Upregulation of SLC38A2 by melatonin attenuates hippocampal ferroptosis in Alzheimer’s disease

Background

Ferroptosis has been implicated in the pathogenesis of Alzheimer’s disease (AD), yet its upstream neuronal regulators remain unclear. This study investigated the mechanisms by which melatonin exerts neuroprotection, focusing on the glutamine transporter solute carrier family 38 member 2 (SLC38A2), which inhibits ferroptosis in AD.

Methods

AD models included C57BL/6 mice injected intracerebroventricularly with beta-amyloid (Aβ)_1-42 oligomers, organotypic hippocampal slices treated with Aβ_1-42 oligomers, and HT22 hippocampal neurons treated with hydrogen peroxide. Cognition and hippocampal pathology were assessed by open field test, Morris water maze (MWM), shuttle box tests, and histology. Cross-species multi-omics comprised single-cell RNA sequencing of human hippocampus, mouse spatial transcriptomics, human bulk RNA sequencing and proteomics. Protein changes were measured by immunohistochemistry, immunofluorescence and Western blotting. Biochemical assays quantified glutathione (GSH), malondialdehyde (MDA), iron, and reactive oxygen species (ROS).

Results

In Aβ_1-42 oligomer-injected C57BL/6 mice, melatonin improved cognition in MWM and shuttle box tests without affecting locomotion, reduced hippocampal neuronal loss, and lowered Aβ/Tau/p-Tau. Single-cell RNA sequencing of human hippocampus and spatial transcriptomics in the mouse brain demonstrated pronounced ferroptosis in the AD hippocampal region. Integrative human proteomics, bulk RNA sequencing and spatial transcriptomics prioritized the downregulation of the glutamine-family amino-acid catabolic process and identified SLC38A2 as a key node. In vivo and in organotypic hippocampal slices, melatonin reversed the Aβ_1-42 oligomer-induced ferroptosis signature. Immunohistochemistry, immunofluorescence, and Western blotting showed that melatonin upregulated SLC38A2 in the Aβ-treated hippocampus and slices. Functionally, shRNA-mediated SLC38A2 knockdown in HT22 neurons reduced GSH and glutathione peroxidase 4 (GPX4), increased iron and ROS, enhanced erastin-induced ferroptosis, and abolished the protective effect of melatonin.

Conclusions

These findings identify SLC38A2 as a pivotal regulator of hippocampal ferroptosis and demonstrate that melatonin confers neuroprotection by stabilizing the SLC38A2-GSH-GPX4 axis, supporting SLC38A2 as a therapeutic target in AD.

Supplementary Information

The online version contains supplementary material available at 10.1186/s13195-026-01990-0.

Highlights

SLC38A2 downregulation promotes ferroptosis in Alzheimer’s disease.

Melatonin restores SLC38A2 to suppress hippocampal ferroptosis.

SLC38A2 is essential for melatonin’s neuroprotection against ferroptosis.

Supplementary Information

The online version contains supplementary material available at 10.1186/s13195-026-01990-0.

Introduction

Alzheimer’s disease (AD) is the most prevalent neurodegenerative disorder characterized by progressive cognitive decline and neuronal loss. Despite decades of research, effective disease-modifying therapies remain limited [1]. While beta-amyloid (Aβ) and hyperphosphorylated tau are hallmark pathologies [2], the mechanisms leading to neuronal death are complex and not fully understood. Identifying the specific cell death pathways associated with Aβ is therefore crucial for developing targeted therapeutic strategies.

Recent evidence underscores a critical role of ferroptosis—an iron-dependent regulated cell death driven by excessive lipid peroxidation—in AD pathogenesis [3, 4]. Mechanistically, ferroptosis reflects an imbalance between oxidative lipid damage and antioxidant defense, and is hallmarked by collapse of the glutathione (GSH)-dependent detoxification system, leading to the accumulation of lethal lipid reactive oxygen species (ROS) [5]. A central protective module is the GSH-glutathione peroxidase 4 (GPX4) axis: GPX4 uses GSH as a reducing cofactor to convert phospholipid hydroperoxides into non-toxic lipid alcohols, thereby terminating lipid peroxidation chain reactions [6]. In parallel, GSH availability is constrained by amino-acid transport (cysteine/cystine, glutamate/glutamine, and glycine) and connected metabolic pathways [7, 8]. Conversely, the accumulation of lethal lipid peroxides is driven by the iron-dependent Fenton reaction and the peroxidation of polyunsaturated fatty acids (PUFAs), a process often catalyzed by enzymes such as acyl-CoA synthetase long chain family member 4 (ACSL4) and lipoxygenases (ALOXs) [9–11].

In AD brains, amyloidogenic processing of APP and subsequent Aβ accumulation have been shown to impair iron efflux and trigger ROS-mediated GSH depletion, thereby creating a permissive environment for ferroptosis [2]. Mechanistically, this ferroptotic cascade drives the irreversible destruction of membrane lipids, leading to synaptic failure and profound neuronal loss in brain regions essential for cognition [12, 13]. Consistent with a pathogenic role of ferroptosis in AD, molecular markers of ferroptosis, including elevated iron and lipid peroxidation and reduced GPX4, strongly correlate with the severity of disease pathology in both human patients and animal models [14, 15]. Consequently, inhibitors of ferroptosis, such as Ferrostatin-1, have demonstrated the ability to reduce neuronal death and improve memory in AD models [16]. These findings suggest that targeting the molecular machinery of ferroptosis represents a promising avenue for AD intervention.

Melatonin, a neurohormone primarily secreted by the pineal gland, is a potent endogenous antioxidant with high blood-brain barrier permeability [17]. Beyond its regulation of circadian rhythms, melatonin exerts robust neuroprotective effects by scavenging free radicals and upregulating antioxidant enzymes, such as SOD, GPx, and catalase [18, 19]. Although melatonin has been shown to reduce Aβ burden and improve cognition [17, 20], its potential as a specific inhibitor of ferroptosis in AD remains under-investigated. Notably, emerging evidence has already substantiated the anti-ferroptotic capacity of melatonin in other pathological contexts. For instance, melatonin was reported to attenuate traumatic brain injury and sleep deprivation-induced cognitive impairment by inhibiting neuronal ferroptosis via the regulation of lipid peroxidation and iron homeostasis [21, 22]. In a sleep deprivation model, melatonin alleviated cognitive deficits by restoring hippocampal iron homeostasis and upregulating GPX4 via MT2/ERK/Nrf2 signaling [23]. Given these precedents, melatonin is logically positioned as a promising candidate to intercept the ferroptotic cascade in AD. However, the precise molecular targets—particularly regarding upstream metabolic regulation—through which melatonin exerts this anti-ferroptotic effect in AD neurons require further elucidation.

Solute carrier family 38 member 2 (SLC38A2) is a sodium-coupled neutral amino acid transporter that mediates cellular uptake of glutamine and other neutral amino acids [24]. Glutamine is a key metabolic precursor for glutamate, which is indispensable for GSH synthesis [25]. A recent study further suggested that SLC38A2-mediates the uptake of cysteine, glycine, and glutamine may help maintain redox homeostasis and intracellular GPX4 activity, thereby conferring anti-ferroptotic capacity [26]. Importantly, recent studies have linked SLC38A2 to hippocampal vulnerability in AD [27, 28]. Collectively, these findings suggest that SLC38A2 may be a key factor influencing neuronal ferroptosis in AD-relevant neurons.

