Abstract
Brain proteins function in their soluble, native conformation and cease to function when transformed into insoluble aggregates, also known as amyloids. Biophysically, the soluble-to-insoluble phase transformation represents a process of polymerization, similar to crystallization, dependent on such extrinsic factors as concentration, pH, and a nucleation surface. The resulting cross-β conformation of the insoluble amyloid is markedly stable, making it an unlikely source of toxicity. The spread of brain amyloidosis can be fully explained by mechanisms of spontaneous or catalyzed polymerization and phase transformation instead of active replication, which is an enzyme- and energy-requiring process dependent on a specific nucleic acid code for the transfer of biological information with high fidelity. Early neuronal toxicity in Alzheimer’s disease may therefore be mediated to a greater extent by a reduction in the pool of soluble, normal-functioning protein than its accumulation in the polymerized state. This alternative loss-of-function hypothesis of pathogenicity can be examined by assessing the clinical and neuroimaging effects of administering non-aggregating peptide analogs to replace soluble amyloid-β levels above the threshold below which neuronal toxicity may occur. Correcting the depletion of soluble amyloid-β, however, would only exemplify ‘rescue medicine.’ Precision medicine will necessitate identifying the pathogenic factors catalyzing the protein aggregation in each affected individual. Only then can we stratify patients for etiology-specific treatments and launch precision medicine for Alzheimer’s disease and other neurodegenerative disorders.
It is certain that brain peptides must be in their native configuration to carry out their normal function, which ceases once they misfold. It is also certain that the accumulation of amyloid-β (Aβ) in the brain precedes the development of dementia in Alzheimer’s disease (AD). What remains uncertain, however, is whether the Aβ aggregates themselves are necessary or sufficient for the subsequent events in the amyloid hypothesis, namely the increase in tau-filled neurofibrillary tangles (NFT) and neuronal degeneration. Is brain amyloidosis the cause of disease or one of its consequences, a pathophysiologic driver or a marker of biological distress? On the answer hinges whether future therapeutic efforts should remain focused on anti-Aβ aggregation or be reconfigured to correct the soluble Aβ being depleted in the aggregation process.
This Hypothesis article is aimed at providing a bird’s perspective to the frog’s in the ecological balance Freeman Dyson advocated for science [1].
Birds fly high in the air and survey broad vistas out to the far horizon. They delight in concepts that unify our thinking and bring together diverse problems from different parts of the landscape. Frogs live in the mud below and see only the flowers that grow nearby. They delight in the details of particular objects, and they solve problems one at a time ( . . . .). It is stupid to claim that birds are better than frogs because they see farther, or that frogs are better than birds because they see deeper. The world is both broad and deep, and we need birds and frogs working together to explore it.
Dyson could have used the following example. A storm ravages the landscape, littering it with broken, uprooted trees. A bird might ask what to do about the next storm to avoid damaging more trees. Without the input from a bird, a frog posits that removing the damaged trees will reverse the problem and prevent additional harm.
ANTI-AMYLOID TRIALS: A BIRD’S PERSPECTIVE
Admitting it as a bird’s oversimplification, the misfolding of proteins (the “damaged trees”), and their presumed replication and spread, are widely considered the pivotal steps in the pathogenesis of AD. The therapeutic pipeline for disease modification in AD has heavily focused on anti-Aβ aggregation (Table 1). Despite demonstration of target engagement, that is, Aβ reduction, in nearly 70% of the 41 trials reported to date, clinical outcomes have been null or truly negative, that is, yielding changes contrary to those hypothesized. In about 40% of these trials, the active arm has shown significant increase in cognitive worsening, brain atrophy, or both, compared to placebo.
The 40 Phase II/III randomized anti-amyloid clinical trials in Alzheimer’s disease
The 40 Phase II/III randomized anti-amyloid clinical trials in Alzheimer’s disease
Against the argument that these outcomes can in part be explained by an approach enacted “too late,” 8 of the trials have targeted prodromal (mild cognitive impairment; n = 5) or preclinical populations (amyloid-positive or mutation-positive individuals with normal cognition “at risk” for AD, n = 3). Cognitive decline and reduction in brain volume have been observed in the treated arm in half of these studies (Table 1).
