Genetically Engineered Mouse Models for Alzheimer Disease and Frontotemporal Dementia

Abstract

Neurodegenerative diseases, including Alzheimer disease, frontotemporal dementia, Parkinson disease, Huntington disease, and amyotrophic lateral sclerosis, are often casually linked to protein aggregation and inclusion. As the origins of those proteinopathies have been biochemically traced and genetically mapped, genetically engineered animal models carrying the specific mutations or variants are widely used for investigating the etiology of these diseases, as well as for testing potential therapeutics. This article focuses on the mouse models of Alzheimer disease and closely related frontotemporal dementia, particularly the ones that have provided most valuable knowledge, or are in a trajectory of doing so. More importantly, some of the major findings from these models are summarized, based on the recent single-cell transcriptomics, multiomics, and spatial transcriptomics studies. While no model is perfect, it is hoped that the new insights from these models and the practical use of these models will continue to help to establish a path forward.

Alzheimer disease (AD) is the leading age-related irreversible neurodegenerative disease, manifesting as the accumulation of β-amyloid (Aβ) plaques in the brain parenchyma and deposits on the blood vessel walls, excessive neurofibrillary tangles in the degenerating neurons, and the destruction of the neurovascular unit.1, 2, 3 Since the diagnosis of the first patient in the early 1900s by Dr. Alois Alzheimer, a few hypotheses for the cause of the disease have been proposed and extensively tested. The cholinergic hypothesis⁴ from the late 1970s was based on biochemical and histopathologic observation of neurotransmitter changes in AD patients, particularly the deficits in excitatory amino acid neurotransmission, and the beneficial effects of cholinomimetic drugs and cholinesterase inhibitors in delaying the cognitive deterioration in AD.⁵ While cholinesterase inhibitors such as donepezil, rivastigmine, and galantamine are still used, their effects are often mild and clinically nonsignificant.

The amyloid hypothesis⁶ is commonly recognized, as it is based on the fact that amyloid plaques and deposits are the result of the imbalance between the production and clearance of Aβ peptides, particularly the less soluble and more toxic Aβ-42 over the normal and shorter forms Aβ-38 and Aβ-40. This occurs in the early-onset familial AD (EOAD) caused by autosomal-dominant mutations (also known as ADRD).⁷ So far, >40 mutations in the human amyloid precursor protein gene (APP), and >200 mutations in the genes encoding the presenilins 1 and 2 (PSEN1 and PSEN2) subunits of γ-secretase complex, which catalyzes the APP to produce the Aβ peptides, have been confirmed to cause ADRD.⁸ Yet, more de novo mutations are still identified through new whole-genome sequencing.⁹ This hypothesis is also supported by the notion that the duplication of the APP gene in Down syndrome also causes the early onset of amyloidosis even in teens, and the apolipoprotein (apo)-E4 gene (APOE4),⁶ the most significant genetic risk factor for sporadic late-onset AD (LOAD), impairs Aβ clearance from the brain.¹⁰ The amyloid hypothesis also takes into consideration that Aβ oligomers induce human Tau hyperphosphorylation, which is known as a major driving force of neuron dysfunction and death, or the real “bullet”.¹¹ The over three decades of work around amyloid and Tau as the biomarkers for AD laid the foundation for the National Institute on Aging and Alzheimer's Association's new Amyloid, Tau, and Neurodegeneration (A/T/N) Classification,¹² for more accurate diagnosis of AD.

Nonetheless, the amyloid hypothesis is not perfect. In fact, it is increasingly questioned due to the failures of large-scale clinical trials of therapeutics that target the amyloid cascades, including the γ-secretase inhibitors and monoclonal antibodies against different Aβ aggragates.¹³ At the same time, piling evidence from both patients and animal models has clearly indicated that the disease is more complex at the genetic, molecular, and cellular levels.¹⁴ Genetic studies targeting the more complex LOAD populations, such as the Genome-Wide Association Studies,¹⁵ whole-exome genotyping¹⁶ and whole-genome sequencing projects,¹⁷ predicted the association of AD risk with hundreds of genetic variants and their related gene loci. These studies not only confirmed that APOE loci, particularly the APOE ε4 allele, are strongly associated with increased AD risk but also identified new variants or mutations that significantly increase the risk, such as the TREM2 R47H variant at rs75932628.¹⁸ Rare protective variants that lower the risk have also been discovered, such as the APOE3 Christchurch mutation (R136S), which even offers resistance in an autosomal-dominant Alzheimer disease mutation carrier.¹⁹ These genetic data largely support the existing alternative hypotheses, such as the vascular hypothesis^2,20 and the neuroimmune hypothesis^21,22 and its related infectious origin of neurodegeneration.^23,24

It was somehow surprising that, despite the relevance of tauopathy in AD, no variants near the MAPT loci were identified through those large-scale genetic studies, which encoded the microtubule-associated protein Tau. In fact, tauopathies are also prominent in frontotemporal dementia (FTD), corticobasal degeneration, progressive supranuclear palsy, and chronic traumatic encephalopathy,²⁵ and MAPT mutations near exons 9 to 13 that affect its mRNA splicing and expression are found in approximately 30% of familial FTD cases, including Pick disease.²⁶ FTD is the second most common type of dementia, and has significant pathologic and genetic overlaps with AD. Therefore, MAPT/Tau models will be discussed in Transgenic Models of Frontotemporal Dementia, with other familial genes including C9orf72, GRN/progranulin, and TARDBP/transactive response DNA-binding protein (TDP)-43.²⁷