Therefore, the objectives of this study were: (i) to determine whether early melatonin treatment can attenuate Aβ-induced hippocampal neuronal injury and ferroptosis; (ii) to explore how SLC38A2 is altered during AD and to define its impact on neuronal ferroptosis, and (iii) to test whether SLC38A2 modulation is required for the anti-ferroptotic and neuroprotective effects of melatonin. By addressing these questions, our study provides a mechanistic rationale for targeting an upstream amino-acid transport step to reinforce the core ferroptosis defense machinery in AD.

Materials and methods

Animals

Male C57BL/6 mice (6 months old, 28 ± 1 g) were obtained from the Laboratory Animal Center of the Second Affiliated Hospital of Harbin Medical University. Mice were housed under standard conditions (12-h light/dark cycle) with ad libitum access to chow and water and were acclimated for 1 week before experimentation. All procedures were approved by the Experimental Animal Ethics Committee of Harbin Medical University (Protocol No. HMUIRB2023018) and conducted in accordance with the NIH Guide for the Care and Use of Laboratory Animals (Publication No. 85 − 23, revised 1996).

In vivo study design and grouping

Mice were randomly assigned to one of four groups (n = 5/group) using an Excel-generated random number table. The sample size was determined based on previous studies utilizing similar AD models [29, 30] to balance statistical validity with ethical guidelines for minimizing animal use. The groups were defined as follows: (1) Control, no intervention; (2) Sham, intracerebroventricular (ICV) injection of vehicle; (3) Aβ, ICV injection of oligomeric Aβ₁₋₄₂ + intraperitoneal (IP) saline; and (4) Aβ + melatonin, ICV injection of oligomeric Aβ₁₋₄₂ + IP melatonin.

Model establishment and drug administration

The experimental timeline was anchored to the acclimatization period (Days 1–7). On Day 8, AD modeling was performed via ICV injection of Aβ₁₋₄₂ oligomers which were prepared following a previously established protocol [31]. Briefly, lyophilized Aβ₁₋₄₂ peptide (A118755, Aladdin, Shanghai, China) was initially dissolved in ice-cold 1,1,1,3,3,3-hexafluoro-2-propanol (HFIP) to a concentration of 1 mM and vortexed to ensure complete monomerization. The solvent was then evaporated under vacuum to yield a peptide film, which was stored at −80 °C. Prior to injection, the peptide film was resuspended in DMSO to 5 mM and subsequently diluted with sterile PBS to a final concentration of 150 µM. This solution was incubated at 4 °C for 12 h to facilitate the formation of soluble Aβ₁₋₄₂ oligomers before use.

Mice were anesthetized with isoflurane (045742, Yizejia, Beijing, China) inhalation (2.0–2.5.0.5% for induction and 1.0–1.5.0.5% for maintenance) and positioned in a stereotaxic apparatus. Following scalp preparation, a microsyringe was inserted into the lateral ventricle at the following coordinates relative to the bregma: AP = −0.94 mm, ML = ±1.7 mm, DV = −2.0 mm [32]. Each mouse in the Aβ groups received a bilateral ICV injection of the Aβ₁₋₄₂ oligomer solution (4 µL/side). The Sham group received an equivalent volume of vehicle.

Pharmacological treatment commenced 24 h post-surgery (Day 9). Melatonin (10 mg/kg) or physiological saline was administered via IP injection daily in the afternoon for three weeks. The selection of this dosage and treatment duration was based on previously established protocols demonstrating the efficacy of melatonin in ameliorating AD-related pathology [33, 34].

Behavioral testing and sampling

Behavioral assessments were conducted from Day 19 to Day 29 (starting 11 days post-surgery), including the open field test (OFT), Morris water maze (MWM), and shuttle box test. The entire experimental period from model establishment to endpoint was 21 days. Brain tissues were harvested for biochemical analysis on Day 30, 24 h after the conclusion of behavioral testing. To minimize confounding factors, the order of treatments, behavioral testing, and tissue collection was randomized, and cage positions were rotated weekly. Investigators conducting behavioral assessments and histological analyses were blinded to group allocation.

Western blot

The Western blot was conducted as described previously [35]. The whole cellular protein was isolated from the mouse hippocampus or HT22 cells. A quantity of 10 µg of protein was separated using 12% SDS-PAGE, followed by transfer onto PVDF membranes. The membranes were blocked with TBST containing 5% nonfat milk for 2 h at room temperature. The membranes were then subjected to an overnight incubation at 4 °C with rabbit polyclonal antibodies against Aβ (sc-9129, SANTA Cruz, CA, USA), Tau (sc-5587, Santa Cruz, CA, USA), p-Tau (82568-1-RR, Proteintech, IL, USA), GPX4 (DF6701, Affinity, Jiangsu, China), ACSL4 (DF12141, Affinity, Jiangsu, China), arachidonate 15-lipoxygenase (ALOX15) (sc-133085, SANTA Cruz, CA, USA), SLC38A2 (DF4673, Affinity, Jiangsu, China), and β-actin (E-AB-40338, Elabscience, Wuhan, China). After washing, the membranes were incubated with HRP-conjugated goat-anti-rabbit IgG for 1 h at room temperature. The immuno-band signal was detected using ECL reagent (29050, Engreen, Beijing, China) and the Tanon Imaging System (Tanon, Shanghai, China). The relative optical density values relative to β-actin were quantified using ImageJ v1.46r.

Cell culture and treatment

HT22 cells, derived from the hippocampus of mice (SCC129, Merck KGaA, Darmstadt, Germany), were cultured in minimum essential medium supplemented with 10% fetal bovine serum (164210, Pricella, Wuhan, China) and antibiotics (streptomycin at 100 U/mL and penicillin at 100 U/mL, PB180120, Pricella, Wuhan, China). The cells were maintained at 37 °C in a humidified incubator with a 5% CO₂ and 95% air mixture. The cells were sub-culturing or subjected to pharmacological treatment upon reaching 90% confluence. The cells were treated for 24 h with 200 µM H₂O₂ (BD5025, Qiaosun Fine Chemical, Guangdong, China), 50 µM melatonin (M813985, Macklin, Shanghai, China), 2–20 µM erastin (HY-15763, MedChemExpress, NJ, USA), or 5–40 µM ferrostatin-1 (Fer-1) (HY-100579, MedChemExpress, NJ, USA).

Intracellular Fe²⁺ staining

HT22 cells were seeded in 24-well plates. After drug treatment, the cells were incubated at 37 °C with 5% CO₂ for 24 h. The medium was then replaced by 1 µM FerroOrange (F374, Dojindo, Kumamoto, Japan), and the cells were incubated at 37 °C with 5% CO₂ for 30 min. The cells were then stained with Hoechst 33,342 (P0133, Beyotime Biotechnology, Shanghai, China) at room temperature for 5 min. The fluorescence was observed and photographed using a laser confocal microscope.

The quantification of ROS and SA within cells

HT22 cells were grown in 24-well plates to 90% confluence. ROS and superoxide anion (SA) detection was performed using fluorescent probes DCFH-DA (E004-1-1, Nanjing Jiancheng Bioengineering Institute, Nanjing, China) and dihydroethidium (DHE) (50102ES02, Yeasen, Shanghai, China), following the manufacturer’s instructions. Fluorescence was directly observed and captured using a fluorescence microscope (Nikon, Tokyo, Japan).

