What is the clinical spectrum of Alzheimer's Disease?
Alzheimer’s disease (AD) is a highly debilitating progressive neurodegenerative disorder that affects cognitive function, mood, and behaviour.
According to the World Health Organisation, AD is the 7th leading cause of death and a major cause of disability among older individuals globally 1. Its prevalence is projected to triple worldwide by 2050, posing a major societal and health care burden.
According to frameworks established by organizations like the National Institute on Aging and Alzheimer’s Association (NIA-AA), AD progresses along a continuum, and based on the level of AD-associated impairment, it can be classified as preclinical/presymptomatic stage, mild cognitive impairment, and dementia stage (that is further subdivided into mild, moderate, and severe dementia) 2. However, several aspects of this framework remain controversial, such as reliance on ATN (amyloid, tau, neurodegeneration) biomarkers as diagnostic and prognostic tools 3.
AD exists in various forms that present a high degree of clinical heterogeneity in phenotypes, age-at-onset, etiology, progression rates, biomarkers, and responses to treatments, both between and within these forms.
Sporadic AD represents about 95% of the AD cases and is categorized as late-onset AD (LOAD), occurring after age 65 4. Commonly, LOAD has an amnestic presentation characterized by early deficits in episodic memory, with non-amnestic symptoms (language impairment, behavioural changes) occurring less frequently than in early onset AD.
Early onset AD (EOAD) can be sporadic or familial and the individuals are commonly affected before age 65 with both typical amnestic (memory loss, cognitive impairment) and atypical impairments (language deficits, visuospatial dysfunction, behavioural dysregulation), that pursue a faster course than in LOAD 5.
Systematic review and meta-analysis of clinical features of LOAD and EOAD showed that individuals with EOAD had significantly poorer baseline cognitive performance and faster cognitive decline but longer survival times than those with LOAD 6.
The genetically inherited forms of AD, familial AD (FAD), are usually tied to an early disease onset before age 65, although there are also late-onset FAD cases. Besides amnestic symptoms, non-amnestic symptoms (language deficits, motor symptoms, visuospatial deficits) were frequently observed in FAD 7,8.
Comorbidities, including cerebrovascular diseases such as cerebral amyloid angiopathy (CAA), appear to occur at higher frequency in LOAD than in EOAD, influencing its progression and severity 9.
The presence of CAA in AD, characterised by Aβ deposits in the walls of small-to-medium-sized arteries, arterioles and venous vessels of the cerebral cortex and pia mater, is associated with faster cognitive decline in AD patients 10.
Depending on the AD subtype, genetic profile, lifestyle and resilience, comorbidities and other factors, AD duration can range from 2 to 20 years from diagnosis to death.
The typical histopathology of AD is characterized by the presence of extracellular neuritic plaques (NP) featuring a core of amyloid beta-peptide (Aβ) deposits, intracellular neurofibrillary tangles (NFT) composed of hyperphosphorylated and misfolded tau protein aggregates, and neurodegeneration 11.
Aβ peptides are fragments of the transmembrane amyloid precursor protein (APP), whereas tau is an axon-enriched microtubule-associated protein.
Besides Aβ, other components of NP include dystrophic neurites (axons or dendrites), activated microglia, reactive astrocytes, and complement proteins that in AD state amplify neuroinflammation and loss of synapses 12.
The spatiotemporal distribution of Aβ and tau varies depending on the AD subtype. For instance, neuroimaging studies in asymptomatic autosomal dominant AD mutation carriers (ADAD) showed that spatial distribution of Aβ and tau in ADAD diverges from the patterns typically observed in sporadic AD 13.
Inter-individual and inter-species differences in the spatiotemporal distribution of Aβ and tau lead to differential microglial responses and gene expression profiles, impacting responses to treatments for AD 14.
Interestingly, immunocytochemistry and staining of Aβ and tau showed that the amount of tau deposits in the brains of patients with AD correlated better with dementia severity than the amount of amyloid deposits 15.
Understanding the links between different components of AD histopathology, as well as between AD and other comorbidities (cerebrovascular diseases, Lewy body disease, TDP-43 proteinopathies) is currently an area of active research.
Alzheimer's disease
ICD-10 Code G30.0, G30.1, G30.8, G30.9
What do we know about the etiology of Alzheimer's Disease?
Alzheimer’s disease (AD) is a multifactorial condition associated with many risk factors, including age, cardiovascular disease, diabetes, traumatic head injury, depression, and genetic predisposition 11, 16, 17.
AD is a uniquely human disorder. The use of experimentally-induced animal models of AD, that are not representative of human physiology, and that cannot recapitulate the complex involvement of multiple causative/contributing factors, hinder progress in understanding the mechanisms of AD and bear responsibility for the lack of effective treatments for AD.
In LOAD, the pathophysiological processes arise on a background of preexisting age-related decline (synaptic loss, accumulation of oxidative stress, chronic neuroinflammation, declining vascular integrity), as suggested by cross-sectional observational study in cognitively normal individuals 18 and GWAS studies 19.
Neuropathologic examinations in deceased AD participants from longitudinal clinical studies demonstrated that, although amyloid load, tangle density, infarcts, and Lewy bodies contributed to the variation in late life cognitive decline, much of this variation appears to be caused by other key determinants, highlighting the need to better understand AD pathological and protective pathways 20.
According to the prevalent amyloid hypothesis of AD pathogenesis, accumulation of pathological forms of Aβ produced by cleavage of the amyloid precursor protein (APP) by the β- and γ-secretase enzymes in the brain is the primary pathological process, occurring upstream of tau pathology 21.
Two peptides derived from APP, Aβ40 and Aβ42, have been the focus of attention, although it is unclear whether other Aβ peptides may also play a role in AD. Given that the more hydrophobic Aβ42 decreases as it deposits in the brain tissue, a low Aβ42/Aβ40 ratio in cerebrospinal fluid and plasma is used as an indicator of early amyloid accumulation 22. Measurement of Aβ42 and Aβ40 production and clearance rate in nervous systems of AD patients, suggested that it was the lack of clearance, rather than excess production of Aβ, that is responsible for Aβ deposits in LOAD 23.
Neuroimaging studies in asymptomatic autosomal dominant AD mutation carriers showed that the temporal dynamics of AD pathology supports a ‘trigger and bullet’ hypothesis, according to which Aβ deposition triggers the conversion of tau to a toxic state, and tau leads to neurodegeneration 13.
Soluble forms of Aβ, in particular Aβ oligomers and protofibrils, that are localised outside Aβ plaques (interstitial fluid, cerebrospinal fluid) and within/near Aβ plaques, have been identified as targets for slowing down neuronal damage in AD.
In support of the hypothesis that more soluble Aβ oligomers might be more neurotoxic than highly compacted fibrillar Aβ, human-based in vitro studies indicated that Aβ oligomers purified from brains of AD patients induced hyperphosphorylation of tau, ultimately causing neuritic dystrophy 21.
It has been postulated that both neuritic plaques and neurofibrillary tangles trigger microglial cell activation, release of proinflammatory cytokines, neuronal death, synaptic alteration, oxidative stress, mitochondrial disturbance, neuroinflammation, alterations in the permeability of the blood–brain barrier, and neurovascular unit dysfunction 24.
Nonetheless, the amyloid cascade hypothesis, according to which Aβ pathology is the exclusive or the primary driver of AD pathogenesis, is regarded with scepticism, since Aβ plaque burden does not necessarily correlate with the presence of AD nor the severity of AD symptoms, and since therapies that targeted various Aβ forms did not produce the desired effect of reversal of cognitive decline in AD patients 3. Consequently, researchers are increasingly investigating the contribution of the multitude of other pathological processes, such as neuroinflammation, mitochondrial dysfunction, and vascular damage 25.
The fact that multiple common variants identified by GWAS are involved in more than one mechanism supports the hypothesis that AD and its treatment should be considered from the multi-convergent theory and not through isolated mechanisms. For example, PICALM is involved in endocytosis, autophagy, and tau propagation, CLU in lipid metabolism, immune modulation, and Aβ clearance, and CD2AP in synaptic pruning and inflammation.
To date, 76 genome-wide significant loci associated with AD risk were identified 23, 26.
Familial EOAD are caused by autosomal dominant mutations with high penetrance in genes like APP, (encoding the precursor of Aβ), PSEN1 (encoding presenilin-1 γ-secretase subunit), and less commonly PSEN2 27. Over 300 mutations (221 pathogenic) of the PSEN1 gene and 80 mutations (19 pathogenic) of the PSEN2 gene have been found 28.
Apart from the above rare variants involved in early-onset AD, familial AD (FAD) may involve common variants that have modest individual effects but which can be integrated in a polygenic risk score (PGS) to estimate the overall genetic risk, as well as rare variants with reduced penetrance.
An example of a common variant strongly associated with familial LOAD is the APOE ε4 isoform of the APOE gene which codes for glycoprotein involved in lipid transport and modulates Aβ aggregation and clearance. Depending on the mutation type (premature stop codon, SNP), ABCA7, that plays a role in lipid transport and phagocytosis, can be both a common and a rare variant found in early-onset and late-onset FAD.
Sporadic AD is likely to be driven by a complex interaction between genetic and environmental factors.
Analysis of associations between AD PGS and AD risk, age at onset, and CSF biomarkers, suggests the existence of two independent genetic entities for sporadic AD: one associated with APOE ε4 genetic variant (APOE4) and the other associated with the combination of about 75 other genetic variants. In contrast to the latter, the impact of the APOE gene, that can have a protective (APOE2) or a pathogenic (APOE4) effect, varies significantly between populations, likely due to genetic differences within the genomic region containing the APOE gene 29.
The multiple genetic variants that contribute to the risk of AD are involved in various neuropathological events, including in microglial activation, neuroinflammation, mitochondrial dysfunction, oxidative stress, dendritic structure, axonal growth and transport, synaptic dysfunction, and vascular damage 24.
In sporadic AD, the APOE4-associated form is more prevalent, since about 70% of sporadic AD patients carry at least one APOE4 allele 21. APOE4 carriers tend to present with higher amyloid/tau burden and greater amnestic impairments than non-APOE4 carriers, while in non-APOE4 carriers non-amnestic impairments tend to be more pronounced.
Epigenetic mechanisms, like DNA methylation, histone modification and chromatin regulation by long noncoding RNA, that can alter the expression of AD-causative/associated genes have also been implicated in AD 30.
How similar are human and animal nervous systems?
This is not an exhaustive list of species-specific differences, nor can one be made given their unknown full extent, but rather an example of how these differences impact the face, construct, predictive, and intrinsic validity of animal models.
Not all species-specific differences can be accounted for in animal models, as there are hundreds of them, their relevance to Alzheimer’s disease (AD) is unclear, and their interaction with other organ systems in the animal model makes it difficult to predict how they would have behaved within the human system.
Species-specific differences in brain structure and function
The idea that human neurodegenerative disorders could be modelled in animals was initially based on coarse anatomical similarities between human and animal brains. Novel tools such as multi-omics, mathematical modelling, immunostaining and neuroimaging, that enable comparison of finer traits, have revealed human-specific features of brain anatomy, morphology and physiology that confer unique vulnerability to Alzheimer's disease.
The cerebral cortex is the primary site of neural integration, playing a key role in memory processing, decision-making, conscious thought, attention, language, sensory integration, and behavioural regulation. In AD, the human cerebral cortex suffers a devastating neuronal and synaptic loss.
Neurons produce and transmit electrical signals known as action potential. Excitatory neurons release neurotransmitters, such as glutamic acid, that stimulate the postsynaptic neuron to produce an action potential and inhibitory neurons release neurotransmitters, such as gamma-aminobutyric acid (GABA), which prevent the firing of an action potential. Electrical signal processing networks, known as neural circuits, are regulated by activity-dependent feedback systems that maintain an excitatory-to-inhibitory balance and perturbations of this balance disrupt information processing 31. The way the regulatory feedback system is thought to function is that the excitatory pyramidal neurons excite inhibitory GABAergic interneurons through glutamatergic synaptic transmissions, which in turn inhibit other excitatory pyramidal neurons.
In the human neocortex, excitatory principal cells have the capacity to elicit firing in local inhibitory interneurons with a single action potential via very large glutamatergic excitatory postsynaptic potentials (VLE). The similar strength of connections has not been found in animals, suggesting that VLE are involved in human-specific cognitive functions 32.
A specific subclass of human excitatory neurons, called von Economo neurons (VEN), primarily found in layer 5 of the frontoinsular and anterior cingulate cortex, are rare or absent in most animal species 33. The frontoinsular cortex and anterior cingulate cortex are the core hubs of the salience network that is crucial for self-awareness, decision-making and social behaviour. The AD-related neurodegeneration affecting VEN is therefore likely to produce uniquely human features of cognitive decline.
The cortical supragranular layers 2/3 are expanded in humans compared to mice, reflecting enhanced cortical connectivity in humans. This expansion includes diversification of neuron types and in particular those involved in long-range communications between different regions of the human brain.
Comparison of electrophysiological and morphological properties of layer 2/3 pyramidal neurons across species indicated that human pyramidal neurons had a higher action potential threshold voltage, a lower input resistance, and larger dendritic arbors 34.
Comparison of patch-seq and transcriptomics data in mice, marmoset monkeys and humans revealed the presence of two glutamatergic pyramidal neuron types, expressing high levels of non-phosphorylated neurofilament H, in deep layer 3 of the human cortex that do not have clear homologues in mice 35. These neurons are also selectively vulnerable in AD, leading to disruption of the human cortical networks.
Postmortem human tissue and neuroimaging analysis showed particular vulnerability of cortical layers 2/3 and 5 parvalbumin and somatostatin inhibitory interneurons to AD pathology, leading to loss of inhibitory control, cortical hyperexcitability, impaired memory encoding and retrieval, increased risk of seizures, and accelerated neurodegeneration through excitotoxicity. The fact that the human neocortex contains a significantly higher number and more diverse forms of GABAergic interneurons compared to rodents 36, 37 is likely to play a part in inadequacy of animal models of AD and their inability to reliably predict responses of AD patients to treatments.
Not only does the brain function differ between humans and most mammalian species, but these differences are also visible when compared to our closest primate relative - the chimpanzee. Comparative analysis of brain network architecture in humans and chimpanzee found increased connectivity between higher-order multimodal association areas in humans. Graph theory analysis showed that, in comparison to chimpanzee, long-range projections contribute significantly more to global network integration in humans 38, improving human information processing efficiency, cognitive flexibility, and higher order cognition. This is significant because the loss of global integration in AD manifests as memory lapses and executive dysfunction early in the process, before significant neuronal loss.
Comparison between human and mouse cortex structure and gene expression data, revealed marked inter-species differences in proportions, laminar distributions, morphology, and transcriptional profiles of neuronal and non-neuronal cell types. Non-neuronal cells, and astrocytes in particular, which are morphologically and functionally specialized compared to rodents, had the most divergent expression 39.
