finding the causative agents of memory and cognitive dysfunction in Alzheimer disease has been the overriding focus of research in my laboratory. To that end, we have taken a molecular approach to studying the functional consequences of the tau and Aβ proteins in the brain. Our approach differs from the approach taken by many other scientific laboratories, which investigate the structural consequences of these proteins; amyloid plaques, neurofibrillary tangles, and neuronal death are examples of the structural changes in the brain that are commonly studied.

    Since abnormal brain and cognitive function is the focus of research in my laboratory, all our work has begun with the generation of transgenic mice. One line, Tg2576, develops amyloid plaques and memory loss, and is used to explore the properties of Aβ (Hsiao et al., 1996). Another, rTg4510, develops neurofibrillary tangles, neuronal loss, and memory impairment, and is useful for examining the tau protein ((Ramsden et al., 2005, in press). Tg2576 and rTg4510 mimic different stages of Alzheimer disease: early, before neuronal loss occurs (Tg2576); and late, when loss of neurons and brain atrophy has developed (rTg4510). Neither mouse is a complete model, but each one is useful for exploring the progression of cognitive decline and the pathology of Alzheimer disease. Investigations of these mice have helped recast the plaques and tangles from being causes to being indications of Alzheimer disease (Hsiao et al., 1995; Chapman et al., 1999; Ashe, 2001; Westerman et al., 2002; Kotilinek et al., 2002; SantaCruz et al., 2005).

    An early, unexpected observation was that memory loss in Tg2576 occurs independently of neuronal or synaptic loss (Irizarry et al., 1997). The lack of significant neuronal or synaptic loss in Tg2576 mice, despite reproducible evidence of memory decline (Hsiao et al., 1996; Westerman et al., 2002; Kotilinek et al., 2002), indicates that memory loss in plaque-forming Tg2576 mice occurs independently of neurodegeneration. Furthermore, the accumulation of Aβ dimers in Tg2576 mice (Kawarabayashi et al., 2004) is not sufficient to induce significant death of neurons.

    We and others showed that plaques are not the cause of memory loss in Tg2576 and PDAPP mice. Because Aβ antibodies rapidly reverse preexisting memory dysfunction in Tg2576 and PDAPP mice, with no changes in plaque load or levels of SDS-soluble or SDS-insoluble Aβ (Kotilinek et al., 2002; Dodart et al., 2002), we concluded that some neurons in these mice are dormant but not dead, and that the dormant neurons are reactivated by the removal of or neutralization of specific memory-impairing Aβ species by Aβ antibodies. We therefore began the search for elusive, heretofore undiscovered forms of the Aβ protein, called Aβ star (Aβ*) molecules, that disrupt memory function independently of amyloidosis or neuronal loss.

    We employed a three-step procedure for determining Aβ*: (a) identification; (b) isolation; and (c) application. Candidate Aβ* proteins had to satisfy two criteria: (a) they appeared at 6 months of age, not before, coinciding with the first appearance of memory loss in Tg2576 mice; and (b) their levels remained stable for the next 6 to 8 months (from 6 to 14 months of age), when the memory impairment in Tg2576 remained unchanged. We excluded intramembranous Aβ dimers as a candidate for Aβ* because its levels increased from 6 to 10 months of age, and even more substantially thereafter, when there was a stable memory deficit in Tg2576 mice (Fig. 2, Kawarabayashi et al., 2004).

    Recently, we found a candidate Aβ* protein which satisfies both criteria, and whose levels correlate very highly with memory deficits (r2 >0.5) (Lesné et al., submitted). Purified Aβ* disrupts cognitive function when applied to young, healthy rats in a dose-related manner. Thus, Aβ* appears to be necessary and sufficient for Aβ to impair cognitive function in mice and rats. Aβ* impairs memory by a mechanism that does not involve amyloidosis or neuronal loss. In contrast, while intramembranous Aβ dimers may be involved in aspects of Alzheimer disease pathogenesis, they do not appear to play a direct role in disrupting memory or cognitive function in mice. Moreover, they do not induce neuronal death. Yet, their presence in an important intramembranous structure, lipid rafts, and their association with critical molecules in Alzheimer disease pathogenesis make one reluctant to disregard their importance altogether. I agree with the other commentators that Dr. Marchesi's hypothesis deserves additional consideration.


    . Correlative memory deficits, Abeta elevation, and amyloid plaques in transgenic mice. Science. 1996 Oct 4;274(5284):99-102. PubMed.

