Integrating neuroinflammation biomarkers into the ATN(X) framework: Advances in Alzheimer's pathogenesis, diagnosis, and insights from non‐human primate models

Abstract

The amyloid/tau/neurodegeneration (ATN) biomarker framework has greatly progressed the diagnosis and staging of Alzheimer's disease (AD). However, recent research highlights neuroinflammation as an equally critical factor in AD pathology across humans, rodents, and non‐human primates (NHPs). This review evaluates the combined use of ATN and neuroinflammatory biomarkers—such as glial fibrillary acidic protein (GFAP) (astrocytic marker) and triggering receptor expressed on myeloid cells 2 (TREM2)/ ionized calcium‐binding adapter molecule 1 (IBA‐1) (microglial markers)—in elucidating AD mechanisms, promoting early diagnosis, and shaping therapeutic strategies. It also summarizes the key features and translational potential of NHP models that closely mimic human AD pathology, highlighting the promising prospects of integrating these models with the ATN(X) biomarker system. These insights strengthen the link between biomarkers, NHP research, and clinical practice, opening new avenues for the early detection and treatment strategies of AD.

Highlights

• Neuroinflammation biomarkers, including glial fibrillary acidic protein (GFAP), triggering receptor expressed on myeloid cells 2 (TREM2)/sTREM2, and YKL‐40, show strong clinical potential in Alzheimer's disease (AD).

• Incorporating neuroinflammation biomarkers into the ATN(X) framework may enhance diagnostic precision.

• Advanced non‐human primate (NHP) models closely replicate human brain pathology, addressing key limitations of mouse models.

• Measuring ATN(X) biomarkers in NHPs may improve clinical translation and support early diagnosis of AD.

• Optimizing NHP models—including ApoE4 status, injection protocols, and gene‐editing approaches—is crucial for reproducibility and efficiency.

1. INTRODUCTION

Alzheimer's disease (AD) is a neurodegenerative disease that aggravates over time, representing the most common cause of dementia. ¹ , ² It has stood out as a major public concern for its high mortality rate, ³ large patient population, ⁴ and substantial economic burden. ⁵ Due to increasing life expectancy and an aging global population, the number of global dementia cases is estimated to triple by 2050 compared to 50 million in 2018, most of which are composed of patients over 65 years old. ⁶ , ⁷

Aging is the primary risk factor for sporadic AD, with the risk doubling every 5 years after age 65. ⁸ Apolipoprotein E4 (ApoE4) remains the strongest genetic risk factor for sporadic AD, ⁹ present in over 50% of cases. ¹⁰ Familial AD is defined by mutations in amyloid precursor protein (APP) and PSEN1 and PSEN2 genes, while additional risk factors include high cholesterol diets, ¹¹ alcohol consumption, ¹² and stress. ¹³ Clinically, AD manifests as deficits in memory, language, learning, and thinking, which typically emerge during the mild cognitive impairment (MCI) stage caused by synaptic and neuronal damage. As the disease advances, dementia develops, characterized by severe memory deficits, reduced independence, and eventual complete dependency.

This review represents a pioneering effort to integrate neuroinflammation‐related biomarkers into the ATN(X) diagnostic framework while comprehensively categorizing recent advancements in AD NHP models based on pathological and biomarker outcomes. It adopts a novel perspective, exploring how these models contribute to understanding AD pathogenesis and critically summarizing their strengths and limitations. Through an in‐depth exploration of changes in key biomarkers across different disease stages, this review emphasizes the integration of biomarkers, especially neuroinflammation markers, into NHP models, thereby bridging the gap between preclinical research and clinical applications and advancing the development of more effective diagnostic and therapeutic strategies for AD.

2. HUMAN AD PATHOGENESIS AND BIOMARKER TYPES ACROSS DISEASE STAGES

2.1. Tracing amyloid aggregation: From toxic oligomers, fibrils, to senile plaques

APP is a transmembrane glycoprotein that plays a major role in amyloid pathology. ¹⁴ In the physiological state, α‐secretase cleaves APP that relates to normal synaptic transmission. ¹⁵ Pathologically, APP is sequentially cleaved by β‐secretase and γ‐secretase, releasing the amyloid‐beta (Aβ) peptide. ¹⁶ The presenilin genes are responsible for encoding the active component of the γ‐secretase enzyme, which performs intramembrane cleavage. ¹⁷ Mutations in the presenilin 1 and 2 genes perturb the enzymatic activity, increasing Aβ42 production and a higher Aβ42/Aβ40 ratio, ¹⁸ , ¹⁹ promoting amyloid plaque deposition. Variations within the APP gene have been identified to facilitate the production and self‐assembly of pathological Aβ peptides. ²⁰ Collectively, these factors synergize to trigger the early onset of AD.

RESEARCH IN CONTEXT

1. Systematic review: The authors conducted a rigorous literature search using stringent criteria to identify studies pertaining to the pathological mechanisms and biomarkers associated with Alzheimer's disease (AD). Subsequently, a comprehensive search was conducted to identify studies focusing on the diagnostic and therapeutic assessment of biomarkers in the context of the disease. Additionally, a systematic search and screening process was applied to the literature on non‐human primate (NHP) models related to AD over the past three decades.

2. Interpretation: The study underscores neuroinflammation as a crucial component in the pathology of AD, supported by experimental and clinical evidence. Integrating neuroinflammatory biomarkers, particularly ATN(X) framework (TREM2), ionized calcium‐binding adapter molecule 1 (Iba1), glial fibrillary acidic protein (GFAP), and YKL‐40, into the amyloid/tau/neurodegeneration (ATN) framework holds significant potential for early diagnosis of AD. NHP models exhibit a high degree of similarity to humans. This review systematically summarizes the characteristics of various models (categorized into biomarker‐based outcomes and other results), their strengths and limitations, and proposes key questions for the building and improvement of these models.

3. Future directions: Once the feasibility of combining the proposed neuroinflammatory biomarkers into the ATN framework is confirmed, this will significantly enhance the accuracy and likelihood of early clinical diagnosis. Moreover, incorporating biomarker monitoring in NHP models and moving beyond a simple binary classification (positive or negative) will be more conducive to exploring the links between biomarkers and brain tissue pathology through quantitative analysis in positive models. Given the substantial advantages offered by NHP models, further research is needed to optimize these models, improve their reproducibility, shorten the modeling period, and ensure comprehensive expression of AD‐related pathological products.

Aβ40 and Aβ42 are two major components in Aβ deposits; notably, the structure of Aβ42 with two extra amino acids confers a greater tendency for aggregation and higher toxicity compared to Aβ40. ²¹ Aβ peptides are critical, as they can exist as monomers or aggregate into different structures, such as oligomers, protofibrils, and amyloid fibrils. ²² Oligomers are diminutive, soluble, and diffusible protein assemblies that subsequently transform into entities characterized by extensive β‐structures. ²³ They are key contributors to the pathology of AD, as they demonstrate greater toxicity than other Aβ conformations. ²² As the oligomers develop into larger and more stable aggregates, the toxicity diminishes. ²⁴ Aβ oligomers disrupt membrane’ integrity and organelle functions, spreading via cell‐to‐cell transmission. ²⁵ Oligomeric Aβ disrupts the N‐methyl‐D‐aspartate receptor (NMDAR) signaling, leading to reactive oxygen species (ROS) activation, triggering calcium dysregulation, mitochondrial impairment, and synaptic dysfunction. ²⁶ , ²⁷ , ²⁸ Notably, Aβ oligomers can exist without forming fibrils and exhibit significant neurotoxicity. Aβ oligomers and fibrils share β‐structures. ²³ These β‐structured forms nucleate fibril formation, a repetitive substructure composed of β‐strands that run perpendicular to the fibril axis, forming cross‐β sheets aligned parallel to the fibril axis, ²⁹ , ³⁰ which further aggregate and deposit into plaques visible by light microscopy. According to later studies, Aβ1‐42 is the predominant component of amyloid‐β deposits found in the central cores of neuritic plaques as well as within the vascular tissues of the brain. ³¹ As Aβ1‐42 aggregates into plaques at an accelerated rate, the plasma and cerebral spinal fluid (CSF) Aβ1‐42/Aβ1‐40 ratios decline in individuals at early stages of subthreshold Aβ accumulation until the onset of dementia. ³² , ³³ , ³⁴ , ³⁵ , ³⁶ Since plaque build‐up is a complicated process that requires a high kinetic barrier and needs to undergo two phases of nucleation, the duration of developing such plaque is quite long, at an estimation of 15–20 years ³⁷ , ³⁸ (Figure 1).

FIGURE 1.

A schematic depiction of amyloid pathology and its progressive transformations in AD. The figure displays the sequential formation and accumulation of various Aβ species, initiated by the cleavage of APP via β‐ and γ‐secretases, resulting in the production of monomeric Aβ peptides. These monomers are categorized into on‐pathway and off‐pathway processes, with on‐pathway species displaying the capacity to aggregate into soluble oligomers, which are considered pivotal contributors to synaptic dysfunction. Off‐pathway processes involve certain Aβ monomers deviating from the conventional aggregation pathway, resulting in the failure to form typical oligomers or fibrils. Instead, these monomers may remain in their monomeric state or form alternative structures. Additional aggregation leads to the formation of protofibrils and eventually insoluble amyloid fibrils that deposit as amyloid plaques within the brain parenchyma, a defining feature of Alzheimer's pathology. The dynamic interconversions among these species, their potential reversibility, and their respective roles in neuronal toxicity are illustrated to emphasize the multifaceted nature of amyloid pathogenesis. Aβ, amyloid‐beta; AD, Alzheimer's disease; APP, amyloid precursor protein

According to the amyloid cascade hypothesis, the imbalance between the production and clearance of Aβ is a fundamental driver of Aβ aggregation, neurodegeneration, and cognitive impairment. ³⁸ Glia, particularly microglia and astrocytes, are crucial in Aβ clearance, demonstrated by one study that degradation of Aβ by a synergistic interplay between microglia and astrocytes via cell‐to‐cell communication, ³⁹ with microglia being more effective. While this crosstalk is beneficial for Aβ clearance, it may also pose a risk of aggravating Aβ pathology. ⁴⁰ In the brain, Aβ spreads in a prion‐like manner, with Aβ deposits initially exhibiting robust metabolic activity in regions of the cortex, including the precuneus, medial orbitofrontal cortex, and posterior cingulate cortex. ⁴¹ Subsequently, these deposits extend throughout the neocortex, infiltrating the brainstem and subcortical nuclei. ⁴²

With a deeper understanding of the disease, the Aβ cascade is now involved not only in its own pathogenesis but also in its interplay with tau pathology, neuroinflammation, and neurodegeneration. ⁴³ , ⁴⁴ , ⁴⁵ Their intricate relationship between Aβ and tau has been extensively studied recently. It is renowned that Aβ functions as an upstream mechanism in tau pathology, ⁴⁶ in the human brain, tau production positively correlates with amyloid plaque number. ⁴⁷ Rodent studies with APP/PSEN1 mutations have confirmed tau as a critical factor, as knocking out the tau gene reduced neurotoxicity and memory impairment. ⁴⁸ , ⁴⁹ Additionally, a decrease in endogenous tau saved hAPP mice from cognitive deficits and plaque burden in the hippocampal area. ⁵⁰ Furthermore, a study by Hurtado compared mice with mutations in frontotemporal dementia with Parkinsonism (PS19 mice), familial early‐onset AD (PDAPP mice), or their hybrids, showing that combined mutations accelerated tangle formation, while amyloid plaques remained unaffected. ⁴⁸ , ⁵¹ Aβ initiates Aβ‐driven tau pathology by stimulating abnormal tau generation through phosphorylation, activated by kinases including GSK‐3β and CDK‐5. ⁵² , ⁵³ Experimental studies in P301 mouse models have demonstrated that Aβ42 fibrils specifically accelerate tangle formation. ⁵⁴ Notably, Aβ accumulation impacts tau pathology and triggers neurodegeneration through either functional loss or tau‐mediated toxicity, which is validated in cross‐sectional tau and amyloid positron emission tomography (PET) studies. ⁵⁵ Collectively, these findings indicate an Aβ‐induced, tau‐dependent mechanism for pathological tau production. The loss of neurons and synapses forms the basis for tau‐driven cognitive impairment (Figure 2). After that, the joint accumulation of Aβ and tau triggers a cascade of deleterious events from synaptic loss and neuronal death to brain atrophy, which is thought to underlie the cognitive deficits that characterize the disease. ⁵⁶

FIGURE 2.

