These are my notes from week 2 of Harvard’s Neurobiology 305qc course “Biochemistry and Biology of Neurodegenerative Diseases”, held by Michael Wolfe and Dominic Walsh on November 10, 2014.

This week’s class is on the role of amyloid beta in Alzheimer’s disease. The readings are one review [Benilova 2012] and two original research papers [Shankar 2008, Melnikova 2013].


Early on, it was recognized that Alzheimer’s brains contain plaques composed of “amyloid.” The constituent protein was eventually identified as amyloid beta (Aβ) [Glenner & Wong 1984, Masters 1985]. However, “Aβ” is not a single thing - there are over 30 different peptides called Aβ, differing by cleavage site (and therefore length), familial mutations and post-translational modifications, though all are cleaved from amyloid precursor protein, encoded by the gene APP. It has traditionally been held that Aβ peptides differ in C terminal cleavage site, but recent evidence suggests there are also different N terminal endpoints [Welzel 2014].

All available evidence from human genetics gives amyloid beta a central role in Alzheimer’s disease. Mendelian early onset Alzheimer’s is associated with mutations in APP that either alter Aβ’s amino acid sequence or increase its production, or mutations in PSEN1 and PSEN2 that increase its production. In fact, the presenilin mutations increase the Aβ42:Aβ40 ratio. (APP mutations account for ~10% of early-onset AD, while PSEN1 and PSEN2 account for 90%). Duplication of APP causes early-onset Alzheimer’s [Rovelet-Lecrux 2006], as does trisomy 21 (Down’s syndrome). By far the strongest genetic risk factor for sporadic late onset Alzheimer’s is the APOE ε4 allele, which reduces clearance of Aβ. A rare variant that protects against Alzheimer’s, APP A673T, reduces beta secretase’s production of Aβ.

Alzheimer’s disease is also associated with neurofibrillary tangles of Tau protein. There exists an excellent spatial and temporal correlation between Tau tangles and cognitive impairment. However, according to [Benilova 2012], however, neuropathologists do not find a clear correlation between Aβ plaque deposition and cognitive impairment. This has led to a hypothesis that soluble oligomers, rather than insoluble plaques, mediate toxicity. This is supported by a number of observations. If you use formic acid to extract Aβ peptides from brain tissue, the level correlates with disease severity [Naslund 2000]. Aβ oligomers inhibit long-term potentiation (LTP) in rat hippocampal neurons [Walsh 2002]. See further discussion below under “Why oligomers?”

Nonetheless, the study of Aβ toxicity is incredibly fraught. Aβ in the human brain is a heterogeneous mix of different peptides and different conformations, and Aβ and is highly prone to oligomerize or aggregate. It is therefore hard to study this peptide without changing it: the exact conditions in which natural Aβ is extracted, or in which synthetic Aβ is prepared, are likely to change its conformation and multimericity in ways that are hard to measure. Perhaps for this reason, many important results have proven hard to reproduce. Adding to the difficulty, Alzheimer’s is poorly modeled in mice. Mice with mutant APP transgenes experience β-amyloidosis but this does not lead to Tau pathology as it does in humans, though co-expression of mutant APP with mutant MAPT in mice does accelerate the formation of Tau tangles [citation needed].

The order of events in Alzheimer’s disease is well agreed-upon. The Aβ cascade hypothesis [Hardy & Allsop 1991, Selkoe 1991], which is more controversial, places causality between each step.

Why oligomers?

