Abstract
Alzheimer’s disease (AD) is a clinicopathologically defined syndrome leading to cognitive impairment. Following the recent failures of amyloid-based randomized controlled trials to change the course of AD, there are growing calls for a re-evaluation of basic AD research. Epidemiology offers one approach to integrating the available evidence. Here we examine relationships between evidence from population-based, clinicopathological studies of brain aging and a range of hypotheses from all areas of AD research. We identify various problems, including a lack of systematic approach to measurement of clinical and neuropathological factors associated with dementia in experimental and clinical settings, poor understanding of the strengths and weaknesses of different observational and experimental designs, a lack of clarity in relation to disease definitions from the clinical, neuropathological, and molecular perspectives, inadequate characterization of brain aging in the human population, difficulties in translation between laboratory-based and population-based evidence bases, and a lack of communication between different sections of the dementia research community. Population studies highlight complexity and predict that therapeutic approaches based on single disease features will not be successful. Better characterization of brain aging in the human population is urgently required to select biomarkers and therapeutic targets that are meaningful to human disease. The generation of detailed and reliable evidence must be addressed before progress toward therapeutic interventions can be made.
Keywords
INTRODUCTION
Alzheimer’s disease (AD) is a complex, clinicopathologically defined dementia syndrome leading to cognitive impairment [1] and is thought to be the most common form of dementia in the older population. There is no accepted cause and the search for therapeutic interventions continues without much success [2]. Many clinical features of AD are shared with other dementing disorders and a clinical diagnosis of AD is uncertain [1, 3]. Clinical diagnosis is confirmed after death neuropathologically by the deposition of amyloid-β protein (Aβ) as plaques and cerebral amyloid angiopathy and the microtubule associated protein tau as neuritic plaques, neurites, and neurofibrillary tangles [4–6]. Familial AD (FAD), accounting for 1% of cases in populations [7], is defined by possession of a fully penetrant mutation in either the presenilins (PS) (PSEN1 and PSEN2) or the amyloid precursor protein (APP) [8], whereas sporadic AD (SAD), associated with a range of factors including increasing age [9], possession of the APOEɛ4 allele [10], vascular disease [11, 12], and metabolic syndromes [13], has no qualitative diagnostic feature.
Following the recent failures of amyloid-based randomized controlled trials to change the course of AD, there are growing calls for a re-evaluation of basic AD research. Epidemiology offers one approach to integrating the available evidence to guide dementia research strategy. Here we critically review relationships between evidence from population-based, clinicopathological studies of brain aging and a range of hypotheses from all areas of AD research to examine how the various hypotheses relate to evidence and how they may be combined to rationally select avenues of future investigation. We highlight significant areas of uncertainty in dementia research and suggest that population studies are essential to place evidence generated to date from different study designs including laboratory based, clinical trials and population studies, in a meaningful context.
AN EPIDEMIOLOGICAL, POPULATION-BASED APPROACH TO AD RESEARCH
Epidemiological approaches to AD research (Fig. 1) aim to make reliable generalizations about A) the understanding of disease processes and B) the efficacy of AD interventions [14]. We will examine these two areas separately. Observational population-based studies of aging with a brain donation program [15] are required for the characterization of AD in humans and are vital to hypothesis testing, biomarker validation, and identification of trends over time [16]. Population studies may include the whole defined population of a given geographical area in a given time period [17] or include a subset (a defined cohort) of participants that is representative of the entire population (‘population-based cohort studies’) [18]. These studies require a clear population provenance, with known sampling including consideration of non-participants, longitudinal attrition, and clearly described acquisition of data during life and after death. Such studies’ generalizability can only be known with such information. In addition, results need to take into account age, gender, ethnicity, and other key sociodemographic features which might influence their interpretation. There are only six such studies worldwide [15] (Table 1), including Cambridge City over 75 s Cohort (CC75C) [19, 20], the Cache County study [21, 22], the Cognitive Function and Ageing Study I (CFAS) [23, 24], the Honolulu Asia Aging study (HAAS) [25, 26], the Hisayama Study [27–30], and Vantaa 85+ [31]. While each population study is unique and depends on the population from which it is drawn, results arising from these studies are generalizable at the population level and where population studies agree, these results can be understood as reliable.
