The analysis of the molecular development of AD (Alzheimer's disease) is technically challenging, due to the chronic nature of the disease, the lack of early and definitive clinical diagnosis, and the fact that the abnormal molecular pathology occurs in the brain. Therefore appropriate animal models of AD are essential if we are to dissect the processes leading to molecular pathology, and ultimately to test the efficacy of potential therapies before clinical studies. Unfortunately, there is controversy over the benefits of the available models, the only consensus of opinion being that no perfect model currently exists. The investigation of animal models is extremely costly and time-consuming, therefore researchers tend to focus on one or two models. For scientists entering the AD research field, it can be difficult to identify the most appropriate model for their needs. Therefore the Models of Dementia: the Good, the Bad and the Future Biochemical Society Focused Meeting provided a platform for discussion and debate on the use and limitations of current models, the most appropriate methods for their characterization and identification of the most pressing needs of the field in general.
- Alzheimer's disease
- amyloid β-peptide (Aβ)
- disease model
- molecular pathology
Background to the meeting
There are many reasons to generate animal models of human disease. For example, the detailed study of the development of AD (Alzheimer's disease) in human patients is practically impossible. The disease is not trivial to diagnose in its early stages, takes many years to fully develop, has genetic and environmental influences (therefore is heterogeneous in nature), and there are ethical problems associated with obtaining permission for inclusion in human studies. In addition, the fact that most of the molecular pathology develops in important areas of the CNS (central nervous system) makes the study of its initiation and progression at the molecular level in humans challenging to say the least. Therefore many research scientists have attempted to develop animal models ranging from the nematode worm (Caenorhabditis elegans), through the fruitfly (Drosophila melanogaster) to rodents and primates. In each case, an aspect of what is known for sure about the human condition (usually a pathological aspect) is induced in the model. Then the subsequent effects on biochemistry, cell biology, physiology and behaviour, as well as the temporal progression of health problems, of the model are studied in great detail. This has generated a lot of insight into the pathology of the disease, but has also resulted in great debate on the benefits and problems associated with each of the many available models. Therefore it is important to have as much information available as possible when deciding on the most appropriate model to use in research into this disease.
Identifying the appropriate model of AD
A few years ago, we wanted to study the post-translational regulation of CRMP2 (collapsin response mediator protein 2; a protein we had identified to be abnormally modified in human AD post-mortem brain) during different stages of development of the disease (not trivial in human cohorts) [1–3]. These data would provide clues as to whether the CRMP2 modification was a cause or consequence of the disease and hence whether it was a possible therapeutic or diagnostic target . The ideal approach would have been to analyse the modification before, during and after the development of the disease in an animal model, and correlate changes in CRMP2 modification with AD pathology and symptoms (cognitive deficits). However, the literature provided multiple possible models to study with little objective information on the benefits or limitations of each model, meaning that potentially dozens of models would have to be examined to cover all eventualities. In large part thanks to the generosity of colleagues, we were able to collect small amounts of tissue from a range of models . Although it meant that the ‘perfect’ study was not possible, I was at least able to establish the relationship of CRMP2 modification to genotype in seven or eight different models. This relied heavily on the availability of models and staff in various laboratories around the world. Our study is still criticized by some due to the fact that the tissue was not all collected by the same group, the background genotype of the different models was variable and the best controls for each were not always available, and we could not study an age-related affect in all models as we had to accept what was actually available at the time from our collaborators. So is there a single model of AD to study the molecular progression of the disease?
Mouse models of AD
The mouse is relatively amenable to genetic manipulation, has a short breeding cycle, and its mammalian physiology (including a relatively developed brain and cognitive readouts) suggests that behavioural deficits may have relevance to human disease. Most current models overexpress mutant human proteins that have been identified in familial human dementia [e.g. APP (amyloid precursor protein), tau, PS1 (presenilin 1) and α-synuclein] in the CNS of the mouse. In general, this results in the development of pathology that has similarities to that seen in the human condition and often includes deleterious effects on electrophysiology and cognitive function (Table 1). These approaches have provided valuable information on the effects of abnormal expression of mutant proteins associated with dementia. However, even these ‘humanized’ models are not regarded by many as the best way to study the human disease. The main criticisms are: (i) none of the current mouse models of AD actually exhibit major neurodegeneration, (ii) almost all are based on familial forms of human dementia (which is relatively poorly represented in the human population), (iii) the plaque and tangle pathology is not identical with that observed in human AD even though they are composed of primarily human protein, and (iv) there is a great deal of variability in the characterization and development of models between laboratories and between models expressing different mutations of the same proteins.
Unfortunately, even models expressing the same transgene will express the protein at different levels or with distinct regional expression, resulting in distinct phenotypes. In addition, even the same genetic model when studied in different laboratories can result in different phenotypes (Table 1), presumably due to variations in housing, husbandry, breeding or phenotyping practices.