Single-cell transcriptome data processing

The single-cell transcriptome is derived from GEO datasets GSE175814. Based on the sample information, the samples were classified into two groups: normal group and the AD group. The R package “Seurat” (v 4.4.1) was employed to process the single-cell data. All functions were executed using default parameters, unless otherwise instructed. Low-quality cells which with fewer than 200 detected features or more than 20% mitochondrial features were excluded. Potential doublets were removed by filtering out cells that had more than three median absolute deviations (MADs) of expressed genes. To mitigate batch effects, we utilized the IntegrateLayers with Harmony algorithm to integrate the gene-cell matrices derived from different samples. After batch effect correction, a shared nearest-neighbor graph was generated using the FindNeighbors function with the parameter: dims = 1:30, and clusters were identified using the FindClusters function. The resulting clusters were visualized using the Uniform Manifold Approximation and Projection (UMAP). We annotated the clusters into specific cell types by comparing the cluster-specific genes with canonical cell markers.

Spatial transcriptome data analysis

The spatial transcriptome data is derived from GEO datasets GSE222981. We obtained Spatial transcriptome datasets from the Gene Expression Omnibus. The gene expression count matrices and corresponding histology images for the Spatial transcriptome dataset were downloaded and processed using the R package Seurat. The raw count matrices were normalized using the function SCTransform in Seurat. The subsequent processing procedures are consistent with the single-cell transcriptome data.

RNA-seq data processing

RNA-seq data were obtained from the GEO datasets GSE28146, GSE29378, GSE5281, GSE48350, and GSE184942. A DGEList object was constructed using the edgeR package, and lowly expressed genes (retaining only genes expressed in at least 50% of the samples) were filtered out for subsequent analyses. The voom function in the limma package was applied to transform the count data into normally distributed expression values suitable for linear model analysis, while simultaneously estimating sample weights to correct for variance. A linear model (lmFit) was then constructed to analyze the association between gene expression and disease status (AD vs. Normal). Group comparison matrices (AD-Normal) were set using makeContrasts, and the empirical Bayes method (eBayes) was used to shrink standard errors and improve statistical power. Genes with a P < 0.05 and a fold change > 0.26 were considered differentially expressed genes.

Gene set scoring

We used the AddmoduleScore function to score the Ferrdb.Driver and Ferrdb.Supprossor of different cell types. AddmoduleScore calculates the average expression levels of each cell types on single cell level, subtracted by the aggregated expression of control feature sets. All analyzed features are binned based on averaged expression, and the control features are randomly selected from each bin.

Proteomic data analysis

The proteomic data used in this study were obtained from the PXD014557 dataset and represent processed results from the original publication, comprising 5,311 proteins quantified across all 10 samples. Differential protein expression between AD and control groups was analyzed using the DEqMS package in R, which builds upon the limma framework and incorporates the number of peptide-spectrum matches/peptide when calculating t-statistics. Multiple hypothesis testing was conducted using the Benjamini-Hochberg procedure, with a false discovery rate (FDR) threshold set at 10% [36]. Proteins with P < 0.01 were considered significant, ranked by log₂ fold change (log₂FC), and the top 50 proteins were selected for subsequent functional enrichment analysis.

Other methods

Other detailed methods are described in the supplementary materials and methods.

Statistical analysis

All data were presented as mean ± SEM. The normality of data distribution was assessed using the Shapiro-Wilk test on residuals. For MWM escape latency across training days, data were analyzed using a two-way repeated-measures ANOVA (group × day), followed by Sidak’s multiple comparisons test. In addition, escape latency on the final training day (Day 5) was analyzed as a prespecified endpoint. Student’s t-test was used to determine significant differences between two groups if the data were normally distributed with equal variances; otherwise, Welch’s t-test was applied, or the Mann-Whitney U test was used for non-normally distributed data. For comparisons among multiple groups, one-way ANOVA with Tukey’s HSD post-hoc test was used when assumptions of normality and homogeneity of variances were met. When normality was acceptable but variances were unequal, Welch’s ANOVA followed by Dunnett’s T3 multiple comparisons test was applied. For non-normally distributed data, the Kruskal-Wallis test followed by Dunn’s multiple comparisons test was used. GraphPad Prism version 9.0.0 (GraphPad Software, San Diego, CA, USA) was used for statistical analysis. A P value less than 0.05 was considered statistically significant.

Results

Melatonin ameliorates Aβ-induced cognitive defects in C57BL/6 mice

We aimed to determine whether early melatonin intervention can suppress the progression of AD. After a 24 h ICV injection of Aβ₁₋₄₂ oligomers, melatonin was administered continuously, and relevant assessments were performed at certain time points (Fig. 1A). In the OFT, total locomotor distance and locomotor distance in the center area were comparable across groups (P = 0.6869 and P = 0.2817, respectively; Fig. 1B).

Fig. 1.

Melatonin ameliorates Aβ-induced cognitive defects in C57BL/6 mice. A In vivo study design diagram. Mice received bilateral intracerebroventricular (ICV) injection of beta-amyloid (Aβ)₁₋₄₂ oligomers (or vehicle) followed by daily melatonin (Mel.) administration. Behavioral assessments (open field test (OFT), Morris water maze (MWM), and shuttle box test) were then performed, followed by tissue and serum collection for downstream analyses. Experimental groups included Control, Sham (ICV vehicle), Aβ (ICV Aβ_­1−42 + intraperitoneal (IP) saline), and Aβ + Mel. (ICV Aβ_­1−42 + IP Mel.), as indicated. B Assessment of locomotor activity and anxiety-like behavior in the OFT. Representative movement trajectories are shown (red color intensity indicates longer dwell time). Left: representative movement trajectoriess (warmer colors indicate longer dwell time). Right: quantification of total locomotor distance and locomotor distance in the center area. Total locomotor distance: Kruskal-Wallis, H(3) = 1.48, P = 0.6869. Center locomotor distance: one-way ANOVA, F(3,16) = 1.391, P = 0.2817. C MWM training stage. Representative swim paths during training (warmer colors indicate longer dwell time). D MWM acquisition training stage. Escape latency across training days (Day 1-Day 5). Across days: two-way repeated-measures ANOVA (group × day), interaction F(12,64) = 1.633, P = 0.1046; day main effect F(3.110,49.76) = 33.97, P < 0.0001; group main effect F(3,16) = 2.025, P = 0.1509. Day 5: one-way ANOVA with Tukey’s HSD, F(3,16) = 3.3357, P = 0.0460; Aβ vs. Aβ + Mel., P = 0.0366. E MWM test stage. Representative swim paths with the platform removed (warmer colors indicate longer dwell time). F Latency to first enter the platform zone during MWM probe test. One-way ANOVA with Tukey’s HSD, F(3,16) = 11.56, P = 0.0003; sham vs. Aβ, P = 0.0005; Aβ vs. Aβ + Mel., P = 0.0034. G Number of platform crossings during MWM probe test. One-way ANOVA with Tukey’s HSD, F(3,16) = 6.612, P = 0.0041; sham vs. Aβ, P = 0.0237; Aβ vs. Aβ + Mel., P = 0.0034. H Swimming speed during MWM probe test. One-way ANOVA, F(3,16) = 1.234, P = 0.33. I Shuttle box test. Active avoidance latency. One-way ANOVA with Tukey’s HSD, F(3,16) = 9.336, P = 0.0008; sham vs. Aβ, P = 0.0084; Aβ vs. Aβ + Mel., P = 0.0009. J Shuttle box test. Number of active avoidance responses. One-way ANOVA with Tukey’s HSD, F(3,16) = 7.108, P = 0.0030; sham vs. Aβ, P = 0.0147; Aβ vs. Aβ + Mel., P = 0.0029. K Shuttle box test. Percentage of active avoidance responses. One-way ANOVA with Tukey’s HSD, F(3,16) = 7.108, P = 0.0030; sham vs. Aβ, P = 0.0147; Aβ vs. Aβ + Mel., P = 0.0029. Data are presented as mean ± SEM, with individual symbols representing individual mice. n = 5 mice/group. *P < 0.05, *P < 0.01. See also Supplementary Table S2 for statistical details

The MWM test was used to assess the effects of melatonin on cognitive function. During the training stage, escape latency decreased robustly across training days (P < 0.0001), while neither the overall group effect nor the group × day interaction reached significance (P = 0.1509 and P = 0.1046, respectively; Fig. 1C, D). However, analysis of Day 5 specifically revealed that melatonin treatment significantly reduced escape latency compared to the Aβ group (P = 0.0366; Fig. 1D). In the test stage, Aβ-treated mice exhibited increased escape latency and decreased platform crossings compared to the sham group (P = 0.0005 and P = 0.0237, respectively), whereas these deficits were alleviated by melatonin treatment (P = 0.0034 for latency; P = 0.0034 for crossings; Fig. 1E-G). Swimming speeds remained similar across groups (P = 0.3300), indicating no overt motor impairment (Fig. 1H).

