A short perspective on the long road to effective treatments for Alzheimer's disease

Abstract

Globally, there are approximately 47 million people living with dementia, and about two thirds of those have Alzheimer's disease (AD). Age is the single biggest risk factor for the vast majority of sporadic AD cases, and because the world's population is aging, the number of people living with AD is set to rise dramatically over the coming decades. There are currently no disease‐modifying treatments for AD, and the few symptomatic agents available have limited impact on the disease. Perhaps surprisingly, there is relatively little activity in the AD research and development field compared with other diseases with a high mortality burden, such as cancer. There is enormous economic incentive to discover and develop the first disease‐modifying treatment, but previous failure has significantly reduced further industrial investment in this field. The short review looks at the historical path trodden to develop treatments and reflects on the journey down the road to truly effective treatments for people living with AD.

Linked Articles

This article is part of a themed section on Therapeutics for Dementia and Alzheimer's Disease: New Directions for Precision Medicine. To view the other articles in this section visit http://onlinelibrary.wiley.com/doi/10.1111/bph.v176.18/issuetoc

Abbreviations

AD

Alzheimer's disease

ApoE

apolipoprotein E

APP

amyloid precursor protein

Aβ

amyloid peptide

Aβ_x

amyloid peptide of a specific length (i.e., x amino acids)

BACE 1/2

β‐secretase 1/2

FAD

familial Alzheimer's disease

1. INTRODUCTION

Alzheimer's disease (AD) is the most common form of dementia accounting for approximately two thirds of all cases, with vascular dementia, fronto‐temporal dementia, dementia with Lewy bodies, and a host of rarer dementias making up the rest. Different countries have different ways of recording disease and its impact on mortality and morbidity, and AD is not always broken out separately. The World Health Organization lists dementia as the fifth largest cause of death across the globe (World Health Organization, 2018). Dementia has recently become the single biggest cause of death in the United Kingdom responsible for approximately 13% of all deaths (UK Office of National Statistics, 2018). There are approximately 47 million people globally living with dementia (World Alzheimer Report, 2015). Given that age is the single biggest risk factor for developing most forms of dementia, including AD, and the world's population is aging, that number is set to rise to approximately 131 million by 2050 (World Alzheimer Report, 2015). AD, like most forms of dementia, is progressive, slowly robbing the patient of their cognitive abilities, which then impacts on daily activities and their ability to live independently. Not surprisingly, this places a huge burden on health and social care, with the estimated global cost hitting $1 trillion in 2018 and set to rise to $2 trillion by 2030 (World Alzheimer Report, 2015). Unlike many major diseases, the direct health care cost makes up a relatively small proportion of these costs, social care and unpaid care (e.g., a family member giving up work to look after someone) being a much larger component (Prince, Comas‐Herrera, Knapp, Guerchet, & Karagiannidou, 2016). AD and dementia are frequently referred to as the most pressing health care challenge of our time, and the numbers above demonstrate why. The large, and growing, patient population coupled with high impact on quality of life for patients and their families clearly makes the treatment of dementia an attractive opportunity for pharmaceutical companies. Not surprisingly, most drug discovery efforts have been focussed on AD because the size of patient population and this article will discuss these primarily. Therefore, it is perhaps surprising that there is such a dearth of approved medications and comparatively little R&D activity in this area compared with cancer or heart disease. Figure 1 indicates that scientific productivity, as measured by number of publications, is 28‐fold and eightfold lower for AD, compared with cancer or heart disease. Whilst publications alone do not lead to new therapies, they do act as a good surrogate of the level of investment and research activity for a given disease. The dearth of AD literature compared with cancer or heart disease is likely why these latter diseases have many effective treatments and AD does not. A look back at the efforts, activities, and subsequent failures in AD research is informative for where the field stands today and how to achieve success in the future.

Figure 1.

Comparison of the number of publications in different disease areas. The graph shows the total number of publications in PubMed in 3‐year time bins using the search terms “cancer,” “heart disease,” “HIV,” “dementia,” and “Alzheimer.” All diseases show that the number of publications has consistently increased over the last 30 years, with cancer showing the steepest rise. It is also clear that the volume of research activity is far higher in other diseases that have effective drug therapies than it is for Alzheimer's disease (AD) or dementia. Whilst the number of publications is not a measure of the amount of drug discovery or development activity for a given disease, it does act as a surrogate for how much is understood about a given disease, which is very important for successful drug discovery

2. EXISTING TREATMENTS FOR AD

In common medical practice, clinical AD diagnosis has typically been made based on symptoms, rather than via pathophysiological measurements such as MRI, PET, or biomarker analyses of blood or CSF samples. It should be noted that the situation is rapidly changing with increased use of biomarker assessment. Availability of new treatments will likely accelerate that change further. Biomarker studies and amyloid PET scans have shown that disease pathology is evident many years before the onset of overt cognitive decline (Jack et al., 2010), such that AD is now usually described as having preclinical (i.e., presymptomatic), prodromal, mild, moderate, and severe stages. Definitions vary in the literature, but preclinical AD is typically defined as someone with positive AD biomarkers but not yet showing cognitive impairment. This progresses to the prodromal stage once mild cognitive impairment is observed, and this may progress on to mild AD and subsequently to moderate and severe AD followed by death (Amieva et al., 2008; Ballard et al., 2011; Weintraub et al., 2018). The annual conversion rate from mild cognitive impairment to mild AD varies considerably across studies because of differences in definitions of clinical symptoms and use of biomarker data, but meta‐analyses suggest that it is approximately 10% (Ward, Tardiff, Dye, & Arrighi, 2013).

