Testing the oligomer hypothesis is certainly an important goal for the field, however, from a broader perspective, the ACH will continue to be debated. One might pose the question as to whether any future negative trial would nullify the ACH. As detailed elsewhere in this manuscript, the various therapeutic approaches generally considered relevant to the ACH operate by mechanisms that have clear differences. Small molecule inhibitors of gamma-secretase and BACE were unsuccessful, and in fact, caused slight cognitive worsening. Monoclonal antibodies have generated more promising results in certain cohorts, but even these therapeutics have important differences and should not be considered a monolithic class. For example, solanezumab targets monomers rather than plaque, does not reduce plaque and does not cause ARIA. Aducanumab and donanemab target and reduce amyloid plaques based on PET imaging, however, they cause ARIA as a result. Lecanemab may target protofibrils preferentially compared with monomers, however, it also exhibits appreciable binding to amyloid fibrils (Englund et al., 2007; Lord et al., 2009; Magnusson et al., 2013). As expected, plaque reduction and ARIA are observed, though existing data indicate that the ARIA incidence may be less than that observed for aducanumab and donanemab treatments (Logovinsky et al., 2016). As discussed in the section “ACU193: AD Immunotherapy that Selectively Targets Amyloid β Oligomers,” plaque binding and ARIA are not expected for ACU193, based on its binding selectivity for AβOs; however, clinical data confirming this expectation are not yet available. Thus, it is important to recognize important differences in mechanism of action of the varied therapeutics presumed to test ACH. Over time, failed clinical trials may result in formulation of new or modified hypotheses, which eventually become accepted when supporting clinical data emerge.

An important advance in a number of therapeutic antibody trials has been the inclusion of individuals assessed to have early AD i.e. mild cognitive impairment (MCI) or mild dementia due to AD. In the past, some have speculated that any amyloid-related treatment would need to be given as secondary prevention to individuals who are cognitively normal but harbor amyloid plaques. Such studies include public–private partnerships such as the Anti-Amyloid Treatment in Asymptomatic Alzheimer’s (A4) study and the AHEAD 3-45 study. By definition, these studies are conducted with cognitively normal individuals and require very sensitive cognitive measure. Trial design parameters for these secondary prevention studies are not particularly well understood, so while prevention may be a worthy goal, negative trial results may emerge for a therapeutic that could be efficacious in a symptomatic population. The recent emergence of promising efficacy signals in certain early AD clinical cohorts receiving high doses of aducanumab, donanemab (Mintun et al., 2021), and lecanemab (Siemers et al., 2022) is an encouraging development.

The near-complete failure of anti-plaque therapeutics suggests that amyloid plaques and deposits are not the ideal target for AD therapeutics. Critics of the ACH have argued that deposition of Aβ protein is not the cause of AD (Lee et al., 2004; Teich and Arancio, 2012; Mullane and Williams, 2013, 2018; Morris et al., 2014, 2018; Behl, 2017; Makin, 2018; Panza et al., 2019; Høilund-Carlsen et al., 2020; Tolar et al., 2020). To be fair, however, the ACH relied heavily on compelling and irrefutable evidence that early-onset AD-causative mutations increased production of Aβ42, the most abundant plaque protein (Hardy and Allsop, 1991; Hardy and Higgins, 1992). The obvious inference made by Hardy was that elevated Aβ42 monomer deposited into plaques, which initiated the damage that occurs in AD. What Hardy and colleagues could not have known or anticipated in the early 1990s, was that Aβ42 monomer was capable of self-assembly into soluble, non-fibrillar neurotoxic oligomers possessing the ability to interfere directly with synaptic function and learning and memory.