    . Age-related CNS disorder and early death in transgenic FVB/N mice overexpressing Alzheimer amyloid precursor proteins. Neuron. 1995 Nov;15(5):1203-18. PubMed.

    . Impaired synaptic plasticity and learning in aged amyloid precursor protein transgenic mice. Nat Neurosci. 1999 Mar;2(3):271-6. PubMed.

    . Learning and memory in transgenic mice modeling Alzheimer's disease. Learn Mem. 2001 Nov-Dec;8(6):301-8. PubMed.

    . The relationship between Abeta and memory in the Tg2576 mouse model of Alzheimer's disease. J Neurosci. 2002 Mar 1;22(5):1858-67. PubMed.

    . Reversible memory loss in a mouse transgenic model of Alzheimer's disease. J Neurosci. 2002 Aug 1;22(15):6331-5. PubMed.

    . Tau suppression in a neurodegenerative mouse model improves memory function. Science. 2005 Jul 15;309(5733):476-81. PubMed.

    . APPSw transgenic mice develop age-related A beta deposits and neuropil abnormalities, but no neuronal loss in CA1. J Neuropathol Exp Neurol. 1997 Sep;56(9):965-73. PubMed.

    . Dimeric amyloid beta protein rapidly accumulates in lipid rafts followed by apolipoprotein E and phosphorylated tau accumulation in the Tg2576 mouse model of Alzheimer's disease. J Neurosci. 2004 Apr 14;24(15):3801-9. PubMed.

    . Immunization reverses memory deficits without reducing brain Abeta burden in Alzheimer's disease model. Nat Neurosci. 2002 May;5(5):452-7. PubMed.

  16. Reply by Vincent Marchesi to comment by Karen Ashe

    Based on her recent studies, Dr. Ashe states that "the intramembranous Aβ dimers do not contribute substantially to cause memory and cognitive dysfunction or neuronal loss in mice." I have the following reply:

    Dr. Ashe suggests, and most observers would agree, that the transgenic mice she and her coworkers have generated are useful animal models to study various changes in mouse brains that mimic to some extent what appear to be comparable changes in the brains of human patients with AD. She goes on to suggest that memory loss of affected Tg 2576 animals occurs without "significant" neuronal or synaptic loss and cites various publications which support this view. Here I have serious reservations for a number of reasons. First, and most obvious, we simply don't know anything about the pathophysiology of the early events that lead to memory loss, in either mice or people. This being the case, there is no way to rule out the possibility that small amounts of Aβ, or some as yet undetected substance, might affect critical neurons in ways that compromise their functions, long before large-scale neuronal destruction or histologically visible loss of synapses is evident.

    Dr. Ashe goes on to suggest that anti-Aβ antibodies can reverse memory loss in some animals and proposes that removal of a previously undetected form of Aβ (called Aβ*) can reactivate neurons that are dormant but not dead. This Aβ* is described as a specific memory-impairing Aβ species. This Aβ* material has not been isolated, but is assumed to be present in the brains of Tg 2576 mice who are between 6 and 14 months of age, at a time when both the level of the peptide material and the degree of memory loss are described as being unchanged. The thrust of this argument is that since neither the Aβ* material nor memory loss appears to be changing during this period, they might well be related. Dr. Ashe rules out the possibility that intramembranous Aβ dimers might be involved, since their levels increase during the period from 6 to 9 months of age, when the memory deficit appeared to be stable. I don't find this argument persuasive for the same reasons I stated above. Without knowing more about the pathophysiology of memory impairment in mice, events that coincide temporally may have nothing to do with primary causality.

  17. The hypothesis by Vincent Marchesi is attractive and timely. Given that considerable effort in investigating the pathogenesis of AD is spent on the effects of plaque-associated extracellular Aβ, the article makes one rethink the direction of current investigations. We add two points to the ongoing debate.

    1. GxxxG motif of Aβ and channel formation
    In assessing Marchesi's hypothesis that Aβ forms heterodimers with APP and thereby interferes with the yet-to-be-identified physiological function of APP, the field might consider a recent paper by James Bowie's group (Kim et al., 2005). In it, the authors present evidence for yet another alternative hypothesis. They have identified a motif present in many membrane proteins, called "transmembrane glycine zippers." Its basic unit is the GxxxG motif, similar to what Marchesi proposed. The Bowie group shows that Aβ and PrP peptides contain the glycine zipper motif and that the glycines in the GxxxG motif of Aβ are critical for channel formation and for neuronal cell death in vitro. The GxxxG motif thus appears to mediate self-association of Aβ peptides. This does not address specifically the heterodimerization that Marchesi proposes, but it does support his ideas insofar as it also implies that Aβ might exert an effect by staying inside the membrane by means of oligomerization. Whether this effect is the work of a pool of Aβ that is generated in the bilayer and never leaves it, or of Aβ that reinserts into the membrane to form channels after having been first secreted, is an open question.