The pathological hallmarks of AD were first described by Alois Alzheimer in 1906 consisting of extracellular plaques, intracellular tangles, and widespread neurodegeneration (Figure 2), although it was not until many decades later that amyloid and tau were identified as the main constituents of these aggregates (Brion, Passareiro, Nunez, & Flament‐Durand, 1985; Glenner, Wong, Quaranta, & Eanes, 1984). In the 1970s, the relative vulnerability of the septal cholinergic system was first identified (Davies & Maloney, 1976; Whitehouse et al., 1982), and this led to the first effective drugs for treating the cognitive symptoms of AD that were launched around the turn of the last century (Francis, Palmer, Snape, & Wilcock, 1999). Donepezil, rivastigmine, and galantamine are all AChE inhibitors approved for the symptomatic treatment of mild‐to‐moderate AD patients (Birks, 2006). AChE inhibitors prevent the breakdown of synaptically released ACh and thereby prolong postsynaptic activation of nicotinic and muscarinic receptors. It is thought that the therapeutic benefit largely derives from potentiation of septo‐hippocampal cholinergic activity improving attention and thus enabling better cognitive performance (Muir, Everitt, & Robbins, 1994; Sahakian & Coull, 1993). However, the cholinergic system is also involved in many other, notably autonomic, functions, which is the reason these drugs have a range of adverse effects (e.g., nausea, diarrhoea, and vomiting). The clinical dose of these drugs has been empirically selected to balance their tolerability against therapeutic benefit (Gauthier, 2001; Thompson, Lanctôt, & Herrmann, 2004), although it is likely that greater efficacy could be achieved if tolerability could be improved (Felder et al., 2018). The only other approved symptomatic treatment for AD is memantine, which is a non‐competitive NMDA glutamate receptor blocker, although its mechanism of action was only discovered during its clinical development, which was driven by empirical observation of symptomatic benefit (Lipton, 2006; McShane, Areosa Sastre, & Minakaran, 2006). None of these drugs have any significant impact on the progression of AD, instead just offering modest amelioration of some symptoms.

Figure 2.

The long road to anti‐amyloid therapies. Starting with the first description of Alzheimer's disease (AD) in 1906 through to ongoing clinical trials with amyloid‐lowering therapies, this timeline indicates some of the key discoveries and clinical trial outcomes. Relatively little progress was made through the 20th century until genetic studies clearly linked amyloid processing to disease causation. Since the 1990s, therapeutic research and development has explored several different approaches to lowering amyloid in the brain with the aim of providing therapeutic benefit in the form of slowing, or stopping, disease progression. Most of these approaches have been stopped due to either safety issues or lack of robust efficacy. Several anti‐amyloid antibodies are in late stage clinical trials and will hopefully prove to be effective in patients. FAD: familial Alzheimer's disease; MCI: mild cognitive impairment; APP: amyloid precursor protein

3. THE AMYLOID CASCADE HYPOTHESIS

The search for disease‐modifying treatments for AD is the primary goal of most therapeutic research efforts, and it was a field that grew rapidly in the 1990s but then began to wane over the next 10 years (Figure 2). Two key factors drove the large‐scale pharmaceutical investment; the first was economic and the second scientific advances. The 1990s saw enormous economic success of neuroscience drugs such as the atypical antipsychotics for schizophrenia (e.g., olanzapine and risperidone), selective 5‐HT (serotonin) reuptake inhibitors for depression (e.g., fluoxetine and paroxetine), α2δ ligands for neuropathic pain (gabapentin and pregabalin), and a range of new mechanisms to treat epilepsy (e.g., lamotrigine and topiramate). These successes instilled the (not unreasonable) belief in senior pharmaceutical executives that psychiatric and neurological conditions were lucrative opportunities for R&D investment given the large patient populations and significant unmet patient need, along with advancing scientific understanding of the molecular underpinnings of these diseases. AD was top of the neuroscience priority list for many companies. The key scientific advance that increased confidence that effective disease‐modifying drugs could be developed was the identification of causative genes for human disease. Discovery of the genes responsible for autosomal dominant diseases for the first time directly linked potential drug targets with human disease bypassing the need for less direct linkage provided by animal models, functional changes, and empirical observations. At that time, it appeared that genetics had paved a clear path to effective medicines.

