Historical highlights

It would be difficult to overstate the urgency of finding solutions to the Alzheimer disease (AD) problem. Alzheimer Disease International, in their 2010 World Alzheimer Report, estimates that there are 35 million people suffering from this disorder at an annual cost of $604 billion (http://www.alz.co.uk). The 2013 United States Alzheimer’s Association Report estimates that there are 5.2 million cases in that country alone, with a new case being identified every 68 s. The cost in the US is estimated to be $ 203 billion, not including the unpaid care costs of patients, at $216 billion [156]. More ominously, both reports conclude that unless effective measures of prevention and treatment are discovered, the numbers of cases may increase two- to threefold by 2050.

Such a situation could hardly have been imagined by Alois Alzheimer when, in November 1907, he reported at a meeting in Tübingen “Ueber eine eigenartige Erkrankung der Hirnrinde” (On an unusual illness of the cerebral cortex). His two-page paper is worth reading by every AD researcher today. A translation into English is available [151]. The paper gives a succinct description of the clinical history in a case of dementia with onset at age 51 and the unique pathological findings of plaques and tangles. It concludes with a warning that researchers should not be classifying clinically undetermined cases by forcing them into categories of recognized illnesses. His report ends with this disappointing footnote: “Keine Diskussion”.

The failure to pay much attention to this hallmark paper continued for a further seven decades. It fell to Robert Katzman to ring the alarm bells. In a 1976 editorial he pointed out that AD and senile dementia were indistinguishable and ranked as the fourth or fifth most common cause of death in the US [83].

The decade from 1982 to 1992 was a groundbreaking era for unraveling the mysteries of AD. Up to then, age was the only known risk factor. Below age 65 AD was found to be rare, but the prevalence had been reported to double every 5 years thereafter, reaching about 25 % in the 80–85 year age bracket [79]. There was no explanation then, nor is there now, as to why age is so critical.

The most important development of that decade was Glenner and Wong’s [48] identification of a cerebrovascular amyloid protein which turned out to be Abeta. They then showed the same protein could be found in Down’s syndrome cases, correctly predicting that the genetic defect causing AD would be found on chromosome 21 [49].

The gene for amyloid precursor protein (APP) was soon cloned by three groups [52, 109, 133]. The full structure of Abeta was then apparent. It was concluded that an abundant presence of Abeta immunoreactive plaque lesions throughout the cortex and subcortical gray matter structures was typical of AD [70].

A point mutation in Abeta was then found that produced cerebral hemorrhages and heavy deposits of Abeta in blood vessels and brain [99, 158]. In further screening of chromosome 21, Goate et al. [50] found an AD causing point mutation in Abeta, providing a direct link between Abeta metabolism and AD.

The second pathological hallmark of AD is the presence of intracellular neurofibrillary tangles. The core protein was identified as tau [51, 54, 88] and its incorporation into tangles involved phosphorylation [69]. Since tangles were known to appear in diseases where plaques were absent, tau could be discounted as the fundamental cause [171]. Nevertheless tangled neurons die, so they are an extremely harmful consequence of AD pathogenesis even if they are a secondary phenomenon. The dead neurons leave behind their tau deposits as ghost tangles.

Braak and Braak [16] then published their classic paper on the neuropathological staging of AD in 1991. It was not based on Abeta deposits, but on the hierarchical development of neurofibrillary tangles (NFTs) and neuropil threads (NTs). This was a paradoxical finding, since genetic studies had already established that Abeta abnormalities induced tau abnormalities and not vice versa. In the Braak and Braak series of cases, amyloid deposits, even though variable, frequently appeared prior to tau deposits in pre-dementia stages, so there may not have been an incompatibility between genetic and early pathological findings.

How amyloid deposits spur tau aggregation is still a mystery. However, it is known that extracellular Abeta deposits induce inflammation. That inflammatory stimulation may enhance tau upregulation, which, in turn, may promote intracellular tau aggregation in vulnerable neurons. Tau aggregation can be self-sustaining. There is increasing evidence that such self-propagating tangle formation occurs following concussions in sports such as boxing and football. Dementia developing many years later is the end result [117, 119]. These data have profound implications for therapeutic strategies. If tau aggregation, once initiated, develops a life of its own independently of Abeta aggregation, then therapy based only upon limiting Abeta production or clearance will be of limited value.

Selkoe [143] first integrated the accumulated data into the amyloid cascade hypothesis of AD. This concept was similarly advanced by Hardy and Allsop [57]. The amyloid cascade hypothesis is nevertheless incomplete since it fails to take into proper account the neurotoxicity associated with the resulting neuroinflammation. Eikelenboom and Stam [37] first drew attention to an inflammatory process by reporting that AD senile plaques were positive for the opsonizing components of complement. Griffin et al. [53] noted that the inflammatory cytokine IL-1 was elevated in the brains of both AD and Down syndrome cases. When complement activation was further explored, a fundamental dichotomy was discovered. It was found that, while the opsonizing components decorated senile plaques, the terminal components, which form the membrane attack complex (MAC), were associated with dystrophic neurites [112, 113].

The reason why the complement system is so important in AD came with the discovery by Rogers et al. that Abeta and its N-terminal fragments bind to C1q. Thus Abeta deposits directly activate the complement system, initiating all the inflammatory consequences which occur in AD [134].

These results, as illustrated in Fig. 1, show the features of neuroinflammation in AD as revealed by immunohistochemistry. They demonstrate that deposits of Abeta initiate activation of the classical complement pathway in AD, causing opsonization of the plaques. Microglia then become activated and aggregate over the opsonized material. While such opsonization promotes their phagocytosis, continuation of the complement cascade to formation of the MAC results in self-damage by the MAC inserting itself into dystrophic neurites (Fig. 1c). These dystrophic neurites are characterized by intracellular tau aggregates (Fig. 1f). The MAC, which is intended to destroy invaders, instead turns on the host, destroying viable cells and their processes. As inflammation intensifies, so does the process of neuronal damage.