In the shuttle box test, Aβ-treated mice showed longer active avoidance latency (P = 0.0084; Fig. 1I), and performed fewer times and percentage of active avoidance responses (both P = 0.0147; Fig. 1J, K) than the sham group. Melatonin significantly improved avoidance latency (P = 0.0009; Fig. 1I), and increased both the number and percentage of active avoidance responses compared with the Aβ group (both P = 0.0029; Fig. 1J, K).

Overall, these behavioral assessments indicate that melatonin mitigates Aβ-associated cognitive impairments without altering general locomotor performance.

Melatonin attenuates Aβ-induced pathologies in the hippocampus of C57BL/6 mice

Given the critical role of the hippocampus in cognitive function, we examined the neuroprotective effects of melatonin on hippocampal pathology.The Aβ group displayed marked neuronal damage, evidenced by a disorganized arrangement, irregular cell morphology, and reduced neuronal density (P < 0.0001 for both CA1 and CA3; Fig. 2A, B). Melatonin treatment markedly alleviated these changes and increased Nissl-positive neuronal counts relative to the Aβ group (P < 0.0001 for both CA1 and CA3; Fig. 2A, B). Consistent with these observations, H&E staining showed similar neuroprotective effects (Supplementary Fig. S1).

Fig. 2.

Melatonin attenuates Aβ-induced pathologies in the hippocampus of C57BL/6 mice. A Representative Nissl staining images of the hippocampus, with higher-magnification views of the CA1, CA3, and dentate gyrus (DG) regions. Scale bars: 500 μm (whole hippocampus), 50 μm (CA1 and CA3), and 100 μm (DG). B Quantification of Nissl-positive neurons in CA1, CA3, and DG, presented as fold change relative to the Control group. Each symbol represents one section; n = 9/group (3 sections/mouse, 3 mice/group). CA1: one-way ANOVA, F(3,32) = 49.52, P < 0.0001; Tukey’s HSD post-hoc: Control vs. Sham, P = 0.0616; Sham vs. Aβ, P < 0.0001; Aβ vs. Aβ + Mel., P < 0.0001. CA3: Welch’s ANOVA, W = 91.34, df = 3.000, 16.52, P < 0.0001; Dunnett T3 post-hoc: Sham vs. Aβ, P < 0.0001; Aβ vs. Aβ + Mel., P < 0.0001. DG: one-way ANOVA, F(3,32) = 2.955, P = 0.0472. C Representative immunohistochemical staining for Aβ in the hippocampus with higher-magnification views of CA1, CA3, and DG. Scale bars: 500 μm (whole hippocampus) and 100 μm (CA1/CA3/DG). n = 3 mice/group. D Representative immunohistochemical staining for Tau in the hippocampus with higher-magnification views of CA1, CA3, and DG. Scale bars: 500 μm (whole hippocampus) and 100 μm (CA1/CA3/DG). n = 3 mice/group. E Western blot analysis of Aβ, total Tau, and phosphorylated Tau (p-Tau, Ser202/Thr205) in the RIPA-soluble hippocampal fraction, with β-actin as a loading control, followed by densitometric quantification. Protein levels were normalized to β-actin and expressed relative to the Control group. Each symbol represents one mouse; n = 5 mice/group. Aβ: one-way ANOVA, F(3,16) = 40.50, P < 0.0001; Tukey’s HSD post-hoc: Sham vs. Aβ, P < 0.0001; Aβ vs. Aβ + Mel., P < 0.0001. Tau: one-way ANOVA, F(3,16) = 8.746, P = 0.0012; Tukey’s HSD post-hoc: Sham vs. Aβ, P = 0.0042; Aβ vs. Aβ + Mel., P = 0.0062. p-Tau (Ser202/Thr205): one-way ANOVA, F(3,16) = 8.078, P = 0.0017; Tukey’s HSD post-hoc: Sham vs. Aβ, P = 0.0042; Aβ vs. Aβ + Mel., P = 0.0082. Data are presented as mean ± SEM. *P < 0.05, **P < 0.01. See also Supplementary Table S2 for statistical details

Immunohistochemistry further showed increased hippocampal Aβ and tau immunoreactivity in the Aβ group, which appeared reduced by melatonin treatment (Fig. 2C, D). Western blot analysis corroborated these findings, revealing significantly elevated levels of Aβ, total Tau, and phosphorylated Tau (p-Tau at Ser202/Thr205) in the soluble protein fraction of the Aβ group that were significantly attenuated by melatonin (Aβ: sham vs. Aβ, P < 0.0001; Aβ vs. Aβ + Mel., P < 0.0001; Tau: sham vs. Aβ, P = 0.0042; Aβ vs. Aβ + Mel., P = 0.0062; p-Tau: sham vs. Aβ, P = 0.0042; Aβ vs. Aβ + Mel., P = 0.0082; Fig. 2E).

Collectively, these results indicate that melatonin treatment effectively suppresses Aβ-induced neuronal loss and reduces pathological Aβ and Tau proteins in the hippocampus.

Single-cell and Spatial transcriptomics reveal pronounced ferroptosis in the AD hippocampus

We next examined whether ferroptosis is engaged in the AD hippocampus. Single-cell technology enables high-resolution and comprehensive investigation of gene expression and cell-specific changes. We analyzed a human AD hippocampal single-cell RNA sequencing dataset (GSE175814). After quality control, 24,584 cells were retained. The batch-corrected transcriptomic data were then subjected to dimensionality reduction and clustering using UMAP. Clusters were annotated based on canonical marker genes, resulting in the identification of eight major hippocampal cell types (Fig. 3 A, B).

Fig. 3.