Animal models, particularly genetically modified murine models, have been a cornerstone in modern biomedical research. It began with the efforts of introducing viral DNAs in very early stages in mouse embryos and developing the oncomice.²⁸ Meanwhile, the knock-out/knock-in system of manipulating endogenous genes became possible with the establishment of homologous recombination in mouse embryonic stem cells.²⁹ Therefore, mouse models of AD and FTD also started with transgenic models carrying the mutated human transgenes, and later evolved into knock-in models. For example, the first transgenic AD model, reported in 1995, was transgenic mice overexpressing mutant human amyloid precursor protein V717F (PDAPP mice), in which the human APP with Indiana mutation V717F was controlled by a platelet-derived growth factor (PDGF)-β promoter,³⁰ followed by the ApoE targeted replacement model in 1999,³¹ and the first APP knock-in model in 2014.³² The development of these models targeting either the drivers of the proteinopathies or associated-risk genes was propelled later by large-scale consortium projects, such as the Knockout Mouse Project²⁹ and the MODEL-AD Consortium,³³ and further facilitated by recent technological development, particularly the clustered regularly interspaced short palindromic repeats and CRISPR-associated protein 9 (CRISPR/Cas9)–mediated gene-editing system.³⁴ While it is impossible to cover all the models over the course of half-century in one review, there is focus on the commonly used models in AD and FTD research, and more importantly discuss the recent advances from these models with single-cell transcriptomics, multiomics, and spatial transcriptomics related approaches (Figure 1).

Figure 1.

Diagram showing the development of key animal models for AD in the past 50 years. Nonhuman primates (blue), including chimpanzee, canine, and guinea pig models, were examined between the 1970s and 1990s, before transgenic mouse models were available (blue). The generation of APP models, such as Tg2576 (1996), APP23 (1997), APP/PS1 (2001), 3xTg (2003), APPSwDI (2004), 5xFAD (2006), APP^NL-G-F (2017), and APP^SAA (2022), substantially validated the amyloid hypothesis, and demonstrated the neuroimmune dysregulation and neurovascular impairment associated with amyloid pathologies (green). The generation of LOAD models, including TR-APOEs (1997), APOE KIs (2004, 2019), and other LOAD risk genes such as TREM2 models (2015, 2017), further defined how LOAD risk genes can influence neuroimmune and neurovascular functions in the context of AD pathogenesis (red). The rise of human induced pluripotent stem cell (iPSC)-derived cells from AD patients (2011) and three-dimensional models, such as brain organoids and microphysiological systems (2013), complements the animal models in different ways (orange). AD, Alzheimer disease; APOE, apolipoprotein E; APP, amyloid precursor protein; LOAD, late-onset familial AD; KI, knock-in; KO, knock-out; NL-G-F, Swedish, Arctic, and Beyreuther/Iberian; PS, presenilin; SAA, Swedish, Arctic, Austrian; SwDI, Swedish, Dutch, Iowa; TR, targeted replacement; TREM, triggering receptor expressed on myeloid cells protein.

Transgenic Models for Alzheimer Disease

Classic transgenic AD animal models carrying different forms of EOAD mutations found in familial cases, such as APP and PSEN1, concordantly have demonstrated that amyloidosis can be inherited through these genetic mutations (Table 135, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57). While neuron loss or Tau pathologies are usually not seen in these models, or at least are not prominent, amyloid pathologies are tightly linked to gliosis, neuroinflammation, neurovascular dysfunction, and cognitive changes associated with memory. In addition to amyloid plaque formation in the brain parenchymal, cerebral amyloid angiopathy, and neurovascular dysfunction are also recapitulated in some of the APP models.⁵⁸

Table 1.

