It is known that Aβ plaques occur earliest in neocortex in cognitively normal subjects where they are relatively evenly distributed [17]. Neurofibrillary tangles appear first in limbic areas such as the transentorhinal cortex and progress in a predictable pattern of regional distribution to the neocortex [18]. Arnold and coworkers mapped the distribution of NFTs and neuritic plaques (Aβ plaques surrounded by dystrophic neurites) in the brains of patients with AD [19]. Compared to NFTs, neuritic plaques were, in general, more evenly distributed throughout the cortex, with the exceptions of notably fewer neuritic plaques in limbic periallocortex and allocortex (the areas with greatest NFT density such as the hippocampus). This is supported by the findings of Price and Morris who found little Aβ pathology in the hippocampus of elderly controls or very mild AD patients [12]. Thus, while limbic areas have early and severe NFT pathology, the mesial temporal lobe has relatively little neuritic plaque pathology early in the course of AD. The cerebellum is notably free of neuritic plaques in AD (Fig 3), although diffuse Aβ deposits which do not label with fibrillar dyes, such as Congo red and PiB, are commonly observed [10,20,21]. Thus from an AD diagnostic perspective, one would look for the presence of Aβ deposits or the binding of an Aβ imaging agent in specific brain regions, and the absence of binding in other regions. However, the presence of significant Aβ deposits in specific brain regions does not necessarily indicate the presence or severity of clinical AD, as some control and MCI subjects contain concentrations of Aβ plaques in their brains as high as contained in the brains of AD subjects (Fig 5) [14,22,23]. The coincidence of high Aβ loads and the absence or lack of extensive cognitive deficits in some subjects could be interpreted as high resistance to Aβ damage (i.e., a high cognitive reserve) or could indicate that excessive Aβ has not been present for a sufficiently long time in some subjects to cause as much damage as in AD subjects. Comprehensive longitudinal studies using Aβ imaging agents likely will help sort out these possibilities that single time point cross-sectional studies are not able to distinguish.

A second and closely related potential use of Aβ imaging agents is to help test the “amyloid cascade hypothesis” of AD [24,25,26]. A growing consensus points to the overabundance of extracellular Aβ protein in the human brain as the central event in the pathogenesis of AD [27]. The amyloid cascade hypothesis holds that the overproduction of Aβ, or the failure to clear this protein from the brain, leads to AD (Fig 6). Proponents of this hypothesis posit that high levels of Aβ subsequently lead to a series of downstream neuropathological events, including the production of extensive intracellular NFT deposits, inflammation, oxidative damage, excitotoxicity, loss of synaptic connections, neuronal cell death, and, eventually, clinical symptoms such as memory impairment and AD [28]. The single, most important piece of evidence supporting the amyloid cascade hypothesis of AD is the demonstration that many different mutations in the Aβ precursor protein (APP) gene on chromosome 21, all lying in or near the Aβ peptide region, cause early-onset or familial forms of AD [28,29,30]. Further genetic support for the amyloid cascade hypothesis comes from the finding that the most common form of early-onset, autosomal dominant familial AD – the chromosome 14 mutations – is caused by mutations in the presenilin-1 (PS1) gene, which codes for a protein that is strongly implicated to be an essential component of the “γ-secretase” enzyme complex responsible for C-terminal cleavage of Aβ from its precursor, APP (Fig 6) [31].

Studies in triple transgenic mice overexpressing both APP and tau have provided insights into the mechanistic relationship between Aβ and tau pathology in AD. These so-called 3xTg-AD mice bear both the Swedish (KM670/671NL) APP mutation and the PS1 (M146V) mutation leading to overproduction and deposition of Aβ as well as the human four-repeat tau (P301L) mutation leading to overproduction and deposition of tau [32,33]. Billings and coworkers [34] have summarized the neuropathological aspects of these mice that are reminiscent of AD: (1) Aβ plaques and neurofibrillary pathology develop in a hierarchical manner in AD-relevant brain regions, mainly the hippocampus, cortex, and amygdala; (2) Aβ plaque pathology precedes tangle formation, and plaques consist of the longer, more amyloidogenic Aβ42; (3) the pattern of conformational and phosphorylation changes that the tau protein undergoes parallels the sequence in the human AD brain; and (4) the 3xTg-AD mice show selective loss of nicotinic α7 receptors in the hippocampus and cortex. It also has been shown that both Aβ immunotherapy and γ-secretase treatment lead to clearance not only of Aβ, but also early-stage tau lesions [35]. Importantly, Aβ immunotherapy did not reverse late, well-established tau aggregation, pointing out the importance of preventing this pathology very early before synaptic and neuronal losses have occurred (see below). These and other studies [36,37] suggest that, at least in the mouse model, Aβ deposition precedes and exacerbates NFT formation.