Epidemiological approach applied to AD (adapted from [14]). A) Understanding AD: genetic and other risk factors trigger disease (induction) and initiate causal disease pathways leading to disease onset (promotion) with an unknown latency in AD that could be decades. Clinical dementia status is the disease outcome (expression) in AD. Clinical evaluations including cognitive function and associated genetic factors, molecular biomarker and neuropathological distributions contribute to disease characterization. B) AD interventions: primary, secondary and tertiary interventions that prevent, postpone, shorten, ameliorate or cure build on the understanding of AD pathways and can involve preventative lifestyle interventions, symptomatic treatments, management of care and ultimately aim to find a cure.
Core funding status of population studies of aging with a brain donation program [15]
Population studies are an essential resource to understand how emerging hypotheses relate to human expression of disease and to rigorously assess biomarkers and therapeutic interventions. However, they are expensive and of the 6 studies included here, only two have continuing core funding. These are in Hawaii (USA) and Hisayama (Japan), with only the Hisayama study including both genders. The remaining four no longer have any core funding and have therefore lost their previous ability to respond efficiently to new research opportunities despite the decades spent in creating them. Many excellent population-based studies exist, e.g., [33], but only those noted above have brain donation associated with them.
Population studies aim to minimize selection bias; however, the lack of selection with reference to AD due to a lack of qualitative diagnostic features in the older population, can lead to lower than expected estimates of relationships due to information bias including problems with inaccurate measurement, missing data, and poor or changing diagnostic criteria [14, 34]. The impact of factors such as attrition, non-participation, and survivor bias must also be carefully assessed [35]. Age, the biggest risk factor in SAD, has been shown to alter the associations between pathological features and dementia status [9] and therefore findings from younger cohorts may not translate to older cohorts for all disease associations. Population studies are challenging in terms of participant recruitment, longitudinal follow up with stable protocols over decades, and the organization of tissue donation and storage. The engagement and contributions of the participants themselves is invaluable to the success of these studies.
A range of other study designs contribute to dementia research in humans. Cases and controls are often selected on the basis of presence of amyloid with MRI, with PiB positron emission tomography (PET), presence of biomarkers of tau or Aβ and after death, the neuropathological Aβ or tau deposition, leading potentially to quasi-circular experimental designs where the associations between disease and neuropathology are stronger than should be expected due to selection bias. Another common study design, autopsy series, may be biased toward younger age groups and rely on attendance at specialized clinics with brain donation infrastructure whereas older people with dementia may be cared for in the family home or a care home without regular or indeed any attendance at a memory clinic for diagnosis and care with implications for interpreting any results [36]. Community-based study designs are population derived but are not population representative. One example, the 90+ study, included all those accepted in a specific retirement community; however, because the criteria for acceptance into that community are unknown, the relationships between this study and the general population cannot be quantified [37]. All human-based experimental designs therefore have limitations. In combination, these approaches can be used iteratively to generate and then test hypotheses to clarify those areas of uncertainty that remain. All study types can contribute to the selection of valid therapeutic targets, with each study type having different biases, some more quantifiable than others, that impact on the precision of their findings and to which populations any findings may be applicable.
The key to understanding neurodegenerative disorders such as AD is to be able to translate evidence between genetic, molecular, neuropathological, and clinical domains in a clear and comprehensive way. Such work ideally requires integration of biomarkers measured during life (functional and structural imaging, blood, CSF, other biomaterials) and then examination of the brain after death. In vivo biomarkers still require validation through longitudinal follow up for clinical progression (or not) and what is found in the brain at death.
Many hypotheses to account for risk factors, causal processes, disease initiation, and disease progression in AD have been proposed (Tables 2–4) and can be broadly grouped by main area of focus within genetic, molecular, cellular, and physiological levels (Fig. 2). While this simplistic approach is useful to illustrate the breadth of areas contributing to the understanding of AD, nearly all areas and levels considered here involve multiple avenues of cross-talk and feedback via common cellular signaling pathways and some factors in Table 2 may be relevant in more than one area. These signaling pathways contribute synergistically to a dynamic and iterative homeostatic system spanning the entire body that underlies normal functions including cognition. This interconnectedness alone suggests that an approach based on considerations of complexity has great value [38, 39]. While we aim to consider many diverse areas to illustrate the issues, our approach cannot be completely comprehensive and some areas will necessarily be covered only briefly.