In 1991, the first Alzheimer's autosomal dominant mutation (V717I) was discovered in a family in London in the gene that codes for the amyloid precursor protein (APP; Goate et al., 1991). It was quickly followed by a whole series of other disease‐causative mutations in this gene (Tcw & Goate, 2017). Biochemical studies determined that APP was preferentially cleaved by enzymes α‐secretase and β‐secretase (more typically referred to as BACE) as well as the γ‐secretase complex, to produce peptide fragments of varying sequences (De Strooper, Vassar, & Golde, 2010), the most important of which was β‐amyloid (Aβ), which was already known to be the primary constituent of amyloid plaques (Glenner et al., 1984). The cleavage of APP into various fragments with a range of biological functions is complex and has been reviewed in detail elsewhere (Gralle & Ferreira, 2007; Ludewig & Korte, 2017). Sequential BACE and γ‐secretase cleavage are required to produce Aβ peptides (De Strooper et al., 2010), whereas α‐secretase cleaves in the middle of the Aβ sequence and yields nonamyloidogenic fragments (Jorissen et al., 2010; Kuhn et al., 2010). Intriguingly, nearly all of the disease‐causing mutations on APP cluster around the BACE and γ‐secretase cleavage sites, implicating inappropriate processing as the pathophysiological mechanism in AD. Two further genes linked to familial AD (FAD), presenilin 1 and 2 (Bertram, Lill, & Tanzi, 2010; Lendon, Ashall, & Goate, 1997; Sherrington et al., 1995), were a few years later shown to be the critical aspartyl protease components in the γ‐secretase complex (De Strooper et al., 2010; Sisodia & St George‐Hyslop, 2002). FAD cases make up <1% of all AD cases, but the clinical symptoms, disease progression, and biochemical and neuropathological changes are all remarkably similar to typical late‐onset sporadic AD, except that symptom onset is usually in the third to fifth decades of life rather than in old age (Ryman et al., 2014).

The amyloid cascade hypothesis of AD was proposed by Hardy in the early 1990s, in which build‐up of longer forms of Aβ, particularly the aggregation‐prone Aβ₄₂, led sequentially to the formation of neurotoxic oligomers, insoluble amyloid fibrils, and ultimately amyloid plaques (Hardy & Allsop, 1991; Hardy & Selkoe, 2002). Most FAD mutations shift the ratio of Aβ peptides towards the longer, more aggregation‐prone forms. In sporadic forms, it was not clear if there was over production of Aβ generally, an increase in the ratio Aβ₄₂/Aβ₄₀, or alternatively, insufficient clearance from the brain of Aβ₄₂, any of which could lead to accumulation and in turn aggregation. The critical involvement of BACE and γ‐secretase in disease pathology presented two obvious drug targets, especially as enzyme inhibitors are a well‐precedented class of drugs (Vassar et al., 2014).

γ‐Secretase was the target most heavily pursued initially, largely because assays of γ‐secretase activity existed before the complete make‐up of the enzymatic complex was elucidated and presenilins identified as the catalytic subunit (Sisodia & St George‐Hyslop, 2002). γ‐Secretase inhibitors, such as semagacestat from Eli Lilly, were relatively quickly identified and advanced into large clinical trials on AD patients. Unfortunately, γ‐secretase has around 40 cellular substrates, which resulted in significant on‐target toxicity. Notch, a key component of the Wnt signalling pathway, is a substrate of γ‐secretase (Geling, Steiner, Willem, Bally‐Cuif, & Haass, 2002; Micchelli et al., 2003), and blocking Notch cleavage is likely the reason increases in skin cancers and infections were observed in these clinical trials (Chávez‐Gutiérrez et al., 2012; Doody et al., 2013). These adverse effects, plus the fact that the semagacestat‐treated patients actually had worse cognitive decline than those in the placebo group, resulted in its discontinuation in 2011 (Doody et al., 2013). Several companies pursued other γ‐secretase inhibitors into the clinic and experienced similar challenges, eventually resulting in them all being stopped. Two alternative approaches to γ‐secretase have been proposed and to some extent persued: (a) γ‐secretase modulators (Bursavich, Harrison, & Blain, 2016; Hall & Patel, 2014) that reduce APP cleavage but have reduced effect on the cleavage of other substrates and (b) γ‐secretase stabilizers that shift cleavage to shorter, less toxic Aβ peptides such as Aβ₃₈ (Szaruga et al., 2017). γ‐Secretase modulators have reached clinical trials, but clinical benefit has not yet been demonstrated, whereas γ‐secretase stabilizers are still at the preclinical discovery stage.

