A Cerebral Organoid Model of Familial Alzheimer’s Disease Using Amyloid Precursor Protein Mutation, Val669Leu (APPSeoul) Authors: Won Hee Jung, Seung Tack Oh, Mu Seog Choe, Yu Jin Kwon, So Jin Kim, Ye Seong Jeon, Kyung Min Baek, Woochul Chang, Kyung Seob Lim, Joong Sun Kim, Seung Pil Yun, Min Young Lee Published: 2026 ABSTRACT Various animal and cellular Alzheimer’s disease (AD) models harboring familial AD (fAD) mutations have been developed and widely used for AD research. In this study, we established an AD cerebral organoid (CO) model using a novel Val669Leu (APPSeoul) mutation in the APP gene. We generated a human embryonic stem cell (hESC) line overexpressing APPSeoul, referred to as the fAD-S hESC line. Using this line, we produced COs and confirmed robust AD-associated pathologies, including amyloid-β (Aβ) accumulation and tau phosphorylation. In addition, increased expression of β-secretase was observed in this model. Based on these findings, we investigated the effects of BACE1 inhibitor IV, a β-secretase inhibitor, in the CO model. Treatment with BACE1 inhibitor IV significantly reduced Aβ levels and tau phosphorylation. Furthermore, we differentiated the fAD-S hESC line into cortical neurons (fAD-S neurons) to establish a 2D cellular AD model. Consistent with the CO results, fAD-S neurons exhibited elevated levels of Aβ and phosphorylated tau, which were also significantly attenuated by BACE1 inhibitor IV treatment. Collectively, these results demonstrate the successful establishment of hESC-derived 2D and 3D AD models based on the APPSeoul fAD mutation. Introduction Alzheimer’s disease (AD) is a neurodegenerative disorder characterized by the progressive loss of memory and cognitive function. As the global population ages, the number of patients with AD is increasing worldwide at an alarming rate (1-4). AD is classified into sporadic AD (sAD) and familial AD (fAD) based on etiology and age of onset (5). sAD typically develops after the age of 65 years and is therefore considered late-onset AD, accounting for more than 95% of all AD cases (6). In contrast, fAD usually manifests before the age of 60 years and is categorized as early-onset AD (6). Although fAD is relatively rare, accounting for less than 1% of dementia cases, the familial mutations associated with fAD have formed the basis of the predominant molecular models of AD, upon which many current therapeutic strategies are being developed (7). Fully penetrant mutations in the amyloid precursor protein (APP), presenilin 1 (PSEN1), and presenilin 2 (PSEN2) have been widely used to generate various animal models of AD (8). These models have provided valuable insights into AD pathophysiology and have facilitated the development of candidate therapeutics. Nevertheless, animal models have inherent limitations due to genetic and physiological differences between humans and animals (9). Therefore, additional model systems are required to more accurately recapitulate human-specific aspects of AD biology. Cerebral organoids (COs) derived from human pluripotent stem cells (hPSCs) recapitulate key aspects of human brain development in three-dimensional (3D) structures composed of diverse neuronal and glial cell populations (10, 11). These characteristics make COs a powerful human-specific model for investigating AD. Indeed, AD CO models generated from human induced pluripotent stem cells (hiPSCs) derived from patients carrying the APOE-ε3/ε3 or APOE-ε4/ε4 genotypes have been reported, and these models exhibit increased levels of AD-related pathologies, including amyloid-β (Aβ) and phosphorylated tau (p-tau), supporting their utility as AD disease models (12, 13). In the present study, we employed an alternative strategy for AD CO modeling by overexpressing fAD mutations in normal hPSCs. This approach avoids the complex and time-consuming procedures required for patient-derived hiPSCs, such as the acquisition of somatic cells, ethical approval processes, and cellular