GPT 5.6 Sol using GPT Work, with Jared Edward Reser

Working scholarly draft; support-led revision and critical citations updated 12 July 2026.

Abstract

Alzheimer’s disease is defined by a patterned convergence of glucose hypometabolism, tau and amyloid biology, synaptic failure, inflammation, and neuronal loss. The adaptive cerebral-thrift hypothesis proposes that part of this sequence begins as regulated withdrawal from metabolically expensive neural plasticity and becomes disease when recovery, containment, and substrate compensation fail. Recent evidence makes this model more mechanistically specific. Hibernation produces large reversible increases at Alzheimer-relevant tau phosphosites without the microtubule-binding-region fragments associated with mature aggregation. Site-specific tau phosphorylation is now known to perform normal circuit functions: T205 phosphorylation induced during memory encoding constrains extraneous neuronal recruitment and supports remote-memory formation, while S396 phosphorylation is required for AMPA-receptor internalization during hippocampal long-term depression. T205 phosphorylation also protects Alzheimer-model mice from amyloid-β excitotoxic signaling. Amyloidogenic APP processing is causally required for homeostatic synaptic scaling in healthy hippocampal tissue, establishing physiological Aβ as part of a synaptic feedback controller. Most directly, food restriction in mouse visual cortex lowers AMPA-receptor conductance and synaptic ATP expenditure by approximately 29%, preserves spike output, selectively reduces fine coding precision, and reverses with restored weight or leptin signaling. Natural sleep independently removes synaptic AMPA receptors and shrinks mostly smaller synapses while restoring human LTP-like plastic capacity. Resting synaptic vesicle pools impose a continuous energetic burden, with vesicular proton leakage accounting for up to 44% of resting terminal ATP consumption, providing a concrete energetic payoff for synaptic disassembly beyond reduced firing. Cooling–rewarming experiments identify an explicit failure-of-recovery mechanism: Alzheimer-model mice lose RBM3 induction and synapse-regeneration capacity before established synaptic loss, while RBM3 restoration rescues reassembly and prevents behavioral and neuronal deterioration. Early-Alzheimer mouse engrams can retain memories that are inaccessible to natural cues but recoverable by targeted activation. In humans, aging selectively loses youthful aerobic glycolysis in developmentally prolonged association regions; Alzheimer glucose deficits coexist with preserved ketone uptake; randomized ketogenic interventions increase brain energy supply without restoring glucose uptake; and noradrenergic augmentation in biomarker-positive MCI shifts soluble tau and metabolic markers in a recovery-compatible direction. Environment-dependent cognitive and fertility advantages of APOE4 supply a concrete early-benefit/late-cost tradeoff. These findings support a failure-of-exit model in which physiological restriction, disassembly, feedback, and forgetting mechanisms become persistent or mistargeted. They do not yet prove positive selection for age-related human cerebral thrift: whole-brain and life-course energy savings, the adaptive brain phenotype linking APOE4’s early benefits to its late costs, and the ordering of these states in sporadic disease remain unresolved. We propose a phosphosite-resolved entry–exit experiment combining engram tracking, synaptic ATP accounting, RBM3 and arousal manipulation, APOE genotype, and alternative-fuel rescue to distinguish regulated economization from generic stress and degeneration.

Keywords: Alzheimer’s disease; aging; aerobic glycolysis; amyloid-β; cerebral metabolism; cognitive aging; hibernation; synaptic pruning; tau; torpor

1. Introduction: pattern is not proof of purpose

Alzheimer’s disease (AD) does not affect all neural systems simultaneously or equally. Its early clinical expression emphasizes episodic memory and flexible learning; its neuroimaging signature concentrates in medial temporal and association networks; and its molecular pathology unfolds across decades. Primary sensory and motor capacities, familiar routines, and some procedural abilities often remain functional after the capacity for rapid encoding and flexible recombination has weakened. This pattern raises a biological question that is distinct from the question of molecular toxicity: why does aging fail in this particular sequence?

The adaptive cerebral-thrift hypothesis offers a life-history answer. The developing human brain invests heavily in learning, synaptic remodeling, biosynthesis, and construction of culturally informed representations. After maturation, an increasing share of behavior can be guided by established knowledge, practiced skills, and stable policies. If plasticity is energetically expensive, natural selection might favor a gradual reduction in the cost of maintaining an indefinitely youthful learning system, provided that consolidated function remains useful. Clinical AD would not be the adaptation. It would be the late pathological extreme of an earlier, compensated trajectory.

This idea was initially formulated as the proposal that natural cognitive aging and AD may represent metabolism-reduction programs (Reser, 2009). A recent reformulation describes “adaptive cerebral thrift”: reduced investment in plastic metabolism, synaptic turnover, and flexible cognition, followed in susceptible brains by failed compensation, persistent tau and amyloid pathology, inflammation, and degeneration (Reser, 2026). The conceptual appeal is clear. It connects cognitive aging, regional vulnerability, energetic demand, and molecular pathways that are also engaged during starvation, torpor, and hibernation.

Appeal is not confirmation. Similarity between AD and a low-energy physiological state can arise for at least four reasons. First, AD may actively recruit a conserved energy-management mechanism. Second, degenerating neurons may mount a compensatory stress response. Third, both states may converge on temperature- or energy-sensitive enzymes without sharing an adaptive function. Fourth, pathology may selectively damage high-demand systems, creating the appearance of strategic downregulation. Any serious evaluation must distinguish these alternatives.

We therefore separate five evidentiary levels: (1) phenomenological overlap; (2) molecular homology; (3) causal energetic function; (4) selective functional preservation; and (5) positive evolutionary benefit. The cerebral-thrift hypothesis becomes progressively stronger as it moves through these levels. Evidence at one level cannot be treated as proof of the next. Our goal is to determine which levels are currently supported, where the hypothesis requires revision, and which experiments could provide decisive evidence.

2. A formal cerebral-thrift model

The strongest testable formulation contains eight steps:

1. Human development sustains unusually high investment in cortical plasticity, biosynthesis, synaptic remodeling, and cultural acquisition.
2. After maturation, the marginal return from open-ended acquisition declines relative to the return from established knowledge and skill.
3. Normal aging includes regulated withdrawal from parts of the youthful plastic metabolic phenotype.
4. State-sensitive tau phosphorylation, synaptic downscaling, activity-linked Aβ signaling, and glial remodeling may implement this withdrawal.
5. Moderate expression reduces neural expenditure while preserving consolidated representations and essential behavior.
6. Proteostasis, phosphatase activity, sleep, arousal systems, vascular support, substrate flexibility, and glial containment permit compensation or recovery.
7. Aging and AD risk impair these recovery mechanisms.
8. Persistent remodeling becomes protein aggregation, inflammatory pruning, network failure, synaptic loss, and dementia.

The model makes predictions that are not required by a generic degeneration account. Early human p-tau should resemble reversible low-energy tau more closely than mature tangles. The phosphorylation state should have a measurable energetic or structural function. Downscaling should preferentially affect labile, high-turnover synapses while preserving consolidated function. Young healthy animals should exit efficiently, whereas aged or AD-risk animals should show delayed tau dephosphorylation, incomplete synaptic restoration, or failed metabolic rebound. Finally, the savings must be large enough to matter to whole-organism fitness.

These are demanding predictions. They are valuable precisely because the hypothesis can fail them.

3. Reversible tau phosphorylation: the strongest molecular parallel

3.1 Alzheimer-relevant phosphosites during hibernation and torpor

The most compelling evidence for a connection between AD biology and adaptive low-energy physiology concerns tau. European ground squirrels show strong AT8 immunoreactivity during torpor that reverses after arousal, demonstrating that paired-helical-filament-like phospho-epitopes can participate in a nondegenerative state (Arendt et al., 2003). Subsequent studies extended the observation across species and sites. Arctic ground squirrels exhibit high phosphorylation at multiple tau sites during torpor, with site-specific dephosphorylation during interbout arousal and the lowest levels in summer-active animals (Su et al., 2008). Syrian hamsters, arctic ground squirrels, and black bears show a relationship between metabolic depression and phosphorylation at AD-relevant sites (Stieler et al., 2011). Synthetic torpor induces reversible AT8-positive tau in the brain and spinal cord of nonhibernating rats, showing that the basic response is not restricted to seasonal hibernators (Luppi et al., 2019; Hitrec et al., 2021).

A recent mass-spectrometry study provides the clearest molecular boundary between adaptive and pathological tau states. In paired samples from free-ranging brown bears, plasma p-tau181 and p-tau217 increased by median values of 362% and 294% during hibernation, and p-tau205 and p-tau231 also rose. In hamster brain, p-tau217 and p-tau231 increased during torpor. Crucially, microtubule-binding-region (MTBR) fragments associated with tangle aggregation did not increase. Human AD brain measured with the same platform showed much larger p-tau abnormalities and pronounced MTBR-fragment elevations (Brum et al., 2025). Hibernation therefore reproduces part of the soluble phosphorylation signature but stops before the aggregation state characteristic of AD.