BACE inhibitor development has also been an area of intense focus once BACE was cloned by five independent groups in 1999 (Hussain et al., 1999; Lin et al., 2000; Sinha et al., 1999; Vassar et al., 1999; Yan et al., 1999). There are two BACE isoforms (BACE1 and BACE2), which initially appeared to have fewer alternative substrates than γ‐secretase. It was therefore reasoned that BACE inhibitors may be safer than γ‐secretase inhibitors; however, subsequent investigation demonstrated a range of important substrates for both BACE1 and BACE2 (Dislich & Lichtenthaler, 2012). The active site of BACE is very open compared with other aspartyl proteases (Hong et al., 2000), which made it very difficult to find potent and selective inhibitors that also possessed drug‐like properties (Ghosh, Brindisi, & Tang, 2012; Vassar et al., 2014). Many of the early inhibitors were peptide mimetic molecules with poor absorption, stability, and brain or cell penetration making them unsuitable even for rodent in vivo studies let alone for clinical development (Ghosh et al., 2012). Concerted medicinal chemistry efforts eventually yielded appropriate molecules, and Merck Inc. led the way into large‐scale clinical trials with verubecestat (MK‐8931) in 2012 (Scott et al., 2016). BACE inhibitors have demonstrated an acceptable safety profile for large‐scale clinical trials in AD patients at doses that robustly reduce Aβ levels in plasma and CSF (Kennedy et al., 2016). Verubecestat has been tested in prodromal, mild, and moderate AD patient populations, but despite lowering CSF Aβ levels (Kennedy et al., 2016), no cognitive or functional benefit was observed in patients (Alzforum, 2018c; Egan et al., 2018). Since the negative Phase III results of verubecestat, other BACE inhibitor studies have also been terminated (Alzforum, 2018b,d).

The third major approach to therapeutically target Aβ directly is active or passive immunization. Encouraging efficacy was first shown by Schenk et al. (1999) at Elan Corporation using the PDAPP Alzheimer's mouse model. PDAPP mice overexpress a mutant form of human APP (APP V717F) and show age‐dependent deposition of amyloid plaques (Masliah et al., 1996). Active immunization of these mice with Aβ₄₂ peptide and QS‐21 adjuvant blocked the development of amyloid plaque pathology when given to young mice (6 weeks of age) and markedly reduced established pathology in 11‐month‐old mice (Schenk et al., 1999). Elan then progressed this agent, known as AN‐1792, into a clinical trial in mild‐to‐moderate AD patients. Unfortunately, a small number of patients developed serious brain inflammation that was later shown to be aseptic meningoencephalitis (Nicoll et al., 2003). Given the serious nature of this adverse event and no means of predicting/excluding patients at greatest risk, development of AN‐1792 was terminated in 2002 (Check, 2002). Only a subset of patients in the trial mounted a significant antibody response to the treatment, mostly directed to the N‐terminus portion of Aβ, and this was not correlated with those experiencing aseptic meningoencephalitis (Lee et al., 2005). Importantly, post‐mortem analysis of the patients who mounted a response showed that some clearance of amyloid plaques from their brains had occurred (Nicoll et al., 2003). However, there were conflicting results as to whether the patients experienced clinical benefit or not (Holmes et al., 2008; Vellas et al., 2009).

The mechanistic success of AN‐1792 (i.e., clearance of amyloid plaques in humans) encouraged the field to find similar approaches with an improved safety profile. Several companies developed passive immunotherapies where humanized antibodies targeting Aβ peptides were administered to patients. The major safety advantage to this approach is that the host immune system is not directly stimulated, thereby avoiding the inflammatory response seen with AN‐1792. Additionally, the epitope of the therapeutic antibody is defined so that therapies aimed at soluble monomers, oligomers, or fibrillar Aβ could be designed. The disadvantage of the passive immunotherapy response is that the antibodies administered will be cleared from the body necessitating regular dosing (usually every 4 weeks). Antibodies are also very poorly brain penetrant with only approximately 0.1% of the dose getting into the brain (Banks et al., 2002; Levites et al., 2006), resulting in a high peripheral load of antibody and consequently a higher dose needing to be given, which greatly increases the cost of manufacturing the therapy. In the decade following AN‐1792 termination, a number of Aβ‐targeting antibodies, including solanezumab (Eli Lilly), crenezumab (Roche), gantenerumab (Roche), and bapineuzumab (Johnson & Johnson/Pfizer), entered clinical trials and demonstrated an improved safety profile. Some instances of amyloid‐related imaging abnormalities associated with either microhaemorrhage or vasogenic oedema were observed in subsets of patients, but these have proved manageable for the running of large clinical trials aimed at establishing efficacy in AD patients (Ostrowitzki et al., 2012; Salloway et al., 2014; Sperling et al., 2012). Sadly, the therapeutic benefit in terms of rate of cognitive or functional decline has been minimal in prodromal, mild, and moderate patient AD populations for many of these antibodies (Cummings, Cohen, et al., 2018; Doody, Farlow, et al., 2014; Doody, Thomas, et al., 2014; Ostrowitzki et al., 2017; Siemers et al., 2016). Subgroup analyses, usually performed post hoc, have indicated some cognitive benefits, but these have failed to be robustly reproduced.