Astrocytes are one of the key players in AD, since they help clear the Aβ plaques, provide mechanical support to neurons, modulate glucose metabolism, and participate in signaling to endothelial cells, and their dysfunction in AD contributes to neuroinflammation, Aβ accumulation, impaired energy supply and vascular integrity, and neuronal loss.
In the human neocortex, astrocytes are 2.6-fold larger in diameter and extend 10-fold more glial fibrillary acidic protein-positive primary processes compared to rodents. There are several anatomically defined subclasses of astrocytes that are not represented in the neocortex of rodents 40. Moreover, in comparison to chimpanzee, the subclass of interlaminar astrocytes is more abundant. In light of these evolutionary divergences, it is believed that the increased astrocytic complexity in humans is directly related to the increased functional competence of the human brain and increased vulnerability to human neurodegenerative disorders. These features are poorly modeled in animals, limiting the value of AD animal models.
Microglia, the resident immune cells of the central nervous system that play a key role in AD, are also more diverse in humans compared to other mammalian species 41. Single-cell transcriptomics across over 10 species spanning more than 450 million years of evolution revealed larger heterogeneity among human microglia, as well as species-specific gene expression pathways related to complement system, phagocytosis, and metabolic pathways. Of particular significance for modelling human neurodegenerative disorders, analyses of transcriptomic signatures revealed that, in comparison to human microglia, rodent microglia express only a fraction of genes related to susceptibility to AD 42.
Considerable inter-species differences are also evident in the brain vasculature, that supports blood flow and maintains the blood-brain barrier. Transcriptional profiles of brain vascular cells (endothelial, pericytes, vascular smooth muscle cells) and perivascular cells (astrocytes, macrophages, microglia) differ between humans and mice. Individuals with AD exhibit human-specific vulnerabilities and perturbations across the brain vasculature that do not overlap with mouse models of AD 43.
Comparison of transcriptomics data in human and mouse cerebral cortex also demonstrated that the most-divergent gene families include neurotransmitter receptors, ion channels, extracellular matrix (ECM) elements and cell-adhesion molecules 39. Each of these play a role in AD and contribute to poor translation of mouse studies to the clinical setting.
For instance, AD involves dysfunction in glutamatergic, GABAergic, and cholinergic signaling and mouse models of AD may not replicate human synaptic vulnerability. Hyperexcitability in AD is shaped by voltage-gated sodium, potassium, and calcium channels, therefore inter-species differences in expression, diversity of channel subtypes, density on the neuronal membrane, and gating kinetics may produce fundamentally different responses to AD pathology in humans compared to AD mouse models. Divergences in ECM components may affect synaptic plasticity and the efficacy of remodelling in response to AD-related neuronal loss. Cell adhesion molecules are crucial for synapse formation and neuronal-glial interactions, and their species-specific expression may misrepresent synaptic loss and immune cell infiltration in AD mouse models.
Species-specific differences in aging mechanisms
Age is the main risk factor for AD. Biological drivers of aging can exert their effects through a wide variety of mechanisms, including structural damage (DNA, proteins and lipids oxidation), disruption of signaling pathways, mitochondrial dysfunction, proteostasis collapse, disruption of immune homeostasis, and change in gene expression.
Gene expression can be altered through transcriptional, post-transcriptional and epigenetic mechanisms. Potential biological drivers of age-related gene expression change in humans include chronic low-grade inflammation (inflammaging), accumulated oxidative stress, chronic infection, microbiome dysbiosis, reduced regenerative capacity, and accumulated epigenetic changes caused by environmental and lifestyle factors.
Rodents, who have evolved for rapid reproduction within their short lifespan, do not recapitulate the interactive effects that progressively cumulate across the human lifespan.
In additions, rodent-specific features of genetic background (regulatory elements, lncRNA, transposable elements), immune system 44, metabolic rate 45, susceptibility to pathogens and host-pathogens interactions 46, glucose regulation 47, hypothalamic-pituitary-thyroid axis dynamics 48, gut microbiome 49, lipid metabolism 50, stress response capacity etc. can have a protective effect against the effects of biological drivers of age-related decline to which humans are commonly vulnerable.
Change in gene expression can contribute to age-related decline characteristic of humans by altering neural networks. Comparison between human, rhesus macaque and murine gene expression changes that occur in the cerebral cortex during normal aging, showed that mice and men share only a small subset of age-regulated changes 51.
Notably, in humans, age-dependent repression of neuronal genes is dramatically increased compared to mice. Gene ontology analysis showed that genes involved in GABA-mediated inhibitory neurotransmission, including GABA receptor subunits and GABA vesicular transporter, were more robustly downregulated in aging humans than in aging mice, and that this age-related downregulation was more significant in GABA-mediated inhibitory neurotransmission than in glutamate-mediated excitatory neurotransmission. This finding suggests that mouse models of AD are likely to show increased resistance to excitotoxicity and neuronal loss compared to humans with AD.
Species-specific differences in genetics and gene expression regulation
Researchers who rely on mouse models often minimize the extent of species-specific differences between animal and human genomes. For instance, it is often said that 99% of protein-coding human genes have a counterpart in mice, which is misleading for several reasons: the protein-coding DNA represents only a tiny 1-2% fraction of total DNA that contains vital regulatory non-coding genomic elements, the gene orthologs can have a different structure and function across species, and the 1% human-specific protein-coding genes confer properties that are unique to humans.
The majority of GWAS-identified risk variants for AD reside in non-coding regions of the human genome - non-coding DNA (promoters, enhancers, silencers) and non-protein-coding genes (microRNA - miRNA, long non-coding RNA - lncRNA) - that have a crucial role in gene expression regulation.
Many of these non-coding genomic regions show little evolutionary conservation between humans and model organisms, meaning that their sequence, structure, and regulatory function is likely to significantly diverge across species.
The implications for the use of “humanized” mouse models is that, the human regulatory context (chromatin architecture, lncRNA, epigenetic changes, transcription factors, lncRNA-miRNA interactions, neighbouring genes) cannot be faithfully recapitulated in mice, even when human non-coding regions are inserted into the mouse genome.
Comparison between human, rhesus macaque, mouse, rat, rabbit, opossum and chicken expression patterns of lncRNA, across developmental time points in seven major organs, showed that sequences and expression profiles of lncRNA vary significantly across species 52.
Human-specific lncRNA rewire gene expression through a wide variety of regulatory mechanisms - transcriptional (by recruiting transcription factors to promoters and repression factors to silencers or by stabilizing promoter-enhancer looping), epigenetic (by recruiting epigenetic modifiers to specific genomic regions) and post-transcriptional (by influencing the stability, translation, or degradation of mRNA or by sequestering miRNA and preventing them from binding to their target mRNA) 53.
Even when genes are evolutionary conserved across species, the intron-exon composition of mRNA transcripts can vary across species due to differences in alternative splicing machinery.
Alternatively spliced exons contribute to the molecular diversity, differentiation and function of brain cell types 54. Yet, only about a quarter of alternatively spliced exons for a given transcript is conserved between humans and rodents 55.
This results in inter-species differences in protein isoforms, that can have distinct protein structure (binding sites, localisation signals) and function.
For instance, in humans, alternative splicing of MAPT gene produces human tau isoforms that differ from the murine tau in the number of repeated microtubule-binding sequences (R) and in the sequence of the N-terminal domain 56.
In the human brain, both 3R and 4R isoforms are co-expressed, whereas only the tau 4R isoform is expressed in mice. This inter-species difference might affect the propagation dynamics of tau in AD.
Since the N-terminal domain of tau is relevant for interacting with synaptic proteins, like NMDA ionotropic glutamate receptors, the murine tau isoform may not produce the same physiological or pathological effect (excitotoxicity, synaptic dysfunction) as the human tau isoform.
Unlike in the human brain, KPI domain‐containing APP isoforms, APP751 and APP770, are not expressed in the mouse brain, owing to species-specific differences in intron 7 sequence of APP gene and alternative splicing 57.
This has implications for AD, since KPI (Kunitz protease inhibitor) domain inhibits serine protease enzymes that break down Aβ into smaller non-toxic fragments. In AD patients, the ratio of KPI-containing/KPI-lacking APP mRNA tends to increase in the brain, whereas in mice, the KPI-lacking APP695 isoform predominates in the brain, conferring a protective effect against Aβ accumulation.
Another example that underscores the importance of not minimizing species-specific differences in gene orthologs is that of human-specific APOE isoforms.
Due to evolutionary divergence in DNA sequence of APOE gene orthologs, mouse Apoe shares about 70-80% amino acid identity with human APOE. However, it is not the 20-30% missing protein sequence that affects significantly the APOE function, but only 2 single nucleotide polymorphisms (Cys112Arg, Arg158Arg) 58.
Indeed, human APOE isoforms differ between each other by just 2 out of total 299 amino acids, yet this small difference has a massive impact on APOE’s tertiary structure. While the APOE2 isoform is protective against AD and the APOE3 isoform is neutral in regard to AD, the APOE4 isoform increases risk for late-onset AD by enhancing pathological interactions with Aβ, reducing delivery of sterols/lipid nutrients to neurons, and promoting microglia activation.
In contrast to humans, mice have only one Apoe isoform, which in terms of impact on AD is more similar to human APOE3.
APOE4 is believed to be the ancestral form of ApoE across mammals, meaning that the uniquely human vulnerability to AD associated with APOE4 is expressed in the context of other human-specific genetic, physiological, and environmental traits 59.
To circumvent the inter-species divergence in APOE isoforms, transgenic mice that express human APOE isoforms were designed. However, these “humanized” mouse models still fall sort of faithfully recapitulating human Aβ aggregation and clearance because of myriad species-specific differences in brain structure, metabolic rates, sterol turnover, enzymatic profiles, bile acids composition, and lipoprotein transport.
For instance, cholesterol-rich lipid rafts in cell membrane provide a favourable environment for production of aggregation-prone Aβ, and since mice clear and recycle brain sterols over 10 times faster than humans 60, transgenic mice carrying human APO4 are likely to miss the downstream effects of persistence of cholesterol-rich domains in human neurons and their prolonged exposure to toxic Aβ forms.
Many other species-specific differences that will not be covered in this section, such as immune response, neurotropism and host-pathogens interaction, gut microbiota, glucose regulation, and lipid metabolism, equally contribute to the fact that animals cannot adequately model AD.
Face validity - How well do animal models replicate the human disease phenotype?
Out of over 200 experimentally-induced animal models of Alzheimer’s disease (AD) generated so far 61, none were able to fully recapitulate the disease phenotype.
Several groups have attempted to model AD in species other than mice, including chicks, dogs, rats, guinea pigs, rabbits, dolphins, and non-human primates, but without obtaining a more human-relevant AD phenotype.
Another major limitation of AD animal models is that they do not recapitulate the inter-individual heterogeneity in AD (sub)types, pathophysiology, comorbidities, disease mechanisms, age-at-onset, and progression 3.
Due to the technical complexity of replicating the interplay between dozens of sporadic AD mutations and environmental factors in mouse models, most animal research has instead focused on reproducing core pathological hallmarks shared by both sporadic and familial AD - abundant amyloid plaques, neurofibrillary tangles (NFT), and neuronal loss.
As a result, the pathophysiological landscape of sporadic AD - metabolic dysregulation, vascular dysfunction, chronic low-grade inflammation and oxidative damage - remains under-studied, despite the fact that sporadic AD represents over 95% of AD cases.
In reaction to repeated failures in modeling AD in animals, in 2015 the NIH had initiated a new program called Model Organism Development and Evaluation for Late-Onset Alzheimer’s Disease (MODEL-AD). However, the initiative of “humanizing” animal models is unlikely to yield substantial benefits for AD patients, since the sheer number of species-specific differences that could be of relevance for AD by far exceeds the capacity of experimental “humanization”.
The use of multiple animal models to model distinct features of AD prevents the ability to trace the spatiotemporal evolution of mutually interactive pathological events.
Amyloid beta injection-induced animal models
The method of injections of soluble Aβ42 oligomers derived from brain of AD patients into hippocampus of awake mice was reported to reduce dendritic spine density, and impair memory in maze tests 3. However, while Aβ42 injections induced tau hyperphosphorylation and tangles-like pathology near the injection site, it did not produce widespread NFT and neurodegeneration.
Chemical injection-induced animal models
Acetyl-cholinesterase inhibitors, like donepezil, tacrine, galantamine and rivastigmine, originally indicated for neuromuscular disorders, were validated in scopolamine-induced and basal forebrain lesion-induced animal models.
In rodents, scopolamine induces reversible cholinergic deficit mimicking AD-like impairment of short-term memory 62, while ibotenic acid, 192 IgG-saporin, and mechanical lesions produce irreversible neuronal loss that mimics AD-like long-term cognitive deficits 63.
However, these models are not representative of the gradual progression of AD and do not recapitulate the key pathophysiological hallmarks of AD - Aβ deposits, NFT, and neurodegeneration.
Senescence-accelerated animal models
Among senescence-accelerated mouse prone (SAMP) strains, SAMP8 is the most commonly used strain in AD research 64, 65.
SAMP8 mice showed early cognitive decline at 4 months of age, neuroinflammation, Aβ deposition around blood vessels, tau hyperphosphorylation, and hippocampal neuronal loss, but without NFT, extensive neurodegeneration, and brain atrophy characteristic of AD.
Genetically-induced animal models
Overexpression models
The first generation single and double transgenic mouse models engineered to overexpress human APP, with or without familial AD (FAD) mutations, showed amyloid pathology within months of birth, which is roughly 2-3 times sooner than in familial and non-familial early-onset AD (EOAD). As a result, these mouse models did not capture the preclinical and intermediary stages of AD, preventing the possibility of identifying therapeutic targets for early spatiotemporal disease progression.
Despite the presence of large depositions of human Aβ, these APP overexpression mouse models did not recapitulate key pathological hallmarks of AD - NFT and substantial neurodegeneration 3, 66.
Furthermore, histological comparison between lesions in AD patients and lesions in single APP-Tg AD mouse models, APP23 and TG2576, that carry the Swedish KM670-671 NL mutation under control of overexpressing mouse Thy-1 and hamster prion protein promoters, showed that inflammatory responses were stronger in AD patients 67, underscoring the lack of relevance of these AD mouse models for studying neuroinflammation.
Double transgenic overexpression 5xFAD mice, that co-overexpress 3 FAD mutations in APP and 2 FAD mutations in PSEN1 under Th1 promoter, produced a very early and aggressive amyloid pathology, accompanies by neuronal loss, and progressive cognitive deficits, however still without NFT 68.
Examples of triple transgenic mouse models of AD obtained by overexpressing FAD/FTD mutations include 3xTg-AD and TauPS2APP mice.
The 3xTg-AD mouse model, that contains Swedish APP and MPT (Tau) mutations overexpressed under Thy1.2 promoter and PSEN1 mutation under endogenous promoter, showed aged-related Aβ deposition, tau phosphorylation, synaptic dysfunction, mild cognitive deficits, and impaired long-term potentiation, however without NFT, extensive neuronal loss and brain atrophy 69, 70.