This result supports a central distinction: the adaptive candidate is not the mature neurofibrillary tangle. It is an earlier, soluble, reversible phosphorylation state. Mature aggregation may represent what happens when phosphorylation, mislocalization, proteostatic failure, and seeding persist.

3.2 T205 phosphorylation constrains excess recruitment and supports remote memory

The strongest recent evidence that an Alzheimer-linked tau modification has a normal circuit function concerns threonine 205. Kosonen et al. (2026) found that memory encoding induces T205 phosphorylation and that phosphorylation-competent tau regulates recruitment into hippocampal engrams while suppressing extraneous local activation. Preventing T205 phosphorylation specifically in encoding ensembles impaired remote spatial and fear-memory retrieval even when initial learning and nearer-term recall were intact. Optogenetic activation of the targeted engram bypassed the retrieval deficit. Tau phosphorylation at this site therefore helps create a sparse, enduring representation rather than merely marking neuronal injury.

This result converges with earlier work in Alzheimer-model mice. P38γ-mediated T205 phosphorylation disrupted postsynaptic excitotoxic signaling complexes engaged by Aβ; loss of p38γ worsened circuit abnormalities, cognitive impairment, and premature lethality, whereas increasing p38γ activity or mimicking the phosphorylation was protective (Ittner et al., 2016). T205 is also strongly responsive during hibernation. Brum et al. (2025) estimated a 791% increase in bear plasma p-tau205, while Su et al. (2008) found that T205 remained elevated into arousal even as several torpor-responsive sites began to dephosphorylate.

These findings suggest that T205 should not be treated as a unitary shutdown marker. It may participate in protection, sparse allocation, or recovery-phase stabilization. The cerebral-thrift model consequently needs a phosphosite-by-compartment-by-time account rather than the claim that bulk hyperphosphorylation implements economization.

3.3 Tau phosphorylation directly participates in physiological synaptic weakening

Other phosphosites provide a direct bridge to downscaling. Hippocampal long-term depression induces tau phosphorylation at S396 and S404. Tau deletion disrupts the GluA2–PICK1 interaction required for AMPA-receptor internalization, while phosphorylation-null S396/S404 tau prevents LTD but leaves LTP intact (Regan et al., 2015). Site-specific tau phosphorylation can therefore be causally required for receptor removal and synaptic weakening.

This is not evidence that aging or AD uses LTD to save whole-brain energy. It is nevertheless an experimentally demonstrated molecular route by which tau can reduce excitatory synaptic efficacy. The next step is to determine whether the same site-specific machinery is recruited during low-energy states and whether its ATP savings are measurable.

3.4 Reversible phosphorylation does not yet demonstrate energy conservation

Tau overlap is striking, but a strong alternative explanation remains. Protein phosphatases are temperature sensitive. Anesthesia-induced hypothermia causes rapid, massive tau hyperphosphorylation by inhibiting PP2A; maintaining normothermia prevents the effect (Planel et al., 2007). Repeated cold acclimation improves thermogenesis in aged 3xTg-AD mice and completely prevents tau hyperphosphorylation during a subsequent cold challenge (Tournissac et al., 2019). These results demonstrate regulation and reversibility, but they also show that AD-like p-tau can arise as a physicochemical consequence of hypothermia.

No study located in this review directly demonstrates that tau phosphorylation lowers ATP expenditure, suspends axonal transport in a way that produces net savings, or improves survival during scarcity. Tau may participate in remodeling, but energetic function remains inferred. This missing causal step is decisive. Without it, hibernation establishes molecular homology, not cerebral thrift.

3.5 Human tau chronology complicates the analogy

Human tau is also site- and stage-specific. In dominantly inherited AD, soluble p-tau217 and p-tau181 rise with initial amyloid aggregation, whereas p-tau205 and total tau rise closer to atrophy and hypometabolism (Barthélemy et al., 2020). In sporadic AD, longitudinal biomarker studies generally place Aβ changes before p-tau, neurofilament changes, hippocampal atrophy, and cognitive decline (Jia et al., 2024). Thus, some early human p-tau is tightly linked to amyloid biology rather than a documented age-programmed metabolic transition.

The productive question is not whether “tau phosphorylation is adaptive” in the abstract. It is whether a subset of soluble human tau states participates in reversible remodeling before crossing into self-propagating aggregation. That question requires state-resolved assays, not a binary distinction between normal tau and tangles.

4. Synaptic disassembly, restoration, and memory

4.1 Hibernators disassemble rather than destroy synapses

Hibernation produces dramatic but reversible neural remodeling. In European hamsters, CA3 apical dendrites become shorter, less branched, and less spiny during torpor, while mossy-fiber terminals show fewer synaptic vesicles and reduced contact with spine profiles. Within approximately two hours of induced arousal, dendritic length, branching, spine density, and terminal morphology return toward active-state values (Magariños et al., 2006). Ground-squirrel neurons across hippocampus, cortex, and thalamus show temperature-dependent retraction and rapid regrowth, with recovery converging on a stable structural set point despite different depths of torpor (von der Ohe et al., 2006).

Synaptic proteins are not simply destroyed. Entry into torpor produces an estimated 50–65% reduction in colocalized presynaptic and postsynaptic markers without corresponding protein loss. Proteins disperse from active-zone and postsynaptic-density clusters, creating a reservoir that can be rapidly reused during arousal (von der Ohe et al., 2007). This is precisely the sort of reversible economization that a thrift model requires: functional disassembly with retention of reconstruction materials.

4.2 Preserved memory despite spine regression

Bullmann et al. (2016) connected soluble p-tau, synaptic structure, and behavior in hibernating golden hamsters. Spine reduction was concentrated in CA3 apical dendrites where phosphorylated tau was abundant, while basal dendrites showed much less of either change. Despite this regression, tested hippocampal-dependent maze memory was not significantly impaired. The authors proposed that tau-associated regression may preferentially affect unstable or dynamic spines while sparing established stable spines.

This is one of the strongest pieces of evidence compatible with selective cerebral thrift. Yet the key mechanism remains unproven. The study did not longitudinally mark individual spines by age or prior learning involvement. Memory retention in a maze does not establish preservation across all consolidated representations, and negative behavioral results may reflect assay sensitivity. A decisive extension would label newly formed and long-lived spines before torpor, manipulate tau phosphorylation, and measure both energy and memory.

4.3 Torpor–arousal can enhance subsequent learning in an AD model

Daily torpor in laboratory mice provides a second important result. De Veij Mestdagh et al. (2021) used mild ambient cooling and food restriction to induce approximately nine-hour torpor bouts in which oxygen consumption fell from more than 120 mL/h to approximately 20 mL/h. Unlike seasonal hibernators, these mice did not exhibit large dendritic or spine losses. During arousal, however, postsynaptic and mitochondrial proteins showed an overshoot, hippocampal plasticity increased, and contextual fear learning was enhanced. APP/PS1 mice conditioned during late arousal subsequently performed at approximately wild-type levels.

This result supports the idea that AD-model circuitry retains recoverable plastic capacity and that arousal biology can produce a synaptic and mitochondrial rebound. It is frequently overstated. The memory was encoded after torpor, not recovered from storage before torpor. Tau was not measured, and the study did not demonstrate p-tau clearance. Cooling, food restriction, fasting, refeeding, torpor, and arousal were not fully isolated as causal components. The experiment is a near-clincher for recovery biology, not proof that AD is failed torpor.

4.4 Neurodegeneration can disable synapse rebuilding before established loss

Cooling–rewarming experiments identify a causal failure-of-recovery mechanism. Wild-type mice reduced synapse number during cooling and rebuilt those contacts after rewarming. Prion-infected and 5xFAD mice initially retained this response but lost it early in disease, before their established phase of synapse loss. The failure tracked loss of induction of the cold-shock protein RBM3. Hippocampal RBM3 overexpression restored synapse reassembly, protected synaptic transmission, prevented behavioral decline and neuronal loss, and prolonged survival; RBM3 knockdown accelerated deterioration and blocked cooling-related protection (Peretti et al., 2015).

This is among the most discriminating findings for a failure-of-exit model. It demonstrates that neurodegeneration involves not only accelerated destruction but loss of a physiological reconstruction program, and that restoring one component of that program changes disease outcome.

4.5 Early Alzheimer-like amnesia can leave the engram present but inaccessible

Roy et al. (2016) labeled dentate-gyrus engram cells during memory formation in early-AD mouse models. The animals later failed recall using natural contextual cues, but direct optogenetic activation of the original engram produced memory expression. The impairment therefore included failure of natural access to retained information rather than immediate erasure of the representation. The result does not imply that advanced Alzheimer memories remain intact; storage deteriorates as synapses and neurons are lost. It does show that functional silencing or disconnection can precede destruction.

