Understanding Alzheimer’s Disease Through Neurodevelopment: Insights from Human Cerebral Organoids

Abstract

Alzheimer’s disease (AD) is a progressive neurodegenerative disorder and the leading cause of dementia, for which there is currently no cure. The causes of AD are still not well understood, although 5% of cases are known to have a genetic origin, associated with pathogenic genetic variants of the APP and PSEN1/2 genes. There is growing evidence that both APP and PSEN1/2 are also essential for proper human brain development and neural/neuronal function. This implies that abnormalities in early brain development could increase neuronal vulnerability to AD later in life. Human cerebral organoids (hCOs), generated from induced pluripotent stem cells (iPSCs) from AD patients, provide an exceptional model for better understanding the cellular and molecular mechanisms involved in human brain development, as well as early neurological alterations in the evolution of AD. This review compiles the main studies in which hCOs are used as a model for studying AD and for the discovery of new biomarkers. We also discuss the advantages and applications of these hCOs for studying the early stages of AD from a neurodevelopmental perspective. Finally, we mention the main current challenges in the use of hCOs for future research into AD.

1. Introduction

Alzheimer’s disease (AD) is a progressive neurodegenerative disorder and the most prevalent cause of dementia globally. It is characterized by a gradual deterioration in cognitive functions [1,2] and constitutes a substantial public health and socioeconomic burden. From a neuropathological standpoint, AD is defined by extracellular deposition of amyloid-β (Aβ) plaques, intracellular accumulation of hyperphosphorylated Tau protein, synaptic impairment, neuroinflammation, and neuronal loss [3]. Additional hallmark features include cytoskeletal abnormalities, most notably neurofibrillary tangles (NFTs), as well as widespread axonal degeneration [4,5].

As an age-related neurodegenerative disorder, AD is classified according to the age at onset into early-onset AD (EOAD, <65 years) and late-onset AD (LOAD, ≥65 years). Most patients develop LOAD, a multifactorial and heterogeneous disorder influenced by genetics, aging, environmental factors, lifestyle, and comorbidities such as obesity [6,7]. EOAD accounts for approximately 10% of all AD cases and typically manifests before 65 years of age. Among EOAD cases, only about 5% are attributed to pathogenic variants in the APP, PSEN1, or PSEN2 genes, which encode the amyloid precursor protein and presenilins 1 and 2, or to the APOE ε4 allele, whereas the underlying mechanisms remain unknown for the majority of patients.

This classification parallels the traditional distinction between familial AD (fAD) and sporadic AD (sAD). Familial AD, which often corresponds to EOAD, arises from autosomal dominant mutations in the APP, PSEN1, and PSEN2 genes that enhance amyloid-β production and aggregation [2,8]. In contrast, sporadic AD, typically overlapping with LOAD, constitutes over 90% of cases and is associated with multiple risk factors, including aging, APOE genotype, microglial dysfunction, and alterations in brain metabolism [6,7]. Despite these differences, both subtypes share common neuropathological features, such as Aβ plaques and Tau inclusions [9,10].

Although significant advances have been made in elucidating AD pathogenesis, the mechanisms that initiate and sustain neurodegeneration remain unclear, and current therapeutic interventions are unable to halt disease progression. Conventional pharmacological treatments, including acetylcholinesterase inhibitors and NMDA receptor antagonists, only offer symptomatic relief [11].

In recent years, disease-modifying anti-Aβ therapies have emerged. Lecanemab, a humanized IgG1 monoclonal antibody targeting soluble Aβ protofibrils, enhances their clearance and was approved by the FDA in 2023 for patients with mild or moderate cognitive impairment [12]. Although generally well tolerated, it can induce amyloid-related imaging abnormalities (ARIA), particularly in APOE ε4 homozygous individuals or those with a history of cerebral microhemorrhages [13,14]. Similarly, Donanemab, another recently approved IgG1 monoclonal antibody, binds the truncated pyroglutamate form of Aβ at position 3 (pGlu3-Aβ) and demonstrates comparable efficacy and safety outcomes [15,16]. Despite representing a significant milestone toward disease-modifying therapy, the overall clinical benefit of these antibodies remains limited, underscoring the need for more predictive and physiologically relevant preclinical models that can recapitulate the complexity of human AD pathology.