The two core arguments for Aβ oligomers as the driver of Alzheimer’s are:

  1. The genetic imperative: all of the human genetics points to Aβ, so if not plaques, then smaller species must be responsible.
  2. Spatial distribution: because Alzheimer’s pathology spreads, it must be mediated by soluble species (such as oligomers) rather than insoluble plaques

However, many of the specific arguments for oligomers can still be debated:

point counterpoint
oligomer levels correlate with disease severity CSF levels of Aβ42 rise long before symptoms occur
synthetic Aβ forms oligomers so what? the process from a monomer to an aggregate has to involve an oligomer intermediate.
oligomers have been shown to be toxic the toxicity assays might not be relevant to the disease. assays in which Aβ acutely kills neurons cannot possibly model Alzheimer’s, which takes years to progress
APP mutations that cause EOAD alter the amount or amyloidogenicity of Aβ they also change other properties and other cleavage products of APP. So really the genetic evidence points to APP and not Aβ per se.
APOE haplotypes differ in Aβ clearance APOE is involved in other neurodegenerative diseases too [citation needed] so APOE involvement does not necessarily pinpoint Aβ.
formic acid-extractable Aβ correlates with disease severity this is based on one study and that study looked only at one brain region
injection of synthetic Aβ is toxic that’s not necessarily a model of the real disease

Alternatives to the oligomer hypothesis

Full-length APP or other metabolites of APP could actually be responsible for the disease, and Aβ could then be merely an imperfect correlate of those. This is the view favored by [Melnikova 2013], discussed below.

Some people still think APP might not be important at all, or might be important only in genetic Alzheimer’s. But this requires ignoring a lot of the evidence.

Aside: native function of APP and presenilins

APP appears to be involved in neuritic outgrowth, neural migration, and so on. It has two paralogs, APLP1 and APLP2. Knockout of any two of the three is viable; knockout of all three is lethal. However it is not clear whether the native function is important to understanding Alzheimer’s disease.

The most important role of gamma secretase and presenilins is cleavage of Notch. Knockout of presenilins is lethal for this reason.

Shankar 2008: Amyloid-β protein dimers isolated directly from Alzheimer’s brains impair synaptic plasticity and memory

Questions for discussion

Q. Discuss whether the observed Aβ dimers are actually present in Alzheimer brains or are dissociated forms resulting from the solubilization protocol and gel electrophoresis method.

A. These bands were observed with three different extraction methods (TBS without detergent, TBS with detergent, and 5M GdnHCl). The TBS treatment is extremely moderate and unlikely to create artifacts. However, after extraction, all of these extracts were subjected to LDS-PAGE, which is very likely to create artifacts. The exact procedure used in creating the Western blots in Fig 1A-C is not stated, instead it is cited to [Walsh 2000]. The protocol from that paper is as follows:

we devised an immunoprecipitation/Western blot (ip/wb) protocol that allowed the highly sensitive detection of unlabeled Aβ species… Samples were immunoprecipitated so as to avoid reconstitution procedures which might alter the assembly form or recovery of Aβ. Following immunoprecipitation (described above), samples were electrophoresed on 16% Tricine gels and transferred onto 0.2 μM nitrocellulose membranes at 400 mA for 2 h.

After reading this, I wasn’t clear whether this procedure actually included the standard step of boiling the extracts in SDS (or LDS). Dr. Walsh clarified that yes, these samples were boiled in detergent before running on the gel. For this reason it is difficult to be certain that the monomers, dimers and trimers in Figure 1 were really present in the original brains. All that Figure 1 can confidently tell you is that the presence of this much soluble Aβ, in whatever multimericity, is specific to AD and Down’s syndrome brains - on thes gels, soluble Aβ is far, far less abundant in brains from other neurodegenerative diseases. A “P-AD” brain with pathology of Alzheimer’s but no clinical symptoms had insoluble plaques (Fig 1C) but not soluble Aβ. An additional limitation is that all of the AD brains examined in this paper were from terminal-stage AD, representing a worst case.

Q. What is long-term potentiation (LTP), and why is it relevant to the study of Alzheimer disease mechanisms?

A. LTP is enhancement of a neuron’s response to stimulus as a result of repeated stimuli. LTP is an important process for learning and cognition. A strength of this paper is that they measure changes in LTP, a more moderate phenotype that could plausibly be caused by Aβ in vivo. This is in contrast to studies that measure neuronal death in response to Aβ, which is a far more rapid and severe phenotype than Alzheimer’s disease progression actually reflects. To measure LTP, you stimulate a presynaptic neuron with a stimulating electode, and measure response in its postsynaptic neuron using a recording electrode.