The convergence with T205-dependent engram allocation is notable. Physiological tau helps restrict and stabilize the encoding ensemble, while pathological tau within active ensembles produces aberrant activation and amnesia. A thrift model can now be tested at the engram level: does a low-energy state silence access while preserving ensemble connectivity, and does failed arousal leave the representation latent?

4.6 Noradrenergic augmentation moves human tau and metabolic biomarkers

The locus coeruleus is both an arousal controller and an early site of Alzheimer-related tau pathology. A small biomarker-positive MCI crossover trial provides preliminary human intervention evidence. Six months of atomoxetine robustly increased central norepinephrine and reduced CSF total tau by 6.0% and p-tau181 by 4.7% relative to placebo, without changing CSF Aβ42. Exploratory FDG-PET and proteomic analyses pointed toward increased temporal-limbic metabolism and shifts in synaptic and metabolic proteins, although cognition did not change over the short trial (Levey et al., 2022).

This does not establish that norepinephrine is an “exit switch.” Changes in soluble tau may reflect secretion or turnover, the sample contained only 39 participants, and the trial was not powered for clinical efficacy. It nevertheless yields the correct joint direction for a recovery hypothesis: increasing an arousal transmitter simultaneously moved tau, metabolism, and synaptic-metabolic biology rather than merely producing transient alertness. A decisive replication would combine locus-coeruleus imaging, sleep–wake measurement, FDG and ketone PET, SV2A PET, phosphosite-resolved tau, and delayed memory outcomes.

4.7 The closest mechanistic analogue: scarcity-triggered cortical energy economy

An invertebrate experiment supplies unusually direct evidence that selective cognitive austerity can improve fitness. After 24 hours of starvation, Drosophila selectively stopped forming protein-synthesis-dependent aversive long-term memory by silencing slow activity in two pairs of dopaminergic neurons. A cheaper anesthesia-resistant consolidated memory remained available, as did food-relevant appetitive memory. Artificially reactivating the neurons restored aversive long-term memory formation but shortened survival under starvation (Plaçais & Preat, 2013). This is a causal adaptive result: overriding the cognitive restriction improved one form of memory at a measurable survival cost. Its relevance to mammals and Alzheimer aging is indirect, and it concerns formation of new memory rather than preservation of memories already stored.

Padamsey et al. (2022) directly asked whether mammalian cortex economizes during food scarcity. Male mice restricted to 85% of baseline weight showed a 37% decrease in single-channel AMPA-receptor conductance and a 29% decrease in visually evoked excitatory current. Sodium-flux accounting therefore predicted 29% lower ATP expenditure on excitatory postsynaptic currents. This was independently corroborated in awake cortex with a FRET ATP sensor: after ATP-synthesis inhibition, cortical ATP decayed more slowly in food-restricted animals.

The functional tradeoff was strikingly selective. Increased input resistance and a depolarized resting potential preserved average spike rate despite the smaller currents. Coarse differentiation of very different natural scenes remained possible, but trial-to-trial variability increased, orientation tuning broadened by 32%, decoding of similar scenes worsened, and fine visual discrimination was impaired. Coding precision recovered after several days of refeeding and weight restoration; ten days of leptin replacement also restored tuning and scene decoding while the mice remained food restricted.

This is the strongest direct support presently identified for the core cerebral-thrift mechanism. A body-energy signal induced a reversible cortical state that measurably saved ATP, preserved basic output, and sacrificed costly precision. It does not establish an age-scheduled program, memory preservation, or Alzheimer causation. It was conducted in the visual cortex of young male mice, and the calculated synaptic saving is not a whole-brain or whole-body energy balance.

The sex-specific follow-up is theoretically important. Under the same proportional weight loss, young female mice maintained leptin and largely maintained V1 ATP use and orientation selectivity, whereas males showed a 72% leptin decrease, energy-pathway reprogramming, and reduced ATP use and coding precision (Padamsey et al., 2024). Females required greater food restriction and appeared to conserve energy through other tissues. Cerebral economization is therefore conditional on sex and life-history allocation rather than being an automatic consequence of fewer calories. The age prediction is unusually sharp: as ovarian signaling changes, the female cortical response may become more male-like, a possibility the authors explicitly identify but have not tested.

4.8 Sleep implements reversible synaptic down-selection

Natural sleep provides a second convergent system. Three-dimensional electron microscopy of 6,920 cortical synapses found that axon–spine contact area was approximately 18% smaller after sleep. Downscaling was selective: the largest synapses and synapses containing recycling endosomes were spared (de Vivo et al., 2017). Parallel proteomic, biochemical, genetic, and imaging experiments showed widespread removal and dephosphorylation of synaptic AMPA receptors during sleep. Homer1a was required for this process and entered postsynaptic densities when noradrenergic tone fell and sleep-pressure-related adenosine signaling rose; the same pathway participated in contextual-memory consolidation (Diering et al., 2017).

Human physiology shows the complementary exit benefit. After one night of sleep deprivation, healthy young adults had higher cortical excitability but reduced inducibility of LTP-like plasticity and poorer acquisition of new word pairs. Seventy percent expressed LTP-like potentiation after normal sleep, compared with 25% after deprivation (Kuhn et al., 2016). Sleep-related down-selection therefore does not simply reduce function: it restores headroom for new plasticity.

An early AD-model study provides a pathological echo. APP/PS1 double-knockin mice developed age-related postsynaptic AMPAR downscaling, confirmed electrophysiologically and ultrastructurally, alongside impaired LTP, LTD, and flexible learning of a changed platform location; related testing found working memory comparatively intact (Chang et al., 2006). More recently, restoring slow-wave power through frequency-matched astrocyte stimulation in APP/PS1 mice reduced amyloid deposition, prevented neuronal calcium elevation, and improved memory (Lee et al., 2023). These results do not prove that Alzheimer AMPAR loss is “stuck sleep.” Aβ-driven endocytosis can be directly toxic. They do, however, establish a concrete mechanistic convergence: physiological sleep and early Alzheimer pathology both regulate AMPAR availability, while successful recovery rhythms restore plastic capacity and disease models lose it.

5. Human aging as loss of youthful metabolic plasticity

5.1 Aerobic glycolysis provides a more specific metabolic phenotype than FDG alone

Most discussions of AD metabolism emphasize reduced FDG uptake. FDG is important but does not distinguish glucose used through oxidative metabolism from glucose used outside oxidative phosphorylation. Aerobic glycolysis (AG)—glycolytic glucose use despite available oxygen—supports biosynthesis, lipid production, neurite growth, redox functions, and plasticity. It is prominent in developmentally prolonged human association regions.

Goyal et al. (2017) measured glucose uptake, oxygen utilization, and blood flow in 205 cognitively normal adults aged 20–82. Age-related glucose decline exceeded oxygen-use decline, producing a global loss of AG. The spatial pattern changed most in regions with high AG in young adults and with prolonged developmental transcriptional features. By contrast, the topographies of total glucose use, oxygen consumption, and blood flow were relatively stable. Aging therefore does not produce only a uniform reduction in fuel use. It selectively reduces a youthful metabolic signature associated with plasticity and neoteny.

This is strong evidence for patterned metabolic aging and is among the best matches to the cerebral-thrift prediction. It still does not reveal purpose. Reduced AG could reflect regulated withdrawal, vascular injury, loss of astrocytic support, impaired biosynthesis, or declining cellular health.

5.2 Preservation of youthful glycolysis predicts resilience

The direction of benefit is especially important. In a later multitracer PET study, cognitive impairment was strongly associated with loss of the youthful AG pattern. Cognitively unimpaired amyloid-positive adults showed relative preservation of youthful AG, exceeding that seen in some amyloid-negative older adults. White-matter-hyperintensity burden was associated with loss of youthful AG, whereas analogous patterns were not observed for oxygen metabolism or blood flow (Goyal et al., 2023). In another study, lower AG among amyloid-positive individuals was associated with higher tau deposition, while total glucose and oxygen metabolism did not show the same relationship (Vlassenko et al., 2018).

These results argue against a monotonic model in which progressively lower plastic metabolism remains adaptive. They favor a biphasic interpretation. A moderate adult reduction from developmentally extreme AG might reduce cost while established function remains sufficient. Under amyloid, vascular, or inflammatory stress, however, preservation of residual metabolic flexibility becomes protective. Excessive withdrawal is failure of resilience, not successful thrift.

5.3 Standard hypometabolism does not generally precede amyloid

The original model emphasized hypometabolism as an initiating process. Current longitudinal evidence requires correction. In a 20-year nested case-control study, CSF Aβ42 diverged from controls approximately 18 years before clinical diagnosis, followed by Aβ42/Aβ40, p-tau181, total tau, neurofilament light, hippocampal atrophy, and cognition (Jia et al., 2024). Autosomal-dominant and ADNI cohorts similarly place amyloid alteration early, with neurodegeneration and standard hypometabolism emerging later in the canonical sequence (Gordon et al., 2018; Schaap et al., 2024).

An upstream thrift state could still involve declining AG, altered substrate allocation, or network regulation that FDG does not capture. Alternatively, early hyperactivity and amyloid may trigger compensatory downregulation. What is no longer defensible is the broad assertion that measured AD hypometabolism generally precedes amyloid and tau.