reprogramming. In our previous work, we successfully established an AD CO model using a normal human embryonic stem cell (hESC) line overexpressing APP and PSEN1 harboring fAD mutations (14). Multiple pathogenic APP mutations have been identified to date (15). Recently, a novel APP mutation, Val669Leu (APPSeoul), was reported in a Korean female patient with early-onset AD (16). In this study, we aimed to establish an AD CO model incorporating the APPSeoul mutation. Using this model, we confirmed the manifestation of AD-associated pathologies and further demonstrated its potential applicability as a platform for drug testing. Collectively, we anticipate that this CO model will serve as a valuable tool for future AD research. Materials and Methods Human embryonic stem cell (hESC) culture All experiments involving hESCs were approved by the Public Institutional Bioethics Committee (approval numbers: P01-202110-41-001) designated by the Ministry of Health and Welfare and were approved by the Korea Disease Control and Prevention Agency. The hESC line (SNUhES31) was obtained from the Institute of Reproductive Medicine and Population, Medical Research Center, Seoul National University Hospital, South Korea. The cells were cultured on 10 μg/mL mitomycin-C-treated STO cells (American Type Culture Collection, ATCC) in hESC medium (1% minimum essential medium–non-essential amino acids [MEM-NEAA; Thermo Fisher Scientific], 1% GlutaMAXTM [Thermo Fisher Scientific], 7 μL/L β-mercaptoethanol, and 20% knockout serum replacement in Dulbecco’s Modified Eagle Medium/Nutrient Mixture F-12 [DMEM/F-12]) containing 20 ng/mL basic fibroblast growth factor (bFGF; PeproTech). HESCs were cultured in Essential-8 medium (Thermo Fisher Scientific) on GeltrexTM (Thermo Fisher Scientific)-coated culture plates for feeder-free. hESCs were passaged every four days using 500 μM ethylenediaminetetraacetic acid (EDTA)–phosphate buffered saline (PBS). Establishment of hESC line with the Val669Leu APP mutation (APPSeoul) To generate the APP clone with a mutation, we synthesized a partial length of the fAD Val669Leu (APPSeoul, GTG to CTG) gene mutation with restriction enzyme sites (Macrogen Inc.). Next, the synthesized fragment was subcloned into the downstream part of the pre-existing APP expression vector (CAG-APPSweFlLon, CA) using Mlu I and Xho I, creating the APPSeoul mutant gene (14). Finally, the expression of the APP gene containing the Seoul mutation (APPSeoul) was driven by the CAG promoter in the constructed lentiviral plasmid (CAG-APPSeoul, fAD-Seoul [fAD-S]). All newly built vectors were confirmed by sequencing (Enzynomics). To produce fAD-S lentiviral particles, we co-transfected the plasmid with packaging plasmids (an envelope plasmid harboring the VSVg gene and another plasmid harboring the gag–Pol genes) into 85% confluent HEK293T cells (ATCC) in a 10 mm tissue culture plate using the X-tremeGENETM HP transfection reagent (Roche). The medium was changed 24 hours after transfection, and 10 mL of virus-containing medium was harvested daily for three days. Each aliquot of the collected medium (30 mL) was concentrated to 200 μL by ultracentrifugation (25,000 rpm) at 4℃ for 2 hours (Hitachi). Next, feeder-free cultured hESCs in a well of a 12-well tissue culture plate were transfected with fAD-S viral particles by changing the medium to Essential-8 medium containing 2×107 IU/mL-concentrated viral particles and 2 μg/mL hexadimethrine bromide (Sigma-Aldrich). The medium was changed daily for three days. Then, we dissociated hESCs, transfected with fAD-S viral particles, into single cells with AccutaseⓇ and 500 hESCs were plated into a STO-plated well of a 6-well plate (Corning). After 7 days of culture, every single cell-derived colony was manually dissected under the stereomicroscope (NSZ818, Nexcope) and transferred into the wells of a Matrigel-coated 12-well culture plate. After expansion of transferred cells under feeder-free conditions, the APP expression levels