Summary of Commonly Used EOAD and LOAD Mouse Models

Models	Description	AD-like pathology	Neuron loss	Cognitive impairment and behavioral changes	References
Transgenic EOAD
Tg2576	Transgenic mouse expressing human APP gene with Swedish mutations	Plaques at 11 months	Limited	Dendritic spine loss at 4.5 months, LTP decline at 5 months; impaired spatial learning and memory at 6 months, progressively worse at 12 months	Hsiao et al, 199635; Frautschy et al, 199836; Lanz et al, 200337
APP23	Expresses human APP with Swedish double mutations	Plaques at 6 months	14%–28% loss in CA1 at 14–18 months	No synaptic loss or LTP changes; impaired spatial learning and memory at 3 months, worse in aging	Jankowsky et al, 200138; Jankowsky and Zheng, 201739; Onos et al, 201940
APP/PS1	Expresses human APP and PS1 transgenes with familial AD mutations	Plaques at 6 weeks in cortex and 3 months in hippocampus	Modest loss at 17 months	Dendritic spine loss at 4 weeks, LTP impairment at 8 months; deficits in spatial learning and memory at 7–8 months	Radde et al, 200641; Gengler et al, 201042
5xFAD	Expresses five familial AD mutations (three in APP, two in PS1)	Plaques at 2 months	Cortical layer V and subiculum at 12 months	Synaptic loss at 4 months, LTP decline at 4 months; impaired working memory and memory retrieval at 4–5 months	Oakley et al, 200643; Oblak et al, 202144
APPSwDI	Expresses human APP Swedish, Dutch and Iowa mutations	Vascular deposits at 3 months	N.A.	Synaptic loss or LTP has not been reported; impaired learning and memory at 3 months	Park et al, 201445
3xTg-AD	Triple transgenic model expressing mutant forms of APP, PS1, and Tau, exhibiting both amyloid and Tau pathology.	Plaques at 6 months; tauopathy at 12 months	N.A.	LTP decline at 6 months; cognitive impairment at 4 months	Oddo et al, 200346
TauPS2APP	Cross of TauP301L and PS2APP transgenic lines, resulting in both amyloid and Tau pathologies	Plaques and tauopathy at 4 months	Absent	Synaptic loss or LTP has not been reported; cognitive impairment at 4 months, but no age-dependent progression	Grueninger et al, 201047; Lee et al, 202148
KOs and KIs EOAD and LOAD
APPNL−G-F KI	Expresses humanized APP with NL-G-F mutations	Plaques at 2 months in homozygotes, 4 months in hemizygotes	Absent	Synaptic loss around plaques; memory impairment in homozygous mice at 6 months	Saito et al, 201432; Sasaguri et al, 201749; Watamura et al, 202250
APPSAA KI	Expresses a chimeric mouse/human APP with Swedish, Arctic and Austrian mutations	Plaques at 4 months in homozygotes, 16 months in hemizygotes	N.A.	Synaptic loss or LTP has not been reported; behavioral assessment showed locomotor hyperactivity and habituation deficits, particularly in females	Xia et al, 202251
APOE targeted replacement	Expresses human APOE2, -E3, or -E4 isoforms in the mouse Apoe allele	Absent	Interneurons decrease in TR-APOE4 mice	TR-APOE4 mice exhibit decreased GABAergic interneurons at 6 months and loss of GABAergic synapses at 16 months, and cerebrovascular impairment with blood–brain barrier breakdown at 8 months	Piedrahita et al, 199252
Polygenic LOAD
TR-APOE; 5xFAD (EFAD)	APOE KI models crossed with 5xFAD	Accelerated amyloid pathology in E4FAD mice	As in 5xFAD	E4FAD mice showed the greatest age-dependent behavioral deficits relative to E3FAD and E2FAD animals	Liu et al, 202153
Trem2 KO; 5xFAD	Trem2 KO crossed with 5xFAD	Plaques at 4 months	Cortical layer V neuron loss at 8 months	N.A.	Wang et al, 201554; Ulland et al, 201755
Trem2 KO; APP/PS1	Trem2 KO crossed with APP/PS1	Plaques at 2 months, but less in Trem2 KO	N.A.	N.A.	Jay et al, 201556; Jay et al, 201757
APOE4 R136S KI; PS19	APOE4 mice carrying the R136S Christchurch mutation, crossed with PS19 Tau	Tauopathy at 10 months, but reduced in APOE4 R136S KI	N.A.	N.A.	Nelson et al, 202319

AD, Alzheimer disease; APOE, apolipoprotein E; APP, amyloid precursor protein; EOAD, early-onset familial AD; LOAD, late-onset familial AD; KI, knock-in; KO, knock-out; LTP, long-term potentiation; N.A., not applicable; NL-G-F, Swedish, Arctic, and Beyreuther/Iberian; PS, presenilin; SAA, Swedish, Arctic, Austrian; SwDI, Swedish, Dutch, Iowa; TR, targeted replacement; TREM, triggering receptor expressed on myeloid cells protein.

Tg2576

The Tg2576 mouse model was one of the earliest models and is most commonly used to study amyloid plaque formation and its effects on neuropathology and cognitive deficits. It utilizes the hamster prion protein promoter to drive overexpression of a mutant form of APP with the double Swedish mutations K670N and M671L.³⁵ The mice develop early neuron phenotypes, including the loss of dendritic spine in the hippocampal CA1 region at 4.5 months of age,³⁷ and decreased synaptic functions in the dentate gyrus based on measurements of long-term potentiation (LTP) at 5 months. Microglia activation and gliosis become prominent at around 10 months,³⁶ followed by substantial parenchymal Aβ plaque appearance at 11 months.³⁷ The impairments in spatial learning and working memory were observed in this model at as early as 6 months, and often became progressively worsened at around 12 months. The finding of Aβ∗56 species from this model has been controversial. The lack of a Tau pathology in this model may present a significant limitation,³⁵ but Tg2576 is still a highly reliable and well-characterized model of AD.