5.4 Preserved ketone use reveals a glucose-specific deficit in viable tissue

Dual-tracer PET separates failure to use glucose from failure to metabolize fuel generally. Cognitively healthy older adults, people with MCI, and people with early AD show a disease-related glucose deficit while acetoacetate uptake remains preserved or relatively increased (Croteau et al., 2018). This substrate dissociation suggests that affected networks contain tissue capable of importing and oxidizing alternative fuel even while glucose utilization is suppressed.

Intervention studies test whether the retained pathway is functionally meaningful. In a six-month randomized trial, ketogenic MCT increased brain ketone metabolism by approximately 230% without changing glucose uptake. Improvements in episodic memory, language, executive function, and processing speed were related to increased ketone uptake (Fortier et al., 2019). The small 2026 BREAK-AD pilot increased ketone uptake 2.4-fold in gray matter and 3.1–3.6-fold in selected white-matter tracts. Glucose uptake again did not change. Fornix myelin measures declined by as much as 10% on placebo but were preserved in the active ketone-plus-nicotinamide-riboside group; processing speed covaried with ketone uptake and myelin measures (Roy et al., 2026).

These trials do not establish that glucose hypometabolism is deliberately adaptive. Ketones and nicotinamide riboside have signaling, inflammatory, and redox effects, and the BREAK-AD sample contained only 15 completers per arm with baseline imbalance and tolerability problems. The results nevertheless show that the metabolic phenotype is partly bypassable without repairing glucose uptake. That is consistent with viable, underfueled, or selectively throttled infrastructure rather than uniform terminal metabolic incapacity.

6. Amyloid-β: feedback, facilitation, toxicity, and containment

6.1 Physiological Aβ is not a uniform brake

Aβ is generated through normal APP processing and is coupled to neuronal activity. Increasing neuronal activity raises interstitial Aβ, and chronic reduction of activity reduces amyloid deposition in transgenic models (Bero et al., 2011; Yuan and Grutzendler, 2016). This relationship makes feedback regulation plausible, but the direction of Aβ’s effects depends on concentration, assembly state, duration, and circuit context.

The feedback interpretation now has direct causal evidence. In organotypic hippocampal tissue, APP-null dentate granule cells failed to strengthen their excitatory synapses homeostatically after prolonged activity blockade. Blocking either β- or γ-secretase, or scavenging endogenous Aβ, also prevented scaling in wild-type tissue. Aβ40 or Aβ42 restored the response through NMDAR-, CaMKII-, and synaptopodin-dependent machinery (Galanis et al., 2021). Separate depletion-and-rescue experiments found that antibody removal of endogenous Aβ or APP knockdown impaired CA1 LTP, contextual fear memory, and reference memory, while human Aβ42 restored these functions (Puzzo et al., 2011). Synaptic activity itself changes interstitial Aβ within minutes to hours through vesicle exocytosis (Cirrito et al., 2005), and changing endogenous Aβ in either direction perturbs short-term facilitation (Abramov et al., 2009).

Together these results identify Aβ as part of a physiological synaptic controller with an optimal range. The Galanis experiment concerns compensatory up-scaling after inactivity, not energy-saving down-scaling, so it is not a thrift clincher by itself. Its importance is more precise: amyloidogenic APP processing is causally embedded in homeostatic plasticity, making it biologically plausible that chronic Alzheimer pathology is an overdriven, mistimed, or uncoupled version of normal feedback rather than an entirely foreign toxic process.

Low-dose Aβ42 reduced excitability and network synchronization in rat prefrontal slices, whereas prolonged high-dose exposure produced increased excitability and tonic firing (Wang et al., 2009). Brief picomolar oligomer exposure can enhance plasticity and contextual memory; longer exposure impairs plasticity while increasing spontaneous neurotransmitter release and functional presynaptic sites (Koppensteiner et al., 2016). Low picomolar Aβ can increase synaptic-vesicle recycling, whereas depletion or micromolar exposure reduces it (Lazarevic et al., 2017). Endogenous monomeric Aβ is also required for stress-induced neuronal aerobic glycolysis in culture (Santangelo et al., 2021).

These findings are incompatible with a simple claim that Aβ evolved to suppress synapses and save energy. They support a hormetic regulator whose chronic dysregulation can produce either excessive inhibition or hyperexcitability. A cerebral-thrift model must specify Aβ species and temporal stage.

6.2 Aggregated amyloid contributes causally to clinical progression

Randomized anti-amyloid trials constrain any interpretation that treats amyloid as merely protective or incidental. In the Clarity AD trial, lecanemab reduced amyloid burden by 59.1 centiloids relative to placebo and slowed worsening on the Clinical Dementia Rating–Sum of Boxes by 0.45 points over 18 months, while causing amyloid-related imaging edema or effusions in 12.6% of treated participants (van Dyck et al., 2023). Donanemab likewise produced substantial plaque clearance and significantly slowed integrated cognitive and functional decline, particularly in participants with lower or intermediate tau burden, while also producing ARIA and serious adverse events (Sims et al., 2023).

The clinical effects are modest relative to the magnitude of plaque removal, demonstrating that amyloid is neither sufficient to explain the whole disease nor irrelevant. The most compatible integrated interpretation is that physiological Aβ has normal functions, pathogenic assemblies contribute causally to downstream disease, and later tau, vascular, immune, and proteostatic processes limit the benefit of amyloid removal.

6.3 Plaque compaction can be protective within a pathogenic process

Plaque biology is stage dependent. Microglial processes form a barrier around plaques, increasing compaction and limiting outward protofibrillar Aβ42 hotspots. Uncovered regions bind soluble Aβ42 more avidly and exhibit more severe axonal dystrophy. Aging reduces microglial coverage, enlarging protofibrillar halos and worsening local injury (Condello et al., 2015). Recent depletion and xenograft experiments clarify a dual role: homeostatic microglia help seed early plaques, whereas activated microglia later compact plaques and limit neuritic dystrophy. Human microglia carrying TREM2 R47H show impaired late protective responses (Baligács et al., 2024).

Plaque compaction therefore can contain more toxic species without making the overall amyloid cascade adaptive. Local containment is compatible with chronic pathology. This distinction is more defensible than claiming that plaques themselves are energy-saving structures.

6.4 Proteostasis failure offers a strong alternative account of plaque genesis

Autophagy and lysosomal biology supply an alternative route from stress response to amyloid pathology. In five APP-based AD mouse models, autolysosome acidification declined before extracellular amyloid deposition. Aβ and APP-βCTF accumulated in enlarged, poorly acidified autolysosomes; advanced neurons developed PANTHOS structures, lysosomal leakage, glial invasion, cell death, and extracellular plaques. Related profiles were observed in human AD tissue (Lee et al., 2022). Earlier work similarly found autophagy activation before extracellular plaques, followed by accumulation of immature autophagic vacuoles rich in Aβ-generating machinery (Yu et al., 2005).

This evidence can explain plaque formation through failed clearance rather than purposeful deposition. It remains compatible with the general motif “initially protective response becomes pathological,” but it does not require energy conservation. The thrift hypothesis must add discriminating evidence: reduced expenditure with selective functional preservation.

7. Glial pruning: ancient machinery, experimentally harmful activation

Microglia and complement reuse mechanisms that sculpt developing neural circuits. In AD mouse models, C1q rises at synapses before overt plaques. Blocking C1q, C3, or the microglial complement receptor CR3 reduces microglial engulfment and early synapse loss; C1q is required for Aβ-oligomer-induced LTP impairment (Hong et al., 2016). Astrocytes also eliminate excitatory and inhibitory synapses in a C1q-dependent manner, and C1q deletion rescues synaptic density in tau and combined amyloid–tau models (Dejanovic et al., 2022).

These studies demonstrate reuse of ancient pruning pathways. They do not currently demonstrate a beneficial moderate range in AD. In the experimental conditions, pathway activation is harmful, and inhibition is protective. The evidence therefore supports inappropriate reactivation or chronic overrun more directly than strategic energy conservation.

Glial effects are also stage dependent. Microglia can remove synapses, seed plaques, compact plaques, clear debris, or generate inflammatory injury. A useful cerebral-thrift theory must predict when and where pruning is beneficial, identify the function retained, and quantify the energy saved. Invoking developmental machinery alone is insufficient because many diseases reactivate developmental pathways without recapitulating their original adaptive role.

8. Cognitive selectivity: acquisition, retention, and access

8.1 Average-domain dissociations are real but coarse

Fluid reasoning, episodic encoding, working memory, and rapid learning generally decline earlier than vocabulary, accumulated knowledge, and practiced skills. Longitudinal structural work has found fluid decline alongside relative stability of crystallized measures, with fluid ability more closely associated with cortical thickness (Yuan et al., 2018). A meta-analysis of 17 studies found that procedural-learning differences between aMCI/AD groups and healthy controls were statistically and clinically trivial, although formal equivalence was not established (De Wit et al., 2021).