On a more positive note, two other amyloid antibodies, aducanumab and more recently BAN2401, have reported positive effects (Alzforum, 2018a; Sevigny et al., 2016). Aducanumab demonstrated dose‐dependent clearance of amyloid plaques, as determined by amyloid PET, in a 12‐month Phase Ib safety study with approximately 25–30 prodromal and mild AD patients. Impressively, it also showed largely dose‐dependent reductions in the Clinical Dementia Rating Scale sum of boxes after 12 months of dosing (Sevigny et al., 2016). Presentation of data from subsequent open‐label extension phases from this study shows that those benefits are sustained over 4 years of dosing. This is a small study, and therefore, replication of these findings in large Phase III trials is required to establish robust safety and efficacy data. Two Phase III studies completed enrolment in mid‐2018, and the outcomes are awaited. BAN2401 reported the results of an 856 patient dose‐ranging study in prodromal and mild AD (Alzforum, 2018a). Again, there was a dose‐dependent reduction in amyloid PET signal, but only the highest dose (10 mg·kg⁻¹ biweekly) showed a statistically significant improvement in cogntive decline. Interpretation of the results is complicated by the adaptive randomization design of the study and a change in regulatory safety guidance part way through the study resulting in an inbalance in the percentage of apolipoprotein E (ApoE) ε4 carriers in the different arms (Alzforum, 2018a). These results are encouraging because of the size of the study but will likely require replication in pivotal Phase III studies before approval. The reasons why aducanumab may be effective when several other antibodies have failed have not been determined.

4. Aβ‐LOWERING THERAPIES: HAVE WE REACHED A DEAD END?

The lack of reproducible success of amyloid‐targeting therapeutics to date is clearly disappointing for researchers, pharmaceutical companies, and most of all to patients who are still without an effective treatment to slow down or stop the course of their disease. However, does this mean that the amyloid hypothesis of AD is incorrect? Certainly, there are voices saying that it is, but is this really true? A careful look at the research data and clinical trial findings is required to determine how thoroughly the hypothesis has actually been tested. Three key questions need to be answered to understand how well we have tested the amyloid hypothesis. These are (a) have we been treating the right patients, (b) when do we need to treat with Aβ‐lowering therapies to be effective, and (c) how long do we need to treat for? Over the last 20 years of clinical studies in AD, we have learnt a huge amount about the disease course of AD, improved the tools available for studying the disease in patients (Blennow & Zetterberg, 2018; Pietrzak, Czarnecka, Mikiciuk‐Olasik, & Szymanski, 2018; Rice & Bisdas, 2017), and increased our understanding of which endpoints are the most informative (Aisen et al., 2017; Reiman et al., 2016). It is therefore much more accurate to say that these large clinical development programmes have been negative (i.e., failed to meet their primary endpoint), rather than failed (i.e., the trial design or execution was flawed such that the results are uninterpretable).

The question of whether we have been treating the right patients may seem a simple one to answer, but for dementias, it is actually far from trivial. In fact, in the earlier antibody trials where an amyloid PET scan was not available and therefore was not an entry criterion, it is estimated that up to a third of patients enrolled did not have elevated amyloid levels (Siemers et al., 2016; Vellas et al., 2013). It is critical to get the inclusion criteria for enrolment into a clinical trial correct such that only patients with a high amyloid load resulting in AD are studied in the trial. From a scientific point of view, these criteria need to be as stringent as possible to avoid recruiting inappropriate patients, as those not possessing an amyloid pathology will only add to the noise and potentially obscure a genuine positive effect. However, that stringency needs to be balanced against the practical reality that enough patients need to be identified to make the running of the trial feasible from both a time and cost perspective. The gold‐standard method of diagnosis is neuropathological confirmation of plaques and tangles post‐mortem, which is obviously not applicable. Cognitive assessment alone determines that the patient has cognitive deficits consistent with AD, but given the wide variation in symptoms for AD and other dementias such as vascular dementia and fronto‐temporal dementia (Pasquier, 1999), those findings alone are insufficient (Fox & Rossor, 1999). Assessment of plasma Aβ concentrations, whether Aβ₄₂ or the ratio of Aβ₄₂/Aβ₄₀, generally has a poor correlation to brain Aβ load (Beach, Monsell, Phillips, & Kukull, 2012), although recent methodological advances may be starting to overcome that issue (Nakamura et al., 2018; Ovod et al., 2017). CSF Aβ concentrations are much more predictive of the brain amyloid load (Clark et al., 2003; Galasko et al., 1998), but there are concerns over patient willingness to undergo a lumbar puncture, which has limited its usage. A key advance in biomarker tools for accurate diagnosis of AD patients was the development of amyloid PET ligands over the last 10 years (Chételat et al., 2013; Morris et al., 2016). Several agents such as florbetapir and flutemetamol are available for use in clinical trials and approved for diagnostic use in medical practice. These ligands bind to aggregated Aβ (Klunk et al., 2001) and allow the clear determination of elevated amyloid load. Their use in screening patients for AD trials is now very wide spread and ensures that patients are appropriate for amyloid‐lowering therapy trials. Trials run before these ligands were widely available recruited a high proportion of amyloid negative subjects (Siemers et al., 2016; Vellas et al., 2013), which was initially thought to be a contributory reason for the lack of robust clinical efficacy. However, a more recent solanezumab trial that included a positive amyloid PET scan as an inclusion criterion also did not show robust efficacy (Honig et al., 2018).