A good example of the variable phenotype of a model is the triple transgenic mouse (3×Tg) . This model develops both plaque and tangle pathology due to the presence of APP, PS1 and tau transgenes. Currently, there are at least 50 published papers which use the 3×Tg mouse to study features as diverse as amyloid deposition, neuronal functioning, cognitive and behavioural activity, stress and metabolism. In addition, a number of emerging treatments designed to slow or reverse the development of pathological features observed with aging have been tested in this model. The 3×Tg mouse accumulates intraneuronal Aβ (amyloid β-peptide) followed by the development of extracellular plaques; this occurs initially in the cortex before progressing to the hippocampus . In tandem with this process, hyperphosphorylation of tau occurs along with the development of neurofibrillary tangles. In contrast with Aβ pathology, the tau deposition initiates in the hippocampus and spreads to the cortex, with tangles present only in relatively aged mice. Unfortunately, temporal and spatial variation in the progression of these features has been widely reported (Table 1). The original characterization of the model reported the presence of intraneuronal Aβ reactivity at 3–4 months in the cortex and by 6 months in the CA1 region of the hippocampus. Extracellular plaques then begin to develop at 6 months in the frontal cortex and are clearly apparent by 12 months, when plaques begin to develop in other regions of the cortex and hippocampus . Meanwhile, intraneuronal Aβ was also reported at an even earlier age within midbrain and brainstem regions . España et al.  reported a slightly slower progression with intraneuronal Aβ at 6 months in the cortex and hippocampus, and plaques present by 12 months, whereas Hirata-Fukae et al.  observed intraneuronal Aβ at 6 months in the cortex and 9 months in the hippocampus, a full 3 months later than the original report. This group reported a sex difference in amyloid deposition, but not tau pathology, something not reported in the original study. In addition, there was a later onset of plaque pathology (>14 months of age). Around the same time, other researchers found early intraneuronal Aβ (from the age of 2 months), but late onset plaque pathology within the hippocampus (>15 months of age), and no plaque pathology within the cortex until 18 months of age . Whether these variable observations of pathology in the same genetic strain reflects differences in the practicalities of assessing intraneuronal Aβ and plaque deposition or real variations is unclear.
In this same model, increased tau phosphorylation at sites recognized by the AT100 and 12E8 antibodies (Thr212/Ser214 and Ser262) was observed at 6 months, whereas phosphorylation at the AT8 (Ser202 and Thr205) and AT180 (Thr231) sites was not detected until 12–15 months (along with conformational changes detected by the MC1 antibody) [28,40]. Neurofibrillary tangles were not evident until 12–18 months of age [28,40]. As with plaque pathology, there are some discrepancies reported in the temporal and spatial development of tangles between groups. Mastrangelo and Bowers  observed increased AT180 staining (suggesting phosphorylation of Thr231 of tau) in the hippocampus of 6-month-old 3×Tg mice, 6 months earlier than reported previously . This group also failed to observe substantial tangle pathology until the late age of 23 months, although they did detect some PHF-1 (paired helical filament 1) immunopositivity at 15 months within the hippocampus. More recently Oh et al.  reported multiple phosphorylation events on tau, including increased AT180 staining as early as 3 weeks of age in the amygdala and cortex of the 3×Tg mice, although tangles did not appear until the age of 23 months.
Deficits in non-pathological measures such as the electrophysiology of neuronal populations in 3×Tg mice have also been found to vary between colonies of mice. The original report of the 3×Tg found a marked reduction in LTP (long-term potentiation) in 6-month-old mice , but a subsequent study reported that LTP is normal in 10-month-old 3×Tg mice .
Cognitive and behavioural alterations have been widely observed in transgenic mouse models expressing APP, PS1 and tau mutants, including the 3×Tg mouse. These can be a useful feature to study the relationship of pathology to the clinical symptoms of AD, such as decline in memory function and specific changes in behaviour. The challenge faced by researchers is to choose the appropriate behavioural task and to link the phenotype to the development of the pathological features within the appropriate brain regions; this task is made more difficult in the 3×Tg mouse by the discrepancies in the time course of these biochemical events.
The general consensus is that the 3×Tg mice show limited exploratory behaviour and freezing during multiple tasks, which may be a sign of increased emotionality or anxiety [36,42,43]. However, a number of different paradigms can be used which each test a slightly different aspect of cognitive function. It is not clear whether testing hippocampal-dependent memory  or contextual fear conditioning, which is thought to be amygdala-dependent , is most appropriate as a symptom of AD. Deficits in each occur in the 3×Tg at different ages (Table 1), but whether this is related to progression of the disease or variability in plaque or tangle load between colonies is unclear.