Fig. 1
figure 1

Alzheimer disease markers illustrating the inflammatory reaction process. a Alzheimer senile plaques immunostained with the Abeta antibody 6F3D (l0X). b Alzheimer senile plaques immunostained with C4d antibody (10X). Notice the similar staining indicating that the classical complement pathway has been activated on senile plaques, thus opsonizing them. c Alzheimer brain tissue immunostained with an antibody to the membrane attack complex of complement C5b-9 (20X). The immunostaining is completely different. Dystrophic neurites within and around a senile plaque are immunostained, indicating that the autodestructive MAC has inserted itself into viable, but damaged neuronal processes. d Shows Alzheimer disease senile plaques doubly immunostained for Abeta (brown) and microglia (CD 11b, blue, ×10). The microglia are activated and agglomerated over plaques but are not activated in the surround. e Alzheimer brain tissue doubly immunostained for C4d (brown) and microglia (CD 11b, blue, ×10). The immunostaining is similar to D, indicating that it is the opsonization of the plaque that has attracted and activated the microglia. f Alzheimer tissue doubly immunostained for phosphorylated tau (AT8, brown) and C5b-9 (blue, ×10). Phosphorylated tau deposits are noticeable as many brown neuropile threads, while some, colored blue, are under attack from the MAC. g Alzheimer brain tissue doubly immunostained for microglial cells (IbA1, blue) and Abeta (6F3D, brown, ×40). The senile plaque is being attacked by agglomerated and activated microglia in contrast to the surround. h Control brain tissue doubly immunostained for microglia (IbA1, blue) and Abeta (6F3D, brown, ×10). Only inactive microglia appear, as there are no brown Abeta deposits. i Control tissue (10X) doubly immunostained for microglia (IbA1, blue) and phosphorylated tau (AT8, brown). Again, only resting microglia appear as there are no phosphorylated tau deposits

Full size image

Since self-damage from the MAC is prominent in AD, it is logical to suppose that those taking antiinflammatory agents should be relatively spared from the disease. Rheumatoid arthritis were a good population to study, since disease onset occurs earlier than in AD, and patients are known to be chronic users of antiinflammatory agents. An initial epidemiological study was carried out in which the prevalence of AD amongst rheumatoid arthritics over the age of 65 was compared with the prevalence of AD reported to occur in age-matched general populations. An apparent sixfold sparing was noted [115]. Numerous epidemiological studies have since been carried out with some variation in the results. Nevertheless, the overwhelming consensus is that antiinflammatory agents, particularly NSAIDs, if started early enough, have a sparing effect on AD. These studies are summarized in Table 1.

Table 1 Epidemiological studies of NSAIDs and AD
Full size table

During the 1982–92 decade, these key discoveries provided insight into how successful treatments for AD could be developed. Yet, 20 years later, no disease-modifying therapies have reached the bedside. Failures have too often been blamed on the amyloid cascade hypothesis being in error, when the evidence indicates that inappropriate attempts to apply it may be at fault. As Schneider and Lahiri [141] observed “It seems—as if both large and small companies keep repeating each others’ mistakes, deriving their development program from an earlier, often unproven one”. A comprehensive list of clinical trials, either completed or underway as of 2011, is summarized by Mangialasche et al. [108].

Mutations causing Alzheimer disease

The central role of Abeta in AD pathology has been further strengthened by the discovery of additional autosomal dominant mutations in APP, either within Abeta itself, or close to the N- and C-terminal regions. These include the regions Abeta minus1-2, Abeta6-7, Abeta22-25, and Abeta43-46 [65]. It is unclear why these mutations cause disease. There are additional polymorphisms and mutations which enhance the risk of AD without directly causing it. These provide clues as to additional factors which may be involved in AD pathogenesis. By far the most important of these is apolipoprotein E (ApoE). One copy of ApoE4 has been shown to enhance age-specific AD prevalence by three- to fourfold and two alleles by about 15-fold [136]. In contrast, ApoE2 is mildly protective. ApoE is involved in cholesterol metabolism, indicating an important, but not fully understood role in Abeta metabolism. Significant, but much less important, polymorphisms in other genes have been identified as risk factors. These include complement receptor 1 (CR1) [30, 60] and the triggering receptor on myeloid cells (TREM 2) [55, 78], both of which are involved in responses to inflammation. The phosphatidylinositol binding clathrin assembly binding protein (PICALM) has also been identified as a risk factor [58].

Neurotoxicity associated with Abeta

Despite the evidence provided by AD causing mutations, there are significant weaknesses associated with the amyloid cascade hypothesis. More than 99 % of all AD cases fall into the wild-type category, also known as sporadic or late onset AD (LOAD), for which no explanation exists. Moreover, the hypothesis fails to explain how Abeta in any of its forms causes neuronal death. That is why an extension to an amyloid cascade-inflammatory hypothesis is appropriate. If neuronal death is to be arrested or prevented, therapeutic strategies based on blocking the resulting neuroinflammation may need to be followed.

The levels of Abeta in brain are very low, being in the pmol/g range. Xia et al. [176] found the in vivo levels for Abeta 1-40 to be 1.1 pmol/g in AD brain compared with 0.7 pmol/g tissue for controls. They found Abeta 1-42 levels in AD brain to be comparable to Abeta 1-40 levels, but Abeta 1-42 levels in controls to be much lower. Steinerman et al. [150] found soluble Abeta 1-40 in AD patients to be 2.7 pmol/g, while control levels were only 0.8 pmol/g. They reported Abeta 1-42 levels to be 14.6 pmol/g, while controls were only 0.8 pmol/g. Brody et al. [20] measured Abeta 1-x levels in vivo by dialysis of interstitial fluid in the white matter of brain injured patients. They found levels of 0.44 pmol/ml in patients with axonal brain injury but in other types of brain injury they were less than 0.22 pmol/g. While there is some variability in these results, they provide convincing evidence that soluble levels of Abeta species in human brain in vivo are in the very low pmol range.

Hypotheses have been put forward that Abeta itself, a toxic fraction such as Abeta 25-35, fibrillar Abeta, or oligomers of Abeta, are directly neurotoxic. In all cases, neurotoxicity has been in the micromolar range, more than a million times too high to lend credibility to the theory that they are neurotoxic in vivo. Yankner and colleagues [178] initially proposed that a fraction of Abeta was directly toxic to primary hippocampal cultured neurons. May et al. [111] then found the required concentration to produce significant toxicity was 10 μm, while Forloni et al. [42] found that repeated additions of 25–100 μm of Abeta 25-35 were required.