Genetic evidence from population studies listed in Table 1
Genetic evidence from population studies listed in Table 1
ACH, amyloid cascade hypothesis; AD, Alzheimer’s disease; AMA, AβPP matrix approach; Aβ, amyloid-β; AβPP, amyloid-β protein precursor; FAD, familial Alzheimer’s disease; NP, neuritic plaques; PS, presenilin; PSH, presenilin hypothesis; SAD, sporadic Alzheimer’s disease.
Evidence from population studies listed in Table 1 relating to hypotheses and features relating to cellular systems and functions
AD, Alzheimer’s disease; GFAP, glial fibrillary acidic protein; NFT, neurofibrillary tangles; NP, neuritic plaques, NSAID, nonsteroidal anti-inflammatory drugs.
Evidence from population studies listed in Table 1 relating to hypotheses and features relating to physiological systems and behavior
AD, Alzheimer’s disease; Aβ, amyloid-β; CAA, cerebral amyloid angiopathy; MCI, mild cognitive impairment; NP, neuritic plaques; TBI, traumatic brain injury; VaD, vascular dementia.
AD hypotheses grouped by main areas of focus. Yellow: ACH - main area of focus is β-cleavage and main outcomes investigated are levels of Aβ; orange: PSH – main area of focus is γ-cleavage and main outcomes investigated are levels of Aβ overlap with ACH indicated by area of light orange; white: AMA - main area of focus is whole AβPP cleavage system main outcomes would include all proteolytic fragments including Aβ; light blue: various cellular system and functions – main outcomes depend on system being investigated and include levels of Aβ; dark blue: Various physiological and behavioral systems - main outcomes depend on system being investigated and include levels of Aβ.
AβPP and FAD related
Several hypotheses relate to understanding the genetic and molecular evidence associated with dementia (Table 2). While the amyloid cascade hypothesis (ACH) focusing on the contributions of Aβ to AD has held a dominant position for decades, it has never been fully accepted and concerns remain [2, 280]. Alternative hypotheses focused on understanding molecular and genetic evidence include the presenilin hypothesis (PSH) [43, 44], which emphasizes loss or altered γ-secretase function and the amyloid-β protein precursor (AβPP) matrix approach (AMA) [48–52], which considers the dynamic behavior of the entire AβPP proteolytic system as a synergistic whole, have not been investigated in the same degree of detail as the ACH.
Understanding genetic evidence from FAD is vital as mutation in APP or PSEN1 and PSEN2 is a qualitative diagnostic feature. Rather than highlighting a fundamental disease process involving Aβ as suggested by the ACH, the genetic evidence can also be interpreted from the perspective of the AMA as highlighting various interactions and pathways, supporting the idea of AD as a syndrome of different but related pathways, not all of which may be relevant to SAD. Differences in levels of Aβ 40 and Aβ 42 [281] and differences in the AβPP beta carboxy terminal fragment [46] between PSEN associated FAD and SAD support this multiple pathways approach. These differences make a definition of AD at the molecular level difficult.
The mutations could be investigated from a population perspective, where possession of a particular mutation defines a distinct population which is rigorously characterized in terms of all the proteolytic fragments arising from the AβPP proteolytic system and careful analysis of how the distributions of proteolytic fragments relate to disease features such as age at onset, specific clinical features such as seizure, etc., and neuropathological characterization. Families with these mutations represent a unique and invaluable resource to investigate this complex proteolytic system as the various mutations represent natural knock in or complete/partial knock out models that are more easily translated to human disease than laboratory-based, reductionist investigations. Detailed investigations are required to identify which disease features and pathways are shared in FAD and SAD and which are unique to specific mutations.
Evidence from population studies relating to neuropathological Aβ deposition does not illustrate a straightforward correspondence that would unquestioningly support the ACH and instead finds that the relationships between Aβ deposition specifically, AD-related pathology in general, and dementia are complex [13, 42]. This evidence fundamentally questions the relevance of the ACH to the understanding of AD in the older population where most dementia occurs.