The TauPS2APP triple transgenic mouse model, containing APP Swedish, PSEN2 and MPT (Tau) mutations driven by Thy1 and prion protein promoters, showed age-related accumulation of Aβ plaques and NFT, but still without extensive neurodegeneration and without progressive cognitive decline 71.
It was estimated that about 60% of the phenotypes observed in AD mouse models created by overexpression of mutant APP/APP+PSEN were artifacts 72. These artifacts include synaptic dysfunction and behavioural defects induced by toxicity of excessive Aβ production, as opposed to gradual accumulation of Aβ in AD.
It is believed that Aβ and non-Aβ fragments produced in excess by experimental overexpression may interact unphysiologically with cellular proteins, trigger downstream pathways that are not specific of AD, or deposit in brain regions and aggregate forms that are not representative of progression stages in familial and sporadic AD 68.
Knock-in models
In knock-in (KI) mouse models of AD the wild type/mutated human gene is expressed under the endogenous mouse promoter, producing physiological instead of overexpressed levels of human wild type/mutated protein.
Although these second generation AD mouse models bypass certain drawbacks of overexpression in first-generation mice, critical divergence with human AD remain, such as failure to produce Aβ neuropathology in human-relevant single APP mutations and absence of several AD hallmarks.
Typically, KI mouse strains artificially designed to carry a single or multiple APP FAD mutations are used. The strain NL-F carries the Swedish and Beyruthian/Iberian APP FAD mutations and the strain NL-G-F contains an additional Arctic APP FAD mutation.
The Swedish APP mutation (NL) is adjacent to the β-secretase cleavage site on APP, making APP more susceptible to β-secretase cleavage, and increasing total Aβ production. Located near the γ-secretase cleavage site, the Iberian mutation (F) alters γ-secretase processing of APP, thus increasing levels of more aggregation prone Aβ42. The Arctic (G) mutation, located near the α-secretase cleavage site in APP, promotes Aβ protofibril formation and aggregation.
In contrast to single APP-Tg mice, single KI mice that carry the Swedish APP mutation failed to produce Aβ deposits in the brain at up to 22 months of age, which corresponds roughly to 60-70 years of age in humans. When crossbred with human mutant PSEN1 KI mice, the resulting double knock-in (DKI) APP swe/PSEN1 mice presented with Aβ pathology. As in single/double transgenic mice, and in contrast to AD patients, NFT and neurodegeneration were not observed in these DKI mice 68, 73.
While DKI NL-F and triple KI (TKI) NL-G-F mice recapitulated early onset plaque deposition, microglia and astrocyte activation, hyperphosphorylated tau, and cognitive deficits, they also failed to recapitulate NFT and widespread neurodegeneration which define the full clinical picture of AD 68. Similarly, the TKI AppSAA mouse that harbors Swedish, Arctic, and Austrian APP FAD mutations did not develop NFT nor overt signs of neurodegeneration.
The hybrid AD mouse model, that combines DKI NL-G-F with overexpressed MAPT P301S mutation, showed Aβ accumulation, neuroinflammation, enhanced tau pathology and increased neurodegeneration 74.
It should be emphasized that, since AD patients individually carry single and not double/triple APP mutations, there is concern that interactions between double/triple APP mutations in DKI and TKI mice could produce inaccurate clinical representations 68.
The third generation AD mouse models combine KI APP FAD mutations with KI PSEN1 FAD P117L mutation - APP^NL-F or APP^NL-G-F x PSEN1^P117L 75. The obtained phenotype exhibits an earlier (3 months) and more aggressive Aβ deposition than in 2nd generation DKI APP mice, as well as astrogliosis and microglial activation, hyperphosphorylated tau and misfolded tau aggregates, however without NFT, widespread neuronal loss and brain atrophy typical of late-stage AD.
The absence of NFT in AD animal models in spite of Aβ pathology is problematic since it was shown that in AD patients the quantity of tau deposits correlated better with dementia severity than the quantity of amyloid deposits 15.
Another widespread limitation across APP-based AD mouse models is that cognitive and behavioural deficits are often mild and inconsistent, highlighting the lack of relevance of mouse models for studying human-relevant cognitive decline.
While genetic mouse models of AD typically rely on overproduction of Aβ, this diverges from sporadic AD in which it is the impaired clearance, and not overproduction, that is believed to be the main driver of Aβ accumulation.
Genetic AD mouse models based on FAD mutations do not recapitulate the spatiotemporal distribution of Aβ and tau in sporadic AD. While in sporadic AD, Aβ begins in neocortex and spreads slowly over decades, mouse models show early and aggressive Aβ deposition, often starting in the hippocampus.
Similarly, while in sporadic AD tau pathology starts in entorhinal cortex and follows progression stages described by Braak, in transgenic mouse models harbouring the FTD MPAT tau pathology follows a different distribution pattern that lacks Braak-like progression.
As a result of reliance on mouse models to better understanding AD, the pathophysiology of gradual progression, the chronology of events, and the complex interactions at molecular and cellular level remain poorly understood.
The focus on Aβ pathology as the main driver of AD is of limited relevance to sporadic AD in which multiple pathogenic pathways are thought to converge 56.
The co-occurrence of multiple brain pathologies that are strongly associated with aging and with cognitive decline in humans, such as proteinopathies and cerebrovascular diseases, are also missing in mouse models.
Owing to myriad human-specific differences that confer unique susceptibility to AD, including in immune system, metabolism, and brain physiology, the age-related changes that shape disease progression in humans cannot be reliably studied in animals.
In attempt to model SAD, ApoE4 KI mice, expressing humanized ApoE4, were subjected to lipopolysaccharide injections 76. Despite decrease in dendritic spine density, spatial learning and memory were intact, and neuroinflammation was not detected in this mouse model.
Animal models of AD that do not show the full spectrum of AD symptoms are oftentimes presented as models of early or prodromal AD, however, these is insufficient evidence of construct validity and predictive validity to support that notion.
To date, the search for AD models with a more faithful recapitulation of AD phenotype and a better concordance with human clinical trials still continues.
Construct validity - How well do the mechanisms of disease induction in animals reflect the currently understood etiology of the human disease?
As it the case for many animal models designed to model human diseases, animal research into Alzheimer’s disease (AD) is driven by hypothesis-led model validation rather than unbiased hypothesis testing. For many decades, the ability to mimic amyloid aggregates in mice was the central criteria for judging the validity of preclinical models and for predicting the clinical success of anti-amyloid therapies. As the dominant amyloid hypothesis gained traction, concerns about the human-relevance of animal models focused on amyloid aggregation were largely overlooked.
In the context of predominant amyloid cascade hypothesis, the strategy of experimental induction of AD in animals is mainly reliant on introduction of FAD mutations to induce amyloid beta (Aβ) deposits.
The multiple disease mechanisms that mutually interact to trigger sporadic AD are not recapitulated in animal models of AD.
Animal models of AD are not representative either of distinct disease mechanisms that are responsible for heterogenous EOAD and LOAD presentations.
The complex interactions between internal and external risk factors that intervene in sporadic AD are not captured in animals, and even if they were, they would be heavily influenced by species-specific differences, producing pathomechanisms that are irrelevant for humans.
Amyloid beta injection-induced animal models
The method involving the direct injection of Aβ oligomers extracted from the brains of AD patients into the mouse hippocampus has served as the basis for demonstrating the role of Aβ in synapse loss 77, 78.
This method, however, is likely to lead to erroneous interpretations of AD mechanisms. Apart from the obvious fact that, in humans, Aβ does not come from the environment and does not start in the hippocampus, but instead builds slowly over time following Braak stages starting from the neocortex,
this method does not allow to understand the upstream factors that lead to Aβ accumulation in humans.
The mechanical injury itself is likely to produce inflammation, making it difficult to determine the exact role and contribution of inflammatory responses that are specific to AD.
Depending on the precision of injection, quality of Aβ (synthetic versus physiological) and volume injected, hippocampal injection may lead to off target effects, non-specific cytotoxicity, or lack of Aβ toxicity.
The non-specific neuronal damage in animals that can be caused by this procedure is likely to affect the outcomes of behavioural tests that are commonly used to assess anxiety, sociability, and cognitive function.
Chemical injection-induced animal models
Scopolamine-induced amnesia and basal forebrain lesion modelling in rats, mice, and non-human primates are still actively used in AD research.
Basal forebrain lesion models, often created using ibotenic acid and 192 IgG-saporin, simulate long-term cholinergic neuron degeneration 79. The muscarinic acetylcholine receptor antagonist scopolamine temporarily disrupts cholinergic signaling, mimicking the memory deficits 62.
The chemically-induced models do not mimic the genetic and non-genetic causes of familial and sporadic AD. It is therefore not surprising that treatments developed in chemically-induced animal models can help manage certain AD symptoms but do not reverse the disease course.
In animals, these chemicals are typically administered locally, through intraparenchymal injection into nucleus basalis magnocellularis and through intracerebroventricular injection into cerebral ventricles.
As in Aβ injection-induced animal models of AD, the acute localized insult does not mimic the upstream triggering factors and associated pathways. The slow cholinergic degeneration in AD with its intermediary stages and compensatory mechanisms is not replicated. The injected chemicals can produce off target effects and trigger non-specific inflammatory responses, and the procedure can cause motor and sensory disturbances in animals producing misleading results in behavioural tests.
Senescence-accelerated animal models
The commonly used senescence-accelerated mouse prone (SAMP) strains are derived from inbred AKR/J lines that were selected for traits of accelerated aging 64, 80.
The exact mechanisms of aging in humans are not well understood, and are believed to involve genetic factors, environmental factors or a combination of both. SAMP mice are not representative of any of these causes, since they are artificially uniform and are not exposed to the same environmental risk factors as humans.
In contrast to genetically edited mice with FAD mutations, the SAMP8 strain does not carry the known rare variants associated with FAD, and there is insufficient evidence that SAMP8 genetically mirrors sporadic AD in terms of GWAS-identified common variants.
In addition, unlike humans, SAMP strains do not mimic the heterogeneous genetic background that can predispose individuals to age-related decline and that are believed to contribute to inter-individual differences in human AD phenotype and underlying mechanisms.
Even if certain dysregulated pathways did overlap between SAMP mice and AD patients, species-specific differences in gene expression regulation, epigenetic modification, and splicing would still stand in the way of direct translation to age-related decline in human AD.
Although the SAMP8 strain exhibits AD-like vascular amyloid deposition and neuroinflammation, the murine immune system and brain vasculature are fundamentally different in humans, which might explain why SAMP8 mice do not show extensive neurodegeneration.
Genetically-induced animal models
Overexpression models
In AD mouse models that overexpress human genes involved in familial AD (FAD), the mutated gene is driven by a strong promoter, such as cytomegalovirus, hamster prion protein, and mouse Th1 promoters, which produce several fold higher levels of the target protein.
Overexpression of mutated genes is not representative of mutations found in humans with familial and sporadic AD, except perhaps for rare cases of familial EOAD in which the APP locus is duplicated 78.
In overexpression animal models of AD, the transgene, that typically carries APP and PSEN mutations found in FAD, is integrated into the mouse genome outside the endogenous gene locus, bypassing the native regulatory mechanisms of gene expression. Since the transgene is integrated randomly, it can interrupt or alter the expression of nearby genes.
The genetic construct does not recapitulate the spatiotemporal regulation of gene expression in humans, which is likely to produce a regional and chronological distribution of target proteins that does not match the pathology of AD in humans.
Excessive levels of target protein tend to produce features (localization of the protein, cellular stress) that are due to a general overwhelming of cellular capacities (protein folding, autophagy) rather than to the pathogenic properties of the target protein.
The downstream cascade of events triggered by excessive protein production is anticipated to be different from the one triggered by physiological levels of a dysfunctional protein. For instance, in overexpression models, non-specific loss of neuronal and non-neuronal cells might be driven by widespread cytotoxicity that triggers mechanisms of ER stress, unfolding protein response, and necrosis, whereas in AD-related synaptic and neuronal loss, progressive amyloid and tau pathology might trigger pathogenic receptor-mediated signaling pathways, microglia activation, and PANoptosis.
This type of construct also makes it difficult to identify proteins responsible for neurodegeneration, since overexpression of genes encoding APP with FAD mutations produces several Aβ fragments in excess 56. It is also unclear whether the endogenous mouse App contributes to the amyloid pathology in ways that are not representative of the amyloid pathology in AD patients.
In attempt to compensate for the inability of genetically edited mouse models to recapitulate some of the major hallmarks of AD such as NFT and widespread neurodegeneration, in certain mouse models mutated MAPT (tau) was overexpressed.
This approach poses several issues, one of which being that MAPT (tau) mutations, such as P301L, P301S, and R406W, are not found in AD patients but in patients with frontotemporal dementia, a disorder with distinct clinical features and mechanisms 81. In humans with AD, tau pathology arises through entirely different mechanisms of post-translation modifications and isoforms imbalance.
Furthermore, the overexpressed tau transgene can produce a tau pathology independently of the tau mutations type 56, indicating that the mechanisms at play in mouse models with overexpressed MAPT mutations might not be relevant for AD.
Age is considered as the main risk factor of AD, yet, in the overwhelming majority of mouse models of AD the aggressive amyloid deposition and tau pathology appear only a few months after birth. This scenario is not found in sporadic AD and only rarely in early onset familial AD, preventing investigation of the cascades of pathways that are actioned over time as the disease slowly progresses.
The multiple mechanisms related to aging, including oxidative stress from accumulation of ROS, decrease in efficiency of mitochondrial ATP production, impaired autophagic clearance of damaged proteins and organelles, altered synaptic plasticity, chronic low-grade inflammation, and stem cell exhaustion, are not recapitulated in animal models of AD. Even if these mechanisms were captured in animals, they would still not be directly translatable to humans owing to species-specific differences.
It is also noteworthy that, in spite of presence of fully penetrant APP and PSEN FAD mutations in genetically edited mouse models, inter-species differences in genetic background can still affect the severity of pathology, inflammatory response, and response to treatment 78.
Knock-in models
In knock-in (KI) mouse models of AD, the gene of interest in inserted in the native genomic locus and expressed under the endogenous mouse promoter, producing physiological levels of the target human protein.
The newer generation KI AD mouse models combine several AD-associated mutations using simultaneous multi-locus editing, in which mutations are introduced in a single CRISPR/Cas9 step, reducing mosaicism.
However, even though this genetic engineering strategy addresses several limitations of the overexpression mouse models, it still falls short of recapitulating human-specific gene expression regulation, epigenetic modifications, post-transcriptional regulation, and modifier genes.