Next it was proposed that fibrillized Abeta or soluble oligomers of Abeta were the neurotoxins [104]. Numerous publications have since appeared in which the comparative neurotoxicity of Abeta, fibrillized Abeta, and oligomerized Abeta have been tested against primary cultured neurons [34, 85, 126, 167] with toxicity always in the micromolar range.

The logical conclusion is that it is deposits of Abeta that are the problem. They are neurotoxic due to the inflammatory reaction they stimulate. As Fig. 1 shows, each consolidated Abeta deposit is a focus of inflammation. While the numbers of such deposits in AD have never been counted, Bussiere et al. [24] have estimated their volume to be as high as one-third of the total volume in highly vulnerable brain regions such as the subiculum. Accordingly, therapeutic strategies that ignore the inflammation these deposits provoke are unlikely to be successful.

Enzymatic cleavage of APP

Since inability to handle an Abeta load must be a precipitating factor in AD, it is critical to have a clear understanding of the key enzymes involved. Those responsible for cleaving APP have been named α, β, and γ secretases. α-Secretase is actually a family of disintegrin and metalloprotease domain (ADAM) enzymes which cleave APP at the glutamine–lysine site within Abeta (Abeta 16-17). Such a cleavage prevents Abeta formation. ADAM 10 is considered to be the most important member of this family, but stimulation of its activity is considered a risky therapeutic strategy due to cleavage of other substrates that may stimulate tumorigenesis [129].

After APP is generated, it is transported to the cell surface. The N-terminal region traverses the external membrane and is cleaved by β-secretase or BACE-1 [65, 67, 162]. For Abeta to form, this enzyme must have preferred access to APP compared with α-secretase. Cleavage by BACE-1 occurs within the cellular membrane leaving the C-terminal region containing Abeta 1-29 still within the membrane [65].

Here it can be cleaved by the third enzyme, γ-secretase. It is a trans-membrane protease complex with multiple components embedded in the membrane. Knowledge concerning this complex is still not complete, but it is known to include several proteins. Initially, two genes linked to the catalytic subunits of γ-secretase were identified by genetic screening of familial AD cases. They were named presenilin-1 (PS-1), located on chromosome 14 [146], and presenilin-2 (PS-2), located on chromosome 1 [98]. Multiple mutations in these genes were found to cause autosomal dominant AD. By 2012, their numbers had reached 185 for PS-1 and 13 for PS-2 [26]. For reasons unknown, the cleavage process is variable. The main product is Abeta40, followed by Abeta42, but the cleavage also occurs at Abeta38 and other sites beyond Abeta42. The cleaved products are all released extracellularly.

In a series of experiments Haass and colleagues demonstrated the assembly route of γ-secretase. It commences in the endoplasmic reticulum where nicastrin, anterior pharynx-defective 1 (APH-1), and presenilin enhancer 2 (PEN2) are separately expressed and bind together in the Golgi apparatus. Trafficking then takes the complex to the transmembrane region and lysosomes where it becomes activated [80].

It would logically be presumed that the presenilin mutations would produce a gain of function, thus enhancing the production of Abeta. But at least one-third are nonsense mutations or substitutions which clearly produce a loss of function. If Abeta overproduction is causative for AD, how is it possible that reducing γ-secretase through one incompetent gene can result in autosomal dominant AD? If γ-secretase is a target for Abeta therapy, should its action be enhanced or inhibited? These and other questions regarding the viability of Abeta-degrading enzymes remain unanswered [81, 123, 138].

Abeta deposition can be prevented through cleavage by extracellular enzymes. Neprilysin is the major peptidase carrying out this function [38, 74]. Both the mRNA [179] and the protein [73] are down regulated in vulnerable areas of AD brain, suggesting this may be an initiating factor in the disease. Insulin-degrading enzyme and several minor peptidases can also cleave extracellular Abeta [123]. Up-regulation of Abeta-cleaving enzymes is an attractive strategy, but no practical methods for doing so have so far been developed.

Transgenic animals

A new dimension was added to AD research when Games et al. [45] published the first transgenic mouse (Tg) model that developed amyloid plaque pathology. They transfected mice with a minigene encoding the human disease-causing mutant APPV717F (PDAPP) behind a platelet-derived growth beta promoter. The mice developed plaque deposits at 6–9 months of age. This success was soon confirmed by the creation of new models [68] which involved transfecting mice with single, or multiple AD causing genetic mutations behind various promoters. Within the first decade, eight had been developed incorporating APP mutations, six incorporating PS-1 mutations, and three incorporating PS-2 mutations, plus various combinations [116]. The most aggressive so far is the B6SJL-5X Tg model, developed by Jackson Laboratories, which produces plaques starting at 6–8 weeks of age.

These mouse models have been utilized as the gold standard for developing new AD treatments. But it must be kept in mind that they do not fully replicate the pathology of AD. They are, therefore, incomplete models [87, 142]. The most obvious differences are the failure to develop neurofibrillary tangles and the failure to demonstrate significant neuronal loss. A more subtle difference is the relatively weak inflammatory response in the mouse models.

Abeta deposits in mouse models are readily soluble in SDS, while AD deposits can only be solubilized by such harsh treatment as hot formic acid. Human Abeta deposits undergo such post-translational modifications as N-terminal degradation, pyroglutamyl formation, and cross-linking, while mouse deposits do not [82, 91].

There is also a major difference in the response of the complement system. Mouse C1q poorly recognizes human Abeta deposits in Tg mice, so the overall response is weak compared with human AD. There are differences in human and mouse Abeta at residues 5, 10, and 13 which may play a role in this weak recognition [41, 169, 184].

There is no expression of the MAC in Tg mouse models [82, 142], while it plays a prominent part in AD pathogenesis [112, 113, 168, 181]. While stimulation of an inflammatory response in Tg mice might be helpful by promoting phagocytosis without inducing the self-damaging MAC, comparable stimulation in AD could be anticipated to produce highly damaging results [114].

