Pushing the boundaries of brain organoids to study Alzheimer’s disease

Abstract

Progression of Alzheimer’s disease (AD) entails deterioration or aberrant function of multiple brain cell types, eventually leading to neurodegeneration and cognitive decline. Defining how complex cell-cell interactions become dysregulated in AD requires novel human cell-based in vitro platforms that could recapitulate the intricate cytoarchitecture and cell diversity of the human brain. Brain organoids (BOs) are three-dimensional self-organizing tissues that partially resemble the human brain architecture and can recapitulate AD-relevant pathology. In this Review, we highlight the versatile applications of different types of BOs to model AD pathogenesis, including amyloid-β and tau aggregation, neuroinflammation, myelin breakdown, vascular dysfunction, and other phenotypes, as well as to accelerate therapeutic development for AD.

Brain organoids recapitulate cellular diversity of the human brain

A groundbreaking development of the induced pluripotent stem cell (iPSC) technology has enabled disease modeling in vitro using patient-derived cellular models [1–3]. Human iPSC-based disease models overcome the difficulty of obtaining primary human tissues, such as brain tissue, as well as the issue of species divergence when using animal models [4, 5]. Although two-dimensional (2D) cell culture models have dominated in vitro research, more complex cellular assemblies are required to faithfully recapitulate multifaceted pathogenesis of Alzheimer’s disease (AD). The emergence of human organoid technology bridges the gap between 2D models and the complex three-dimensional (3D) environment in vivo. Brain organoids (BOs) are 3D stem cell-derived tissues that mimic the cellular composition and structure of the human brain [6–9]. BOs can be engineered to contain the diversity of human brain cell types, including neurons, astrocytes, oligodendrocytes, microglia, pericytes, and epithelial cells (see Box 1). BOs also exhibit complex cytoarchitecture and electrophysiological properties, facilitating the study of neurodegeneration and cognitive decline (see Box 2).

Box 1: Techniques for brain organoid differentiation.

Numerous BO differentiation protocols to derive neural structures of the desired brain region and incorporate different cellular components have been developed [9, 103–105]. Generally, pluripotent stem cells are first aggregated into embryoid bodies and induced towards the neuroectoderm lineage [39]. Signaling molecules can be added to promote guided differentiation, patterning the BO towards a specified regional identity; alternatively, unguided differentiation yields neural organoids that contain patches of tissue of various brain regional identities [39, 105].

Enrichment of BOs with glial cells necessitates cell type-specific strategies, given the variable developmental timing and ontogeny of glial cells [104]. Astrocytes arise only after the neurogenic-to-gliogenic transition of radial glia in the developing human brain [104]. Similarly, astrocytes spontaneously arise following long-term culture of BOs [62, 106]. Given the protracted nature of human brain myelination, additional signaling cues are required to induce oligodendrocyte progenitor cell (OPC) and oligodendrocyte development in BOs [23]. Supplementation of signaling molecules required for oligodendrogenesis, including platelet-derived growth factor AA (PDGF-AA), insulin-like growth factor 1 (IGF-1), and thyroid hormone (T3), promotes OPC and oligodendrocyte differentiation as well as neuron myelination in specialized BOs termed myelinoids [66–68]. Alternatively, ectopic expression of the oligodendrocyte transcription factor 2 (OLIG2) promotes directed differentiation of OPCs and oligodendrocytes in the myelinoid [65]. Myelination can further be improved by patterning the myelinoid identity towards the ventral spinal cord, where oligodendrogenesis begins during brain development [68]. Microglia and brain vasculature arise from the mesoderm, whereas mesoderm specification is generally suppressed during guided BO differentiation. Therefore, iPSC-derived microglia can be seeded onto differentiated BOs to allow infiltration [70]. Microglia-containing neuroimmune organoids can also be derived by ectopic expression of the microglial fate-determining transcription factor PU.1 during BO differentiation [71]. Alternatively, iPSC-derived microglia progenitors can be mixed with neural progenitors before BO differentiation [73]. Microglia may also develop innately during unguided BO differentiation; however, the reproducibility of this approach remains to be clarified [107]. Vascularized BOs can be obtained by ectopic expression of the Ets variant transcription factor 2 (ETV2) during BO differentiation [78]. Alternatively, integration of pericytes into differentiating BOs yields pericyte-containing BOs [81]. Self-organized iBBB can be obtained by mixing astrocytes, pericytes, and endothelial cells in a 3D hydrogel or assembling these cells into a spheroid [30, 79]. Finally, a combination of WNT and BMP signaling pathway activation has been applied to generate choroid plexus organoids (ChPOs) [83, 84].

Box 2: Modeling electrophysiology of Alzheimer’s disease using brain organoids.

Human brain electrical activity reflects neuronal network connectivity and cognitive functions. Although electrophysiological recordings are cost-effective and non-invasive, identification of electrophysiological markers as a diagnostic tool for AD remains a challenge [108]. Nevertheless, novel insights into electrophysiological activity of AD-afflicted neuronal networks may be obtained by recording electrical activity of 3D human BOs that form relatively complex neuronal networks as compared to 2D neuronal cultures. Electrical activity of cortical BOs increases over the course of differentiation and exhibits alternating periods of synchronized events and quiescence, reminiscent of early brain development in vivo [109]. Other modifications, such as BO vascularization or microglia infiltration, also facilitate BO electrical maturation [71, 72, 78, 110, 111]. For example, neuroimmune organoids exhibit increased frequency of oscillatory bursts and synchronization as compared to microglia-lacking BOs [71, 72, 110].