In addition, the mouse orthologues of genes involved in human AD may have a different structure and function. For example, due to amino acid substitutions in the Aβ sequence, the murine Aβ is less prone to aggregation than the human Aβ, and in mouse models where endogenous murine APP gene is retained, the murine Aβ could have a protective effect on amyloid and tau pathology 82. While the human TREM2 plays a key role in AD by regulating microglial activation and phagocytosis, the murine Trem2 has divergent ligand binding and signaling dynamics (cytokine release, metabolic reprogramming), limiting its mechanistic relevance to human pathology 83. In humans, the APOE gene is polymorphic and its APOE4 isoform is considered as the primary genetic risk factor of sporadic AD, however, mice do not naturally possess the equivalent of this human isoform 84.
Despite the fact that KI mouse models of AD make use of known FAD mutations, they fail to recapitulate NFT and severe neurodegeneration 69, 75, 76, pointing to the critical role of other human-specific factors that are poorly understood and that cannot be recapitulated in mice.
Only a handful out of several dozen pathogenic variants linked to AD 27, 28 were used in overexpression and KI mouse models of AD. Different pathogenic variants are likely to produce inter-individual differences in type/location/structure of Aβ aggregates and qualitatively different pathologies. The commonly used mouse models are therefore not representative of inter-individual differences that affect responses to treatments.
Furthermore, mouse models that combine distinct double or triple APP mutations to enhance amyloid pathology, such as DKI NL-F and TKI NL-G-F mice, do not mimic the etiology of FAD, since in FAD patients these mutations are not cumulative but occur individually in separate families or individuals. Since Swedish, Arctic, Iberian and other APP mutations each have a distinct pathogenic mechanism, the use of several mutations in vivo is unlikely to lead to an accurate understanding of FAD disease mechanisms.
Most KI mouse models of AD employ FAD mutations which do not reflect the genetic architecture of sporadic AD, causing misleading interpretations 56.
So far, 75 common variants have been identified for sporadic AD, and while these variants have small effect sizes, their cumulative impact is significant since they involve dozens of sporadic AD-related mechanisms that are believed to converge together 25, 26.
Depending on their genetic background, an average person carries between 20 and 30 risk alleles for sporadic AD. It is not feasible to generate hundreds of mouse models that recapitulate various combination of risk alleles. Consequently, the multitude of pathways that intervene in triggering and progression of sporadic AD (oxidative stress, mitochondrial dysfunction, inflammation) cannot be recapitulated and studied in animals.
The impact of human environmental, lifestyle and comorbidities adds an additional layer of complexity in replicating the multifactorial nature of sporadic AD.
Internal (diabetes, cardiovascular diseases), lifestyle (sleep, diet, exercise) and external (infections, stress) risk factors can alter gene expression through epigenetic changes, through altering of transcription factors activity or through change in mRNA processing.
Epigenetic imprints of decades of accumulation of combined human risk factors cannot be captured in mouse models, which have a much shorter life span, species-specific physiology, and laboratory-specific exposure to environmental factors.
Similarly to transgenic mice that overexpress FAD mutations, knock-in mouse models with DKI/TKI APP mutations result in overproduction of Aβ, whereas sporadic AD is believed to result from reduction of Aβ clearance.
In effort to model reduced Aβ clearance in vivo, mice with impaired LRP1 function were used 85, however, there is no evidence of existence of rare or common variants impairing LRP1 in AD patients, and it is unclear whether certain compensation pathways of Aβ clearance might be specific to mice.
In addition, given that AD heterogeneity in severity, location, pathophysiology, and resilience is determined by complex interactions of genetic, co-pathology, lifestyle, and environmental factors, animal models are inadequate for designing therapies tailored for specific AD subtypes.
Behavioural test validity - how well do behavioural tests in animals translate to human behaviour?
Alzheimer’s disease (AD) affects both cognitive function and neuropsychiatric health, and clinicians use a combination of assessments methods to measure the degree of impairment in these domains - mini-mental state examination, Montreal cognitive assessment, behavioural pathology in Alzheimer’s disease rating scale, Cornell scale for depression in dementia, clinical dementia rating scale sum of boxes etc. 86, 87.
Behavioural tests in rodent models of AD typically assess learning and spatial memory (Morris water maze, Barnes maze, Y-maze spontaneous alternation), fear-associated memory (contextual and cued fear conditioning), recognition memory (novel object recognition), exploration (rearing), anxiety (elevated plus maze, open field), depression (forced swim test), sociability and social interaction (three-chamber test).
Results of behavioural tests in animals for a given drug candidate are often presented as a promise of cognitive improvement in AD patients, despite the fact that it was demonstrated over and over again that reversal of cognitive decline was not observed in AD patients, revealing a profound disconnect between the faith in preclinical behavioural tests and the cruel reality of clinical failure.
Given that no single animal model can faithfully mimic human behavioural psychiatric symptoms 88, researchers tend to focus on a reductionist approach of modeling components of psychiatric syndromes 89.
As it is not feasible to measure the same components of cognitive and psychiatric health in humans and in rodents, behavioural proxies are used. For instance the open field test is a proxy for human curiosity, anxiety or apathy, based on spatial behaviour in rodents, that poses difficulties in interpretation of rodent motivations and translation to human emotions.
In addition to the lack of direct equivalents of behaviour, the molecular and cellular mechanisms that underly animal behaviour differ in humans.
Rodents have evolved to adapt to specificities of their environment under constraints of their anatomy and physiology. Since it is very difficult to understand the meaning of a certain rodent behaviour in a certain context, researchers often resort to subjective interpretations.
Behavioural measurements in animal models are ridden with anthropomorphic fallacies, ultimately producing erroneous hypotheses about efficacy of candidate treatments 90.
Behaviours that are not homologous to those in humans are particularly likely to be misleading 91. An example of rodent behaviour that is not homologues to any human behaviour is rearing in rodents (standing on hind legs), which is viewed as an exploratory and low anxiety behaviour even though a rodent might rear in a new environment despite feeling anxious or not be anxious but nevertheless not rear because he does not sense anything interesting to explore.
The open field test, that measures activity of rodents near the center square of a brightly lid open space, reflects a belief that, depending on their level of anxiety, rodents will choose between exploring open areas or staying close to the wall of the field. This premise has been heavily criticized for its over-simplification and for not taking into consideration relevant species-specific aspects of behaviour. In addition, the behaviour this test captures is a snapshot of a response under stress, and not a demonstration of persistent anxiety disorder observed in humans with AD 92.
The Morris water maze test measures the time a rodent takes to locate a small platform hidden below the water surface. However, rodents do not necessarily act in a way that humans would and can look for alternative ways to escape their situation. In contrast to humans, many animal species do not encounter such situations in the wild and may have anxiety in the water environment, ultimately affecting their cognition in a way that is not specific to AD.
Analysis of individual differences in Morris water maze swim paths of over 1000 mice demonstrated that the innate tendency of mice to stay closer to the walls (thigmotaxis) explained 49% of the variability, passive floating or slow swimming (passivity) accounted for 19% of the variability, and only 13% of the variability reflected the search time spent in the probe trial target quadrant (memory), indicating that the majority of reported changes in behaviour of AD mouse models were in reality not related to cognitive function 93.
More recently, Rosso et al. 94 assessed the sensitivity of common behavioural tests of anxiety in mice, including elevated plus maze test and open field test, to detect anxiolytic effects of drugs prescribed to treat anxiety in humans. Their findings indicate a general lack of sensitivity of those behavioural tests and cast serious doubt on their construct and predictive validity.
The novel object recognition memory test is based on the assumption that rodents have an innate preference for novelty and will spend most of their time with a novel object, under condition that they recognize the familial object. However, an animal may have an innate preference, an aversion, or an indifference for a specific object for reasons that are not related to memory decline 95.
The three chamber social preference test is based on the supposition that healthy mice will spend more time in the same chamber as a new conspecific than in an empty chamber. Reduced time with the new conspecific is interpreted as a sign of social withdrawal. Similarly, the three chamber social recognition tests assumes that healthy mice will spend more time in the same chamber as a new conspecific than in the chamber with the already known conspecific. Equal time spent in each chamber is depicted as impaired social recognition 96.
The main problem of this experimental concept is that it was designed according to how humans perceive the meaning of social interactions. In contrast to humans, who engage for companionship, emotional bonding and social approval, the social behaviour of rodents is mainly shaped by dominance, territory and mating. A mouse is naturally inclined to spend less time with a conspecific, not because he is socially withdrawn or because he has impaired memory, but simply because he perceives the conspecific as a threat.
Despite ample evidence that forced swim test (FST) is not predictive of human responses, FST is still used in AD research to assess depressive-like behaviour 97. In the forced swim and tail suspension tests, increasing the latency to stop struggling is interpreted as sign that the animal is less depressed. However, animals may stop struggling, not because they have fallen into a state of despair, but because they understood that it was not energy-efficient to persist and that, if they stayed still, they would be scooped up by the examiner.
In their retrospective review assessing whether 109 compounds tested in FST were shown to have antidepressant effects in humans, Trunnell and Carvalho 98 conclude that FST has poor accuracy for identifying novel antidepressants.
Since December 2023, Australia’s National Health and Medical Research Council no longer funds research using FST, judging that FST does not meet scientific and ethical standards 99. The U.S. National Institute of Mental Health (NIMH) has discouraged use of the FST while the UK Home office has announced restriction on this test in March 2024 100.
The methods that are used for artificial induction of AD symptoms in animals can produce behavioural test outcomes for reasons that are not related to AD. For example, intracerebral injection of Aβ peptides can affect visual and olfactory areas, depriving rodents of sensory cues necessary for object recognition, Morris water maze, or fear conditioning, and scopolamine can cause general malaise, affecting locomotion, motivation, and sensory processing.
Since rodents are sentient beings, the abnormal conditions of housing, in a laboratory instead of in the wild, produce a baseline of chronic psychological distress affecting their behaviour 101.
In addition, certain strains of mice have higher innate levels of anxiety, which could impact locomotor activity and exploration time.
Analyses of behavioural outcomes across mouse strains and laboratories showed significant variation in all behavioural tests, despite efforts to standardize laboratory settings and procedures 102, highlighting the need to redirect resources to human-based testing methods as to avoid inflicting further harm on patients and experimental animals.
To add a supplementary lay of complexity to inter-species and inter-strain differences in behaviour, the genetic background of animals and the environmental effects (test situation, lab environment, handling, post-partum pregnancy, water) are often interactive, such that some genotypes respond more than others to specific features of the controlled environment in which they are assessed. Failures to replicate patterns of genetic results across laboratories can arise from any or all of these factors 103.
Despite decades of investment in open-science collaboration, software development and data sharing, standardization in mouse neurobehavioral genetics is still lacking and most tests are still done in a manner that is unique to each laboratory.
Researchers who conduct behavioural tests believe that using a battery of several tests can improve the reliability of findings, however, these tests remain individually unreliable and combining several tests with unclear and contradictory findings does not necessarily improve the overall predictivity.
Even if robustness of individual behavioural tests was improved, these tests would still fail to recapitulate human-specific cognitive capacities, behavioural patterns and molecular underpinnings.
Predictive validity – How well do animal models predict safety and efficacy of therapies in patients?
Although it was argued that animal research was instrumental in developing new treatments 104, preclinical research using animal models of Alzheimer’s disease (AD) has consistently failed to produce significant benefits for patients suffering from AD.
Clinical trials for AD have a very high failure rate of 99.6%. Indeed, between 2002 and 2012, 413 AD trials were performed with an overall approval rate of only 0.4% 124.
Since preclinical trials lack a mandatory registry, statistics on preclinical studies for AD are unavailable. Nevertheless, it can be estimated that therapeutic leads were tested in over thousand preclinical studies so far. Due to species-specific differences, it cannot be excluded that some of the leads that did not show benefit in animal models of AD would have been beneficial for AD patients.
At least 200 unique compounds were validated in AD animal models in the past two decades. As of January 2022 there were 143 agents in clinical trials for AD. Disease-modifying therapies represent 83% of the total number of agents in trials and 37% of the candidate agents in the pipeline are repurposed drugs approved for other indications 25.
According to the Common Alzheimer's Disease Research Ontology (CADRO) analysis of disease-modifying targets, about 30% of drugs in the AD pipeline target amyloid plaques and tau aggregates, while the remaining drugs target CADRO categories like synaptic plasticity, oxidative stress, neuroinflammation, neurotransmitter receptors, and others. Yet, in spite of proliferation of mechanistic treatment approaches, very few treatments for AD were approved in phase 3 trials.
Current treatments for AD include symptomatic therapies, that ease the symptoms of AD without addressing its underlying causes (cholinesterase inhibitors and partial NMDA antagonists), and therapies that focus on the underlying causes (aducanemab, lecanemab, and donanemab immunotherapy) 11, 105.
It’s particularly noteworthy that cholinesterase inhibitors and partial NMDA antagonists were originally developed for other indications before being repurposed for AD, highlighting the value of clinical observations of off-label benefits and patient-centric neurobiological explorations.
Acetyl-cholinesterase inhibitors (donepezil, galantamine and rivastigmine), that help counter deficit in acetylcholine by increasing its availability in the synapse, thereby boosting communication between neurons, demonstrated beneficial effect at the mild to moderate stages of AD.
Memantine, a low affinity noncompetitive NMDA antagonist, that works by reducing excitotoxicity caused by excessive glutamate signaling, was shown to have a small but clinically appreciable benefit on cognition and functional decline in patients with moderate to severe AD 4.
Unfortunately, although these symptomatic drugs offer some relief for AD patients, they do not halt disease progression.
In accordance with the prevalent amyloid cascade hypothesis, development of therapies that target Aβ took center stage for many decades. However, not all immunotherapy molecules that target Aβ and tau produced beneficial effects in clinical trials. Therapies targeting Aβ that demonstrated a modest effect in reducing the rate of clinical decline, did not succeed in halting and reversing it.
Semagacestat, an inhibitors of γ-secretase that cleaves several substrates, including Aβ and Notch receptors, did not show a significant impact on Aβ levels and the patients experienced cognitive decline in clinical trials, presumably through disruption of Notch-signaling 106.
The first-generation monoclonal antibodies (mAb) that target various Aβ forms - solanezumab, bapineuzumab, and crenezumab - failed to demonstrate clinical benefit for AD 107.
Solanezumab that binds monomeric Aβ, which is not the most neurotoxic Aβ species, failed to reduce plaque burden and to slow cognitive decline in patients with preclinical to moderate AD.
Bapineuzumab, designed to clear primarily the fibrillar, but also monomer and oligomer forms of Aβ by targeting the N-terminal sequence of Aβ, did not demonstrate significant cognitive improvement and had high rates of amyloid related imaging abnormalities (ARIA) adverse effects.
Even though crenezumab binds to mid-domain of Aβ oligomers and fibrils, it still failed to show benefit in 2019 and 2022 phase 2 and 3 clinical trials.
Semorinemab, a first-generation mAb targeting the N-terminal region of all six isoforms of human tau, had shown mixed results 108. After failing to meet primary and secondary endpoints in Tauriel trials for prodromal-to-mild AD, and meeting one of the two co-primary endpoints of slowing-down cognitive decline, albeit without demonstrating reduction in tau accumulation, its development was discontinued.