    2. Rafts and oligomerization
    Evidence that Aβ oligomers are raft-associated is based on detergent extraction (Kawarabayashi et al., 2004). Lipid rafts are dynamic, liquid-ordered assemblies in the membrane that require cholesterol and sphingomyelin for their stability (Rajendran and Simons, 2005). Several transmembrane proteins, lipid-modified proteins (such as GPI-anchored proteins that are tethered to the exoplasmic leaflet and fatty acid-modified proteins that are anchored to the cytoplasmic leaflet), as well as peptides, can partition into raft domains. However, in the resting state the raft assemblies are continuously associating and dissociating, thus evading visualization or detection by current technology. Only after raft clustering has been induced [oligomerization is one driving force for raft clustering (Paladino et al., 2004; Schuck and Simons, 2004)] do rafts become accessible to microscopy or other detection methods. So clearly, the use of detergents normally leads to artificial raft coalescence and does not provide information on the in-vivo scenario.

    In the case of APP and β-secretase, they seem to come together in raft clusters formed after internalization from the cell surface by endocytosis (Ehehalt et al., 2003). Also, γ-secretase can be raft-associated (Vetrivel et al., 2004) and probably encounters its substrate in the endosomal raft clusters to generate Aβ where the peptides could undergo oligomerization. An additional contribution from the biosynthetic pathway for Aβ production cannot be excluded. If amyloidogenic cleavage of APP occurs in endosomal raft clusters, only the fraction of APP that partitions into rafts will be cleaved. By heterodimerizing with the raft-associated Aβ through the GxxxG motif, more APP could be dragged into rafts by raft-associated Aβ and thus more Aβ could be produced by β-secretase cleavage of Aβ-bound APP. Recent studies in proteoliposomes have demonstrated that β-secretase activity is enhanced by raft lipids. This supports Marchesi's proposal that an Aβ-APP heterodimer could lead to increased β-cleavage. It is our contention that an important experimental venue will be to reconstruct these processes in model membranes and to establish cell-free assays that recapitulate what happens in cells. These experiments will not be able to elucidate how AD comes about, but will lay solid foundations for further work.


    . Transmembrane glycine zippers: physiological and pathological roles in membrane proteins. Proc Natl Acad Sci U S A. 2005 Oct 4;102(40):14278-83. PubMed.

    . Dimeric amyloid beta protein rapidly accumulates in lipid rafts followed by apolipoprotein E and phosphorylated tau accumulation in the Tg2576 mouse model of Alzheimer's disease. J Neurosci. 2004 Apr 14;24(15):3801-9. PubMed.

    . Lipid rafts and membrane dynamics. J Cell Sci. 2005 Mar 15;118(Pt 6):1099-102. PubMed.

    . Protein oligomerization modulates raft partitioning and apical sorting of GPI-anchored proteins. J Cell Biol. 2004 Nov 22;167(4):699-709. PubMed.

    . Polarized sorting in epithelial cells: raft clustering and the biogenesis of the apical membrane. J Cell Sci. 2004 Dec 1;117(Pt 25):5955-64. PubMed.

    . Amyloidogenic processing of the Alzheimer beta-amyloid precursor protein depends on lipid rafts. J Cell Biol. 2003 Jan 6;160(1):113-23. PubMed.

    . Association of gamma-secretase with lipid rafts in post-Golgi and endosome membranes. J Biol Chem. 2004 Oct 22;279(43):44945-54. PubMed.

    . Lipids as modulators of proteolytic activity of BACE: involvement of cholesterol, glycosphingolipids, and anionic phospholipids in vitro. J Biol Chem. 2005 Nov 4;280(44):36815-23. PubMed.