The discovery by Schenk et al. [140] that Abeta deposits in PDAPP mice could be attenuated by vaccination with full-length Abeta opened up a revolutionary approach to treating AD. The beneficial effect of active immunization in mice, including behavioral improvement, was soon confirmed by others [77, 120, 175]. These findings were then extended to passive immunization, where peripherally administered Abeta antibodies were effective in assisting efflux of Abeta from the brains of mice, although a heavy Abeta load was less effective [32, 33].

Translational experiments

Clinical trials efforts based on immunization

The ease of plaque clearance by active and passive immunization of mice raised hopes that a similar strategy would be successful in treating human AD. They were quickly dampened when the first clinical trial of an Abeta vaccine named AN 1792 was halted because of dangerous side effects. These included death, stroke, and sterile meningoencephalitis in about 6 % of patients [127]. Nicoll et al. [124] then reported widespread encephalitis in an autopsy case although there was plaque clearance. But this clearance could have occurred due to macrophage invasion after remaining dystrophic neurites had been destroyed [5, 6]. Consistent with this interpretation, it was concluded in a follow-up of 80 patients, including post-mortem analysis of eight cases, that there was no slowing of disease progression and no clearance of tau deposits [66].

One explanation offered for the dangerous side effects in the AN 1792 trial was adjuvant-dependent modulation of Th-1 and Th2 responses. Novartis then developed a vaccine designated avagacestat, against gamma secretase. In an initial study involving 46 AD patients, no cases of meningoencephalitis were identified, suggesting that it had an acceptable safety profile [29]. But Bristol-Myers Squib terminated clinical trials in December 2012 due to lack of efficacy in AD. The fundamental problem is that an autoimmune disease is induced by active immunization, and it has not been demonstrated that this can be done without causing self-damage from the MAC. It has even been suggested that vaccination with DNA is a possibility and that immune-suppressive therapy should accompany active vaccination attempts [31].

One possibility for the future is to combat any such self-damage with a specific inhibitor of MAC formation, such as aurin tricarboxylic acid. In a pilot trial using 5X B6SJL-Tg mice it was found to be without toxicity and effective in preserving memory [95].

Passive immunization attempts have also been made, using antibodies against specific regions of Abeta. While such a strategy avoids the complication of inducing an autoimmune disease, it does require the relatively invasive and highly costly procedure of repeated infusions of antibody. Again, the theoretical basis for believing it could be effective in AD is tenuous, since the transgenic mouse models where they have been tested are not a reliable indicator of what will take place in humans. So far all attempts at passive immunization have failed.

Bapineuzumab is an example. It is a humanized monoclonal antibody directed at the N-terminal sequence of Abeta. Four clinical trials have been undertaken. After the failure of two of these trials, development by Johnson and Johnson and Pfizer was discontinued. Altogether 4,500 mild to moderate AD patients, randomized to the antibody or placebo, were tested over 18 months. Side effects included vasogenic edema at higher dose levels. In the CSF, no difference was noted in Abeta levels between test and control patients, although bapineuzumab patients showed a decrease in total tau and p-tau [15].

Yet another antibody is solanezumab, developed by Eli Lilly. This antibody is directed at the mid-region of Abeta and is thought to have an advantage over bapineuzumab by binding with much higher affinity and only to soluble Abeta. Its safety profile is higher [71] but, as with bapineuzumab, it failed in clinical trials [155]. In this case more than 2,000 patients did not meet the primary outcome measures of cognition and function. Nevertheless, Eli Lilly has plans to carry on with clinical trials.

Ponezumab, developed by Pfizer, is a humanized monoclonal antibody which binds specifically to the C-terminus of Abeta40. As with the other humanized antibodies, it showed efficacy in Tg mice [94]. But, after a phase 2 clinical trial, it was halted by Pfizer for unspecified reasons.

The most ambitious trial of all is that being undertaken by Genentech of the humanized monoclonal antibody against Abeta40 and Abeta42 named crenezumab [122]. The trial, projected to cost over $100 million, will test the effects on 300 individuals from Colombia known to have a PSEN-1 mutation.

There are continuing plans to develop a successful treatment of AD with active or passive immunization strategies. These are based on the belief that the science behind the strategy is sound, but failures have also been interpreted as being due to incorrectness of the amyloid cascade hypothesis itself [155]. They overlook the role of inflammation and the fact that immunization, whether active or passive, will stimulate further inflammation rather than compensate for its effects. According to this interpretation, future immunization trials are unlikely to succeed.

Clinical trial efforts based on inhibiting Abeta levels

Weggen et al. [170] found that certain NSAIDS such as flurbiprofen and ibuprofen modified production of Abeta42 in favor of Abeta1-38. These profens exist in S and R enantiomeric forms. Only the S form inhibits COX, but both forms equally affect Abeta production [121]. The authors proposed that the R-form might be an effective treatment for AD and showed that chronic administration was able to attenuate spatial learning deficits in Tg2576 mice [90]. The R form of flurbiprofen, produced as Flurizan, was tested in a trial involving 1,800 AD patients over an 18-month period by Myriad Genetics. It showed no positive effect and was abandoned. A major problem was that the concentrations required to produce the effect on Abeta42 production were more than 100-fold higher than those required to inhibit COX activity [86]. It is not surprising that Szekely et al. [154] found no advantage of Abeta-42-lowering NSAIDs, compared with non-lowering NSAIDs, for prevention of dementia in a data set of six pooled epidemiological studies.

Semagacestat is a γ-secretase inhibitor which was shown to lower plasma, CSF, and Abeta levels in animals [62]. A large international trial of semagacestat was halted before completion because those on the drug fared worse than the placebo patients, and, some, receiving the higher dose of 140 mg/day rather than 100 mg/day, had adverse effects such as skin cancer and infections [35]. The problems have been attributed, at least in part, to inhibition of notch cleavage by semagacestat. Notch is a highly conserved protein with a multitude of essential physiological functions. Inhibitors of γ-secretase cleavage will have dangerous side effects if they also inhibit notch cleavage [36, 80, 173]. Bristol Myers Squib is currently testing their BMS-708163 which is more selective for γ-secretase compared with notch. But, in view of the fact that many of the AD causative PS-1 mutations are nonsense mutations, even specific inhibitors of γ-secretase may have adverse effects.