Emerging evidence indicates that BOs derived from AD patients also recapitulate neuronal network dysfunction characteristic of AD patients. For example, neuronal hyperexcitability and increased prevalence of epileptiform activity are evident in AD patients [112]. Indeed, neuronal hyperexcitability has also been document in fAD, sAD, and ApoE4 BOs [50, 113, 114]. Furthermore, cortical and ganglionic eminence assembloids integrate excitatory and inhibitory neurons, enabling the emergence of more complex electrophysiology [115]. Such assembloids exhibit epileptiform-like activity in a disease setting, whereas aberrant neuronal activity can be modulated pharmacologically [115]. Deterioration of BO electrical activity may also be used as a readout for cognitive decline; therefore, it will be important to characterize how various AD-relevant insults affect BO electrical activity over the course of disease progression. Notably, serum exposure and HSV-1 infection both result in impaired electrical activity of BOs [62, 89].

2D microelectrode arrays (MEAs) are often used to record neuronal activity in whole or sliced BOs [62, 109, 116]. However, engineered solutions to record BO electrical activity across the BO surface area and without the need for slicing would facilitate spatiotemporal recording during BO differentiation and beyond. Novel self-folding MEA caps overcome the constraints of 2D MEAs and can wrap around the organoids of different sizes in 3D space [117]. Moreover, cyborg organoids integrate stretchable mesh nanoelectronics arrays across the organoid to further increase the interface area between neurons and electrodes [118]. Such 3D electrode interfaces will enable investigators to assess how various AD-relevant genetic and environmental insults drive the deterioration of neuronal networks over time.

In this Review, we begin by considering the immense complexity of cell-cell interactions that deteriorate during AD progression; we also discuss the importance and advantages of using human cell-based models for studying AD. Subsequently, we draw on the recent advances in developing various BO models, including neural organoids, myelinoids, neuroimmune organoids (NIOs), blood-brain barrier (BBB) organoids, and choroid plexus organoids (ChPOs), to highlight their potential for modeling AD pathogenesis and developing novel therapeutics for AD.

Cell-cell interactions in Alzheimer’s disease

AD is a progressive neurodegenerative disease that presents with cognitive decline and is the most common cause of dementia [10]. Most AD cases are sporadic (sAD), whereas autosomal dominant familial AD (fAD) comprises about 1% of all AD cases and is caused by mutations in presenilin 1 (PSEN1), presenilin 2 (PSEN2), and amyloid precursor protein (APP) genes [11]. The neuron-centric Amyloid Hypothesis has been the leading theory for how AD develops and progresses, postulating that amyloid-β (Aβ) accumulation and toxicity initiate a cascade of events culminating in neurodegeneration and cognitive decline [11]. Extensive efforts to target Aβ pathology have resulted in recent accelerated approvals for monoclonal antibodies aducanumab and lecanemab by the Food and Drug Administration (FDA) [12–14]. Hyperphosphorylation and aggregation of the microtubule-associated protein tau is another hallmark of AD [10]. In recent years, tau has become a major focus in AD research, given the emerging evidence that tau pathology can develop and progress independently of Aβ [15]. Moreover, spatial and temporal correlation between tau pathology and neurodegeneration is much stronger than that between Aβ pathology and neurodegeneration, highlighting the importance of tau in disease progression [15]. In addition to neuronal Aβ and tau pathology, genome-wide association studies (GWAS, see Glossary), high-throughput profiling of gene expression, and functional experiments have revealed remarkable complexity of genetic risk factors, aging, cellular states, and cell-cell interactions that together drive AD pathogenesis [16–22]. Novel hypotheses for AD progression, including the Cellular Phase, ApoE Cascade, Myelin Breakdown and Neuroinflammation hypotheses have emerged [17, 18, 23, 24]. As the strongest genetic risk factor for AD, the E4 variant of the apolipoprotein E (APOE) gene has been linked to dysfunctional lipid metabolism, impaired myelination, heightened neuroinflammation, and vascular breakdown, among other phenotypes, exemplifying the immense complexity of cellular interactions that go awry in AD [18, 21, 25]. As the primary source of the secreted ApoE protein, astrocytes orchestrate brain lipid metabolism that is essential for neuronal physiology, injury repair, and myelin remodeling [17]. Astrocytes also maintain systemic homeostasis of the neuronal circuitry, remodel synapses, and form the glia limitans membrane at the BBB. In AD, astrocytes undergo reactive astrogliosis, leading to deterioration of their supportive functions, increased pro-inflammatory cytokine secretion, and neurotoxicity [17, 26]. Oligodendrocytes provide metabolic support to neurons and produce the myelin sheath for neuron electrical insulation [17]. The Myelin Breakdown Hypothesis postulates the central importance of myelin integrity to brain homeostasis, whereas myelin breakdown drives neurodegeneration and cognitive decline in AD [23, 27]. Moreover, excessive myelin debris may activate microglia, initiating neuroinflammation [27]. As the brain resident macrophages, microglia survey the brain environment, prune synapses, phagocytose debris, and mount an immune response to a pathogenic stimulus [17]. Microglial ability to remove toxic Aβ debris indicates a protective role in AD; however, at later stages of the disease, microglia drive neuroinflammation and may even propagate Aβ into unaffected brain regions [28]. Finally, brain vasculature coordinates energy metabolism and forms the BBB that protects the privileged brain environment from peripheral neurotoxic factors [17]. Moreover, the choroid plexus (ChP) forms the blood-cerebrospinal fluid (B-CSF) barrier and secretes CSF that facilitates metabolite exchange across the brain and serves a protective function [29]. Yet, deterioration of the brain vasculature in AD manifests as vascular loss, cortical hypoperfusion, BBB breakdown, and cerebral amyloid angiopathy, among other phenotypes [22, 30, 31]. Oligomeric Aβ drives capillary constriction and pericyte dysfunction, whereas profiling of the CSF proteome reveals dozens of dysregulated proteins in AD patients [32, 33]. All these cellular phenotypes converge to drive neuronal network degeneration, neuron cell death, and ultimately, cognitive decline. Therefore, defining how diverse cell-cell interactions deteriorate in AD and contribute to synaptic loss and neuronal cell death is a pivotal goal for future therapeutic development.