  18. Vincent Marchesi: Summary and Impressions

    It is a measure of the maturity of the Alzheimer disease field that many of its most prominent investigators seem willing, at least provisionally, to consider an alternative interpretation to the Aβ hypothesis that rests on so little experimental support. I take as one of the compelling reasons the fact that so many of the respondents share my view that the precise means by which Aβ peptides damage neurons and induce synaptic dysfunction are still very much unsettled issues. Even more unclear, in my opinion, is how the physicochemical changes that we measure can affect human memory, or what we attribute to memory in experimental animals.

    By drawing attention to the possibility that Aβ peptides might concentrate within the membrane lipid bilayer itself, with the capacity to affect vital membrane functions, we have a whole new biochemistry to deal with and an opportunity to explore the biological consequences of hydrophobic interactions within a nonaqueous milieu. A large fraction of our functioning genome codes for polypeptide chains that reside within lipid bilayers of a diverse set of membranous elements, yet, apart from the analysis of a few channel proteins, we know little of the forces that determine how intramembranous segments of proteins affect each other. Recently, it has become clear that proteolytic cleavage of transmembrane proteins is far more prevalent than previously realized, and a new field, referred to as regulated intramembranous proteolysis (RIP), is providing new insights into how transmembrane signaling is mediated. And, like the cleavage of APP to generate Aβ peptides, it is likely that many other intramembranous cleavages may also be linked to pathogenic mechanisms.

    An attractive feature of the prevailing "Aβ/amyloid" hypothesis is the possibility that small aggregates of Aβ peptides, termed Aβ oligomers, are toxic to neurons and other cells through their ability to create channel-like perturbations in plasma membranes. Adding Aβ oligomers to cultured cells does result in cell death, in the hands of some investigators, and it is conceivable that such peptides might insert into cell membranes if their concentrations were high enough and they were not otherwise bound to other proteins. But it is unclear exactly how extracellular peptides that are folded into β-like strands can insert back into membranes and regain the stability of α helices. A much simpler question to ask is whether Aβ dimers that are already in the membrane affect natural channel proteins.

    Looking at possible functions of Aβ peptides inside membranes may also help us sort out the confusing and often contradictory claims regarding the role of cholesterol and other lipids in AD pathogenesis. Similarly, the recently described effects of NSAIDs on Aβ production might better be explored by developing assays that can measure their effects on the ability of intramembranous Aβ peptides or partially degraded fragments of them to interact with other membrane elements. Many published model systems now exist that can measure interactions between transmembrane domains of membrane proteins quantitatively.

    The suggestion that dimeric forms of the Aβ peptides might exist within membranes was based in part on a published study that reported the presence of Aβ dimers in membrane fragments isolated as putative lipid rafts, which are subdomains of membranes enriched in cholesterol and sphingolipids. But the significance of this observation has been questioned by several respondents, and by a recent publication which claims that Aβ peptides might bind non-specifically to such membrane fragments as a consequence of detergent lysis. It has also been pointed out that preparations of rafts are likely to be artifactually "coalesced" structures that may not represent the distribution of membrane-associated proteins under in vivo conditions. It is conceded that more Aβ could be generated if Aβ-APP heterodimers accumulate in raft structures that are rich in β-secretase, but whether this association between Aβ and APP is mediated by the GxxxG motif, or some other mechanism, remains to be determined. At the present time, we have no way to decide whether rafts, as they are now isolated, are real or artifactual aggregates, but they do have operational value and are a useful way to study APP processing.

    My congratulations to the Alzforum for sponsoring such a productive exchange of views.

  19. We have done experimental work on the GxxxG motif of APP since we realized, almost 3 years ago, that it is part of the Aβ domain. To contribute to the theoretical discussion on a possible impact of the GxxxG motif on APP processing and Aβ toxicity, we would like to summarize shortly the data we have produced so far and that were presented at the last Society for Neuroscience conference in Washington.

    This motif has a dual role for APP and Aβ. When we analyzed Aβ generation, we found strong evidence that the γ-secretase cleavage site depends on the oligomeric form of its substrate, which is strongly influenced by the GxxxG motif. Our results show that APP dimerization mediated by the transmembrane sequence decides on Aβ42 generation. In Aβ, the same motif determines if mature fibrils are formed. The enclosed abstract provides information as given at the SfN conference on 14 November in Washington.