While there is considerable agreement that Aβ plays a key causative role in at least early-onset, autosomal dominant forms of familial AD, there is disagreement about the molecular and cellular mechanisms through which Aβ exerts its pathophysiological effects. Butterfield and Bush have reviewed the evidence that factors such as oxygen, the single methionine-35 residue of Aβ42, and redox metal ions (Zn⁺² and Cu⁺²) are important for the oxidative stress and neurotoxic properties of Aβ (Butterfield 2004). Others have pointed out that the assembly of Aβ42 into toxic species is intrinsic to the primary structure of Aβ42 and does not require chemical modification of the peptide or the invocation of peptide-associated enzymatic activity [38]. This latter discussion has focused increasing attention on the important, and perhaps unique, role of small, soluble, oligomeric assemblies of Aβ42 in the cascade of synaptic dysfunction and neurotoxicity (Fig 6). That is, it may be the soluble Aβ42 oligomers, more than insoluble fibrillar deposits, that most contribute to the neurotoxic effects of high extracellular concentrations of Aβ. Oligomeric Aβ42 has been shown to be 10- to 100-fold more toxic than fibrillar and monomeric Aβ42 when incubated with Neuro-2A cells [38,39]. However, at 10 μM concentrations, all “starting” aggregation states of Aβ42 (i.e., monomeric, oligomeric, and fibrillar) produce equivalent toxicity suggesting either that aggregation state-dependent toxicity is a dose-related and not an absolute phenomenon or, more likely, that at higher concentrations an equilibrium is established under physiologic conditions that produces a spectrum of toxic species independent of the “starting” aggregation state.

Regardless of the equilibrium state under any given set of physiological conditions, all of the protein forms (monomer, soluble oligomer, and fibril) are separated by energy barriers that can be crossed in both directions. Studies of immunotherapies in transgenic mice [40,41] and autopsy studies of humans treated with active immunization in the AN-1792 trial [42,43,44], strongly suggest that Aβ in fibrillar deposits can be mobilized and cleared. The equilibrium-based nature of this clearance is perhaps best exemplified by the observation that immunization of Aβ-depositing transgenic mice in a manner that produces antibodies specific for oligomeric Aβ led to marked reduction of not only oligomeric forms of Aβ, but also resulted in clearance of thioflavin-S positive insoluble plaque forms of Aβ as well [45,46]. This most likely occurred by shifting the equilibrium away from the fibrillar species toward the oligomeric forms that could be cleared by the antibodies.