Single-cell and spatial transcriptomics reveal pronounced ferroptosis in the AD hippocampus. A Uniform manifold approximation and projection (UMAP) plot of cells from GSE175814 human hippocampal single-cell RNA-seq dataset after quality control and integration, color coded to indicate the major cell lineages. B Bubble heatmap displaying the average expression and expressed percentage of representative markers of each cell lineages. Dot size indicates fraction of expressing cells, colored based on normalized expression levels. C Differentially expressed ferroptosis Driver and Suppressor genes in AD identified by integrating five bulk transcriptomic datasets (GSE184942, GSE28146, GSE29378, GSE48350, and GSE5281) and intersecting Alzheimer‘s disease (AD)-associated DEGs with FerrDb V2 database. Triangles denote ferroptosis Driver genes and circles denote ferroptosis Suppressor genes; red indicates upregulation in AD and blue indicates downregulation in AD. Differential expression in each bulk dataset was computed using a limma-style moderated (Bayesian) t-test. Genes meeting|log₂FC| ≥ 0.58 (fold change ≥ 1.5) and P < 0.05 were included. D Boxplots of “FerrDb Driver” module scores across major cell lineages in GSE175814, comparing AD versus control cells within each lineage. For the box plots, the centre line shows the median, the box limits show the first and third quartiles, the upper and lower whiskers extend from the hinge to the largest or the lowest value no further than 1.5 × the interquartile range (IQR) from the hinge. Two-sided Wilcoxon rank-sum tests (AD vs control)/lineage: Endothelial, P = 8.629 × 10⁻⁵; SMC, P= 0.0130; Astrocyte, P = 3.621 × 10⁻¹⁶; Oligodendrocyte, P= 0.517; OPC, P = 0.309; Microglia, P = 0.00335; Inhibitory neuron, P = 0.00264; Excitatory neuron, P = 5.305 × 10⁻¹⁸.E Boxplots of “FerrDb Suppressor” module scores across major cell lineages in GSE175814, comparing AD versus control cells within each lineage (boxplot elements defined as in panel D). Two-sided Wilcoxon rank-sum tests (AD vs control)/lineage: Endothelial, P = 0.0288; SMC, P = 1.460 × 10⁻¹⁷; Astrocyte, P = 3.049 × 10⁻³⁷; Oligodendrocyte, P = 5.978 × 10⁻¹⁰; OPC, P = 8.390 × 10⁻¹⁷; Microglia, P= 1.700 × 10⁻⁶; Inhibitory neuron, P = 2.626 × 10⁻⁹; Excitatory neuron, P = 6.380 × 10⁻¹⁶. F Mouse spatial transcriptomics (GSE222981) showing spatial clustering (“molecular niches”) and hippocampal regional expression patterns of anti-ferroptotic genes (glutamate-cysteine ligase catalytic subunit (Gclc), glutamate-cysteine ligase modifier subunit (Gclm), and glutathione peroxidase 4 (Gpx4)). The hippocampal region was defined by anatomical correspondence to the H&E image and unsupervised clustering (cluster 4), and gene expression within this hippocampal region was compared between groups as indicated. Two-sided Wilcoxon rank-sum test: Gclc, P = 3.540537 × 10⁻¹⁵; Gclm,P = 1.832519 × 10⁻⁷; Gpx4, P = 0 in the output (below machine precision). See also Supplementary Table S2 for statistical details

To derive a ferroptosis-relevant gene set for single-cell scoring, we integrated five human AD bulk transcriptomic datasets (GSE184942, GSE28146, GSE29378, GSE48350, GSE5281) and identified highly variable genes, which were then cross-referenced with the ferroptosis database (FerrDb V2). This yielded 39 upregulated ferroptosis driver genes and 39 downregulated ferroptosis suppressor genes in AD. These 78 genes were subsequently used to quantify ferroptosis activity in single-cell data (Fig. 3 C).

We applied the AddModuleScore function to the GSE175814 single-cell dataset to evaluate the expression patterns of these ferroptosis driver and suppressor genes across different cell types. The results revealed significant upregulation of ferroptosis driver genes in astrocytes and excitatory neurons under AD conditions, while ferroptosis suppressor genes were markedly downregulated in astrocytes and excitatory neurons (both P < 0.0001; Fig. 3D, E). Given the abundance and vulnerability of excitatory neurons in the hippocampus, these signatures support a prominent ferroptosis-prone state in AD.

To further assess ferroptosis in the hippocampus during AD, we analyzed the mouse spatial transcriptomics dataset GSE222981. Based on HE staining images and unsupervised clustering of transcriptomic expression, cluster 4 was identified as corresponding to the hippocampal region. The results showed that anti-ferroptotic genes—glutamate-cysteine ligase catalytic subunit (Gclc), glutamate-cysteine ligase modifier subunit (Gclm), and Ggx4—were markedly downregulated in the hippocampal region of AD mice (all P ≤ 1.83 × 10^− 7; Fig. 3 F).

Integrative multi-omics identifies significant alterations in the glutamine metabolism pathway in AD and nominates SLC38A2 as a candidate regulator

To identify upstream pathways that may shape ferroptosis vulnerability in AD, we analyzed proteomic profiles from AD and control hippocampal samples (PXD014557; n = 5/group). The top 50 upregulated and top 50 downregulated proteins (ranked by logFC) were subjected to functional enrichment analysis using Metascape (www.metascape.org/). Based on the top five pathways ranked by -log₁₀(P value), the enrichment analysis revealed that the hippocampus exhibits upregulation of GSH transmembrane transport alongside downregulation of the glutamine family amino acid catabolic process in AD (P = 8.36 × 10^− 6, FDR q = 0.030; Fig. 4A), implicating a glutamine-glutathione axis that is tightly linked to ferroptosis regulation.

Fig. 4.

Multi-omics analyses highlight that glutamine amino acid catabolism is significantly altered in AD, and identify SLC38A2 as a potential regulator. A Pathway enrichment of dysregulated proteins in AD based on proteomic data (PXD014557; AD n = 5, Control n = 5). The top 50 upregulated and top 50 downregulated proteins (ranked by logFC) were submitted to Metascape for enrichment analysis, and the top 5 enriched terms are displayed as -log₁₀(P), with red bars indicating terms enriched among proteins upregulated in AD and blue bars indicating terms enriched among proteins downregulated in AD. PPathway enrichment: hypergeometric test; glutamine family amino acid catabolic process, P = 8.36 × 10^− 6, FDR q = 0.030. B Density plot showing the distribution of glutamine family amino acid catabolic process signature scores across major hippocampal lineages in the single-cell RNA-seq dataset (GSE175814). Scores were calculated at the single-cell level using an expression-based module-scoring approach (gene set derived from GO:0009065; e.g., FAH, GOT1, ASRGL1, FAHD1, FOLH1, PEPD). C Histogram and density plot of glutamine family amino acid catabolic process scores in excitatory neurons (Excit.Neuron), stratified by sample type (AD vs. Normal) in GSE175814. D-E Pearson correlation analysis between glutamine family amino acid catabolic process scores and ferroptosis-related gene sets in AD Excit.Neuron cells (GSE175814). D Correlation with ferroptosis suppressor genes; E correlation with ferroptosis driver genes. Black lines represent regression fits with 95% confidence intervals shown as gray bands. Two-sided P values were calculated using Pearson’s correlation test. F Functional enrichment network of genes downregulated in excitatory neurons with low ferroptosis suppressor scores (bottom quartile, AD) compared with those with high suppressor scores (top quartile, Normal) in GSE175814. The network summarizes the top enriched biological themes and highlights genes shared across the three leading terms; rectangular nodes indicate GO biological processes and circular nodes indicate differentially expressed genes. Node color indicates the number of pathways in which the gene appears: orange (3), blue (2), green (1). Solute carrier family 38 member 2 (SLC38A2), a glutamine transporter gene, was identified among the shared genes. Enrichment: hypergeometric test; top terms include modulation of chemical synaptic transmission (P = 3.12 × 10^− 45, FDR q = 3.92 × 10^− 41), Transmission across Chemical Synapses (P = 1.97 × 10^− 36, FDR q = 1.45 × 10^− 32), and vesicle-mediated transport in synapse (P = 1.14 × 10^− 35, FDR q = 6.28 × 10^− 32). G Spatial expression pattern of Slc38a2 in the hippocampal region of AD mouse brain from the GSE222981 spatial transcriptomics dataset. Unsupervised clustering and histological alignment were used to define the hippocampal region (cluster 4), and Slc38a2 expression was visualized within this region across groups. Group comparison: Wilcoxon rank-sum test, P = 1.70 × 10^− 25. See also Supplementary Table S2 for statistical details

We further evaluated the glutamine family amino acid catabolic process signature scores across different cell types at the single-cell level (GSE175814). Excitatory neurons (Excit.Neuron) displayed a multimodal distribution of pathway scores, suggesting substantial heterogeneity in this metabolic process among Excit.Neuron subpopulations (Fig. 4B). Comparison of this pathway between Excit.Neuron cells from AD and normal samples revealed significant differences in score distributions, indicating altered pathway activity in AD (Fig. 4C).

