The pharmaceutical industry has invested a great deal of time and finance in the development of therapeutics targeting amyloid generation, signalling and plaque stability. This has been based on the amyloid cascade hypothesis which states that abnormal amyloid precursor protein processing and the formation of amyloid plaques is the central process in the development of the symptoms of Alzheimer's disease. However, most clinical trials in this area have been disappointing; therefore the attendees of the Models of Dementia: the Good, the Bad and the Future meeting were given the opportunity to openly debate the proposal ‘the amyloid cascade has misled the pharmaceutical industry’, with the main contributions from Professor John Hardy and Professor John Mayer. The present article is a representation of the debate.

Introductory talks

John Hardy: I am here to talk about the amyloid hypothesis and whether it has been misleading. I should say that, although I get credit and blame for the hypothesis, I don't think of it as mine. I give credit to George Glenner and Colin Masters for the discovery of the amyloid hypothesis.

In the 1980s, there was no structure to AD (Alzheimer's disease) research with dozens of theories. It is generally accepted that there are three parts to AD: plaques, tangles and cell damage. The three elements are considered to have different importance by different scientists. As a geneticist, I considered myself unbiased, as I knew little about the disease, so when we found amyloid mutations, I thought that it was clear-cut, but we predicted that more mutations influencing amyloid would have the same effect as amyloid mutations, and presenilin mutations confirmed this. Other pathological features must be downstream and tau mutations were then found. I am convinced that, for early onset of the familial type of AD, the hypothesis is proven. The situation remains less clear for sporadic late-onset disease.

The current models are models of the early-onset form of AD. Are these models useful for sporadic disease? I think to some extent they are; for example, amyloid immunization works on pathology in mice and humans. Have drugs developed in these mice been successful in preventing AD in humans? We simply don't know yet. This may be due to the lack of information as the trials in humans have been pulled before valuable results have been received. We need positive controls for clinical trials for drug tests. The fact that we have not included early-onset familial disease (which got this research going) is bad science, as including a small number of individuals with APP (amyloid precursor protein) or presenilin mutations within each trial would provide solid positive controls.

The collection of these individuals would act as an intermediate model between mice and late-onset AD. Although challenging this is vital.

John Mayer: I take one exception to what John said, hallmark AD pathology is not end-point disease, they don't become demented because of hallmark AD, they die with something else, neuronal death. In almost all cases, the current models don't have neuronal death; however, cell death is clearly at the heart of the disease (whether it's end-stage or not).

Therefore I believe that we should focus on cell death for modelling not just this disease, but the spectrum of diseases including other dementias, and, of course, co-morbidities. In AD, there are pathologies other than plaques and tangles, e.g. Lewy bodies. Therefore AD is not a straightforward disease.

Although AD is clearly genetic, could there also be another reason that neurons within the brain die regionally in order to cause disease? Clinical scientists are concerned that the treatments for different types of AD should be developed. Neurons may be dying for many reasons within the cortex which may lead to the development of Lewy bodies. We simply don't understand neuronal death. If hallmark pathology is part of a defence mechanism, then targeting it would be dangerous. What if we only see plaques and tangles around and in surviving neurons.

There are many ways neurons could be dying, so you have to be able to model neuronal death (and gliosis) as well as hallmark pathology.

Parkinson's disease can be mimicked by depletion of 26S proteosomes specifically in dopaminergic neurons and we see Lewy-like bodies, similar results are found when proteasomes are depleted in the forebrain.

Similarly, if you delete autophagy genes, you see neurodegeneration, actually in the same parts of brain where neurodegeneration occurs in humans.

So, I am not disputing the importance of amyloidogenic proteins, but there is more involved: you do not lose cells by overexpression of amyloidogenic proteins in models.

I strongly urge you all to consider more than the amyloid cascade hypothesis in order to understand what is really going on in idiopathic disease.

Comments and questions from the floor

Simon Lovestone (chair): Why don't mice get AD?

JH: I don't know. The system which worries me most with respect to sporadic AD is the vascular system. There is a lot of evidence pointing towards the neurovasculature playing a key role in sporadic disease. The vascular systems of a mouse are very different to that of humans and therefore this may have a strong effect on why mice don't get AD?

JM: Generally, people blame the mice due to its shorter lifespan. However, this is not enough reason. There must be something else. We simply do not have the understanding of the death of neurons in mice or humans.

Contribution from unidentified audience member: In response to the suggestion that the current models are not useful for establishing mechanism, I would argue that we have identified some of the signalling steps involved in neuronal death using model organisms, molecules such as HDAC3 (histone deacetylase 3) and GSK3 (glycogen synthase kinase 3) for example.