APPswe/PSEN1dE9

The generation of APP models made it possible to investigate APP processing and amyloid deposition in vivo, as well as the interaction between APP mutation and γ-secretase. The APPswe/PSEN1dE9 mice are among the first models to harbor mutations in both APP and PSEN1. It is achieved by separately co-injecting two vectors containing APP with K670N and M671L mutations and PSEN1 lacking exon 9 (dE9).^38,59 The model demonstrated accelerated amyloidosis beginning at 6 months,³⁸ which is accompanied by astrogliosis. This model has been bred to different genetic backgrounds, including a hybrid C57BL/6 × C3H background and even multiple wild-derived strains such as CAST, WSB, and PWK,⁴⁰ and provided strong experimental evidence that genetic background or heterogeneity can have a huge impact on the onset and progression of amyloid pathology.³⁹ This model lacks tangles/Tau pathology and falls short in expressing neuron loss in the cortical and hippocampal regions up until 8 months in the PWK strain. Even in the CAST and WSB strains, whether the neuron differences lead to neurodegeneration remains undetermined.⁴⁰

APP/PS1

The APP/PS1 (APPPS1-21) mice also harbor mutations in both APP and PSEN1, but use a more neuron-specific Thy1 promoter to drive the expression of APP Swedish mutations and the PSEN1 L166P mutation, resulting in a very robust and early onset of amyloidosis and cognitive dysfunction. This model has been extensively used to study the interaction between Aβ and neuron loss. However, the absence of Tau accumulation and synaptic dysfunction are notable drawbacks.⁶⁰ Aβ deposition can be detected in the cortex at 6 weeks of age, and later in the hippocampus at around 3 months.⁴¹ Phosphorylated Tau can be observed in neurites near plaques, but not in mature neurofibrillary tangle forms.⁴¹ The mice develop dendritic spine loss around plaques, but LTP impairment in the hippocampus occurs much later, at 8 months of age, and neuron loss is very minimal.⁴² The model has shown early memory impairment by 6 months, which provides value in studying the development of therapeutic approaches to amyloid and microglial dysfunction that are related to memory deficits.⁶¹

5xFAD

5xFAD mice are another widely used transgenic model, harboring five EOAD mutations, including three APP mutations (Swedish K670N/M671L, Florida I716V, and London V717I), and double PSEN1 mutations (M146L and L286V). The current Tg6799 strain was selected from the three founder lines with the highest levels of transgene expression and rapid development of amyloid pathology.⁴³ The model was initially established on a mixed C57BL/6 × SJL background,⁴³ and is available on a congenital C57BL/6 background.⁴⁴ The neuropathology starts with intraneuronal accumulation of amyloid, followed by extracellular deposition before 2 months.⁴³ The highest levels of amyloidosis are often seen in the subiculum and cortical layer V,⁴³ and the levels in thalamus catch up after 4 months.⁴⁴ The female mice in this model develop more severe amyloid pathologies compared to aged-matched males.⁴⁴ This model exhibits earlier and pronounced amyloid deposition closely mimicking familial forms of AD. Nonetheless, similar to the Tg2576, the 5xFAD model does not incorporate Tau pathology. The model, due to five familial AD mutations, leads to an artificial acceleration of pathology that may not accurately represent the slow, progressive nature of the disease in humans. This rapid accumulation can distort the neuroinflammatory response and may overlook the slower disease mechanisms present in human AD.⁴³

Tg-SwDI

The APP Flemish A692G, Arctic E693G, Dutch E693Q, Italian E693K, and Iowa D694N mutations are closely positioned between the residues 21 to 23 of Aβ and linked to familial forms of cerebral amyloid angiopathy.⁶² The Tg-SwDI (Swedish, Dutch, and Iowa) is the first model developed to produce mutant Aβ carrying these mutations. It develops fibrillar amyloid deposits around the cerebral microvasculature at between 3 and 6 months, and profound cerebrovascular dysfunction, including cerebral blood flow reduction, endothelial dysfunction, and mural cell loss.^45,63 Therefore, it is uniquely positioned as a model for vascular cognitive impairment and dementia.

Triple Transgenic Models

Due to the lack of tauopathy in nearly every APP model, triple transgenic models with mutations in APP, PSEN1, or PSEN2, and Tau/MAPT were generated. The 3xTg-AD model uniquely incorporates the APP Swedish, PSEN1 M146V, and MAPT P301L mutations, and exhibits progressive Aβ pathologies beginning at 3 months, and tauopathy at 12 months.⁴⁶ The 3xTg-AD model is used to study the interplay of amyloid and Tau pathologies. The TauPS2APP triple transgenic model was established by crossing the PS2APP model (line B6.152H) with the Tau model (line B6.TauP301L),⁴⁸ in which tauopathy and amyloid plaques both appear at as early as 4 months.⁴⁷ The dual pathologies in these models provide a platform to evaluate drugs targeting both types of neurodegenerative conditions.

Knock-In and Knock-Out Models for Alzheimer Disease

APP^NL-G-F, APP^NL-F, and APP^G-F KI Models

The RIKEN Center for Brain Science (Tokyo, Japan) first developed these new APP models by introducing the Swedish (NL), Arctic (G) and Beyreuther/Iberian (F) mutations through knock-in strategies.⁴⁹ By mitigating the overexpression issue, these models overall demonstrated that endogenous expression levels of APP mutants are sufficient to drive the typical Aβ pathology, neuroinflammation, and cognitive impairment in mice. The APP^NL-G-F model has been adopted in many laboratories and is well-characterized with amyloid pathology and gliosis at 6 months, and cognitive impairment at 18 months,³² while the APP^G-F model mitigates the issue of the Swedish mutation of inhibiting β-site APP cleaving enzyme (BACE)-1 activities, and potentially will be useful for testing BACE1 inhibitors.⁵⁰

