GLP‐1R agonist semaglutide protects human forebrain organoids from cholesterol‐induced neurotoxicity

Abstract

INTRODUCTION

Dysregulated cholesterol metabolism represents a critical metabolic stressor in the central nervous system, contributing to neuronal injury across multiple neurological conditions, including Alzheimer's disease (AD). While glucagon‐like peptide‐1 receptor (GLP‐1R) agonists show neuroprotective potential, their capacity to directly mitigate cholesterol‐induced metabolic stress in human neural cells remains to be established.

METHODS

Here, we modeled metabolic stress using human induced pluripotent stem cell–derived forebrain organoids exposed to cholesterol overload. The protective effects of the GLP‐1R agonist semaglutide were characterized by integrating bulk/single‐cell transcriptomics, calcium imaging, and biochemical assays. Furthermore, clinical translatability was supported by mapping organoid expression signatures against human post mortem neurodegenerative brain datasets.

RESULTS

Cholesterol overload induced cellular stress and transcriptomic alterations that partially overlapped with AD‐associated signatures. Semaglutide protected neural cells from lipotoxic injury. Intracellularly, GLP‐1R activation engaged the cyclic adenosine 3',5'‐monophosphate–protein kinase A and phosphoinositide 3 kinase–protein kinase B–mechanistic target of rapamycin signaling pathways, consistent with reduced lipid droplet accumulation and oxidative stress. Intercellularly, single‐cell analysis indicated partial preservation of disrupted cellular communication, including the neurotrophic midkine signaling network. Functionally, semaglutide stabilized cellular calcium activity patterns under metabolic stress.

DISCUSSION

These findings highlight the role of cholesterol homeostasis in maintaining neuronal integrity and position GLP‐1R signaling as a candidate protective axis under metabolic stress, offering insights into therapeutic strategies for AD and broader neurodegenerative disorders.

Highlights

• Cholesterol overload triggers metabolic stress in forebrain organoids.

• Cholesterol‐induced transcriptomic profiles overlap with Alzheimer's disease–associated signatures.

• Glucagon‐like peptide‐1 receptor activation reduces lipid droplet accumulation and oxidative stress.

• Semaglutide preserves midkine intercellular communication.

• Semaglutide stabilizes cellular calcium activity patterns under metabolic stress.

1. BACKGROUND

Neurological damage is classically characterized by progressive neuronal loss, synaptic dysfunction, and the collapse of neural networks. ¹ , ² Beyond these structural impairments, emerging evidence highlights disrupted lipid metabolism as a critical hallmark accompanying neural injury. ³ , ⁴ , ⁵ , ⁶ , ⁷ Epidemiological evidence underscores a robust correlation between cholesterol‐induced metabolic stress and neurological health across the lifespan. In early life, offspring of mothers with obesity‐associated hypercholesterolemia face a nearly 2‐fold increased risk of neurodevelopmental disorders, ⁸ , ⁹ , ¹⁰ , ¹¹ such as autism spectrum disorder (ASD) ¹² and developmental delays, ¹³ as excess cholesterol can traverse the placental barrier to disrupt fetal brain development. ¹⁴ , ¹⁵ Meanwhile, in aging populations, elevated midlife cholesterol serves as an independent predictor of Alzheimer's disease (AD), increasing risk by up to 57%. ¹⁶ Thus, aberrant cholesterol metabolism represents a common pathological denominator underlying both early neurodevelopmental defects and late‐life cognitive decline.

To address such metabolic derangements, glucagon‐like peptide‐1 receptor (GLP‐1R) agonists are widely used. ¹⁷ Semaglutide, a long‐acting GLP‐1R agonist, is well established for reducing body weight and improving lipid profiles in patients with type 2 diabetes mellitus (T2DM) and obesity. ¹⁸ , ¹⁹ Recently, its neuroprotective potential has garnered significant attention. ²⁰ , ²¹ Large‐scale real‐world data demonstrate that semaglutide reduces the risk of a first‐time AD diagnosis by 40% to 70% compared to other antidiabetics. ²² In patients with T2DM and obesity, it significantly lowers risks of dementia (hazard ratio [HR] 0.63) and stroke. ²³ Furthermore, a pooled analysis of clinical trials reported a 53% reduction in dementia risk (HR 0.47). ²⁴

While semaglutide is proven to improve systemic lipid profiles and exhibits robust clinical neuroprotective potential, the cellular and molecular mechanisms underlying its direct action on neural cells remain poorly understood. Specifically, it remains elusive how semaglutide modulates cellular responses against lipid metabolic stress. Resolving this mechanistic gap is essential for optimizing its therapeutic utility in metabolism‐related neurological disorders.

Investigating these direct drug–cell interactions is critically hindered by a scarcity of controlled studies using humanized models. Although 2D human cell lines (e.g., SH‐SY5Y) have demonstrated semaglutide's anti‐apoptotic effects, ²⁵ , ²⁶ their lack of 3D architecture and intercellular complexity severely limits their physiological and translational relevance. In this regard, human induced pluripotent stem cell (hiPSC)‐derived brain organoids offer a transformative platform. ²⁷ These organoids faithfully recapitulate key structural, cellular, and molecular features of the human brain, providing a physiologically relevant platform for modeling neurological disorders, investigating neural injury, and evaluating therapeutic interventions under controlled conditions.

Here, we used hiPSC‐derived forebrain organoids to model cholesterol‐induced metabolic stress and investigate direct drug–cell interactions. Our findings establish semaglutide as a protective agent against cholesterol‐driven neural injury and position GLP‐1R signaling as a candidate protective axis under metabolic stress, offering insights into therapeutic strategies for AD and broader neurodegenerative disorders.

2. METHODS

2.1. hiPSC culture and forebrain organoid differentiation

hiPSC lines (DXR0109B and DYR0100) were cultured and differentiated into neural progenitor cells (NPCs) and dorsal forebrain organoids via a previously established stepwise protocol. ²⁸ Detailed step‐by‐step differentiation protocols, culture medium compositions, and reagent specifications are provided in the supporting information (Supplementary Methods 3.1–3.2).