Safety and efficacy of second generation tau-targeting mAb, such as bepranemab and BMS-986446, designed to target pathogenic tau domains is currently under evaluation 109.
All three second-generation mAb - aducanemab, lecanemab, and donanemab - that target oligomer, protofibril and fibril forms of Aβ with improved specificity and binding affinity, received accelerated FDA approval based on evidence of reduction of amyloid plaques in AD patients.
Lecanemab, that primarily targets Aβ protofibril soluble aggregates, reduced markers of amyloid in early AD and resulted in 27% slowing of disease progression on measures of cognition and function 110, 111.
Donanemab, that targets the pyroglutamate-modified Aβ plaque core, delayed cognitive and functional decline among patients with mild-to-moderate AD by 32% 112, 113. Other researchers have, nonetheless, pointed out that the absolute difference between treatment and placebo groups on the Clinical Dementia Rating-Sum of Boxes scale was just under 0.7, below a full point that would have represented a significant effect on disease severity 114.
Aducanemab that targets Aβ fibrils and oligomers had received accelerated FDA approval in June 2020. However, since it missed the primary endpoint of cognitive or functional benefits in phase 3 trial, it was discontinued in November 2024 115.
In terms of safety of therapies for AD, in spite of improvement in mAb target specificity, there remains a significant risk of ARIA adverse effects that occur when antibodies bind to amyloid plaques in cerebral blood vessels, triggering inflammation and vascular leakage that can result in vasogenic edema and intracerebral hemorrhage 116, 117.
Out of the three immunotherapies targeting Aβ, only lecanemab is currently authorized for use in the EU and only in patients with early AD who carry one or no copies of the APOE4 gene, as to minimize the risk of ARIA in these particularly susceptible AD populations 118.
Due to high frequency of ARIA with serious events, even in non-APOE4 carriers, the EMA issued a negative opinion on the marketing authorization for donanemab in March 2025 119.
There are equally growing concerns about the potential adverse effects of anti-amyloid mAb on cognitive function in AD patients, particularly since these mAb have been reported to accelerate brain atrophy 120.
Apart from therapies for AD that focus on Aβ and tau, other therapeutic leads currently under investigation include several pathological (neuroinflammation, oxidative stress) and protective processes (synapse preservation, vascular repair) 121.
Devising strategies so that drugs can more effectively cross the human blood-brain-barrier to reach its target in sufficient concentration (antibody-drug conjugates, nanoparticles, receptor-mediated transport) is another active area of research.
The approach of using combinatory therapies as to treat several underlying mechanisms is also being actively studied.
Given large inter-individual differences in AD characteristics and causes, it is equally important to adjust therapies to distinct AD subtypes.
A patient-centered approach using human-based research methods can improve understanding of heterogenous disease mechanisms, opening avenues for therapies that are tailored to patient-specific AD phenotype and etiology.
Ethical validity - How well do animal experiments align with human ethical principles?
Preclinical
Ethics is a human-specific philosophical concept. Animal experimentation is unethical in essence by human standards, since it involves physical constraint, psychological suffering and deprivation of freedom, social interactions, natural environment, and life purpose.
In addition to this baseline, animal experiments inflict severe clinical harm in animals 122.
Table S4: Severity classification of clinical signs
Neurological signs impairing ability to move and function normally - up to severe clinical signs
Table S5: Severity classification of chemical disease models
Neurology and sensory organs: up to severe clinical signs
Table S10: Severity classification of behavioural tests
Contextual and cue fear conditioning: moderate
Morris water maze, elevated plus maze, Y maze, novel object recognition, social preference/social novelty, and open field exploration: mild
Forced swim or exercise tests with exhaustion as the end point: severe
Table S11: Severity classification of function measurements
Intracerebral injections, electrophysiology, microdialysis etc.: up to moderate
Table S13: Severity classification of genetically altered (GA) lines
Mortality - GA lines resulting in lethality from 2 weeks post-partum on: Severe
GA lines resulting in long-term moderate pain or short-term severe pain: Severe
GA lines resulting in long-term moderate anxiety or short-term severe anxiety: Severe
Clinical
While there is no consensus on whether an unethical act can be justified by a pursuit of a hypothetically ethical outcome, it was suggested that animal research was necessary to advance treatments for human diseases.
However, statistics consistently show that clinical success rates of drugs developed and tested in animals is very low, raising the question of whether it is ethical to put the health of patients at risk.
Over 2011-2020, the overall likelihood of approval of drugs for neurology was 5.9%. The Phase I to II transition success rate was 47.7%, below the 52% average of all indications, while Phase II to III transition success rate was 26.8% versus 28.9% average for all indications 123.
These numbers are much lower for Alzheimer’s (AD), since only 0.4% of drug candidates were approved between 2002 and 2012 124.
Existing therapies for AD carry a significant risk of adverse events, including ARIA and brain atrophy, for a very modest clinical benefit.
Intrinsic validity - How well do animal models capture the clinical heterogeneity of the human disease?
Animal models do not recapitulate the clinical heterogeneity in Alzheimer’s disease phenotypes, age-at-onset, etiology, progression rates, and responses to treatments.
Extrinsic validity - How well does animal experimentation generate reliable and reproducible outcomes?
It was often argued that although animal models have severe limitations, animal research enables to gather insights that may be valuable. However, the basic precondition for a hypothetical benefit is not met since the majority of animal experiments is irreproducible 125.
Contributing factors include flawed experimental design, variation in animal strains and experimental conditions, and lack of transparency on methodology and results of animal studies.
In a bid to improve the quality of reporting of animal experiments, the ARRIVE - Animal Research: Reporting of In Vivo Experiments - guidelines, were published in 2010 and updated to ARRIVE 2.0 in 2020 126.
Nonetheless, and in spite of significant investment in dissemination, various incentives and training of animal researchers, the Arrive guidelines remain poorly implemented 127, 128.
As a result of weak relevance, rigor and reliability of animal studies, erroneous and misleading hypothesis are generated, animal and human lives are needlessly sacrificed and dozens of billions of dollars are annually wasted 129.
Beyond the problem of experimental design and reporting of animal studies, there are deep-seeded cultural reasons that are not likely to be addressed any time soon, such as the "publish or perish" culture that discourages from unbiased analysis and reporting of negative results and rewards exaggerated and overhyped claims 130.
In Summary
Alzheimer’s disease (AD) is a highly debilitating progressive neurodegenerative disorder affecting cognitive function and behaviour. Its prevalence is projected to triple worldwide by 2050, posing a major societal and health care burden.
Critical species-specific differences between animals and humans in brain structure, cognition, behaviour, aging mechanisms, and gene expression regulation preclude faithful recapitulation in animals of symptoms of this uniquely human disorder.
Due to decades old reliance on animal research to study AD pathogenesis, the exact mechanisms of AD remain unclear.
The failure rate in clinical trials of treatments developed and validated in animal models of AD is as high as 99.6%. To date, none of the treatments tested in AD animal models were able to halt or reverse cognitive decline in AD patients.
A switch from animal-based to human-based research is urgently needed to explore the mechanisms of AD, identify new therapeutic targets and develop effective therapies adapted to heterogenous AD subtypes.
How is Human-Based In Vitro Testing the Answer to Advance Biomedical Research into Alzheimer's Disease
*To model AD subtypes using AD patient iPSC-derived brain tissues. To identify proteins and pathways associated with AD donor neuropathological score, cognitive score, and polygenic risk scores by comparing in vitro measures to patient data 131.
*To dissect the functional impact of sporadic AD-associated genetic variants by parallel editing of multiple loci in human iPSC neurons/organoids/brain-on-chip using multiplexed CRISPR systems.
*To determine the localisation and temporal evolution of AD molecular and cellular abnormalities (patterns of gene expression, vulnerable cell subpopulations, trajectory inference, environmental interactions, inter-individual variations), by using single cell sequencing with spatial multi-omics (sc/snRNA-seq, ATAC-seq, FISSEC/MERFISH, CODEX, electrophysiology, morphology) in AD patient-derived brain tissues at different stages of the disease in large scale cohort studies 132, 133, 134, 135.
*To study the impact of alternative splicing events on AD (KPI-containing APP isoforms, 3R/4T tau isoforms, clusterin isoforms), by leveraging sQTL in AD patient-derived iPSC/brain organoids containing patient-specific alternative splicing machinery (genetic variants, splicing factors, hnRNP, lncRNA).
*To determine the impact of epigenetic modifiers on late-onset sporadic AD and late-onset familial AD, by leveraging CRISPR/dCas9 epigenetic editing to introduce DNA methylation/histone acetylation at loci relevant to AD (APP, BACE1, APOE, TREM2) in AD patient-derived iPSC/brain organoids.
*To model human aging and to study features of aged human neurons in late-onset AD (DNA methylation, gene expression, reduced synaptic plasticity, mitochondrial defects, accumulated DNA damage, REST proteins), by using direct fibroblast-to-induced neuron conversion that preserves elderly AD donor-specific epigenetic signatures 136 and by accelerating aging in human iPSC by exposing them to chronic oxidative stress.
*To investigate the effect of neuroinflammation on AD progression, by using human cortical organoid microphysiological systems cocultured with AD patient-derived monocytes 137.
*To gain mechanistic insights into how specific dietary components influence AD pathology, by analysing neuroinflammation, barrier integrity, oxidative stress, Aβ expression etc. in human gut-liver-brain-on-chip.
*To study the effect of human microbiome dysbiosis on AD (microglial activation, BBB disruption, Aβ expression, tau phosphorylation) by studying
microbiome-BBB interactions, nutrient absorption, pathogenic signaling pathways, immune tolerance etc., in immunocompetent AD patient-derived tissues brain-gut microbiome-on-chip 138.
*To study mechanisms by which comorbidities (endocrine, cardiovascular) affect AD pathogenesis and progression, by using AD patient-derived multi-organs-on-chip with brain compartment connected to other organ compartments (liver, heart).
*To investigate mechanisms of structural breakdown, impaired Aβ clearance, and immune infiltration of human BBB in AD state. To develop strategies for more effective delivery of AD therapies across the human BBB. To predict inter-patient variability in BBB functions, using human/AD patient-derived BBB-on-chip 139.
*To assess the contribution of each cell type to AD phenotype, by using chimeric bioprinted tissues containing one or more diseased cell types in the context of a healthy background of other cell types - excitatory neurons, inhibitory neurons, microglia, astrocytes, oligodendrocytes, endothelial cells - in healthy and AD patient--derived brain tissues 140.
*To investigate mechanisms of senescence-associated secretory phenotype (SASP)-driven neurotoxicity by using human non-neuronal cells from aged donors, and by co-culturing human non-neuronal cells with human neuronal cells to study paracrine effects. To investigate non-neuronal cell type-specific effects (astrocytes, microglia) on cellular senescence. To investigate the impact of AD subtype-specific genetic background on SASP. To test efficacy of senolytics in AD patient-derived in vitro models.
*To study human-relevant mechanisms of neurotropic infection, pathogenesis, adaptation and latency, using healthy human/AD patient-derived brain tissues/organoids/brain-on-chip that mimic chronic neuroinflammatory environment 141, 142.
*To investigate the role of human neurotropic viruses in selective vulnerability of cell types and brain regions to AD in human CNS tissues/organoids/brain-on-chip 143, 144, 145.
*To identify biomarkers of AD subtypes and improve biomarker-guided stratification, beyond Aβ and tau, such as neuroinflammation and neurodegeneration markers, by AI-powered multi-omics analysis in large-scale cohort studies.
*To identify biomarkers of human cognitive dysfunction, by AI-aided analysis of correlations between biomarkers of human cognitive impairment (ribosomal proteins, transcriptions factors, neurotrophic factors, astrocytes & microglia activation markers, synaptic proteins, excitotoxicity, lipid peroxidation, glucose metabolism) and results of cognitive testing in large-scale longitudinal studies. To assess the degree of cognitive impairment in an individual by predictive analysis of patient-derived iPSC.
*To develop and test new combinations of antiviral drugs for AD patients who tested positive for antibodies against neurotropic viruses 146.
*To build predictive in vitro models of response to treatment, by AI-powered corelation between data collected in parallel from human brain organoids/brain-on-chip/human-on-chip (morphology, function, electrophysiology, multi-omics) and data from AD patients (mobile neuroimaging, behavioural tracking, cognitive score, CSF/plasma biomarkers).
*To improve predictivity of systemic toxic effects of mAb, by using human primary cells assays (cardiovascular, hepatic, immune, renal), co-culture systems, and multi-organ-on-chip/human-on-chip. Real-time measurement of mAb safety in vitro can be combined with PBPK models that simulate how mAb distribute throughout the human body.
*To test inhibitors of aβ and tau aggregation by real-time tracking of aggregation kinetics in familial AD scenario (patient-derived brain organoids carrying familial AD mutations) 147 and in sporadic AD scenario (healthy human brain-organoids chemically-induced to produce aβ) 148.
*To test the efficacy of inhibiting and disassembling tau aggregates, by high-throughput screening of compounds in cell-free tau biochemical assays 149 and optogenetically-induced tau in human neural progenitor cells 150.
*To identify AD subtype-specific molecular pathogenic pathways, To discover new therapeutic targets, and To assess the efficacy of multi-target drug combinations, by combining in vitro assays in human cerebral organoids with mathematical modeling (dynamic network model, regulation of pathways, in silico perturbation analysis) 151.
*To test safety and efficacy of personalized treatments for familial and sporadic AD in patient-specific 3D tissues/organoids/organs-on-chip.
Last Updated: September 2025
Although in vitro methods have inherent limitations, their relevance to human biology far exceeds that of animal research.
Animal model organisms were never comprehensively compared to humans and scientifically validated. Complementing in vitro methods with animal
experiments is not effective for human patients, because species-specific differences prevent reliable integration and translation of results to humans.
While animal research benefits from experimenting on a complete organism, model organisms fail to replicate the interplay of thousands of human-specific features, from molecular level to organism level, and are therefore not representative of the complete human organism.
To address the challenges of individual human-based in vitro models, they can be integrated with other human-based in vitro methods, AI-driven analysis, clinical data, and real-world patient data.
Have you leveraged in vitro methods in unique ways? We would love to hear how! Join the conversation to exchange ideas, collaborate and inspire new directions in human-based science!