of each cell lines were analyzed by Western blotting. The transfected hESCs were named as the fAD-S hESC line. Generation of cerebral organoids COs were generated from hESCs using a previously established protocol (10). Briefly, feeder-free hESCs were dissociated into single cells through incubation with 0.5 mM EDTA for 4 minutes and AccutaseⓇ for 4 minutes and then plated onto an ultra-low-attachment 96-well U-bottom plate at a cell density of 9×103 in low-bFGF hESC medium (1% MEM-NEAA [Thermo Fisher Scientific], 1% GlutaMAXTM, 7 μL/L β-mercaptoethanol, and 20% knockout serum replacement in DMEM/F-12) containing 5 ng/mL bFGF) with 50 μM Y27632, a Rho-associated protein kinase inhibitor; this was recorded as day 0. The medium was changed every other day. On day 2, the medium was replaced with low-bFGF hESC medium with 50 μM Y27632. On day 4, the medium was replaced with hESC medium without Y27632. On day 6, each embryoid body (EB) was transferred to one well of an ultra-low-attachment 24-well plate containing neural induction medium (1% MEM-NEAA, 1% GlutaMAXTM, 1% N2 supplement [Thermo Fisher Scientific], and 1 μg/mL of heparin [Sigma-Aldrich] in DMEM/F-12). On day 8, the EBs were fed by adding a fresh neural induction medium. On day 10, neuroepithelial tissues were transferred to Matrigel droplets containing CO differentiation medium without vitamin A (0.5% MEM-NEAA, 1% Glutamax, 1% B27 supplement without vitamin A, 0.5% N2 supplement, 2.5 μg/mL human insulin [Roche], and 3.5 μL/L β-mercaptoethanol in a 1:1 mixture of neurobasal medium [Thermo Fisher Scientific]). On day 12, the medium was replaced with a fresh CO differentiation medium without vitamin A. On day 14, the Matrigel droplets containing the organoid were transferred to a disposable spinner flask (Corning) with CO differentiation medium (0.5% MEM-NEAA, 1% Glutamax, 1% B27 supplement, 0.5% N2 supplement, 2.5 μg/mL human insulin [Roche], and 3.5 μL/L β-mercaptoethanol in a 1:1 mixture of neurobasal medium [Thermo Fisher Scientific]) and changed every 7 days thereafter. Neuronal differentiation Single cell-dissociated 1×105 hESCs were plated in Matrigel-coated 24-well plates in mTeSRTM1 supplemented with 5 μM of Y27632 (day −4). On day −3, the medium was replaced with mTeSRTM1 without Y27632, and the medium was changed daily. On day 0, the mTeSRTM1 medium was replaced with neuronal cell induction medium (1% MEM-NEAA, 1% GlutaMaxTM, 2% B27 supplement, 1% N2 supplement, and 20 μg/mL of human insulin in a 1:1 mixture of neurobasal medium and DMEM/F-12), and the medium was replaced every day until day 10. On day 11, the dissociated cells were resuspended in a neuronal cell induction medium supplemented with 5 μM of Y27632 and replated onto a Matrigel-coated 24-well plate (3×105 cells/well). On day 13, the medium was changed to neuronal cell induction medium supplemented with 100 nM of LDN-193,189, 10 μM of SB431542, and 2 μM of XAV 939. The medium was refreshed every other day until day 17. On day 19, the medium was changed to a neuronal cell maturation medium (1% N2 supplement, 2% B27 supplement, and 25 ng/mL of BDNF in the neurobasal medium). The medium was again changed every other day. On day 25, the cells were dissociated and replated in a Matrigel-coated 24-well plate (1×105 cells/well), and the medium was replaced every other day. The cells were subcultured every 7 days thereafter. Enzyme-linked immunosorbent assay The Aβ 1∼40 and Aβ 1∼42 levels were determined using human amyloid-β assay kits (IBL). CO samples or neural cell samples were sonicated in radioimmunoprecipitation assay (RIPA) buffer on ice. Protein concentrations were quantified using the bicinchoninic acid method. Briefly, 10 μg samples and serially diluted standards were plated in triplicate onto 96-well plates pre-coated with the specific capture antibody, and incubated overnight at 4℃. After four washes using wash buffer, horseradish peroxidase (HRP)–conjugated anti-human Aβ 1∼40 or 1∼42 antibody was