Interaction between Aβ pathology and tau pathology in AD. Aβ aggregation triggers a cascade that promotes tau hyperphosphorylation, aggregation, and the formation of NFTs, thereby accelerating neurodegeneration and cognitive decline. In turn, tau pathology establishes a feedback mechanism that exacerbates Aβ toxicity through its effects on NMDA receptor activity, glial cell activation, and mitochondrial dysfunction, forming a synergistic cycle of neurodegeneration. Aβ, amyloid‐beta; AD, Alzheimer's disease; NFT, neurofibrillary tangle; NMDA, N‐methyl‐D‐aspartate

2.2. Mapping tau dynamics: Biomarkers from fibrils to tangles

Tau was primarily found in neuronal axons as a microtubule‐associated protein. ⁵⁷ In humans, the tau protein is encoded by the microtubule‐associated protein tau (MAPT) gene, which produces tau isoforms 0N, 1N, and 2N, as well as 3R and 4R, through alternative splicing. ⁵⁸ In physiological conditions, tau is involved in microtubule assembly, transportation, and stabilization of neuronal axons. ⁵² During the disease process, tau proteins undergo hyperphosphorylation, a post‐translational modification that occurs at common phosphorylation sites such as threonine and serine in AD. ⁵⁹ In human CSF, phosphorylated tau (p‐tau) rises first as p‐tau217, then p‐tau181 and p‐tau205, which are crucial for disease differentiation. ⁶⁰ Along with total tau (t‐tau) and other tau isoforms, they collectively reflect the severity of tau pathology in AD. ⁶⁰

Structural modifications in tau protein cause a transition from its native unfolded state to the formation of paired helical filaments or larger aggregates. Alternations detach tau from microtubules, often followed by aberrant aggregation. ⁶¹ Synaptic dysfunction therefore arises from loss of tau's microtubule stabilization, obstruction from tau aggregates, or downstream toxicity. ⁶² , ⁶³ As AD progresses, tau proteins aggregate into insoluble neurofibrillary tangles (NFTs), which is a critical diagnostic hallmark. Tau deposition follows a stereotypical pattern, beginning in the entorhinal cortex and spreading to limbic areas and eventually the neocortex. ⁶⁴ The spread of tau thus damages neurons and causes neurodegeneration. ⁶⁵ , ⁶⁶ In AD, neurodegeneration is widely detected using tau‐PET, which is highly sensitive for early and differential diagnosis of AD and provides valuable prognostic information. ¹ Moreover, an analysis involving 982 participants in 2021 found that tau pathology interacts with vascular changes related to neurodegeneration, explaining the potential mechanism by which tau contributes to cognitive impairment in AD. ⁶⁷ Tau‐PET is often used alongside the Braak staging system, a widely respected method that classifies AD by the spatial progression of pathological tau. This approach distinguishes different phases of AD and shows strong correlations with clinical, fluid, and imaging biomarkers. ⁶⁸ , ⁶⁹

2.3. Charting neurodegeneration: From synaptic dysfunction to neuron death

In AD, neurodegeneration plays a central role throughout the disease course, acting as a central driver of pathological progression and clinical symptoms. In the early stage, changes begin with Aβ plaques and early tau pathology, as tau pathology spreads, it leads to synaptic loss, brain atrophy, disrupted neural networks, and cognitive decline. ⁶⁶ , ⁷⁰ Longitudinal tau‐PET studies in human AD cases have been employed to quantify tau accumulation, showing a significant increase over time, ranging from 3.0% to 8.0% annually, among Aβ‐positive clinically diagnosed AD patients. ⁵⁵ In addition to the respective effects of Aβ and tau, their combined interactions further accelerate neurodegenerative changes. ⁷¹ A recent clinical study involving elderly subjects with a family history of sporadic AD but no cognitive impairment revealed that tau deposition following Aβ accumulation triggered a swift transition in neuronal activity from hyperactivity to hypoactivity, ⁵⁶ which was closely linked to cognitive decline, demonstrating the combined impacts of Aβ and tau on neurodegeneration.

NMDA receptors are critical for normal synaptic function and neuronal signal integration. In AD, pathological Aβ and tau proteins interact via fyn kinase, ⁷² with tau promoting fyn‐mediated NMDA receptor activity, ⁷³ causing excessive Ca²⁺ influx, disrupted cellular signaling, mitochondrial dysfunction, ⁷⁴ , ⁷⁵ increased ROS production, ultimately resulting in neuronal damage and cell death. ⁵¹ Dysregulated Ca²⁺ not only directly impacts tau and Aβ but also creates a vicious cycle that exacerbates pathology. ⁷⁶ Furthermore, two in‐depth clinical studies by van der Kant et al and Harrison revealed that Aβ accelerates tau progression, and together they worsen cognitive impairment and brain atrophy. ⁷⁷ , ⁷⁸

Neurofilament light (NfL) is a non‐specific biomarker indicative of axonal injury in neurodegenerative diseases, which was initially detected in CSF by enzyme‐linked immunosorbent assay (ELISA) and now is measurable in blood. ⁷⁹ NfL levels rise in the early stage of AD and can forecast disease severity through CSF and serum. ⁸⁰ , ⁸¹ However, given its broad nature, NfL should be complemented with other specific pathological markers for an accurate diagnosis of AD. 18F‐fluorodeoxyglucose positron emission tomography (¹⁸FDG‐PET) scanning is vital for monitoring AD progression, as it reflects glucose metabolism in regions such as the temporoparietal association cortex, anterior cingulate cortex, and posterior cingulate cortex. ⁸² This technique not only differentiates AD from other neurodegenerative diseases but also confirms clinical subtypes, evaluates disease severity, and predicts cognitive decline. ⁸² Magnetic resonance imaging (MRI) detects early brain changes, like hippocampal and cortical atrophy, before clinical symptoms emerge, ⁸³ supporting differentiation from other dementias, ⁸⁴ and enhancing diagnostic accuracy and individualized treatment guidance.

2.4. Neuroinflammatory signals: A key driver of neurodegeneration

2.4.1. Pathophysiological mechanisms of neuroinflammation in AD

Neuroinflammation refers to the inflammatory response within the central nervous system (CNS) and is integral to AD progression. ⁸⁵ Astrocytes and microglia, the CNS's primary glial cells, form a critical line of defense ⁸⁶ , ⁸⁷ and are essential for clearing insoluble Aβ and tau deposits. ⁸⁸ , ⁸⁹ However, in AD, the overactivation of glial cells triggers uncontrolled inflammatory responses, which can be neurotoxic, furthering the disease progression. ⁹⁰ They contribute by enhancing Aβ aggregation, ⁹¹ reducing neurotrophic support, causing inflammation‐induced neuronal damage, releasing abnormal phagocytic signals, ⁹² producing oxidative stress from ROS, and producing excitotoxicity from neurotoxic factors.

During AD, neuroinflammatory microglia induce a reactive astrocyte subtype termed A1 through cytokines like interleukin (IL) ‐1α, tumor necrosis factor (TNF), and C1q, which leads to neuronal death. ⁹³ Communication between microglia and astrocytes may generate a positive feedback loop, resulting in a dysregulated and self‐sustaining inflammatory response. ⁹⁴ Proinflammatory cytokines, notably plasma IL‐6 and TNF‐α, are elevated in Alzheimer's patients, driving neuroinflammation and neurotoxicity, providing compelling scientific evidence that supports the involvement of the inflammatory response in AD. ⁹⁵ , ⁹⁶ GFAP is an astrocyte‐specific and well‐established biomarker of neuroinflammation, with greater plasma increases in AD than in CSF. ⁹⁷ Besides, YKL‐40, a glycoprotein produced by active astrocytes in AD and reflecting neuroinflammation, is also a promising biomarker for diagnosis and monitoring. ⁹⁸ , ⁹⁹ Although important in AD, there is still a lack of fluid markers truly specific to microglia, as proteins (such as TREM2) related to AD lack brain specificity. ¹⁰⁰

2.4.2. Crosstalk between neuroinflammation and other AD pathologies

In AD pathogenesis, the activation of glial cells is also influenced by Aβ and tau. ¹⁰¹ , ¹⁰² The post mortem human brain study by Carrano et al observed activated microglia expressing NOX‐2 around Aβ‐burdened brain capillaries, suggesting a potential connection between microglia and aggregation of Aβ. ¹⁰³ Research has shown that Aβ not only proliferates but also activates microglia through a process known as “priming,” which sensitizes microglia to secondary inflammatory stimuli, especially in the elderly, potentially triggering an exaggerated inflammatory response. ¹⁰⁴ Activated microglia produce toxic ROS and reactive nitrogen species (RNS) through NADPH oxidase, contributing to amyloid formation and decreasing amyloid clearance. ¹⁰⁵

Both Aβ and tau disrupt microglia homeostasis, ¹⁰⁶ leading to distinct microglial subtypes, as confirmed in animal studies. ¹⁰⁷ In AD, microglia dynamically transform from homeostatic to disease‐associated microglia (DAM) via sequential stages: early interferon‐responsiveness, mid‐stage DAM activation (Aβ clearance), and late major histocompatibility complex (MHC) ‐II ^hi proliferating states driving chronic inflammation and synaptic loss, highlighting the need for collaborative efforts to more precisely understand microglial identities and states. ¹⁰⁸ , ¹⁰⁹ , ¹¹⁰ Furthermore, tau pathology also intensifies glial activity, prompting innate immune cell activation and the release of both pro‐ and anti‐inflammatory molecules ¹¹¹ that create a self‐amplifying neuroinflammation cycle. In addition to responding to Aβ and tau, microglia and astrocytes also promote Aβ and tau pathology, further accelerating AD progression. While microglia and astrocytes help Aβ clearance, they might increase pathological spread within and between interconnected networks, acting as a double‐edged sword. ³⁹ Additionally, astrocytes lead to Aβ‐dependent tau pathology, as shown in a biomarker study on cognitively unimpaired AD patients, ¹¹² indicating Aβ was linked to elevated levels of plasma p‐tau only in individuals with astrocyte reactivity (Ast+).