References
1. World Health Organization: WHO, World Health Organization: WHO. Dementia. Mar 2025. https://www.who.int/en/news-room/fact-sheets/detail/dementia
2. Posteinsson AP, R.S. Isaacson, Sean Knox, M.N. Sabbagh, I. Rubino. Diagnosis of Early Alzheimer’s Disease: Clinical Practice in 2021. The Journal of Prevention of Alzheimer’s Disease. Jul 2021. https://doi.org/10.14283/jpad.2021.23
3. Morris GP, Clark IA, Vissel B. Questions concerning the role of amyloid-β in the definition, aetiology and diagnosis of Alzheimer’s disease. Acta Neuropathologica. 2018; 136(5):663–689. https://doi.org/10.1007/s00401-018-1918-8
4. Lane CA, Hardy J, Schott JM. Alzheimer’s disease. European Journal of Neurology. 2017; 25(1):59–70. https://doi.org/10.1111/ene.13439
5. Sirkis DW, Bonham LW, Johnson TP, La Joie R, Yokoyama JS. Dissecting the clinical heterogeneity of early-onset Alzheimer’s disease. Molecular Psychiatry. 2022; 27(6):2674–2688. https://doi.org/10.1038/s41380-022-01531-9
6. Seath P, Macedo-Orrego LE, Velayudhan L. Clinical characteristics of early-onset versus late-onset Alzheimer’s disease: a systematic review and meta-analysis. International Psychogeriatrics. 2023; 36(12):1093–1109. https://doi.org/10.1017/s1041610223000509
7. Ryan NS, Willumsen N, Nicholas JM, Poole T, Pavisic IM, Walsh P, Rice H, Lu K, O’Connor A, Weston PS, et al. Atypical presentations of autosomal dominant familial Alzheimer’s disease: insights into genetic, neuropathological and clinical heterogeneity. Alzheimer S & Dementia. 2023; 19(S12). https://doi.org/10.1002/alz.076186
8. Van Der Flier Wiesje M. Clinical heterogeneity in familial Alzheimer’s disease. The Lancet Neurology. Volume 15, Issue 13p, 1296-1298, Dec 2016
https://doi.org/10.1016/S1474-4422(16)30275-7
9. Attems J, Jellinger KA. The overlap between vascular disease and Alzheimer’s disease - lessons from pathology. BMC Medicine. 2014; 12(1). https://doi.org/10.1186/s12916-014-0206-2
10. Banerjee G, Carare R, Cordonnier C, Greenberg SM, Schneider JA, Smith EE, Van Buchem M, Van Der Grond J, Verbeek MM, Werring DJ. The increasing impact of cerebral amyloid angiopathy: essential new insights for clinical practice. Journal of Neurology Neurosurgery & Psychiatry. 2017; 88(11):982–994. https://doi.org/10.1136/jnnp-2016-314697
11. Kumar A, Sidhu J, Lui F, Tsao JW. Alzheimer Disease. StatPearls - NCBI Bookshelf. Feb 2024. https://www.ncbi.nlm.nih.gov/books/NBK499922/
12. DeTure MA, Dickson DW. The neuropathological diagnosis of Alzheimer’s disease. Molecular Neurodegeneration. 2019; 14(1). https://doi.org/10.1186/s13024-019-0333-5
13. Zhou Z, Wang Q, Liu L, Wang Q, Zhang X, Li C, Liu J, Wei Y, Gao J, Fu L, et al. Investigating the Aβ and tau pathology in autosomal dominant Alzheimer’s disease: insights from hybrid PET/MRI and network mapping. Alzheimer S Research & Therapy. 2025; 17(1). https://doi.org/10.1186/s13195-025-01690-1
14. Wachter A, Woodbury ME, Lombardo S, Abdourahman A, Wuest C, McGlame E, Pastika T, Tamm J, Romanul N, Yanamandra K, et al. Landscape of brain myeloid cell transcriptome along the spatiotemporal progression of Alzheimer’s disease reveals distinct sequential responses to Aβ and tau. Acta Neuropathologica. 2024; 147(1). https://doi.org/10.1007/s00401-024-02704-2
15. Braak H., Thal D. R., Ghebremedhin E., Tredici K. D. Stages of the pathologic process in Alzheimer disease: age categories from 1 to 100 years. Journal of Neuropathology & Experimental Neurology, Volume 70, Issue 11, Nov 2011, Pages 960–969. Nov. 2011. https://doi.org/10.1097/NEN.0b013e318232a379
16. Gatz Margaret, Reynolds Chandra, Fratiglioni Laura et al. Role of genes and environments for explaining Alzheimer disease. JAMA Psychiatry Vol. 63, No. 2, 2006; 63;(2):168-174. https://doi.org/10.1001/archpsyc.63.2.168
17. Barnard N., Bush A. I., Ceccarelli A., Cooper J., De Jager C.., Erickson K. I., Fraser G., Kesler S., Levin S. M., Lucey B., Morris M. C., Squitti R.. Dietary and lifestyle guidelines for the prevention of Alzheimer’s disease. Neurobiology of Aging, Volume 35, Supplement 2, Sep 2014, Pages S74-S78. https://doi.org/10.1016/j.neurobiolaging.2014.03.033
18. Jack Jr Clifford R., Wiste Heather J., Weigand Stephen D. et al. Age, sex, and APOE Ε4 effects on memory, brain structure, and Β-Amyloid across the adult life span. JAMA Neurology Vol. 72, No. 5, 72;(5):511-519. May 2015 https://doi.org/10.1001/jamaneurol.2014.4821
19. Vialle RA, De Paiva Lopes K, Li Y, Ng B, Schneider JA, Buchman AS, Wang Y, Farfel JM, Barnes LL, Wingo AP, et al. Structural variants linked to Alzheimer’s disease and other common age-related clinical and neuropathologic traits. Genome Medicine. 2025; 17(1). https://doi.org/10.1186/s13073-025-01444-6
20. Boyle PA, Wilson RS, Yu L, Barr AM, Honer WG, Schneider JA, Bennett DA. Much of late life cognitive decline is not due to common neurodegenerative pathologies. Annals of Neurology. 2013; 74(3):478–489. https://doi.org/10.1002/ana.23964
21. Lane CA, Hardy J, Schott JM. Alzheimer’s disease. European Journal of Neurology. 2017;25(1):59–70. https://doi.org/10.1111/ene.13439
22. Bun S, Ito D, Tezuka T, Kubota M, Ueda R, Takahata K, Moriguchi S, Kurose S, Momota Y, Suzuki N, et al. Performance of plasma Aβ42/40, measured using a fully automated immunoassay, across a broad patient population in identifying amyloid status. Alzheimer S Research & Therapy. 2023; 15(1). https://doi.org/10.1186/s13195-023-01296-5
23. Mawuenyega KG, Sigurdson W, Ovod V, Munsell L, Kasten T, Morris JC, Yarasheski KE, Bateman RJ. Decreased clearance of CNS Β-Amyloid in Alzheimer’s disease. Science. 2010; 330(6012):1774. https://doi.org/10.1126/science.1197623
24. Andrade-Guerrero J., Santiago-Balmaseda A., Jeronimo-Aguilar P., Vargas-Rodríguez I., Cadena-Suárez A. R, Sánchez-Garibay C., Pozo-Molina G., Méndez-Catalá C.F., Cardenas-Aguayo M-del-C., Diaz-Cintra S., Pacheco-Herrero M., Luna-Muñoz J., Soto-Rojas L. O. Alzheimer’s Disease: An updated overview of its genetics. International Journal of Molecular Sciences. 24(4), 3754. Feb 2023. https://doi.org/10.3390/ijms24043754
25. Cummings J, Lee G, Nahed P, Kambar Mezn, Zhong K, Fonseca J, Taghva K. Alzheimer’s disease drug development pipeline: 2022. Alzheimer S & Dementia Translational Research & Clinical Interventions. 2022; 8(1). https://doi.org/10.1002/trc2.12295
26. Alzheimer's Disease Sequencing Project, List of AD Loci and Genes with Genetic Evidence Compiled by ADSP Gene Verification Committee – ADSP, Aug 2023
27. Hoogmartens J, Cacace R, Van Broeckhoven C. Insight into the genetic etiology of Alzheimer’s disease: A comprehensive review of the role of rare variants. Alzheimer S & Dementia Diagnosis Assessment & Disease Monitoring. 2021; 13(1). https://doi.org/10.1002/dad2.12155
28. Lanoiselée H-M, Nicolas G, Wallon D, Rovelet-Lecrux A, Lacour M, Rousseau S, Richard A-C, Pasquier F, Rollin-Sillaire A, Martinaud O, et al. APP, PSEN1, and PSEN2 mutations in early-onset Alzheimer disease: A genetic screening study of familial and sporadic cases. PLoS Medicine. 2017; 14(3):e1002270. https://doi.org/10.1371/journal.pmed.1002270
29. Nicolas A, Sherva R, Grenier-Boley B, Kim Y, Kikuchi M, Timsina J, De Rojas I, Dalmasso MC, Zhou X, Guen YL, et al. Transferability of European-derived Alzheimer’s disease polygenic risk scores across multiancestry populations. Nature Genetics. Jun 2025. https://doi.org/10.1038/s41588-025-02227-w
30. De Jager PL, Srivastava G, Lunnon K, Burgess J, Schalkwyk LC, Yu L, Eaton ML, Keenan BT, Ernst J, McCabe C, et al. Alzheimer’s disease: early alterations in brain DNA methylation at ANK1, BIN1, RHBDF2 and other loci. Nature Neuroscience. 2014; 17(9) :1156–1163. https://doi.org/10.1038/nn.3786
31. Marín O. Interneuron dysfunction in psychiatric disorders. Nature Reviews. Neuroscience. 2012; 13(2) :107–120. https://doi.org/10.1038/nrn3155
32. Szegedi V, Paizs M, Csakvari E, Molnar G, Barzo P, Tamas G, Lamsa K. Plasticity in single axon glutamatergic connection to GABAergic interneurons regulates complex events in the human neocortex. PLoS Biology. 2016; 14(11):e2000237. https://doi.org/10.1371/journal.pbio.2000237
33. González‐Acosta CA, Ortiz‐Muñoz D, Becerra‐Hernández LV, Casanova MF, Buriticá E. Von Economo neurons: Cellular specialization of human limbic cortices? Journal of Anatomy. 2022; 241(1):20–32. https://doi.org/10.1111/joa.13642
34. Mihaljević B, Larrañaga P, Bielza C. Comparing the electrophysiology and morphology of human and mouse layer 2/3 pyramidal neurons with Bayesian networks. Frontiers in Neuroinformatics. 2021; 15. https://doi.org/10.3389/fninf.2021.580873
35. Berg J, Sorensen SA, Ting JT, Miller JA, Chartrand T, Buchin A, Bakken TE, Budzillo A, Dee N, Ding S-L, et al. Human neocortical expansion involves glutamatergic neuron diversification. Nature. 2021 ; 598(7879):151–158. https://doi.org/10.1038/s41586-021-03813-8
36. Jones Edward G. The origins of Cortical interneurons: Mouse versus Monkey and Human. Cerebral Cortex, Volume 19, Issue 9, September 2009, Pages 1953–1956. May 2009. https://doi.org/10.1093/cercor/bhp088
37. Lee BR, Dalley R, Miller JA, Chartrand T, Close J, Mann R, Mukora A, Ng L, Alfiler L, Baker K, et al. Signature morphoelectric properties of diverse GABAergic interneurons in the human neocortex. Science. 2023; 382(6667). https://doi.org/10.1126/science.adf6484
38. Ardesch DJ, Scholtens LH, Li L, Preuss TM, Rilling JK, Van Den Heuvel MP. Evolutionary expansion of connectivity between multimodal association areas in the human brain compared with chimpanzees. Proceedings of the National Academy of Sciences. 2019; 116(14):7101–7106. https://doi.org/10.1073/pnas.1818512116
39. Hodge RD, Bakken TE, Miller JA, Smith KA, Barkan ER, Graybuck LT, Close JL, Long B, Johansen N, Penn O, et al. Conserved cell types with divergent features in human versus mouse cortex. Nature. 2019; 573(7772):61–68. https://doi.org/10.1038/s41586-019-1506-7
40. Oberheim NA, Takano T, Han X, He W, Lin JHC, Wang F, Xu Q, Wyatt JD, Pilcher W, Ojemann JG, et al. Uniquely hominid features of adult human astrocytes. Journal of Neuroscience. 2009; 29(10):3276–3287. https://doi.org/10.1523/jneurosci.4707-08.2009
41. Masuda T, Sankowski R, Staszewski O, Böttcher C, Amann L, Sagar N, Scheiwe C, Nessler S, Kunz P, Van Loo G, et al. Spatial and temporal heterogeneity of mouse and human microglia at single-cell resolution. Nature. 2019; 566(7744):388–392. https://doi.org/10.1038/s41586-019-0924-x
42. Geirsdottir L., David E., Keren-Shaul H., Weiner A., Bohlen S. C., Neuber J., Balic A., Giladi A., Sheban F., Dutertre C-A, Prefile C., Peri F. et al. Cross-Species Single-Cell analysis reveals divergence of the Primate Microglia program. Cell, Volume 179, Issue 7p 1609-1622.e16 Dec 2019. https://doi.org/10.1016/j.cell.2019.11.010
43. Yang AC, Vest RT, Kern F, Lee DP, Agam M, Maat CA, Losada PM, Chen MB, Schaum N, Khoury N, et al. A human brain vascular atlas reveals diverse mediators of Alzheimer’s risk. Nature. 2022; 603(7903):885–892. https://doi.org/10.1038/s41586-021-04369-3
44. Mestas Javier, Hughes Christopher C W. Of mice and not men: Differences between mouse and human immunology. The Journal of Immunology, Volume 172, Issue 5, Mar 2004, Pages 2731–2738. https://doi.org/10.4049/jimmunol.172.5.2731
45. Yegian AK, Heymsfield SB, Castillo ER, Müller MJ, Redman LM, Lieberman DE. Metabolic scaling, energy allocation tradeoffs, and the evolution of humans’ unique metabolism. Proceedings of the National Academy of Sciences. 2024; 121(48). https://doi.org/10.1073/pnas.2409674121
46. Shang Z, Li X. Human cytomegalovirus: pathogenesis, prevention, and treatment. Molecular Biomedicine. 2024; 5(1). https://doi.org/10.1186/s43556-024-00226-7
47. Chandrasekera PC, Pippin JJ. Of rodents and men: Species-specific glucose regulation and type 2 diabetes research. ALTEX. 2014 Jan 1:157–176. https://doi.org/10.14573/altex.1309231
48. Zoeller RT, Tan SW, Tyl RW. General background on the Hypothalamic-Pituitary-Thyroid (HPT) axis. Critical Reviews in Toxicology. 2007; 37(1–2):11–53. https://doi.org/10.1080/10408440601123446
49. Kieser S, Zdobnov EM, Trajkovski M. Comprehensive mouse microbiota genome catalog reveals major difference to its human counterpart. PLoS Computational Biology. 2022; 18(3):e1009947. https://doi.org/10.1371/journal.pcbi.1009947
50. Bergen Werner G., Mersmann Harry J. Comparative aspects of lipid metabolism: impact on contemporary research and use of animal models. The Journal of Nutrition, Volume 135, Issue 11, Nov 2005, Pages 2499-2502. https://doi.org/10.1093/jn/135.11.2499