Animal models carrying human AD-associated mutations have been widely used to investigate disease mechanisms and evaluate potential therapeutic strategies. However, these models often fail to reproduce the full complexity of human pathology. Most of them represent familial rather than sporadic AD and display substantial anatomical and physiological differences from the human brain, including cortical organization and ventricular zone architecture [5,17,18,19,20,21]. To address these limitations, human cell-based systems have been developed. Two-dimensional (2D) cultures derived from patient-specific induced pluripotent stem cells (iPSCs) have provided valuable insights into AD pathology [22,23,24]. Nevertheless, 2D models lack the structural and cellular complexity of the human brain and cannot fully reproduce the dynamic cell–cell interactions that occur in vivo [25,26].

In this context, human cerebral organoids (hCOs) have emerged as a promising alternative. These three-dimensional (3D) structures more accurately reproduce the cytoarchitecture, cellular diversity, and developmental processes of the human brain than conventional in vitro systems. hCOs provide an innovative platform for investigating early- and late-onset forms of AD by overcoming the species-specificity limitations of animal models and the lack of tissue architecture in 2D cultures. Their application holds significant potential to deepen our understanding of AD mechanisms and to support the development of more effective and predictive therapeutic strategies.

2. Genetics of Familial (Early Onset) Alzheimer’s Disease

Among the most prevalent and well-characterized genes implicated in fAD are the amyloid precursor protein (APP), presenilin 1 (PSEN1), and presenilin 2 (PSEN2).

2.1. Amyloid Precursor Protein (APP)

APP is a highly conserved type I transmembrane protein encoded by the APP human gene located on chromosome 21 (Hsa21). It undergoes several alternative splicing events, resulting in three major isoforms (APP695, APP751 and APP770). In the central nervous system (CNS), the predominantly expressed isoform is APP695.

Structurally, APP comprises three functionally distinct domains: an N-terminal ectodomain, a single transmembrane domain, and a cytoplasmic C-terminal domain [27]. The N-terminal ectodomain serves as an extracellular receptor capable of binding multiple ligands from the Wnt family [28], netrin-1 [29], F-spondin [30], and BRI3 [31], while functioning as a source of bioactive processed products sAPPα and Aβ [32,33]. The transmembrane domain is essential for proper membrane orientation and stability, mediating specific interactions with α-secretase (ADAM10) [34] and constituting the direct binding site for γ-secretase. The cytoplasmic C-terminal domain, whose γ-secretase-cleaved fragment (AICD) regulates transcription and APP processing through interactions with adaptor proteins such as Fe65 and X11 [35], functions analogously to the Notch intracellular domain.

2.1.1. Amyloid Precursor Protein (APP) Pathogenic Variants and Processing in Alzheimer’s Disease

APP processing occurs via two mutually exclusive pathways. In the non-amyloidogenic pathway, α-secretase cleavage releases sAPPα and precludes Aβ generation. In contrast, the amyloidogenic pathway begins with β-secretase cleavage, producing sAPPβ; the remaining C-terminal fragment is subsequently processed by γ-secretase to release Aβ peptide. The AICD generated from this latter pathway can translocate to the nucleus to regulate the transcription of different target genes such as BACE1, ADAM10, and GSK3B [36]. The diverse metabolites generated through these pathways, including sAPPα, sAPPβ, Aβ peptides and AICD, exert different biological functions determined by their structural properties, equilibrium, and concentration relative to other APP-derived fragments [37,38].

To date, over 110 mutations have been identified in the APP gene, predominantly localized in coding regions. These regions correspond to the DNA sequences that are transcribed and translated into proteins; therefore, mutations located here often result in changes to the amino acid sequence. In the context of APP, over 25 of these mutations are classified as pathogenic and directly linked to Alzheimer’s disease [39]. Mechanistically, these coding variants alter the proteolytic processing of the APP, favoring the production and aggregation of neurotoxic Aβ peptides, a specific pathological hallmark that distinguishes AD from other neurodegenerative diseases.

Pathogenic variants cluster mainly in exons 16 and 17, proximal to the Aβ sequence or secretase cleavage sites, with family studies conclusively demonstrating that overexpression of the normal APP sequence (trisomy 21 or APP duplication) or mutations leading to elevated total Aβ, elevated Aβ42, or increased Aβ aggregation result in dementia and AD neuropathology [40].