Maria-Grazia Spillantini (University of Cambridge): Mice and humans are quite different in tau patterns. The difference is in isoforms/splicing. Clearly, there is a difference in the function of tau between the species and therefore the neurons don't function in the same way.

Frances Edwards (UCL): I agree that humans are definitely the best models, but this is not practical. Cell death is probably the least important point of AD because it is so far along in the progression of the disease. We should try and identify the pathway of degeneration from the start and the changes which occur earliest in the disease. Therefore if, as is likely, the animal models represent similar changes to those occurring early in human disease, then they are useful. The biggest problem with the familial animal model is that they may only represent the small proportion of AD cases with these mutations and therefore developing sporadic models is very high priority. Also, all of the familial models overexpress the genes in high doses and therefore it's like giving drugs in high doses.

We need sporadic models!

JH: I agree, but I believe we need a deep understanding of APP function. Understanding the function would lead to its effects in late-onset disease.

Andrew Doig (University of Manchester): In response to where are we going wrong with models, I think that the concentration of Aβ (amyloid β-peptide) is wrong in models. Aβ is toxic when it aggregates and this is highly sensitive to concentration. We know from in vivo evidence that anything that lowers rates of degradation (or when we increase the concentration of Aβ) in the brain leads to early-onset AD. Mechanisms for toxicity are not clear, but are all very concentration-dependent, so we need models with lower concentrations of Aβ and timescales that are much longer.

SL: Would you not accept that, whereas they may not model the disease accurately because the concentration is so high, if you develop a drug which can clear the pathology that's induced by a really high amount, then that ought to be a really potent drug when the pathology is produced by a small amount?

AD: The pathology may be different in each case because, if its forming ion channels, these might only form at high concentrations of Aβ which is never encountered in the brain.

Yadong Huang (Gladstone Institute of Neurological Disease): I think that we have underestimated the ApoE (apolipoprotein E) effect in late-onset AD. This is the major genetic risk factor in late-onset disease. We can see clinically ApoE-driven effects in young individuals. Even if we compare human and mouse proteins, they are usually quite similar; however, human and mouse ApoE are quite different, especially regarding the Aβ metabolism. If crossing APP mouse with ApoE-knockout mouse, you will get less Aβ accumulation, suggesting that mouse ApoE strongly stimulates Aβ deposition. But human ApoE actually reduces Aβ accumulation in this model, suggesting that human ApoE actually stimulates Aβ clearance. We are actually dealing with two different effects on Aβ accumulation depending on the presence of human or mouse ApoE in many mouse models. When a drug is developed, especially for late-onset disease, ApoE4 should considered either as an independent target or as a combined target when testing Aβ or tau-related drugs.

SL: Why is it that this is the most powerful gene effect that we're ever likely to see for a complex disorder, and yet it's one of the most under-researched proteins/gene in AD with the exception of your group?

YH: From the day ApoE was identified as a risk factor, it has been identified as a risk factor only. However, I believe that it's much more than a risk factor: it's one of the main causative factors of late-onset AD. It may be that it provides a combined affect with Aβ or tau. AD may not be a ‘disease’, but may be a syndrome. If we will have successful drugs, it should be a cocktail of drugs targeting a number of proteins, including ApoE4. Clearly, ApoE4 is understudied.

Thomas Bayer (University of Göttingen): I think both Johns are right. From a genetic point of view, the β-amyloid hypothesis is central. But have we been misled by the animal models we have used? I think so. If you ask a molecular biologists or cell biologist what is AD as a concept, they might say that if my cell expresses Aβ, then that is AD. Or if I have plaques in a mouse, this represents AD. Many mouse models are now available with a lot of plaques, but with no neuron loss. If we get a drug that removes the plaques from these mice, then we have presumed in the past that this is an anti-AD drug. One example is active or passive immunization treatments of AD mice. This worked very well and also very well in subsequent clinical trials in humans, but only at getting rid of plaques. When we have a drug with a strong disease effect, we should at least rescue neuron death in the second-generation mouse models (with neurodegeneration, neuron loss or robust behavioural deficits). Only then should we go into clinical trials. Another issue that we need to address is that we produce lots of Aβ daily, so what is the physiological consequence of the rising and falling levels? What is the normal function of Aβ? Also, we completely neglected that Aβ is not Aβ. For example, in human brains, there are a variety of different Aβs. Truncated peptides are more toxic than the physiological 1–40 and 1–42.