Many lessons from past trial failures have been generated [2, 3]. Yet none of them have amounted to a rejection of the hypothesis that Aβ pathology is pathogenic in AD. In fact, a pooled analysis of 14 randomized controlled trials of anti-amyloid drugs for the prevention or treatment of AD showed that lowering brain Aβ had no significant effect on cognition, “within the timeframe of typical clinical trials” [4]. In other words, the results are not as expected because we are not observing patients long enough. Or, as with other criticisms, we are not concurrently using anti-tau strategies or co-targeting associated mechanisms, such as vascular disease, mitochondrial dysfunction, diabetes, sleep fragmentation, neuroinflammation, and gut microbiome dysbiosis. And, above all, we are not recruiting subjects at the earliest, ideally preclinical state. While the “more” and “early” approaches to therapy are sensible, the futility of trials also offers the opportunity to reconsider the directionality of the amyloid hypothesis. We here propose that the harm may be inflicted not by the accumulation of misfolded, insoluble Aβ but by the consumption of its native, soluble precursor.
PERFECT HYPOTHESIS, IMPERFECT TRIALS
A “perfect” clinical trial is one defined by the following characteristics: it targets early, prodromal, or, ideally, preclinical patients; uses sensitive endpoints, responsive to clinical and imaging changes attributable to the intervention; achieves adequate target engagement; and applies the right dose. If such study had been conducted, the consensus goes, we would have demonstrated the hypothesized alignment between reducing brain Aβ and improving clinical function.
Since Aβ aggregates are conceived as invariably toxic, negative anti-Aβ trials can only be the result of imperfections in clinical trials, not in the underlying hypothesis. The conclusions of two trials published in 2018, both of which met target engagement by demonstrating a reduction of brain Aβ, provide a poignant illustration:
Crenezumab, a monoclonal anti-amyloid antibody, did not change cognitive endpoints. Conclusion: further studies should examine the effect of higher doses [5]. Verubecestat, a β-site amyloid precursor protein cleaving enzyme-1 (BACE1) inhibitor, worsened cognitive endpoints and accelerated hippocampal atrophy. Conclusion: further studies should examine the effect of lower doses [6].
The reader is left to conclude that the treatment did not work enough in the former but worked excessively in the latter, with a discrepancy between the extent of amyloid reduction and the clinical endpoints accounted for by errors in dose calculation.
A possible exception to the string of failures was recently suggested. On March 21, 2019, Biogen announced it would discontinue the ENGAGE (NCT 02477800) and EMERGE (NCT 02484547) Phase III trials of the anti-Aβ human monoclonal antibody aducanumab in AD after the analysis of an independent data monitoring committee determined that both were unlikely to meet their primary efficacy endpoint. On October 22, 2019, Biogen notified investors that the data now “suggested a different possible outcome than that predicted by the futility analysis.” A 15% increase in sample size provided by 318 additional subjects enrolled between March and October, preferentially allocated to the high-dose arm, allowed one of the two clinical trials, EMERGE, to meet the primary (Clinical Dementia Rating-Sum of Boxes, [CDR-SB]) and secondary endpoints (Mini-Mental State Examination, AD Assessment Scale-Cognitive Subscale 13 Items [ADAS-Cog 13], and the AD Cooperative Study-Activities of Daily Living Inventory Mild Cognitive Impairment Version). In July 2020, Biogen submitted a Biological Licensing Application to the Food and Drug Administration to gain regulatory approval for aducanumab as a therapy to reduce clinical decline in AD.
The large reduction of Aβ by aducanumab stands in contrast with the small magnitude of change in CDR-SB relative to placebo reported as significant in the EMERGE trial, which at only –0.39 points is well below the 2-point minimal clinically important treatment difference for patients with mild AD [7]. The hope for any success in our fight against AD favors overlooking a biased evaluation of the data from the aducanumab trials. It heightens our trust in the subgroup results of a trial with only 60% of completers (EMERGE) and in the significance of a p value of < 0.05 despite the post-hoc nature of the multiple comparisons made. It also allows us to discard the opposing evidence from an identical trial (ENGAGE), which yielded a nonsignificant 0.03 CDR-SB point difference favoring placebo. Critically, it leads us to accept as rational the key comparison between the 20% subset of compliant, high-dose completers, who were mostly non-APOE ɛ4 carriers, and the larger randomly assigned placebo group, who were mostly APOE ɛ4 carriers [8]. To compound matters, because most patients in the high-dose group who stopped their treatment due to amyloid-related imaging abnormalities (ARIA) were allowed to resume study participation, blinding was compromised. The “Lazarus-like return” of aducanumab in the background of incomplete and delayed reporting despite analyses available since June 2019 has raised important questions about the validity and clinical significance of the results [9]. Regardless of these uncertainties, the FDA advisory committee in November 6, 2020 supported aducanumab’s pharmacodynamic effect on plaques but found no evidence for effectiveness in either of the trials. The FDA is expected to issue a final decision on the drug application by June*.