However, these variations are not unique to the 3×Tg mouse (Table 1). Therefore designing prospective studies to tease out the effects of intraneuronal Aβ or plaque or tangle on behaviour or neurobiology using the models becomes quite tricky. Clearly, the appearance of pathology needs to be assessed in every mouse, and the study has the potential to take anything between 2 and 18 months, making costing and staffing difficult to plan.
It has been suggested that the presence of multiple individual breeding colonies, each descended from multiple generations of mice, has resulted in a degree of drift in the phenotype which has resulted in the pathological features being expressed at a later time point than originally reported. It is therefore possible that individual colonies of transgenic mice can be characterized to minimize subsequent variation and aid the planning of experiments. Many of these issues are addressed in much more detail in the subsequent papers in this issue of Biochemical Society Transactions.
Other models of dementia
The large number of variations of models based on expression of mutant proteins linked to the pathology of AD makes the choice of model difficult. However, as the majority of human AD is sporadic and not familial, one could argue that none of these models are relevant to the study of the majority of human disease. However, what are the alternatives? Recent information from human genome-wide association studies and epidemiology of sporadic AD has suggested that the presence of co-morbidities greatly influences the development of AD pathology and symptoms. The incorporation of these co-morbidities into modelling of AD is only just being realized and is also covered in the following papers.
Until recently, the use of non-mammalian organisms, such as C. elegans and Drosophila, to model human brain disease would have been considered inappropriate. However, groundbreaking work in these organisms is establishing them as excellent models of specific aspects of the molecular development of AD, and their inherent advantages of low cost and rapid data generation is enhancing their development in this field.
Aims and conclusions of meeting
The initial aims of the meeting were: (i) to debate the benefits and limitations of current models of dementia and neurodegeneration, (ii) to identify areas of research that require support in order to improve modelling of dementia, and (iii) to provide a published record of the meeting to inform researchers and funding bodies.
Despite a general acknowledgement that there is no perfect model of all aspects of dementia, it is clear that there are good models of specific aspects of the disease. However, it is vital to be aware of the limitations as well as the benefits of the various models, and what each is actually modelling. One major conclusion of the meeting was an emphasis to develop models that are not based purely on the familial form of AD. For example, it is quite likely that sporadic AD does not occur in humans as a pure disease, but is strongly influenced by co-morbidity, such as inflammation, metabolic disease and neurological stress. Therefore the scientific community has to be more ingenious in their attempts to model (and visualize) early biochemical and cellular processes that lead to neurodegeneration, rather than simply overexpressing a mutated form of APP or tau. This would be aided by a more fundamental understanding of the biology (neurochemistry, neuroanatomy and psychology) of different forms of dementia. Although the recent genome-wide association studies will focus more attention on biological processes that influence the risk of developing dementia, there is also a desperate need for funding of pure neuroscience research to understand how different risk factors (e.g. aging, stress, infection and diabetes) exacerbate neurodegenerative (synaptic dysfunction and neuronal death) processes and how they interact with the amyloid cascade and tangle formation. Finally, it is also quite clear that generating accurate models of dementia requires more accurate and earlier diagnosis of the different forms of dementia in the human population. What are we actually trying to model? This would permit more informative biochemical and imaging analysis of the early stages of specific forms of dementia in the best model of all, the human being. Only then will the most appropriate strategies for modelling in animals or cells become apparent.
The following articles provide more detailed information on models based on pathology, novel models of dementia and relevance of the models to human disease. The organizers hope that this proves useful to the research community.
Work in the C.S. laboratory is supported by the Medical Research Council, Diabetes UK and Alzheimer's Research UK, whereas L.G. is funded by the Medical Research Council and Tenovus.
We greatly appreciate the work of our co-organizers of the meeting, Dr Stuart Allen, Dr Christian Hölscher, Dr Karen Horsburgh and Professor Simon Lovestone, but make special mention of Elizabeth Faircliffe for the highly professional and smooth organization of the whole conference.
Models of Dementia: the Good, the Bad and the Future: A Biochemical Society Focused Meeting held at Robinson College, Cambridge, U.K., 15–17 December 2010. Organized and Edited by Stuart Allan (Manchester, U.K.), Christian Hölscher (University of Ulster, Coleraine, U.K.), Karen Horsburgh (Edinburgh, U.K.), Simon Lovestone (King's College London, U.K.) and Calum Sutherland (Dundee, U.K.).
Abbreviations: Aβ, amyloid β-peptide; AD, Alzheimer's disease; APP, amyloid precursor protein; CNS, central nervous system; CRMP2, collapsin response mediator protein 2; LTP, long-term potentiation; PS1, presenilin 1
- © The Authors Journal compilation © 2011 Biochemical Society