added and incubated for 1 hour at 4℃. After five washes using wash buffer, substrate solution was added for color development and incubated for 30 minutes. The reaction was stopped with stop solution. The absorbance was measured at 450 nm using a microplate reader (Infinite M200 Pro, Tecan) and the concentration of Aβ 1∼40 or 1∼42 in the samples was calculated by interpolating the data against the standard curve. Reverse transcription-quantitative polymerase chain reaction Total RNA was extracted using the AccuPrepⓇ Universal RNA Extraction Kit (Bioneer). Reverse transcription (RT) was performed using the iScriptTM cDNA Synthesis Kit (Bio-Rad). The resulting cDNA (2.5 ng) was amplified using the FastStart Essential DNA Green Master polymerase chain reaction (PCR) Kit (Roche). Quantitative PCR (qPCR) was carried out using the LightCyclerⓇ 96 real-time PCR system (Roche) for 35 cycles under the following conditions: initial denaturation at 95℃ for 5 minutes, followed by denaturation at 95℃ for 30 seconds, annealing at 58℃ for 30 seconds, and extension at 72℃ for 30 seconds. All RT-qPCR experiments were performed in triplicate. Cycle threshold (Ct) values were calculated using the default settings of the LightCyclerⓇ 96 software (Roche). Immunohistochemistry The COs were fixed with a 4% paraformaldehyde (PFA) solution in PBS. COs were allowed to sink in 15% and 30% sucrose solutions sequentially, followed by embedding in optimal cutting temperature (OCT) compound (Leica). Next, COs were cryosectioned into 10 μm thick sections using a cryotome (CM1850 cryostat; Leica). The sections were permeabilized and blocked with 10% normal goat serum (NGS) in PBS with Tween 20 (PBST) for immunohistochemistry. Then, the sections were incubated with primary antibodies in PBST containing 2% NGS. Primary antibodies against the following proteins were used: 6E10 (mouse, BioLegend), 4G8 (mouse, BioLegend), p-tau (AT8; mouse, Thermo Fisher Scientific), class III beta-tubulin 1 (TUJ1; mouse, R&D Systems), doublecortin (DCX; rabbit, Cell Signaling), paired box 6 (PAX6; rabbit, BioLegend), microtubule-associated protein 2 (MAP2; rabbit, Cell Signaling), and N-cadherin (rabbit, Santa Cruz). After three washes using PBST, the tissues were incubated with secondary antibodies, goat anti-rabbit or anti-mouse Alexa Fluor 488 and 647 conjugates (Thermo Fisher Scientific). Images were obtained using confocal microscopy (Leica TCS SP5 II; Leica). Western blot analysis COs or neuronal cells were sonicated in RIPA buffer (iNtRON Biotechnology) on ice (Vibra-CellTM). The proteins obtained from each sample were separated using 10%∼12% sodium dodecyl sulfate (SDS)-polyacrylamide gel resolving gels and transferred to poly vinylidene fluoride (PVDF) transfer membranes (Millipore). The membranes were washed using TBST (150 mM NaCl, 10 mM Tris-HCl [pH 7.6], and 0.1% Tween-20), blocked with 5% skimmed milk (Millipore) for 1 hour, and incubated with primary antibodies. Primary antibodies were used against the following proteins: APP (mouse, Cell Signaling), β-secretase (rabbit, Cell Signaling), p-tau (AT8; mouse, Thermo Fisher Scientific), T-tau (mouse, Cell Signaling), and β-actin (mouse, Santa Cruz). The membranes were washed using TBST and incubated with secondary antibodies, HRP-linked goat anti-rabbit (Santa Cruz) or goat anti-mouse (Santa Cruz) IgG. An enhanced chemiluminescence solution (Thermo Fisher Scientific) was used to visualize the bands. Images were obtained using a ChemidocTM imaging system (Bio-Rad). Densitometric analysis was performed using ImageJ software from the National Institute of Health (NIH). Bielschowsky’s silver staining Bielschowsky’s silver staining was performed using VitroViewTM Bielschowsky’s Silver Stain kit (VitroVivo Biotech). Briefly, the cryosectioned slides were incubated in a pre-warmed (40℃) silver nitrate solution for 15 minutes before rinsing in distilled water (DW). The slides were incubated in ammonium silver solution at 40℃ for 