The pitfalls in the donanemab trial report
A recently completed phase II randomized, placebo-controlled trial of donanemab, a humanized IgG1 antibody directed at an N-terminal pyroglutamate Aβ epitope, in 257 patients with early symptomatic AD (Eli Lilly-funded TRAILBLAZER-ALZ) “resulted in a better [\dots] ability to perform activities of daily living than placebo at 76 weeks” [10]. Review of the full dataset suggests an alternative outcome. First, it is important to note that donanemab met target engagement: At 76 weeks, the reduction in the amyloid plaque levels assessed by florbetapir positron emission tomography (PET) was 85.06 centiloids greater in the donanemab group than in the placebo group (–84.13 versus 0.93 centiloids). On the other hand, the primary outcome, the composite Integrated Alzheimer’s Disease Rating Scale (iADRS; range, 0 to 144), showed a difference of only 3.20 points in favor of donanemab. While statistically significant (p = 0.04), the trial was powered for a difference of 6.0 points, a goal not reached. A survivorship bias contributed to the very modest iADRS change: nearly 40% of patients developed ARIA and most of them were allowed to continue, compromising the blinding. Tellingly, all secondary outcomes were negative. Significant but underemphasized was the reduction in brain volume and the increase in ventricular volume (a surrogate of brain volume) in donanemab-treated patients compared to placebo. Finally, the slopes of decline for donanemab and placebo were parallel, not divergent, suggesting an absence of disease-modifying effects. Nevertheless, this “promising” phase II study has justified moving on to a future phase III trial.
The clinical trial experience notwithstanding, aggregated Aβ remains the main target for disease-modifying therapeutic efforts, with a major emphasis toward initiating anti-Aβ clinical trials as early as possible, ideally in cognitively normal amyloid-positive individuals at midlife. This comes with the recognition that initiating anti-Aβ therapies in the absence of cognitive deficits “will inevitably result in some individuals receiving unnecessary treatment” [11]. In this article, we propose that the same futility or harm to anti-Aβ therapies shown in those already symptomatic will be achieved in otherwise healthy individuals, as early data already indicate [12]. We also propose that Aβ aggregation is a consequence of a range of many pathogenic events (expected to be different in different individuals) and that the consumption of the soluble Aβ precursor may be what is most detrimental for brain function.
THE LOGIC AND PITFALLS OF THE TOXIC GAIN-OF-FUNCTION HYPOTHESIS IN ALZHEIMER’S DISEASE
The driving force behind the anti-Aβ efforts in AD trials is the assumed toxicity of amyloid as it accumulates, a gain-of-function (GOF) mechanism of disease pathogenesis. For over a century, AD has been conceived as a heterogeneous dementing disorder arising from the misfolding of Aβ into plaques and tau into NFT, as documented under the microscope at autopsy. This clinicopathologic framework has supported two major conclusions: The identification of Aβ plaques and NFT at autopsy provide definite confirmation that a presumed clinical impression of AD was, in fact, AD. The identification of Aβ plaques and NFT confirm that the disease is a proteinopathy; these proteins must have caused or contributed to the disease.
