The amyloid hypothesis of Alzheimer's disease proposes that deposition of beta-amyloid (Aβ) protein is central to the pathogenesis of the disorder. Reference Hardy and Higgins1 It has been the foundation of efforts to understand the disease for almost 25 years. However, recently its validity has been called into question after the failure of Aβ-targeting therapies in clinical trials. Some have even suggested that the causal link between Aβ and Alzheimer's disease has been refuted. Reference Drachman2
The pathology of Alzheimer's disease
The classical histological features of Alzheimer's disease are a triad of Aβ plaques, neurofibrillary tangles and neuronal cell loss. Reference Selkoe3 The first of these are insoluble extracellular plaques consisting of Aβ, which accumulates in very high levels in the brains of those with Alzheimer's disease. Aβ is derived from a larger molecule, amyloid precursor protein (APP). APP is a trans-membrane protein, with a long extracellular N-terminal and a shorter intracellular C-terminal. The Aβ sequence consists of some of the extracellular portion and part of the trans-membrane domain, and is 39–42 amino acids in length. The protein has a β-pleated sheet structure, and demonstrates Congo red birefringence and resistance to proteolysis. Reference Hardy and Allsop4 In Alzheimer's disease, Aβ is deposited in abundant extracellular plaques typically composed of straight fibrils, 6–10 nm in diameter. These structures are also found in normal ageing but in less profusion and are sometimes referred to as senile plaques. They are associated with dystrophic neurites and changes in microglia and astrocytes. Reference Selkoe3 Non-fibrillar, diffuse Aβ deposits, which are not associated with dystrophic neurites or reactive glial cells, are also found in Alzheimer's disease and these may represent an early stage plaque formation. In Alzheimer's disease these diffuse plaques are found throughout the central nervous system, whereas typical Aβ plaques are not present in regions such as the spinal cord and cerebellum. Reference Hardy and Allsop4
The second pathological structure found in Alzheimer's disease is the neurofibrillary tangle. These consist of dystrophic neurites containing paired helical filaments, 10 nm in diameter. These paired helical filaments consist of a phosphorylated microtubule-associated protein, tau (MAPT). Reference Selkoe3 In the 1980s there was much debate as to which one of these is the primary driver of Alzheimer's disease pathogenesis. The issue seemed to be resolved with the advent of a new generation of molecular genetic studies. Reference Tanzi5
Early molecular genetic studies of Alzheimer's disease focused on rare families where the disorder occurs exceptionally early and follows an autosomal dominant mode of inheritance. It was discovered that so-called familial Alzheimer's disease is caused by mutations either in the APP gene itself, or in presenilin 1 and 2 (PS1 and PS2) that are involved in cleaving Aβ from APP. Reference Tanzi5 In addition, Alzheimer's disease frequently affects those with trisomy 21, who have a triplication of the APP gene. Reference Tanzi5 On these grounds, Hardy & Allsop Reference Hardy and Allsop4 postulated that APP mismetabolism and Aβ deposition are the primary events in the disease process with tau phosphorylation and neurofibrillary tangle formation occurring downstream. This became known as the amyloid hypothesis. It later transpired that the familial Alzheimer's disease genes increase levels of 42 amino acid Aβ (Aβ42) relative to the shorter 40 amino acid protein, and this form of Aβ aggregates more readily into plaques. Reference Sambamurti, Vasudevaraju, Mintzer, Bachman and Granholm6
The hypothesis has received further support from widely replicated biomarker studies. Brain Aβ deposition in Alzheimer's disease can be demonstrated in vivo using biomarkers such as cerebrospinal fluid (CSF) Aβ42 and Aβ positron emission tomography (PET) imaging. Reference Jack, Knopman, Jagust, Shaw, Aisen and Weiner7 Clinical diagnoses of Alzheimer's disease and Aβ pathology at autopsy correlate with low concentrations of CSF Aβ42. Most patients with a diagnosis of Alzheimer's have increased retention of radioligands for Aβ on PET. Moreover, low CSF Aβ and positive Aβ PET show nearly 100% concordance. Reference Jack, Knopman, Jagust, Shaw, Aisen and Weiner7
Thus, a substantial body of evidence appears to support a causative, pathogenic link between Aβ and Alzheimer's disease. However, there are a few pieces of the Alzheimer's jigsaw that do not quite fit.