The blood–brain barrier (BBB) is a critical defense mechanism in the brain, maintaining homeostasis for neuronal signaling by regulating substance exchange and clearing pathological products. ¹¹³ In AD, BBB dysfunction is strongly linked to iron accumulation, which is observed from a young age and interferes with normal brain maturation. ¹¹⁴ Impaired BBB integrity allows excess iron influx, resulting in iron‐dependent programmed cell death, reinforcing oxidative stress, and intensifying chronic inflammation in the CNS, ¹¹⁵ , ¹¹⁶ which damages neurons and worsens BBB integrity. ¹¹⁷ , ¹¹⁸ A deficiency in BBB clearance promotes the aggregation of pathological products. This vicious cycle is exacerbated by genetic risk factors such as ApoE ε4, ¹¹⁹ which increases BBB permeability and iron overload, leading to accelerated cognitive decline. ¹²⁰ , ¹²¹ , ¹²² , ¹²³ Furthermore, Yuto Uchida's team also introduced advanced imaging methods, including quantitative susceptibility mapping (QSM) and dynamic contrast‐enhanced MRI (DCE‐MRI), ¹¹⁷ , ¹²¹ which now allow for non‐invasive detection and monitoring of BBB breakdown and iron deposition, serving as potential early biomarkers for monitoring AD.

Notably, abnormal neuroinflammation should not be regarded solely as a result of the disease; it is also a significant triggering factor. Biomarkers appearing according to disease stages have significantly advanced our understanding of the pathogenesis, diagnosis, and monitoring of AD. In the future, integrating multidimensional biomarkers within the ATN framework could enhance the accuracy of early diagnosis and treatment monitoring.

3. CURRENT USES OF BIOMARKERS IN HUMAN AD

3.1. Biomarkers in AD diagnosis: Integrating the ATN framework

Earlier diagnostic criteria for AD relied on clinical symptoms, such as the Clinical Dementia Rating (CDR), and PET imaging to detect NFTs and senile Aβ plaques, as approved by the U.S. Food and Drug Administration (FDA). ¹²⁴ However, pathological changes may occur as early as 15 to 25 years before the onset of memory symptoms, often delaying diagnosis until the disease is advanced and treatment options are suboptimal, as they are unable to stop disease progression. ¹ Therefore, early diagnosis of AD is desperately needed for earlier interventions. ¹²⁵

Biomarker refers to specific, measurable indicators within a biological system that reflect normal physiological functions, pathological alterations, or pharmacological responses to therapeutic interventions. It is important that diagnostic AD biomarker tests introduced be accurate (a sensitivity and specificity of >85%, and a positive predictive value of >80%) and suitable for general medical practice in various routine clinical settings. ¹²⁶ The National Institute on Aging‐Alzheimer's Association's (NIA‐AA's) A(amyloid)T(tau)N (neurodegeneration) framework, which is revised in 2024, provides a biomarker‐based approach for diagnosing AD, ¹²⁷ , ¹²⁸ with tau and Aβ recognized as two core pathological markers reflecting the key features of the disease.

Currently, the ATN framework has proven effective in identifying individuals with a higher risk of AD in clinical practice, with studies demonstrating that biomarker levels correlate with worse cognitive trajectories in individuals with MCI and those who are cognitively normal, irrespective of whether they report subjective memory complaints. ¹²⁹ , ¹³⁰ , ¹³¹ , ¹³² Furthermore, the framework can also predict the onset of dementia, potentially up to 5 years from MCI or cognitively normal states. ¹³³ , ¹³⁴ However, further in‐depth and extensive research is necessary before the ATN framework can be widely applied in clinical practice. Integrating multi‐omics biomarkers and advanced neuroimaging may provide the potential to refine the ATX(N) framework, thereby enhancing the accuracy of future classification algorithms across the spatial and temporal dimensions of AD. ¹³⁵

3.1.1. Amyloid‐β (A): Marker of early pathological plaque deposition

Amyloid‐β serves as a primary component of senile plaques in the brain. Aside from amyloid‐PET, Aβ1‐42 (rather than Aβ1‐40 or total Aβ alone) in CSF correlates more strongly with a diagnosis of AD. ⁹⁹ Blennow and Hampel's study included 13 independent studies, encompassing approximately 600 AD patients and 450 healthy controls. With a specificity of 90%, the average sensitivity of CSF Aβ1‐42 for distinguishing AD from normal aging is 86%. ¹³⁶ Previous studies have demonstrated that Aβ42 is more likely to form plaques compared to Aβ40, which explains the strong correlation between the reduced Aβ42/Aβ40 ratio in plasma and CSF and plaque deposition. ¹³⁷ This ratio is also closely associated with amyloid PET imaging results, ¹³⁸ which surpass standalone CSF Aβ measurements in predicting amyloid positivity, highlighting its potential as a diagnostic marker for AD.

3.1.2. Tau (T): Indicator of tangle pathology and Braak staging

CSF p‐tau detection remains a standard in clinical practice. For the diagnosis of AD, tau phosphorylation at Thr181, Ser199, Thr231, Ser205, Thr217, and Ser404 have been identified as key isoforms, and they each offer unique pathological and diagnostic implications. ¹³⁹ Among these, p‐tau231, p‐tau181, and p‐tau217 have shown their potential in distinguishing early AD in individuals with Aβ positivity. ¹⁴⁰ p‐Tau217, in particular, increases during the early stages of amyloid plaque formation, whereas p‐tau205 demonstrates greater specificity and sensitivity in detecting the onset of neuronal dysfunction. ⁶⁰ Furthermore, p‐tau217 is particularly noteworthy due to its strong correlation with longitudinal tau‐PET measurements. ¹⁴¹ , ¹⁴² In addition, CSF p‐tau262 and p‐tau356 enable sensitive detection of NFTs, allowing AD to be distinguished from other tauopathies and providing both diagnostic and therapeutic value. ¹⁴³

Blood biomarkers are also highly valued for their non‐invasive nature, with comparable accuracy to CSF tests, lower cost, and greater acceptance among patients in clinical settings. Plasma p‐tau181, p‐tau217, and p‐tau231 prove strong performances in diagnosing AD, with p‐tau217 standing out for its superior accuracy in both early detection and differentiation ¹⁴⁴ , ¹⁴⁵ of AD compared to other established markers. ¹⁴⁶ , ¹⁴⁷ , ¹⁴⁸ , ¹⁴⁹ p‐Tau181 also effectively identifies Aβ‐PET positivity and predicts future AD in cognitively unimpaired and MCI individuals. ¹⁴⁶ Recently, the plasma microtubule‐binding region (MTBR) containing residue 243 (eMTBR‐tau243) performed with high accuracy in identifying tau tangle pathology and late‐stage AD, with levels rising from the MCI stage and continuing to increase until dementia. ¹⁵⁰

Tau‐PET, a non‐invasive method for visualizing tau protein accumulation in the brain, stands out as a priority biomarker for differential diagnosis with its high accuracy and specificity of over 90%. ¹⁵¹ , ¹⁵² By incorporating the advanced tau tracers (such as [18F]‐AV‐1451, MK‐6240, and PI‐2620), tau‐PET enables even more precise localization and quantification of tau deposits in specific brain regions, thereby supporting pathological staging based on the Braak staging scale. A longitudinal neuroimaging study on patients in the early stages of AD showed that initial tau‐PET measurements correlated with subsequent MRI‐detected neurodegeneration and forecasted cortical atrophy and cognitive performance over the following year. ¹⁵³ , ¹⁵⁴ Notably, Vogel's clinical study demonstrated tau‐PET in identifying both the typical pattern of tau pathology progression in AD (from the entorhinal cortex to the neocortex) and atypical tau distribution patterns associated with non‐typical clinical presentations, with reproducibility of these findings across different samples. ¹⁵⁵ These discoveries highlight the diversity of AD, challenge the widely recognized model of tau propagation, and underscore the critical role of tau‐PET in individualized assessments.

3.1.3. Neurodegeneration (N): Clinical relevance with neuronal damage and functional decline

In the context of fluid biomarkers, elevated soluble t‐tau in CSF may result from the passive release of tau protein triggered by neurodegeneration, which is linked to Aβ‐related pathology. ¹⁰¹ Increased NfL levels in blood and CSF indicate neuroaxonal damage. ¹⁵⁶ Mattsson's study using the ATN framework found that plasma NfL concentrations were higher in patients with AD dementia compared to those with MCI, and these levels correlated with disease severity. Furthermore, changes in NfL were also associated with the degree of alignment with the ATN criteria, suggesting the potential of plasma NfL as a marker for tracking disease progression and assessing pathophysiological processes in AD. ¹³⁹ Brain‐derived neurotrophic factor (BDNF), which supports synaptic plasticity associated with learning and memory, is typically decreased in AD. ¹⁵⁷ However, conflicting findings regarding plasma BDNF levels highlight the need for further investigation before its clinical application can be considered. ¹⁵⁸ , ¹⁵⁹

MRI demonstrates brain atrophy in AD patients, especially in the hippocampus and medial temporal lobe (MTL), providing a reliable basis for early diagnosis. Similarly, fluorodeoxyglucose ¹⁸FDG‐PET scans measure cerebral glucose metabolic function, and reduced metabolic activity in the bilateral temporoparietal regions is a diagnostic marker for AD, ¹⁶⁰ helping to distinguish it from other types of dementia when combined with additional biomarkers.

3.1.4. Neuroinflammation (X): Emerging role in Alzheimer's pathology

GFAP serves as an indicator of astrocyte activation, with an increase in its levels detectable in both CSF and plasma. ¹⁶¹ Plasma GFAP levels are markedly elevated in preclinical AD, even in cognitively normal, Aβ‐positive individuals, and continue to increase as the disease progresses. ¹⁶² , ¹⁶³ , ¹⁶⁴ This elevation closely correlates with Aβ pathology, evidenced by decreased levels of Aβ1‐42 and a strong correlation with Aβ neuroimaging biomarkers. ¹⁶⁵ , ¹⁶⁶ However, since GFAP also fluctuates in various neurodegenerative diseases and its diagnostic accuracy is lower than that of Aβ and tau‐related biomarkers, ⁹⁹ its reliability as a standalone biomarker is limited. In contrast, plasma GFAP changes are minimal in Parkinson's disease across cognitive stages ¹⁶⁷ and in pure dementia with Lewy bodies without AD co‐pathology, ¹⁶⁸ , ¹⁶⁹ indicating a distinct AD‐specific, Aβ‐driven astrocyte activation pattern. GFAP levels are also significantly higher in AD compared to frontotemporal dementia, providing moderate effectiveness for differential diagnosis (area under the curve [AUC] 0.818). ¹⁶¹ In Alexander disease, GFAP elevation is primarily due to core GFAP pathology, whereas in AD, it is secondary to underlying disease processes, resulting in lower levels and slower GFAP increases. ¹⁷⁰ , ¹⁷¹ Furthermore, GFAP levels may also rise after acute brain injury or systemic inflammation, complicating clinical interpretation. Notably, given the potential of GFAP in identifying early AD and elucidating Aβ changes and neuroinflammation, ¹⁷² its integration into biomarker panels alongside other core AD markers is promising to improve early and differential diagnosis. ¹⁷³ YKL‐40 stands for the activation of microglia and astrocytes. ¹⁷⁴ , ¹⁷⁵ In AD, its levels in plasma and CSF show a moderate to large effect size (1.28–1.47). ⁹⁹ While not directly reflecting core AD pathology, YKL‐40 can predict neuroinflammation and is involved in Aβ‐induced tau pathology and tau‐related neuronal injury. ¹⁷³ It holds potential as a biomarker for neurodegeneration and glial activation and holds the potential of integration into neuroinflammation biomarkers in clinical assessments. ¹⁷⁶