Cause or effect
For the following discussion, and according to the literature, amyloid positivity is defined as a brain positron emission tomography (PET) amyloid radiotracer uptake binding above a certain ratio of capture or a cerebrospinal fluid (CSF) level measurement below a cutoff of 192 pg/ml [13]. Among cognitively normal individuals, everyone (100%) with no measurable brain amyloid has CSF Aβ42 levels above 192 pg/ml (as per Luminex platform) whereas ∼87% of those with positive brain amyloid have CSF Aβ42 levels below 192 pg/ml [14]. In the absence of symptoms, amyloid positivity enables the diagnosis of “Alzheimer’s disease continuum” [15]. The longitudinal amyloid PET and CSF data that have emerged in the last ∼5 years showed that a proportion of otherwise cognitively normal individuals with abnormal amyloid biomarkers are likely to decline when followed for a substantial period of time [14]. This is also true for cognitively normal individuals with subthreshold amounts of fibrillar amyloid [16, 17]. Clearly, amyloid positivity reflects an abnormality in the brain. But is it a cause or an effect?
The long lag between the increase in extraneuronal amyloid and the onset of dementia has been explained on the basis that brain amyloidosis must first turn toxic with isocortical tau deposition in the form of NFT, an intraneuronal process. This argument requires allowing for major exceptions, such NFT accumulation in chronic traumatic encephalopathy, a disorder in which most cases show no aggregated Aβ42 [18] and primary age-related tauopathy (PART), in which NFT in the mesial temporal lobe among cognitively healthy aged individuals includes few or no amyloid deposits [19]. While Aβ-associated neurodegeneration of the entorhinal cortex occurs only in the presence of high p-tau [20], a reduction in CSF Aβ42 is the strongest predictor of conversion to AD (hazard ratio [HR], 16), an effect size four times greater than the increase in t-tau (HR, 2.8) or p-tau 2.6 (HR, 2.6) [21].
As a marker of neuronal death, tau accumulation is considered a downstream event [22]. As such, it is temporally closer to the clinical manifestations of, and therefore better correlated with, dementia, but with lower specificity than Aβ42 (70.5% versus 91.8%) [23]. Nevertheless, under a GOF framework of protein toxicity, negative anti-amyloid trials have inspired a robust focus on anti-tau approaches. Foreboding the outcome of 12 anti-tau therapies in development, the first phase II clinical trial of an IgG4 humanized anti-tau antibody against the N-terminal epitope (semorinemab, TAURIEL trial) showed no differences in the rates of cognitive decline in any subgroup analyses of prodromal or mild AD patients compared with placebo [24]. Further suggesting that tau accumulation may be a terminal event, two anti-tau therapies examined in patients with progressive supranuclear palsy, a “primary” tauopathy, have already shown futility [25].
THE CENTENARIANS’ PARADOXES
Although by the generous life span of 85 years about 60% of the population is amyloid-positive, the prevalence of dementia is only 10%, 5-fold lower than predicted (Fig. 1) [26]. That is, while we are more likely than not to be in the “AD continuum” toward the end our lives, conspicuously few among us will develop AD. If aggregated Aβ were toxic, the ratio of amyloid positivity to dementia should approach 1 by the age of 75 years, represented by the Aβ and dementia curves converging within a normal lifetime.
Biologically versus clinically defined Alzheimer’s disease. Figure adapted from the population-based Mayo Clinic Study of Aging (5,213 individuals from Olmsted County) [26] showing the estimated prevalence of amyloid (by PET, n = 1,524) compared to that of dementia of Alzheimer’s type across the age span from 60 to 90 years. The prevalence of dementia in amyloid-positive individuals is about 5 times lower than expected by the age of 85 years in this dataset. Notice that the amyloid and dementia curves diverge for most of the first two decades (area represented by a trapezoid) but become parallel only after the age of 85 (area represented by a parallelogram). These curves never converge, as it would have been predicted according to a toxic, gain-of-function mechanism for amyloid.