Species divergence necessitates human iPSC-based models of Alzheimer’s disease

Species differences between human and mouse brains have been a significant hurdle in translating rodent-based findings into viable AD therapies in the clinic [16]. High-throughput profiling studies have revealed considerable divergence in transcriptional programs of the brain cell types, differences in cortical cell organization, and notably increased cell-cell contacts in the human cortex [34, 35]. The divergence in gene expression profiles is particularly apparent in non-neuronal cells; for example, human but not mouse microglia are enriched for expression of AD risk genes [36]. Transcriptional divergence also implies phenotypic differences. Indeed, human and mouse astrocytes exhibit contrasting programs of energy metabolism, response to inflammatory stimulation, and susceptibility to oxidative stress [37]. Divergent oligodendrocyte gene expression profiles manifest in notable differences in myelin proteomes [38]. Species differences are also evident in the brain vasculature. Brain endothelial cells (BECs) and pericytes exhibit among the most divergent transcriptional profiles between human and mouse brain cell types [22]. These differences are amplified in a disease setting; for example, there is little overlap of differentially expressed genes between BECs isolated from AD patients and an AD mouse model when compared to healthy controls [22]. Consequently, considerable species divergence has hindered the elucidation of human-specific disease mechanisms and, by extension, therapeutic development for AD. BO technology enables modeling of AD in all-human cellular systems, so that human-relevant molecular mechanisms of AD pathogenesis can be uncovered.

Brain organoid models of Alzheimer’s disease

Unravelling the complexity of AD pathogenesis requires diverse BO models that resemble distinct structures of the human brain and incorporate specialized cell types. In this section, we highlight the applications of different BO models to study various aspects of AD pathogenesis.

Neural organoids

Neural or cerebral organoids are the most widely used 3D pluripotent stem cell-derived models of the human brain tissue to study brain diseases, including AD [6, 39]. Neural organoids derived from iPSCs of fAD patients robustly develop AD-relevant pathology, including Aβ aggregation and tau hyperphosphorylation (reviewed in [40]). These neural organoids also exhibit synaptic loss, impaired neuronal activity (Box 2), and neuronal cell death, in addition to other phenotypes, such as endosomal dysfunction [40]. Tau pathology is also evident in neural organoids that express mutant tau variants associated with frontotemporal dementia but relevant in a broader context of tauopathies, including AD [41]. For example, mutant tau-V337M neural organoids accumulate phosphorylated tau (p-tau) and exhibit impaired proteostasis, neuroinflammation, dysregulated lipid metabolism, and selective loss of glutamatergic neurons [42, 43]. Moreover, Aβ and tau pathology can be induced in a controllable manner, such as by treating neural organoids with an amyloidogenesis activator Aftin-5 or by injecting adeno-associated virus (AAV) carrying a mutant tau expression cassette [44, 45]. In this way, both independent and synergistic effects of Aβ and tau pathology on AD progression may be investigated using neural organoids [46, 47]. An important consideration is whether the conformation of tau aggregates in neural organoids resembles that of in vivo inclusions. Indeed, iPSC-derived neurons lack tau isoform diversity of the human brain, which may affect tau oligomerization and fibrillization properties in neural organoids [48]. AAV-mediated expression of a mutant tau-P301L that is more prone to aggregation than wild-type tau may overcome this limitation, leading to formation of tau oligomers and fibrils reminiscent of late tau pathology in AD [45]. Finally, self-propagation of aggregated tau can be modeled using astrocyte-neuron spheroids, where p-tau pathology is readily propagated from neurons pre-treated with tau to untreated neurons [49].