Microglia activation markers, including ionized calcium‐binding adapter molecule 1 (IBA‐1) and triggering receptors expressed on myeloid cells 2 (TREM2), typically increase during the AD continuum. ¹⁷⁷ TREM2 is a microglial receptor that detects damage and regulates immune and metabolic responses, existing both as a membrane protein and as soluble sTREM2 (sTREM2) detectable in CSF and plasma. ¹⁷⁸ However, sTREM2 is not supported as a stand‐alone biomarker due to its variability and limited disease specificity. ¹⁷⁹ , ¹⁸⁰ , ¹⁸¹ Additionally, CSF sTREM2 levels peak during the MCI stage and then plateau or decline during dementia, forming an inverted “U” curve. ¹⁷⁸ Given that TREM2 and sTREM2 are also influenced by various factors such as NLRP3 ¹⁸² and microglia subtypes, ¹⁸³ their independent use remains controversial. However, it is undeniable that the TREM2 molecule is solely expressed by myeloid cells in the brain and serves as a direct marker of their activity. ¹⁸⁴ , ¹⁸⁵ Recent clinical studies also demonstrated its significant potential in understanding myeloid cell responses, predicting cognitive impairment, and potential as therapeutic targets. ¹⁸⁶ , ¹⁸⁷ , ¹⁸⁸ , ¹⁸⁹ , ¹⁹⁰ , ¹⁹¹ Integrating TREM2/sTREM2 with other markers within the ATN(X) framework may improve diagnostic accuracy, enhance differential diagnosis, and support clinical monitoring, especially with advanced nanosensing platforms. Furthermore, proinflammatory cytokines such as IL‐6 and TNF‐α within the innate immune system are released by activated glial cells and have been shown to correlate with disease progression. ¹⁹² However, the high variability and low disease specificity of these biomarkers limit their clinical usefulness, as levels are easily affected by various systemic conditions. ¹⁹³ Therefore, we excluded them from our clinical framework in favor of more specific markers.

Based on current pathological mechanisms and clinical research findings, we propose integrating neuroinflammation biomarkers as the “X” component within the ATN(X) framework, with the potential to enhance diagnostic accuracy and sensitivity for AD. Further investigations into neuroinflammation are crucial to better understand its role in AD and to refine the utility of these biomarkers for broader clinical application, which may offer a promising avenue for early diagnosis in AD. Biomarkers hold great potential in the screening, diagnosis, staging, and progression prediction of AD (Table 1). Although research on biomarkers has been rapidly advancing in recent years, current definitions based solely on biomarkers are not yet fully applicable to clinical practice and require further validation. For more accurate and personalized diagnosis in clinical settings, especially in cognitively unimpaired individuals, biomarkers should be considered closely alongside clinical phenotypic presentations. ¹⁹⁴

TABLE 1.

Summary of molecular and imaging biomarkers associated with Alzheimer's disease

Characteristic	Origin	Marker	Significance
Aβ
	CSF, plasma	Aβ1‐40	Aβ1‐40↑: differentiates between AD(+) and AD(−) individuals 18 , 19 , 99
		Aβ1‐42	Aβ1‐42↓: Aβ plaque formation 32 , 33 , 34 , 35 , 36
		Aβ42/40	Aβ42/40↓: correlates with plaque status, amyloid‐PET; surpasses standalone CSF Aβ 32 , 33 , 34 , 35 , 36 , 137 , 138
	Brain	Amyloid‐PET	Detection of amyloid deposition 99
Tau
	CSF	p‐tau181	p‐tau181↑: the gold standard in evaluating phosphorylation, helps diagnose AD; differentiates between Aβ(+) and Aβ(−) patients 139 , 140
		p‐tau205	p‐tau205↑: the onset of neuronal dysfunction 60 , 139
		p‐tau217 (best performance)	p‐tau217↑: differentiates between Aβ(+) and Aβ(−) patients; predicts Aβ plaque formation; strong correlation with tau‐PET 139 , 140 , 141 , 142
		p‐tau231	p‐tau231↑: differentiates between Aβ(+) and Aβ(−) patients 139 , 140
		p‐tau262	p‐tau262↑: diagnostic value in AD; sensitive detection of NFTs 143
		p‐tau356	p‐tau356↑: helps diagnose AD; correlates with NFTs 143
		MTBR‐tau species 243, 299, 354	MTBR‐tau species 243, 299, 354↑: associate with tau‐PET imaging, cognitive decline 150
	Plasma	p‐tau181,231,217	p‐tau181,231,217↑: promising diagnostic markers of AD; p‐tau217 owes the best diagnostic performance with accuracy on differentiation and early diagnosis 144 , 145 , 146 , 147 , 148 , 149
	Brain	TAU‐PET	Localization and quantification of tau deposits; PET‐based Braak staging in clinical setting 68 , 69 , 151 , 152 , 153 , 154 , 155
Neurodegeneration
	CSF, plasma	NfL	NfL↑: neuroaxonal damage; ↑with disease severity; more obvious upregulation happens due to Aβ and tau in advanced stages 79 , 80 , 81 , 139 , 156
	CSF	BDNF	BDNF↓: impaired spatial memory, neuronal damage 157 , 158 , 159
	CSF, plasma	t‐tau	t‐tau↑: neurodegeneration; dependent on Aβ‐related pathology 101
	Brain	MRI brain volume	MRI↓brain volume: atrophy; neurodegeneration; early diagnosis of AD 83 , 84
	Brain	18FDG‐PET reactivity	↓18F‐FDGPET reactivity:↓glucose metabolism; AD diagnostic indicator; differentiate between AD(+) and AD(−) patients 82 , 160
Neuroinflammation
	CSF, plasma	GFAP	GFAP↑: astrocyte activation; correlates with Aβ pathology 97 , 161 , 162 , 163 , 164 , 165 , 166 , 172 , 173
		YKL‐40	YKL‐40↑: reflects glial activation, neuroinflammation; mediates Aβ‐induced tau pathology 99 , 173 , 174 , 175 , 176
		IBA‐1	IBA‐1↑: microglia activation 99 , 177
		TREM2	TREM2↑: microglia activation; associates with Aβ plaque deposition 177 , 178 , 179 , 180 , 181 , 184 , 185 , 186 , 187 , 188 , 189 , 190 , 191

Abbreviations: 18FDG‐PET, 18F‐fluorodeoxyglucose positron emission tomography; Aβ, amyloid‐beta; AD, Alzheimer's disease; BDNF, brain‐derived neurotrophic factor; CSF, cerebral spinal fluid; GFAP, glial fibrillary acidic protein; MRI, magnetic resonance imaging; MTBR, microtubule‐binding region; NfL, neurofilament light chain; NFT, neurofibrillary tangle; PET, positron emission tomography.

3.2. Targeting therapy via biomarker in AD: Unmet needs

The complexity and multifactorial nature of AD pose significant challenges for effective treatment. To date, only nine FDA‐approved drugs exist for AD. Among these, cholinesterase inhibitors—tacrine, donepezil, galantamine (capable of crossing the BBB), and rivastigmine—and the NMDA receptor antagonist memantine ¹⁹⁵ are utilized to mitigate cognitive decline. Additionally, suvorexant and brexpiprazole are prescribed to manage non‐cognitive symptoms such as sleep disturbances and agitation, respectively. Despite their use, clinical trials have demonstrated limited efficacy in altering disease progression or improving MCI. ¹⁹⁶ , ¹⁹⁷ This has shifted the focus toward disease‐modifying therapies and immunotherapies targeting specific pathological products, marking a new era in AD treatment.

Therapeutic strategies mainly fall into two categories: one focuses on preventing the formation of pathological products to inhibit disease progression, while the other aims to facilitate the clearance of existing pathological products through pharmacological or other therapeutic interventions, accelerating their degradation. Immunotherapy has emerged as a promising strategy in this domain. Biomarkers are invaluable tools for disease targeting and monitoring of treatment efficacy, and a deeper understanding of the molecular mechanisms underlying AD will further clarify the significance of dynamic changes in these biomarkers.

3.2.1. Conformation‐specific Aβ immunotherapy: Efficacy and amyloid‐related imaging abnormalities risks

Therapeutic strategies targeting Aβ have undergone extensive investigation. Clinical studies have indicated that approaches aimed at interfering with β‐ and γ‐secretases, which are responsible for the production of Aβ peptides, have not yielded significant outcomes. Recently, Aβ‐targeted immunotherapies, such as aducanumab and lecanemab, which target soluble Aβ oligomers to insoluble plaques, received approval from the FDA in 2021 and 2023, respectively. Phase III clinical trials have demonstrated that these therapies effectively reduce amyloid plaque burden, with improvements in cognitive and functional outcomes. ¹⁹⁸ For instance, aducanumab binds to aggregated forms of Aβ, reducing plaque deposition and slowing disease progression by approximately 27%. The ENGAGE and EMERGE clinical trials of aducanumab demonstrated 13%–16% reductions in plasma p‐tau181 levels in high‐ and low‐dose groups compared to placebo after 56 weeks of treatment. These results were consistent with amyloid‐PET findings, supporting the drug's efficacy in reducing amyloid pathology and the potential of biomarkers in monitoring treatment effects. ¹⁹⁹

Similarly, the TRAILBLAZER‐ALZ randomized clinical trial led by Michael Pontecorvo showed that donanemab, targeting a specific form of Aβ, namely Aβ with pyroglutamate at the N‐terminal 3rd position (abbreviated as N3pG‐Aβ), significantly reduced plasma p‐tau217 levels by 24% and also decreased GFAP levels compared to placebo. These effects were in line with reductions in amyloid burden observed in Aβ‐PET imaging, as demonstrated in Sergey Shcherbinin's study. ²⁰⁰ , ²⁰¹ In contrast, the phase III trial of solanezumab, a monoclonal antibody targeting monomeric amyloid, showed its ability to slow Aβ aggregation as observed using amyloid‐PET over 5 years but failed to influence tau propagation (measured by tau‐PET), cognitive decline, or brain atrophy (measured by MRI). ²⁰² Despite these limitations, solanezumab provides insights into the selective targeting of Aβ pathology. In a recent phase III randomized clinical trial, both solanezumab and gantenerumab demonstrated reduction of Aβ fibrils, as well as lowering levels of CSF sTREM2, GFAP, and NfL in both blood and CSF, biomarkers associated with neuroinflammation and neurodegeneration. ²⁰³

It is important to note that Aβ immunotherapies carry significant risks. In phase III trials of Aducanumab, approximately one‐third of participants experienced amyloid‐related imaging abnormalities (ARIAs), ²⁰⁴ with ApoE4 carriers at heightened risk. ²⁰⁵ ARIAs include ARIA‐E (vasogenic edema or subarachnoid/subcortical effusion) and ARIA‐H (cerebral microbleeds or hemosiderin deposits). Although often asymptomatic, ARIAs can occasionally lead to severe or even fatal cerebral hemorrhage. Despite representing a significant milestone in the treatment of AD, immunotherapies carry serious side effects, necessitating the adoption of stringent clinical usage criteria.