β-Secretase (BACE 1) inhibitors do not have the systemic drawbacks of γ-secretase inhibitors, but there are already disappointing results. Eli Lilly discontinued development of its β-secretase candidate LY2886721, citing abnormal liver tests as the reason. Merck has initiated a 78-week phase II/III trial of its candidate MK 8931, and others are under development. Mullard [122] has summarized the latest β- and γ-secretase inhibitors, as well as monoclonal antibodies and vaccines that are under trial. There is optimism that β-secretase inhibitors will eventually be successful [47], but, to be clinically effective, an inhibitor will need to be highly selective, very potent, and administered in the early stages of the disease.

Clinical trials based on inhibiting Abeta aggregation

In theory, inhibiting Abeta aggregation is a more attractive strategy than inhibiting its production with enzyme inhibitors. There are no concerns about interfering with any normal functions of Abeta throughout the body, and no concerns about the inevitable side effects of enzyme inhibition. Unfortunately, the aggregation inhibitors that have been tested to date have failed to demonstrate efficacy.

Tramiprosate (homotaurine), an agent which binds to Abeta, thus preventing fibrillization, failed to show efficacy in a phase III trial of 1,052 patients [2]. Scyllo-inositol, a neutralizer of Abeta42 trimers, showed promise in animal experiments [106]. The drug was abandoned, following a trial involving 350 patients, after high doses resulted in infections and death, and low doses failed to show efficacy [137].

Our laboratory has developed a simple in vitro screening assay to replicate as closely as possible the events presumed to be occurring in AD brain [56]. The technique is to take slices of frozen post mortem AD brain tissue and incubate them with a fluorophore (HiLyte488) labeled Abeta42. The Abeta42 derivative is then visualized binding selectively to senile plaques in the AD brain. Candidate inhibitors are next added to determine their relative ability to inhibit the Abeta42 build-up. By this test, tramiprosate and scyllo-inositol were inactive, even at very high concentrations. Agents that showed some promise included 1,2,3,4,6-penta-O-galloyl-d-glucopyranose (PGG), epigallocatechin gallate, and resveratrol. Of potentially greater interest was the inhibitory effect of a number of food products. These included ginger, rhubarb, blueberries, cinnamon, and turmeric. The implication is that each of these products contains some Abeta42 aggregation inhibitors. Of course, such inhibitors would need to reach the brain to be effective in vivo. Many other possible inhibitors have been suggested, with some in various stages of development [172].

Antiinflammatory approaches

Antiinflammatory approaches are based on pathological evidence of neuroinflammation in AD, coupled with epidemiological evidence of AD sparing in individuals consuming NSAIDs. Following the initial report of a sparing of AD in rheumatoid arthritis patients [115], multiple epidemiological studies have been carried out (Table 1). These include incident, population-based, and case–control methodologies.

Seven incidence studies have been carried out (Table 1) [10, 44, 72, 103, 152, 165, 182]. The most comprehensive study to date is that of Vlad et al. [165]. They compared a population from the US Veterans Affairs Health Care System of 49,359 cases of AD with 196,850 controls. The base was large enough to permit analysis of length of any NSAID use and to compare the effects of various NSAIDs.

They found that for more than 5 years of use, the most effective preventative NSAIDs were indomethacin (OR 0.45), piroxicam (OR 0.54), and ibuprofen (0.56). Naproxen (OR 0.78) was considerably less effective. In contrast, COX-2 inhibitors substantially enhanced the risk of developing AD. The odds ratio for rofecoxib climbed from 0.90 for less than 1 year of use to 5.89 for more than 5 years of use. For celecoxib the odds ratio for less than 1 year of use was 0.94, climbing to 1.27 for more than 5 years of use. Such a negative result for preferential COX-2 inhibitors was to be expected given that brain is one of the few organs where COX-2 is constitutively expressed at high levels. It is particularly concentrated in hippocampal neurons, implying a significant role in synaptic transmission [84].

As for the traditional NSAIDs, those with a high COX1/COX 2 inhibitory ratio [43] appeared to give the most protection. Naproxen, utilized in two failed clinical trials, has a relatively unfavorable odds ratio of 0.78.

Int Veld et al. [72] had previously shown in the Rotterdam prospective study of 6,989 subjects that the relative risk of AD amongst NSAID users was 0.95 for 1 month use or less, 0.83 for 1–2 years of use, and 0.20 for longer than 2 years of use. Diclofenac, ibuprofen, and naproxen were the most widely used NSAIDs. Similarly, Zandi et al. [182], in the Cache County Utah study of 3,227 participants, found a risk ratio of 0.75 for less than 2 years of use, decreasing to 0.45 for longer than 2 years of use. Steward et al., in the Baltimore Longitudinal study of 1,686 participants [154], found the odds ratio to be 0.40 for more than 2 years use of NSAIDs but 0.65 for less than 2 years. Ibuprofen was the most commonly used NSAID. Lindsay et al., in the Canadian Study on Health and Aging, found an odds ratio of 0.65 for NSAID use, but the amount and type were not recorded [103].

In contrast to these studies, Arvenitakis et al. [10], in a much smaller population of 1,019 Catholics enrolled in the Religious Orders Study, found no apparent relationship between NSAID use and incident AD. No data were provided on duration of use. Fourier et al. [44] in the French Paquid cohort found no sparing of AD in 53 NSAID users compared with 1,199 non-users, but only for 1 year of use.

The six case–control studies shown in Table 1 are of smaller size [7, 13, 18, 19, 174, 180]. Where the duration of use is recorded, the results are in line with the incidence data in showing a lower odds ratio for longer NSAID use. Duration of use is not given in the three population-based prevalence studies [8, 21, 93].

The overall conclusion from these epidemiological studies is that significant sparing of AD occurs only when NSAIDs have been used for at least 2 years prior to onset of clinical symptoms. Selective COX-2 inhibitors are harmful, and those traditional NSAIDs with the highest COX-1/COX-2 inhibitory ratio are the most beneficial.

Tg mouse studies involving NSAIDs

Table 2 gives a list of 14 Tg mouse studies where the effects of NSAIDs have been tested [27, 61, 63, 76, 89, 101, 102, 118, 131, 153, 159–161, 177]. These studies have the advantage over human epidemiological studies in that the dose, duration, and NSAID involved are known. They suffer the disadvantage of being tested in a heterogeneous collection of six mouse models.