To determine whether this pathway is involved in ferroptosis regulation within excitatory neurons, we extracted Excit.Neuron cells from AD samples and assessed the Pearson correlation between genes involved in the glutamine family amino acid catabolic process and ferroptosis-related genes. The analysis revealed a significant positive correlation between the pathway and genes annotated as “FerrDb Suppressor” (Fig. 4D), but no correlation with ferroptosis driver genes (Fig. 4E). These results suggest that reduced activity of the glutamine family amino acid catabolic process may lead to decreased expression of ferroptosis-suppressor genes, potentially contributing to ferroptosis in excitatory neurons.

To further nominate candidate regulators, we contrasted excitatory neurons with high ferroptosis suppressor scores (top quartile) from normal samples against those with low scores (bottom quartile) from AD samples and performed differential expression analysis followed by enrichment analysis. Among the top three enriched functions/pathways, we identified 22 genes commonly involved across all three leading terms (Fig. 4F), including SLC38A2, a key glutamine transporter. Spatial transcriptomic analysis of the GSE222981 dataset further confirmed that Slc38a2 expression was significantly reduced in the hippocampal region of AD mice (P = 1.70 × 10^− 25; Fig. 4G).

Melatonin ameliorates hippocampal ferroptosis induced by Aβ

We further examined whether melatonin modulates ferroptosis in the mouse hippocampus in vivo. GPX4 immunoreactivity appeared reduced after Aβ injection and was visibly restored by melatonin treatment, as shown by GPX4 IHC in hippocampal subregions (CA1, CA3, and DG) and by GPX4 immunofluorescence in the hippocampus (Fig. 5A, B). Western blot analysis of hippocampal lysates showed that Aβ administration markedly reduced GPX4 levels compared with the sham group (P = 0.0016; Fig. 5C) and significantly increased ACSL4 (P = 0.0342; Fig. 5D) and ALOX15 levels (P < 0.0001; Fig. 5E), whereas melatonin treatment reversed these alterations relative to the Aβ group (GPX4, P = 0.0018; ACSL4, P = 0.0025; ALOX15, P < 0.0001; Fig. 5C-E).

Fig. 5.

Melatonin ameliorates ferroptosis induced by Aβ in the hippocampus.  A Representative immunohistochemical (IHC) staining of GPX4 in the hippocampus with higher-magnification views of CA1, CA3, and dentate gyrus (DG). Scale bars: 500 μm (whole hippocampus) and 100 μm (CA1/CA3/DG). n = 3 mice/group. B Representative immunofluorescence (IF) images of GPX4 in the hippocampus. Scale bar: 100 μm. n = 4 mice/group. C-E Western blot analysis of GPX4 (C), acyl-CoA synthetase long chain family member 4 (ACSL4) (D), and arachidonate 15-lipoxygenase (ALOX15) (E) in hippocampal lysates (β-actin as loading control) with densitometric quantification normalized to β-actin and expressed relative to Control. n = 5 mice/group. GPX4 : one-way ANOVA, F(3,16) = 10.17, P = 0.0005; Tukey’s HSD: Sham vs. Aβ, P = 0.0016; Aβ vs. Aβ + Mel., P = 0.0018. ACSL4: one-way ANOVA, F(3,16) = 6.801, P = 0.0036; Tukey’s HSD: Sham vs. Aβ, P = 0.0342; Aβ vs. Aβ + Mel., P = 0.0025. ALOX15: one-way ANOVA, F(3,16) = 24.17, P < 0.0001; Tukey’s HSD: Sham vs. Aβ, P < 0.0001; Aβ vs. Aβ + Mel., P < 0.0001. F Hippocampal glutathione (GSH) levels measured using a commercial assay kit and normalized to protein content (µg/mg protein). n = 5 mice/group. One-way ANOVA, F(3,16) = 6.446, P = 0.0046; Tukey’s HSD: Sham vs. Aβ, P = 0.0094; Aβ vs. Aβ + Mel., P = 0.0271. G Hippocampal malondialdehyde (MDA) levels measured using a commercial assay kit and normalized to protein content (nmol/mg protein). n = 5 mice/group. One-way ANOVA, F(3,16) = 42.64, P < 0.0001; Tukey’s HSD: Sham vs. Aβ, P < 0.0001; Aβ vs. Aβ + Mel., P < 0.0001. H Representative dual immunofluorescence staining images showing GPX4 (green) and ALOX15 (red) with DAPI (blue) in organotypic hippocampal slice cultures (OHSCs) treated with Aβ (500 µM) or Aβ + melatonin (200 µM) for 24 h. Scale bar: 50 μm. n = 3/group. Data are presented as mean ± SEM. *P < 0.05, **P < 0.01. See also Supplementary Table S2 for statistical details

Consistent with the protein changes, hippocampal GSH levels were decreased and malondialdehyde (MDA) levels were increased in the Aβ group (P = 0.0094 and P < 0.0001, respectively), and melatonin significantly restored GSH and reduced MDA compared with Aβ alone (P = 0.0271 and P < 0.0001, respectively; Fig. 5F, G). Similar trends were observed in serum, where Aβ reduced GSH levels and increased MDA levels, whereas melatonin effectively inhibited these systemic ferroptotic changes (Supplementary Fig. S2).

To further validate these effects, organotypic hippocampal slice cultures (OHSCs) were treated with Aβ (500 µM) or Aβ + melatonin (200 µM) for 24 h. Dual-immunofluorescence staining revealed that Aβ decreased GPX4 expression and concurrently increased ALOX15 expression. However, melatonin administration mitigated these effects (Fig. 5H).

Melatonin upregulates SLC38A2 in Aβ-treated hippocampus andorganotypic slices

We then tested whether melatonin acts by regulating SLC38A2. In the mouse hippocampus, SLC38A2 immunoreactivity appeared reduced after Aβ injection across CA1, CA3, and DG subregions and was visibly restored by melatonin (Fig. 6 A). Quantification of IHC signals confirmed a significant reduction in SLC38A2 in the Aβ group compared with the sham group in CA1, CA3, and DG (all P < 0.0001), whereas melatonin significantly attenuated this decrease (Aβ vs. Aβ + melatonin, all P < 0.0001; Fig. 6B). Consistently, western blot results exhibited a significant reduction of SLC38A2 in the Aβ group (P < 0.0001), which was significantly counteracted by melatonin (P < 0.0001; Fig. 6 C).

Fig. 6.