30 minutes. The slides were placed in the Developer Stock Solution for 1 minute, after which the reaction was halted by immersing the slides in 1% ammonium hydroxide solution for 1 minute. After three washes with DW, the slides were put in a 5% sodium thiosulfate solution for 5 minutes before washing three times with DW. After dehydration with 100% ethanol, the slides were incubated in xylene. Then, the slides were mounted with coverslips using a mounting medium. Images were obtained using a microscope (AXIO, Zeiss). Immunocytochemistry Dissociated cells were fixed with a 4% PFA solution in PBS (Wako). The cells were permeabilized and blocked with 10% NGS (Vector Laboratories) in 0.1% PBST (vol/vol, Triton X-100 in PBS). The cells were incubated with primary antibodies in PBST containing 2% NGS. Primary antibodies against the following proteins were used: NANOG (rabbit, Cell Signaling), octamer-binding transcription factor 4 (OCT4; mouse, Santa Cruz), TRA-1-60 (mouse, Santa Cruz), stage-specific embryonic antigen-4 (SSEA4; mouse, Santa Cruz), TUJ1 (mouse, R&D Systems), DCX (rabbit, Cell Signaling), MAP2 (mouse, Abcam), 4G8 (mouse, BioLegend), 6E10 (mouse, BioLegend), and p-tau (AT8) (mouse, Thermo Fisher Scientific). After three washes with PBST, the cells were incubated with the secondary antibodies, goat anti-rabbit or goat anti-mouse Alexa Fluor 488 and 647 conjugates (Thermo Fisher Scientific). Images were obtained through confocal microscopy (Leica TCS SP5 II; Leica). Statistical analysis All results are expressed as the mean±standard deviation. Differences between the mean values were analyzed using the Student’s t-test. A value of p<0.05 was considered statistically significant. Results Establishment of familial Alzheimer’s disease–Seoul (fAD-S) hESC line Lentiviral constructs were designed and generated to express mCherry or APPSeoul, as previously reported (17), under the control of the CAG promoter. After generating CAG-mCherry (Con) and CAG-APPSeoul (fAD-S) lentiviral particles, hESCs were transfected with the Con or fAD-S lentiviral particles. Three single cell-derived mCherry-expressing hESC colonies and six single cell-derived fAD-S hESC colonies were manually picked and expanded (Fig. 1A). Among the six cell lines, one fAD-S hESC line (#5) was selected based on the APP expression level determined by Western blotting (Fig. 1B, 1C). Immunofluorescence staining of hESC markers, including OCT4, NANOG, SSEA4, and TRA-1-60, showed that the undifferentiation of the chosen fAD-S hESC line was not affected by the fAD gene transduction (Fig. 1D). Consistently, qPCR analysis revealed no significant differences in the expression levels of OCT4, SOX2, and NANOG between wild-type and fAD-S hESCs (Fig. 1E). These results demonstrate the successful establishment of the fAD hESC lines, fAD-S. Subsequent studies used these cell lines with Con hESCs as the controls. Next, we generated COs using Con or fAD-S hESCs (Fig. 1F). We examined the expression of several neural markers including neuron-specific TUJ1 (neuronal marker), DCX (neuronal marker), PAX6 (radial glial stem cell marker), neural-specific N-cadherin, SOX2 (a neural progenitor marker), MAP2 (an immature and mature neuronal marker), NeuN (a mature neuronal marker), synapsin 1 (a synaptic functional marker), vesicular glutamate transporter 2 (VGLUT2; an excitatory neuronal marker), glial fibrillary acidic protein (GFAP; an astrocyte marker) using immunohistochemical staining with 70-day-old Con and fAD-S COs (Fig. 1G). These proteins were commonly recapitulated in Con and fAD-S COs. Based on these results, we concluded that normal CO formation is not affected by the transduction of fAD genes, and subsequent experiments were performed. Expression of Alzheimer’s disease pathologies in fAD-S cerebral organoids The Aβ 1∼40 and 1∼42 isoform levels in 70-day-old control and fAD-S COs were measured. As shown in Fig. 2A, the Aβ 1∼40 and 1∼42 levels were significantly increased in the