Key pathogenic examples include the London (V717I) mutation near the γ-secretase site, which increases the Aβ42/Aβ40 ratio [41], and the Swedish (KM670/671NL) mutation in the N-terminal Aβ region, enhancing total Aβ via altered β-secretase cleavage site [42]. The Arctic (E693G) and Flemish (A692G) variants impact α-secretase processing: Arctic reduces plasma Aβ40/Aβ42 levels but boosts protofibril formation [43], while Flemish slows fibril aggregation despite doubling Aβ40/Aβ42 output [44]. The Osaka (E693D) mutation accelerates toxic soluble oligomer formation, inhibits fibrils, and promotes intraneuronal Aβ accumulation [45]. Notably, one APP variant provides exceptional insight into protective mechanisms: the A673T mutation, identified in the Icelandic population and located near the BACE1 site, exerts a protective effect against AD and age-related cognitive decline by reducing Aβ production [46].

2.1.2. Importance of APP for Neural Development and Function

APP is highly conserved and widely expressed during nervous system development, although its physiological function, especially in humans, is still unclear. The APP expression in the brain is critical for many biological processes including neurite outgrowth [47], synaptogenesis [48], neuronal survival [49], axonal growth after injury, the proper functioning of the endolysosomal pathway [50], cell signaling, brain development and plasticity [36,51].

Previous studies on neurogenesis suggest that APP may also play a subtle role in cortical neurogenesis in mice, although the precise nature of this role is unclear [52,53,54,55]. APP is highly expressed in human telencephalic neurospheres and during cortical neuron differentiation and migration [56], suggesting that it may be involved in neural stem cell proliferation, differentiation, and/or maturation [57].

More recently, some authors have found that APP plays a decisive role in maintaining human neural progenitors during the neurogenesis process, whereas APP is dispensable in mouse neural progenitors, where neurogenesis occurs more rapidly. In this sense, APP would contribute autonomously to neurogenesis by suppressing the transcription factor proneurogenic AP-1 and facilitating canonical WNT signaling. The authors propose that APP acts by regulating the balance between self-renewal and differentiation, which may contribute to the temporal patterns of neurogenesis specific to humans [58,59]. These observations suggest that human brain development may be particularly sensitive to alterations in APP levels or its loss. Therefore, it can be speculated that APP mutations in fAD could affect normal cortical development, which could cause reduced stress resistance in neurons and glial cells throughout life, contributing to premature neurodegeneration [59].

2.2. Presenilin 1 (PSEN1) and Presenilin 2 (PSEN2) Function and Pathogenic Variants in Alzheimer’s Disease

PSEN1 (chromosome 14) and PSEN2 (chromosome 2) are homologous genes encoding the catalytic subunits of the γ-secretase complex, a membrane-bound heterotetramer essential for APP cleavage and Aβ generation, as well as proteolysis of other substrates like Notch, critical for brain development [60]. Both presenilins are polytopic transmembrane proteins organized into nine helical domains, with transmembrane domains 6 and 7 containing catalytic activity [61,62]. Despite structural similarities, PSEN1 and PSEN2 differ in enzymatic efficiency, with PSEN2 being generally less efficient, and subcellular localization, with PSEN1 primarily at the plasma membrane and PSEN2 enriched in endosomal compartments [63,64]. These differences, combined with their roles in neurogenesis and neural progenitor maintenance, have clinical relevance for understanding how mutations impact AD pathology.

PSEN1 is the most common causal factor in early-onset familial Alzheimer’s disease, with over 300 mutations identified to date [65]. Patients develop aggressive disease with earlier onset and atypical manifestations including motor dysfunction [66]. PSEN1 mutations are also associated with Parkinson’s disease, Lewy body dementia, and frontotemporal dementia [67,68,69]. Most PSEN1 mutations increase Aβ42/Aβ40 ratio through gain-of-function effects, although some reduce γ-secretase activity; however, both mechanisms disrupt APP cleavage favoring Aβ42 release and impair processing of other substrates like Notch, affecting neuronal homeostasis [70].

PSEN2 mutations are less frequent, with only 91 reported and many lacking functional evidence [65]. These mutations tend to produce milder clinical phenotypes with incomplete penetrance and later disease onset [71]. Similar to PSEN1, some PSEN2 mutations increase Aβ42/Aβ40 ratio, though effects are typically smaller, possibly due to lower γ-secretase processivity or preferential expression in microglia; for example, N141I increases Aβ42/Aβ40 primarily through Aβ40 reduction rather than Aβ42 elevation [65].

Importance of PSEN1/2 for Neural Development and Function

There is growing evidence that the genes involved in the development of fAD, like PSENs, also play an essential role in the proper mammalian brain development and neural/neuronal function. This implies that abnormalities in early brain development could increase neuronal vulnerability to AD later in life.