Challenges to the amyloid hypothesis
Alzheimer's is not an all-or-none phenomenon even at the neuropathological level. Moreover, autopsy studies find sufficient numbers of Aβ plaques and neurofibrillary tangles to meet criteria for a diagnosis of Alzheimer's disease in around a third of cognitively intact elderly people. Reference Rodrigue, Kennedy, Devous, Rieck, Hebrank and Diaz-Arrastia8 This is corroborated by biomarker studies, which suggest that 20–40% of elderly people without cognitive impairment show significant brain Aβ load, either on Aβ PET or CSF Aβ42 concentrations. Reference Rodrigue, Kennedy, Devous, Rieck, Hebrank and Diaz-Arrastia8
The topographic distribution of Aβ plaques differs from neurofibrillary tangle deposition and neurodegenerative changes. In early Alzheimer's disease, neural loss is predominantly in the hippocampus and entorhinal cortex, whereas plaques are first found in frontal regions, basal ganglia or elsewhere. Reference Jack, Knopman, Jagust, Shaw, Aisen and Weiner7,Reference Braak and Braak9 Clinical symptoms are more closely associated with neurofibrillary tangles than Aβ burden. However, cerebral atrophy, representing neuron and synapse loss, corresponds best to cognitive impairment. Reference Jack, Knopman, Jagust, Petersen, Weiner and Aisen10
How distant Aβ plaques might induce neurofibrillary tangles or damage neurons is unclear. It has been proposed that soluble oligomers of Aβ could be neurotoxic. Although soluble oligomers cannot be seen in vivo or post-mortem, they have been found to interfere with postsynaptic potentiation in tissue culture studies. However, the concentration of Aβ oligomers shown to have this effect is greater than usual physiological levels. Reference Karran, Mercken and De Strooper11 Another suggestion is that Aβ plaques could act as a ‘reservoir’ eluting soluble Aβ, but Aβ has a strong tendency to polymerise and fix fragments to plaques, which makes this less likely. Reference Karran, Mercken and De Strooper11 Furthermore, many animal models based on APP and PS1 mutations have not shown progression to synaptic loss, neurofibrillary tangle formation and neurodegeneration. Reference Sambamurti, Vasudevaraju, Mintzer, Bachman and Granholm6
Critics of the amyloid hypothesis also point out that familial Alzheimer's disease, where the aetiological link with APP is strong, is rare and might be an atypical form of the disorder. They point to recent genome-wide association studies (GWAS), which have implicated many novel genes as containing risk factors for typical Alzheimer's disease but not APP or its metabolising enzymes. In defence of the amyloid hypothesis, GWAS only assess common genetic variation and failure to find association does not exclude an important role for a protein in disease. Moreover, some of the genes implicated by GWAS may be involved in Aβ processing. For example, CLU encodes clusterin, which binds soluble Aβ in animal models, forming complexes that can cross the blood–brain barrier, and PICALM encodes phosphatidylinositol binding clathrin assembly protein, which has been postulated to increase Alzheimer's disease risk through APP processing via endocytic pathways, resulting in changes in Aβ levels. Reference Harold, Abraham, Hollingworth, Sims, Gerrish and Hamshere12
The biggest challenge to the amyloid hypothesis has come from the failure of phase III trials of anti-Aβ therapies Reference Drachman2 despite promising results in animal models. Reference Sambamurti, Vasudevaraju, Mintzer, Bachman and Granholm6 Critics point to notable therapeutic failures, such as semagacestat, an inhibitor of gamma (γ)-secretase, an enzyme responsible for the cleavage of APP to produce pathogenic Aβ. Reference Doody, Raman, Farlow, Iwatsubo, Vellas and Joffe13 Participants in the active treatment arm had poorer cognitive outcomes, and the trial was terminated. The trial of immunisation with aggregated human Aβ, AN1792, was halted when some participants developed autoimmune encephalopathy, and there was no effect on disease progression. Reference Gilman, Koller, Black, Jenkins, Griffith and Fox14
This raises a key question. Does the failure of these trials effectively refute the amyloid hypothesis? There are two general reasons why this conclusion might be premature. First, there could have been insufficient target engagement. Second, the drugs may have been administered too late in the disease process.
In an analysis of six programmes testing anti-Aβ therapies, Karran & Hardy Reference Karran and Hardy15 identified various deficiencies. This suggested that the failure of phase III trials of anti-Aβ agents might be because of problems with pharmacokinetics, dosing, outcome measures etc., rather than shortcomings of the amyloid hypothesis. Reference Karran and Hardy15 An analysis by pharmaceutical industry investigators Reference Toyn and Ahlijanian16 reached a similar verdict: negative trials did not demonstrate sufficient target engagement to assess whether reducing Aβ load could modify the course of Alzheimer's disease.