Although AD-relevant pathology can be induced by overt dysfunction or manipulation of the amyloid and tau cascades, neural organoids can also reveal subtle contributions of genetic and environmental risk factors for sAD. Neural organoids derived from iPSCs of sAD patients secrete more Aβ₄₀, Aβ₄₂, and p-tau than do neural organoids derived from healthy controls, indicating neuronal dysfunction [50]. Interestingly, the secreted amounts of Aβ₄₀, Aβ₄₂ and p-tau correlate with the Aβ burden detected by positron emission tomography in corresponding sAD patients, indicating the sensitivity of the BO technology for modeling sAD [50]. Neural organoids carrying genetic risk variants, such as APOE4, and their isogenic corrected controls can also be engineered by CRISPR/Cas9 editing (see Figure 1). Isogenic disease models eliminate the confounding effects of patient-to-patient variation and reveal genetic risk factor-specific phenotypes; this technique is often applied to generate isogenic iPSC lines carrying different APOE variants [51]. To study the impact of ApoE isoforms on AD progression using neural organoids, robust astrocyte differentiation is required (see Figure 2A, Key Figure); indeed, immature neural organoids express little APOE because they lack astrocytes [51]. As the organoids mature, radial glia and astrocytes secrete ApoE that is diffusively distributed throughout the 3D space of the organoid [51, 52]. APOE4 neural organoids accumulate more Aβ and p-tau than do APOE3 neural organoids although this phenotype may depend on neural organoid maturation [51, 53]. Neural organoids may thus be used to investigate how ApoE4 interacts with tau to drive AD-relevant pathology, given the recent observation that removal of neuronal ApoE4 suppresses tau pathology in vivo [25]. Neural organoids are also amenable to pharmacological manipulation, indicating their utility for testing novel therapeutic candidates [42, 49, 54]. Therefore, tau-targeting therapies, such as the recently developed tau-degrading intrabodies, as well as ApoE-targeting antibodies or antisense nucleotides could be tested in neural organoids [55–58]. However, it is important to note that different neural organoid differentiation protocols and variable organoid-to-organoid reproducibility (see Box 3) may affect the experimental findings obtained from neural organoid studies. For example, Hernández et al. found that ApoE, Aβ₄₂, and p-tau levels were highly variable within the groups of ApoE3 and ApoE4 neural organoids; as a result, there was no statistically significant difference between ApoE, Aβ₄₂, and p-tau levels of isogenic ApoE3 and ApoE4 neural organoids [59]. This discrepancy may also be attributed to the relatively mild ApoE isoform-dependent phenotypes as compared to those stemming from causal disease mutations.

Figure 1: Generation and applications of isogenic APOE3 and APOE4 brain organoids.

Patient-derived fibroblasts are first obtained from a skin biopsy of a carrier of APOE4. Fibroblasts are then reprogrammed into induced pluripotent stem cells (iPSCs), and a single nucleotide point mutation (cytosine-to-thymine) is introduced using CRISPR/Cas9-based genetic editing, resulting in a missense mutation and exchange of arginine (Arg) for cysteine (Cys), generating the APOE3 variant. Subsequently, unedited APOE4 iPSCs and isogenic edited APOE3 iPSCs are differentiated into brain organoids (BOs) following well-established protocols. The resulting BOs share the same genetic background except for the APOE variant, enabling mechanistic studies of the roles of ApoE isoforms in lipid metabolism, neuroinflammation, cellular stress, and other processes, without the confounding effects of patient-to-patient variation.

Figure 2 (Key Figure): Strategies for obtaining glial cell-enriched brain organoids.

(A) Long-term culture of neural organoids promotes astrocyte differentiation as radial glia progenitors undergo the neurogenic-to-gliogenic transition. Long-term maintenance of neural organoids can be further improved by slicing spherical organoids and maintaining sliced organoids to allow efficient nutrient and oxygen exchange. Astrocyte-enriched neural organoids can be used to study astrocyte lipid metabolism and astrogliosis in AD, among other roles. (B) Supplying key signaling molecules required for oligodendrocyte progenitor cell (OPC) and oligodendrocyte differentiation, including platelet-derived growth factor AA (PDGF-AA), insulin-like growth factor 1 (IGF-1), and thyroid hormone (T3), promotes OPC and oligodendrocyte differentiation in myelinoids. Patterning myelinoids towards ventral spinal cord identity can further improve myelination. Oligodendrocyte-enriched myelinoids can be used to study the effects of myelination and myelin breakdown in AD. (C) Given that microglia arise from a different germ layer (mesoderm) than do neural tissues (ectoderm), generation of microglia-containing neuroimmune organoids (NIOs) can be achieved by seeding induced pluripotent stem cell (iPSC)-derived microglia into differentiated BOs. To obtain microglia, iPSCs are differentiated into an erythromyeloid progenitor-like intermediate followed by microglia differentiation and maturation in the presence of transforming growth factor β (TGF-β), interleukin 34 (IL-34), and macrophage colony-stimulating factor (M-CSF). Microglia-enriched NIOs can be used to study phagocytosis of toxic debris and microgliosis in AD.

Box 3: Rigor and reproducibility of brain organoid technology.

To reach the full potential of modeling AD using BO technology, it is critical to ensure rigorous differentiation of organoids that faithfully and reproducibly recapitulate key features of the human brain. However, self-organization in vitro, especially when relying on unguided BO differentiation, exhibits substantial organoid-to-organoid variation [59, 119, 120]. BOs derived by unguided differentiation not only contain an assortment of cells from various brain regions, but also develop non-ectodermal cell types, including microglia [39, 107]. Thus, while unguided differentiation may increase the cell type diversity, it also affects reproducibility across individual BOs and may confound experimental findings. On the contrary, guided BO differentiation yields highly reproducible BOs. Single-cell sequencing of forebrain organoids reveals remarkable reproducibility of cell types with variation comparable to that of human brain development [119–121]. Therefore, unguided BO differentiation may be more suitable for exploratory experiments, whereas guided BO differentiation should be used for single-cell high-throughput profiling studies and drug screening applications, among others.