APP^SAA KI Model

The APP^SAA knock-in (KI) model was recently developed by Denali Therapeutics (South San Francisco, CA) and is available through The Jackson Laboratory (Bar Harbor, ME).⁵¹ It is engineered to express the APP Swedish KM670/671NL, Arctic E693G, and Austrian T714I mutations (SAA), and homozygous mice develop amyloid pathology at 4 months of age, while heterozygous mice can take up to 16 months. In addition to the age-dependent formation of amyloid plaques, cerebral amyloid angiopathy, and microglia activation toward the disease-associated microglia subtype have also been reported. In addition, the model manifests behavioral disinhibition and altered lipid metabolism.⁵¹

APOE Models

Due to the difference between the murine Apoe gene and human APOE with three isoforms (ε2, ε3 and ε4), both murine Apoe knock-out mice⁵² and targeted replacement of murine Apoe with different human APOE isoforms were generated.³¹ While the initial application was in atherosclerosis studies, they were subsequently employed to understand APOE isoform–specific effects in AD models. For example, Holtzman's group^64,65 showed that the APOE ε4 isoform significantly exacerbated neuropathology and degeneration in APP/PS1 or P301S Tau transgenic mice. As a vascular and atherosclerosis risk factor, APOE ε4 is known to cause cerebrovascular dysfunction and blood–brain barrier impairment, and to negatively impact the clearance of amyloid,^10,66 which can also be recapitulated in targeted-replacement APOE4 mice.^67,68

Other APOE models were recently developed, including the APOE (ε2, ε3, or ε4) knock-in mice from The Jackson Laboratory,⁶⁹ and the floxed APOE (ε2, ε3, or ε4) KI models from the Cure Alzheimer's Fund.⁷⁰ These models are now widely utilized to understand the APOE isoform–specific impact (ε2 versus ε3, ε3 versus ε4) on amyloid and Tau pathologies and neuroinflammation. In addition to the common variants, rare protective APOE3 mutations, such as the Jacksonville V236E and Christchurch R136S variants, were recently identified. The Jacksonville V236E is associated with a much-reduced LOAD risk,^53,71 the Christchurch R136S was found in a cognitively normal PSEN1 mutation carrier,⁷² and a mouse model carrying the R136S mutation via CRISPR/Cas9-mediated knock-in editing in the APOE4 allele showed strong protection against Tau pathology, neuroinflammation, and degeneration.¹⁹ Further studies are still required to validate the biological function of these new APOE variants by comparison between human cases and transgenic models. Therefore, these APOE mouse models are truly valuable for understanding the pathogenesis in LOAD associated with APOE variants.

Other LOAD Models

In addition to APOE, large-scale Genome-Wide Association Studies¹⁵ have predicted an additional 74 loci with significant association with LOAD.¹⁵ While most of these genetic variants are common variants and do not directly alter protein sequence or function, there are rare AD variants predicted to impact gene regulation and expression and protein functions. While classic knock-out models targeting these genes offer important insights into protein functions, knock-in models are still needed to tease out the specific roles of individual variants in vivo. In addition to individual research groups, the MODEL-AD consortium has provided a list of these knock-out and knock-in models,³³ and these models are usually crossed with an EOAD model, such as 5xFAD or APP/PS1, to investigate the variant function in the context of AD pathologies.

For example, the TREM2 rs75932628 variant encodes a R47H mutation, and is associated with a high risk for AD (odds ratio, 4.59).¹⁸ While TREM2 R47H knock-in mice are still being studied using the 5xFAD model,⁷³ a study in Trem2 knock-out mice showed disease-associated microglia (DAM) activation is at least partly dependent on the triggering receptor expressed on myeloid cells protein (Trem)-2, and that Trem2 deficiency in general mitigated the neuroinflammation related to amyloid pathologies in 8-month–old 5xFAD mice,^54,55 or tauopathy in the PS19 Tau model⁷⁴ or the TauPS2APP triple transgenic model.⁴⁸ In APP/PS1 mice, Trem2 deficiency eliminated a subset of inflammatory macrophages, and reduced the initial plaque burden at 2 to 4 months of age^56,57; however, this finding may be attributable to the remaining neomycin-resistance gene driven by the human ubiquitin C promoter, which also enhanced the downstream Treml1 gene expression.⁷⁵ Nonetheless, the administration of Trem2-activating antibodies showed a promising beneficial effect in 5XFAD mice,^76,77 validating TREM2 as a therapeutic target in AD.⁷⁸

In addition to TREM2, there are other new knock-in models available through the MODEL-AD consortium and The Jackson Laboratory, including ABCA7 A1527G, PLCG2 M28L, and SORL1 A528L, as well as the humanized Aβ knock-in model (hAβ-KI)⁷⁹ and the APP^SFL knock-in model carrying the Swedish, Florida I716V, and London V717I mutations. Some of the models are bred together to generate new LOAD models: APOE4 and Trem2 R47H knock-ins as LOAD1; hAβ, APOE4, and Trem2 R47H knock-ins as LOAD2; and hAβ, APOE4, and MAPT (H1.0) knock-ins as LOAD3. These new models offer unprecedented opportunities to study AD pathogenesis in complex genetic backgrounds, particularly heterogeneity in LOAD conditions.