Upon autopsy findings of Aβ aggregation and NFT of a cognitively normal under-60 individual, the usual conclusion is that such person was ‘on the path to developing AD.’ But what if these findings occurred in a cognitively normal over-90 person? Evidence from autopsy studies of well-characterized individuals who died long past a normal life span challenge the sensitivity and specificity of Aβ plaques and NFT as diagnostic markers of AD—and the very concept that pathology equals pathogenesis. Viewed without current assumptions of toxicity, the data from centenarians suggest, instead, that amyloid positivity may be the common denominator of a brain under biological stress. This oldest-old literature offers many paradoxes to the prevailing disease model built on normal life spans: 1) half of those with dementia have neither Aβ plaques nor NFT and half of those without dementia have both Aβ plaques and NFT [27]; 2) severity staging of brain Aβ plaques and NFT either does not correlate [28] or is inversely associated with dementia [29]; and 3) the protective APOE ɛ2 genotype may be more prevalent in those with greater density of Aβ plaques and NFT compared to APOE ɛ4, the detrimental variant [30]. Combining the centenarians data, AD pathology yields an odds ratio of dementia equal to 0.3, with a confidence interval solidly in the protective range [31]. Nearly half of all over-80 “cognitively stable” participants from the population-based Mayo Clinic Study of Aging (defined by the presence of normal cognition over 5 years of follow up) are amyloid positive [32]. Because such findings are at odds with the GOF hypothesis, a line of research on “cognitive resilience” or “cognitive reserve” has emerged as an attempt to reconcile the discrepancy. The overwhelming evidence suggests that any variability in cognitive reserve cannot be predicted by (and is independent from) the burden of AD pathology [33].
BIOPHYSICAL BASIS OF AMYLOID AGGREGATION: THE LOSS-OF-FUNCTION HYPOTHESIS
A reduction in CSF Aβ42 reflects increased deposition of Aβ in the brain. The prevailing hypothesis models the toxicity as a function of the increase in insoluble Aβ42 (amyloid) rather than the decrease in soluble CSF Aβ42 (Fig. 2) [34]. The data from which the model was created showed a decrease in CSF Aβ42, not an increase. What increases is the insoluble, aggregated form of Aβ42 in the brain. The soluble Aβ42 depletion means a loss of normal Aβ42 function. This loss-of-function (LOF) mechanism of toxicity is best explained by biophysics.
The model versus the data. The top half of the figure illustrates the standard “A/T/N System” [15] (amyloid, tau, and neurodegeneration) of Alzheimer’s disease, with sequential accumulation of CSF Aβ42 followed by PET Aβ42, tau, atrophy, and eventually the expression of symptoms as mild cognitive impairment (MCI) and dementia. These are all noted as going up, implying a gain-of-function toxicity [34]. However, the data from which the model was drawn (Figure 1 in [55], from the DIAN database), illustrated in the bottom half, showed an early and steep reduction in soluble Aβ42, matching the rising insoluble Aβ42 and mirroring hippocampal atrophy. (Increased atrophy of hippocampi is detected in mutation carriers 15 years before symptom onset [55], which is why volume starts below zero.) Aβ42, amyloid β, 42 residues long.
From autopsy-based conclusions to biophysical experiments
The emphasis on the importance of the insoluble Aβ accumulation rather than the loss in its soluble counterpart is an artifact of the clinicopathologic model of neurodegenerative diseases. Autopsy studies that defined boundaries between clinical constructs could only quantify the insoluble, aggregated proteins since their soluble precursors had become invisible at postmortem. It was logical to conclude that abnormally folded proteins must have inflicted toxicity to the brain. With the assistance of cell based and animal studies, the GOF principle was rooted into a theory of conformational replication, which posits that amyloids are proteins that behave like viruses in terms of replication and cytotoxicity, despite the lack of nucleic acid [35].
However, recent studies on the structure and biophysical mechanisms of amyloid aggregation contradict these virus-like assumptions about the behavior of proteins. The term amyloid does not refer to a particular protein; it denotes a general state of proteins polymerized into solid, insoluble fibrillar structures adopting a cross-β conformation [36]. This process of amyloid transformation or polymerization affects as many as 35 proteins in human diseases well beyond AD [37] and is similar to other inorganic and organic phase transformation phenomena, such as crystallization. The transformation occurs at three levels: 1) structural, from monomeric or oligomeric to cross-β conformation; 2) physical, from soluble to insoluble; and 3) biological, from functional to non-functional [38]. The phase transformation from liquid to solid requires an initial nucleation event. Because this “nucleation” reaction is thermodynamically unfavorable, it is usually catalyzed by a preformed nucleus (seed) or an exogenous surface, such as a lipid membrane [38, 39]. Once the solid, amyloid phase is reached, its marked stability makes it very difficult to turn back to the soluble phase. In fact, the amyloid state is considered the most stable state any protein can reach [40]. This extreme stability can explain many features of amyloid fibrils, such as their resistance to degradation and solubilization and their relatively inert nature. The amyloid state is too stable and non-reactive to be toxic.