Jean Manson (The Roslin Institute): Once you start dealing with models of chronic neurodegeneration, things become infinitely more complex. In a single model, what it's starting to look like is that there are many ways to kill a neuron. In different regions of the brain, the neuron can die in very different ways. We have developed models in which we can protect neurons which are normally the target of disease. However, in these models, protecting the neuron does not protect the animal from dying of neurodegenerative disease; it just takes a lot longer to kill it. It is critically important to understand how other cells interact with neurons. We stopped the accumulation of abnormal protein in the neuron, but the other cell types in the brain then take over the process, and the animal ultimately dies with a neurodegenerative disease. So targeting the neurons may not cure the disease. Someone suggested that early time points should be where we're looking, which I agree with, but only when you know that the end point in your model is neurodegeneration

Calum Sutherland (University of Dundee): Is there a danger in describing these models as ‘early-stage disease’, is it possible that we're just modelling a risk factor? The danger in trying to treat a risk factor is that you're not really helping too much, just reducing risk and you will not cure disease. The relationship with age is very dramatic in AD, and so is it possible that these potential risk factors need kick-starting by something else. The well-touted two-hit hypothesis? Also, if you are to treat a risk factor, the treatment will normally have to start so early that it's unrealistic in AD. In other words, is the risk factor of amyloid the wrong target for the drug companies?

JH: I take what you say, Calum. One of the things I've thought about is that Aβ's function is to act as a rapid sealant for haemorrhaged blood vessels and that is actually its physiological function. In familial cases, what you have is a hyper-primed system, so familial cases with APP mutations only need a trivial vascular accident to set them off in a cycle of super-inflammation of the blood vessel wall. What Calum is saying is that maybe there is a trigger, such as vascular damage, for AD, and I think that is a reasonable thing to suggest.

SL: If it was a dual-hit hypothesis in AD, then the second event cannot be a random event, not with the age of onset in the family being so tight.

CS: I would like that to be debated. What could the second hit be as it seems to be so age-dependent?

SL: It would be unusual to have the second hit to be so age-dependent.

JM: I am worried whether removing amyloid or stopping neurofibrillary tangle formation with drugs is really the way to go at all? Do you think all neurons in the brain die at the same rate or regionally the same? I'm not sure that neurons die at the same rate, and so treating early may affect certain areas, but not all areas of the brain. So where is the target?

CS: My point is that with models based solely on familial disease, we won't get that second hit.

JH: I'd like to emphasize that we have not tested this yet. We need more time for the clinical trials to establish whether the amyloid clearance is effective on the disease.

Bettina Platt (Aberdeen University): I feel we need to make a clearer distinction between the amyloid hypothesis and the plaque hypothesis. I think we don't understand what plaque formation really means. I think in some people it may be beneficial to deposit amyloid into plaques, which is why we see elderly people who are doing cognitively well and they have masses of plaques, but for other patients this may not be the case. I think some of the blame has to lie with scientific communities, but also pharmaceutical companies who are still pushing for reduction of plaque load, which forces people to work with models which have high plaque loads, whether they have a cognitive phenotype or not. Obviously, we don't have time to work with animal models which need to age for 15/20 months, no funding body will give you money for such a model, which causes a big problem in the AD field.

Colm Cunningham (Trinity College Dublin): I would like some feedback on the idea of co-morbidity. We know that, as people age, many of them have amyloid in the brain and don't suffer dementia. There is a lot of co-morbidity in those that go on to dementia. It's interesting that in the animal models which haven't really delivered neurodegeneration, the answer typically is to put another mutation in or overexpress another gene, so that we now have animal models with lots of mutations, but no AD patients carry all of these mutations. There are very few investigations into whether there is a single mutation and then a co-morbidity, such as diabetes.

JH: I agree. Five years ago, none of these factors would have been discussed; however, a number of the talks at this meeting have discussed the issue of co-morbidity. I believe that co-morbidity is the rule and not the exception.

Christian Hölscher (Ulster University): I agree that AD is a multifactorial disease and it has many facets, and so all of these points are important and there may be a huge collection of patients who are affected on a number of different levels. I think that it is clear that we need to investigate all of the different mechanisms that lead to the common clinical symptoms.

SL: can the two Johns finish by commenting on this. The drug discovery programme has been using these animal models for 15 years or so. If you knew then what we know now, would you have done anything different?

JH: No, I wouldn't. If I was a young researcher now, I would try to work out what the function of APP is.

JM: I don't think I would have done what they did, because they didn't have enough knowledge and we still don't. Part of the purpose of this meeting is to make you all go away and think. I would like to know about synaptic homoeostasis in these diseases, as well as neuronal death, because everyone says synaptic dysfunction comes first, but are they right? If they are, how do synapses work and how do they communicate with each other and, if they do disappear, could you slow it down? And also, I would like to understand biochemical mechanisms which are involved with cell survival and death. If you knew much more about these processes, then you could identify new targets for disease.