Transgenic Models of Frontotemporal Dementia

FTD (also known as Pick disease), the second leading cause of dementia, is a spectrum of neurodegenerative conditions including behavioral variant FTD, primary progressive aphasia, corticobasal syndrome, progressive supranuclear palsy, and FTD with amyotrophic lateral sclerosis.⁸⁰ These conditions are caused by pathologic proteinopathies such as tauopathy and aberrant protein accumulation of TDP-43 or fused in sarcoma (FUS) in neurons.⁸¹ Autosomal-dominant mutations, including P301L and P301S, in the Tau/MAPT gene were first identified in familial cases.⁸² Mutations in 10 other genes have also been determined and contribute to Tau, TDP-43, and FUS proteinopathies, including C9orf72, CCNF, CHCHD10, CHMP2B, FUS, OPTN, progranulin/GRN, SQSTM1, TBK1, TDP-43/TARDBP, TIA1, and VCP.⁸³ As mutations in Tau/MAPT, progranulin/GRN, and C9orf72 are most frequently found in FTD, these models are included in this section (Table 284, 85, 86, 87, 88, 89, 90).

Table 2.

Summary of Commonly Used FTD Mouse Models

Models	Description	AD-like pathology	Neuron loss	Cognitive impairment and behavioral changes	References
Tau/MAPT models
Tg4510	Transgenic model with tetracycline-inducible expression of human Tau with P301L mutation	Tauopathy at 2.5 months	∼60% CA1 neuron loss at 5.5 months	Impaired spatial learning and memory at 2.5 months; LTP impairment at 4.5 months; motor impairment starting at 6 months; ∼30% dendritic spine loss at 8 months; forebrain atrophy at 10 months	Ramsden et al, 200584
Tau (P301L)	Transgenic model carrying human Tau with P301L mutation	Tauopathy at 8 months	N.A.	Spine density is higher in young mice; cognitive impairment begins at 5 months; LTP deficit at 6 months	Lee et al, 202148; Buchholz and Zempel, 202485
PS19-P301S	Transgenic model carrying human Tau with P301S mutation	Tauopathy at 6 months	Hippocampal neuron loss at 9 months	Synaptic loss at 3 mo, LTP decline at 6 months; impairment in spatial learning and memory at 6 months; motor deficit progression to paralysis at 7–10 months	Schindowski et al, 200686
Progranulin/GRN models
Grn KO	Progranulin/GRN knock-out model	TDP-43 pathology in knock-out mice at 12–18 months	Absent	Social behavior deficits at 1–2 months, depression and disinhibition behavior at 3–4 months, cognitive impairment at 12 months	Filiano et al, 201387
GrnR493X KI	Knock-in mice carrying R493X mutation	TDP-43 pathology in homozygotes mice at 12 months	Neuron loss in motor cortex in end-stage mice	Cognitive impairment in learning, memory, social, and emotion in homozygote mice at 11 months	Nguyen et al, 201888
C9orf72 models
BAC (G4C2)450	BAC transgenic carrying a partial human C9orf72 gene with a (G4C2)450 repeat region	pTDP-43 increase in 22 months	No motor neuron loss	Increased anxiety and cognitive impairment starting at 4–6 months, worse at 12–18 months	Jiang et al, 201689
BAC (G4C2)500	BAC transgenic carrying full human C9orf72 gene with a (G4C2)500 repeat region	TDP-43 inclusions in symptomatic mice at 18 months	Cortical layer V neuron loss at 8 months	Motor impairment at 4 months, paralysis and mortality between 5 and 8 months, survived mice showed anxiety-like behavior at 1 year	Jiang et al, 201689; Liu et al, 201690

AD, Alzheimer disease; BAC, bacterial artificial chromosome approach; LTP, long-term potentiation; N.A., not applicable; TDP, transactive response DNA-binding protein.

Tau/MAPT Models

The human MAPT pre-mRNA can be alternatively spliced around exons 2, 3, and 10, generating six isoforms with different combinations of repeats in the amino-terminal half of the protein (0N, 1N, or 2N) or the carboxyl-terminal half (3R or 4R).⁸⁵ The 3R and 4R isoforms are nearly at equilibrium in the normal brain (approximately 50:50 ratio), but increased 4R/3R ratios are found in AD and FTD. Therefore, the majority of Tau transgenic models are based on 4R isoforms and contain exon 10, in addition to the FTD mutations, such as G272V, N297K, P301L, P301S, V337M, and R406W. These models also differentially express various amino-terminal repeats under different promoters.⁹¹ For example, at least six different monogenic models carry the P301L mutation: The Tg4510 model expresses the 4R/0N isoform under the control of a tetracycline-inducible calcium/calmodulin-dependent protein kinase II (CaMKII) promoter⁸⁴; the Tau (P301L) line carries a 4R/2N isoform under the control of the neuron-specific murine Thy1 promoter.⁴⁸ Nonetheless, all of these transgenic lines show an age-dependent increase in Tau phosphorylation and tauopathies in various brain regions, as well as impairment in sensorimotor functions and cognition.⁹¹ The more severe ones, such as the JNPL3-P301L and PS19-P301S models, develop spinal cord degeneration and motor deficits, and even paraparesis. On the other hand, the THY-Tau22 model carrying both G272V and P301S develops only learning and memory deficits, but not motor impairment.⁸⁶ Mentioned Tau/MAPT animal models replicate Tau pathology, which is linked to neurodegeneration and cognitive decline. The models are helpful in clarifying the role of Tau independent of amyloid, which supports both AD and FTD research.