    Dimeric assembly of the APP membrane-spanning domain defines a selective γ-secretase cleavage site

    LM Munter (FU Berlin), P Voigt (UKBF), E Lindner (TU München), M Schaefer (UKBF), D Langosch (TU München), G Multhaup (FU Berlin)

    Background: The amyloid precursor protein (APP) is sequentially processed by the α- or β-secretase, followed by γ-secretase cleavage within the transmembrane domain. The resulting fragments are the Aβ peptides ending at residues 38, 40, 42, 43, or 46, with Aβ42 being the major risk factor for Alzheimer disease (AD) and the APP intracellular domain (AICD) which is generated by a cleavage at Aβ49. In contrast to other known γ-secretase substrates, the intramembranous cleavage of APP-derived substrates C83 and C99 is less precise.

    Objective: To identify the motif(s) for PS-dependent recognition/cleavage of APP that determine(s) the intramembranous cleavage site.

    Results: A specific motif mediating transmembrane helix-helix interactions within the transmembrane sequence (TMS) of APP was identified to be the driving force for the homodimerization of APP. This site represents a novel third contact site of APP in addition to the extracellular loop region 91-111 (Scheuermann et al., 2001) and the collagen-binding site 448-465 (Beher et al., 1996). Dimerization of the C-terminal fragment of APP, C99, occurs independent from the extracellular contact sites and suggests that not only presenilin fragments form functional dimers, but also that the γ-secretase substrate C99 is dimeric. We further examined the level of Aβ species depending on the dimerization of C99. We find excellent correspondence between dimerization and Aβ42 production, which is selectively lowered by single point mutations within the dimerization motif.

    Conclusion: We propose that most of the γ-secretase cleavage sites are independent of the dimer stability of the substrate. But most importantly, Aβ42 formation strongly correlates with the dimeric state of the substrate. Since Aβ42 production has been linked to the development of the disease, compounds interfering with the dimerization of γ-secretase substrate might decrease Aβ production in humans.


    . Regulation of amyloid protein precursor (APP) binding to collagen and mapping of the binding sites on APP and collagen type I. J Biol Chem. 1996 Jan 19;271(3):1613-20. PubMed.

    . Homodimerization of amyloid precursor protein and its implication in the amyloidogenic pathway of Alzheimer's disease. J Biol Chem. 2001 Sep 7;276(36):33923-9. PubMed.

  20. It is refreshing to read about alternatives to the amyloid hypothesis. Vincent Marchesi's is provocative, as any new hypothesis should be. The Alzforum discussion was fantastic. As a computational physicist, I like to take a minimalist approach when dealing with the unknown (by unknown I mean cleavage of APP, Aβ formation, and early aggregation events), and can add a different perspective.

    Even though I found each of the comments very helpful, Dennis Selkoe answered most of the questions that I had while reading the Marchesi paper. One of his comments was that we need to distinguish between the normal and pathogenic processes involved in Aβ secretion. Thus, I would say we need first a hypothesis on the normal processing and function of Aβ. Are the cleaved APP dimer and the resulting Aβ dimer a part of a normal or pathogenic process? Most importantly, we need to find ways to test this hypothesis. This cannot be done in transgenic mouse models, which complicates the problem.

    When bringing up pro- and counter examples for an AD hypothesis (be it the amyloid hypothesis involving toxic extracellular Aβ oligomers or the Marchesi hypothesis of toxic intramembranous Aβ aggregates), I find it confusing that researchers mix the experimental facts obtained on different transgenic mouse models with the ones obtained on human tissue. Here, again, Dennis Selkoe comes to the rescue by pointing out that a specific environment (be it extracellular, intramembranous, or one of many cell compartments) may be very different in different mouse models, not to mention that the human situation should differ significantly. Aβ is very sensitive to small changes in environment. Our group has shown that folding of a decapeptide fragment of Aβ is very sensitive to small changes in the solvent (see Cruz et al., 2005). For this reason, even just a different proportion of ions in the solvent, or a different concentration of Aβ or other proteins that interact with it, may be enough to completely change the Aβ pathway(s). Thus, my suggestion would be to test each of the hypotheses carefully on one model at a time. Karen Hsiao Ashe stepped in the correct direction by forming her comments using the results of her research on one transgenic mouse model. I find her response very much on target.

    As an expert on Aβ modeling, I can contribute to the discussion of what kind of interactions would lead to formation of Aβ aggregates in an extracellular versus intramembranous environment.

    In the extracellular aqueous environment, the hydrophobic/hydrophilic nature of residues plays an essential role in Aβ folding and aggregation. Thus, initially, Aβ oligomers should have globular structure with hydrophobic C-termini on the inside (away from water) and hydrophilic N-termini on the outside (exposed to water). Such aggregates would not be porous but rather packed in the core. We recently published our simulations on Aβ40 and Aβ42 oligomer formation under these "extracellular" conditions (Urbanc et al., 2004). They provide insight into the 3D structure of oligomers and a "molecular" explanation of the observed difference in oligomerization pathways for Aβ40 and 42 (Bitan et al., 2003).