One should also consider that insoluble Aβ exceeds soluble forms of Aβ by a factor of about 100-fold in AD brain [47], so even a 100-fold greater toxic potency of the oligomeric form leaves questions about the relative toxicities of various aggregation states in vivo. Kuo and coworkers reported that the soluble pool of Aβ displays a continuous distribution of monomeric and oligomeric Aβ [47]. Centricon membrane molecular weight fractionation of the soluble pool showed that approximately 75% of the soluble Aβ pool in human brain was oligomeric. It also is notable that Kuo found the parenchymal levels of soluble oligomeric Aβ to be significantly higher (~50-fold) than CSF Aβ levels (and higher still than blood levels), suggesting that there is some barrier to free diffusion from the brain parenchyma to CSF (and blood). In summary, the evidence that oligomeric Aβ42 is the most toxic species of Aβ is convincing. Studies strongly support the following: 1) all else being equal, enriched oligomeric preparations of Aβ42 appear to be the most toxic species [38,39]; 2) oligomeric species of Aβ42 can inhibit long term potentiation (LTP) and lead to synaptic dysfunction [48,49]; 3) memory loss occurs prior to extensive deposition of fibrillar Aβ42 in transgenic mice and can be reversed by targeting soluble species of Aβ42 [50,51,52,53]; and 4) oligomeric Aβ42 species exist in AD brain [54]. However, it is important to appreciate that the removal or reduction of the soluble, toxic oligomeric forms of Aβ42 from brain parenchyma will require the concomitant reduction and removal of the insoluble (fibrillar) Aβ42 pool as well. Chemists can readily appreciate the analogous situation in the case of a mass of a slightly soluble precipitate at the bottom of a beaker covered by a saturated aqueous solution of the precipitated compound. One can remove the saturated solution and replace it with a fresh supply of water, but the water solution above the precipitate will become saturated again in short order (when equilibrium is reached). Only when the precipitate is completely removed, or dissolved, will the aqueous solution be able to contain lower quantities of the precipitate than contained in a saturated solution above the precipitate. Likewise if the concentration of the solute exceeds its solubility limit, new precipitate will form. The relationship between soluble and insoluble forms of Aβ in the extracellular space of parenchymal tissues may not be this direct, but likely is analogous; the permanent lowering of soluble Aβ in the extracellular solution will depend upon a proportional lowering of insoluble (or more correctly, slightly soluble) forms of Aβ (i.e., Aβ plaques) as well. These Aβ deposits provide for the storage of the vast majority of Aβ in AD brain (>99%), and only when they are removed will the concentration of the more toxic, soluble oligomeric forms of Aβ be permanently reduced. Therefore an implication of the amyloid cascade hypothesis is that an anti-amyloid therapy directed at removing insoluble Aβ stores will likely be effective in reducing soluble Aβ levels as well.

The third potential use of Aβ imaging agents would be to assist with the development of AD therapeutic agents, specifically with drugs aimed at halting or reversing the increased concentration of toxic soluble Aβ oligomers in brain and their subsequent deposition and storage in the forms of Aβ fibrils and plaques. Aβ imaging agents could be used to identify subjects who would benefit from anti-Aβ therapies, i.e., subjects with Aβ deposits in their brains, as well as to assess the efficacy of various treatment approaches in halting or reversing Aβ deposition in these subjects. It is not surprising that the metabolism of Aβ has become an important therapeutic target in AD research. A corollary of the amyloid cascade hypothesis is that prevention of Aβ accumulation in oligomers or plaques should prevent AD. Approaches to “anti-amyloid” therapy have focused on decreasing synthesis, increasing clearance, or decreasing the aggregation/toxicity of Aβ.

Because soluble forms of Aβ in AD brain comprise ~1% of the total Aβ (on a molar basis) [47] imaging agents targeting soluble Aβ will have a much lower target density (B_max) than agents targeting Aβ plaques. In addition, the development of imaging agents specific for soluble, oligomeric Aβ binding sites relative to insoluble Aβ binding sites could prove to be problematic, not only because of the much higher density of insoluble Aβ binding sites in AD brain, but also because of the likely difficulty of developing a small molecule imaging agent specific for binding only soluble Aβ oligomers in the presence of high concentrations of fibrillar, insoluble Aβ deposits. Fortunately, PiB (Fig 2) has been shown to bind specifically to Aβ40 and Aβ42 synthetic fibrils and insoluble Aβ plaques containing Aβ42 and Aβ40 found in AD brain. In contrast, PiB does not bind appreciably to soluble Aβ and probably does not bind to oligomeric forms of Aβ until they reach some critical size (yet to be determined). PiB binding requires an extended β-pleated sheet structure found in Aβ fibrils and plaques in order to bind with high affinity. Hence if one wishes to follow the amount of insoluble Aβ in AD brain tissue, a radiotracer such as [¹¹C]PiB with a high selectivity for insoluble Aβ likely would prove useful in vivo in human brain [16,22].