51. Loerch PM, Lu T, Dakin KA, Vann JM, Isaacs A, Geula C, Wang J, Pan Y, Gabuzda DH, Li C, et al. Evolution of the aging brain transcriptome and synaptic regulation. PLoS ONE. 2008; 3(10):e3329. https://doi.org/10.1371/journal.pone.0003329
52. Sarropoulos I, Marin R, Cardoso-Moreira M, Kaessmann H. Developmental dynamics of lncRNAs across mammalian organs and species. Nature. 2019; 571(7766):510–514. https://doi.org/10.1038/s41586-019-1341-x
53. Lin J., Wen Y., Tang J., Zhang X., Zhang H., Zhu H. Human-specific LNCRNAs contributed critically to human evolution by distinctly regulating gene expression. eLife, Jul 2025. https://doi.org/10.7554/eLife.89001.3
54. Zhao F, Yan Y, Wang Y, Liu Y, Yang R. Splicing complexity as a pivotal feature of alternative exons in mammalian species. BMC Genomics. 2023; 24(1). https://doi.org/10.1186/s12864-023-09247-y
55. Modrek B, Lee CJ. Alternative splicing in the human, mouse and rat genomes is associated with an increased frequency of exon creation and/or loss. Nature Genetics. 2003; 34(2):177–180. https://doi.org/10.1038/ng1159
56. Granzotto A, Vissel B, Sensi SL. Lost in translation: Inconvenient truths on the utility of mouse models in Alzheimer’s disease research. eLife. 2024;13. https://doi.org/10.7554/elife.90633
57. Li L, Ohman T, Deeb SS, Fukuchp K-I. Analysis of Mouse Intron 7 DNA Sequence of the APP Gene: Comparison with the Human Homologue. DNA Sequence. 1999; 10(4–5):219–228. https://doi.org/10.3109/10425179909033951
58. Frieden C, Garai K. Structural differences between apoE3 and apoE4 may be useful in developing therapeutic agents for Alzheimer’s disease. Proceedings of the National Academy of Sciences. 2012; 109(23):8913–8918. https://doi.org/10.1073/pnas.1207022109
59. Huebbe P., Rimbach G.. Evolution of human apolipoprotein E (APOE) isoforms: gene structure, protein function and interaction with dietary factors. Ageing Research Reviews, Volume 37, Aug 2017, Pages 146-161. https://doi.org/10.1016/j.arr.2017.06.002
60. Dietschy John M., Turley Stephen D. Brain lipids. cholesterol metabolism in the central nervous system during early development and in the mature animal. Journal of Lipid Research, Volume 45, Issue 8, Aug 2004, Pages 1375-1397. https://doi.org/10.1194/jlr.R400004-JLR200
61. Alzheimer’s Disease Research Models | ALZ forum Database, 1999-2025 https://www.alzforum.org/research-models/alzheimers-disease
62. Okaichi H, Oshima Y, Jarrard L. E. Scopolamine impairs both working and reference memory in rats: a replication and extension. Pharmacology Biochemistry and Behavior Volume 34, Issue 3, Nov 1989, Pages 599-602. https://doi.org/10.1016/0091-3057(89)90565-0
63. Robinson JK, Wenk GL, Wiley RG, Lappi DA, Crawley JN. 192IgG-Saporin immunotoxin and ibotenic acid lesions of nucleus basalis and medial septum produce comparable deficits on delayed nonmatching to position in rats. Psychobiology. 1996; 24(3):179–186. https://doi.org/10.3758/bf03327034
64. Yagi H, Katoh S, Akiguchi I, Takeda T. Age-related deterioration of ability of acquisition in memory and learning in senescence Accelerated mouse: SAM-P/8 as an animal model of disturbances in recent memory. Brain Research Volume 474, Issue 1, 22 Nov 1988, Pages 86-93. https://doi.org/10.1016/0006-8993(88)90671-3
65. Fernández A, Quintana E, Velasco P, Moreno-Jimenez B, De Andrés B, Gaspar ML, Liste I, Vilar M, Mira H, Cano E. Senescent accelerated prone 8 (SAMP8) mice as a model of age dependent neuroinflammation. Journal of Neuroinflammation. 2021;18(1). https://doi.org/10.1186/s12974-021-02104-3
66. Kalback W., Watson M.D, Kokjohn T.A, Kuo Y-M, Weiss N., Luehrs D.C, Lopez J., Brune D., Sisodia S. S, Staufenbiel M., Emmerling M, Roher a. E. APP transgenic mice TG2576 accumulate AΒ peptides that are distinct from the chemically modified and insoluble peptides deposited in Alzheimer’s disease senile plaques. Biochemistry, Vol 41 - Issue 3, Dec 2001. https://doi.org/10.1021/bi015685+
67. Schwab C., Hosokawa M., McGeer P.L. Transgenic mice overexpressing amyloid beta protein are an incomplete model of Alzheimer disease. Experimental Neurology Volume 188, Issue 1, Jul 2004, Pages 52-64. https://doi.org/10.1016/j.expneurol.2004.03.016
68. Sasaguri H, Nilsson P, Hashimoto S, Nagata K, Saito T, De Strooper B, Hardy J, Vassar R, Winblad B, Saido TC. APP mouse models for Alzheimer’s disease preclinical studies. The EMBO Journal. 2017;36(17):2473–2487. https://doi.org/10.15252/embj.201797397
69. Oddo S., Caccamo a., Shepherd J. D, Murphy M. P, Golde T. E, Kayed R., Metherate R., Mattson M. P, Akbari Y., LaFerla F. Triple-Transgenic model of Alzheimer’s disease with plaques and tangles. Neuron, Volume 39, Issue 3p409-421, Jul 2003. https://doi.org/10.1016/S0896-6273(03)00434-3
70. Li L, Jiang Y, Hu W, Tung YC, Dai C, Chu D, Gong C-X, Iqbal K, Liu F. Pathological alterations of TAU in Alzheimer’s disease and 3xTG-AD mouse brains. Molecular Neurobiology. 2019; 56(9):6168–6183. https://doi.org/10.1007/s12035-019-1507-4
71. Grueninger F., Bohrmann B., Czech C., Ballard T. M., Frey J. R., Weidensteiner C., Von Kienlin M., Ozmen L. Phosphorylation of TAU at S422 is enhanced by AΒ in TAUPS2APP triple transgenic mice. Neurobiology of Disease Volume 37, Issue 2, Feb 2010, Pages 294-306. https://doi.org/10.1016/j.nbd.2009.09.004
72. Saito T, Matsuba Y, Mihira N, Takano J, Nilsson P, Itohara S, Iwata N, Saido TC. Single App knock-in mouse models of Alzheimer’s disease. Nature Neuroscience. 2014; 17(5):661–663. https://doi.org/10.1038/nn.3697
73. Malthankar-Phatak GH, Lin Y-G, Giovannone N, Siman R. Amyloid deposition and advanced age fails to induce Alzheimer’s type progression in a double Knock-In mouse model. Jul 2011. https://pmc.ncbi.nlm.nih.gov/articles/PMC3377826/#b53-ad-3-2-141
74. Jiang L, Roberts R, Wong M, Zhang L, Webber CJ, Libera J, Wang Z, Kilci A, Jenkins M, Ortiz AR, et al. β-amyloid accumulation enhances microtubule associated protein tau pathology in an APPNL-G-F/MAPTP301S mouse model of Alzheimer’s disease. Frontiers in Neuroscience. 2024; 18. https://doi.org/10.3389/fnins.2024.1372297
75. Sato K., Watamura N., Fujioka R., Mihira N., Sekiguchi M., Nagata K., Ohshima T., Saito T., Saido T. C, Sasaguri H. New app knock-in mice that accumulate wild-type human AΒ as rapidly as APpNL-G-F mice exhibit intensive cored plaque pathology and neuroinflammation. BioRxiv, Apr 2021. https://doi.org/10.1101/2021.04.30.442122
76. Ganesan K, Rentsch P, Langdon A, Milham LT, Vissel B. Modeling sporadic Alzheimer’s disease in mice by combining Apolipoprotein E4 risk gene with environmental risk factors. Frontiers in Aging Neuroscience. 2024; 16. https://doi.org/10.3389/fnagi.2024.1357405
77. Shankar GM, Li S, Mehta TH, Garcia-Munoz A, Shepardson NE, Smith I, Brett FM, Farrell MA, Rowan MJ, Lemere CA, et al. Amyloid-β protein dimers isolated directly from Alzheimer’s brains impair synaptic plasticity and memory. Nature Medicine. 2008; 14(8):837–842. https://doi.org/10.1038/nm1782
78. Morris GP, Clark IA, Vissel B. Inconsistencies and controversies surrounding the amyloid hypothesis of Alzheimer’s disease. Acta Neuropathologica Communications. 2014; 2(1). https://doi.org/10.1186/s40478-014-0135-5
79. Robinson JK, Wenk GL, Wiley RG, Lappi DA, Crawley JN. 192IgG-Saporin immunotoxin and ibotenic acid lesions of nucleus basalis and medial septum produce comparable deficits on delayed nonmatching to position in rats. Psychobiology. 1996; 24(3):179–186. https://doi.org/10.3758/bf03327034
80. Fernández A, Quintana E, Velasco P, Moreno-Jimenez B, De Andrés B, Gaspar ML, Liste I, Vilar M, Mira H, Cano E. Senescent accelerated prone 8 (SAMP8) mice as a model of age dependent neuroinflammation. Journal of Neuroinflammation. 2021; 18(1). https://doi.org/10.1186/s12974-021-02104-3
81. Khan I, De Jesus O. Frontotemporal lobe dementia. StatPearls - NCBI Bookshelf. Aug 2023. https://www.ncbi.nlm.nih.gov/books/NBK559286/
82. Ando K, Kosa A-C, Mehadji Y, Lasri H, Lopez-Gutierrez L, Quintanilla-Sánchez C, Aydin E, Doeraene E, Wathelet-Depauw a, Nagaraj S, Brion J-P, Leroy K. Deletion of murine APP aggravates TAU and amyloid pathologies in the 5xFADXTG30 Alzheimer’s Disease model. Biomolecules, 15(2), 159, Jan 2025. https://doi.org/10.3390/biom15020159
83. Yang H, Kim D, Yang YS, Bagyinszky E, Soo AS. TREM2 in Neurodegenerative Disorders: mutation spectrum, pathophysiology, and therapeutic targeting. International Journal of Molecular Sciences, 26(15), 7057, Jul 2025. https://doi.org/10.3390/ijms26157057
84. Van Heuvelen MJG , Van Der Lei MB, Alferink PM, Roemers P, Van Der Zee EA. Cognitive Deficits in human APOE4 knock-in mice: A Systematic Review and Meta-analysis. Behavioural Brain Research Volume 471, 5 Aug 2024, 115123. https://doi.org/10.1016/j.bbr.2024.115123
85. Van Gool B, Storck SE, Reekmans SM, Lechat B, Gordts PLSM, Pradier L, Pietrzik CU, Roebroek AJM. LRP1 Has a Predominant Role in Production over Clearance of Aβ in a Mouse Model of Alzheimer’s Disease. Molecular Neurobiology. 2019; 56(10):7234–7245. https://doi.org/10.1007/s12035-019-1594-2
86. Warren SL, Reid E, Whitfield P, Helal AM, Hamza EGA, Tindle R, Moustafa AA, Hamid MS. Cognitive and behavioral abnormalities in individuals with Alzheimer’s disease, mild cognitive impairment, and subjective memory complaints. Current Psychology. 2023; 43(1):800–810. https://doi.org/10.1007/s12144-023-04281-1
87. Eikelboom WS, Van Den Berg E, Singleton EH, Baart SJ, Coesmans M, Leeuwis AE, Teunissen CE, Van Berckel BNM, Pijnenburg YAL, Scheltens P, et al. Neuropsychiatric and cognitive symptoms across the Alzheimer disease clinical spectrum. Neurology. 2021; 97(13). https://doi.org/10.1212/wnl.0000000000012598
88. Willner P. Animal models as simulations of depression. Trends in Pharmacological Science. 1991 Apr; 12(4):131-6. https://doi.org/10.1016/0165-6147(91)90529-2
89. Van Den Berg H. Evaluating the validity of animal models of mental disorder: from modeling syndromes to modeling endophenotypes. History & Philosophy of the Life Sciences. 2022; 44(4). https://doi.org/10.1007/s40656-022-00537-4
90. Paul SM, Potter WZ. Finding new and better treatments for psychiatric disorders. Neuropsychopharmacology. 2023;49(1):3–9. https://doi.org/10.1038/s41386-023-01690-5
91. Nithianantharajah J., Grant S. G.N. Cognitive components in mice and humans: Combining genetics and touchscreens for medical translation. Neurobiology of Learning and Memory Volume 105, Oct 2013, Pages 13-19. https://doi.org/10.1016/j.nlm.2013.06.006
92. De Figueiredo Cerqueira MM, Castro MML, Vieira AA, Kurosawa JAA, Amaral FLD Junior, De Carvalho Chaves De Siqueira Mendes F, Sosthenes MCK. Comparative analysis between Open Field and Elevated Plus Maze tests as a method for evaluating anxiety-like behavior in mice. Heliyon. 2023; 9(4):e14522. https://doi.org/10.1016/j.heliyon.2023.e14522
93. Wolfer DP, Stagljar-Bozicevic M, Errington ML, Lipp H-P. Spatial memory and learning in transgenic mice: fact or artifact? Physiology. 1998; 13(3):118–123. https://doi.org/10.1152/physiologyonline.1998.13.3.118
94. Rosso M, Wirz R, Loretan AV, Sutter NA, Da Cunha CTP, Jaric I, Würbel H, Voelkl B. Reliability of common mouse behavioural tests of anxiety: A Systematic review and Meta-analysis on the effects of anxiolytics. Neuroscience & Biobehavioral Reviews Volume 143, Dec 2022, 104928. https://doi.org/10.1016/j.neubiorev.2022.104928
95. Lueptow LM. Novel Object Recognition test for the investigation of learning and memory in mice. Journal of Visualized Experiments. 2017;(126). https://doi.org/10.3791/55718
96. Jabarin R, Netser S, Wagner S. Beyond the three-chamber test: toward a multimodal and objective assessment of social behavior in rodents. Molecular Autism. 2022; 13(1). https://doi.org/10.1186/s13229-022-00521-6
97. Várkonyi D, Török B, Sipos E, Fazekas CL, Bánrévi K, Correia P, Chaves T, Farkas S, Szabó a, Martínez-Bellver S, Hangya B, Zelena D. Investigation of anxiety- and depressive-like symptoms in 4- and 8-Month-Old male triple transgenic mouse models of Alzheimer’s disease. International Journal of Molecular Sciences, 2022, 23(18), 10816. https://doi.org/10.3390/ijms231810816
98. Trunnell ER, Carvalho C. The forced swim test has poor accuracy for identifying novel antidepressants. Drug Discovery Today Volume 26, Issue 12, Dec 2021, Pages 2898-2904. https://doi.org/10.1016/j.drudis.2021.08.003