2.2. Histological, phenotypic, and cell viability assays

Experimental procedures and tissue preparations for cell viability (Cell Counting Kit‐ 8 [CCK‐8], Calcein‐AM/PI live/dead staining), lipid accumulation (Oil Red O staining), mitochondrial and reactive oxygen species (ROS) staining, western blot, and immunofluorescence are fully detailed in the supporting information (see Supplementary Methods 3.3–3.10).

2.3. Pharmacological interventions

Water‐soluble cholesterol (10 mg/mL stock in H₂O) was added to culture medium at a final concentration of 10 µg/mL to induce cholesterol overload. Semaglutide (250 µM stock in phosphate‐buffered saline) and the GLP‐1R antagonist C5d (10 mM stock in dimethyl sulfoxide [DMSO]) were diluted in culture medium to final working concentrations of 100 nM and 1 µM, respectively. Controls received equivalent volumes of respective vehicles (DMSO < 0.1%).

2.4. Total RNA extraction and quantitative polymerase chain reaction

Total RNA was isolated from forebrain organoids using the Eastep Super Total RNA Extraction Kit, strictly adhering to the protocol provided by the manufacturer. Relative gene expression levels were subsequently determined using the 2^−ΔΔCt method, with normalization performed against the housekeeping gene GAPDH. Detailed oligonucleotide information is listed in Table S1 in supporting information.

RESEARCH IN CONTEXT

1. Systematic review: Dysregulated cholesterol metabolism represents a critical metabolic stressor in the central nervous system associated with neuronal injury across neurodegenerative contexts, including Alzheimer's disease (AD). Although glucagon‐like peptide‐1 receptor (GLP‐1R) agonists demonstrate clinical potential in lowering dementia risk, their capacity to directly safeguard human neural cells against cholesterol‐induced metabolic stress remains poorly defined.

2. Interpretation: We demonstrate that cholesterol overload in human forebrain organoids induces cellular stress accompanied by transcriptomic profiles that partially overlap with selected AD‐associated signatures. Semaglutide directly counteracts this lipotoxicity by modulating lipid clearance and reactive oxygen species levels, preserving cell–cell communication, and stabilizing cellular calcium activity. These findings position GLP‐1R signaling as a candidate protective axis under metabolic stress, offering mechanistic insights into cholesterol‐driven neural injury.

3. Future directions: Further validation in vascularized organoid assembloids incorporating microglia and astrocytes, alongside in vivo models and clinical biomarker analyses, will be essential to translate these GLP‐1R–mediated neuroprotective effects into targeted AD therapies.

2.5. Bulk RNA sequencing analysis

Total RNA was extracted using the RNeasy Mini Kit (Qiagen) and quality checked via Agilent 4150 TapeStation. Stranded RNA sequencing was performed on the DNBSEQ platform (BGI/Innomics). Reads were trimmed, aligned to GRCh38, and quantified. Differential expression was analyzed using DESeq2 (|log₂FC| > 0.58, false discovery rate [FDR] < 0.10). Gene set enrichment analysis (GSEA) was performed using clusterProfiler based on log₂FC ranking. Full pipeline details, parameter settings, and software versions are provided in the supporting information (Supplementary Methods 3.11).

2.6. Single‐cell RNA sequencing analysis

Day 30 forebrain organoids from control group (CTR), cholesterol treatment (CHOL), and cholesterol + semaglutide (CHOL+SEMA) groups (two biological replicates per group) were dissociated into single‐cell suspensions. High‐throughput single‐cell encapsulation, library preparation, and paired‐end sequencing were performed on the MGI DNBelab C‐TaiM4 and DNBSEQ‐T7 platforms (BGI/Innomics). Sequence data were aligned, quantified, and analyzed using Seurat. Quality‐filtered cells were normalized, dimensionally reduced via principal component analysis and uniform manifold approximation and projection (UMAP), clustered, and annotated based on canonical cell‐type markers. Cluster‐specific differential expression, pathway enrichment, and lineage trajectory analyses (via Slingshot) were subsequently performed. Detailed protocols for tissue dissociation, quality control thresholds, software versions, and trajectory parameters are provided in the supporting information (Supplementary Methods 3.12–3.14).

2.7. Calcium imaging analysis

Organoids were incubated with Fluo‐4 AM (37°C, 40 minutes in the dark) and washed with Hank's Balanced Salt Solution. Dynamic intracellular calcium changes were recorded via fluorescence microscopy. Time‐lapse movie stacks were processed using Fiji and custom Python scripts (Pandas/NumPy). Photobleaching was corrected via exponential fitting, and maximum intensity projections (MIPs) were generated. Individual neuronal somas (regions of interest [ROIs]) were segmented using Gaussian blur (σ = 2.0), auto‐thresholding, and Watershed algorithms. To eliminate artifacts from slow background drifts, the baseline fluorescence for each ROI was calculated dynamically. Specifically, F ₀(t) was defined as the 10th percentile of the intensity trace within a 60‐second rolling window. The relative fluorescence change was calculated as ΔF/F ₀ = (F[t] – F ₀[t]) / F ₀(t), with calcium transients defined by ΔF/F ₀ > 0.2. Network functional parameters were quantified: firing frequency (events per cell), active cell ratio (percentage of responsive cells), and synchronization index (mean Pearson correlation coefficient of ΔF/F₀ traces across all cell pairs).

2.8. Transcriptomic similarity and concordance analysis with human AD datasets

Differential expression between CHOL‐treated and CTR organoids was analyzed using DESeq2 (P _adj < 0.05) to define the organoid cholesterol‐effect signature. Human post mortem samples from GSE33000 were categorized into AD, non‐demented control, and Huntington's disease groups. Clinical samples were standardized using the gene‐wise median and median absolute deviation of the control group, and Spearman correlation between each sample's expression deviation and the organoid cholesterol‐effect signature was defined as the similarity score. Group differences were evaluated using multivariable linear regression adjusted for age and sex. Patient AD versus control differential expression was analyzed using limma, and directional concordance with organoid log₂FC values was evaluated for shared significant genes.