The second question of when patients need to be treated with amyloid‐lowering therapies to have a beneficial effect on disease is still somewhat of an open question because there are no approved therapies. Genetic findings from FAD mutations indicate that increased production of Aβ₄₂ is sufficient to cause AD (Tcw & Goate, 2017). Additionally, a protective mutation, A673T which is close to the BACE cleavage site, has also been discovered that reduces Aβ production by approximately 40% in cell culture models (Jonsson et al., 2012; Maloney et al., 2014) and protects human carriers from developing AD or cognitive decline (Jonsson et al., 2012). The protective genetic mutation is carried from birth, suggesting that in the worst‐case scenario, life‐long reduction in Aβ production is required for protection. On the flip‐side even in FAD patients, the disease is rarely diagnosed before the age of 30 (Ryman et al., 2014) and more usually in their 40s or 50s, suggesting that the brain is resilient for several decades before the onset of cognitive decline. The use of amyloid PET ligands allowed the longitudinal mapping of amyloid deposition and, in combination with other scanning techniques (Weiner et al., 2010), built a relatively clear picture of the sequence of events leading to cognitive decline and a clinical diagnosis of AD (Jack, Holtzman, & Holtzman, 2013; Jack et al., 2010). Amyloid levels start to rise at least 20 years prior and actually plateau about 10–15 years before the onset of overt clinical symptoms resulting in a diagnosis of AD (Schott & Petersen, 2015). Additional scanning techniques such as fluorodeoxyglucose uptake to look at metabolic activity (O'Brien et al., 2014), volumetric MRI to assess atrophy (Jack et al., 2010), microglial activation (Edison, Donat, & Sastre, 2018; Varley, Brooks, & Edison, 2015), and tau (Okamura et al., 2018) PET ligands have built up a picture of the sequence of events in the brain leading up to cognitive decline. These can be plotted as a series of sequential but overlapping curves first clearly summarized by Jack et al. (2010) and Jack et al. (2013): Amyloid build up occurs first, followed by metabolic decline, then tau accumulation leading to reductions in brain volume. Only at this relatively late stage in terms of brain pathology are obvious symptoms of cognitive decline observed and a clinical diagnosis of AD typically given. The precise role of amyloid in AD is yet to be fully determined, but accumulating evidence suggests that it may be more likely to act as a trigger for other pathologies and may be less important, or even irrelevant, later in the disease course (Karran, Mercken, & De Strooper, 2011).

Our understanding of this sequence of progressive brain pathology in advance of overt clinical symptoms points to the most likely reason that amyloid‐lowering therapies have not been beneficial in most trials run to date: They were administered too late (van Dyck, 2018). Amyloid load plateaus many years before a patient would be given a diagnosis of prodromal or mild AD, by which time amyloid may not be the primary driver of disease progression (Jack et al., 2010; Weiner et al., 2010). The ability of amyloid therapies to lower brain amyloid load, even to the point of being classified as amyloid negative by PET scanning and yet have little to no benefit on cognitive decline, would certainly be consistent with this hypothesis (Egan et al., 2018; Panza et al., 2014). Clinical intervention earlier in the disease course is the obvious next step for these therapies, and several studies in prodromal or preclinical disease are being conducted (Reiman et al., 2016; Sperling, Mormino, & Johnson, 2014). Recommended inclusion criteria for such studies include a positive amyloid PET scan, lack of significant cognitive impairment, and one or more risk factors such as family history or being a carrier of the ApoEε4 allele (Dubois et al., 2016). The primary outcome of these studies is a slowing of the rate of cognitive decline as measured by a composite cognitive score. The challenge with such studies is the length of time that they take to conduct and the number of patients required given the variability in rates of disease progression. Both of these factors make such trials prohibitively expensive, but thankfully, industrial and philanthropic funding is allowing a small number of such studies to be carried out. An alternative approach to test the early intervention hypothesis is selection of a much more defined patient population, that is, those with FAD mutations. These patients will definitely progress to AD, and we know that their APP/PS mutation is the driving cause of their disease. Additionally, they are often much more motivated to participate in clinical trials than healthy older individuals in the general population who are currently symptom free. Research networks of FAD carriers have been established, and amyloid‐lowering therapy preventive trials, such as DIAN‐TU, have been ongoing for the last few years (Mills et al., 2013; Sperling, Rentz, et al., 2014).

The third question of how long we need to treat for is also something of an unknown and likely to depend on the exact mechanism of the therapy. PET studies in mild‐to‐moderate AD patients using antiamyloid antibodies have shown that amyloid load can be reduced over a period of 6–18 months in a dose‐dependent manner (Ostrowitzki et al., 2017; Sevigny et al., 2016). The relatively slow clearance of already deposited amyloid is probably a combination of preventing further deposition plus removal of plaques via activation of microglial clearance mechanisms (Ostrowitzki et al., 2012; Sevigny et al., 2016). Verubecestat has likewise shown a gradual clearance of plaques over several months even though a reduction of CSF or plasma Aβ occurs much more quickly. Understanding the kinetics of amyloid plaque clearance is interesting from a scientific point of view, but evidence to date shows that it offers limited patient benefit. Instead, prevention of deposition in the first place may be a more relevant goal, which hopefully, the ongoing secondary prevention trials will tell us. The duration of treatment for sustained beneficial effect is likely to be life long as the longitudinal data suggest that amyloid deposition is probably the triggering event for AD rather than necessary throughout the disease (Karran et al., 2011). As such amyloid levels will need to be kept low for long periods to avoid the disease processes reinitiating. This in turn requires that therapies have a very good safety profile such that the benefit–risk is still favourable even in people who have no overt disease. Any ongoing safety monitoring requirements, such as regular MRI scans for signs of amyloid‐related imaging abnormalities, could be detrimental to the real‐world uptake of these therapies for long‐term secondary prevention.