3.2.2. Tau‐targeted therapies: From phosphorylation to clearance, triumphs, and translation gaps

p‐Tau has emerged as a valuable tool for monitoring therapeutic responses, as it reflects both tau and Aβ pathologies. ²⁰⁶ Meanwhile, tau is also highly promising as a therapeutic target since it directly associates with synaptic loss and neurodegeneration. ⁶⁶

Glycogen synthase kinase‐3 (GSK‐3), a critical kinase involved in tau phosphorylation in AD, has been explored as a therapeutic target. Blocking GSK‐3 activity in transgenic mouse models reduced tau phosphorylation and mitigated axonal transport deficits from Aβ pathology. ²⁰⁷ Consequently, GSK‐3 is considered a highly promising therapeutic target. Currently, drugs targeting GSK‐3 are categorized into adenosine triphosphate (ATP) ‐competitive inhibitors, exhibiting high activity and well‐defined mechanisms but with limited kinase selectivity, and ATP‐noncompetitive inhibitors, offering high selectivity but possessing less‐understood mechanisms. ²⁰⁸ Among these agents, only lithium carbonate and tideglusib have advanced to clinical trials. Preclinical studies on tideglusib using double transgenic mice expressing human APP and tau reported its ability to lower tau phosphorylation, reduce amyloid deposition, attenuate neuronal loss, and improve cognitive function. ²⁰⁹ However, as an irreversible, time‐dependent inhibitor, ²¹⁰ , ²¹¹ tideglusib showed limited efficacy in reducing cognitive and functional decline in a phase II trial conducted on patients with mild‐to‐moderate AD. ²¹²

Another approach to reducing tau protein expression targets tau mRNA transcripts. For example, antisense oligonucleotides (ASOs) regulate tau mRNA through multiple mechanisms, including direct degradation of mRNA, modulation of mRNA splicing, regulation of mRNA stability, and inhibition of mRNA translation. ²¹³ ASOs decrease the production of pathological tau and possibly reverse existing tau productions. ¹⁰¹ Besides, ASOs also effectively reduced human tau mRNA and protein levels in P301S mutant tau‐expressing mice and cynomolgus monkeys, achieving reductions in brain and CSF tau. ²¹⁴ This was associated with lowered tau inclusions, reversal of p‐tau, prevention of MRI‐detected hippocampal atrophy, and cognitive improvement, which highlight the therapeutic potential of tau‐lowering ASOs for patients with tau‐positive inclusions, even in advanced stages of tau pathology. A phase 1b trial investigating ASOs targeting MAPT demonstrated their safety in humans and a dose‐dependent reduction in t‐tau and p‐tau181 levels. ²¹⁵ In a Phase 1b trial involving participants with mild AD, the MAPT‐targeting ASO BIIB080 demonstrated favorable tolerability and efficacy. The treatment led to reductions in CSF t‐tau, p‐tau181, and tau PET signals, highlighting its potential to mitigate cognitive decline. ²¹⁶ Evaluating the potential of this mechanism for AD treatment is crucially dependent on elucidating the role and extent of extracellular tau in pathological propagation, as well as demonstrating the efficacy of therapeutic interventions targeting this process. ²¹⁷

As an alternative and highly competitive therapeutic approach, current immunotherapy primarily targets the N‐terminal and MTBR domains of tau protein, which are critical functional regions involved in tau's interaction with microtubules. ¹⁵² These regions can be conceptualized as “staplers” facilitating this binding process. Immunotherapy can be classified into active and passive immunization, both targeting the inhibition of tau propagation between neurons and the prevention of its aggregation. Active immunization strategies, including AADvac‐1, which targets the N‐terminal region of tau, and ACI‐35, which targets the phosphorylated tau epitope p‐tau396/404, have completed phase 1 clinical trials and are currently undergoing phase 2 trials. ²¹⁸ Therapeutic approaches directed at the N‐terminal region, such as semorinemab, and those targeting the MTBR, including zagotenemab, have demonstrated reductions in CSF p‐tau levels during phase 2 clinical trials. ²¹⁹ Specifically, semorinemab has exhibited limited effectiveness in the early stages of AD. However, none of these trials have shown significant reductions in tau‐PET signals, a key marker of pathological tau burden. The underlying mechanisms remain poorly understood, with several hypotheses suggesting that these therapies may fail to target the majority of pathological tau forms or adequately address the most toxic tau species. ¹⁵² Further research is needed to better delineate tau's pathophysiological role and optimize therapeutic strategies.

4. CHALLENGES IN BIOMARKER RESEARCH: BETWEEN HUMANS AND RODENTS

4.1. The biomarker dilemma in human AD: Latency, dynamics, and heterogeneity

The study of AD biomarkers in humans faces several limitations. First, the disease has a prolonged course with pathological changes likely initiating years before the onset of noticeable clinical symptoms, making early intervention challenging. ²²⁰ Furthermore, the invasive nature of CSF sampling in clinical practice raises ethical concerns and substantially affects patient compliance. Second, AD is an immensely complex disorder. Relying on a single biomarker provides limited insight due to constrained sensitivity and specificity. ²²¹ Additionally, some biomarkers exhibit non‐linear dynamics during the disease course, plateauing in the later stages and reducing their accessibility for monitoring disease progression. ¹⁰¹ AD also displays marked heterogeneity across individuals, especially during the MCI stage, where patients may present with diverse clinical symptoms. ²²² , ²²³ This high degree of variability may arise from the inherent complexity of AD itself or the overlapping contributions of multiple neurodegenerative and non‐neurodegenerative conditions, complicating biomarkers’ interpretation in clinical and research contexts. ¹⁰¹

Finally, the role of the APOE ε4 allele, an established risk factor for AD, introduces further complexity to AD by shaping the clinical phenotypes and imparting distinct patterns of regional brain vulnerability. ¹¹⁶ Therefore, incorporating APOE ε4 genotype information is critical for improving the diagnostic accuracy and risk prediction associated with biomarker analyses. Clinical research faces challenges such as low follow‐up rates due to participant mortality, compromising the completeness and reliability of the findings. Additionally, autopsy‐based brain tissue analysis provides only a static snapshot of the disease at a specific time point, restricting insights into the dynamic processes of disease progression.

4.2. The cross‐species challenges in rodents AD modeling: Temporal disconnect, pathological mismatch, and recapitulation failure

As the most commonly used and well‐established animal models in AD research, rodents have undergone various transgenic modifications to generate models exhibiting key phenotypes associated with AD. For example, rodents have been developed to accelerate amyloid pathology through the introduction of human APP or APP+PSEN1 mutations, to model tau pathology with human MAPT mutations, and to combine genetic factors in 3xTg models (APP/PS1/Tau transgenic) to capture multiple hallmarks of human AD. ²²⁴ Recently, models such as 5xFAD mice (carrying APP Swe, Flo, Lon, and PS1 M146L, L286V mutations), which exhibit rapid‐onset amyloidosis, neurodegeneration, and memory impairment, ²²⁵ as well as 3xTg‐AD models (expressing human APP695(Swe), MAPT4R0N(P301L), and knock‐in Psen1M146V mutations) that replicate the interplay between Aβ, tau pathology, and synaptic dysfunction, ²²⁶ have been used for studying AD pathology. With the advancement of study, transgenic models targeting the APOE genetic background of human AD have been developed. ²²⁷ , ²²⁸ These models simulate the genetic context of human APOE by expressing human APOEε3 or APOEε4 while knocking out the rodent APOE gene. Additionally, by knocking out the TREM2 gene, researchers can replicate the functional loss or specific mutations of TREM2 observed in human AD, thereby enabling in‐depth investigation into its role in the disease's pathogenesis. ²²⁹

Despite their significance, rodent models have an alarming failure rate as high as 99.6% when translating successful therapeutic strategies to humans during drug development, underscoring the immense challenges inherent in cross‐species research and raising significant doubts about their validity. ²³⁰ , ²³¹ One of the inherent limitations of using rodent models in AD research is their inability to fully replicate the broader pathological and clinical characteristics observed in human AD. ²³² For instance, while rodents can develop amyloid plaque formation, they often fail to exhibit the cognitive deficits typical in humans. ²³³ Additionally, rodents express fewer isoforms of key proteins, and their tau protein sequence lacks certain human‐specific variants or post‐translational modifications critical for the pathology of human AD. This discrepancy hinders the ability of rodent models to accurately reflect key aspects of tau protein phosphorylation and the resulting pathological folding and aggregation observed in humans. Another issue lies in the fact that rodent models exhibit pathological features not consistent with human AD, possibly due to overexpression of pathological markers that potentially lead to the mislocalization of amyloid plaques, ²³⁴ a phenomenon rarely observed in human AD.

A critical consideration in the evaluation of models is the interplay between amyloid and tau proteins, which, in human AD, progress in a closely interconnected and mutually influential manner. In rodents, these pathological progressions often remain relatively independent and disconnected, limiting their ability to fully represent human AD pathogenesis. ²³⁵ , ²³⁶ Introducing Aβ or tau proteins into rodents can accelerate pathological accumulation but offers little insight into initial mechanisms. Moreover, it is critical to note that over 95% of AD cases are sporadic with poorly understood pathology, ²³⁷ whereas the most commonly used transgenic rodent models carry mutations associated with early‐onset familial AD, which accounts for less than 5% of cases. ²³⁸ Future improvements in mouse models still have vast potential in experimental design, technological optimization, and applications.

5. UNLOCKING AD INSIGHTS: NON‐HUMAN PRIMATE MODELS FOR AD

Difficulty in developing effective therapies for AD underscores the urgent need for animal models that fully reflect disease pathology and progression while serving as platforms for exploring potential treatments. NHP models provide invaluable advantages for AD research due to their physiological congruence with humans. For instance, rhesus macaques share approximately 93% genomic similarity with humans, ²³⁹ making them an ideal model for studying molecular and cellular biological processes, immune responses, and signal transduction mechanisms. Moreover, NHPs show 100% human Aβ sequence homology, express both 3R and 4R tau isoforms, and exhibit robust large‐scale neurogenesis. ²⁴⁰ Advanced imaging enabled by their larger brains offers insights into lesion dynamics, while genetic engineering techniques further refine their relevance by replicating specific human AD pathologies.

Furthermore, NHPs exhibit behavior changes similar to humans, as has been systematically summarized in previous studies, such as a reduction in gait speed. These characteristics make NHPs an ideal model for studying brain pathology, as they allow for the MCI simultaneous evaluation of behavioral changes associated with AD and offer deeper insights into the potential connections between pathological changes and behavioral manifestations.

In AD research, the clinical relevance of biomarkers in diagnostic and treatment monitoring has been strongly supported by NHP models. These models demonstrate pathological features and behavioral changes within a biological and physiological context highly analogous to human conditions. These features and changes serve not only as diagnostic indicators but also as measures of pathological progression, thereby enhancing the predictive validity of preclinical studies. Over the past two decades, the application of NHP models in AD research has shown a steady increasing trend (Table 2). At the same time, the evaluation of biomarkers in NHP models has also witnessed increased utilization.

TABLE 2.