Presenilins (PSEN1 and PSEN2) are essential components of γ-secretase, an aspartate protease that cleaves type I transmembrane proteins, including APP and Notch [72]. It has been observed that the ventricular zone of PSEN1^−/− mouse brains is markedly thinner at embryonic day (E) 14.5, indicating impaired neurogenesis as compared to controls [73]. Another report suggests that the deficiency of PSEN1 in mice causes the loss of Cajal-Retzius neurons and cortical hyperplasia, similar to human lissencephaly [74].

Moreover, recent studies have shown that PSEN1 mutations cause age-dependent neurodegeneration via an Aβ-independent mechanism. This was demonstrated by breeding PSEN mutant mice with mice lacking the APP gene (an ‘APP-null background’). This strategy eliminates the substrate required for Aβ production. The fact that neurodegeneration still occurred in these APP-deficient models suggests a pathogenic mechanism caused by a loss of PSEN function, rather than amyloid accumulation [75].

The development of conditional double knockout (cDKO) PSEN mice demonstrated that selective inactivation of PSEN in excitatory or inhibitory neurons leads to age-dependent brain atrophy, inflammatory responses and accumulation of pathological Tau in neurons and glial cells. Memory and synaptic alterations were also observed in these animals, as well as the presence of a marked age-dependent neurodegeneration in the cerebral cortex [73,76]. All these data are raising the possibility that PSEN mutations may cause fAD through a loss-of-function mechanism.

In summary, the requirement for PSEN1 and PSEN2 in Notch receptor activation and APP processing clearly implies that these proteins play an important role in various stages of brain development, from neurogenesis to neuronal migration and axonal and dendritic arborization [77].

PSEN1 mutations cause most cases of familial AD, possibly by disrupting proper Notch signaling, leading to early cellular changes that go unnoticed and affect the subsequent progression of AD. Individual PSEN1 mutations may differentially affect neurodevelopment and may give insight into fAD progression to provide earlier time points for more effective treatments.

3. Human Cerebral Organoids

Traditional 2D stem cell cultures have long been the standard in biomedical research and have significantly contributed to our biological understanding. However, in the context of neurodegenerative diseases such as AD, these models fall short: they lack the tissue-like structure, cellular organization, architecture, and developmental trajectory observed in vivo, which limits their ability to reproduce key pathological events such as the spatial and temporal progression of the disease or the aggregation of extracellular peptides [78,79]. In contrast, the emergence of 3D models (particularly cerebral organoids) made possible by advances in stem cell technology represents a major step forward. Given the complexity and inaccessibility of the human brain, especially during neurodevelopment, these organoid-based approaches offer a more faithful approximation of in vivo conditions and open new avenues for understanding the mechanisms underlying AD [25,26].

The term “organoid” refers to 3D cell aggregates derived from primary adult tissues or stem cells that can self-organize into organ-like structures in vitro. They are defined by their composition of multiple relevant cell types, their ability to reproduce organ-specific functions, and their structural organization resembling that of the native organ [80].

To generate hCOs, two categories of pluripotent stem cells (PSCs) can be used: embryonic stem cells (ESCs), which are obtained from the inner cell mass of the blastocyst, and induced pluripotent stem cells (iPSCs), which originate from the reprogramming of somatic cells through the introduction of transcription factors [81]. Unlike many other types of organoids that can be derived from adult tissue stem cells (AdSCs), hCOs are produced solely from PSCs, largely due to the low regenerative capacity of the adult brain [82].

In terms of existing methodologies for hCOs generation, two main methodological frameworks have emerged: unguided protocols and guided protocols. On the one hand, unguided approaches rely on the intrinsic self-organization capacity of PSCs, allowing organoids to develop spontaneously in the absence of exogenous patterning cues. As a result, a single hCOs commonly contains multiple interrelated brain regions. This strategy was the basis for the earliest hCOs models, which recapitulated diverse neuroepithelial domains within the same structure [83,84,85]. In contrast, guided approaches introduce defined morphogenetic or patterning factors into the culture media to steer differentiation toward a specific regional identity. Using this approach, researchers have developed organoids that faithfully model discrete brain territories (such as the cerebral cortex, midbrain, or hypothalamus) by directing early fate decisions during hCOs formation [86,87,88,89].

The versatility of organoid technology has expanded significantly with the development of region-specific protocols. Current guided methods have successfully generated organoids representing the entire CNS axis, ranging from forebrain, midbrain, and hindbrain structures to specific sub-regions like the hippocampus, thalamus, pituitary, and spinal cord, each recapitulating key cellular and functional features of their in vivo counterparts [81]. Moreover, addressing the cellular heterogeneity of the brain is crucial. Recent protocols have incorporated non-neuronal lineages, particularly microglia, to mimic the neuroimmune environment [90].