Q. What is the evidence that soluble Aβ assemblies inhibit LTP?

A. In Figure 2, the TBS extracts of AD brains, but not of non-AD brains, reduce LTP signal. Importantly, the samples were heavily manipulated at many points between the original brain and the data shown in Figs 2 and 3. It is therefore hard to say that dimers present in AD brains inhibit LTP. Rather, you could say that dimers present at the end of this procedure (whether or not they were really present in the original brain) do inhibit LTP.

Q. What is long-term depression (LTD), and why is it relevant to the study of Alzheimer disease mechanisms?

A. A neuron’s desensitization to stimulus following repeated stimuli. Like LTP this is important for cognition.

Q. What is the evidence that soluble Aβ assemblies facilitate LTD? What receptors might mediate this effect?

A. Evidence in Fig 2E. MCPG and SIB, both mGluR antagonists, prevented this facilitation, whereas AP-V, an NMDAR antagonist, had no effect (Fig 2F). Thus, mGluR might mediate the effect.

Q. TBS extract from human AD brains impaired rat memory function (Figure 2G) but extract that had been immunodepleted with anti-Aβ antibody did not. What other control(s) might be appropriate for this experiment?

A. Mock immunodepletion with a random IgG not against Aβ. Also, other fractions of the AD brain extract that contain different Aβ species.

Q. What is the evidence that Aβ dimers specifically can affect neuronal function?

A. To confirm that Aβ dimers are sufficient to affect LTP, they created synthetic Aβ40 and forced dimerization by introducing a disulfide bond. These dimers were toxic, though [Benilova 2012] raises the concern that there is no quantitative comparison of the degree of toxicity or the dose required to achieve this toxicity, of the synthetic vs. natural Aβ dimers. In any event, the evidence here at most shows that Aβ dimers are sufficient for toxicity, not that they are necessary, as they haven’t ruled out trimers or other species also being toxic.

Q. [What is the evidence] that amyloid plaques might contain neurotoxic Aβ dimers?

A. In Figure 4 they centrifuged the brain extracts, washed the pellet in TBS and repeated several times. Eventually the TBS extract ceased to contain detectable Aβ monomers or dimers. The insoluble Aβ plaques did not inhibit LTP, but they could be dissociated into dimers using formic acid, and then those dimers did inhibit LTP.

Extra discussion

Benilova’s critiques of [Shankar 2008] are:

Aβ ran as a dimer in SDS-PAGE and interfered with synaptic function in different paradigms. The authors therefore defined Aβ dimers as the minimal toxic species in vivo, but whether the synaptotoxicity observed in their preparations was indeed directly caused by the dimers present in their purified fractions remains unproven. Furthermore, the exact amino acid composition of the putative ‘dimeric’ species has not been definitively clarified. Additional contamination by unknown protein or lipid, covalent modification of Aβ or the rapid aggregation of the putative dimers into larger structures during assays might provide alternative explanations. The authors confirmed that Aβ dimers made by cross-linking synthetic Aβ were synaptotoxic, although no direct quantitative comparison with the brain-derived dimers was made. Moreover, in a later study, the authors found that these synthetic dimers rapidly aggregate in metastable protofibrils, suggesting that a more compli- cated interpretation of the available evidence is needed.

A couple of other issues we discussed:

  • Fig 2F uses AP-V to antagonize NMDAR, wherease Fig 2H uses CPP, a different antagonist. Different small molecules were used because these two experiments were done in two different labs (Selkoe and Sabatini labs respectively).
  • A comparison is made between two antibodies to the N terminus of Aβ which precipitate soluble Aβ and prevent LTP disruption, and two antibodies to the C terminus, which do neither. It is not clear if this is an inherent property of N vs. C terminal mAbs or if it’s just that they had good N-terminal antibodies and not-as-good C terminal antibodies.