Studies in transgenic mice that overexpress mutant human APP have been interpreted to suggest that small soluble forms of Aβ are a critical toxic species. In Tg2576 mice, no obvious correspondence between memory and insoluble Aβ levels was apparent [53]. Furthermore, passive immunization studies of this same Tg2576 strain of mice showed that memory deficits and disruption of LTP can be reversed without affecting the levels of soluble or insoluble Aβ, suggesting “neutralization” of soluble Aβ by antibodies may be sufficient to aid in cognitive improvement [51,52]. These results were interpreted to imply that insoluble Aβ is a surrogate marker for “small assemblies” of Aβ that disrupt cognition and occur as intermediates during the formation of insoluble Aβ. This finding was extended to the PDAPP strain of mice by Dodart and colleagues [50]. This group previously reported that chronic treatment of PDAPP mice with the m266 anti-Aβ antibody (every 2 weeks from 4 to 9 months of age) reduced Aβ burden at least in part by increasing peripheral clearance [55]. In the subsequent study in 24 month-old PDAPP mice [50], they found that a “subchronic” six-week course of m266 immunotherapy reversed the cognitive deficits measured, but did not decrease the Aβ immunohistochemical burden (brain Aβ levels were not determined by ELISA). More surprisingly, they found that the cognitive deficits measured in 11 month-old PDAPP mice could be reversed within days of a single dose of the m266 anti-Aβ antibody. Dodart and coworkers did not determine if this acute improvement was transient or long-lasting. This experiment was subsequently performed by Billings et al. in the 3xTg-AD mouse [34]. At four months of age, intraneuronal soluble and insoluble Aβ accumulation appeared to lead to early cognitive deficits prior to extracellular plaque and tangle deposition in the 3xTg-AD mouse. Intracerebroventricular injection of an anti-Aβ antibody into 4 month-old 3xTg-AD mice cleared the intraneuronal Aβ and reversed the early cognitive deficit when tested 1 week later. However, 1 month after this single antibody treatment, intraneuronal Aβ pathology returned along with the early cognitive deficits typical of untreated mice [35]. This important experiment points to the transient nature of acute immunotherapy on early Aβ accumulation and the need to effect and maintain a stable shift in the equilibrium of Aβ accumulation. Taken together, these data suggest that immunotherapy leads to cognitive improvement in these mouse studies via actions primarily directed at soluble oligomeric forms of Aβ.

Several studies suggest that if anti-Aβ mediated Aβ clearance happens in the brain, it is not solely via a microglial Fc receptor-dependent mechanism. Administration of the F(ab′)₂ fragment of an anti-Aβ antibody (i.e., an Aβ binding agent which lacks the Fc region of the antibody) led to a reduction in brain Aβ to the same extent as the full anti-Aβ antibody [74]. This suggests that Aβ binding alone, not microglia-mediated Aβ clearance, is sufficient to reduce Aβ load, although this does not necessarily imply a peripheral effect. Solomon and coworkers have shown that antibodies to Aβ can bind to fibrils and cause disaggregation [75]. More directly, APP transgenic mice crossed with Fc receptor knockout mice responded to vaccination as well as mice with intact Fc receptors [76] showing that Fc receptor mediated phagocytosis was not necessary for Aβ clearance in this mouse. An alternative hypothesis suggests that antibodies present in the peripheral blood alter the central:peripheral Aβ equilibrium [55]. This “peripheral sink hypothesis” is supported by the finding of marked elevations in plasma Aβ levels after passive immunization [55] and active immunization [77].

One of the most encouraging findings from the AN-1792 active immunization trial was that the autopsy data strongly suggested that immunotherapy can have a profound Aβ-clearing effect in humans. Although this finding can be considered a proof-of-concept in many ways, enthusiasm must be tempered not only by the incidence of meningoencephalitis, but also by the relatively modest clinical effect of AN-1792. However, careful consideration of the state of the pathology when AN-1792 treatment was initiated may be instructive as to why this anti-amyloid therapy had such modest clinical effects. Figure 7 shows a small field-of-view from the frontal cortex of the AN-1792 autopsy case reported by Masliah and coworkers (Masliah 2005). The brain has been stained with X-34 (see Fig 1), a highly fluorescent Congo red derivative that stains β-sheet deposits including both fibrillar tau and Aβ pathology [89]. As confirmed by immunohistochemistry with anti-Aβ antibodies (data not shown), this area of the frontal cortex has been cleared of Aβ deposits, but extensive neuritic and neurofibrillary pathology remain untouched. The asterisk points to an area void of Aβ or neuritic pathology and has the appearance of an area that once may have been occupied by an Aβ plaque. This figure graphically points out how Aβ immunotherapy failed to affect tau pathology once it is well-established and the neuronal dysfunction associated with this tau pathology. Evidence from transgenic mouse models of plaque and tangle deposition suggest that plaque deposits precede and enhance tangle pathology at least in these animals [35]. This suggests that effective anti-amyloid therapy should be initiated early in the pathological process of AD in order to be optimally effective. Evidence suggests that Aβ pathology begins a decade or more before clinical symptoms are apparent [13]. Imaging technologies such as [¹¹C]PiB that can detect Aβ deposits are under development by several groups [16,90,91,92]. The exact point in the evolution of Aβ pathology at which these imaging technologies will become useful remains to be determined, but early evidence suggests that detection of extensive Aβ deposition is possible during the mild cognitive impairment (MCI) stage [22] and perhaps prior to the onset of symptoms (Fig 5) [14,23].