2.9. Transcriptomic similarity and concordance analysis with multi‐disease signatures

Stage‐associated cortical AD signatures were constructed from GSE131617 and GSE84422 using cohort‐specific covariate‐adjusted empirical‐Bayes moderated linear models followed by random‐effects meta‐analysis, and were independently evaluated in GSE118553. Three signatures meeting the prespecified validation criterion (FDR < 0.10)—Cortical_Early_AD_up, Cortical_Late_progression_up, and Cortical_Late_progression_down—were retained for the primary organoid analyses. Huntington's disease (GSE64810) and Parkinson's disease (GSE68719) reference signatures were constructed separately using DESeq2 Wald statistics, with the top 250 genes in each direction retained. To evaluate disease‐specific transcriptomic concordance while controlling for general injury responses, stress‐depleted signatures were generated by removing core Hallmark stress pathways and organoid stress Gene Ontology (GO) terms. Preranked GSEA, Spearman rank correlations, and rank–rank hypergeometric overlap (RRHO) analyses were conducted across organoid contrasts (CHOL vs. CTR, CHOL+SEMA vs. CHOL, and Interaction) using both original and stress‐depleted signatures. Detailed cohort specifications, meta‐analysis pipelines, stress‐depletion procedures, and sensitivity analyses are provided in the supporting information (Supplementary Methods 3.15–3.16).

2.10. Statistics and reproducibility

All statistical analyses were performed using GraphPad Prism (GraphPad Software Inc.), with sample sizes (n) and experimental replicates detailed in the respective figure legends. Quantitative data are presented as mean ± standard error of the mean. Statistical significance was defined as P < 0.05. Detailed statistical testing procedures and parameter selections are provided in the supporting information (Supplementary Methods 3.17).

3. RESULTS

3.1. Establishment and characterization of human forebrain organoids expressing GLP‐1R

We used our previously established protocol ²⁸ , ²⁹ to generate forebrain‐specific organoids from hiPSCs (Figure 1A). Organoids underwent neural induction, forebrain patterning, transient Matrigel embedding, and long‐term maturation in shaking culture.

FIGURE 1.

Establishment of a cholesterol‐induced metabolic injury model in human forebrain organoids. A, Schematic illustration of the differentiation protocol for generating human forebrain organoids from hiPSCs. (Diagram created with BioRender.com). B, Representative bright‐field images showing the morphological development of organoids at Day 2, 14, and 30. Scale bar, 500 µm. C–F, Immunofluorescence staining of forebrain organoids at Day 30. C, FOXG1 (red) confirms the forebrain regional identity. Scale bar, 100 µm. D, KI67 (green) marks proliferating cells. Scale bar, 100 µm. (E) SOX2 (red) identifies neural progenitor cells, while MAP2 (green) marks mature neurons. Scale bar, 100 µm. F, GLP1R (green) shows the distribution of GLP‐1R. Nuclei are stained with DAPI (blue). Scale bar, 50 µm. G, Cell viability of NPCs treated with varying concentrations of cholesterol (1 µg/mL, 5 µg/mL, 10 µg/mL, 15 µg/mL) for 3 days, measured by CCK‐8 assay to determine the optimal cytotoxic concentration. n = 4 independent experiments. Statistical significance was determined using one‐way analysis of variance. H–I, Evaluation of lipid accumulation. H, Representative Oil Red O staining showing lipid droplet accumulation (red) in CTR and CHOL (10 µg/mL) treated organoids, and hematoxylin (blue) counterstains nuclei. Scale bar, 50 µm. I, Quantification of lipid droplet area normalized to nuclear area. n = 5 independent experiments. Unpaired Welch t test was used for statistical analysis. J, KEGG pathway enrichment analysis reveals signaling and functional pathways damaged by cholesterol overload, identified through RNA sequencing. K, Sankey diagram (left) showing the mapping of cholesterol‐responsive genes to the indicated functional categories. Dot plot (right) showing the gene ratio, gene count, and enrichment significance for each category. L, Preranked gene set enrichment analysis of cholesterol‐treated organoids against cortical AD signatures. Profiles for batch‐adjusted CHOL versus CTR differential expression showing significant positive enrichment for early AD up and late progression up signatures, alongside significant negative enrichment for the late progression down signature. Vertical dashed lines indicate the peak enrichment score. AD, Alzheimer's disease; cAMP, cyclic adenosine 3',5'‐monophosphate; CCK‐8, Cell Counting Kit‐ 8; CHOL, cholesterol treatment; CTR, control group; DAPI, 4′,6‐diamidino‐2‐phenylindole; FDR, Benjamini–Hochberg adjusted false discovery rate; FOXG1, forkhead box protein G1; GLP‐1R, glucagon‐like peptide‐1 receptor; hiPSC, human‐induced pluripotent stem cell; KEGG, Kyoto Encyclopedia of Genes and Genomes; MAP2, microtubule associated protein 2; NES, normalized enrichment score; NPC, neural progenitor cell; SOX2, sex‐determining region Y box 2.

The organoids showed progressive morphogenesis with neuroepithelial rosettes becoming pronounced by Day 30 (Figure 1B). Immunofluorescence confirmed a forebrain fate at Day 30 through strong expression of the transcription factor forkhead box protein G1 (FOXG1; Figure 1C). Furthermore, the organoids established cortical‐like architecture with Ki‐67⁺ proliferating cells forming a ventricular‐like zone (VZ; Figure 1D), sex‐determining region Y box 2 (SOX2⁺) neural progenitors residing in inner layers, and microtubule associated protein 2 (MAP2⁺) neurons occupying outer regions (Figure 1E), confirming established apical–basal polarity within the neuroepithelium.

To verify the model's suitability for investigating semaglutide effects, we confirmed robust GLP‐1R expression in Day 30 organoids (Figure 1F). We additionally established hiPSC‐derived 2D forebrain NPCs as a complementary 2D model for cytotoxicity assessment (Figure S1A–E in supporting information). Collectively, these models provide a validated system for investigating the direct neurological impacts of semaglutide.