Melatonin upregulates SLC38A2 expression in Aβ-treated hippocampus and organotypic slices. A Representative immunohistochemical (IHC) staining of SLC38A2 in the hippocampus with higher-magnification views of CA1, CA3, and dentate gyrus (DG). Scale bars: 500 μm (whole hippocampus) and 50 μm (CA1/CA3/DG). B Quantification of SLC38A2 IHC signal in CA1, CA3, and DG, expressed as relative SLC38A2 expression (IHC). Each symbol represents one section; n = 9/group (3 sections/mouse, 3 mice/group), quantified using ImageJ. CA1: one-way ANOVA, F(3,32) = 19.20, P< 0.0001; Tukey’s HSD: Sham vs Aβ, P < 0.0001; Aβ vs Aβ+Mel., P < 0.0001. CA3: one-way ANOVA, F(3,32) = 20.19, P < 0.0001; Tukey’s HSD: Sham vs Aβ, P < 0.0001; Aβ vs Aβ+Mel., P < 0.0001. DG: one-way ANOVA, F(3,32) = 15.12, P < 0.0001; Tukey’s HSD: Sham vs Aβ, P< 0.0001; Aβ vs Aβ+Mel., P < 0.0001. C Western blot analysis of SLC38A2 in the hippocampus (β-actin as loading control) and expressed relative to Control. n = 5 mice/group. One-way ANOVA, F(3,16) = 32.18, P < 0.0001; Tukey’s HSD: Sham vs Aβ, P < 0.0001; Aβ vs Aβ+Mel., P < 0.0001. D Representative immunofluorescence images of SLC38A2 (red) with DAPI (blue) in OHSCs treated with Aβ or Aβ+melatonin for 24 h. Scale bar: 50 μm. E Quantification of SLC38A2 immunofluorescence intensity in OHSCs, calculated in ImageJ using corrected total fluorescence and normalized to the Control group. Each symbol represents one slice; n = 9/group. One-way ANOVA, F(2,24) = 12.59, P = 0.0002; Tukey’s HSD: Control vs Aβ, P = 0.0003; Aβ vs Aβ+Mel., P = 0.0015. Data are presented as mean ± SEM. All tests were two-sided. *P < 0.05,**P < 0.01. See also Supplementary Table S2 for statistical details

To further evaluate the effects of melatonin on SLC38A2, OHSCs were treated with Aβ or Aβ + melatonin for 24 h. Immunofluorescence analysis demonstrated that Aβ treatment markedly decreased SLC38A2 levels compared to the control group (P < 0.01; Fig. 6D, E). However, melatonin treatment significantly upregulated SLC38A2 expression in these slices (P < 0.01; Fig. 6D, E).

SLC38A2 knockdown induces ferroptosis in HT22 hippocampus cells

To validate the role of SLC38A2 in regulating ferroptosis, HT22 cells were transfected with short hairpin RNA (shRNA) targeting Slc38a2. Western blot analysis confirmed that shRNA-Slc38a2 led to a marked decrease in SLC38A2 expression compared to the shRNA-negative control (NC) group (P = 0.0114; Fig. 7A). These findings were further supported by immunofluorescence staining, which showed a similar reduction in SLC38A2 signal intensity (Fig. 7B).

Fig. 7.

SLC38A2 knockdown induces ferroptosis in HT22 hippocampus cells. A Western blot analysis of SLC38A2 with densitometric quantification normalized to β-actin and expressed relative to Control. n = 14/group. One-way ANOVA, F = 7.105, df = 2,39, P = 0.0023; Tukey’s HSD: shRNA-NC vs. shRNA-Slc38a2, P = 0.0114. B Representative immunofluorescence images of SLC38A2 (red) and DAPI (blue) in HT22 cells. Scale bar: 10 μm. n = 4/group. C Western blot analysis of GPX4 with densitometric quantification normalized to β-actin and expressed relative to Control. n = 13/group. One-way ANOVA, F = 39.75, df = 2,36, P < 0.0001; Tukey’s HSD: shRNA-NC vs. shRNA-Slc38a2, P = 0.0106. D Representative immunofluorescence images of GPX4 (red) and DAPI (blue) in HT22 cells. Scale bar: 10 μm. n = 4/group. E Cell viability measured by CCK-8 assay and expressed as relative viability. n = 5/group. One-way ANOVA, F = 72.85, df = 2,12, P < 0.0001; Tukey’s HSD: shRNA-NC vs. shRNA-Slc38a2, P < 0.0001. F Intracellular reduced GSH content measured by a reduced GSH assay kit. n = 10/group. One-way ANOVA, F = 23.33, df = 2,27, P < 0.0001; Tukey’s HSD: shRNA-NC vs. shRNA-Slc38a2, P = 0.0006. G Total cellular iron measured by an iron assay kit. n = 8/group. One-way ANOVA, F = 4.710, df = 2,21, P = 0.0204; Tukey’s HSD: shRNA-NC vs. shRNA-Slc38a2, P = 0.0479. H Representative fluorescence images of Fe²⁺ levels detected using the FerroOrange probe (orange) with Hoechst 33,342 nuclear staining (blue). Scale bar: 10 μm. n = 4/group. I Representative fluorescence microscopy images of reactive oxygen species (ROS) (green) and superoxide anion (SA) staining (red) in HT22 cells. Scale bar: 100 μm. n = 6/group. J Erastin dose-response. HT22 cells were treated with erastin (2–20 µM) for 24 h and viability was measured by CCK-8. n = 5/group. One-way ANOVA, F = 162.1, df = 10,44, P < 0.0001; Tukey’s HSD pairwise comparisons versus Control are summarized in Supplementary Table S2 (e.g., Control vs. 2 µM erastin, P = 0.0001; Control vs. 4 µM erastin, P = 0.0495; higher doses as listed). K Fer-1 rescue of erastin cytotoxicity. HT22 cells were treated with 16 µM erastin together with Fer-1 (5–40 µM) for 24 h, and viability was measured by CCK-8. n = 5/group. One-way ANOVA, F = 39.88, df = 9,45, P < 0.0001; Tukey’s HSD: Control vs. 20 µM erastin, P < 0.0001; 20 µM erastin vs. Fer-1 (5–40 µM)+erastin, all P < 0.0001 (dose-wise comparisons listed in Supplementary Table S2). L SLC38A2 knockdown exacerbates erastin toxicity and weakens Fer-1 protection. HT22 cells were transfected with or without shRNA-Slc38a2 and then treated with 16 µM erastin or erastin + 20 µM Fer-1; viability was measured by CCK-8. n = 5/group. One-way ANOVA, F = 405.0, df = 4,20, P < 0.0001; Tukey’s HSD: Control vs. erastin, P < 0.0001; erastin vs. erastin + Fer-1, P < 0.0001; erastin vs. shRNA-Slc38a2 + erastin, P < 0.0001; erastin + Fer-1 vs. shRNA-Slc38a2 + erastin + Fer-1, P < 0.0001. Data are presented as mean ± SEM. Significance is indicated on the graphs (*P < 0.05, **P < 0.01). See also Supplementary Table S2 for statistical details

Functionally, SLC38A2 silencing reduced GPX4 protein levels (P = 0.0106; Fig. 7C) and GPX4 immunofluorescence signal (Fig. 7D), accompanied by decreased cell viability (P < 0.0001; Fig. 7E) and reduced intracellular GSH levels (P = 0.0006; Fig. 7F) relative to shRNA-NC. Conversely, SLC38A2 knockdown significantly increased total iron levels (P = 0.0479; Fig. 7G), Fe²⁺ signal (FerroOrange; Fig. 7H), and the production of ROS and superoxide anion (SA) (Fig. 7I). Notably, shRNA-NC treatment did not significantly alter these parameters compared to the control group (Fig. 7A-I).