99. Statement on the forced swim test in rodent models. NHMRC, Dec 2023. https://www.nhmrc.gov.au/research-policy/ethics/statement-forced-swim-test-rodent-models
100. Reardon S. Pressure grows to ditch controversial forced swim test in rodent studies of depression. Science, Vol 383, Issue 6689, Mar 2024. https://doi.org/10.1126/science.zrf1p5t
101. Bailey J. Does the stress of laboratory life and experimentation on animals adversely affect research data? A critical review. Alternatives to Laboratory Animals. 2018; 46(5):291–305. https://doi.org/10.1177/026119291804600501
102. Akhtar A. The flaws and human harms of animal experimentation. Cambridge Quarterly of Healthcare Ethics. 2015; 24(4):407–419. https://doi.org/10.1017/s0963180115000079
103. Wahlsten D. Standardizing tests of mouse Behavior: Reasons, recommendations, and reality. Physiology & Behavior Volume 73, Issue 5, Aug 2001, Pages 695-704. https://doi.org/10.1016/S0031-9384(01)00527-3
104. Götz J, Ittner LM. Animal models of Alzheimer’s disease and frontotemporal dementia. Nature Reviews. Neuroscience. 2008; 9(7):532–544. https://doi.org/10.1038/nrn2420
105. Sims JR, Zimmer JA, Evans CD. Donanemab in early symptomatic Alzheimer disease the TRAILBLAZER-ALZ 2 randomized clinical trial. JAMA Vol. 330, No. 6, Jul 2023. https://doi.org/10.1001/jama.2023.13239
106. De Strooper B. Lessons from a failed Γ-Secretase Alzheimer trial. Cell, Volume 159, Issue 4p721-726, Nov 2014. https://doi.org/10.1016/j.cell.2014.10.016
107. Kim B-H, Kim S, Nam Y, Park YH, Shin SM, Moon M. Second-generation anti-amyloid monoclonal antibodies for Alzheimer’s disease: current landscape and future perspectives. Translational Neurodegeneration. 2025;14(1). https://doi.org/10.1186/s40035-025-00465-w
108. Sa AI. AC immune to regain global rights to crenezumab and semorinemab. GlobeNewswire News Room. Jan 2024. https://www.globenewswire.com/news-release/2024/01/22/2812913/0/en/AC-Immune-to-Regain-Global-Rights-to-Crenezumab-and-Semorinemab.html
109. ClinicalTrials.gov. https://clinicaltrials.gov/study/NCT04867616
110. Van Dyck CH, Swanson CJ, Aisen P, Bateman RJ, Chen C, Gee M, Kanekiyo M, Li D, Reyderman L, Cohen S, et al. Lecanemab in early Alzheimer’s disease. New England Journal of Medicine. 2022; 388(1):9–21. https://doi.org/10.1056/nejmoa2212948
111. Sarkisian CA, Romanov A, Mafi JN. Talking with patients about the new anti-amyloid Alzheimer disease medications. Annals of Internal Medicine. 2024; 177(2):246–248. https://doi.org/10.7326/m23-3377
112. Rashad a, Rasool a, Shaheryar M, Sarfraz a, Sarfraz Z, Robles-Velasco K, Cherrez-Ojeda I. Donanemab for Alzheimer’s Disease: A Systematic review of Clinical trials. Healthcare, 11(1), 32; Dec 2022. https://doi.org/10.3390/healthcare11010032
113. Company ELA. Lilly’s Donanemab significantly slowed cognitive and functional decline in phase 3 study of early Alzheimer’s disease. May 2023. https://www.prnewswire.com/news-releases/lillys-donanemab-significantly-slowed-cognitive-and-functional-decline-in-phase-3-study-of-early-alzheimers-disease-301814001.html
114. Travis J. ‘It’s not a miracle drug’: Eli Lilly’s antibody slows Alzheimer’s disease but safety issues linger. Science, May 2023 https://doi.org/10.1126/science.adi5580
115. Dyer O. Aduhelm: Biogen abandons Alzheimer’s drug after controversial approval left it unfunded by Medicare. BMJ. Feb 2024 :q281. https://doi.org/10.1136/bmj.q281
116. Piller C. Scientists tie third clinical trial death to experimental Alzheimer’s drug. Science, Dec 2022. https://doi.org/10.1126/science.adg4121
117. Tonegawa-Kuji R, Hou Y, Hu B, Lorincz-Comi N, Pieper AA, Tousi B, Leverenz JB, Cheng F. Efficacy and safety of passive immunotherapies targeting amyloid beta in Alzheimer’s disease: A systematic review and meta-analysis. PLoS Medicine. 2025; 22(3):e1004568. https://doi.org/10.1371/journal.pmed.1004568
118. Biogen. European Commission grants marketing Authorization for Leqembi® (LecanemAB). Apr 2025. https://investors.biogen.com/news-releases/news-release-details/european-commission-grants-marketing-authorization-leqembir
119. Kisunla | European Medicines Agency (EMA). Jul 2025. https://www.ema.europa.eu/en/medicines/human/EPAR/kisunla
120. Alves F, Kalinowski P, Ayton S. Accelerated brain volume loss caused by Anti–Β-Amyloid drugs. Neurology. 2023; 100(20). https://doi.org/10.1212/wnl.0000000000207156
121. Zhang J, Kong G, Yang J, Pang L, Li X. Pathological mechanisms and treatment progression of Alzheimer’s disease. European Journal of Medical Research. 2025; 30(1). https://doi.org/10.1186/s40001-025-02886-9
122. De Vleeschauwer S , Lambaerts K, Hernot S, Karlijn Debusschere K. Severity classification of laboratory animal procedures in two Belgian academic institutions. Animals, 13(16), 2581, Aug 2023. https://doi.org/10.3390/ani13162581
123. Clinical development success rates and contributing Factors 2011-2020 | BIO. https://www.bio.org/clinical-development-success-rates-and-contributing-factors-2011-2020
124. Cummings JL, Morstorf T, Zhong K. Alzheimer’s disease drug-development pipeline: few candidates, frequent failures. Alzheimer S Research & Therapy. 2014; 6(4). https://doi.org/10.1186/alzrt269
125. Freedman LP, Cockburn IM, Simcoe TS. The Economics of Reproducibility in Preclinical research. PLoS Biology. 2015; 13(6):e1002165. https://doi.org/10.1371/journal.pbio.1002165
126. ARRIVE Guidelines. https://arriveguidelines.org/
127. Du Sert NP, Hurst V, Ahluwalia A, Alam S, Avey MT, Baker M, Browne WJ, Clark A, Cuthill IC, Dirnagl U, et al. The ARRIVE guidelines 2.0: Updated guidelines for reporting animal research. BMC Veterinary Research. 2020; 16(1). https://doi.org/10.1186/s12917-020-02451-y
128. Mazoit JX. Statistics in Biology: A survey of the three major multidisciplinary journals. BioRxiv, Feb 2025. https://doi.org/10.1101/2025.02.04.636422
129. Yarborough M, Bredenoord A, D’Abramo F, Joyce NC, Kimmelman J, Ogbogu U, Sena E, Strech D, Dirnagl U. The bench is closer to the bedside than we think: Uncovering the ethical ties between preclinical researchers in translational neuroscience and patients in clinical trials. PLoS Biology. 2018; 16(6):e2006343. https://doi.org/10.1371/journal.pbio.2006343
130. Smaldino PE, McElreath R. The natural selection of bad science. Royal Society Open Science. 2016;3(9):160384. https://doi.org/10.1098/rsos.160384
131. Lagomarsino VN, Pearse RV, Liu L, Hsieh Y-C, Fernandez MA Et Al. Stem cell-derived neurons reflect features of protein networks, neuropathology, and cognitive outcome of their aged human donors. Neuron, Volume 109, Issue 21p 3402-3420.e9, Nov 2021. https://doi.org/10.1016/j.neuron.2021.08.003
132. Boldog E, Bakken TE, Hodge RD, Novotny M, Aevermann BD, Baka J, Bordé S, Close JL, Diez-Fuertes F, Ding S-L, et al. Transcriptomic and morphophysiological evidence for a specialized human cortical GABAergic cell type. Nature Neuroscience. 2018 ; 21(9):1185–1195. https://doi.org/10.1038/s41593-018-0205-2
133. Lau S-F, Cao H, Fu AKY, Ip NY. Single-nucleus transcriptome analysis reveals dysregulation of angiogenic endothelial cells and neuroprotective glia in Alzheimer’s disease. Proceedings of the National Academy of Sciences. 2020 ; 117(41):25800–25809. https://doi.org/10.1073/pnas.2008762117
134. Xu D, Zhang C, Bi X, Xu J, Guo S, Li P Et Al. Mapping Enhancer and Chromatin Accessibility Landscapes charts the regulatory network of Alzheimer’s disease. Computers in Biology and Medicine Volume 168, Jan 2024, 107802. https://doi.org/10.1016/j.compbiomed.2023.107802
135. Gabitto MI, Travaglini KJ, Rachleff VM, Kaplan ES, Long B, Ariza J, Ding Y, Mahoney JT, Dee N, Goldy J, et al. Integrated multimodal cell atlas of Alzheimer’s disease. Nature Neuroscience. 2024 Oct 14. https://doi.org/10.1038/s41593-024-01774-5
136. Mertens J, Paquola ACM, Ku M, Yao J, Hetzer MW, Gage FH Et Al. Directly reprogrammed human neurons retain Aging-Associated transcriptomic signatures and reveal Age-Related nucleocytoplasmic defects. Cell Stem Cell, Volume 17, Issue 6p705-718, Dec 2015. https://doi.org/10.1016/j.stem.2015.09.001
137. Tian C, Ao Z, Cerneckis J, Cai H, Chen L, Niu H, Takayama K, Kim J, Shi Y, Gu M, et al. Understanding monocyte-driven neuroinflammation in Alzheimer’s disease using human cortical organoid microphysiological systems. Science Advances. 2025; 11(34). https://doi.org/10.1126/sciadv.adu2708
138. Lee J, Menon N, Lim CT. Dissecting Gut‐Microbial Community Interactions using a Gut Microbiome‐on‐a‐Chip. Advanced Science. 2024; 11(20). https://doi.org/10.1002/advs.202302113
139. Vatine GD, Barrile R, Workman MJ, Chen Z, Van Eyk J, Svendsen CN et al. Human IPSC-Derived Blood-Brain Barrier chips enable disease modeling and personalized medicine applications. Cell Stem Cell, Volume 24, Issue 6p995-1005.e6, Jun 2019. https://doi.org/10.1016/j.stem.2019.05.011
140. Tan PK, Ostertag T, Rosenthal SB, Murphy K, Miner JN, Brenner DA et al. Role of hepatic stellate and liver sinusoidal endothelial cells in a human primary cell Three-Dimensional model of nonalcoholic steatohepatitis. The American Journal of Pathology, Volume 194, Issue 3p353-368, Mar 2024. https://doi.org/10.1016/j.ajpath.2023.12.005
141. Reis RSD, Sant S, Keeney H, Wagner MCE, Ayyavoo V. Modeling HIV-1 neuropathogenesis using three-dimensional human brain organoids (hBORGs) with HIV-1 infected microglia. Scientific Reports. 2020 ;10(1). https://doi.org/10.1038/s41598-020-72214-0
142. Cairns DM, Smiley BM, Smiley JA, Khorsandian Y, Kelly M, Itzhaki RF, Kaplan DL. Repetitive injury induces phenotypes associated with Alzheimer’s disease by reactivating HSV-1 in a human brain tissue model. Science Signaling. 2025; 18(868). https://doi.org/10.1126/scisignal.ado6430
143. Andrews MG, Mukhtar T, Eze UC, Simoneau CR, Ross J, Parikshak N, Wang S, Zhou L, Koontz M, Velmeshev D, et al. Tropism of SARS-CoV-2 for human cortical astrocytes. Proceedings of the National Academy of Sciences. 2022; 119(30). https://doi.org/10.1073/pnas.2122236119
144. Jacob F, Pather SR, Huang W-K, Song H, De La Torre JC, Ming G-L et al. Human pluripotent stem Cell-Derived neural cells and brain organoids reveal SARS-COV-2 neurotropism predominates in choroid plexus epithelium. Cell Stem Cell, Volume 27, Issue 6p937-950.e9, Dec 2020. https://doi.org/10.1016/j.stem.2020.09.016
145. Miller SJ, Dhodapkar RM, Sutova HE, Xue Y, Lee S, Logan R, Ran C, Bhatta S, Gomm A, Ju IG, et al. SARS-CoV-2 induces Alzheimer’s disease–related amyloid-β pathology in ex vivo human retinal explants and retinal organoids. Science Advances. 2025; 11(27). https://doi.org/10.1126/sciadv.ads5006
146. Rybak-Wolf A, Wyler E, Pentimalli TM, Legnini I, Martinez AO, Glažar P, Loewa A, Kim SJ, Kaufer BB, Woehler A, et al. Modelling viral encephalitis caused by herpes simplex virus 1 infection in cerebral organoids. Nature Microbiology. 2023; 8(7):1252–1266.
https://doi.org/10.1038/s41564-023-01405-y
147. Raja WK, Mungenast AE, Lin Y-T, Ko T, Abdurrob F, Seo J, Tsai L-H. Self-Organizing 3D Human Neural Tissue Derived from Induced Pluripotent Stem Cells Recapitulate Alzheimer’s Disease Phenotypes. PLoS ONE. 2016; 11(9):e0161969. https://doi.org/10.1371/journal.pone.0161969
148. Pavoni S, Jarray R, Nassor F, Guyot A-C, Cottin S, Rontard J, Mikol J, Mabondzo A, Deslys J-P, Yates F. Small-molecule induction of Aβ-42 peptide production in human cerebral organoids to model Alzheimer’s disease associated phenotypes. PLoS ONE. 2018; 13(12):e0209150. https://doi.org/10.1371/journal.pone.0209150
149. Haj E, Losev Y, KrishnaKumar VG, Pichinuk E, Engel H, Raveh a, Gazit E, Segal D. Integrating in vitro and in silico approaches to evaluate the “dual functionality” of palmatine chloride in inhibiting and disassembling TAU-derived VQIVYK peptide Fibrils. Biochimica Et Biophysica Acta (BBA) - General Subjects Volume 1862, Issue 7, Jul 2018, Pages 1565-1575. https://doi.org/10.1016/j.bbagen.2018.04.001
150. Hurtle B, Donnelly CJ, Zhang X, Thathiah A. Live-cell visualization of tau aggregation in human neurons. Communications Biology. 2024; 7(1). https://doi.org/10.1038/s42003-024-06840-z
151. Park J-C, Jang S-Y, Lee D, Lee J, Kang U, Chang H, Kim HJ, Han S-H, Seo J, Choi M, et al. A logical network-based drug-screening platform for Alzheimer’s disease representing pathological features of human brain organoids. Nature Communications. 2021 ; 12(1). https://doi.org/10.1038/s41467-020-20440-5