Rigor of the experiments using BO models also depends on how well BOs recapitulate the in vivo environment and whether BOs exhibit any phenotypes that could confound the experimental readout in the context of AD. Notably, BOs exhibit elevated expression of genes that regulate the pathways of endoplasmic reticulum stress and glycolysis [92]. Moreover, the lack of organoid vascularization is associated with internal hypoxia and increased hypoxia-inducible factor 1α (HIF1A) expression as compared to vascularized organoids [78]. These phenotypes have also been implicated in AD progression and thus may confound experimental findings. For example, HIF-1α signaling orchestrates microglia metabolic reprogramming from oxidative phosphorylation to glycolysis upon encounter of Aβ, whereas AD patient-derived neurons undergo Warburg effect-like glycolytic reprogramming that may be dependent on HIF-1α [122, 123]. Internal BO necrosis may also activate inflammatory responses of microglia and astrocytes upon debris encounter. Therefore, investigators should be aware of these intrinsic BO limitations as well as the ways to overcome them. For example, improved nutrient and oxygen exchange using a microfluidics device leads to increased glucose consumption, lactate secretion, cell proliferation, and electrical activity as well as reduced necrosis and HIF-1α⁺ area in BOs [93]. Moreover, BO slicing reduces necrosis and enables long-term organoid maintenance [94, 95].

Neural organoids may also be used to define the effects of non-genetic risk factors on AD pathogenesis. For example, BBB leakage is a well-recognized AD risk factor that may be driven by aging [60, 61]. Exposure of neural organoids to human serum induces accumulation of insoluble Aβ aggregates and p-tau, among other phenotypes, indicating that BBB leakage may precipitate AD-relevant pathology [62]. Because AD is an age-related neurodegenerative disorder, the study of BBB leakage that occurs in aged brains offers a way to incorporate age-relevant events when modeling AD in vitro [60–62]. Moreover, given that serum exposure results in widespread AD-relevant pathology, it will be important to determine what specific peripheral factors induce neuronal toxicity upon BBB breakdown [62]. For example, liver-derived ApoE4 has recently been shown to trigger BBB breakdown, induce astrogliosis, and impair memory in vivo [63]. Emerging reports also indicate a role for peripheral immune cells in AD progression [31, 64]. We anticipate that complex organoid models that integrate peripheral effectors will reveal novel mediators of neurotoxicity originating from outside the brain.

Myelinoids

Although neural organoids robustly develop neurons and astrocytes, additional strategies to introduce other AD-relevant cell types are required (Box 1). Cortical myelinoids contain oligodendrocyte progenitor cells (OPCs) and oligodendrocytes as well as exhibit a degree of myelination [65–67]. Ventral spinal cord myelinoids (Figure 2B) further develop lamellae of compact myelin wrapped around neurons with comparable myelin thickness to that in vivo as well as nodes of Ranvier and paranodal junctions [68]. Yet, whether myelinoids with improved myelination but patterned towards the spinal cord identity are applicable to AD research remains to be shown. Given the accumulating evidence for the importance of myelin breakdown in AD, we anticipate that myelinoids will become an effective platform for examining disease-associated phenotypes, underlying molecular mechanisms, and therapeutic strategies targeting myelination programs [23, 27]. For example, APOE4 expression is associated with impaired myelination, whereas 2-hydroxypropyl-β-cyclodextrin, a compound that promotes cholesterol efflux, has recently been shown to improve myelination in APOE4-targeted replacement mice in vivo [69]. Whether such a beneficial effect can be demonstrated in human cell-derived 3D myelinoids that exhibit complex myelination patterns remains to be determined.

Neuroimmune organoids

Neuroimmune organoids (NIOs) (Figure 2C, Box 1) contain microglia that survey their environment, prune synapses, phagocytose debris, and respond to injury [70–73]. NIOs can be used to model clearance of AD-relevant debris as well as inflammatory responses of microglia that contribute to neuroinflammation in the brain. Upon oligomeric Aβ₄₂ treatment, microglia become activated and phagocytose Aβ₄₂ in NIOs, whereas gene expression profiling reveals enrichment for lysosome and phagocytosis-related pathways [71]. Given that several genetic risk factors for sAD, such as variants of the triggering receptor expressed on myeloid cells 2 (TREM2), are microglia-specific, manipulation of their expression in NIOs may reveal novel disease phenotypes [74]. Cakir et al. developed a CRISPR interference (CRISPRi)-based strategy to suppress expression of various sAD risk genes in microglia in a high-throughput manner [71]. The authors found that suppression of TREM2 or the sortilin-related receptor 1 (SORL1) was associated with increased organoid injury upon oligomeric Aβ₄₂ treatment, indicating protective roles of TREM2 and SORL1 against amyloid-related toxicity [71]. Engineered microfluidic devices could further improve characterization of microglia behavior in response to AD-relevant pathology. Tubular NIOs with a hollow core can be constantly perfused to improve nutrient and oxygen exchange as well as deliver toxic substances to the organoid [75]. Moreover, NIOs may be used to clarify how microglia propagate tau pathology, whereas exposure of NIOs to primary or iPSC-derived peripheral immune cells may reveal microglia-immune cell interactions that contribute to AD progression [76, 77].