Summary of NHP models for AD

Characteristic	Methods	Reference	Primate selection	A	T	N	NI	ATN(X) biomarkers	Other findings*
Aβ
	Sporadic
		Souder et al.,2021 241	Rhesus monkeys (∼31 year)	–	–	–	–	None	Aβ plaques [temporal cortex]; mitochondrial distribution around plaques; lipofusin particles[brain]; IBA‐1↑; GFAP↑
		Stonebarger et al.,2020 242	Rhesus monkeys (22–44 year, average 31.8 year)	–	–	–	–	None	Aβ plaques [Brodmann areas 32 & 46 of PFC]; NEUN+↓
		Zhang et al.,2019 243	Aged rhesus monkeys (19–31year)	–	–	–	–	None	Aβ plaques [cerebral cortex]; IBA‐1↑, GFAP↑ [cerebral cortex]
		Zhou et al.,2018 244	Rhesus monkeys (> 30); cynomolgus monkeys (29–32 year)	–	–	–	–	None	Aβ plaques [cerebral cortex; hippocampus]
		Lardenoije et al.,2018 245	Caribbean vervets (12.2–32 year)	–	–	–	–	None	Aβ plaques [TC; hippocampus]
		Poduri et al.,1994 246	rhesus monkeys (3–15 year; 20–35 year)	–	–	–	–	None	Neuritic Aβ plaques [cortex; hippocampus]; Bergmann's astrocytes; APOE e4(PCR); APOE immunoreactivity [ERC; neocortex; hippocampus]
		Bons et al.,1994 247	Microcebus murinus (> 8y)	–	–	–	–	None	Amyloid deposits [cortical neuropil; walls of selected meningeal and cerebral vessels (CAA)]
		Podlisny et al.,1991 248	Cynomolgus monkeys	–	–	–	–	None	Aβ plaques[PFC]; CAA [cortex]; GFAP (around plaques); complete Amino Acid Homology to human
	Injection
	AβOs(100 µg) once/3d for 24d(data last collected on day30)	Beckman et al.,2019 249	Rhesus monkeys (female11–19 year; 22–28 year)	+	–	–	–	Aβ1‐42↑(CSF);AβOs↑(CSF)	Oligomeric Aβ[postsynaptic densities]; selective loss of highly plastic thin spines [area 46 of DLPFC]; IBA‐1↑(microglia activity, priming; ↑microglia volume); TNF‐α↑(CSF)
	Aβ fibrils+ lipopolysaccharide (LPS);(5months)	Philippens et al.,2017 250	Marmosets (Callithrix jacchus)	–	–	–	–	None	Aβ plaques (61% Aβ1‐42 positive); IBA‐1+ [around plaques];↑CD95 and CD45RA expression on CD3+CD4+ cells(blood)
	Intracortical injections of Aβ (200 pg)	Sani et al.,2003 251	Rhesus monkeys (5 year; 25–30 year)	–	–	–	–	None	Aβ deposits& plaques [cerebral cortex]; CAA [cerebral cortex; occipital cortice];plaque composition: Aβ40/Aβ42≈2.1
Tau
	Sporadic
		Leslie et al.,2021 252	Rhesus monkeys (18–30 year)	–	–	–	–	None	WB: pSer235,396‐tau [brain]
		Sharma et al.,2019 253	Common marmosets (Callithrix jacchus; 0–5 year)	–	–	–	–	None	Immunoblotting: tau at pSer‐202,404,235,396; Thr231 [PFC]
	Injection
	AAV1‐P301L/S320F (data collected monthly for 6 m)	Beckman et al.,2024 254	Rhesus macaques (female 10–16 year)	–	+	+	+	pSer199, 396, pThr‐181, 231(CSF)↑; t‐tau(CSF)↑; sTrem2↑(CSF); NfL↑, BDNF↓(CSF)	pretangle states (NEUN+/AT8+/ThioS–); mature tangles (NEUN+/AT8+/ThioS+); IBA‐1↑, GFAP↑ [hippocampus, ERC]; NEUN+↓ [left CA3/Hilus]
	AAV‐P301L/S320F(data collected at1m, 2 m, and 3 m)	Beckman et al.,2021 255	Rhesus macaques (10–15 year)	–	+	+	+	pSer‐199,396, pThr‐231(CSF)↑; t‐tau(CSF)↑; sTrem2↑(CSF); NfL↑(CSF; plasma); BDNF↓(CSF)	NFTs(ThioS+ in left hippocampus); IBA1+↑[left hippocampus]); NEUN+↓(hippocampus); TNF‐α, IL‐6↑(CSF, plasma)
Aβ&Tau
	Sporadic
		Huang et al.,2024 256	Cynomolgus monkeys with T2DM (17–20 year)	+	–	+	‐	Aβ1‐40,42↓(plasma); NfL↑(CSF)	Aβ plaques [PFC; FC; TC; ERC]; NFTs [FC; TC; hippocampus]; IBA‐1+microglia↑, GFAP+ astrocytes↑; synaptic damage(↓PSD95 immunoreactivity in ERC); TNF‐α↑(peripheral blood)
		Latimer et al.,2018 257	African green monkeys (mean 11.2 year (middle‐aged); 21.7[aged])	+	+	+	–	Aβ1‐42↓(CSF); p‐tau181↑(CSF); MRI volumes↓(right prefrontal, left inferior, left posterior temporal cortex); CMRg(↓18F‐FDGPET)	Aβ plaques[lateral temporal lobe]; PHF‐tau[temporal cortex]; gait speed↓
		Darusman et al.,2019,2014,2013 258 , 259 , 260	Cynomolgus monkeys (9–15; > 20 year)	+	–	+	–	Aβ1‐42↓(CSF; compared to young); MRI atrophy (cerebral cortex; hippocampus)	Aβ1‐42 plaques [frontal, temporal, parietal, occipital lobe; hippocampus.]; CAA; tauopathy [p‐Tau231 in cytoplasm of the neuron cell]; granulovacuolar degeneration (GVD)
		Paspalas et al.,2018 261	Rhesus macaques (4.5–31 year)	–	–	–	–	None	Aβ1‐42 plaques [layer5(ERC)]; CAA; NFTs, PHF‐tau [DLPFC; ERC]; pSer214‐tau [layerII ERC cell]; Ca2+ dysregulation (pSer2808‐RyR2[DSR cisterns])
	Injection
	7.5 µL hTau‐expressing AAVs into the hippocampus; 59 weeks	Jiang et al.,2024 262	Rhesus macaques (7–15 year)	+	+	+	–	Aβ42 peptide↓(CSF); Aβ40/42↓; p‐tau181,231↑(CSF); tau‐PET (hippocampus to cerebellum); t‐tau↑(CSF); NfL↑; CMRg (↓18F‐FDG PET); MRI volume↓(hippocampus)	Aβ plaques; intra/extracellular Aβ deposits [hippocampus]; IBA‐1+microglia↑, GFAP+ astrocytes↑; CD68; NFTs[hippocampus]; ↓NEUN+ reactivity [hippocampal DG/CA2/CA3/CA4 regions]; cognitive decline (DR task; (DMTS/DNMTS) task)
	(0 month)tau aggregates purified from human AD brains(ERC) + (6th month)ICV injections of oligomeric Aβ(18 m (data collect))	Darricau et al.,2024 263	Rhesus macaques (14.7–15.2 year)	–	+	+	–	p‐tau181↑(CSF); t‐tau↑(CSF)	Aβ plaques[ERC]; CAA(exclusively in AβO injected monkeys); IBA+ microglia↑, GFAP+ astrocytes↑; neuropril threads; NFTs [ERC]; hippocampus];↓NEUN+ reactivity [CA1 layers(exclusively in AβO injected monkeys)]
	ICV or IT injection of 100‐µg AβO, Q3W*4 weeks(12 week(data collect))	Wakeman et al.,2022 264	African green monkeys (9.9+‐0.6 year)	–	–	+	–	MRI volume↓(hippocampus)	diffuse‐like plaques [6E10(MTL)]; p‐tau [hippocampus]; IBA+ microglia↑, GFAP+ astrocytes↑ [DG]
	Streptozotocin treated T1DM adult male vervet monkeys receiving BID exogenous insulin injections for 8–20 weeks	Morales‐Corraliza et al.,2016 265	Vervet monkeys (Chlorocebus aethiops) (6.9+_0.2) years	–	–	–	–	None	soluble Aβ40, 42↑[hippocampus; TC]; ↓Aβ‐degrading enzyme neprilysin (NEP) [hippocampus]; p‐tau [CP13; PHF‐1]; tau‐kinase activity [↑ERK1/2 widespread in the brain]
	Transgenic
	PSEN1‐ΔE9(genomic deletion of PSEN1 exon9 by CRISPR/Cas9)	Li et al.,2024 266	Cynomolgus monkeys	+	+	–	–	Aβ42↑(CSF); Aβ42/40↑(CSF); tau217/t‐tau↑(CSF); tau181/t‐tau↑(CSF); tau‐PET	Alterations in plasma inflammatory and immune molecules.
	Injection of human MAPT gene with Tau p301L mutation (0N4R P301L);(42month(last data))	Tu et al.,2023 267	Cynomolgus monkeys	+	+	+	–	Aβ42 Oligomers↑(spinal cord); Aβ1‐42↑(spinal cord); pThr231‐tau↑(CSF, plasma); NfL↑(CSF, plasma); t‐tau↑(CSF, plasma); MRI volume↓ (hippocampus); ↓glucose metabolism(PET([18F] FDG)‐MRI)	NFTs; ↓NEUN+ reactivity [hippocampus]; IBA+ microglia↑, GFAP+ astrocytes↑; cognitive dysfunction((1)delayed response[short‐term& long‐term memory tests]; (2)sleep behavior changes; (3)↓motor function[fine motor coordination test]; (4)anxiety‐like behaviors& ↓exploration activity))
	APPtg (Swedish K670N/M671L+ Indiana V717F mutations (APPSWE/IND)); sacraficed at 10‐year‐old	Chan et al.,2022 268	Rhesus monkeys	+	+	+	–	Aβ42↑(4y, CSF); p‐tau/Aβ42↓(4y, CSF); t‐tau↑(2–4 year, CSF)	Aβ plaques; CAA[occipital, parietal, and caudal temporal neocortices]; IBA+ microglia↑(slightly);↓Barrier Detour task at 16 months, Object‐in‐Place task(9 years of age);↑hostility(6years)

Note: “+” and “−” to indicate the presence or absence of results. Detailed findings are demonstrated in the rightmost column: “ATN(X) biomarkers”.

The table headers A, T, N, and NI represent amyloid, tau, neurodegeneration, and neuroinflammation, respectively. *The “other findings” section includes biomarker results excluded from ATN(X), post mortem brain tissue‐related results, and behavioral tests.

Abbreviations: 18FDG‐PET, 18F‐fluorodeoxyglucose positron emission tomography; Aβ, amyloid‐beta; AβO, amyloid‐beta oligomer; AD, Alzheimer's disease; APP, amyloid precursor protein; ATN, amyloid/tau/neurodegeneration; BDNF, brain‐derived neurotrophic factor; CAA, cerebral amyloid angiopathy; CSF, cerebral spinal fluid; GFAP, glial fibrillary acidic protein; IBA‐1, ionized calcium‐binding adapter molecule 1; ICV, intracerebroventricular; MAPT, microtubule‐associated protein tau; MRI, magnetic resonance imaging; MTL, medial temporal lobe; NfL, neurofilament light chain; NFT, neurofibrillary tangle; PET, positron emission tomography; TC, temporal cortex; TREM2, triggering receptor expressed on myeloid cells 2.