Cellular systems and functions
Hypotheses relating to the contributions at the level of cellular systems and functions (Table 3) such as Ca2 + regulation [79, 80], neurotransmitters [82–89], cholesterol homeostasis [53–56], mitochondrial functions [106–113], oxidative stress [117–121], immune system [126–128], senescence pathways [50, 146–149], synaptic plasticity [156–158], metal ion homeostasis [160, 161], and the cell cycle [164–166] are inter-related by multiple intra- and extracellular signaling pathways. Different cell types may have different organizations of signaling cascades and may express different arrays of receptors so that neuronal or glial subtypes, may not respond the same way to the same stimulus and further, homeostatic responses over time may depend on the integration of multiple stimuli and regulation by complex feedback pathways.
Evidence from pathological, epidemiological, and genome wide association studies implicate a wide range of cellular processes in AD. However, target identification has not been straightforward. Here we illustrate challenges faced by the AD research community using cholesterol homeostasis in AD [53–56, 105] as an example. In the population, the APOEɛ4 allele, involved in cholesterol transport, has long been recognized as a significant risk factor for AD [57–63], is associated with elevated cholesterol [68] and with AD type pathologies [59, 64] but not with vascular dementia [60, 64]. Abnormal lipid metabolism as defined by high levels of total cholesterol, triglycerides, high-density lipoprotein cholesterol and low-density lipoprotein cholesterol was associated with neuritic plaques but not neurofibrillary tangles [98], elevated late life HDL cholesterol was associated with both neuritic plaques and neurofibrillary tangles [102] and development of dementia may be marked by reduced serum cholesterol levels [104] supporting an etiological role for cholesterol homeostasis in AD. However, translating this approach into meaningful interventions to prevent, delay or control disease progression has not yet been successful. Statins were found to delay functional decline in one population study [99] but not in another [100] and the effect was not supported in systematic review [101]. The complexity and interconnectedness of the immune and cholesterol systems as illustrated by the multiple roles of ApoE in both [105, 282] and the additional roles of cholesterol in the cardiovascular system with implications for the development of dementia, listed in Table 4, suggest that we need to clarify our understanding of the relationships between these systems and AD progression before therapeutic targets aimed at cholesterol homeostasis can be identified with any certainty.
Physiological systems and behavior
As with hypotheses relating to cellular systems and functions, those relating to whole physiological systems and behavior listed in Table 4, such as the vascular system [179–184], diabetes, infection [129, 130], stressful life events [283], cognitive [249, 284] and metabolic [269–271] reserve are also connected by multiple pathways and additionally may be affected by human lifelong experience [285], wider genetic background and environmental factors [286, 287]. Population studies show that general health relates to cognition [265–268], comorbidity is more serious in those with dementia [256], sociological/economic factors are important [267, 289], that dementia incidence and prevalence estimates change over time [30, 290] and differ between populations [29, 291] and by sex [155, 293].
Further, the prevalence of cognitive and functional impairment may be more common in women [150, 175] and some studies report cognitive decline is faster in women than men [293]. Functional decline is associated more with stroke in men and AD in women [177] and responses to dementia medication may be different in women compared to men [176]. This suggests that gender differences may represent different disease pathways in men and women that further complicate the search for therapeutic intervention and the design of randomized controlled trials.
All the above taken together suggests that AD is a complex disorder relating to the whole person and the context in which we live and that focus on one particular part requires an understanding of the wider context.
Integrating the available evidence
In order to fully understand progress so far and future directions in dementia research, it is essential that evidence is translatable between clinical studies in humans, neuropathological diagnostics, and molecular investigations. This evidence must be reliable and assumptions must be transparent and testable if we are to be able to tease apart the complexities.