These findings are compatible with selective economization. They are not uniquely predicted by it. Procedural systems depend more heavily on basal-ganglia and cerebellar circuits that are affected later in typical AD. Regional disease anatomy can generate the same dissociation.

8.2 Crystallized knowledge is not an independent compensatory system

Two large longitudinal cohorts showed that rates of fluid and crystallized change are strongly correlated within individuals. People with greater fluid decline showed smaller crystallized gains or actual crystallized decline (Tucker-Drob et al., 2022). Semantic fluency can decline early in people at elevated AD risk and tracks hippocampal and association-cortex atrophy and hypometabolism (Vonk et al., 2020). Broad claims that semantic or crystallized function is “spared” are therefore misleading.

Some apparent semantic loss reflects impaired access, executive search, processing speed, or episodic support rather than erasure of stored representations. That distinction is central to the refined prediction. The appropriate contrast is not “fluid versus crystallized” in the abstract. It is retention and execution of already consolidated policies versus acquisition, updating, recombination, and flexible access, after matching tasks for difficulty, motivation, sensory-motor requirements, and regional pathology.

8.3 A decisive cognitive design

A strong test would teach participants or animals matched task families years, months, days, or minutes before metabolic manipulation. Performance would then separate storage, retrieval, new acquisition, reversal learning, schema updating, and transfer. Synaptic-density PET or longitudinal spine imaging would identify whether labile connections are preferentially removed. The cerebral-thrift prediction is specific: energy reduction should impair new construction and updating more than execution of consolidated behavior, even when task difficulty is matched.

9. Comparative biology: conserved components and divergent endpoints

9.1 Hibernators reveal reversible machinery

Hibernators provide the strongest comparative evidence because the low-energy state, molecular changes, structural remodeling, and recovery are observed within the same biological cycle. Their relevance lies in entry and exit mechanisms, not simply in the presence of an AD-associated marker.

9.2 Naked mole-rats show aggregation resistance, not plaque tolerance

Naked mole-rat brains contain high soluble Aβ levels comparable to those of 3xTg-AD mice but show no extracellular plaques or age-related Aβ increase in the examined animals. Naked mole-rat Aβ differs from human Aβ by one residue and aggregates less readily, although both peptides were similarly toxic to mouse hippocampal neurons in vitro (Edrey et al., 2013). Naked mole-rats are therefore models of aggregation resistance and tolerance. They should not be described as carrying abundant plaques without impairment.

9.3 Degus exhibit heterogeneous natural susceptibility

AD-like findings in Octodon degus vary across colonies and genetic backgrounds. A recent study found amyloid, tau, and glial pathology in a cognitively impaired subset of aged outbred degus, while age-matched behaviorally normal animals showed less pathology (Tan et al., 2022). This is valuable as a model of spontaneous heterogeneous vulnerability, but it is not evidence for a universal species-wide thrift program.

9.4 Nonhuman primates separate amyloid from severe human AD

Aged nonhuman primates commonly develop diffuse amyloid plaques or cerebral amyloid angiopathy, but severe tau pathology and dementia are uncommon. Comparative analyses suggest that amyloid occurrence scales with maximum lifespan, whereas tau tangles remain rare and tend to appear near the limits of nonhuman-primate life expectancy (Bouras et al., 2025). Full amyloid–tau coupling and widespread neurodegeneration are unusually human.

Rhesus macaques also support a network-demand rival. Association-cortex neurons require calcium–cAMP–PKA signaling for persistent firing. With age, calcium leak, AD-relevant p-tau, impaired firing, and cognitive decline become linked; ryanodine-receptor blockade improves neuronal activity and performance (Datta et al., 2021). The same mechanisms that enable high-level cognition may generate vulnerability when regulation fails, without being strategically downregulated.

Comparative evidence therefore supports ancient molecular components and species-specific endpoints. Ubiquity is not evidence of adaptive value. The critical comparative variable is recovery capacity.

10. The evolutionary premise requires major revision

10.1 Ancestral adults did not normally die at 55

The claim that modern longevity exposes a program that ancestral adults rarely lived long enough to experience has often relied on an average “terminal age” near 55. Cross-cultural demographic evidence does not support that value. Gurven and Kaplan (2007) estimated an average modal adult age at death of approximately 72 years across small-scale populations, with a range near 68–78. Adults who reached 45 averaged about two further decades of life. Modern populations have greatly increased survival into the eighth through tenth decades, but meaningful survival beyond 55 is not uniquely modern.

Selection against effects expressed after reproduction remains weak, especially at advanced ages. It is not necessarily absent. “Modern-lifespan overrun” is most defensible as increased frequency and duration of exposure to ages 75–100, not as a claim that ancestral humans usually died by 55.

10.2 Older adults often remain net producers

The second problem concerns productivity. Tsimane hunting efficiency peaks around the 40s, but net family production is often positive from approximately age 40 into the 70s or later. Older women can be highly productive as childcare constraints diminish, and middle-aged or older adults contribute food, manufacturing skill, childcare, social knowledge, and conflict mediation (Gurven et al., 2012, 2017; Hooper et al., 2015). Embodied capital generates late returns.

These data are consistent with a shift from acquisition toward exploitation. They oppose a simple claim that older brains became net caloric liabilities after 45. Selection should preserve the expertise enabling late-life transfers. Any adaptive reduction must therefore be narrow: reduced cost of open-ended plasticity without loss of productive established knowledge.

10.3 APOE4 supplies a measurable early-benefit/late-cost tradeoff

The strongest common genetic risk factor for late-onset AD has documented environment-dependent benefits. Among Tsimane adults, APOE4 interacted positively with parasite burden: older carriers under high eosinophil counts maintained or slightly improved cognitive performance, whereas noncarriers declined (Trumble et al., 2017). In rural Ghana, APOE4 showed no significant survival disadvantage and, among women exposed to high pathogen levels, one allele predicted approximately one additional child and two alleles approximately 3.5 additional children (van Exel et al., 2017). A larger Tsimane natural-fertility study found that carriers had 0.3–0.5 more children, began reproduction 0.8 years earlier, and had birth intervals about 0.23 years shorter; homozygotes had an estimated 1.4–2.1 additional children (Trumble et al., 2023).

These findings offer real antagonistic pleiotropy: an allele can improve fitness or cognitive resilience in infection-rich environments while increasing late-life dementia risk in industrial settings. They do not show that cerebral thrift itself is the advantageous trait. The benefit may be mediated by lipid transport, immunity, infection tolerance, or reproduction. The sharper hypothesis is therefore genotype-by-environment: APOE4 may alter the threshold, gain, or recoverability of neural resource-allocation and immune-containment programs that were beneficial under ancestral infection and fertility regimes but become damaging under long modern lifespans and low-pathogen, metabolically different conditions.

10.4 Selection shadow is not positive adaptation

Late expression can explain why destructive pathology persists under weak selection. It does not show that the earlier phenotype was favored. Positive adaptation requires evidence that moderate cerebral thrift improved survival, reproduction, or inclusive fitness. A complete evolutionary demonstration would need age-specific energy savings, preserved production or care, corrected mortality and transfer schedules, and formal selection modeling. None currently exists.

11. Is the energetic benefit large enough?

Neural signaling is expensive. Updated cortical energy budgets estimate that synaptic processes consume approximately 59% of signaling energy and that signaling constitutes roughly 75% of total cortical energy when housekeeping is included (Howarth et al., 2012). This places synaptic processes near 44% of the modeled cortical total. Large savings are physiologically possible.

Direct cellular measurements add an important piece that model-based budgets miss. Resting hippocampal nerve terminals consumed approximately 3,100 ATP molecules per terminal per second even when action potentials were blocked. Continuous proton leakage from synaptic vesicles drove V-ATPase activity accounting for up to 44% of resting terminal ATP consumption; suppressing this burden improved tolerance of fuel withdrawal (Pulido & Ryan, 2021). Thus a maintained but electrically quiet synapse is not metabolically free. Disassembling vesicle pools can, in principle, save continuous maintenance expenditure in addition to activity-related ion pumping.

An illustrative calculation shows the scale, but should not be mistaken for evidence. A 20-W human brain uses approximately 415 kcal/day. If synaptic processes account for 44%, they would use approximately 183 kcal/day. Reducing that expenditure by 20% would save roughly 37 kcal/day; a 50% reduction would save approximately 91 kcal/day. Such values could matter during sustained scarcity.

The calculation rests on unverified assumptions. Rodent cortical budgets may not scale directly to humans. Synapse loss may reduce some costs while increasing inflammatory, repair, and proteostatic expenditure. FDG decline does not necessarily equal lower ATP turnover, and energy may be reallocated rather than removed from whole-body expenditure. Direct resting-terminal measurements establish a real cost but do not reveal how much aging, torpor, or Alzheimer-related remodeling saves in vivo. No study located here simultaneously measures synaptic density, absolute glucose and oxygen use, alternative substrates, whole-body calorimetry, and cognition during a proposed thrift transition. The whole-brain energetic claim therefore remains open, but it now has a measurable cellular foundation.