Blood-brain barrier organoids

3D models of the brain vasculature include vascularized BOs, pericyte-containing BOs, in vitro BBB (iBBB), and organs-on-a-chip (see Figure 3A & 3C and Box 1) [30, 78–81]. Vascularized BOs develop perfusable CD31⁺ endothelial tubes while maintaining normal cortical organization, contain tight junction proteins and have higher trans-endothelial resistance than control BOs [78]. Similarly, iBBB exhibits restricted paracellular permeability and active export of small-molecule compounds [30, 79]. Therefore, BBB organoid models can be used to evaluate drug permeability across the BBB as well as determine the effects of a particular drug on the BBB integrity [30, 79]. BBB organoid models also develop AD-relevant pathology: exposure of vascularized BOs and the iBBB to Aβ disrupts tight junctions and results in amyloid accumulation reminiscent of cerebral amyloid angiopathy, respectively [30, 78]. Moreover, APOE4 iBBB accumulates more amyloid deposits than APOE3 or ApoE-depleted controls [30]. A BBB organ-on-a-chip enables further functional compartmentalization of cell types that form the neurovascular unit, whereas uncoupled perfusion of vascular and brain compartments facilitates modeling of nutrient exchange characteristic of the BBB [80, 82]. Overall, in vitro BBB models may be used to study the molecular mechanisms driving pericyte and endothelial cell dysfunction, breakdown of the selective barrier, impaired glucose metabolism, and other phenotypes relevant to AD.

Figure 3: Human organoid technology for modeling brain vascular systems.

(A) Vascularized BOs can be engineered by transcription factor-driven directed endothelial cell differentiation during BO generation. Alternatively, spheroids containing astrocytes, pericytes, and endothelial cells self-organize into distinct layers that form a blood-brain barrier (BBB)-like structure. Finally, organ-on-a-chip devices combine microfluidics with in vitro cell culture, enabling the desired layering of different BBB cell types to be achieved. (B) Differentiation of choroid plexus organoids (ChPOs) enables modeling of the blood-cerebrospinal fluid (B-CSF) barrier and the production of CSF-like fluid. Furthermore, ChPOs may be used to study the changes in CSF composition during AD progression as well as to uncover AD-relevant CSF biomarkers. (C) Organoid models of brain vascular systems recapitulate key elements of the BBB and the B-CSF barrier. For example, BBB models contain endothelial cells, pericytes, and astrocytic end-feet as well as express genes encoding tight junction and channel proteins, such as aquaporin 4 (AQP4). ChPOs contain secretory epithelium producing CSF-like fluid as well as express genes encoding tight junction proteins, such as claudin 1 (CLDN1), and proteins required for CSF secretion, such as aquaporin 1 (AQP1). Therefore, the BBB and B-CSF barrier models can be used to study various aspects of vascular dysfunction associated with AD, and explore novel directions for AD research, such as the roles of peripheral lipoproteins and ApoE as well as the adaptive immune system in AD progression.

Choroid plexus organoids

Recently developed ChPOs form a B-CSF barrier-like structure lined with secretory cuboidal epithelium that produces CSF-like fluid into enclosed cysts [83] (Figure 3B & 3C and Box 1). ChPOs express various genes encoding transporter, signaling, and junction proteins as well as proteins required for CSF secretion, such as the water transporter aquaporin 1 (AQP1) [83, 84]. Remarkably, permeability of the ChPO barrier to various drugs correlates with that of the B-CSF barrier in vivo, highlighting the potential of using ChPOs for testing drug permeability and toxic drug accumulation before proceeding to clinical trials [83]. Moreover, the CSF-like fluid produced by ChPOs contains AD-relevant factors, such as ApoE [83]. Therefore, ChPOs may be used to study how various AD risk factors affect the integrity of the B-CSF barrier as well as the composition of the CSF-like fluid. Given that peripheral ApoE4 has been shown to trigger BBB breakdown, it will be important to define the impact of peripheral ApoE4 on the B-CSF barrier as well [63]. Moreover, co-culturing ChPOs with T cells will facilitate modeling of T cell activity and dysfunction in the CSF of AD patients [64]. ChPOs may also aid identification of novel CSF biomarkers, which could both inform research directions and facilitate more accurate stratification of patients that have different types of dementia but share symptomatology and pathology similarities [10, 16, 33]. However, biomarker discovery based on the CSF-like fluid produced by ChPOs should be rigorously validated using CSF isolated from AD patients, given the limitations of in vitro CSF-like fluid production.