3.2. Semaglutide effects on forebrain organoids under physiological conditions

Semaglutide safety was first evaluated in 2D NPCs, in which CCK‐8 assays showed no cytotoxicity from 10 nM up to 10 µM after 3‐ or 7‐day exposure (Figure S1F). Based on this safety profile and previous reports of GLP‐1R–mediated neuroprotection, 100 nM semaglutide was selected for chronic treatment from Day 14 to Day 30 in forebrain organoids. Semaglutide‐treated (SEMA) organoids maintained normal morphology, intact rosette structures, and showed no significant alterations in average radius or perimeter compared to CTR (Figure S2A–S2C in supporting information). Additionally, cell proliferation (Ki‐67) and apoptosis (cleaved caspase‐3) levels remained unchanged (Figure S2D‐S2F). These results confirm that semaglutide is well tolerated and does not disrupt neurodevelopmental homeostasis.

Then, quantitative polymerase chain reaction analysis revealed that semaglutide significantly elevated mRNA levels of SOX2 (P < 0.001), MAP2 (P < 0.05), and the antioxidant regulator NFE2L2 (P < 0.05), while GLP‐1R levels remained stable, ensuring that the observed downstream effects were not biased by fluctuations in receptor expression (Figure S2G–S2J). GO and Kyoto Encyclopedia of Genes and Genomes (KEGG) analyses further showed enrichment in DNA replication and neurogenesis pathways (Figure S2K–S2L). Notably, increased neurogenic signatures occurred alongside reduced synaptic transmission pathways, suggesting enhanced progenitor commitment with delayed functional maturation.

3.3. Cholesterol overload induces cellular stress and selected AD‐associated transcriptomic signatures

To investigate cholesterol‐induced metabolic stress, we first screened NPCs with a water‐soluble methyl‐β‐cyclodextrin (MβCD)–cholesterol complex at 1, 5, 10, and 15 µg/mL. This range was selected based on physiologically relevant cholesterol levels in humans. ³⁰ , ³¹ , ³² CCK‐8 assays identified 10 µg/mL as the optimal concentration for inducing metabolic insult, as it significantly reduced NPC viability (P < 0.01; Figure 1G). In 3D organoids, the CHOL group (10 µg/mL) induced extensive lipid droplet accumulation (P < 0.01; Figure 1H and 1I) and severely disrupted organoid architecture, characterized by loss of neuroepithelial rosettes and reduced size, indicating impaired morphogenesis (Figure 2B).

FIGURE 2.

Semaglutide ameliorates cholesterol‐induced apoptosis via GLP‐1R activation. A, Mitigation of cholesterol‐induced cytotoxicity by semaglutide. Cell viability of NPCs was measured after co‐treatment with 10 µg/mL cholesterol and varying concentrations of semaglutide (10 nM, 100 nM, 1 µM, 10 µM) for 3 days. n = 6 independent experiments. B–D, Semaglutide effectively mitigates cholesterol‐induced morphological defects. B, Representative bright‐field images showing the morphological changes of forebrain organoids in CTR, CHOL, and CHOL+SEMA groups on Day 30. Scale bar, 500 µm. Quantification of the average radius (C) and perimeter (D) of organoids on Day 30. n = 13 organoids per group. E–G, Mechanistic verification of GLP‐1R dependence. E, Schematic illustration of the GLP‐1R antagonist (C5d) blocking semaglutide action, diagram created with BioRender.com. F, Representative live/dead cell staining of forebrain organoids using calcein‐AM (green) for live cells and propidium iodide (PI, red) for dead cells. Scale bar, 250 µm. G, Viability was quantified as the ratio of calcein‐AM–positive cells (Live) to the total cell count (Live + PI‐positive cells). n = 6 independent experiments. H–J, Semaglutide inhibits the intrinsic apoptosis pathway. H, Representative western blot bands showing the expression of key apoptosis‐related proteins: cleavedcaspase‐3 (17 kDa), Bcl‐2 (26 kDa), and β‐actin (42 kDa) as a loading control. I–J, Quantification of the relative protein levels of Cleaved‐caspase‐3 (I) and Bcl‐2 (J) normalized to β‐actin. n = 4 independent experiments. K–M, Semaglutide protects NPC integrity and suppresses apoptosis. K, Representative immunofluorescence images of organoids at Day 30 stained for KI67 (green, proliferating cells), cleaved‐caspase‐3 (red, apoptosis marker), and SOX2 (magenta, NPC marker). Nuclei are stained with DAPI (blue). Scale bar, 200 µm. L–M, Quantification of the SOX2 (L) and cleaved‐caspase‐3 (M) positive area, normalized to the total DAPI area. n = 7 independent experiments. N–P, Semaglutide protects proliferative capacity and neuronal architecture. N, Representative immunofluorescence images showing organoids stained for KI67 (green, proliferating cells) and MAP2 (magenta, mature neurons). Scale bar, 150 µm. O–P, Quantification of the KI67‐positive area (O) and the MAP2‐positive area (P), both normalized to the total DAPI area. n = 8 independent experiments. All data are presented as mean ± standard error of the mean. Statistical significance was determined using one‐way analysis of variance. ns, not significant. *P  <  0.05, **P  <  0.01, ***P  <  0.001, ****P  <  0.0001. CHOL, cholesterol treatment; CHOL+SEMA, cholesterol + semaglutide treatment; CTR, control group; DAPI, 4′,6‐diamidino‐2‐phenylindole; GLP‐1R, glucagon‐like peptide‐1 receptor; MAP2, microtubule associated protein 2; NPC, neural progenitor cell; SOX2, sex‐determining region Y box 2.

RNA sequencing revealed widespread cellular stress after cholesterol overload. KEGG analysis showed activation of p53 and apoptosis pathways, accompanied by suppression of synaptic vesicle cycle and glutamatergic synapse pathways (Figure 1J). These changes were associated with lipid metabolic dysregulation, including increased expression of apolipoprotein E (APOE) and ABCA1 and reduced expression of endogenous steroid biosynthesis genes (Figure 1K and Figure S3 in supporting information).

To evaluate clinical relevance, we calculated individual‐subject similarity scores between organoid transcriptomes and human clinical cohorts. CHOL‐treated organoids displayed significant transcriptomic similarity to patients with neurodegenerative diseases, particularly AD (P = 1.56 × 10⁻³²; Figure S4A in supporting information). We also observed high gene‐level concordance of differential expression between CHOL organoids and AD‐like profiles (Spearman P = 0.4960; Figure S4B).