There are several implications of Figures 3, ​,5,5, and ​and77 with respect to the application of anti-Aβ therapies and the effective role of Aβ imaging agents in aiding drug development efforts. First, agents such as [¹¹C]PiB can identify subjects who would benefit from anti-Aβ therapies and those who would not. Clearly, AD patients (Figure 3 and ​and5)5) with heavy Aβ depositions are candidates for these therapies. However as shown in Figure 7, these treatments may be too late in advanced AD cases that have significant downstream damage from Aβ deposition such as extensive NFT deposits, dystrophic neurites, neuroinflammation, and synaptic and neuronal cell loses. Instead, MCI subjects with moderate to heavy Aβ loads (Fig 5) might be better candidates for these treatments. These subjects may not have the extensive synaptic and neuronal loses suffered by AD patients. In addition, MCI subjects with no evidence of Aβ deposition (Figs 3 and ​and5)5) would likely not be candidates for these therapies. These MCI subjects might be among the reported 30–40% of MCI subjects who do not progress on to develop AD [93]. Normal control subjects who have evidence of some Aβ deposition (Figs 3 and ​and5)5) also might benefit from anti-Aβ treatments prior to the onset of detectable cognitive symptoms. We have followed an elderly control subject with significant Aβ brain depositions for two years with [¹¹C]PiB imaging (Fig 8). This subject shows increased [¹¹C]PiB binding over two years greater than the test/retest variability in brain regions known to contain high concentrations of Aβ plaques in AD (e.g., precuneus, parietal cortex, frontal cortex, and striatum) and no increase in [¹¹C]PiB binding in brain regions known to contain low concentrations of Aβ plaques in AD (e.g., cerebellum, mesial temporal cortex, sensory motor strip, white matter). This elderly control subject would be an excellent candidate for anti-Aβ therapies at a stage where little neuronal damage is evident.

The second implication of Figures 3, ​,5,5, and ​and77 is that only subjects with detectable Aβ loads will demonstrate decreases in [¹¹C]PiB binding with anti-Aβ therapies. Elderly controls and MCI subjects with no evidence of Aβ deposition are unlikely to show changes in [¹¹C]PiB binding with anti-amyloid therapy. Finally, [¹¹C]PiB and related compounds should be able to detect decreases in insoluble Aβ load about 2-fold greater than the test/retest variability of the imaging method. For [¹¹C]PiB, the test/retest variability depends on the data analysis method and the region of interest [22], but is in the range of 5–10%. Thus 10–20% decreases in Aβ load should prove detectable with [¹¹C]PiB.

The type of anticipated effects of anti-Aβ treatment on Aβ load are shown in Figure 9. This figure shows the result of in vitro [³H]PiB binding assay in post mortem frontal cortex tissue collected from the same patient shown in Figure 7. The amount of [³H]PiB binding in the AD tissue sample is comparable to that found in the frontal cortex of normal control subjects and well below levels found in AD patients. The question is: what level of Aβ plaque deposition did this patient have in frontal cortex before AN-1792 treatments? With a single post mortem sample, one can not address this question. But with amyloid imaging at baseline and after therapy, this question can be addressed. Currently, [¹¹C]PiB is being used in anti-amyloid drug development trials in clinically diagnosed mild to moderate AD subjects to quantitatively assess changes in amyloid plaque densities throughout the brain over the course of the experimental treatments. Accurate pre-treatment, baseline measures of [¹¹C]PiB binding in these subjects will determine a reference point against to judge the effectiveness of the treatments in clearing amyloid plaque burden, as well as confirm the clinical diagnosis of apparent AD.