Organoid models of virus infection

Environmental risk factors may precipitate AD onset or accelerate its progression. BOs have been used extensively to study virus infection, including SARS-CoV-2 most recently [81, 84–86]. Viruses may preferentially infect different cell types of the brain, illustrating the need for diverse BO models to evaluate viral and other environmental risk factors for AD [84, 86]. For example, SARS-CoV-2 selectively infects ChPOs and astrocyte-enriched BOs as well as infects ApoE4 neurons and astrocytes more efficiently than their ApoE3 counterparts [84, 86]. However, it remains to be determined if there is a definitive link between SARS-CoV-2 and AD although reports supporting the idea are emerging [87]. Conversely, viruses of the Herpesviridae family have long been implicated in AD pathogenesis [88]. Infection of a neural progenitor cell-derived 3D brain tissue model with herpes simplex virus 1 (HSV-1) induces formation of Aβ deposits, promotes astrocyte activation, and drives neuroinflammation [89]. Similarly, infection with Zika virus accelerates Aβ aggregation and tau phosphorylation in fAD BOs as well as leads to microglia and astrocyte activation in NIOs [73, 90]. These findings indicate that virus infection evokes AD-relevant pathology and warrant further investigation into the relationships between pathogenic insults, known risk factors for AD, and other effectors that may trigger AD onset or worsen its progression.

High-throughput drug screening using brain organoids

In addition to drug development applications discussed throughout this Review, BOs may also be applied for high-throughput drug screening, if BO scaling and reproducibility are optimized (Box 3). For example, engineered micropillar arrays may serve as scaffolds for efficient and highly reproducible generation of a large number of BOs suitable for drug screening [91]. Moreover, tissue clearing techniques enable high-content screening of organoid pathology by automated imaging acquisition and analysis [50]. Park et al. developed a scalable platform to evaluate accumulation of Aβ and p-tau in BOs using automated imaging in 96-well plates [50]. The authors performed a proof-of-principle screening of a panel of drugs targeting AD-relevant cellular pathways and robustly detected reduction in the Aβ and p-tau burden under most treatment conditions [50]. We anticipate that advancing automated imaging techniques will also be applied to assess AD-relevant pathology in various other organoids discussed throughout this Review, such as myelinoids to evaluate changes in the myelin volume [68].

Limitations of brain organoids for the study of Alzheimer’s disease

Although the exponential growth of the BO field has led to substantial improvements in BO differentiation and characterization protocols in recent years, important limitations relevant to disease modeling remain. BOs are primarily a neurodevelopmental model that recapitulates early human brain development rather than the aging brain when AD manifests. Therefore, the extent to which BOs can be used to model age-related neurodegeneration is not self-evident and requires rigorous validation of the observed phenotypes using alternative approaches. These include immunohistochemical and transcriptomic analysis of primary human brain tissue as well as experiments in vivo. Moreover, development of various techniques to model age-relevant events, such as BBB leakage by serum exposure as discussed earlier, is likely to exacerbate neurodegenerative phenotypes and strengthen the relevance of using BOs for the study of AD [62]. BOs are also a reductionist model, resembling only certain aspects of the brain cytoarchitecture and function; for example, neural organoids largely lack microglia, myelination, and the BBB. Although it remains challenging to introduce these critical components into a single BO model, specialized BOs discussed throughout this Review—myelinoids, NIOs, BBB organoids, and ChPOs—offer a versatile platform for modeling distinct AD-relevant phenotypes. Furthermore, the lack of vascularization in BOs contributes to internal hypoxia and cellular stress that lead to necrosis and impaired cell subtype specification [92]. Oxygen and nutrient exchange may be improved by using spinning bioreactors, engineering organoids-on-a-chip, introducing vascular cells, or simply slicing BOs (Box 3) [8, 78, 93–95]. Finally, organoid-to-organoid and batch-to-batch variation is a major issue that affects reproducibility of BO studies and is a barrier to high-throughput applications, such as drug screening. Variation can be mitigated by selecting guided BO differentiation approaches over those of unguided differentiation (Box 3), generating BOs from a defined number of iPSCs using microscaffolds, and including a quality control step to ensure correct neuroectoderm organization during differentiation [39, 91]. Similarly, careful assessment and reporting of the BO differentiation stage are also required to obtain reproducible results. For example, studying ApoE-related phenotypes requires sufficient BO maturation to establish a prominent astrocyte compartment, given that astrocytes are the primary source of secreted ApoE [21, 52]. Likewise, optimization of NIO differentiation protocols is required to ensure consistent microglia density across different batches of NIOs [73].

Concluding remarks and future directions

Rapidly advancing human BO technology enables modeling of complex cell-cell interactions of the human brain using easily obtainable and genetically modifiable in vitro models. BOs can be engineered to contain the desired combinations of genetic risk variants and cell types, so that scientific hypotheses could be addressed in the most direct manner (see Outstanding Questions). Unique advantages of BO models discussed throughout this Review could be integrated to derive organoids that recapitulate distinct cellular phenotypes associated with AD. For example, integration of microglia into myelinoids would enable modeling of microglia-mediated recycling of myelin debris upon myelin breakdown as well as how re-myelination could be invigorated by therapeutic intervention in AD patients. Novel approaches to promote BO maturation and specification will also facilitate the development of more complex BO models. For example, slicing BOs resolves diffusion constraints and allows formation of an expanded cortical plate and even nerve tracts, whereas integration of a Sonic hedgehog signaling center promotes topographical BO specification [94–96]. The ability to assemble heterotypic regionally-patterned BOs into assembloids will enable modeling of the interactions between distal brain regions during AD progression; for example, cortical-hippocampal assembloids could be used to study tau propagation, mimicking the in vivo pattern of tau spreading from the hippocampus to the neocortex, as well as clarify the roles of ApoE4 and microglia in tau propagation [10, 15, 25, 76]. Development of neural-vascular assembloids could further advance BO vascularization efforts and facilitate the study of the neurovascular unit [97]. Chimeric brain organoids, where only a specific cell type expresses a disease risk variant (reminiscent of conditional transgenic mouse models), may be used to define the cell type specificity in mediating the effects of genetic risk variants for AD [30, 98]. Advances in bioengineering of biomimetic scaffolds and organs-on-chip may enable BO perfusion as well as assembly of multi-organ chips to uncover the effectors originating from distal organs, such as gut microbiota-derived metabolites, that contribute to AD progression [99, 100]. Finally, BO transplantation in vivo will accelerate the efforts to promote BO maturation and functionalization, including circuit integration, vascularization, and microglia infiltration [101, 102]. We envision that BO transplantation models will serve as drug development platforms that combine the advantages of complex in vivo environment with the advantages of using human cell-derived BO models that have the potential to reveal human-specific molecular mechanisms governing AD progression, thus accelerating therapeutic development for AD.