fAD-S COs compared to the Con COs. Immunohistochemistry using 6E10 and 4G8 anti-Aβ antibodies demonstrated Aβ accumulation in the fAD-S COs (Fig. 2B). Additionally, Bielschowsky’s silver staining showed strong amyloid plaque-like silver deposition in fAD-S COs; however, not in the Con COs (Fig. 2C) (18). The hyperphosphorylation of tau protein is another pathological marker of AD (19). As shown in Fig. 2D and 2E, our results revealed that the p-tau levels were significantly increased in the fAD-S COs compared to the Con COs. These results indicate that the pathological features of AD, Aβ, and tau phosphorylation are expressed in fAD-S COs. Application of fAD-S cerebral organoids in drug testing In this study, we determined the usefulness of our AD CO model as a tool for drug testing. Since APP gene mutations were reported to increase β-secretase cleavage (20), we examined the expression level of β-secretase. As shown in Fig. 3A, β-secretase expression was significantly increased in fAD-S COs compared to the Con COs. Next, we examined the effect of a β-secretase inhibitor, BACE1 inhibitor IV (BSI), on AD pathology expression on fAD-S COs. Our results indicated that treatment with BSI markedly decreased Aβ 1∼40 and 1∼42 levels in fAD-S COs (Fig. 3B). Western blot analysis revealed that p-tau and total tau (T-tau) expression in fAD-S COs was also attenuated by BSI (Fig. 3C). These findings are consistent with previous studies reporting that inhibition of BACE1 significantly reduced both p-tau and T-tau levels in hiPSCs-derived neurons as well as in primary neuronal cultures (7, 8). These results suggest that the fAD-S CO model reflects theoretical background concerning the relationship between APP mutations and β-secretase activity and can be a useful tool for drug development studies. Cellular Alzheimer’s disease modeling using fAD-S hESC line In this study, we induced differentiation into cortical neurons from the Con and fAD-S hESC line using a modified protocol from a previously reported method (21, 22). Differentiated cells strongly expressed neuronal markers, including TUJ1, DCX, and MAP2, and we did not observe differences in neuronal differentiation capacity between Con and fAD hESC lines (Fig. 4A). Next, we examined the expression of AD pathologies and the effect of BSI on the fAD-S neuronal cell model. As shown in Fig. 4B, the Aβ 1∼40 and 1∼42 levels were markedly increased in fAD-S neuronal cells compared to the Con neuronal cells, and BSI inhibited this increase. Additionally, Western blot analysis showed that p-tau and T-tau levels also significantly increased in fAD-S neuronal cells compared to the Con neuronal cells, and this increase was attenuated by BSI (Fig. 4C). These results suggest that fAD-S hESCs can also be used as a cellular AD model through differentiation into neuronal cells, and this model can be used as a tool for drug testing. Discussion Numerous studies have demonstrated the utility of COs as a platform for AD modeling (23, 24). Indeed, several reports have shown that hallmark AD pathologies, including Aβ and p-tau, are present in COs derived from patients with early-onset AD (25, 26). However, previous findings have indicated that the generation of COs from early-onset AD patient–derived hiPSCs alone may not be sufficient to robustly recapitulate AD pathology (14). Consistent with this observation, we previously established an AD CO model exhibiting pronounced AD-associated pathologies through the induction of exogenous fAD mutations. In the present study, we applied the same strategy to generate the AD CO model used herein. fAD is most commonly caused by mutations in the APP, PSEN1, and PSEN2 genes (25). Among these, approximately 25 APP mutations have been identified as pathogenic (27), although their precise pathogenic mechanisms remain incompletely understood. Therefore, investigating individual APP mutations may provide important insights into AD pathogenesis. This study focused on a recently