Progranulin/GRN Models

Approximately 70 FTD mutations have been identified within the human GRN gene, and most of them contribute to reduced progranulin levels or loss of function.⁹² The R493X mutation results in a premature stop codon, producing mRNAs that are at risk for nonsense-mediated decay and progranulin haploinsufficiency in FTD,⁹³ which can be largely recapitulated in the heterozygous Grn^+/– models at the behavior level, but not at the neuropathologic level,⁸⁷ as the mice do not develop TDP-43 pathology or significant neuron loss. In different progranulin-deficient mice, cytosolic phosphorylated TDP-43 accumulation was observed in the thalamic neurons of 18-month–old knock-out mice.⁹⁴ Similar outcomes were observed in the Grn^R493X knock-in model, and homozygous Grn^R493X mice showed increased levels of total and phosphorylated TDP-43 in the cytoplasm of thalamic neurons.⁸⁸

C9orf72 Models

Hexanucleotide repeat expansion of GGGGCC (in the human chromosome 9 open reading frame 72 gene; C9orf72) is frequently observed in both FTD and familial amyotrophic lateral sclerosis. So far two C9orf72 transgenic models successfully recapitulated the motor deficits and neurodegeneration related to amyotrophic lateral sclerosis/FTD,⁹⁵ and both utilized the bacterial artificial chromosome approach to express human C9orf72 with 450⁹⁴ or 500 G₄C₂ repeats.⁹⁰ Both models showed typical pathologies related to C9orf72 hexanucleotide repeat expansion, including the formation of sense and antisense RNA foci in neurons and poly(GA) expression from the hexanucleotide repeat expansion and cytoplasmic aggregates in young animals,^89,90 and increased pTDP-43 levels in 22-month–old mice⁸⁹ or TDP-43 inclusions in symptomatic old mice.⁹⁰

Single-Cell Transcriptomics, Multiomics, and Spatial Transcriptomics of AD and FTD Models

Utilizing the cutting-edge genomic technologies like single-cell and single-nucleus RNA sequencing (scRNA/snRNA-seq) and spatial transcriptomics, researchers have made significant discoveries to decipher the intricate molecular mechanisms that underpin AD and FTD. scRNA-seq has revealed that in the APP23 mouse model, there is significant neuronal diversity and a variety of stress responses at the single-cell level,⁹⁶ which could explain the differential susceptibility of neurons to Aβ toxicity. It has highlighted not just the expected neuron dysfunction in AD mouse models, but also offers a nonbiased method for characterizing microglial subtypes and states as they transition from normal conditions to disease.

A recently identified subtype of microglia, DAM, in 5xFAD mice plays a protective role in AD.^48,97, 98, 99, 100, 101, 102, 103 DAMs are strategically located near sites of AD pathology. A detailed examination of gene expression in DAM throughout the progression of the disease showed an increase in lipid metabolism and phagocytic pathways, which are essential for plaque clearance. Further immunohistochemistry and single-nucleus RNA-seq analyses confirmed the presence of DAMs in both mouse and human brain tissues, suggesting their broader relevance to neurodegenerative diseases.^104,105 Crucially, the specific gene-expression profile of DAM includes several genes that are known risk factors and are believed to help mitigate the disease, highlighting their potential therapeutic importance in AD. Additionally, it has been observed that microglia pass through distinct stages as they transition to DAMs, influenced by TREM2, which modifies their activation and gene expression, reflecting the complexity and dynamism of microglial responses in AD.¹⁰¹ The microglia activation phenomenon has also been confirmed in FTD models with transcriptomics studies; for example, microglia cells from the rTg4510 model showed NF-κB signaling activation.¹⁰⁶

Similarly, scRNA-seq studies have identified specific subsets of astrocytes that change expression patterns in response to neurodegeneration, suggesting their potential role in either exacerbating or mitigating disease progression. A recent study utilized snRNA-seq to profile astrocytes from 34 wild-type and 5xFAD mice at various ages, and identified a population of disease-associated astrocytes that increased with disease progression.¹⁰⁷ These disease-associated astrocytes were also found in aging human brains and were often located adjacent to amyloid plaques in regions like the hippocampus and subiculum, where AD manifestations are pronounced. Initially, disease-associated astrocytes may participate in gliosis to help contain damage from accumulating misfolded proteins, but may become destructive as the disease progresses due to an inflammatory and neurotoxic gene-expression profile. Similarly, another study identified a disease-associated subpopulation of oligodendrocytes (DAOs) during the progression of AD-like pathology in male APP^NL-G-F and male 5xFAD AD mouse models, as well as in postmortem human AD brains.¹⁰⁸ They discovered that aberrant Erk1/2 signaling is linked to the activation of DAOs in the App^NL-G-F mouse brains. Importantly, inhibiting Erk1/2 signaling in these DAOs was found to restore axonal myelination, reduce Aβ-associated pathologies, and improve cognitive function in the App^NL-G-F AD mouse model.