We have previously suggested that the AMA [48–52] (Fig. 3) provides a flexible framework with which different areas of dementia research can be inter-related and understood. Rather than focusing on one small part of the AβPP proteolytic system as seen in the ACH with Aβ, the AMA focuses on the complexity of the AβPP proteolytic system as a dynamic whole and emphasizes the contributions from wider cellular systems that affect the balance between the AβPP cleavage pathways via regulation of α-, β-, γ-, and other cleavages. Further, the AMA suggests that the AβPP proteolytic system feeds back to these wider cellular systems via the ratios of all the proteolytic fragments leading to a dynamic cyclic system capable of coordinating cellular responses. The fragments interact synergistically with wider cellular systems in agonistic and antagonistic manners both intra and extracellularly. This approach to understating the AβPP proteolytic system is compatible with recent work showing that cellular communication relies on combinations of small binding proteins, such as Aβ- and P3-type peptides, and receptor expression rather than absolute levels of these proteins [294]. In this respect, the AMA has the potential to integrate current evidence relating to various homeostatic systems such as cholesterol, Ca2 +, immune signaling, cell cycle, senescence, oxidative stress, etc., and the roles of AβPP proteolytic fragments in a way that better represents cellular functions. The AMA suggests that FAD associated mutations affect this homeostatic balance in ways particular to each mutation.
The AβPP matrix approach. Solid black lines represent cleavage pathways; dotted lines represent synergistic interactions of full length AβPP, sAβPPα, and sAβPPβ (top) and P3 and Aβ (bottom). Solid grey lines represent complex synergistic homeostatic interactions between wider cellular systems and the AβPP proteolytic system. Other cleavages, e.g., BACE2 leading to sAβPPβ’ and Aβ’, general catabolism of all fragments and caspase cleavage not shown.
In contrast to the ACH, the AMA suggests that while Aβ has a role in disease, it is only one small part and its expression can either drive disease or be driven by other disease related factors depending on the exact disease context. The AMA is compatible with the PSH if the PSH widened consideration to all fragments released from γ-cleavage and evidence exists to suggest that the production of P3 is affected similarly to Aβ by PSEN1 mutation [295]. The AMA is also compatible with other detailed hypotheses listed in Table 2 as it allows various factors to impact directly on AβPP cleavage, meaning that perturbations in wider cellular systems can both drive and be driven by disease pathways. As such, the AMA is a framework that allows each detailed hypothesis to be placed in relation to others in a synergistic way to see where factors converge or conflict and so allows a more flexible experimental approach to identify therapeutic targets.
The hypotheses listed in Tables 2–4 form a network from which therapeutic targets can be identified and interventions can be designed that prevent, postpone, shorten, cure, or reduce the severity of impairment in AD (Fig. 1). It is not currently possible to tie the various hypotheses listed to pathological processes with the detail required for therapeutic intervention with certainty. The data are confounded [296] and for some hypotheses, data from population representative cohorts are inconclusive or missing. The rational assessment of which features of disease merit more detailed investigation is difficult with the current limited evidence and there are many problems that undermine current understanding in dementia research.
Problems with current dementia research strategy
Within the older population, it is not clear whether we are dealing with one generally applicable or multiple sub-groups of pathways to disease. If we consider the evidence of cellular and physiological systems in the population, we understand that not everyone with dementia will share specific features, e.g., specific neuropathologies or in life factors such as possession of ApoE ɛ4 alleles, gender, diabetes, or hypertension, but that each factor has the potential to exert a partial pressure to modify disease expression. We are not certain that those with diabetes or contributions from other factors such as vascular disease [297] or aggregation of the TAR-DNA binding protein 43 (TDP-43) [154] will necessarily share the same therapeutic targets with those possessing an AD-associated mutation. In addition to differences in the contributions of factors such as inflammation and diabetes, gender differences suggest that at the population level, AD is poorly defined.
Progress in dementia research requires that findings are replicable within and translatable between different experimental systems. However, this presents problems for such a complex disorder. With no qualitative diagnostic feature, the diagnosis of SAD in those with dementia depends on cut off points along continua of features, such as neuropathological variables, biomarkers [298, 299], and 11C-PIB PET [300] that have yet to be validated and do not always agree [301, 302]. There is significant overlap in these features between those with and without dementia [20, 170] so that the selection of cases and controls in SAD for randomized controlled trials or hypothesis testing is uncertain. Within populations, there are individuals with inappropriately high or low burdens of pathology in relation to their clinical dementia status; in CFAS, 25% of respondents were neuropathologically misdiagnosed when assessed blind to clinical status [24, 199]. This evidence suggests that while they are associated with dementia, neuropathological features alone do not clearly define AD.