5. ALTERNATIVE ROADS TO SUCCESS?

The disproportionate amount of academic research effort and industrial investment into the amyloid cascade hypothesis has probably slowed the field down in investigating alternative mechanisms and developing drugs for other targets (Figure 3). The most obvious alternative target is tau: Tau tangles are a diagnostic criterion for AD diagnosis, and their accumulation correlates much better with cognitive decline (Arriagada, Growdon, Hedley‐Whyte, & Hyman, 1992; Nelson et al., 2012; Tomlinson, Blessed, & Roth, 1970). Interestingly, disease‐causing mutations in tau are not linked to AD, rather to various other forms of neurodegenerative diseases such as fronto‐temporal dementia, progressive supranuclear palsy, Pick's disease, and corticobasal degeneration (Ghetti et al., 2015; Hutton et al., 1998; Spillantini et al., 1998). Whilst this may appear to weaken tau as a therapeutic target for AD, it does strengthen the case that aberrant tau and its aggregation into intracellular tangles are sufficient to drive neurodegeneration, albeit with varied clinical presentation. Additionally, tau is necessary for Aβ‐induced neurotoxicity (Rapoport, Dawson, Binder, Vitek, & Ferreira, 2002). The challenge with tau as a drug target has been how to modulate it. The normal function of tau is as an intracellular scaffolding protein forming microtubules (Weingarten, Lockwood, Hwo, & Kirschner, 1975). Hyperphosphorylation of tau leads to aggregation first in the form of paired helical filaments and then into tangles (Grundke‐Iqbal et al., 1986; Iqbal et al., 1986). As a scaffolding protein, it is much less obvious if there is an amenable binding site for small molecule drugs to interfere with its pathophysiological function compared with the active site of an enzyme or ligand binding site on a GPCR. Additionally, being an intracellular protein, it was not thought to be amenable to antibody therapies that do not readily penetrate into cells. It is only in recent years, when it was discovered that tau accumulation follows a specific pattern of spreading through the brain, which is hypothesized to occur via trans‐synaptic transmission of tau “seeds” (Clavaguera et al., 2009, 2013; Frost & Diamond, 2010), that there has been renewed interest in tau as a target. Several anti‐tau antibodies have been developed (Golde, Lewis, & McFarland, 2013) and are moving into Phase II clinical trials (Cummings, Lee, Ritter, & Zhong, 2018). The knowledge gained from negative amyloid trials in terms of patient selection, appropriate endpoints, and effective use of biomarkers including tau PET ligands will all hopefully enable much faster testing of the tau hypothesis in AD patient trials. The evidence with tau suggests that it remains an important driver of disease progression later into the disease course than amyloid, and hence, trials in prodromal or mild‐to‐moderate stage AD patients perhaps have a greater chance of success.

Figure 3.

Alternative routes to effective treatments for Alzheimer's disease (AD). The schematic illustrates nonamyloid mechanisms that have or are being investigated for therapeutic potential in AD over the last 30 years. The neurochemical analyses of 1970s and 1980s led to the development and launch of symptomatic treatments for AD around the turn of the century. AChE inhibitors and the glutamate antagonist memantine are the only approved drugs to treat the cognitive symptoms of AD. Most of the research activity is now focused on disease‐modifying treatments rather than more symptomatic drugs. Tau and apolipoprotein E were recognized in the 1990s as key players in AD pathophysiology but have yet to be successfully targeted for disease‐modifying therapies. Advances in genetic technologies led to genome‐wide association studies (GWAS) uncovering small‐effect size risk alleles, which have fuelled most of the avenues of active research today. These include lipid processing, mitochondrial dysfunction, protein folding, and clearance mechanisms as well as the role of the innate immune system in neurodegeneration

ApoE function is a second area of great interest, as ApoE is the biggest genetic risk factor in the general population (Corder et al., 1993; Holtzman, Herz, & Bu, 2012; Strittmatter et al., 1993). Unlike APP and PS mutations that are causal for FAD, ApoE just increases the risk of developing late onset AD. There are three common allelic forms ε2, ε3, and ε4 with a frequency of 8%, 78%, and 14%, respectively (Farrer et al., 1997). Compared with someone with the ε3/ε3 genotype, the odds ratio of developing AD is 3 and 15 times higher for the ε3/ε4 and ε4/ε4 genotypes, respectively (Farrer et al., 1997). The precise role ApoEε4 plays in AD pathophysiology has been difficult to determine, but it has been shown to increase amyloid deposition (Bales et al., 1997), impair degradation (Jiang et al., 2008), and exacerbate tau‐mediated neurodegeneration (Shi et al., 2017). Finding ways to modulate ApoE function safely as a drug target for AD has so far proved extremely difficult.