Another new strategy for generating hCOs has been described by our research group [91,92]. In this approach, embryoid bodies (EBs) are not forcibly aggregated (as occurs in guided and unguided protocols); instead, neural induction is performed directly on PSCs in culture, transitioning from a 2D to a 3D model, resulting in hCOs with proper spatial organization and high batch-to-batch homogeneity.

The advancements of 3D cultures and patient-derived hCOs are enabling the modelling of neurological diseases, offering a way to surpass the limitations of 2D cultures and animal models, which often do not accurately represent the human condition [93]. In this regard, the emergence of hCOs technology, pioneered by Lancaster and colleagues in 2013 [84], coupled with the possibility of producing hCOs in most laboratories, has created interest and considerable expectations. Given that AD represents the most prevalent neurodegenerative disease of the central nervous system, considerable efforts have been devoted to employing hCOs models to investigate this condition [94]. Below, we describe the most notable and recent works in relation to this, focusing on the modeling of familial Alzheimer’s disease (fAD) using hCOs (derived from hiPSCs obtained from patients with fAD).

4. Human Cerebral Organoids (hCOs) as a Model to Study Alzheimer’s Disease

This section illustrates how hCOs can faithfully model some of the pathological features characteristic of human AD brain (Figure 1). Through the research approach using hCOs, we can systematically examine early pathogenic mechanisms, identify cell-type-specific vulnerabilities to AD processes, and evaluate therapeutic interventions.

hCOs provide an especially appropriate system to investigate the mechanisms of fAD. In a landmark study, Raja et al. [95] generated the first hCOs from fAD patient-derived iPSCs carrying PSEN1 mutations or APP duplication, demonstrating that patient-specific mutations alone are sufficient to induce hallmark AD pathologies in 3D human neural tissue, including increased Aβ aggregation, elevated Aβ42, and mutation-dependent shifts in the Aβ42/Aβ40 ratio. Subsequent studies have validated the robustness of hCOs for modeling core fAD features, particularly Tau hyperphosphorylation [96,97].

One of the most encouraging applications of hCOs in AD research has been the recapitulation of early Tau pathology. hCOs carrying fAD mutations display robust p-Tau accumulation across multiple phosphorylation sites (p-T217, pT181, pT231) (Figure 2), critical for understanding early molecular events in fAD. hCO models enable simultaneous characterization of the AD pathology (amyloid-β deposition, tau phosphorylation, glial activation, and neuronal stress responses) through methodologies including immunostaining, proteomics, and quantitative immunoassays, providing access to early disease mechanisms in a human context (Figure 2).

Beyond these canonical hallmarks, hCOs reveal broader neuronal and synaptic alterations associated with fAD. Reported phenotypes include changes in neuronal morphology [96], dysregulation of synaptic proteins [98] and disturbances in excitatory/inhibitory balance, characterized by increased VGLUT1 and reduced VGAT [99]. Mutations also lead to endosomal dysfunction [95], upregulation of pro-inflammatory cytokines [100], and epigenetic dysregulation, including increased miRNA125b and decreased 5-hydroxymethylcytosine [101]. Notably, fAD-associated traits tend to be more severe compared to single-mutation lines when hCOs are generated from pluripotent stem cells harbouring multiple pathogenic mutations [102].

Human cerebral organoids are also well suited to interrogate the developmental impact of presenilin mutations. Recent findings have observed altered neurogenesis in iPSCs with early-onset fAD harboring PSEN1 mutations compared to non-isogenic controls using 2D cortical differentiation and 3D cerebral organoid generation [103]. Curiously, these cerebral organoids also exhibit developmental and tissue patterning defects, supported by mRNA single-cell sequencing data showing premature neuronal differentiation [104]. Specifically, the PSEN1 L435F mutation affects human cortical spheroid growth and morphology, increasing cortical progenitors and reducing neuron differentiation. Interestingly, specific Notch1 inhibition rescues cellular phenotypes [105].

Emerging evidence further indicates that fAD mutations may trigger pathology from early developmental stages: neural precursor cells carrying fAD variants exhibit increased cellular stress and an imbalance in the production of excitatory versus inhibitory neurons [106]. This aberrant neurogenesis results in reduced signalling in mature neurons and impaired synaptic function, suggesting that molecular features of AD may emerge at very early stages, already during neuronal differentiation [107].