    What happens in the intramembranous environment? Because there is no water, the hydrophobic and hydrophilic nature of residues does not matter. Instead we are left with 1), hydrogen bond interaction, which is the same for all residues, and 2), Coulombic interaction between pairs of charged residues. What would the Aβ conformers look like in such an environment? Aβ monomers would prefer an alpha helix structure as observed in membrane-like environments and as it is also typical for proteins inserted into a membrane (APP). Monomers with alpha helical structure are rather stable because they have few hydrogen bond donors or acceptors (alpha helix takes care of both donors and acceptors). However, there are the first 10-15 residues at the N-terminus, few at the C-terminus, and the bend centered between G25 and G29 (the so-called GxxxG motif). Possibly the most important feature is that the N-terminus is packed with charged residues. In addition, E22, D23, and L28 within or close to the GxxxG motif are charged as well. The Coulombic interaction should be important within the membrane because of the lack of polar molecules (water) that would normally shield the charges and thereby reduce the strength of Coulombic interactions. As for the difference between Aβ40 and 42, in the waterless environment the difference of two hydrophobic residues, Ile and Ala, should be less relevant than it is in the extracellular, water-rich environment. Thus, if Marchesi's hypothesis is true, it would be difficult to explain the stronger association of Aβ42 (compared to Aβ40) with AD by only considering molecular interactions between Aβ peptides.

    There is also an issue with Aβ42 remaining fully in the membrane. Aβ42 (as well as Aβ40) at the normal pH is highly negatively charged. It is well known that a membrane is highly impermeable to charged molecules. This fact also explains why Mobley et al. saw Aβ42 and 40 inserted into the upper part of a bilayer, with the N-terminus (highly charged) never entering the membrane environment (Mobley et al., 2004). Because most of the charge is concentrated at the N-terminus, it is more likely for the p3 peptide to remain within the bilayer. It would be interesting to simulate how Aβ42 assembles in the membrane.


    . Solvent and mutation effects on the nucleation of amyloid beta-protein folding. Proc Natl Acad Sci U S A. 2005 Dec 20;102(51):18258-63. PubMed.

    . In silico study of amyloid beta-protein folding and oligomerization. Proc Natl Acad Sci U S A. 2004 Dec 14;101(50):17345-50. PubMed.

    . Amyloid beta -protein (Abeta) assembly: Abeta 40 and Abeta 42 oligomerize through distinct pathways. Proc Natl Acad Sci U S A. 2003 Jan 7;100(1):330-5. PubMed.

    . Modeling amyloid beta-peptide insertion into lipid bilayers. Biophys J. 2004 Jun;86(6):3585-97. PubMed.

Make a Comment

To make a comment you must login or register.


News Citations

  1. Gabrielle Strobel Interviews Vincent Marchesi

Paper Citations

  1. . An alternative interpretation of the amyloid Abeta hypothesis with regard to the pathogenesis of Alzheimer's disease. Proc Natl Acad Sci U S A. 2005 Jun 28;102(26):9093-8. PubMed.
  2. . Triple-transgenic model of Alzheimer's disease with plaques and tangles: intracellular Abeta and synaptic dysfunction. Neuron. 2003 Jul 31;39(3):409-21. PubMed.
  3. . Homodimerization of amyloid precursor protein and its implication in the amyloidogenic pathway of Alzheimer's disease. J Biol Chem. 2001 Sep 7;276(36):33923-9. PubMed.
  4. . Dimeric amyloid beta protein rapidly accumulates in lipid rafts followed by apolipoprotein E and phosphorylated tau accumulation in the Tg2576 mouse model of Alzheimer's disease. J Neurosci. 2004 Apr 14;24(15):3801-9. PubMed.
  5. . Selection and characterization of small random transmembrane proteins that bind and activate the platelet-derived growth factor beta receptor. J Mol Biol. 2004 May 14;338(5):907-20. PubMed.
  6. . Transmembrane peptides as inhibitors of ErbB receptor signaling. Mol Biol Cell. 2004 Jul;15(7):3464-74. PubMed.

Other Citations

  1. ARF Live Discussion

Further Reading

No Available Further Reading