To test the specificity and directional concordance of these alterations, preranked GSEA was performed using independently validated disease signatures. In the CHOL group, all validated cortical AD programs were directionally concordant, displaying significant enrichment for the early cortical AD upregulated signature (normalized enrichment score [NES] = 1.39, FDR = 0.0183) and late progression upregulated signature (NES = 1.74, FDR = 0.0034), alongside suppression of the late progression downregulated signature (NES = −1.87, FDR = 0.0082; Figure 1L). These enrichments persisted after removal of genes belonging to the predefined core and expanded cellular‐stress sets (Figures S5–S6 in supporting information), supporting an AD‐associated transcriptional component that could not be fully explained by the predefined generic‐stress genes.

However, cholesterol overload also enriched Huntington's disease upregulated (NES = 2.11, FDR = 0.0034) and downregulated signatures (NES = −1.63, FDR = 0.0079), as well as the Parkinson's disease upregulated signature (NES = 2.05, FDR = 0.0034; Figure S7 in supporting information). These comparative analyses indicate that cholesterol‐induced transcriptional changes are not specific to AD and may reflect broader neurodegenerative disease–related responses.

3.4. Semaglutide alleviates cholesterol‐induced neuronal injury and suppresses apoptosis via GLP‐1R activation

Following the establishment of the cholesterol‐induced metabolic injury model, we evaluated the neuroprotective effects of semaglutide. Compared to the CHOL group, co‐treatment with 100 nM semaglutide (CHOL+SEMA) significantly improved cell viability, preserved organoid size (radius and perimeter; P < 0.01), and maintained rosette structures (Figure 2A–2D).

To verify that this protection is mediated by GLP‐1R activation, we performed pharmacological blockade by GLP‐1R antagonist compound 5d (C5d; ³³ Figure 2E). Live/dead staining revealed that while semaglutide effectively prevented cholesterol‐induced cell death (P < 0.01), co‐treatment with C5d completely abolished this protective effect (P < 0.0001; Figure 2F–2G). To further characterize the molecular basis of this neuroprotection, western blot analysis showed that CHOL+SEMA significantly reduced the apoptotic effector cleaved caspase‐3 (P < 0.05) and increased the anti‐apoptotic protein Bcl‐2 (P < 0.01; Figure 2H–2J).

Immunofluorescence demonstrated enhanced progenitor maintenance, with increased SOX2⁺ area and Ki‐67⁺ cells (both P < 0.01; Figure 2K–2M and 2O). Semaglutide partially preserved the MAP2⁺ area (P < 0.05; Figure 2N and 2P).

3.5. Semaglutide preserves mitochondrial integrity and lipid metabolic homeostasis to support neural survival

Given the link between Bcl‐2/caspase‐3 regulation and mitochondrial apoptosis, we next examined mitochondrial integrity. Mito‐Tracker staining revealed an interconnected filamentous mitochondrial network in CTR organoids, whereas cholesterol exposure induced extensive mitochondrial fragmentation (Figure 3A and 3B). Semaglutide treatment maintained mitochondrial integrity and significantly reduced intracellular ROS levels (P < 0.001; Figure 3C and 3D), consistent with improved mitochondrial function.

FIGURE 3.

Semaglutide activates GLP‐1R signaling to preserve mitochondrial integrity and lipid metabolic homeostasis. A–B, Semaglutide restores mitochondrial integrity. A, Representative immunofluorescence images of forebrain organoids stained with Mito‐Tracker (yellow, active mitochondria) and Hoechst 33342 (blue, nuclei). Bottom: magnified views of the boxed regions. Scale bar, 50 µm. B, Quantification of mitochondrial content, presented as the ratio of Mito‐Tracker area to the nuclear area. n = 3 independent experiments. C–D, Semaglutide alleviates cholesterol‐induced oxidative stress. C, Representative immunofluorescence image showing intracellular ROS levels (detected by DCFH‐DA probe, green), with Hoechst (blue). Scale bar, 50 µm. D, Quantification of intracellular ROS levels measured by the DCFH‐DA probe. Bottom: magnified views of the boxed regions. n = 6 independent experiments. E–F, Transcriptomic analysis of semaglutide‐mediated neuroprotection. E, KEGG pathway analysis comparing the CHOL+SEMA and CHOL groups. F, Gene expression profiles of representative genes involved in lipid metabolism (e.g., ABCA1, ABCG1) and LDs regulation (e.g., PLIN2, PLIN3). G–H, Semaglutide attenuates cholesterol‐induced lipid droplet accumulation. G, Representative immunofluorescence images of forebrain organoids stained for nuclei (DAPI, blue), lipid droplets (LD, green), and neurons (TUJ1, magenta). Right: magnified views of the indicated boxed regions. Scale bars, 40 µm. H, Quantification of the number of LDs per cell across the indicated groups. n = 5 independent experiments. I–J, Activation of the PI3K–AKT–mTOR signaling pathway. I, Representative western blot bands showing the total and phosphorylated levels of mTOR (289 kDa), PI3K (85 kDa), AKT (65 kDa), and downstream S6 (32 kDa), with β‐actin (42 kDa) as a loading control. J, Quantitative analysis of the relative levels of phosphorylated proteins: p‐PI3K (n = 5), p‐AKT (n = 4), p‐mTOR (n = 4), and p‐S6 (n = 4), all normalized to β‐actin. Here, n represents the number of independent biological experiments performed using distinct organoid batches. K, Schematic illustration of semaglutide‐mediated GLP‐1R downstream signaling, highlighting the cAMP–PKA and PI3K–AKT–mTOR pathways involved in mitochondrial stabilization, cholesterol metabolism, and neuronal growth. Diagram created with BioRender.com. All data are presented as mean ± standard error of the mean. Statistical significance was determined using one‐way analysis of variance. ns, not significant. *P  <  0.05, **P  <  0.01, ***P  <  0.001, ****P  <  0.0001. AKT, protein kinase B; cAMP, cyclic adenosine 3',5'‐monophosphate; CHOL, cholesterol treatment; CHOL+SEMA, cholesterol + semaglutide treatment; CTR, control group; DAPI, 4′,6‐diamidino‐2‐phenylindole; GLP‐1R, glucagon‐like peptide‐1 receptor; KEGG, Kyoto Encyclopedia of Genes and Genomes; LD, lipid droplet; mTOR, mechanistic target of rapamycin; PI3K, phosphoinositide 3 kinase; PKA, protein kinase A; ROS, reactive oxygen species.