Although PSEN2 mutations are less common [108], hCOs derived from PSEN2-mutant iPSCs underscore the direct pathogenic consequences of presenilin dysfunction. These organoids are typically smaller, show elevated Aβ42/Aβ40 ratios and increased p-Tau, and display asynchronous calcium transients together with heightened neuronal activity [109,110]. They also show increased caspase-3, consistent with greater apoptotic susceptibility [110]. Additional studies report altered cytokine secretion, dysfunctions in microglia and astrocytes [111] and altered TREM2 levels indicative of dysregulated microglial activation [112].

Collectively, these findings position hCOs as a versatile, human-relevant platform for investigating fAD. They not only recapitulate patient-specific phenotypes but also provide a system for therapeutic screening. hCO-based studies have identified compounds that reduce Aβ pathology [95,96], attenuate inflammatory signatures [100], and rescue mitochondrial dysfunction, through diverse mechanisms, including NeuroD1-mediated mitochondrial support [98], autophagy activation [107,113] and modulation of synaptic receptors, with agents such as NitroSynapsin [114]. By enabling precise dissection of the molecular and developmental consequences of APP, PSEN1, and PSEN2 mutations hCOs help identify therapeutically relevant pathways with direct implications for both familial and sporadic AD research.

While findings highlight the utility of the fAD hCO model, it is important to consider its limitations in the context of available AD models. Unlike transgenic animal models, which offer a systemic environment including vasculature, immune response, and behavioral output, hCOs currently lack non-neural lineages (unless co-cultured) and full anatomical inter-regional connectivity. However, hCOs overcome critical hurdles associated with murine models, specifically the species barrier. Transgenic mice often require the overexpression of multiple mutations, such as the combined expression of mutant APP, PSEN1, and MAPT required in the 3xTg-AD model, to recapitulate both amyloid and tau pathology [115]. In contrast, fAD hCOs provide a human-specific cellular context where key hallmarks emerge spontaneously. Thus, fAD hCOs serve as a powerful reductionist model for dissecting the early molecular and cellular initiation of AD.

5. Modeling the Developmental Origin and Molecular Mechanisms of Alzheimer’s Disease Using hCOs

hCOs faithfully reproduce the cellular complexity of the developing human cortex [107], comprising diverse neural progenitors, excitatory and inhibitory neurons, astrocytes, radial glia, and organized proliferative zones [91,116,117]. While postmortem brain tissue reveals the end stage of AD, hCOs offer a complementary approach to study the earliest stages. Their value lies in revealing how developmental defects, such as aberrant signaling or progenitor dysfunction, may establish the developmental origin of later neurodegeneration.

fAD hCOs allow for the investigation of specific developmental pathways that may prime the brain for later pathology. For instance, beyond amyloid accumulation, PSEN1 mutations have been shown to drive reduced Notch signaling and aberrant Wnt pathway activity. This disruption leads to the premature differentiation of Neural Progenitor Cells (NPCs) into neurons, causing an early depletion of the neural stem cell pool and potentially compromising the neurogenic reserve long before clinical onset [103]. Similarly, the Amyloid Precursor Protein (APP) itself plays a critical role in regulating cell fate during development. Studies utilizing human neural stem cells have demonstrated that APP levels orchestrate the balance between neurogenesis and gliogenesis; specifically, alterations in APP expression can skew differentiation trajectories, further supporting the hypothesis that AD pathology is rooted in early neurodevelopmental deviations [118].

Other structural changes in cell organelles can be studied, including disruptions in ion channels and synaptic pathways [119], mitochondrial dysfunction from impaired mitophagy and elevated reactive oxygen species [120,121], endoplasmic reticulum stress due to misfolded proteins [122], and Golgi apparatus fragmentation.

Intercellular communication mediated by exosomes propagates pathological signals in AD [123]. Transcriptomics and proteomics analysis of secretomes from fAD hCOs enables identification and characterization of not only exosomes but also other extracellular vesicles implicated in AD pathology propagation. This approach facilitates the discovery of predictive biomarkers for the transition to neurodegeneration, which can be correlated with those found in patient cerebrospinal fluid (CSF) and plasma, thereby linking early molecular changes to amyloid-Tau pathology. Clinical biomarkers such as p-Tau217, p-Tau231, BACE1, VILIP-1, neurogranin, and neurofilament light (NF-L) reflect neurobiological processes that are evaluable in this model [124,125]. Multi-omics biomarkers identified in hCOs can be further validated against established clinical and neuropathological resources, including the Alzheimer’s Disease Neuroimaging Initiative (ADNI), BioFINDER cohort, Single-cell and Spatial RNA-seq Atlas for Alzheimer’s Disease (ssREAD), and postmortem brain data from the National Centralized Repository for Alzheimer’s and Related Dementias (NCRAD) and National Alzheimer’s Coordinating Center (NACC).