Spatial transcriptomics technology goes beyond conventional RNA-seq by maintaining the spatial context of the tissue being studied, which is particularly beneficial in brain research. This method allows researchers to see not only which genes are up- or down-regulated in disease but also where these changes are happening within the brain. This is crucial for diseases like AD and FTD, where pathologic features such as plaques and tangles are localized in specific brain regions. In a well-characterized App knock-in mouse model,³² researchers observed the progressive development of a multicellular gene-expression network involving 57 plaque-induced genes around amyloid plaques.¹⁰⁹ These plaque-induced genes facilitate intercellular communication primarily among astrocytes and microglia, extending to other cell types. This interaction triggers simultaneous changes in the classic complement system and the endosomal/lysosomal pathways, which are pathways independently associated with AD pathogenesis. Moreover, the study outlined a dynamic response from oligodendrocytes, predominantly related to myelin-associated genes, that varied due to the gradual buildup of amyloid. Contrasting with the relatively uniform plaque-induced gene response, this oligodendrocyte gene response exhibited variation across different brain regions. In situ sequencing of over 200 human plaque-induced genes and oligodendrocyte genes from the frontal cortical lobe provided partial validation of these findings.¹⁰⁹

As these technologies advance, their integration with other methodologies like proteomics and metabolomics could provide even deeper insights into the cellular and molecular basis of neurodegenerative diseases. Future research using scRNA-seq and spatial transcriptomics will likely include longitudinal studies, tracking how gene expression in individual cells and spatial domains changes over time as disease progresses. This could lead to a more dynamic understanding of AD and FTD, opening up possibilities for timely and effective therapeutic interventions. Moreover, research utilizing scRNA-seq and spatial transcriptomics in mouse models of FTD remains limited. Most existing studies have come from human sample or knock-out models.

Limitations, Alternative Models, and Future Directions

Transgenic mouse models are one of the most crucial research tools today. However, they also have intrinsic limitations, mainly due to their genetic modifications, which are often based on targeting constructs with artificial promoters and transgenes. Replacing the mouse genes with human forms or swapping in the human disease variants or mutations may not always comply with the murine gene regulation and post-translational modification systems. The adoption of the new knock-in models with human mutants or variants largely circumvents these limitations. However, there are still big species-related differences between mice and humans in terms of lifespan and sexes, which are very difficult to convert. The anatomic and functional divergence of the nervous and immune systems between humans and mice can be a confounding factor when translating the outcomes from murine studies to clinical applications.³⁹

The AD and FTD models were largely selected solely based on their capacity to mimic the pathologic hallmarks, such as amyloid plaque formation in most of the AD models and tauopathy in others. While most of the animal models perform quite well in recapitulating the diseases caused by autosomal-dominant mutations, they often fall short in modeling the sporadic conditions in AD and FTD, which are often closely linked with other environmental factors. This may potentially be addressed with the new LOAD models from the MODEL-AD consortium. However, fully validating these models could still take years. Cognitive assessments in these models have been centered on learning and memory impairment in AD, and motor functions in FTD as well. However, other cognitive domains, particularly the neuropsychiatric symptoms, have not been paid enough attention. Considering these limitations, nonmurine models have been increasingly considered for their potential to provide complementary insights.

Nonhuman primates share more genetic and physiological similarities with humans, including the natural development of Aβ plaques as they age.¹¹⁰ While these models have been very useful for testing diagnostic and therapeutic agents, ethics concerns, high cost, long lifespan, and logistical challenges often restrict their use in dementia research. Canines, particularly older dogs, naturally exhibit amyloid-β accumulation and cognitive decline.¹¹¹ Aged guinea pigs, especially the Dunkin-Hartley strain, display AD-like pathologies such as Aβ plaques, tangles, and neuroinflammation, and their transcriptomic signatures are also similar to those in AD patients.^112,113 Organoids and microphysiological systems, particularly derived from human induced pluripotent stem cells (iPSCs), are also widely used as in dementia research. It is now possible to generate iPSC-derived organoids to mimic the architecture and physiology of almost every human organ,¹¹⁴ and to understand the impact of human AD and FTD mutations or variants on specific organ functions. Crosstalk between different brain cell types derived isogenic iPSC lines carrying human mutations can also be studied in advanced three-dimensional culture or microphysiological systems.^115,116 However, these models still face challenges in scalability, reproducibility, and relevance to aging. Mouse–human chimeric models are established by introducing human iPSC-derived cells¹¹⁷ or organoids into the mouse models,¹¹⁸ providing novel insights into the human cell responses in physiological and pathologic in vivo environments.

In summary, transgenic mouse models are versatile tools for modeling human diseases, and have provided incredible opportunities in dementia research in the past 3 decades. The knowledge gained from these transgenic, knock-out, knock-in, and humanized models targeting EOAD, LOAD, FTD, and related dementia is irreplaceable, and has laid the foundation for the recent therapeutic development targeting proteinopathies, or related genes and pathways. With the arrival of a new generation of models, the challenges of AD and FTD may be overcome in the near future.

Disclosure Statement

None declared.