Table 2 Transgenic animal studies with NSAIDs
Full size table

Traditional NSAIDs were reported to have an ameliorating effect on Tg pathology or behavior in 13 of the 14 studies where they were tested. Ibuprofen was the NSAID of choice in ten of these studies. The selective COX-2 inhibitors nimusulide and celecoxib were ineffective, and, in one study, celecoxib tripled the plaque load [89]. Although there are correlations between the results of mouse models and epidemiological studies, they must be regarded with caution.

Clinical trials involving NSAIDs

Table 3 lists the nine clinical trials so far conducted using COX-1 and COX-2 inhibiting agents. Initially, a small pilot clinical trial of indomethacin was carried out which appeared to show some benefit in mild to moderate AD [135], while another with diclofenac/misoprostol showed a non-significant trend towards protection [139]. Gastrointestinal problems led to a high dropout. These trials, which suffered from short duration, low power and high drop-out, could only be interpreted as indicating the need to undertake more definitive trials.

Table 3 Clinical trials with COX-1 and COX-2 inhibiting NSAIDs other than aspirin
Full size table

This was followed by four clinical trials with selective COX-2 inhibitors: one with nimesulide [4], one with celecoxib [147], and two with rofecoxib [3, 132]. These trials were undertaken in the absence of any epidemiological evidence of a protective effect of selective COX-2 inhibiting NSAIDs. In contrast, there was evidence of potential harm due to the high constitutive expression of COX-2 in cortical and hippocampal neurons and a presumed role in synaptic signaling [84]. Although the trials were adequately powered and therapeutic doses were utilized, there was no therapeutic effect. The harmful effects of COX-2 inhibition only appeared in later epidemiological [165] and animal model [89] studies. It must be concluded that selective COX-2 inhibitors do not have a future in the treatment of AD.

Naproxen has been utilized in two clinical trials. A multi-center, randomized, double blind, placebo-controlled parallel group trial with 1 year exposure was evaluated [3]. There were 111 placebo patients, 88 of whom completed the study, and 118 naproxen cases, 90 of whom completed the study. Naproxen was administered at a sub-therapeutic dose of 440 mg/day. In this trial, there was not a significant drop-out due to gastrointestinal problems, perhaps because of the relatively low dose of naproxen. No differences in deterioration were noted between the test and control groups.

The National Institute on Aging sponsored the Alzheimer disease Antiinflammatory Prevention Trial (ADAPT) to address the question of NSAID prevention of AD [1]. Cases were drawn from populations at high risk of developing dementia prior to the onset of symptoms. Celecoxib and naproxen were selected as representatives of COX-2 and COX-1 preferential NSAIDs, respectively. It is difficult to understand the selection of these agents, given their poor epidemiological, clinical, and Tg mouse record.

Enrollment commenced in 2001 but was terminated in December 2004 because of perceived cardiovascular side effects [1]. That is also difficult to understand, given that the differences in cardiovascular events between placebo, celecoxib and naproxen were marginal. While naproxen is considered safe enough to be an over-the-counter agent, it has nevertheless been reported that there is a small, but significantly increased risk of myocardial infarction in patients taking naproxen, rofecoxib, celecoxib, diclofenac, and ibuprofen [64]. In the long run, the benefits of preventing as deadly a disease as AD need to be weighed against the low risk of developing heart disease. Follow-ups have been reported on the ADAPT cohort, but they are difficult to interpret, given the termination of treatment followed by a period of non-treatment [96, 97].

The most relevant clinical trial involving COX inhibitors is that of Pasaqualletti et al. [128]. They administered ibuprofen (400 mg bid plus esomeprazole) or placebo to 132 AD cases (MMSE 15-26) over a 1-year period. There were 66 patients in each arm. The only treatment group not showing a decline was the ApoE4 carriers.

These results have raised questions as to whether the theory itself is wrong, or whether NSAIDs can act only as preventative agents if started very early in disease development. The protective effects of NSAIDs revealed in epidemiological studies all involved up to 2 years of use before clinical diagnosis. To be effective, it can be expected that NSAIDs will need to be administered very early in disease development and at doses which are appropriate to the advancing inflammation.

Biomarker studies

Biomarker studies have opened up a new era for Alzheimer research. So far there are three reliable biomarker types: CSF to determine Abeta and tau secretion levels; positron emission tomography (PET) to determine Abeta deposit level (Pittsburgh Compound B), and metabolic rate (FDG); and MRI to determine brain volume. Overall, they show that AD onset occurs a decade or more before clinical symptoms appear.

Sperling et al. [148] proposed a sequential model of the preclinical stages of AD focusing on the 10-year period before the emergence of cognitive decline. Their model proposed a series of steps but no time line was attached to them. Bateman et al. [12], in a landmark study of 128 participants, identified the same steps while including a time line. Concentrations of Abeta42 in the CSF declined 25 years before the expected clinical onset. Abeta deposits in the brain, as revealed by Pittsburgh compound B, were detected 15 years before clinical onset, along with increased tau in the CSF and an increase in brain atrophy. Impaired episodic memory was observed 10 years before the expected clinical diagnosis, and declines in MMSE and the CDRE rating scale were detected 5 years before the expected clinical diagnosis.

Comparable findings were reported by Villemagne et al. [163], who estimated that it took 19.2 years of linear Abeta accumulation, 4.2 years of hippocampal atrophy, and 3.3 years of memory impairment to reach AD clinical diagnostic levels. Seppala et al. [145] correlated CSF findings with cortical biopsy analysis and found that patients with Abeta cortical plaques in biopsy samples had lower Abeta42 CSF levels than those without plaques. Prestia et al. [130], following patients with MCI, found that conversions to dementia increased as patients went from Abeta42 in the CSF, to Abeta42 plus FDG-PET, to Abeta42 plus FDG-PET plus hippocampal atrophy.