That scenario came to light only a few years after the ACH was published, when Oda et al. (1995) observed soluble, neurotoxic Aβ42 structures in incubations of Aβ42 with small amounts of apoJ. Aβ42 fibril formation was unexpectedly inhibited, and instead, highly neurotoxic supernatant solutions were formed. Lambert et al. (1998) carried out extensive follow-up studies involving atomic force microscopy (AFM) and gel analysis to characterize soluble non-fibrillar oligomeric structures, ranging in size from trimer to dodecamer (Krafft et al., 1998). The soluble, globular oligomers exhibited potent neurotoxicity in hippocampal brain slice cultures, and at lower concentrations, the ability to block long-term potentiation (LTP) when added to organotypic brain slice cultures or injected into anesthetized rats. These neurotoxic oligomers were referred to as amyloid β-derived diffusible ligands (ADDLs). Lambert et al. (1998). wrote: “We hypothesize that impaired synaptic plasticity and associated memory dysfunction during early stage Alzheimer’s disease and severe cellular degeneration and dementia during end stage could be caused by the biphasic impact of Aβ-derived diffusible ligands acting upon particular neural signal transduction pathways.”

Further studies demonstrated the ability to generate rabbit polyclonal antibodies exhibiting > 80:1 AβO:monomer selectivity, and the polyclonal serum was used to detect AβOs in homogenates from AD frontal cortex and temporal cortex, but not in cerebellar homogenates (Lambert et al., 2001). The polyclonal serum also exhibited dose-dependent blockage of AβO toxicity in a PC12/MTT assay, a rudimentary demonstration of an AβO-selective immunotherapeutic approach.

In 2001, Klein et al. reiterated the AβO hypothesis and argued the case for targeting Aβ oligomers. In 2002, Kirkitadze et al. recognized the emergence of AβOs as a clear paradigm shift, and Walsh et al. (2002a) emphasized the potential importance of AβOs as a therapeutic target. In the subsequent two decades, many studies across multiple disciplines have gathered extensive data that implicate AβOs as the causative AD structures, as discussed in a number of comprehensive reviews (Klein, 2002, 2006, 2013; Walsh and Selkoe, 2004, 2007; Viola et al., 2008; Krafft and Klein, 2010; Ondrejcak et al., 2010; Ferreira and Klein, 2011; Hayden and Teplow, 2013; Ferreira et al., 2015; Viola and Klein, 2015; Cline et al., 2018; Mroczko et al., 2018; Li and Selkoe, 2020; Tolar et al., 2021).

Several distinct AβO activities and characteristics are particularly noteworthy. AβOs bind with high affinity to a subset of hippocampal and cortical neurons (Lacor et al., 2004, 2007; Barghorn et al., 2005; Koffie et al., 2009; Shughrue et al., 2010; Pickett et al., 2016; Wang et al., 2017), indicative of specific binding to discrete cell surface receptors that are developmentally regulated. In rodent hippocampal slice preparations, AβOs rapidly inhibit LTP (Lambert et al., 1998; Walsh et al., 2002b; Wang et al., 2002; Rowan et al., 2004; Barghorn et al., 2005; Rammes et al., 2011). Direct injection of AβO solutions into the rodent brain leads to reversible impairment of cognitive function (Cleary et al., 2005; Townsend et al., 2006; Poling et al., 2008; Reed et al., 2011) and injection of AβO solutions into the lateral cerebral ventricle of non-human primates leads to formation of neurofibrillary tangles (Forny-Germano et al., 2014) and synapse loss (Batista et al., 2018). Recently, ultra-high resolution microscopy revealed that AβOs do not bind directly at the synaptic cleft, but instead, form distinct nanoscale clusters encircling the postsynaptic membrane (Actor-Engel et al., 2021). Significant binding was also observed at presynaptic axon terminals. This study further demonstrated that LTP was impaired only at AβO-targeted synapses and not at neighboring AβO-free synapses. Importantly, aberrant AβO signaling activates GSK3β-mediated hyperphosphorylation of tau (De Felice et al., 2008; Zempel et al., 2010; Ochalek et al., 2017), which eventually causes tau aggregation, paired helical filament formation, cytoskeletal collapse, and neurodegeneration. Because AβOs appear to act upstream of tau abnormalities and inhibit LTP rapidly after binding, therapies that target aggregated, hyperphosphorylated tau or GSK3β would not be expected to rescue neurons from early LTP compromise that directly and immediately compromises information storage, well before any tau dysfunction occurs.