To elucidate the molecular mechanisms of semaglutide's protection, we performed KEGG enrichment analysis (CHOL+SEMA vs. CHOL; Figure 3E). KEGG analysis revealed enrichment of neuroactive ligand–receptor interaction, cyclic adenosine 3',5'‐monophosphate (cAMP) signaling, and steroid biosynthesis pathways, with suppression of p53 signaling. The cAMP–protein kinase A (PKA) pathway was activated, supported by increased ADCY5, PRKACB, and RAPGEF4 expression, together with phosphoinositide 3 kinase (PI3K)–protein kinase B (AKT) pathway activation (Figure S8‐S9 in supporting information).

Given the role of PI3K–AKT–mechanistic target of rapamycin (mTOR) in lipid homeostasis and survival, we examined related metabolic genes. Cholesterol overload upregulated cholesterol transporters ABCA1 and ABCG1, and lipid droplet (LD)‐coating proteins PLIN2 and PLIN3. Semaglutide reduced LD accumulation confirmed by immunofluorescence (P < 0.05) and visualized by label‐free imaging (Figure 3F–H and Figure S10 in supporting information).

Western blot analysis further showed that the CHOL group markedly reduced both the expression and phosphorylation of PI3K, AKT, and mTOR (Figure 3I–3J and Figure S11 in supporting information). Semaglutide restored PI3K–AKT–mTOR signaling and increased phosphorylation of downstream S6. To determine the relationships among these pathways, we introduced the PI3K inhibitor LY294002 (5 µM) and the PKA inhibitor H‐89 (5 µM) into the model (Figure S12 in supporting information). LY294002 profoundly abrogated the semaglutide‐induced phosphorylation of the entire PI3K–AKT–mTOR–S6 cascade. Furthermore, the addition of H‐89 also attenuated this cascade, suggesting that PKA may act upstream of the PI3K–AKT–mTOR pathway in this context. Synergistic cAMP–PKA and PI3K–AKT–mTOR response may support neuronal growth, cholesterol metabolism, and mitochondrial stabilization under metabolic stress (Figure 3K).

3.6. Single‐cell analysis reveals altered intercellular communication networks in forebrain organoids

To investigate the cell type–specific effects of semaglutide, we performed single‐cell RNA sequencing (scRNA‐seq) on Day 30 organoids. UMAP clustering identified eight major cell populations: radial glial cells (RGs), dividing cells (DCs), dorsal progenitor cells (DPCs), intermediate progenitor cells (IPCs), immature cortical neurons (CN1s), mature cortical neurons (CN2s), interneurons (INs), and Cajal–Retzius cells (CRs; Figure 4A). Cluster identities were validated by canonical markers such as SLC1A3 (RG), MKI67 (DC), and EOMES (IPC; Figure 4B). Although overall cell composition remained comparable, cholesterol exposure reduced DPC and CR populations (Figure S13A in supporting information) and altered genes associated with lipid metabolism (ABCA1, ABCG1) and synaptic function (RELN, NRXN1), which were partially restored by semaglutide (Figure S13B).

FIGURE 4.

Semaglutide remodels intercellular communication and restores midkine signaling. A–B, Identification of major cell populations in forebrain organoids. UMAP visualization of single‐cell RNA‐sequencing data and dot plot showing representative marker genes defining dividing cells (DCs), radial glial cells (RGs), dorsal progenitor cells (DPCs), intermediate progenitor cells (IPCs), immature cortical neurons (CN1s), mature cortical neurons (CN2s), interneurons (INs), and Cajal–Retzius cells (CRs). C–D, Differentiation trajectory analysis across CTR, CHOL, and CHOL+SEMA conditions. UMAP projections with inferred lineage trajectories and pseudotime analysis illustrating transcriptional dynamics along neural progenitor differentiation and neuronal maturation. E, Global comparison of network size and strength. Bar plots summarize the total number of inferred interactions (left) and the overall interaction strength (right) for each experimental group. F, Shifts in significant ligand–receptor interaction proportions. Heatmaps display the log₂ fold change of the percentage of significant ligand–receptor pairs (sig%) for each sender–receiver pair, comparing CHOL versus CTR and CHOL+SEMA versus CHOL. G–H, Chord diagrams illustrating cell–cell communication mediated by midkine signaling (G) and Wnt signaling (H) pathways under CTR, CHOL, and CHOL+SEMA conditions. I, Pathway‐level communication dynamics for MDK signaling. Bubble plots highlight representative ligand‐receptor pairs within the MDK pathway and compare their communication probabilities. Bubble size denotes the communication probability, and the color gradient indicates the direction and magnitude of the change (log₂ fold change) between conditions. CHOL, cholesterol treatment; CHOL+SEMA, cholesterol + semaglutide treatment; CTR, control group; MDK, midkine; UMAP, uniform manifold approximation and projection.

Pseudotime trajectory analysis identified three transcriptional states (P1–P3) characterized by sequential activation of lineage‐associated programs, including RG/cell‐cycle signatures and neuronal maturation/electrophysiological signatures (Figure 4C–D and Figure S14 in supporting information). Cholesterol exposure disrupted these transcriptional transitions, whereas semaglutide promoted a shift toward more mature neuronal states (CN1 and CN2). CellChat analysis demonstrated reduced intercellular communication in the CHOL group, with total interactions decreasing from 7623 to 6849, whereas semaglutide improved network connectivity and strength (Figure 4E and Figure S15A in supporting information). Differential analysis further revealed increased significant interaction proportions (sig%) in the CHOL+SEMA group, particularly among DPC and CN2 populations, while key communication hubs involving DPC and CN1 were preserved (Figure 4F and Figure S15B).