Melnikova 2013: Reversible Pathologic and Cognitive Phenotypes in an Inducible Model of Alzheimer-Amyloidosis

In the prion field, a seminal experiment [Mallucci 2003] showed that turning off the PrP gene just after the onset of prion disease symptoms leads to a full recovery. This is an important proof of principle that if you can find a way to lower PrPC expression, that would be therapeutically helpful. Similarly, since evidence from human genetics places Aβ as the key initiator of Alzheimer’s pathogenesis, you would think that reducing Aβ production therapeutically would also be helpful. This might even prove achievable with β-secretase inhibitors [reviewed in Ghosh 2012]. Yet it is difficult to do for Aβ what Mallucci did for PrP. You want to design an experiment to ask, “if we shut off Aβ production in symptomatic animals, will a recovery ensue, or will the cascade of Tau prions, once initiated, continue unabated?” It is impossible to ask this question because mice with APP mutations don’t get Tau pathology in the first place. Nevertheless, these mice do experience behavioral deficits presumably associated with Aβ itself. Therefore, David Borchelt created a mouse model that expresses mutant APP conditionally, under a Tet-off promoter, and found that amyloid plaques persist after APP expression is turned off [Jankowsky 2005]. In the present paper, they asked whether behavioral deficits could be reversed in these mice after APP expression ceases [Melnikova 2013].

Questions for discussion

Q. What was the rationale for generating an APP transgenic mouse model in which APP expression could be turned off?

A. Broadly, there is the rationale I described above - in an ideal world we would like to know whether Alzheimer’s impairments would continue if we could shut off the production of APP, or of Aβ.

Q. Describe the effect of doxycycline on APP expression and why bigenic animals were used.

A. Doxycycline shuts off transgenic APP production; see above. However, note that these mice are not on a App knockout background, so endogenous mouse APP is still expressed.

Q. Describe the cognitive phenotype of 10-12 month old APPsi:tTA mice.

A. The phenotype is assessed in a series of water maze experiments (bottom right p. 3766). The Y maze measures the mice’s preference for novel spaces. The Morris water maze puts the mice in a chamber with six arms. They first learn that a platform (which they like, because they hate swimming) is in one arm, then they have to unlearn that and figure out that the platform has moved to a new location. APPsi:tTA mice are apparently deficient in both of these tests, though the n is small and the difference is not huge.

Q. Describe how suppression of new APP synthesis altered this phenotype?

A. Behavior improved.

Q. Explain why these authors were careful to examine Aβ in “water-extractable fractions” and in PBS homogenates. What is the evidence that amyloid plaques are unlikely to contribute to the cognitive impairment seen in APPsi:tTA mice? Describe the techniques used to measure Aβ oligomers and comment on whether or not these are appropriate for assessment of native protein assemblies.

A. Soluble Aβ, which has been suggested to be the toxic species, would be expected to be found in these water-extractable fractions. This paper used two different procedures to extract Aβ. One involved sonication in PBS and centrifugation. In the other procedure, to which “water-soluble fractions” refers, they did not homogenize the brains, just let the brains sit in distilled water for one hour. This is a far milder treatment than used in most studies of Aβ. On one hand, this makes it less likely to produce artifacts by solubilizing something that isn’t highly soluble; on the other hand, it would be interesting to know how much total protein is extracted by this method. The two different extraction procedures were performed on mice of different ages.

They also used two different ELISA procedures to measure Aβ:

To determine whether the water-extractable or PBS-soluble fractions contained soluble Aβ42 peptides, we used an ELISA system that uses a C-terminal capture antibody with mAb 4G8 used as the detection antibody.

— p. 3773


In an attempt to detect oligomeric Aβ by ELISA, we used an antibody directed against an N-terminal epitope to both capture and detect an Aβ in the sample, a design similar to that of Xia et al. (2009).

— p. 3774

As mentioned in [Shankar 2008], antibodies to the N terminus of Aβ can cause precipitation - it would be interesting to discuss whether this affects interpretation.

Q. Besides regulatable transgenic expression of APP, what other mouse models approaches could be used to further examine the effects of Aβ on cognition?

A. Conditional knockout of beta secretase.