The biophysical properties of amyloid fibrils and the mechanism of their assembly indicate that amyloids are not proteins behaving like viruses but proteins behaving like proteins. They can polymerize, solidify, and precipitate following well-defined physico-chemical mechanisms. Moreover, the process of polymerization superficially resembles but is opposite to replication, because there is no transmission of biological information, such as that mediated via DNA mechanisms. Polymerization and subsequent polymorphism depend on such physical and microenvironmental conditions as concentration, pH and the presence of surface catalysts [41]. As these extrinsic factors cannot be encoded for in the amyloid architecture, amyloids cannot faithfully replicate as biological strains and may be incapable of exerting relevant cytotoxic effects. Moreover, all amyloids adopt a universal secondary structure, the cross-β conformation, irrespective of the strain (polymorph) or the pathway to aggregation, be it spontaneously via homogeneous nucleation or catalyzed by a surface via heterogeneous nucleation (see below). This indicates that adoption of the secondary cross-β structure is a thermodynamically favorable, spontaneous folding event under the conditions of phase transformation, and does not require templating. Thus, “strain-like transmission” of conformational information is theoretically impossible to be encoded within the amyloid structure, explaining why the search for the long-sought mechanism of “conformational replication” remains elusive [42].
If proteins undergo polymerization rather than replication, the loss of their native conformation and solubility might be more important for pathogenesis than their end-stage of amyloidosis. Not only are proteins known to be functional in their soluble, normal-configuration state, but studies in animal models have demonstrated that, in many amyloid pathologies, knock-out and knock-down animals show disease phenotypes without protein aggregates [43].
HETEROGENOUS NUCLEATION OF AMYLOID-β: PATHOGEN-TRIGGERED SOLUBLE Aβ CONSUMPTION
Exposure to exogenous surfaces can trigger the process of heterogenous nucleation whereby soluble Aβ is transformed into insoluble amyloid. While the pathogenic exogenous surface remains unknown in most AD cases, a recognized surface is provided by herpes simplex virus type 1 (HSV-1) based on epidemiologic [44] and pathology studies demonstrating a high prevalence of HSV-1 infection in AD brains [45]. Within 3 days of applying the antiviral valacyclovir on a 3D bioengineered brain model, the infection was eliminated and the associated Aβ and tau pathology significantly decreased [46]. The standard conclusion has been that antivirals “might be useful for preventing and/or treating AD,” but these are most likely to exert “anti-AD” effects only in those whom HSV-1 has been identified as the pathogen providing the heterogeneous nucleation surface for the soluble-to-insoluble Aβ transformation [47, 48].
In vivo microdialysis studies have shown that increased neuronal and synaptic activity are dynamically associated with increased brain region-specific amyloid [49], on a timescale of minutes to hours [50]. Areas of the human brain that accrue the most amyloid also have the highest basal rates of metabolic and neural activity, as measured by PET and functional magnetic resonance imaging [51]. If a pathogen is controlled (e.g., eradication of HSV-1 infection by an antiviral treatment), hypermetabolism and Aβ aggregation are likely to cease. If a pathogenic injury is unaddressed, as is the case in persistent infections of the central nervous system, more soluble Aβ becomes consumed into aggregated Aβ and clinical dementia eventually appears.
SOLUBLE Aβ CONSUMPTION IN HOMOGENEOUS NUCLEATION OF AMYLOID-β
The hereditary amyloidopathies have been considered major evidence in support of the GOF hypothesis. In these disorders, Aβ species are either overexpressed or prone to aggregation, such as in individuals with amyloid protein precursor (APP) mutations or duplications, presenilin-1 (PSEN1) or presenilin-2 (PSEN2) mutations [52]. Those with these genetic subtypes of AD have been deemed ideal for testing anti-aggregation treatments, especially if deployed at the earliest possible stage. That was the objective of the DIAN-TU study (NCT01760005), a Phase II/III trial of two anti-aggregation therapies (gantenerumab and solanezumab) versus placebo to examine if either one could slow cognitive decline over a mean of 5 years in 194 individuals with a genetic mutation in APP, PSEN1, or PSEN2. The results, reported at the ADPD meeting on April 2020, showed that the trial did not meet its primary endpoint with either of the monoclonal antibodies [53], suggesting that even in hereditary amyloidopathies an anti-amyloid approach may be misguided.