Outstanding Questions.

How does the interplay between neurons and glial cells drive neuroinflammation in brain organoid (BO) models of Alzheimer’s disease (AD)?

What are the key events that promote microglial activation leading to neuroinflammation in AD BOs?

How does a multitude of genetic and environmental risk factors for AD interact to drive disease progression in BOs?

How do common and rare APOE variants affect myelin health and integrity in myelinoids?

How do various genetic and environmental risk factors for AD affect the blood-brain barrier integrity in brain vasculature-containing organoids?

How do APOE variants affect the integrity of the choroid plexus as well as the composition of the cerebrospinal fluid (CSF)-like fluid in choroid plexus organoids?

Can choroid plexus organoids producing CSF-like fluid be used for biomarker discovery?

How can aging-associated phenotypes be induced in BOs to study the impact of aging on AD pathogenesis?

What are the roles of the adaptive immune system in AD pathogenesis, and how can these roles be evaluated experimentally using the organoid technology?

Highlights.

Induced pluripotent stem cell (iPSC) technology has enabled modeling of human diseases using patient-derived cells with relevant genetic background

iPSC-derived 3D brain organoids self-organize and resemble the cytoarchitecture of in vivo brain tissue, making them attractive tools for modeling Alzheimer’s disease (AD)

Brain organoids contain multiple human brain cell types, enabling the study of cell-cell interactions that drive AD progression

Human iPSC-derived brain organoids may reveal human-specific disease phenotypes and molecular mechanisms that are not recapitulated in animal models

Clinician’s Corner.

The etiology of sporadic Alzheimer’s disease (AD) is poorly understood, whereas genome-wide association studies and high-throughput profiling of the human brain have revealed immense complexity of AD pathogenesis that involves multiple cell types and extensive cell-cell interactions.

Human organoid technology provides sophisticated self-organizing human brain-like tissues for modeling these cell-cell interactions in the context of AD. Moreover, the organoid technology overcomes the limitations associated with using rodent models that exhibit substantial species divergence in their cellular programs compared to those of humans.

Brain organoid technology can be scaled to develop high-content screening assays for drug discovery using engineered microarrays and automated imaging platforms to evaluate disease pathology and response to therapeutic intervention.

Modeling viral infection using brain organoids may rapidly inform research and medical community of how novel viruses, including SARS-CoV-2, infect human brain cell types and precipitate secondary pathology, including neurodegeneration.

Organoids and organs-on-a-chip engineered to contain brain vascular systems can be used to quantitatively evaluate drug permeability and toxic drug accumulation in brain tissues before initiating clinical trials. Substantial species divergence between mouse and human brain vasculature necessitates complementary evaluation of drug pharmacokinetics in human iPSC-based cellular models.

Glossary

Astrogliosis

astrocyte activation and proliferation in response to a damaging insult or brain injury.

Cerebral amyloid angiopathy

a medical condition characterized by amyloid accumulation on the walls of the brain vasculature.

Choroid plexus

a brain tissue comprised of a network of blood vessels and secretory epithelium that produces cerebrospinal fluid.

CRISPR interference

a molecular biology technique to suppress gene expression using the CRISPR/Cas9 platform (generally performed in a high-throughput manner).

Genome-wide association studies

a genetics methodology to identify genetic alterations that are linked to disease phenotypes.

Glutamatergic neurons

excitatory neurons that use glutamate as a neurotransmitter.

High-content screening

a methodology to identify cellular and sub-cellular phenotypes in a high-throughput manner.

Hyperexcitability

increased probability of neuronal activation upon stimulation.

Microgliosis

inflammatory activation of microglia in response to a damaging insult or brain injury.

Neuroectoderm

a type of tissue in a developing embryo that gives rises to the central nervous system.

Neuroinflammation

chronic activation of the innate immune system of the brain.

Nodes of Ranvier

intermittent areas of a myelinated axon that lack myelin and promote saltatory conduction of action potentials.

Organ-on-a-chip

a microfluidics device mimicking in vivo tissue architecture.

Oscillatory bursts

rhythmic electrical activity of neuronal networks.

Positron emission tomography

a non-invasive imaging methodology to detect tissue metabolic activity.

Radial glia

progenitor cells of the developing nervous system that give rise to neurons and macroglia.