AD hCOs provide a controllable platform for AD drug high-content screening, facilitating the identification of effective therapies, validation of therapeutic targets, and optimization of interventions prior to clinical trials. Park et al. [126] established an hCO platform evaluating key parameters such as cell viability, pathological protein aggregation, and neuronal alterations, with a focus on FDA-approved agents capable of crossing the blood–brain barrier (BBB). Recent studies demonstrate that lecanemab, an FDA-approved anti-amyloid antibody, significantly reduces amyloid burden and modulates neuroinflammation in hCOs [127].

Beyond therapeutics, hCOs offer a robust system for evaluating environmental neurotoxicity, investigating how stressors interact with genetic predisposition to modulate neurodegeneration, such as exposure to particulate matter, chemical toxins, heavy metals, nanoplastics, and persistent organic pollutants. Recent studies demonstrate that hCOs exposed to cadmium exhibit severe disruptions in neurodevelopmental patterns [128].

Major research applications employing AD (fAD and sAD) hCOs are summarized in

Table 1. Summary of relevant hCO studies modeling AD pathology, including molecular mechanisms, biomarkers, validation strategies, and therapeutic approaches.

First Author (Year)	Genetic Background	AD Relevance	Molecular Mechanisms	Biomarkers Detected	CSF/Plasma/ Tissue Validation	Drug Tested
Raja (2016) [95]	fAD	Proof-of-concept AD modeling	Amyloid accumulation, Tau hyperphosphorylation, endosomal abnormalities	Aβ42, phospho-Tau, endosomal markers	Postmortem fAD brain	β- and γ-secretase inhibitors
Gonzalez (2018) [129]	fAD	Hallmark pathology	Aβ aggregation and endosome abnormalities	Aβ species (oligomers, fibrils), phospho-Tau, APP-CTF fragments	Postmortem fAD brain	β- and γ-secretase inhibitors
Park (2021) [126]	fAD + sAD (mixed patients)	Multi-mutation comparison	Aβ production, Tau phosphorylation, endosomal abnormalities	fAD-specific proteomic profiles	Clinical cohorts	1300 FDA-approved compounds
Arber (2021) [103]	fAD (PSEN1 mutations)	Neurodevelopmental origin of AD	Premature neurogenesis, disrupted neural development	Cleaved Notch (NICD), FABP7 expression, Tau hyperphosphorylation	Postmortem fAD brain	Notch pathway modulation
Choe (2024) [102]	sAD patient- derived	Individual trajectory modeling	Patient-specific proteomic heterogeneity, Aβ/Tau pathology	scRNA-seq and proteomics (patient-specific responses)	scRNA-seq and proteomics correlation	Compound screening (patient-specific responses)
Ji (2025) [127]	sAD (brain extracts)	Intermediate AD stages	Aβ/Tau aggregates, neuroinflammation, microglial pruning, synaptic/neuronal loss, impaired network activity	Pro-inflammatory cytokines, synaptic markers and axonal (SNAP-25, SYT1) markers	CSF biomarkers in clinical literature	Lecanemab
Zeng (2025) [107]	fAD (APP mutations)	Neurodevelopmental origin of AD	Decreased mature neurons, increased cell senescence, elevated Aβ production, reduced neurogenesis	TMSB4X (thymosin β4) downregulation, elevated Aβ, increased cleaved caspase-3	Postmortem AD patient neurons and 5xfAD mice	Thymosin β4
Labra (2025) [106]	fAD (PSEN1 M146V, APP Swe, PSEN1 ΔE9)	Early pathological phenotypes	Synapse loss, neuronal hyperexcitability, excitatory/inhibitory imbalance, failure autophagy, cellular stress, synaptic dysfunction	pT217 and pT181. p21 (CDKN1A), p16 (CDKN2A), SASP. LC3-II, p62	CSF/Plasma biomarkers in clinical literature	mTor pathway: inhibitor: CCT020312 controls: Rapamycin Torin 1

. The table highlights the specific molecular mechanisms identified, relevant biomarkers detected in hCO secretomes, the validation approaches applied (including correlation with postmortem brain, CSF, or plasma data, as well as with datasets from the literature or patient databases), and the therapeutic interventions tested, encompassing FDA-approved drugs and clinical-stage antibodies. sAD represents a multifactorial syndrome often characterized by polygenic risk factors, which involves complex interactions between aging, immunity, and metabolism that fAD hCO models may not fully recapitulate. However, the results obtained using fAD hCOs offer valuable insights into specific mechanisms related to Aβ toxicity and tau phosphorylation, which are integral parts of the full AD spectrum.