Genetic disorders may in fact also obey a LOF mechanism: the genetic mutations render soluble Aβ inherently unstable, that is, prone to aggregation into insoluble amyloid despite the absence of a nucleating catalyst, a process referred to as homogenous nucleation [38]. Thus, APP, PSEN1, and PSEN2 patients may not be overexpressing toxic, insoluble Aβ, but rapidly consuming their soluble functioning precursor. In agreement with this postulate, soluble levels of Aβ42 have been reported to be lower, not higher, among autosomal-dominant AD patients participating in in the Dominantly Inherited Alzheimer Network (DIAN) study [54, 55] and in Down syndrome patients with APP duplication [56]. These findings demonstrate that the LOF hypothesis is compatible with the strong genetic evidence linking Aβ42 to AD and that soluble Aβ42 depletion occurs in both sporadic and genetic forms of AD.
LOSS OF AMYLOID FUNCTION HYPOTHESIS VERSUS “OLIGOMERIC TOXICITY”
To reconcile the contradictions of the amyloid cascade hypothesis, particularly the poor correlation between amyloid plaques and cognitive decline and the ostensible paradox between successful amyloid reduction and lack of cognitive benefit in anti-amyloid trials, toxicity has been attributed to soluble low- and medium-molecular weight amyloid oligomers [57, 58]. However, the bar to demonstrate toxicity for a short-lived intermediary between two relatively stable non-toxic species should be high. The ill-defined structural and biochemical nature of Aβ oligomers and the lack of agreement on the relevant species as well as the optimal detection method for Aβ oligomers have prevented a mechanistic understanding of the relationship between oligomers and cognitive alterations or neuronal death in AD [59], rendering them problematic targets of therapy [60]. From a physicochemical perspective, oligomers are ephemeral (their half-lives are around 1 to 5 hours at physiological pH), cannot rapidly grow by monomer addition, and predominantly dissociate back to monomers rather than proceed to transform into fibrils [61]. Given that one of the defining features of oligomeric species is that they are soluble, they should be detectable in the CSF of patients with AD but not in that of controls. In fact, as with monomeric Aβ, the concentration of both low- and high-molecular weight Aβ oligomers is lower in CSF of AD patients than in non-AD dementias or other neurological disorders, using flow cytometry and fluorescence resonance energy transfer-based methods [62]. Furthermore, severe cognitive decline is strongly associated with a decrease, not an increase, in CSF Aβ-oligomer levels over time [63]. Some studies have even suggested that Aβ42 oligomers can act as antimicrobial agents in what has been called the “Antimicrobial Protection Hypothesis” [64, 65].
REPLACING AMYLOID-β TO CORRECT ITS CONSUMPTION IN ALZHEIMER’S DISEASE
In a revised model of the pathophysiological processes involved in AD and other amyloid pathologies, an important disease driver is the LOF or consumption of the soluble protein (Fig. 3). As the disease progresses, a superimposed GOF mechanism may accrue as a larger burden of insoluble protein accumulates. Any such GOF toxicity, however, would occur at a stage sufficiently advanced to render the removal of insoluble Aβ ineffective for the purpose of disease modification. This model would explain the failure of antibodies targeting amyloids despite adequate target engagement. In its place, replacement therapy with chemically-modified, non-aggregating analogues of the soluble peptides could be used as an early intervention to keep brain levels above the threshold below which neuronal toxicity may occur. Such a replacement approach would be responsive to evidence from biomarker studies demonstrating that soluble peptides decline very early in the disease [54, 66], and from in vitro studies showing that synthetic Aβ42 monomers support the survival of neurons under conditions of trophic deprivation o excitotoxic death [67], rescuing brain glucose consumption by activating type-1 insulin-like growth factor [68].