6. BRIDGING BIOMARKERS: INSIGHTS FROM NHP TO HUMANS

Over the past two decades, NHP models of AD have demonstrated remarkable parallels with human pathology and biomarker profiles (Figure 3). These models now allow assessment using criteria applied in humans, such as the Braak staging. These advances have positioned NHPs as a valuable platform for studying neurodegenerative disease mechanisms.

FIGURE 3.

Overview of non‐human primate models in AD. The right side depicts the further pathological changes in the brain following amyloid plaque and neurofibrillary tangle formation. The left side shows the corresponding biomarker results in the monkey model, including plasma, CSF, and imaging findings. The letters A, T, and N denote amyloid, tau, and neurodegeneration, respectively. The numbered stages on the right correspond to specific pathological stages, with the associated biomarker results indicated on the left. AD, Alzheimer's disease; CSF, cerebrospinal fluid

6.1. Recapitulating amyloid cascade

6.1.1. Sporadic aging models

Aging is a critical risk factor for late‐onset AD. In aged rhesus monkeys (19–30 years), age‐related Aβ deposits and plaque appear in the brain, particularly in the cerebral cortex, prefrontal cortex (PFC), temporal cortex (TC), and hippocampus. ²⁴¹ , ²⁴² , ²⁴³ , ²⁴⁴ , ²⁴⁶ The plaques are mainly comprised of Aβ42, with some exhibiting APOE immunoreactivity, ²⁴⁶ and amyloid‐dependent microglial activity shown by a distribution pattern of IBA‐1+ microglia tightly surrounding Aβ deposits with alterations in both morphology and volume. In Souder's study, Aβ‐driven microglial priming associated with aging was characterized, uncovering remarkable parallels in the regional impact of amyloid pathology compared to late‐onset human AD. Furthermore, positive immunostaining of GFAP surrounding plaques was also observed, ²⁴¹ highlighting the potential of using rhesus monkeys as a model for investigating neuroinflammation in AD.

Research on aged cynomolgus monkeys also identified the presence of senile plaques, ²⁴⁴ , ²⁴⁸ , ²⁵⁸ with microglial and astrocytic activation observed surrounding the plaques, revealing neuroinflammatory responses. ²⁴⁸ The aged monkeys also exhibited CAA within cortical regions, characterized by Aβ deposition in the walls of blood vessels. This vascular abnormality further underscores neuropathological features closely resembling those observed in human AD. ²⁴⁸ With nearly 100% amino acid homology to humans, cynomolgus monkeys are strong models for AD.

Other NHP species, such as vervets and microcebes, ²⁴⁵ , ²⁴⁷ demonstrated plaque burden within the TC and hippocampus. The microcebes, in particular, presented with CAA, while vervets showed potential in developing plaque pathology in the frontal cortex at the early age of 15 due to their propensity. Despite this early onset, their average life span is relatively short, typically not exceeding 30 years. This unique characteristic positions vervets as a valuable model for studying the early development of AD‐related pathologies and their potential progression over a shorter lifespan compared to other NHPs.

6.1.2. Experimental injection models

Injection‐based NHP models have been employed to study AD, particularly by introducing Aβ‐related pathological agents, which resemble the prion‐like propagation of Aβ in the human brain.

In Beckman's study, injection of oligomeric Aβ into rhesus monkeys ²⁴⁹ increases CSF Aβ1‐42, AβOs, and TNF‐α, accompanied by abnormal activation and hypertrophy of microglia in a persistently primed state. Selective loss of highly plastic thin spines in the dorsolateral prefrontal cortex (area 46) was also observed, implicating synaptic plasticity deficits.

Sepher Sani's research adopting intracortical Aβ injections in rhesus monkeys induces Aβ plaques in the cerebral cortex and CAA involving both major vessels and capillaries, primarily in the occipital cortex. ²⁵¹ A higher burden of senile plaques (composed of both Aβ40 and Aβ42) was correlated with more severe CAA, with some plaques showing cholinesterase‐positivity, similar to human Aβ plaques, further supporting the compositional similarity of senile plaques between rhesus monkeys and humans. Additional injections of lipopolysaccharide (LPS) alongside Aβ fibrils ²⁵⁰ conducted by Ingrid showed that monkeys suffered from chronic neuroinflammation and intracerebral plaques, whereas those injected solely with Aβ fibrils did not exhibit significant pathological changes. None of the subjects showed overt behavioral abnormalities. Blood analysis in LPS‐treated monkeys indicated increased expression of CD95 and CD45RA on CD3+CD4+ T‐cells, suggesting potential implications for immune‐based therapeutic strategies.

These findings collectively highlight the utility of NHP models in studying the pathophysiology of AD, particularly in replicating aspects of human Aβ pathology and related inflammatory processes.

6.2. NHPs with tauopathy

6.2.1. Sporadic models

Marmosets, due to their shorter lifespan and ease of maintenance, are studied for tauopathy. Young marmoset brains from newborns to 5‐year‐olds react with tau antibodies AT8 (Ser202/Thr205), AT180 (Thr231/Ser235), and PHF‐1 (Ser396/Ser404) through immunoblotting but decrease with age, ²⁵³ suggesting that this species may not develop tau proteinopathies associated with these phosphorylation sites. Moreover, adult marmosets exclusively express 4R tau isoforms, and tau immunotherapy shows limited efficacy in these animals.

Studies conducted on aged rhesus monkeys identified pSer235 and pSer396 in the PFC, in addition to well‐known phosphorylation sites. ²⁵² Overexpression of tau proves to upregulate GSK‐3b acetylation, ²⁶⁹ which further facilitates tau phosphorylation at Serine235 and 396, ²⁷⁰ , ²⁷¹ leading to a vicious cycle. This offers insights into the pathogenic risk of these sites and guides clinical targeting of therapeutics.

6.2.2. Virus‐induced models

In a novel‐tau‐based study on rhesus monkey models, intracortical injections of adeno‐associated virus (AAV) vectors encoding the P301L/S320F tau mutation into the left ERC of 10‐ to 15‐year‐old rhesus macaques ²⁵⁵ triggered increased CSF t‐tau and several p‐tau species, including p‐tau199, p‐tau396, and p‐tau231 in the CSF, as well as NFTs in the left hippocampus and tau aggregates in ERC‐connected brain regions, including V1, V4, retrosplenial cortex (RSC), hippocampus, and the ERC itself. Concurrently, evidence of neuroinflammation and neurodegeneration markers was observed. However, no significant changes, in CSF Aβ42 or Aβ40 levels were noted.

In Beckman's subsequent study involving intracortical injections of AAV1‐P301L/S320F into the left ERC of female rhesus macaques, a transition of tau pathology in neuronal populations from pretangle states (NEUN+/AT8+/ThioS−) to mature tangles (NEUN+/AT8+/ThioS+), accompanied by a reduction in the number of healthy neurons (fewer NEUN+ cells) ²⁵⁴ were observed over 6 months. Additionally, the study demonstrated activated microglia since the early stages of tau pathology, whereas astrocyte involvement occurred later, coinciding with neuronal loss. The study further highlighted the prion‐like propagation of exogenous human 4R‐tau, which co‐aggregated with endogenous rhesus monkey 3R‐tau during transneuronal spread, underscoring the translational relevance of this NHP model for studying human tauopathies.

NHPs serve as a critical tool for studying tau isoforms and phosphorylation during aging. They allow both cross‐sectional comparisons of similarities and differences across various monkey species and longitudinal biomarker analyses with lower confounding influences, providing a clearer perspective for research. In contrast, experimentally induced NHP models closely replicate human tau pathology, allowing researchers to track disease progression through biomarkers at different time points over shorter periods and support the groundwork for therapeutic development.

6.3. NHPs with both Aβ and tau pathology

6.3.1. Sporadic models

In sporadic rhesus monkeys, tau pathologies progress systematically with advancing age, ²⁶¹ beginning at 7–9 years with PKA‐phosphorylated tau (pSer214‐tau) in layer 2 of the ERC, also in dendrites and transporting endosomes, leading to dysregulated Ca2+ homeostasis. By their twenties, tau fibrils appear in ERC layer II, associated with cognitive deficits and matching Braak stage I/II. In their thirties, widespread fibrillar tau and mature NFTs manifest in layers 2 and 5, alongside progressive cognitive impairment, consistent with Braak stage III/IV. Additionally, amyloid pathology was noted in aged macaques over 30 years, characterized by Aβ1‐42 senile plaques within the ERC. Notably, microglial engulfment and lipofuscin observed in older monkeys indicate aging. The models mirror late‐onset AD, characterized by amyloid and tau pathology that closely replicates the sporadic progression of aggregate deposition observed in humans. A tau‐based staging framework aids cross‐species pathological comparison.

In aged cynomolgus monkeys exhibiting both amyloid and tau changes, CSF Aβ1‐42 decreases, and MRI detects atrophy in the ERC and hippocampus. ²⁵⁸ , ²⁵⁹ , ²⁶⁰ Post mortem brain analysis identified Aβ plaques, intracellular tauopathy, and granulovacuolar cytoplasmic changes, implying neurodegeneration.

In another study, aged cynomolgus monkeys with type 2 diabetes (T2DM) showed reduced plasma Aβ1‐40 and Aβ1‐42, elevated peripheral TNF‐α, and Aβ plaques in PFC, FC, TC, and ERC. ²⁵⁶ Tau pathology ranged from fibrils to mature tangles in the hippocampus, closely resembling the patterns observed in human AD. In addition to these findings, neuroinflammation and synaptic damage were also noted, although neurodegeneration was less prominent. These results highlight aged cynomolgus monkeys as preclinical AD models.

In vervet monkeys, Latimer's group identified smaller MRI‐measured brain volumes correlated with elevated CSF p‐tau181 levels and reduced gait speed with lower CSF Aβ1‐42 levels. ²⁵⁷ Additionally, Aβ plaque distribution was associated with reduced ¹⁸F‐FDG PET signals. These findings support vervets as preclinical AD models for biomarker studies and mechanistic explorations.

6.3.2. Injection‐based models

Diabetes increases the risk of AD by affecting the inflammatory response, protein metabolism, and cell death through disruptions in insulin signaling. ²⁷² , ²⁷³ , ²⁷⁴ In streptozotocin (STZ) ‐induced type 1 diabetes vervet monkeys, early AD‐associated pathologies included increased soluble Aβ1‐40 and Aβ1‐42 in the hippocampus and TC, reduced Aβ‐degrading enzymes, hyperphosphorylated tau fibrils, and elevated tau kinase activity. ²⁶⁵

In African green monkeys, chronic Aβ oligomer injections resulted in diffuse MTL amyloid plaques, accelerated tau pathology, activated microglia and astrocytes, and hippocampal atrophy detected by longitudinal MRI, paralleling cognitive decline by behavioral tests. These results position Aβ oligomers as a critical upstream driver of AD pathology, providing an experimental platform for exploring preventive interventions. ²⁶⁴

Injections of extracted human tau aggregates into the ERC and hippocampus successfully replicated prion‐like tau spread and Braak stage 1–4 pathology in primates, NFTs, and elevated CSF p‐tau181. Co‐administration of Aβ oligomers significantly accelerated these pathological changes, promoting severe NFTs, heightened t‐tau levels, and overt neurodegeneration. This highlights Aβ’s role in advancing tau progression, showing human‐derived pathological substrates retain toxicity in primate brains. ²⁶³

Zhouquan Jiang's group established a rapid and comprehensive primate AD model in rhesus monkeys via a single injection of AAV‐tau in the bilateral hippocampus. ²⁶² Within 2–3 months, models replicated key AD biomarkers: elevated CSF Aβ1‐42 levels, reduced Aβ40/42 ratios, increased p‐tau181, p‐tau231, and t‐tau, elevated CSF NfL, reduced glucose metabolism, and hippocampal atrophy observed by MRI. Histological analyses revealed widespread amyloid plaques, NFTs, and neuroinflammation, and behavioral assessments confirmed significant cognitive decline. Compared to earlier models, this model offers rapid, comprehensive AD pathology for translational research.