Many studies have identified interactions between AβOs and a variety of cellular proteins, however, the identity of the receptor(s) that mediate peri-synaptic AβO binding, aberrant AβO signaling and disrupted synaptic plasticity remains to be established (Laurén et al., 2009; Benilova and De Strooper, 2013; Xia et al., 2016; Ding et al., 2019; Smith et al., 2019; Guo et al., 2020; Lao et al., 2021). The review by Mroczko et al. (2018) provides a succinct summary of many cell surface molecules for which AβOs have apparent affinity. The principal conundrum for many of the putative receptor candidates is a relatively high measured equilibrium dissociation constant (Kd) for the binding interaction between AβOs and the putative receptor.

Amyloid β oligomers are widely reported to be the most neurotoxic form of Aβ, capable of triggering aberrant neuronal signaling which leads to loss of synapses and progressive neurodegeneration (Gong et al., 2003; De Felice et al., 2008; Krafft and Klein, 2010; Ondrejcak et al., 2010; Zempel et al., 2010; Klyubin et al., 2012; Koffie et al., 2012; Hayden and Teplow, 2013; Hefti et al., 2013; Fowler et al., 2014; Goure et al., 2014; Batista et al., 2018; Hong et al., 2018). Wang et al. (2018) described the dysregulation of calcium homeostasis in cultured neurons by 150–180 pM AβOs, the most potent AβO signaling yet reported. Size exclusion chromatography (SEC) analysis of these AβO solutions revealed a predominant peak eluting shortly before the 43 kDa standard, likely corresponding to the 56 kDa dodecamer identified in human AD brain extracts by 2D gel analysis, and frequently observed in synthetic AβO preparations (Gong et al., 2003; Bernstein et al., 2009; Ahmed et al., 2010; Inayathullah and Teplow, 2011; Bisceglia et al., 2018). Lesné et al. (2006) also isolated stable dodecamer (referred to as Aβ*56) from transgenic AD mouse brain extracts, and showed that its appearance correlates closely with cognitive deficits in three different lines of transgenic AD mice (Billings et al., 2007; Cheng et al., 2007).

The compiled experimental evidence implicating AβOs as the cause of AD memory malfunction and neurodegeneration is strong and continues to gain support. Perhaps the strongest support comes from discovery of an inherited APP mutation (E693Δ), known as the Osaka mutation (Tomiyama et al., 2008; Inayathullah and Teplow, 2011), which results in production of an Aβ peptide variant (Aβ E22Δ) lacking glutamic acid residue 22. This Aβ variant possesses the surprising properties of enhanced oligomerization and no fibrillization. Even though the APP E693Δ mutation reduces Aβ secretion by 38%, the enhanced oligomerization of Aβ E22Δ brings about memory loss in affected family members at ages as young as 35 (Tomiyama and Shimada, 2020). PET imaging of Osaka carriers shows very little deposited amyloid, while post-mortem characterization shows extensive tau pathology and neurofibrillary tangles, with few or no amyloid plaques. In other words, AD without amyloid plaques, coupled with an oligomerization-prone Aβ variant, plainly implicates AβOs as the relevant AD neurotoxins.