12. Rival models and the explanatory role of cerebral thrift

The amyloid cascade explains early biomarker ordering and is supported by modest clinical benefit from amyloid removal. Tau propagation explains the spatial relationship between aggregated tau and future atrophy (La Joie et al., 2020). Vascular models explain substrate failure and blood–brain-barrier injury that can appear independently of canonical biomarkers (Nation et al., 2019). Mitochondrial models explain declining bioenergetic capacity and altered APP processing. Insulin-resistance models explain impaired glucose signaling, though a recent meta-analysis of five intranasal-insulin trials found no significant benefit on major cognitive or functional outcomes (Silva et al., 2026). Immune models explain complement-mediated synapse loss. Proteostasis models explain intracellular Aβ buildup and plaque genesis. Network-demand models explain why highly active association systems are vulnerable.

Cerebral thrift need not replace these mechanisms. Its potential contribution is to explain why early disease uses processes that can reduce plasticity and why loss of recovery may convert regulated remodeling into degeneration. To earn that role, it must predict more than a generic stress response. The distinct predictions are measurable energy savings, selective preservation, age-scheduled deployment, and failure of exit.

13. Decisive experiments

13.1 The aged entry–exit experiment

The most informative animal study would cross age (young versus old) with AD risk (wild-type versus amyloid/tau-risk model) and expose animals to factorially separated cooling, fasting, torpor, refeeding, and arousal. Measurements would include exact tau phosphosites and MTBR fragments, PP2A and kinase activity, axonal transport, longitudinally identified stable and labile spines, SV2A, absolute ATP/glucose/oxygen/ketone flux, electrophysiology, and memory encoded both before and after the intervention.

The clinching result would show that entry reduces expenditure, preferentially downscales labile synapses, and preserves pre-existing memory; young animals would reverse the state, whereas aged or risk animals would show delayed dephosphorylation, incomplete structural restoration, or failed metabolic rebound. If p-tau appears without savings or selectivity and is fully explained by temperature, the broader thrift mechanism would be weakened.

13.2 A human longitudinal plastic-metabolism study

Cognitively normal adults spanning early adulthood to late life should be followed with AG, FDG, oxygen, ketone, and perfusion imaging; SV2A PET; amyloid and tau biomarkers; sleep and arousal measures; vascular and insulin markers; and tasks separating acquisition from long-term retention. The key analysis would test whether loss of plastic metabolism precedes selective acquisition decline after controlling for perfusion, pathology, and atrophy, and whether later disease occurs specifically when substrate flexibility, sleep clearance, or synaptic restoration fails.

13.3 Direct energy accounting

An intervention that changes synaptic plasticity without killing neurons should be paired with absolute brain-flux measurement and whole-body calorimetry. The decisive outcome is net energy reduction large enough to matter outside the brain, with established performance preserved and reversibility demonstrated.

13.4 Comparative recovery index

Cross-species work should measure dephosphorylation, MTBR aggregation, lysosomal acidification, plaque compaction, synaptic rebuilding, substrate flexibility, and cognitive recovery. Experimentally weakening a resilience mechanism in a normally protected species should convert reversible remodeling into persistent pathology. This would be more informative than cataloging plaques and tangles.

13.5 Formal evolutionary test

Empirically measured energy savings and performance effects should enter a life-history model using defensible mortality, fertility, productivity, and kin-transfer schedules. The model must show that the moderate phenotype increases inclusive fitness over a plausible range of ancestral ecologies. Failure to produce such an advantage would favor conserved stress physiology plus selection shadow over positive cerebral-thrift adaptation.

14. Conclusion: the critical variable may be exit

Research since 2009 supports an important part of the cerebral-thrift intuition. Early AD-associated molecular features are not intrinsically synonymous with irreversible degeneration. Mammalian brains use Alzheimer-relevant tau phosphosites during reversible low-energy states, disassemble and rebuild synapses within hours, and can preserve tested memory despite substantial structural regression. Site-specific tau phosphorylation can constrain extraneous neuronal recruitment, support remote-memory formation, protect against Aβ excitotoxic signaling, and implement physiological synaptic weakening. Early-AD mouse engrams can retain information that natural cues no longer access. Human aging selectively loses a youthful plastic metabolic phenotype, and resilience depends on retaining metabolic, vascular, glial, and proteostatic capacity.

The evidence does not currently establish that natural cognitive aging is a positively selected energy-conservation program. The original demographic and productivity premises require revision. Physiological Aβ is not a uniform suppressor; amyloid removal modestly slows disease; aggregated tau predicts atrophy; complement pruning is harmful in disease models; and continuing loss of youthful AG predicts impairment rather than benefit. A maintained resting terminal now has a directly measured ATP burden, but the whole-brain energy saved by age-related or tau-linked economization has not been measured.

The strongest evidence-based reformulation is narrower but consequential: AD may overrun molecular machinery that mammalian brains use for reversible neural economization, and aging may increase disease risk by degrading the capacity to exit, rebuild, contain, refuel, and compensate. RBM3 experiments already show that Alzheimer-model mice lose synapse-regeneration capacity before established synapse loss and that restoring the recovery factor changes disease outcome. Preserved human ketone uptake and preliminary fuel-rescue trials show that glucose-hypometabolic tissue retains metabolically usable capacity. This “failure-of-exit” model does not assume adaptive dementia. It identifies a missing dimension in AD research and generates experiments capable of confirming or falsifying the broader evolutionary claim.

The central economization step is no longer hypothetical in mammalian cortex. Food scarcity can reduce AMPAR conductance and cortical ATP use while preserving spike output and sacrificing fine coding precision, and the state is reversible with restored energy signaling. Sleep independently removes AMPARs and shrinks mostly smaller synapses while restoring human plastic capacity. Early Alzheimer models show age-related AMPAR downscaling and selective loss of flexible learning. What remains unproven is that these are stages of one age-scheduled human process rather than convergent uses and abuses of shared synaptic machinery.

If the remaining entry–exit chain is demonstrated—scarcity- or age-sensitive economization, Alzheimer-relevant soluble p-tau, selective labile-synapse downscaling, preserved consolidated function, and failure of leptin, sleep, RBM3, arousal, or substrate-mediated recovery—the adaptive cerebral-thrift hypothesis would become difficult to ignore. If those links fail, the same experiments will clarify how generic stress responses, metabolic injury, and proteostatic failure create superficial resemblance to adaptive states. Either outcome would advance the biology of Alzheimer’s disease.

References

Arendt, T., Stieler, J., Strijkstra, A. M., et al. (2003). Reversible paired helical filament-like phosphorylation of tau is an adaptive process associated with neuronal plasticity in hibernating animals. Journal of Neuroscience, 23, 6972–6981. https://doi.org/10.1523/JNEUROSCI.23-18-06972.2003

Baligács, N., Albertini, G., Borrie, S. C., Serneels, L., Pridans, C., Balusu, S., & De Strooper, B. (2024). Homeostatic microglia initially seed and activated microglia later reshape amyloid plaques in Alzheimer’s disease. Nature Communications, 15, 10634. https://doi.org/10.1038/s41467-024-54779-w

Barthélemy, N. R., et al. (2020). A soluble phosphorylated tau signature links tau, amyloid and the evolution of stages of dominantly inherited Alzheimer’s disease. Nature Medicine, 26, 398–407. https://doi.org/10.1038/s41591-020-0781-z

Bero, A. W., et al. (2011). Neuronal activity regulates the regional vulnerability to amyloid-β deposition. Nature Neuroscience, 14, 750–756. https://doi.org/10.1038/nn.2801

Bouras, E., et al. (2025). The biological scaling of Alzheimer’s disease neuropathological changes across primate species. Alzheimer’s Research & Therapy, 17, 242. https://doi.org/10.1186/s13195-025-01889-2

Bullmann, T., et al. (2016). Tau phosphorylation-associated spine regression does not impair hippocampal-dependent memory in hibernating golden hamsters. Hippocampus, 26, 301–318. https://doi.org/10.1002/hipo.22522

Brum, W. S., Montoliu-Gaya, L., Brinkmalm, G., et al. (2025). Reversible tau hyperphosphorylation in hibernation: A blood biomarker and brain tissue study. Acta Neuropathologica, 150, 36. https://doi.org/10.1007/s00401-025-02930-2

Condello, C., Yuan, P., Schain, A., & Grutzendler, J. (2015). Microglia constitute a barrier that prevents neurotoxic protofibrillar Aβ42 hotspots around plaques. Nature Communications, 6, 6176. https://doi.org/10.1038/ncomms7176

Datta, D., et al. (2021). Age-related calcium dysregulation linked with tau pathology and impaired cognition in non-human primates. Alzheimer’s & Dementia, 17, 920–932. https://doi.org/10.1002/alz.12325