Revised model of Alzheimer’s disease. The reduction of soluble Aβ42 is compensated in the prodromal stage until a threshold is reached, below which symptoms appear. That threshold was tentatively calculated to be 800 pg/ml, based on an analysis of the ADNI dataset [73]. This model emphasizes an early loss-of-function toxicity while accounting for the possibility of a late gain-of-function toxicity from increases in insoluble Aβ42. LOF, loss-of-function; GOF, gain-of-function; Aβ42, amyloid β, 42 residues long.
Important data have emerged to support a replacement approach to AD. An extracellular fragment of AβPP termed secreted AβPP-alpha (AβPPsα) has been shown to have neurotrophic, neuroprotective, and neurogenic effects, stimulating protein synthesis and gene expression while enhancing long-term potentiation and memory [69]. In an AD mouse model, the overexpression of soluble AβPPsα restored synaptic plasticity and rescued spatial memory [70]. Another replacement therapy framework has been demonstrated by islet amyloid polypeptide (iAPP) or amylin analogues, already used as replacement therapy in diabetes [71]. Pramlintide is an amylin analogue that is chemically modified to preserve its function, enhance solubility, and reduce its propensity to form amyloid aggregates [72]. This preclinical evidence in favor of soluble peptide replacement is now supported at a clinical level by an analysis we recently conducted of all amyloid PET-positive individuals participating in the Alzheimer’s Disease Neuroimaging Initiative (ADNI) study. We tested the hypothesis that the maintenance of high Aβ42 levels can explain the paradox of normal cognition and preserved hippocampal volume in the setting of brain amyloidosis. Soluble Aβ42 levels were indeed significantly higher, in a dose-dependent manner, in amyloid-positive individuals with normal cognition than in those with mild cognitive impairment or AD, and were associated with preservation of hippocampal volume [73]. Levels of Aβ42 above 800 pg/ml were associated with normal cognition regardless of brain amyloid burden [73].
A proof-of-concept clinical trial for protein replacement in AD would face the same issues of anti-amyloid trials, namely how early to intervene, for how long, and how to measure success. Because the magnitude of the desired biomarker change in prior trials (reduction in brain PET Aβ42) has not aligned with clinical improvements, optimal candidates for replacement therapy should target symptomatic patients, not individuals at the prodromal or preclinical stage. This would allow measuring whether the desired biomarker change, increase in CSF Aβ42, correlates with the hypothesized clinical improvements. Biological criteria would supersede clinical criteria for recruitment purposes. Patients with any cognitive or behavioral symptoms, regardless of whether they meet strict clinical criteria for AD, would be selected on the basis of having low CSF Aβ42 and moderate burden of Aβ42 on amyloid PET (standardized uptake value ratio [SUVR] between 1.30 and 1.46, encompassing the middle tertile in our recent ADNI-based analysis) [73]. To confirm that the molecule tested acts in a more stable fashion, an increase in CSF Aβ42 should not accompany an increase in Aβ42 on amyloid PET. Finally, the time course of such study will need to be at least 3 years, with interim analyses of clinical, CSF, and PET-based endpoints at 1 and 2 years.
CONCLUSIONS
Biophysically, amyloids are proteins that lost their native conformation, solubility and, hence, their function. The consistently null or detrimental outcomes from anti-amyloid clinical trials likely reflect a fundamental problem with the GOF amyloid hypothesis rather than artifacts of trial design or execution, or therapeutic endeavors enacted “too late” [74]. The time has come to examine the competing LOF hypothesis for Aβ. Launching disease modification for AD will require a paradigm shift from insoluble protein destruction to soluble protein replacement in order to mitigate the LOF toxicity (Fig. 4). Eventually, the individualized identification and targeting of the pathogenic nucleating factors that trigger amyloid aggregation promises to launch precision medicine in AD and other neurodegenerative disorders.
Disease-modifying treatment paradigms for AD. Soluble, mono- to oligomeric functioning peptides require exposure to a nucleating surface, such as that of herpes simplex virus type 1 illustrated in the left upper corner, in order to phase transform via heterogeneous nucleation into amyloid. This loss-of-function hypothesis predicts benefits from two complementary therapeutic strategies: replacing the depleting soluble peptides (rescue therapy, applicable to everyone affected) and targeting pathogenic nucleators (precision medicine, applicable to one biotype at a time).
DISCLOSURE STATEMENT
Authors’ disclosures available online (https://www.j-alz.com/manuscript-disclosures/21-0415r1).