ACU193 exhibits significant preferential binding to AβOs over Aβ monomers. In a competition ELISA assay (Figure 3A), ACU193 binds to AβOs with 650-fold greater affinity than to Aβ monomers (Savage et al., 2014; Goure et al., 2016). Because this value does not account for the larger average size of synthetic AβOs used in these experiments, on a molar basis, the selectivity may be as high as 10,000-fold. Further evidence of ACU193 selectivity for AβOs was obtained in a cell-based AβO binding assay in which a very high concentration (5 μM) of monomeric Aβ did not decrease the ACU193 binding affinity for AβOs (Figure 3B). ACU193’s selectivity for AβOs in the presence of abundant Aβ monomers is expected to be representative of the in vivo levels of these Aβ species in AD patients. Thus, ACU193 does not experience “target distraction” from non-toxic Aβ monomers in an environment simulating brain interstitial fluid.

ACU193 binds AβOs from CSF and brain tissue extracts of AD patients, and it exhibits minimal or no binding to amyloid plaques, as illustrated in Figure 4. Amyloid plaques stained with thioflavin S appear fluorescent green, while the binding of ACU193 appears as red fluorescence (Cline et al., 2019). Substantial ACU193 binding can be observed in regions that are without thioflavin S-stained amyloid plaques (Figures 4B,E), with minimal binding in regions exhibiting thioflavin-S-positive fibrillar Aβ structures (Figure 4D). Close examination does reveal occasional co-localization of ACU193 at the periphery of some amyloid plaques (Figure 4F). It is likely that this ACU193 localization reflects binding to AβOs associated with plaque surfaces (Gaspar et al., 2016). Taken together, these results are consistent with the concept that ACU193 binds endogenous AβOs, does not prevent thioflavin-S binding to plaques, and importantly, preferentially binds AβOs versus plaque-associated Aβ.

ACU193 exhibited no binding to amyloid deposits surrounding blood vessels (cerebral amyloid angiopathy) in AD brain slices, nor did it bind to vascular amyloid when injected into transgenic mice, as illustrated in Figure 5. On the other hand, injected bapineuzumab bound to vascular amyloid at all doses tested (Figures 5E,F).

In view of the linkage between the amyloid plaque-binding properties of multiple amyloid-directed antibodies and the frequent occurrence of ARIA (e.g., aducanumab, gantenerumab, lecanemab, and donanemab), the negligible binding of ACU193 to amyloid plaques, including amyloid plaques associated with cerebral amyloid angiopathy, suggests that ACU193 treatment will not elicit ARIA side effects.

ACU193 binds a broad spectrum of AβOs across various molecular weights. AβOs prepared from synthetic Aβ42 were fractionated by SEC (Figure 6) and characterized by ELISA using ACU193, hu3D6 (bapineuzumab) or hu266 (solanezumab) as the capture antibody and biotinylated anti-human Aβ antibody 82E1 for detection (Goure et al., 2015). These data show ACU193 binds mid- to higher molecular weight AβOs, with preferential binding to mid-molecular weight oligomers compared to hu266. This range of molecular weights includes dodecameric AβOs that are thought to be physiologically relevant (Gong et al., 2003; Lesné et al., 2006).

Collectively, the data show that ACU193 binds AβOs with 650-fold selectivity versus Aβ monomers (perhaps > 10,000-fold on molar basis), exhibits limited to no plaque binding, and binds to a broad range of synthetic and endogenous low, mid, and higher molecular weight AβOs. Based on these and other data, ACU193 is expected to target therapeutically relevant AβOs effectively within the brain in early AD patients.

In ex vivo studies using the murine hippocampal slice LTP model (Rammes et al., 2011), pre-incubation with 100 pM ACU193 or ACU3B3 (murine precursor of ACU193) can prevent AβO-induced LTP blockage, as illustrated in Figure 7 (Cline et al., 2019). As discussed earlier in section “Two Decades of Profiling Bolsters the Amyloid β Oligomer Hypothesis,” AβO disruption of LTP has been widely documented. LTP blockage prevents synaptic strengthening that is essential for learning and memory, and the ability of ACU193 to prevent AβO-LTP disruption may be a key property for preventing further memory deterioration in early AD patients.