Dejanovic, B., et al. (2022). Complement C1q-dependent excitatory and inhibitory synapse elimination by astrocytes and microglia in Alzheimer’s disease mouse models. Nature Aging, 2, 837–850. https://doi.org/10.1038/s43587-022-00281-1

De Wit, L., et al. (2021). Procedural learning in individuals with amnestic mild cognitive impairment and Alzheimer’s dementia: A systematic review and meta-analysis. Neuropsychology Review, 31, 103–114. https://doi.org/10.1007/s11065-020-09449-1

de Veij Mestdagh, C. F., et al. (2021). Torpor enhances synaptic strength and restores memory performance in a mouse model of Alzheimer’s disease. Scientific Reports, 11, 15486. https://doi.org/10.1038/s41598-021-94992-x

Edrey, Y. H., et al. (2013). Amyloid beta and the longest-lived rodent: The naked mole-rat as a model for natural protection from Alzheimer’s disease. Neurobiology of Aging, 34, 2352–2360. https://doi.org/10.1016/j.neurobiolaging.2013.03.032

Gordon, B. A., et al. (2018). Spatial patterns of neuroimaging biomarker change in individuals from families with autosomal dominant Alzheimer’s disease: A longitudinal study. The Lancet Neurology, 17, 241–250. https://doi.org/10.1016/S1474-4422(18)30028-0

Goyal, M. S., et al. (2017). Loss of brain aerobic glycolysis in normal human aging. Cell Metabolism, 26, 353–360.e3. https://doi.org/10.1016/j.cmet.2017.07.010

Goyal, M. S., et al. (2023). Brain aerobic glycolysis and resilience in Alzheimer disease. Proceedings of the National Academy of Sciences, 120, e2212256120. https://doi.org/10.1073/pnas.2212256120

Gurven, M., & Kaplan, H. (2007). Longevity among hunter-gatherers: A cross-cultural examination. Population and Development Review, 33, 321–365. https://doi.org/10.1111/j.1728-4457.2007.00171.x

Gurven, M., et al. (2012). From the womb to the tomb: The role of transfers in shaping the evolved human life history. Experimental Gerontology, 47, 807–813. https://doi.org/10.1016/j.exger.2012.05.006

Gurven, M., et al. (2017). The Tsimane Health and Life History Project: Integrating anthropology and biomedicine. Evolutionary Anthropology, 26, 54–73. https://doi.org/10.1002/evan.21515

Hooper, P. L., et al. (2015). Inclusive fitness and differential productivity across the life course determine intergenerational transfers in a small-scale human society. Proceedings of the Royal Society B, 282, 20142808. https://doi.org/10.1098/rspb.2014.2808

Hong, S., et al. (2016). Complement and microglia mediate early synapse loss in Alzheimer mouse models. Science, 352, 712–716. https://doi.org/10.1126/science.aad8373

Hitrec, T., Squarcio, F., Cerri, M., et al. (2021). Reversible tau phosphorylation induced by synthetic torpor in the spinal cord of the rat. Frontiers in Neuroanatomy, 15, 592288. https://doi.org/10.3389/fnana.2021.592288

Howarth, C., Gleeson, P., & Attwell, D. (2012). Updated energy budgets for neural computation in the neocortex and cerebellum. Journal of Cerebral Blood Flow & Metabolism, 32, 1222–1232. https://doi.org/10.1038/jcbfm.2012.35

Jia, J., et al. (2024). Biomarker changes during 20 years preceding Alzheimer’s disease. New England Journal of Medicine, 390, 712–722. https://doi.org/10.1056/NEJMoa2310168

La Joie, R., et al. (2020). Prospective longitudinal atrophy in Alzheimer’s disease correlates with the intensity and topography of baseline tau-PET. Science Translational Medicine, 12, eaau5732. https://doi.org/10.1126/scitranslmed.aau5732

Koppensteiner, P., et al. (2016). Time-dependent reversal of synaptic plasticity induced by physiological concentrations of oligomeric Aβ42. Scientific Reports, 6, 32553. https://doi.org/10.1038/srep32553

Lazarevic, V., et al. (2017). Physiological concentrations of amyloid beta regulate recycling of synaptic vesicles via α7 acetylcholine receptor and CDK5/calcineurin signaling. Frontiers in Molecular Neuroscience, 10, 221. https://doi.org/10.3389/fnmol.2017.00221

Lee, J.-H., et al. (2022). Faulty autolysosome acidification in Alzheimer’s disease mouse models induces autophagic build-up of Aβ in neurons, yielding senile plaques. Nature Neuroscience, 25, 688–701. https://doi.org/10.1038/s41593-022-01084-8

Luppi, M., et al. (2019). Phosphorylation and dephosphorylation of tau protein during synthetic torpor. Frontiers in Neuroanatomy, 13, 57. https://doi.org/10.3389/fnana.2019.00057

Magariños, A. M., et al. (2006). Rapid and reversible changes in intrahippocampal connectivity during the course of hibernation in European hamsters. Proceedings of the National Academy of Sciences, 103, 18775–18780. https://doi.org/10.1073/pnas.0608785103

Nation, D. A., et al. (2019). Blood–brain barrier breakdown is an early biomarker of human cognitive dysfunction. Nature Medicine, 25, 270–276. https://doi.org/10.1038/s41591-018-0297-y

Planel, E., et al. (2007). Anesthesia leads to tau hyperphosphorylation through inhibition of phosphatase activity by hypothermia. Journal of Neuroscience, 27, 3090–3097. https://doi.org/10.1523/JNEUROSCI.4854-06.2007

Reser, J. E. (2009). Alzheimer’s disease and natural cognitive aging may represent adaptive metabolism reduction programs. Behavioral and Brain Functions, 5, 13. https://doi.org/10.1186/1744-9081-5-13

Reser, J. E. (2026). Adaptive cerebral thrift: Alzheimer’s disease as the modern-lifespan extreme of natural cognitive aging. Iterated Insights. https://iteratedinsights.com/2026/06/26/adaptive-cerebral-thrift-alzheimers-disease-as-the-modern-lifespan-extreme-of-natural-cognitive-aging/

Schaap, T., et al. (2024). Timing of Alzheimer’s disease biomarker progressions: A two-decade observational study from ADNI. Alzheimer’s & Dementia, 20, 9060–9067. https://doi.org/10.1002/alz.14306

Santangelo, R., Giuffrida, M. L., Satriano, C., Tomasello, M. F., Zimbone, S., & Copani, A. (2021). β-Amyloid monomers drive up neuronal aerobic glycolysis in response to energy stressors. Aging, 13, 18033–18050. https://doi.org/10.18632/aging.203330

Silva, A. M. P., et al. (2026). Intranasal insulin for mild cognitive impairment and Alzheimer’s disease: A systematic review and meta-analysis of randomized controlled trials. Revue Neurologique, 182, 36–48. https://doi.org/10.1016/j.neurol.2025.11.005

Sims, J. R., et al. (2023). Donanemab in early symptomatic Alzheimer disease: The TRAILBLAZER-ALZ 2 randomized clinical trial. JAMA, 330, 512–527. https://doi.org/10.1001/jama.2023.13239

Stieler, J. T., Bullmann, T., Kohl, F., Tøien, Ø., Brückner, M. K., Härtig, W., Barnes, B. M., & Arendt, T. (2011). The physiological link between metabolic rate depression and tau phosphorylation in mammalian hibernation. PLoS ONE, 6, e14530. https://doi.org/10.1371/journal.pone.0014530

Su, B., Wang, X., Drew, K. L., Perry, G., Smith, M. A., & Zhu, X. (2008). Physiological regulation of tau phosphorylation during hibernation. Journal of Neurochemistry, 105, 2098–2108. https://doi.org/10.1111/j.1471-4159.2008.05294.x

Tan, Z., Garduño, B. M., Fernández Aburto, P., et al. (2022). Cognitively impaired aged Octodon degus recapitulate major neuropathological features of sporadic Alzheimer’s disease. Acta Neuropathologica Communications, 10, 182. https://doi.org/10.1186/s40478-022-01481-x

Tournissac, M., Bourassa, P., Martinez-Cano, R. D., Vu, T.-M., Hébert, S. S., Planel, E., & Calon, F. (2019). Repeated cold exposures protect a mouse model of Alzheimer’s disease against cold-induced tau phosphorylation. Molecular Metabolism, 22, 110–120. https://doi.org/10.1016/j.molmet.2019.01.008

Tucker-Drob, E. M., et al. (2022). A strong dependency between changes in fluid and crystallized abilities in human cognitive aging. Science Advances, 8, eabj2422. https://doi.org/10.1126/sciadv.abj2422

van Dyck, C. H., et al. (2023). Lecanemab in early Alzheimer’s disease. New England Journal of Medicine, 388, 9–21. https://doi.org/10.1056/NEJMoa2212948

Vlassenko, A. G., et al. (2018). Aerobic glycolysis and tau deposition in preclinical Alzheimer’s disease. Neurobiology of Aging, 67, 95–98. https://doi.org/10.1016/j.neurobiolaging.2018.03.014