The continua of diagnostic features raise concerns relating to the classification and definition of dementia and AD in the older population. Where do we place cut-off points to capture dementia diagnoses as accurately as possible and will these be the same for each study? We cannot know whether those who died with high burdens of pathology but no dementia, defined as prodromal AD, would have developed dementia if they had lived longer. Those diagnosed clinically with AD-type dementia but with no or insufficient pathology for a neuropathological diagnosis of dementia type, suggest that further pathway(s) relating to dementia remain to be found that could potentially contribute to the lack of correspondence between clinical dementia and neuropathological diagnosis in the population. Is the definition of AD in the context of laboratory-based mechanistic studies using animals or cell culture, often operationally reduced to levels of Aβ, applicable to human disease? Is there a reliable molecular definition of AD that is transferable between different experimental approaches or even the different disease categories, FAD and SAD, in humans? To what extent does poor definition of AD in the various experimental contexts contribute to lack of progress— are we investigating the same AD in all approaches? These issues relating to defining AD are fundamental to dementia research and have been raised before [303] but are as yet unanswered.
Dementia in the older old is often mixed with contributions from a range of pathologies contributing to dementia [20, 304] and MCI [203, 297] and the correspondence between clinical diagnosis and neuropathological diagnosis blind to clinical status is not strong [20]. In CC75C, 85% of those with and 76% of those without dementia had sufficient AD-type neuropathology for a diagnosis of AD when assessed blind to clinical status. Multiple pathologies often contribute measures of the overall burden of dementia [20, 92] making it difficult to assign causal roles to specific pathologies with certainty. Is the current strategy of targeting therapeutic interventions at single disease features appropriate?
Despite decades of research, Aβ-related pathologies have yet to be fully characterized in the human population, e.g., Aβ deposition as different plaque types [305], different sequence lengths, aggregation states, and solubilities, and their context within the wider AβPP proteolytic system have not been adequately investigated. We do not know whether pathology is a proxy for other as yet hidden processes or whether pathology is inherently neurotoxic or a mix of both. Similar issues may also apply to other neuropathological features related to protein aggregation and deposition, e.g., tau [20, 31] and TDP-43 [154].
No study has yet measured the contributions of all the AβPP proteolytic fragments so there is a degree of confounding when assigning particular features of disease to any one proteolytic fragment [49, 52]. Since levels of AβPP are rate limiting, any proposed gain of function in one cleavage pathway necessarily leads to loss of function in another in this complex proteolytic system. These contributions will be confounding unless they are controlled for in experimental design. At the level of basic science, confounding arising from cross reactivities of commonly used anti Aβ antibodies in human neuropathological diagnostics and research [296] requires urgent clarification. We need to know which specific peptide sequences released from γ-cleavage are present in amyloid deposits and CSF and how much of each specific aggregation state (monomers, dimers, oligomers, and fibrils) for each specific sequence length is present— detailed information which is entirely missing from the literature base. Definitions of Aβ in practice may not be the same between different research approaches and results from studies using different anti Aβ antibodies are not directly translatable.
Given the complexity of the AβPP system and the uncertainty surrounding anti-Aβ antibody cross reactivities, it is time to address this lack of understanding and accurately describe the AβPP proteolytic system as a synergistic whole in humans. However, the measurement of a dynamic and iteratively changing system leads to a paradox of absolute measurement at one time (cross-section) versus measurement of flow through a pathway (longitudinal). Can a measurement at one point in time, as represented by MRI, sampling biological fluids for biomarkers or examining the brain after death, adequately describe a dynamic system changing in response to multiple perturbations over various time scales, e.g., diurnal variation [306, 307]? Understanding what biomarkers or neuropathological assessments actually represent remains to be fully addressed in AD. We should not be assuming that they are directly neurotoxic and represent therapeutic targets without understanding the complex human context in which they exist. We do not yet have the techniques to non-invasively generate the evidence required to understand dementia pathways in humans. How should we reduce the complexity of human cognitive function to generate laboratory-based models that can be used to dissect the complex processes associated with cognition and its failure and how do we test whether any such reductions are applicable?