While hCOs do not fully recapitulate the macroscopic spatiotemporal propagation of AD pathology (e.g., Braak staging) owing to the absence of complex inter-regional connectivity, they serve as a robust model for the initiation and accumulation phases of the disease. Specifically, the fAD hCO model mirrors the temporal progression of molecular pathology consistent with the early-to-intermediate cellular phases. Characterized by intracellular stress and synaptic dysfunction, this timeline parallels the biochemical sequence underlying the transition from preclinical cellular stress to overt neurodegeneration [95].

6. Challenges for Future Research with hCOs in the Study of Alzheimer’s Disease

In general, hCOs have some limitations that must be addressed methodologically related to the need for a rigorous standardization, robust quality control systems, multiple independent hiPSC lines for biological replication, and in the case of genome editing, identical genetic backgrounds. Ensuring the reproducibility of hCO models requires the implementation of strict controls at the morphological, cellular, and molecular levels. Regarding experimental design, it is recommended to include multiple differentiation batches derived from distinct iPSC clones to account for intrinsic cell line variability. Additionally, the availability of standardized commercial kits offers a strategy to minimize the variability. To mitigate technical noise, researchers are increasingly adopting methods to standardize oxygen and nutrient diffusion, thereby reducing the variability associated with necrotic cores. Furthermore, the validation of homogeneity goes beyond visual inspection; methods such as single-cell RNA sequencing (scRNA-seq) are increasingly used to confirm that the ratio of cell types remains consistent across replicates.

Standard protocols for generating hCOs fail to recapitulate key brain components, including vascular networks, BBB and functional microglia.

The absence of vasculature in hCOs restricts their size, impairs neuronal maturation, and induces hypoxia in core regions, leading to cellular stress and necrosis; to address this, several strategies have been developed, including co-culture with endothelial cells (e.g., HUVECs), incorporation of biomaterials such as Matrigel or hydrogels, supplementation with growth factors like VEGF and FGF, transplantation into immunodeficient mouse brains for host vessel ingrowth, and advanced approaches like microfluidic systems or genetic modifications (e.g., ETV2 overexpression) to promote angiogenesis and blood–brain barrier-like properties [130,131,132,133].

Microglia, the brain’s key immune cells, actively participate in neuronal development, synapse formation and pruning, and plasticity, with their dysfunction implicated in neurological diseases; however, they are absent or poorly represented in standard hCOs due to their mesodermal origin. Strategies to incorporate functional microglia include co-culture with hiPSC-derived microglia or myeloid precursors, and overexpression of the transcription factor PU.1 [90,134], which enhances neural maturation, reduces stress, and promotes network synchronization.

hCOs exhibit transcriptomic profiles with limited capacity to recapitulate age-dependent mechanisms critical for late-onset AD, as maturation criteria (layered cytoarchitecture, synaptic connectivity, electrophysiological activity) remain incomplete even after extended culture. Methods to enhance maturity and physiological relevance include spinning bioreactors for improved oxygen/nutrient diffusion and microfluidic devices that mitigate core hypoxia and promote network synchrony [83,133].

Standard cortical protocols for hCOs yield inefficient oligodendrocyte differentiation, resulting in insufficient myelination levels; to overcome this, specific strategies supplement culture media with factors promoting myelinization such as PDGF-AA, T3, and cAMP [135].

hCOs lack the spatial organization and interregional interactions characteristic of the human brain; to address this, assembloids, fusing hCOs from specific regions, enable modeling of cortical-subcortical circuits and connectivity.

Ultimately, the full potential of hCOs will likely be realized by their integration with emerging technologies. While bioengineering approaches like 3D bioprinting are advancing, the synergy between hCOs and microfluidic “organ-on-a-chip” platforms offers a particularly promising avenue. These systems facilitate the development of a functional BBB and improve nutrient exchange, mitigating core hypoxia and enhancing physiological relevance [136,137]. Finally, to transition hCOs from research models to industrial tools, there is a critical need to incorporate AI-driven high-content screening. By leveraging machine learning algorithms to analyze complex 3D phenotypic data, such as subtle morphological changes or neurite network complexity, it will be possible to scale up hCO applications for high-throughput drug discovery, accelerating the development of the next generation of neurological therapeutics [138].