The third major step towards alternative drug targets for AD has again come from the field of genetics. The advance in sequencing technologies over the last 30 years has enabled genome‐wide association studies so that genes with small effect sizes could be identified, rather than just the large effect size of familial genes first discovered in the 1990s. Several genome wide association studies involving hundreds of thousands of subjects have now identified approximately 30 risk alleles (Guerreiro & Hardy, 2014; Zhu, Tan, Tan, & Yu, 2017) and two protective variants (ApoEε2 and PLCγ2; Corder et al., 1994; Sims et al., 2017). Other than ApoEε4, which has been known as a significant risk factor for many years (Corder et al., 1993; Farrer et al., 1995), few of the newer risk alleles are directly involved in amyloid or tau processing. Instead, pathway and network analyses have shown them to be primarily related to the innate immune system, energy metabolism, and proteostasis/autophagy (De Strooper & Karran, 2016; Zhu et al., 2017). These discoveries have also revealed a convergence of the mechanisms involved in neurodegeneration, so that despite different clinical presentations and neuropathological signatures in other dementia‐causing diseases, the underlying biology shares many commonalities (Brettschneider, Del Tredici, Lee, & Trojanowski, 2015; Salter & Stevens, 2017). The above factors have reinvigorated the dementia research field and industrial drug discovery investment. Although many of the risk alleles are not themselves particularly obvious drug targets, they sit on pathways with tractable targets that are now at the preclinical stages of drug discovery. In the next few years, many of these will move into clinical testing and hopefully yield positive results on disease progression in patients.

Whilst the main drug development focus is on finding a drug that is efficacious, it is most likely that polypharmacy will be required for effective clinical management of AD. The genetic studies indicate that more than one pathophysiological mechanism can cause AD and it is likely that multiple mechanisms will be active in any given patient. Recent post‐mortem studies are increasingly demonstrating that many patients have overlapping neuropathological features of AD, vascular dementia, dementia with Lewy bodies, and so forth (Attems & Jellinger, 2014; Irwin et al., 2017; Spires‐Jones, Attems, & Thal, 2017). These data indicate that mixed pathologies may be relatively common, and therefore, treating the patients' disease with drugs of multiple mechanisms will likely be necessary. The use of biomarkers, MRI/PET scans, and other means of appropriately stratifying patients will need to develop significantly from where they are today to be able to correctly identify which mechanisms are relevant to a specific patient and therefore start treatment with the appropriate drugs. Such precision medicine approaches for cancer therapy are now common place and hopefully will become so for AD and other dementias in the future.

Improvements in the research landscape and drug target pipeline over the last decade have also been matched by changes in the political and funding landscape. Several governments have woken up to the impact of dementia on health and social care systems and developed routes and action plans to address the issue. A prominent example was the G8 summit in 2013 focused on dementia (Fox & Petersen, 2013) that led to the setting of a key therapeutic goal: to identify a cure or disease‐modifying therapy for dementia by 2025 (Vradenburg, 2015). Many organizations around the world have aligned themselves with this goal, which could be achieved by several agents in late stage clinical trials (Cummings, Lee, et al., 2018), although this is by no means a certainty! The setting of political goals has also brought increased government funding into the dementia research field with big increases in the U.S.'s National Institute of Health dementia budget in recent years (Alzforum, 2018e) and the establishment of the Dementia Research Institute in the United Kingdom (https://ukdri.ac.uk/). Increased venture capital funding has also been created through targeted investment mechanisms such the Dementia Discovery Fund, which now has $350 million to invest in dementia drug discovery and development (https://theddfund.com/). Finally, philanthropists with global profile such as Bill Gates (Marais, 2018) and the Chan‐Zuckerberg Initiative (https://www.chanzuckerberg.com/) are increasingly turning their attention to dementia research funding, which attracts further public and political interest in the area.

In summary, we are still travelling down the long road to effective disease‐modifying treatments for dementia, but several alternative routes are hopefully increasing our chance of successfully reaching our destination. Experience in how to run clinical trials should help avoid some of the inefficiencies of previous trials. Rapidly improving understanding of the mechanisms involved in neurodegenerative diseases is opening up new areas of biology and offering a wealth of potential drug targets. Increased political and public awareness and interest in dementia is leading to more funding and increased research as well as drug discovery and development capacity. Hopefully, they will all help us find the urgently needed treatments and better outcomes for people living with dementia.

5.1. Nomenclature of targets and ligands

Key protein targets and ligands in this article are hyperlinked to corresponding entries in http://www.guidetopharmacology.org, the common portal for data from the IUPHAR/BPS Guide to PHARMACOLOGY (Harding et al., 2018), and are permanently archived in the Concise Guide to PHARMACOLOGY 2017/18 (Alexander, Christopoulos et al., 2017; Alexander, Fabbro et al., 2017; Alexander, Kelly et al., 2017; Alexander, Peters et al., 2017).

CONFLICT OF INTEREST

D.R. holds shares in Pfizer Ltd and was a past employee of Pfizer Inc. and Merck Inc.

Reynolds DS. A short perspective on the long road to effective treatments for Alzheimer's disease. Br J Pharmacol. 2019;176:3636–3648. 10.1111/bph.14581