Vonk, J. M. J., Bouteloup, V., Mangin, J.-F., et al. (2020). Semantic loss marks early Alzheimer’s disease-related neurodegeneration in older adults without dementia. Alzheimer’s & Dementia: Diagnosis, Assessment & Disease Monitoring, 12, e12066. https://doi.org/10.1002/dad2.12066

von der Ohe, C. G., et al. (2006). Ubiquitous and temperature-dependent neural plasticity in hibernators. Journal of Neuroscience, 26, 10590–10598. https://doi.org/10.1523/JNEUROSCI.2874-06.2006

von der Ohe, C. G., et al. (2007). Synaptic protein dynamics in hibernation. Journal of Neuroscience, 27, 84–92. https://doi.org/10.1523/JNEUROSCI.4385-06.2007

Yuan, P., & Grutzendler, J. (2016). Attenuation of β-amyloid deposition and neurotoxicity by chemogenetic modulation of neural activity. Journal of Neuroscience, 36, 632–641. https://doi.org/10.1523/JNEUROSCI.2531-15.2016

Yuan, P., Voelkle, M. C., & Raz, N. (2018). Fluid intelligence and gross structural properties of the cerebral cortex in middle-aged and older adults: A multi-occasion longitudinal study. NeuroImage, 172, 21–30. https://doi.org/10.1016/j.neuroimage.2018.01.032

Wang, Y., Zhang, G., Zhou, H., Barakat, A., & Querfurth, H. (2009). Opposite effects of low and high doses of Aβ42 on electrical network and neuronal excitability in the rat prefrontal cortex. PLoS ONE, 4, e8366. https://doi.org/10.1371/journal.pone.0008366

Yu, W. H., et al. (2005). Macroautophagy—A novel β-amyloid peptide-generating pathway activated in Alzheimer’s disease. Journal of Cell Biology, 171, 87–98. https://doi.org/10.1083/jcb.200505082

References added in the 12 July 2026 support-led revision

Croteau, E., Castellano, C.-A., Fortier, M., Bocti, C., Fulop, T., Paquet, N., & Cunnane, S. C. (2018). A cross-sectional comparison of brain glucose and ketone metabolism in cognitively healthy older adults, mild cognitive impairment and early Alzheimer’s disease. Experimental Gerontology, 107, 18–26. https://doi.org/10.1016/j.exger.2017.07.004

Fortier, M., Castellano, C.-A., Croteau, E., et al. (2019). A ketogenic drink improves brain energy and some measures of cognition in mild cognitive impairment. Alzheimer’s & Dementia, 15, 625–634. https://doi.org/10.1016/j.jalz.2018.12.017

Ittner, A., Chua, S. W., Bertz, J., et al. (2016). Site-specific phosphorylation of tau inhibits amyloid-β toxicity in Alzheimer’s mice. Science, 354, 904–908. https://doi.org/10.1126/science.aah6205

Kosonen, R., Stefanoska, K., Lin, Y., et al. (2026). Tau T205 phosphorylation modulates engram cell recruitment and remote memory in mice. Nature Communications. https://doi.org/10.1038/s41467-026-73207-9

Luan, Y., Wang, W., Huang, Q., et al. (2025). Synaptic loss pattern is constrained by brain connectome and modulated by phosphorylated tau in Alzheimer’s disease. Nature Communications, 16, 6356. https://doi.org/10.1038/s41467-025-61497-4

Peretti, D., Bastide, A., Radford, H., et al. (2015). RBM3 mediates structural plasticity and protective effects of cooling in neurodegeneration. Nature, 518, 236–239. https://doi.org/10.1038/nature14142

Pulido, C., & Ryan, T. A. (2021). Synaptic vesicle pools are a major hidden resting metabolic burden of nerve terminals. Science Advances, 7, eabi9027. https://doi.org/10.1126/sciadv.abi9027

Regan, P., Piers, T., Yi, J.-H., et al. (2015). Tau phosphorylation at serine 396 residue is required for hippocampal LTD. Journal of Neuroscience, 35, 4804–4812. https://doi.org/10.1523/JNEUROSCI.2842-14.2015

Roy, D. S., Arons, A., Mitchell, T. I., Pignatelli, M., Ryan, T. J., & Tonegawa, S. (2016). Memory retrieval by activating engram cells in mouse models of early Alzheimer’s disease. Nature, 531, 508–512. https://doi.org/10.1038/nature17172

Roy, M., Fortier, M., St-Pierre, V., et al. (2026). A combination of ketones and NAD+ precursor preserves white matter integrity in mild cognitive impairment. Alzheimer’s & Dementia: Translational Research & Clinical Interventions, 12, e70278. https://doi.org/10.1002/trc2.70278

Abramov, E., Dolev, I., Fogel, H., et al. (2009). Amyloid-β as a positive endogenous regulator of release probability at hippocampal synapses. Nature Neuroscience, 12, 1567–1576. https://doi.org/10.1038/nn.2433

Cirrito, J. R., Yamada, K. A., Finn, M. B., et al. (2005). Synaptic activity regulates interstitial fluid amyloid-β levels in vivo. Neuron, 48, 913–922. https://doi.org/10.1016/j.neuron.2005.10.028

Galanis, C., Fellenz, M., Becker, D., et al. (2021). Amyloid-beta mediates homeostatic synaptic plasticity. Journal of Neuroscience, 41, 5157–5172. https://doi.org/10.1523/JNEUROSCI.1820-20.2021

Levey, A. I., Qiu, D., Zhao, L., et al. (2022). A phase II study repurposing atomoxetine for neuroprotection in mild cognitive impairment. Brain, 145, 1924–1938. https://doi.org/10.1093/brain/awab452

Puzzo, D., Privitera, L., Fa’, M., et al. (2011). Endogenous amyloid-β is necessary for hippocampal synaptic plasticity and memory. Annals of Neurology, 69, 819–830. https://doi.org/10.1002/ana.22313

Trumble, B. C., Stieglitz, J., Blackwell, A. D., et al. (2017). Apolipoprotein E4 is associated with improved cognitive function in Amazonian forager-horticulturalists with a high parasite burden. FASEB Journal, 31, 1508–1515. https://doi.org/10.1096/fj.201601084R

Trumble, B. C., Charifson, M., Kraft, T., et al. (2023). Apolipoprotein-ε4 is associated with higher fecundity in a natural fertility population. Science Advances, 9, eade9797. https://doi.org/10.1126/sciadv.ade9797

van Exel, E., Koopman, J. J. E., van Bodegom, D., et al. (2017). Effect of APOE ε4 allele on survival and fertility in an adverse environment. PLOS ONE, 12, e0179497. https://doi.org/10.1371/journal.pone.0179497

Chang, E. H., Savage, M. J., Flood, D. G., et al. (2006). AMPA receptor downscaling at the onset of Alzheimer’s disease pathology in double knockin mice. Proceedings of the National Academy of Sciences, 103, 3410–3415. https://doi.org/10.1073/pnas.0507313103

de Vivo, L., Bellesi, M., Marshall, W., et al. (2017). Ultrastructural evidence for synaptic scaling across the wake/sleep cycle. Science, 355, 507–510. https://doi.org/10.1126/science.aah5982

Diering, G. H., Nirujogi, R. S., Roth, R. H., et al. (2017). Homer1a drives homeostatic scaling-down of excitatory synapses during sleep. Science, 355, 511–515. https://doi.org/10.1126/science.aai8355

Kuhn, M., Wolf, E., Maier, J. G., et al. (2016). Sleep recalibrates homeostatic and associative synaptic plasticity in the human cortex. Nature Communications, 7, 12455. https://doi.org/10.1038/ncomms12455

Lee, Y. F., Russ, A. N., Zhao, Q., et al. (2023). Optogenetic targeting of astrocytes restores slow brain rhythm function and slows Alzheimer’s disease pathology. Scientific Reports, 13, 13075. https://doi.org/10.1038/s41598-023-40402-3

Padamsey, Z., Katsanevaki, D., Dupuy, N., & Rochefort, N. L. (2022). Neocortex saves energy by reducing coding precision during food scarcity. Neuron, 110, 280–296.e10. https://doi.org/10.1016/j.neuron.2021.10.024

Padamsey, Z., Katsanevaki, D., Maeso, P., et al. (2024). Sex-specific resilience of neocortex to food restriction. eLife, 12, RP93052. https://doi.org/10.7554/eLife.93052

Plaçais, P.-Y., & Preat, T. (2013). To favor survival under food shortage, the brain disables costly memory. Science, 339, 440–442. https://doi.org/10.1126/science.1226018

Wang, J., et al. (2024). Thermoneutral temperature exposure enhances slow-wave sleep with a correlated improvement in amyloid pathology in a triple-transgenic mouse model of Alzheimer’s disease. Sleep, 47, zsae078. https://doi.org/10.1093/sleep/zsae078