Exposure to ACU3B3 prevents AβO-induced calcium overload in cortical neuronal cultures (Figure 8; Wang et al., 2018). Disruptions in calcium homeostasis that cause cellular dysfunction have been implicated in a number of disease states, including myocardial infarction and stroke. Further, AβOs have been shown to cause disruption of calcium homeostasis, and thus, restoration of intracellular calcium to normal levels could serve as a functional indicator of treatment effect in AD. Multiphoton microscopy was used to examine the relationship of AβOs and neuronal calcium homeostasis in vitro (Figure 8). Direct application of AβOs elicited calcium elevations in cortical neuronal cultures. Prior exposure to ACU3B3 prevented this calcium elevation (Figure 8). These results demonstrate that AβOs induce elevated concentrations of intracellular neuronal calcium and that ACU3B3 prevented the AβO-induced calcium overload.

In a blinded study (Figure 9) conducted by QPS Austria GmbH, plaque bearing 9- to 10-month-old, female, APP/SL transgenic (Havas et al., 2011; Faizi et al., 2012) and non-transgenic mice were administered ACU3B3 (20 mg/kg) or vehicle (PBS) by intraperitoneal (IP) injection once a week for 4 weeks. The last injection was given 24 h before the first water maze training trial. Prior to treatment, transgenic mice demonstrated behavioral deficits compared to non-transgenic mice, but as shown in Figure 9B, significantly improved their performance across testing sessions. On the first training day, ACU3B3 treatment improved swim path length in transgenic mice (Figure 9A), and this effect was significant (p < 0.05) after exclusion of outliers. ACU3B3 also improved swim path length performance in session 3 trials (Figure 9B), and restored swim speed in transgenic mice (p < 0.02) to the same level as in wild-type mice (Figure 9C).

In another blinded study (Figure 10) conducted at Stanford University (Dodart et al., 2014), the hyperactivity phenotype of 5- to 7-month-old Thy1-hAPP/SL transgenic mice in the Open Field and Y-Maze tests was also significantly reduced after 4–5 weeks of ACU3B3 treatment (20 and 30 mg/kg, weekly). Prior to dosing, Thy1-hAPP/SL mice showed increased activity in the activity chamber compared to wild-type mice. After ACU3B3 treatment, Thy1-hAPP/SL mice activity fell to a level comparable to wild-type mice, particularly activity in the center of the test arena (Figure 10A). Similar effects of ACU3B3 were found with changes in Y-maze behavior (Figure 10B) and passive avoidance (Figure 10C).

Additional studies were conducted at the Gladstone Institute of Neurological Disease in young hAPP/J20 mice, which express APP carrying the Swedish and Indiana mutations (Mucke et al., 2000). Animals, ages 2–5 months were dosed with ACU3B3 to evaluate its efficacy prior to detectable amyloid deposition. Treatment intervention preceded the appearance of measurable levels of amyloid plaques as well as the progressive decrease in synaptic markers, microglia activation and tau hyperphosphorylation (Shankar and Walsh, 2009) but coincided with the appearance of measurable levels of AβOs (Johnson et al., 2020). ACU3B3 significantly improved behavioral deficits measured in Open Field, Y-Maze, Plus Maze, and Morris Water Maze tests (Cline et al., 2019; Ma et al., 2019), supporting the AβO selectivity of ACU3B3 and its potential for intervening prior to a stage where synaptic loss and the underlying behavioral impairment and memory loss becomes irreversible.

Taken together, the extent of behavioral studies conducted in two different APP transgenic lines with different AD phenotypes and at different stages of disease progression (notably pre- and post-plaque accumulation) provides a comprehensive understanding of the pharmacological activity across age groups. The beneficial behavioral effects observed in animals treated with ACU3B3 indicate that the antibody effectively crosses the blood–brain barrier to engage target AβOs. The dose range selected in the behavioral studies is also representative of the dose range evaluated in the ACU193 phase 1 clinical trial.