identified APP mutation, Val669Leu (APPSeoul), reported in a Korean female patient with early-onset AD (16). The APPSeoul mutation is a point mutation located at the β-secretase cleavage site of APP, proximal to the well-characterized APPSwedish mutation. We established clonal AD hESC lines overexpressing APPSeoul under the control of the CAG promoter and selected one line for subsequent analyses. COs derived from this line exhibited significantly increased Aβ expression and aggregation, as well as elevated levels of p-tau. These findings demonstrate that AD-like pathologies can be recapitulated in COs and that AD organoid modeling is feasible through the overexpression of the APPSeoul mutation. Aβ accumulation and tau hyperphosphorylation are well-established pathological hallmarks of AD. Pathogenic APP mutations have been shown to increase total Aβ production or elevate the Aβ42/Aβ40 ratio (28, 29). APP is a transmembrane protein that can undergo a series of proteolytic cleavages by secretase enzymes. When cleaved by α-secretase in the middle of the β-amyloid domain (Aβ), it is not amyloidogenic. However, when APP is cleaved by β- and γ-secretase enzymes, neurotoxic Aβ peptides are released, which can accumulate into oligomer aggregates. In particular, mutations in the APP gene tend to inhibit cleavage by α-secretase and consequently enable preferential cleavage by β-secretase (20). Indeed, we observed that β-secretase expression was increased in COs with the APPSeoul mutation. While the precise mechanism underlying this upregulation remains to be fully elucidated, previous reports have shown that elevated Aβ levels can induce β-secretase expression through a feed-forward mechanism involving oxidative stress (5) and JNK pathway (6), etc. Therefore, we interpret the increased β-secretase levels in our model as a secondary response to Aβ accumulation rather than a direct transcriptional effect of APPSeoul mutation itself. Additionally, accumulating evidence indicates that excessive Aβ deposition promotes tau hyperphosphorylation (30) by disrupting the balance between tau kinases and phosphatases (31). Specifically, Aβ has been shown to activate kinases such as GSK-3β and CDK5, leading to aberrant phosphorylation of tau (32-34). Our results support this mechanism, as treatment with the BSI significantly reduced both Aβ and p-tau levels in COs carrying APPSeoul mutation. These findings not only demonstrate that tau phosphorylation is mediated by Aβ accumulation but also highlight the value of our AD organoid model for assessing therapeutic drug effects. Although organoid models more closely recapitulate tissue architecture and complex cellular interactions, two-dimensional (2D) cellular models remain widely used for high-throughput drug screening due to their simplicity, scalability, and cost-effectiveness (35). Moreover, various advanced 3D modeling approaches, including spheroids, bioprinting, and organ-on-a-chip systems, rely on 2D-cultured cells as their starting material (36, 37). Numerous small-scale compound screening studies targeting AD-related pathways have been conducted using differentiated neuronal subtypes (38, 39). Notably, a recent large-scale screening of over 1,600 compounds identified multiple candidates that reduce tau phosphorylation, thereby providing important insights into the mechanisms underlying tau pathology (40). In the present study, neuronal cells differentiated from the same fAD hESC line stably expressed Aβ and p-tau, and both markers were significantly reduced following BSI treatment. These findings indicate that our transgenic fAD hESC line can also serve as a robust 2D neuronal AD model. Collectively, this study demonstrates that the APPSeoul mutation induces AD-associated pathologies in both COs and hESC-derived neuronal cells. We anticipate that this modeling strategy will provide a valuable platform for future AD research and for establishing disease models of other genetically driven disorders.