6.3.3. Transgenic models

Transgenic APP rhesus monkeys exhibit prominent AD‐related biomarkers by age 4: elevated CSF Aβ1‐42, and t‐tau, alongside diminished CSF p‐tau/Aβ42 ratios indicating early pathological changes. ²⁶⁸ By age 10, some display widespread CAA, plaque deposition, and microglial activation, similar to human AD but with earlier onset. Behavioral anomalies arise by 16 months, demonstrating the model's value for preclinical studies, preventative interventions, and disease‐modifying therapies.

PSEN1‐mutant cynomolgus monkeys, utilizing CRISPR/Cas9 to delete exon 9 of the PSEN1 gene, yield a live‐birth success rate of 50%. ²⁶⁶ This AD model recapitulates familial AD with key neuropathological biomarkers, including increased CSF Aβ42, p‐tau217, and plasma immune and inflammatory changes, strengthening the model's power in studying familial AD, evaluating early diagnostic biomarkers, and aiding therapeutic interventions.

In cynomolgus monkeys expressing the human MAPT gene with the pathogenic P301L mutation (0N4R P301L), tau promoted Aβ aggregation within the spinal cord, a phenomenon identified for the first time in this model. Models also exhibit accelerated tau aggregation, notable biomarker alterations, neuroinflammation, synaptic degeneration, and motor deficits, faithfully replicating critical aspects of AD. ²⁶⁷ This demonstrates a mechanistic tau‐Aβ link, offering a unique opportunity for the investigation of targeted therapies.

Transgenic NHP models of AD have significantly advanced our understanding of AD pathology and facilitated therapeutic innovations. These models accurately mimic human symptoms, providing a deeper understanding of the intricate and diverse nature of the disease. Consequently, they shed light on advancing the development of personalized and precision medicine approaches.

7. EXPLORING NHP MODELS: PROMISE AND CHALLENGES IN AD RESEARCH

7.1. Key questions for optimizing present NHP models

NHP models provide key advantages in AD research by bridging gaps left by rodent models, particularly regarding age‐related pathology onset, genetic gaps, and the reproducibility of AD phenotypes. To maximize their relevance, it is essential to thoroughly evaluate disease features in NHPs.

In natural aging models, pathological variability among individuals of comparable ages underscores the need to uncover underlying mechanisms. Characterizing key pathological markers, such as 3R and 4R tau isoforms, phosphorylation patterns, and Aβ structure, may help refine the models. Natural models may not fully mirror human disease progression due to species and genetic differences. Identifying optimal NHP species, like rhesus monkeys, and establishing species‐specific timelines for the earliest pathological changes are critical for model development. Induced models—using transgenic techniques or pathological material injection—require careful timing and duration to better simulate human‐like disease progression. Longitudinal studies tracking AD hallmarks—such as Aβ deposition and tau abnormalities—are crucial for replicating the decades‐long progression seen in humans. Recent advances in transgenic NHP models, including multigene modifications (APOE, Aβ, tau), show promise in improving disease fidelity. Chinese NHP cohorts have highlighted the importance of rigorous validation and reproducibility to ensure the reliability and broader applicability of findings.

Addressing these challenges may enhance the precision and utility of NHP models, supporting breakthroughs in understanding AD pathophysiology and expediting the development of effective therapies. Additionally, other models also warrant consideration, such as cholinergic nervous system injury models frequently employed to investigate the effects of damage to the cholinergic pathways and the STZ injection method in the context of studying diabetes and related neurodegenerative processes. ²⁷⁵ These models provide valuable insights into specific pathological mechanisms and contribute to a more comprehensive understanding of the underlying conditions.

7.2. Potential of integrating neuroinflammation biomarkers into NHP models

A growing body of recent studies has specifically evaluated GFAP and sTREM2 as key neuroinflammation biomarkers in AD. ¹⁶¹ , ¹⁶⁵ , ¹⁶⁶ , ¹⁹² Supporting these findings, NHP studies from Beckman's group have reported elevated levels of sTREM2 in the CSF. ²⁵⁴ , ²⁵⁵ Additionally, most NHP studies have observed increased numbers of IBA‐1+ microglia and GFAP+ astrocytes in brain regions affected by AD, further highlighting the role of neuroinflammation in disease pathology. ²⁴¹ , ²⁴³ , ²⁴⁶ , ²⁴⁸ , ²⁵⁰ , ²⁵⁴ , ²⁵⁵ , ²⁵⁶ , ²⁶² Importantly, clinical trials have begun to include neuroinflammation biomarkers in their therapeutic evaluation for reducing the cost and complement to traditional Aβ and tau measurements. ²⁷⁶ , ²⁷⁷ , ²⁷⁸ , ²⁷⁹ Despite advances in biomarker research, there still remains a significant gap in the assessment of neuroinflammation‐based biomarkers in NHP studies, with only two included studies reporting neuroinflammation‐related findings. ²⁵⁴ , ²⁵⁵ This limitation is likely due to the predominant focus on Aβ and tau pathology in most NHP research, which often overlooks the assessment of additional inflammatory markers. While some research has measured general inflammatory indicators such as TNF‐α, specific examination of CNS glial cells is still lacking. ²⁵⁶ This gap may, in part, reflect constraints associated with the disease stage represented in current models. This gap may, in part, reflect constraints associated with the disease stage represented in current models. To address these limitations, incorporating neuroinflammation biomarkers into NHP studies is crucial. This approach may help uncover novel disease mechanisms and improve the relevance of NHP models in human AD, highlighting their potential clinical value.

7.3. Current limitations to wider adoption of NHP models

Although NHP models are indispensable in AD research, they face significant challenges in faithfully replicating the complex pathology and cognitive symptoms seen in humans. In naturally aging NHPs, AD‐related pathological features are often only partially replicated, and these models rarely reproduce the full spectrum or progression of the disease, despite their high genetic similarity to humans. Additionally, behavioral signs of cognitive decline are inconsistently observed.

Genetically modified or experimentally induced NHPs—through the introduction of human AD‐related genes or pathological proteins—can better replicate specific pathological features and align more closely with human clinical biomarkers. However, such models still struggle to reproduce several key pathological characteristics in human AD, such as the expression of human‐specific tau isoforms, characteristic Aβ plaque distribution patterns, and the mechanisms of disease spreading within the brain. From a modeling perspective, although introducing key pathological factors can induce corresponding disease features under ideal conditions, technical challenges such as mosaicism during model creation, low breeding efficiency, and the inability to fully replicate human disease mutations significantly hinder the establishment of highly accurate disease models. ²⁸⁰ Currently, stricter ethical regulations, complex approval processes, ²⁸⁰ and high costs of model generation and maintenance due to longer lifespans further restrict large‐scale application, while the effectiveness of advanced gene‐editing technologies like CRISPR still requires large‐scale validation.

Addressing these challenges is essential for improving the reliability and translational relevance of NHP models. Advances such as employing gene‐editing technologies can more precisely recreate the human genetic context, while multifactorial induction strategies may produce composite models that better emulate the disease's pathological complexity. Furthermore, fostering interdisciplinary collaboration across neuroscience, genetics, and molecular biology will be crucial to refining NHP models, ultimately providing more robust and predictive tools for advancing AD research.

8. DISCUSSION

Biomarkers play a crucial role in the early diagnosis, staging, and therapeutic monitoring of AD. The adoption of the ATN framework has greatly advanced the clinical utility of biomarkers, enabling more precise early intervention. However, existing biomarkers and the ATN framework may insufficiently capture the complex, multifactorial nature of AD, especially in its earliest stages. Diagnostic sensitivity and specificity are still not optimal for all patient populations, and routine use remains limited by cost and accessibility. Building on this concept, this study proposes that neuroinflammation (X) might serve as a promising complement to the ATN(X) framework, as it is now recognized as a major pathological feature in AD and is closely intertwined with amyloid and tau pathology. Currently, neuroinflammatory biomarkers are less established due to their variability in detection techniques and limited longitudinal data. The complex relationship between neuroinflammation and AD core pathology still remains under‐explored, further inhibiting the broader use of neuroinflammation biomarkers. Further research should focus on standardizing neuroinflammation biomarker assays, establishing longitudinal correlations with disease progression, and integrating these markers in large‐scale clinical studies.

NHPs have gained increasing attention in recent years as promising animal models for AD research. Biomarker detection has demonstrated an increasing trend in the study of AD‐related NHP models. However, a standardized primate model equivalent to the widely used 5xFAD mouse model is still lacking. Optimizing pathological induction techniques and gene‐editing approaches may enhance reproducibility and efficiency. Given the evidence of neuroinflammation in monkey brain slices, future research should focus on addressing the complexity of AD by standardizing NHP models and incorporating neuroinflammatory biomarkers. This integration is crucial for capturing disease dynamics and improving model reliability.

In conclusion, we posit that biomarkers serve as a critical thread that links their applications in early diagnosis, therapeutic monitoring, and their utilization in NHP models. As our understanding of biomarkers and their applications in NHP models continues to expand, and as these models are further refined and developed, we anticipate that these combined efforts will significantly enhance our understanding of AD and provide a robust and effective strategy in the ongoing battle against this prolonged war.

AUTHOR CONTRIBUTIONS

Conceptualization: Zhihui Jin, Huifang Tang, and Huashun Cui. Data curation (100%): Zhihui Jin. Formal Analysis (100%): Zhihui Jin. Methodology (100%): Zhihui Jin. Visualization(100%): Zhihui Jin. Investigation(100%): Zhihui Jin. Writing—Original draft (100%): Zhihui Jin. Supervision: Yan Lu and Huashun Cui. Project administration: Yan Lu. Resources: Yan Lu. Writing‐Review & editing: Yan Lu, Huifang Tang, and Huashun Cui. Validation (100%): Huifang Tang. Funding acquisition: Huashun Cui.

CONFLICT OF INTEREST STATEMENT

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. All authors approved the final manuscript. Dr. Zhihui Jin (daughter of the co‐corresponding author Dr. Huashun Cui) is the sole first author. To ensure objectivity, we declare the following content: (1) Search strategy was audited by Prof.Huifang Tang; (2) Final inclusion was approved by all co‐authors with no family ties; (3) Data sources were exclusively public; (4) No industry/pharmacy funding supported this review. Dr. Huashun Cui recused herself from inclusion/exclusion of studies related to Traditional Chinese medicine (her expertise), and quality assessment (using NIA‐AA guidelines). Author disclosures are available in the supporting information.

CONSENT STATEMENT

The consent was not necessary.

Supporting information

Supporting Information

Supporting Information

Jin Z, Lu Y, Tang H, Cui H. Integrating neuroinflammation biomarkers into the ATN(X) framework: Advances in Alzheimer's pathogenesis, diagnosis, and insights from non‐human primate models. Alzheimer's Dement. 2025;21:e70472. 10.1002/alz.70472