The lack of full characterization of dementia in human populations impacts on laboratory-based, between-species comparisons, e.g., given that the role and the organization of G-protein coupled receptor signaling in mice and humans differs in pancreatic islets cells [308] and up to 90% of GPRs may be expressed in the brain [309], we can ask whether G-proteins in the brain have equivalent roles and organization between species. Other differences are suggested such as the role of PS1 in human oligodendrocytes and myelination that is absent in the mouse [310]. Between species differences lead to difficulties and potential failures when directly applying results from animal research to humans. Better characterization of both animal models and human populations will lead to better experimental models, more refined therapeutic target identification, and enable a more detailed understanding of how animal research can be best translated to humans.
Standardization of methodological issues relating to tools, experimental design, scoring and measurement protocols, and reporting of results is essential in order to rationally interpret findings from various experimental approaches in a wider context. While generalizable and qualitative trends can be identified from population studies, different methods to assess dementia status, different neuropathological protocols including the use of different antibodies, different diagnostic cut-offs and different methods of analysis make detailed comparison between studies difficult.
FUTURE DIRECTIONS
We suggest that the current evidence base is too narrow and it is not possible to identify therapeutic targets that have good chances of success to change the course of disease in humans. Several issues undermine a clear research strategy in the immediate future and require clarification. Agreed definitions of AD and Aβ that are transferrable between clinical, neuropathological, and molecular evidence bases are urgently required. There are currently few fully accepted, standardized measures and reporting formats that allow direct comparisons between studies. While qualitative comparisons are valuable, standardization to allow quantitative comparisons, especially for molecular factors, such as specific Aβ-type peptides, is required. We do not have the detailed evidence required to directly translate molecular findings between laboratory-based mechanistic studies and disease in the human population. Several stages of translation need to be developed before this can happen reliably. We need to better characterize the relationships between clinical dementia, biomarkers of AD, neuropathological diagnosis, and specific molecular features in the human population and explore how these relate to laboratory-based experimental models. Better characterization of dementia related factors in the population will increase understanding of how many AD related disease pathways are possible and which pathways share therapeutic targets. There may be groups of disease pathways that can be better defined with better characterization of human disease. Hypotheses guide both experimental design and interpretation of results. The ACH effectively reduces the output of the AβPP proteolytic system to measures of Aβ. This limits considerations of complexity. We should address complexity by characterizing the AβPP proteolytic system in a systematic manner and interpret results within more flexible frameworks that reflect the complex cellular milieu.
Future dementia research strategy depends on clarifying our understanding of current evidence and identifying sources of uncertainty to be corrected. Without such detailed assessments, identifying therapeutic targets and drug discovery strategies may not have the rational basis required.
SUMMARY
AD research has been dominated by the ACH for decades with little advance in our understanding of the role of the AβPP proteolytic system as a whole in disease initiation and progression due to confounding by molecular complexity, misunderstanding of antibody reactivities, and biased experimental designs. The neglect of the PSH, relating to the contributions from γ-secretase, and the AMA, relating to the dynamic balance between all cleavage pathways and products of the AβPP proteolytic system, can be understood as a significant hypothesis bias. The genetic and neuropathological evidence emphasizes the importance of this proteolytic system in AD, and it is now time to re-assess the evidence so far to clarify our understanding.
Population studies are an unrivaled resource to better characterize the myriad factors associated with dementia and be able to translate these findings to better diagnostic protocols that are urgently needed. They also highlight complexity and predict that single therapeutic approaches based on isolated disease features will not be successful.
The importance of population studies for disease characterization, hypothesis testing, and disease marker validation has not been fully acknowledged by the wider AD research community and their essential contributions to the development of efficient research strategies are neglected. Better characterizations of brain aging in the human population will lead to a more rational selection of AD therapeutic targets that are meaningful to human disease. Biomarker validation in the human population will lead to a better understanding their relationship with disease and refine how they can be applied clinically. Research at the population level is significantly hindered by lack of core funding and without it, unquantified bias in experimental designs may mislead the research community. More flexible molecular models such as the AβPP matrix approach may contribute greatly to integrating and understanding evidence from very diverse fields of dementia research.