Pathway analysis identified midkine (MDK) signaling as a neurotrophic pathway associated with neuronal survival and repair. MDK‐mediated intercellular communication was reduced by cholesterol stress but restored by semaglutide, with increased MDK–nucleolin and MDK–low density lipoprotein receptor‐related protein 1 interactions involving DPC, CN1, CN2, and CR populations (Figure 4G and 4I). Wnt signaling alterations induced by cholesterol were also partially rescued by semaglutide (Figure 4H and Figure S15C). Together, these findings indicate that semaglutide restores disrupted cellular communication networks and preserves neuronal homeostasis under cholesterol stress, particularly through the MDK signaling axis.

3.7. Semaglutide exerts partial neuroprotection by modulating metabolic stress and neuronal activity in Day 80 organoids

To determine whether semaglutide's neuroprotective effects persist in more mature neural tissue, we evaluated late‐stage forebrain organoids (Days 65–80). Cholesterol exposure significantly reduced organoid area, whereas semaglutide did not restore overall size (Figure 5A and 5B). Western blot analysis showed that cholesterol increased amyloid precursor protein (P < 0.01) and p‐tau217 (P < 0.05) levels, while semaglutide did not significantly reduce these neurodegeneration‐associated markers (Figure 5C‐5E).

FIGURE 5.

Semaglutide exerts limited neuroprotective effects in Day 80 forebrain organoids. A–B, Morphological evaluation of late‐stage forebrain organoids. Organoids were treated for 15 days (from Day 65 to Day 80) and analyzed on Day 80. A, Representative bright‐field images. Scale bar, 1 mm. B, Quantification of organoid area on Day 80. n = 11 organoids per group. C–E, Western blot analysis of AD‐associated pathological proteins. Representative bands show the expression of APP (≈ 110 kDa), p‐tau217 (≈ 55 kDa), and total tau (≈ 55 kDa), with β‐actin (42 kDa) as a loading control. D–E, Quantitative analysis of (D) relative APP levels and (E) p‐tau217/tau ratio. n = 4 independent experiments. F–H, Representative immunofluorescence images of MAP2 (green) and DAPI (blue) in Day 80 organoids. Middle panels show the MAP2 channel only, and lower panels show high‐magnification views of the boxed regions. Scale bar, 50 µm. G, Quantification of the ratio of MAP2‐positive area to DAPI‐positive area. n = 6 independent experiments. H, Quantification of MAP2⁺ fragmented branch density, calculated as the number of MAP2⁺ fragmented branches normalized to MAP2⁺ area. n = 6 independent experiments. I–M, Calcium imaging analysis of neuronal network activity in Day 80 organoids. I, Schematic of calcium imaging workflow and representative Fluo‐4 AM raw data. Scale bar, 50 µm. J, Representative heatmaps showing calcium transients (ΔF/F₀) in individual ROIs (n = 40 cells per organoid) over a 5‐minute recording period. K–M, Quantitative analysis of (K) spontaneous calcium firing frequency (events/cell), (L) synchronization index quantified by the mean Pearson correlation coefficient of ΔF/F ₀ traces across all cell pairs within each organoid, and (M) percentage of active versus inactive cells. n = 120 ROIs from n = 3 organoids across three independent differentiations. Statistical comparisons were performed based on the biological replicates (n = 3) rather than technical ROIs. All data are presented as mean ± standard error of the mean. Statistical significance was determined using one‐way analysis of variance. ns, not significant. *P  <  0.05, **P  <  0.01, ***P  <  0.001, ****P  <  0.0001. AD, Alzheimer's disease; APP, amyloid precursor protein; CHOL, cholesterol treatment; CHOL+SEMA, cholesterol + semaglutide treatment; CTR, control group; DAPI, 4′,6‐diamidino‐2‐phenylindole; MAP2, microtubule associated protein 2; p‐tau, phosphorylated tau; ROI, region of interest.

At the transcriptional level, cholesterol suppressed HMGCS1 and increased CASP3 and SOD1 expression (Figure S16 in supporting information). Semaglutide partially normalized these stress‐related gene changes, indicating continued modulation of metabolic and oxidative pathways despite limited effects on established pathology.

MAP2 staining showed that semaglutide significantly reduced MAP2⁺ fragmented branch density compared to the CHOL group (P < 0.05), although MAP2⁺ area was not restored (Figure 5F–H). To provide functional evidence, we performed calcium imaging using the Fluo‐4 AM indicator to measure intracellular calcium transients (Figure 5I). Cholesterol exposure drove neurons into a state of aberrant firing, characterized by increased cellular firing frequency (P < 0.0001) and reduced synchronization index (P < 0.0001; Figure 5J–M). Semaglutide normalized cellular firing frequency, while recovery of cellular synchronization remained limited.

4. DISCUSSION

In this study, we used a human forebrain organoid system as a model of cholesterol‐induced metabolic stress and examined the protective effects of the GLP‐1R agonist semaglutide. Cholesterol exposure induced molecular and functional disturbances that overlapped with selected AD‐associated and shared neurodegeneration‐related transcriptomic signatures, without recapitulating the full complexity of AD. Cholesterol overload also impaired cellular homeostasis and functional integrity. Semaglutide activated GLP‐1R and downstream cAMP–PKA and PI3K–AKT–mTOR signaling, which may support cellular homeostasis under metabolic stress. These effects were accompanied by partial preservation of calcium activity patterns and intercellular communication.

The neurotoxicity of excess cholesterol in this model appears to arise from lipid imbalance and subsequent bioenergetic failure. CHOL organoids exhibited severe lipid overload, accompanied by increased cholesterol efflux (e.g., ABCA1, ABCG1) and suppressed endogenous biosynthesis (e.g., SREBF2, HMGCR). However, these responses were insufficient to resolve the accumulation. At the subcellular level, excess cholesterol disrupts mitochondrial membrane fluidity and potential, ³⁴ leading to oxidative stress and fragmentation of the mitochondrial network. This bioenergetic collapse affected metabolically demanding cell populations (e.g., DPC and CR cells). ³⁵ , ³⁶ , ³⁷ Although these structural vulnerabilities emerged during early neural organization, lipid‐induced bioenergetic failure may represent a shared stress mechanism across several neurological contexts. ³⁸ The cross‐disease transcriptomic overlap observed here therefore indicates limited disease specificity. Although early‐life perturbations have been proposed to influence later AD risk, ³⁹ the present findings do not establish such a life‐course relationship. Instead, dysregulated cholesterol homeostasis in this system provides a context for investigating selected AD‐associated consequences of metabolic stress.