Okonko et al. [125] found that abnormal Abeta42, but not tau alterations, were associated with increased risk of AD, and Buchhave et al. [23] reported similar results in a study of patients with MCI followed for a median of 9.2 years. They concluded that 90 % of patients with MCI and pathologic CSF biomarkers develop AD within 9–10 years, and that Abeta42 is fully decreased 5–10 years before appearance of dementia. Shaw et al. measured CSF biomarkers in mild AD and MCI patients compared with controls, as well as autopsy confirmed cases compared with controls. They concluded that Abeta42 plus total tau predicted conversion of MCI to AD and that Abeta42 was the most sensitive marker in the autopsy cases [144]. Similarly, Visser et al. [164], in the DESCRIPA study involving a prospective cohort, found that patients with an AD profile in their CSF were prone to advance from MCI to AD type dementia.

Zetterberg et al. [183] found that CSF biomarkers remained stable over a 2-year period even with cognitive decline. Mattson et al. also found CSF biomarkers to be stable over a 4-year period in MCI patients. They reached a similar conclusion that a CSF profile shifting toward normal would be useful in tracking disease-modifying drugs [110].

As for Abeta deposits in the brain, Vlassenko et al. found that scans with Pittsburgh compound B about 2 years apart in cognitively normal adults showed that those with elevated binding showed enhancement of binding, indicative of increased brain Abeta deposits. They concluded that a major growth in the Abeta burden occurs during a preclinical stage of AD [166]. Bruck et al. [22] compared the prognostic value of Pittsburgh compound B (PIB)-PET, FDG-PET and hippocampal volume MRI for their prognostic value in predicting conversion of MCI to AD. Of the 29 patients, 17 converted to AD after 2 years. They concluded that the PET methods were superior to hippocampal volume methods in predicting the conversion. Hatashita and Yamasaki [59] followed 68 MCI patients by PIB-PET and FDG-PET. Over 19 months, 44 % of patients converted to AD. They found PIB-PET as the most definitive marker of MCI. Jack et al. [75] have proposed a model in which Abeta biomarkers become abnormal first, with neurodegenerative biomarkers becoming abnormal later, correlating with clinical symptom severity.

Despite the biomarker data, Braak et al. [17] have proposed, on the basis that tau aggregations in subcortical and paleocortical nuclei precede Abeta aggregations, that such tau aggregations must induce A-beta aggregations. But, as previously noted, this predicts that at least some tau mutations promoting tau aggregation should cause AD, which they do not. A possible explanation is that the minimal sub-cortical tau aggregations reported by Braak et al. are not decisive in influencing Abeta depositions in the neocortex. It is these much more widespread neocortical Abeta depositions that correlate with biomarker studies indicative of disease onset.

Figure 2 illustrates this principle. It shows the relative development of Abeta and tau deposits in the temporal cortex of three cases from our Kinsmen Laboratory brain bank. They are at different phases of AD progression. Figure 2a, b are immunostained temporal cortex tissue from an 88-year-old lady who died suddenly at home. There are extensive Abeta deposits, but a total absence of tau deposits, indicative of very early disease onset. Figure 2c, d are from the temporal cortex of a 72-year-old male who was admitted to hospital with mild disorientation and a vaguely documented history of some decline in cognitive function. He died in hospital of undetermined causes. They show extensive Abeta deposits as well as some tau deposits, consistent with early cognitive decline. Figure 2e, f are from the temporal cortex of an 87-year-old lady who had suffered from Alzheimer disease for at least 8 years. They show extensive Abeta deposits, and, in addition, very extensive tau deposits, consistent with advanced AD.

Fig. 2
figure 2

Comparative temporal cortex immunostaining for Abeta (6F3D) and tau (tau 0024) at various stages of AD development. a Extensive Abeta deposits in the temporal cortex of an 88-year-old lady who died suddenly at home. b The same area shows a total absence of tau deposits. This corresponds to the expected pathology in phase 1 and early phase 2 of disease development. c Extensive Abeta deposits in the temporal cortex of a 72-year-old male who was admitted to hospital with mild disorientation and a vaguely documented history of some decline in cognitive function. He died in hospital of undetermined causes. d The development of early tangles in the same area of the temporal cortex. This corresponds to the expected pathology in phase 4 of disease development. e Extensive Abeta deposits in the temporal cortex of an 87-year-old lady who had suffered from Alzheimer disease for at least 8 years. f The same area with very extensive tangle development. This corresponds to the expected pathology in phase 6 of disease development

Full size image

In summary, these data indicate that an extended window of opportunity exists for appropriate AD therapy to ameliorate, or even prevent disease development. They establish that disease onset can be detected by a reduction of Abeta42 in the CSF at least ten, and possibly 20 years prior to clinical diagnosis. This is followed by an increase in tau species in the CSF. Progression can be measured by PET and MRI scanning, all of which become abnormal years before AD can be diagnosed clinically.

Therapeutic implications

Table 4 is a theoretical construct of the phases of AD disease development and the implications for therapeutic intervention. Since the measurable prevalence of clinical AD commences at age 65 [79], the measurable prevalence for disease onset can be hypothesized to occur approximately 10 years earlier.

Table 4 Biomarkers and Phases of Alzheimer disease
Full size table

On this basis, the risk of Phase 1, marking disease onset, can be hypothesized to commence at age 55. Thereafter it will double every 5 years [79]. It is characterized by the commencement of Abeta deposition in brain with consequent decreases in the CSF. Therapeutic opportunities are at their highest. Any strategy which limits Abeta production, enhances its clearance, or prevents its aggregation should be disease modifying. Effectiveness of treatment should be measurable by CSF Abeta levels returning towards normal. The current problem is that there are no simple ways of making the diagnosis of wild-type AD at this very early stage.

Biomarker studies suggest that phase 2 sets in about 5 years later. Abeta deposits in the brain have built up to the level where they can be detected by PET scanning with Pittsburgh compound B. They have continued to show decreases in the CSF while tau levels in the CSF have now increased. Therapeutic opportunities have declined because cortical tau aggregation in brain has been induced. Ideally, a tau aggregation inhibitor should now be added to any therapeutic regimen, but so far none have been produced. Given that tau aggregation is the central pathogenic event in more than two dozen neurodegenerative diseases, developing such an agent is an urgent priority. Numerous proposals have been made [149], and at least two are in clinical trials [122]. The next best method of intervention is to administer antiinflammatory agents. NSAIDs are the most widely utilized, but they are only partial inhibitors and their effectiveness will wane as plaque and tangle levels in the brain increase.