Early research assumed that abnormal deposition of Aβ is a proximal cause of neurodegeneration in Alzheimer's disease. An alternative model is that production of Aβ at abnormal levels begins much earlier in life, is integral to the inception of the disease process, but the subsequent pathological cascade becomes autonomous. If this were true, anti-Aβ therapies would only be effective if administered early in the disease process. Support for this view comes from biomarker studies showing a clear sequence of abnormalities as the disease progresses. Reference Jack, Knopman, Jagust, Petersen, Weiner and Aisen10 The earliest markers of brain Aβ deposition are reductions in CSF Aβ42 followed by increased Aβ PET tracer uptake. These changes occur in the ‘preclinical’ phase Reference Sperling, Aisen, Beckett, Bennett, Craft and Fagan17 and, by the time cognitive impairment is clinically detected, Aβ markers have plateaued. Subsequently, neuronal injury and neurodegeneration predominate. These are shown by increased CSF tau and cerebral atrophy on structural magnetic resonance imaging. Decreased fluorodeoxyglucose uptake on PET indicates accompanying synaptic dysfunction. These markers, which become abnormal later in the disease, correlate closely with clinical symptoms Reference Jack, Knopman, Jagust, Shaw, Aisen and Weiner7 (Fig. 1).
The failed therapeutic trials may also support this chronological pattern. For example, the AN1792 immunisation, Reference Gilman, Koller, Black, Jenkins, Griffith and Fox14 which did not improve cognition and had minimal effect on neural loss, gliosis and tau accumulation, did reverse Aβ deposition. The number of patients followed up, however, was small. This is consistent with the view that, in its later stages, the pathogenic process is not dependent on Aβ deposition, even if it is triggered by it. This means that the timing of any intervention in the Alzheimer's pathological process is crucial. It also makes Aβ levels an equivocal proxy end-point for clinical trials at least in the later stages of the process. Reference Hyman and Sorger18
Among the strongest recent evidence that reducing Aβ cleavage may protect against Alzheimer's disease is the discovery of a rare mutation in the APP gene that is associated with decreased Aβ synthesis and which protects against Alzheimer's disease and cognitive decline in elderly people. Reference Choi, Kim, Hebisch, Sliwinski, Lee and D'Avanzo19 Further support for the amyloid hypothesis has come from recent work using human neural stem-cell-derived cultures in a three-dimensional system to model the effects of familial Alzheimer's disease mutations. They deposited Aβ and aggregates of phosphorylated tau. Furthermore, inhibition of Aβ generation with β- or γ-secretase inhibitors decreased both Aβ and tau. Reference Choi, Kim, Hebisch, Sliwinski, Lee and D'Avanzo19
The emerging evidence suggests that Aβ deposition occurs early in Alzheimer's disease, and the correct timing of interventions may turn out to be crucial. Identifying asymptomatic individuals who have Aβ-related neurodegeneration will be important, as not all patients currently selected for Alzheimer's disease drug trials have Aβ-positive PET imaging. Reference Vellas, Carrillo, Sampaio, Brashear, Siemers and Hampel20 Trials of anti-Aβ therapy in cognitively normal elderly people with positive Aβ biomarkers and younger, asymptomatic, individuals who carry a familial Alzheimer's disease mutation are underway Reference Sperling, Rentz, Johnson, Karlawish, Donohue and Salmon21,Reference Mullard22 and, if successful, will support the need for anti-Aβ therapies to be given before cognitive decline has become established.
There is increasing interest in the possibility that the key to treating the disorder may be to identify those at risk long before they develop symptoms, as hyperlipidaemia and hypertension are treated years before myocardial or cerebral infarction. This will require the right anti-Aβ drugs to be given to the right patients, those at high risk and at the right time, before irreversible changes have taken place. Reference Jack, Knopman, Jagust, Petersen, Weiner and Aisen10 This will require much further research. Although biomarkers can detect asymptomatic amyloidosis, large longitudinal studies will be needed to investigate their usefulness as predictive tests. Reference Karran, Mercken and De Strooper11 Genetic profiling may also have a role, using risk profile scores based on panels of single nucleotide polymorphisms that are associated with increased risk. Reference Harold, Abraham, Hollingworth, Sims, Gerrish and Hamshere12
Despite two decades of intensive work, the amyloid hypothesis has not led to the hoped for therapeutic advances. This has caused some to question its validity and to ask whether efforts aimed at reducing Aβ synthesis are ever likely to be successful. The finding of a mutation in APP that protects against Alzheimer's disease and reduces the production of Aβ suggests that it may be premature to write off the amyloid hypothesis. Moreover, the negative results of therapeutic trials should be interpreted in the light of evidence that Aβ deposition occurs early in the preclinical phase of the illness. The emerging paradigm of targeting treatments at asymptomatic high-risk individuals remains untested, but if this gains support, it will signal a sea change in the way in which Alzheimer's disease is treated with a move from tertiary to secondary prevention.