Mechanistically, semaglutide exerts its protective effects through GLP‐1R activation, which engages multiple downstream signaling cascades. KEGG pathway mapping revealed coordinated enrichment of cAMP–PKA and PI3K–AKT signaling, and pharmacological perturbation supported the involvement of both pathways in the response to semaglutide. Activation of cAMP–PKA signaling may contribute to mitochondrial stabilization, as PKA critically regulates mitochondrial bioenergetics and dynamics to reduce ROS levels. ⁴⁰ In parallel, activation of the PI3K–AKT–mTOR axis may facilitate lipid regulation and mitigate lipid accumulation. ⁴¹ , ⁴² The inhibitor responses are also consistent with functional crosstalk between PKA and PI3K–AKT signaling, although they do not establish a strict upstream–downstream relationship. Because brain insulin resistance and dysregulated PI3K–AKT signaling are associated with AD, ⁴³ while aberrant cAMP–PKA signaling can influence tau phosphorylation, ⁴⁴ these findings identify semaglutide‐responsive pathways that may intersect with AD‐associated metabolic dysfunction.

Beyond intracellular signaling, maintenance of metabolic homeostasis may help preserve intercellular communication. Single‐cell analysis indicated that cholesterol exposure disrupted signaling hubs involving DPC and CN1/CN2. Semaglutide partially restored predicted communication patterns, including signaling niches associated with cellular resilience. MDK‐mediated interactions have been linked to neural survival, synaptic plasticity, and repair under severe stress. ⁴⁵ , ⁴⁶ These observations suggest that GLP‐1R activation influenced the intercellular signaling environment, although the inferred communication changes require direct functional validation.

Protective effects were also observed in more mature Day 80 organoids, in which intrinsic hypoxic gradients and necrotic‐core formation increased baseline vulnerability. ⁴⁷ , ⁴⁸ , ⁴⁹ Under these conditions, semaglutide showed limited capacity to reverse early AD‐associated molecular alterations (e.g., APP and p‐tau217). However, it continued to modulate markers of metabolic stress (e.g., HMGCS1, CASP3) and oxidative responses (e.g., SOD1), and functionally attenuated prodromal‐like hyperactivity (aberrant neuronal firing frequency). Because Fluo‐4 AM is not neuron specific, these calcium signals are interpreted as cellular activity recorded from neuronal‐enriched regions. Compared to Day 30 organoids, the partial response in Day 80 organoids suggests that semaglutide acts primarily as a metabolic stabilizer in this model and has limited capacity to reverse established amyloid or tau progress.

Despite the protective effects observed here, several limitations of this study warrant consideration. First, this forebrain organoid system models cholesterol‐induced metabolic stress rather than the full etiology or progression of AD. It lacks AD‐relevant genetic backgrounds, microglia, functional vasculature, and aging‐associated features that shape late‐onset disease. ⁵⁰ , ⁵¹ Second, exogenous cholesterol overload induces direct lipotoxicity but bypasses systemic lipoprotein transport dynamics in vivo. ³⁰ , ⁵² Third, the high semaglutide concentrations used for in vitro mechanistic proof of concept, together with uncertainty regarding blood–brain barrier penetration in humans, ⁵³ , ⁵⁴ limit direct translational inference. Future work should incorporate loss‐of‐function studies (e.g., conditional genetic knockouts) and AD‐relevant genetic backgrounds (e.g., APOE variants with isogenic controls), to define genotype‐specific responses. Vascularized and immune‐competent assembloids, alongside in vivo models, will be required to determine whether these stress‐response mechanisms extend to more AD‐relevant systems and evaluate central drug‐delivery strategies. ⁵⁵

In conclusion, these findings highlight the importance of cholesterol homeostasis in maintaining neuronal integrity, and position GLP‐1R signaling as a candidate protective axis under metabolic stress. By alleviating lipotoxicity, semaglutide offers valuable mechanistic insights into therapeutic strategies for AD and broader neurodegenerative disorders.

AUTHOR CONTRIBUTIONS

Shaohua Ma conceived and supervised the work. Feifei Yu, Jiani Xu, Honghui Zheng, Yibo Hou, Lingyi Guo, and Wenxin Xu performed the experiments. Feifei Yu, Jiani Xu, Yibo Hou, and Davit Khutsishvili analyzed the bulk RNA‐seq data. Jiani Xu, Yibo Hou, and Cheng Hai performed the single‐cell data analysis. Feifei Yu, Jiani Xu, and Lingyi Guo generated the figures. Feifei Yu, Jiani Xu, Davit Khutsishvili, Yibo Hou, Bailing Chen, Zitian Wang, and Shaohua Ma wrote the manuscript.

CONFLICT OF INTEREST STATEMENT

The authors declare no conflicts of interest. Author disclosures are available in the Supporting Information.

CONSENT STATEMENT

Not applicable. This study does not involve human participants.

Supporting information

Supporting Information: alz71888‐sup‐0001‐ICMJE.pdf

Supporting Information: alz71888‐sup‐0002‐SuppMat.pdf

DATA AVAILABILITY STATEMENT

Requests for further information and resources should be directed to, and will be fulfilled by the lead contact, Shaohua Ma (ma.shaohua@sz.tsinghua.edu.cn). All unique reagents generated in this study are available from the lead contact with a completed materials transfer agreement. The bulk RNA‐seq and scRNA‐seq data generated in this study have been deposited in the Gene Expression Omnibus (GEO) under accession numbers GSE324518 and GSE329738, respectively. All data are publicly available as of the date of publication. All other data supporting the findings of this study are available from the lead contact upon request.