In phase 3, typically in another 5 years, there is a slight metabolic decline that can be demonstrated by a reduction in FDG uptake by PET scanning. Presumably this is due to synaptic loss. Cortical tangle and thread development has increased to Braak stage III. Abeta and tau continue to be expressed at the same levels in the CSF. Pittsburgh compound B scanning intensity has increased. Therapeutic opportunities have further declined. Higher doses of therapeutic agents may be required to arrest disease development or to provide significant disease modification.

Phase 4, in another 5 years, represents disease advance to the level where cognitive deficits are detectable as mild cognitive impairment (MCI). Hippocampal atrophy has become evident by MRI scanning. The disease can no longer be totally prevented. Irreversible brain damage has begun to occur. Abeta and tau continue to be expressed at the same levels in the CSF but Braak stage IV pathology has been reached in brain. Nevertheless, substantial slowing, or perhaps even arrest, could be achieved by anti-Abeta agents combined with antiinflammatory agents such as NSAIDs.

Phase 5, in another 5 years, represents the level where AD can be diagnosed clinically. Neuronal damage and loss is evidenced by further PET-FDG metabolic decline and increasing brain volume loss by MRI. Abeta deposits in brain continue to accumulate, and Braak stage V levels of NFTs and NTs have been reached. Opportunities for therapeutic intervention have seriously declined.

Phase 6 defines the period of progressive clinical decline in all aspects of the disease. Cognitive deficits progress from mild to severe. Full-time care of patients becomes necessary. Costs associated with their care escalate. Therapeutic opportunities are minimal and yet it is in this phase that most clinical trials have been conducted.

To be a highly successful disease-modifying agent, a therapeutic candidate needs to deter Abeta42 aggregation, or intervene in its consequences, prior to the time when tau aggregation takes hold. For example, early application of direct inhibitors of Abeta aggregation, or inhibitors of β-secretase, or γ-secretase might succeed. Active immunization is highly questionable, given the dangers of inducing an autoimmune disorder and stimulating self-attack by complement. Passive immunization is also unlikely to succeed, due to the problems of blood–brain barrier penetration. Several failures have already been demonstrated.

NSAIDs offer the best current opportunity for disease modification since they have the capacity to combat the inflammation caused by both Abeta and tau aggregation. But they will need to be administered at higher doses and earlier in disease development than in previous trials. The effectiveness in each subject of such trials could theoretically be monitored by a shift in CSF Abeta and tau levels towards normal.

The difficulty in starting early treatment is in recognizing when disease onset has commenced. Obviously widespread screening of CSF is impractical. Screening for Abeta levels in blood is more practical but much less certain [157]. Screening saliva for Abeta42 levels is a possibility for the future. It has the advantage of being completely non-invasive. There is one report that Abeta42 saliva levels are increased above normal in AD cases [14]. We have been able to confirm this finding in a very small pilot study (unpublished data). Much more investigation will be required to determine if this is a possibility.

Dietary factors

The value of knowledge concerning dietary factors is that they are self-protective and dietary choices can be made long before the age of AD risk commences. The Mediterranean diet is the most widely studied. In a comprehensive review, Lourida et al. [105] reported that nine out of twelve epidemiological studies reported a lower risk of AD amongst those adhering to this diet. Laitenen et al. [92] found that a moderate intake of polyunsaturates decreased the risk of dementia, whereas a moderate intake of saturated fat may increase the risk.

Caffeine reduced plaque development and improved memory in two AD transgenic mouse studies [9, 28]. In the CAIDE study, Eskilinen and Kivipelto [40] found that drinking 3–5 cups of coffee per day at midlife was associated with a decreased risk of dementia/AD by about 65 % at late life. Maia and de Mendonca [107], in a case–control study, found that high caffeine intake was associated with a significantly lower risk of AD (odds ratio 0.40). However, Gelber et al. [46], in the Honolulu-Asia Aging study, found no overall effect of coffee, except for fewer lesions in men in the highest quartile of caffeine intake compared with the lowest.

Ballon et al. [11] analysed four studies for a specific association between AD and serum vitamin D levels. They concluded that those with levels of 50 nmol or higher were at lower risk. Vitamin D may have a mild protective effect.

As for antioxidants, Butterfield et al. [25] pointed out long ago that oxidative damage is characteristic of AD brain and that antioxidant therapy is a promising area of research. Engelhart et al. [39] reported that in the Rotterdam cohort of 5,395 participants, high intake of vitamin C and vitamin E was associated with a lower risk of AD (OR for each 0.82). Li et al. [100] did a meta-analysis of seven studies on the intake of vitamins E, C, and β-carotene, and the risk of AD. They concluded that each antioxidant had a mild protective effect with vitamin E being the most effective.

Summary

It is more than 20 years since the amyloid cascade hypothesis was first put forward. Yet it has not gained universal support because of the many unsuccessful attempts to develop disease-modifying treatments. One problem has been a failure to recognize that it is the inflammatory response to Abeta deposits, and later cortical tau aggregation, which drives the pathology. NSAIDs in general, and ibuprofen in particular, provide the best practical opportunity for therapeutic exploitation of this principle. Ibuprofen and related NSAIDs, which are widely used for the relatively trivial condition of osteoarthritis, have been shown to reduce the risk of AD in multiple epidemiological studies. They have also been shown to reduce the pathology in even more numerous transgenic mouse models of AD. Early administration may be the key to success. Such agents are relatively safe and highly available.

Recent biomarker studies indicate that disease onset commences as much as 10–15 years before clinical symptoms become manifest. If simple methods of preclinical diagnosis can be achieved, a 10-year window may exist for measures to be instituted that prevent, delay, or ameliorate the disease.

Search strategy and selection criteria. References for this Review were identified by searches of PubMed between 1969 and June, 2013. Among the search terms used were “Alzheimer disease and amyloid beta protein, tau, inflammation, complement, epidemiology, transgenic mice, clinical trials, and biomarkers”. There were no language restrictions. The final reference list, which could not include all possible contributions, was generated on the basis of highest relevance to the topics covered in this Review.