To further explore the functional necessity of SLC38A2, we utilized erastin, a classic ferroptosis inducer. HT22 cells were exposed to erastin (2–20 µM) for 24 h, which induced a dose-dependent loss of viability (P < 0.0001; Fig. 7J). While the ferroptosis inhibitor Fer-1 (5–40 µM) significantly rescued the viability of cells treated with 16 µM erastin (P < 0.01; Fig. 7K), silencing of SLC38A2 exacerbated erastin-induced cytotoxicity and significantly compromised the protective capacity of Fer-1 (P < 0.0001; Fig. 7L).

Collectively, these findings highlight the critical role of SLC38A2 in protecting hippocampal cells against ferroptosis.

SLC38A2 is necessary for melatonin’s anti-ferroptotic effect

We further tested whether SLC38A2 is required for melatonin-mediated protection against ferroptosis in HT22 cells. In HT22 cells, H₂O₂ treatment was used to simulate neuronal injury and induce ferroptosis. H₂O₂ reduced SLC38A2 and GPX4 protein levels (SLC38A2: P = 0.0313; GPX4: P = 0.0047), whereas melatonin markedly prevented these decreases (both P = 0.0002; Fig. 8A). However, this protective effect was significantly abolished upon the introduction of shRNA-Slc38a2 (SLC38A2, P = 0.0046; GPX4, P < 0.0001; Fig. 8A). Immunofluorescence for SLC38A2 and GPX4 showed consistent patterns (Fig. 8B, C).

Fig. 8.

SLC38A2 knockdown abolishes melatonin’s anti-ferroptotic effect in HT22 cells. A-E HT22 cells were transfected with or without shRNA-Slc38a2 (shSlc38a2) vector for 48 h, followed by treatment with melatonin (M), H₂O₂ (H), or H₂O₂+melatonin (HM) for 24 h. A Western blot analysis of SLC38A2 and GPX4 with densitometric quantification normalized to β-actin and expressed relative to Control. n = 8/group. SLC38A2: one-way ANOVA, F(4,35) = 9.066, P < 0.0001; Tukey’s HSD: Control vs. H, P = 0.0313; H vs. HM, P = 0.0002; HM vs. HM+shSlc38a2, P = 0.0046. GPX4: one-way ANOVA, F(4,35) = 16.24, P < 0.0001; Tukey’s HSD: Control vs. H, P = 0.0047; H vs. HM, P = 0.0002; HM vs. HM+shSlc38a2, P < 0.0001. B Representative immunofluorescence images of SLC38A2 (red) with DAPI (blue). Scale bar: 10 μm. n = 4/group. C Representative immunofluorescence images of GPX4 (red) with DAPI (blue). Scale bar: 10 μm. n = 4/group. D Representative fluorescence images of Fe²⁺ detected using FerroOrange (orange) with Hoechst 33,342 nuclear staining (blue). Scale bar: 10 μm. n = 4/group. E Representative fluorescence microscopy images of ROS (green) and SA staining (red). Scale bar: 100 μm. n = 6/group. F-H HT22 cells were transfected with or without shSlc38a2 and then treated with erastin or erastin plus melatonin or 24 h. F Cell viability measured by CCK-8 and expressed as relative viability. n = 10/group. One-way ANOVA, F(3,36) = 58.89, P < 0.0001; Tukey’s HSD: Control vs. Erastin, P < 0.0001; Erastin (E) vs. E + M, P < 0.0001; E + M vs. E + M +shSlc38a2, P < 0.0001. G Cellular reduced GSH content measured by a reduced GSH assay kit. n = 6/group. One-way ANOVA, F(3,20) = 45.83, P < 0.0001; Tukey’s HSD: Control vs. Erastin, P < 0.0001; Evs E + M, P < 0.0001; E + M vs. E + M +shSlc38a2, P < 0.0001. H Total iron content measured by an iron assay kit. n = 6/group. One-way ANOVA, F(3,20) = 18.37, P < 0.0001; Tukey’s HSD: Control vs. Erastin, P = 0.0001; Evs E + M, P = 0.0008; E + M vs. E + M+shSlc38a2, P = 0.0004. I Schematic model created with BioRender.com. Melatonin promotes SLC38A2 expression to maintain the GSH-GPX4 axis, thereby inhibiting Aβ-induced hippocampal ferroptosis. Solid lines indicate direct actions; dashed lines indicate indirect effects. Data are presented as mean ± SEM; *P < 0.05, **P < 0.01. See also Supplementary Table S2 for statistical details

Functionally, melatonin significantly inhibited the H₂O₂-induced decrease in cell viability and GSH levels and increase in total iron, yet these effects were substantially attenuated by SLC38A2 knockdown (Supplementary Fig. S3). Furthermore, melatonin treatment mitigated the H₂O₂-induced accumulation of Fe²⁺ signal, ROS fluorescence, and SA staining, but these anti-ferroptotic benefits were largely abrogated following SLC38A2 silencing (Fig. 8D, E).

To more accurately validate these findings, we substituted H₂O₂ with erastin. Erastin significantly reduced cell viability (P < 0.0001; Fig. 8F) and GSH levels (P < 0.0001; Fig. 8G), while increasing total iron content (P < 0.0001; Fig. 8H). While melatonin treatment significantly countered these erastin-induced changes (all P ≤ 0.0008 vs. erastin), its protective efficacy was significantly inhibited by SLC38A2 silencing (all P ≤ 0.0004 vs. erastin+melatonin; Fig. 8F-H).

Together, these data support a causal pathway in which melatonin requires SLC38A2 to sustain the GSH-GPX4 axis and restrain hippocampal ferroptosis.

Discussion

In this study, we identify a novel metabolic axis linking melatonin signaling to the suppression of neuronal ferroptosis in AD. While melatonin is known to exert neuroprotective effects, the precise upstream mechanisms governing its regulation of the ferroptotic cascade have remained elusive. By integrating cross-species multi-omics analysis—spanning human single-cell RNA sequencing, mouse spatial transcriptomics, and proteomics—we identified the glutamine transporter SLC38A2 as a critical metabolic checkpoint that is compromised in AD. We demonstrate that the downregulation of SLC38A2 in excitatory neurons creates a “ferroptosis-prone” metabolic state by limiting the GSH synthesis machinery. Crucially, we show that melatonin treatment effectively restores SLC38A2 expression, thereby reinstating the GSH-GPX4 antioxidant defense system and mitigating hippocampal ferroptosis and cognitive deficits. These findings establish SLC38A2 as an indispensable mediator of melatonin’s anti-ferroptotic efficacy.

Ferroptosis has been increasingly implicated in AD across both experimental models and patient samples [37, 38]. Its canonical biochemical features—iron dysregulation, excessive lipid peroxidation and ROS accumulation, coupled with compromised GPX4 activity and GSH depletion—are well recognized [39, 40], and reductions in key components of GSH biosynthesis (including GCLC and GCLM) have been reported in AD brains [41–44]. Consistent with these findings, we systematically confirmed the occurrence of neuronal ferroptosis in AD pathology by integrating cross-species multi-omics with biochemical validation. Single-cell and spatial transcriptomics revealed a profound “ferroptosis-prone” state, characterized by the enrichment of driver genes and the specific silencing of anti-ferroptotic defenses (Gclc, Gclm, Gpx4) within hippocampal excitatory neurons. Our Aβ₁₋₄₂ oligome-induced models confirmed that this transcriptional susceptibility translates into functional damage, evidenced by the collapse of the GSH-GPX4 antioxidant axis. Concurrently, the upregulation of pro-ferroptotic enzymes ACSL4 and ALOX15, alongside the accumulation of lipid peroxides, signals the initiation of an irreversible cascade. Collectively, our findings—from gene expression profiling to executioner protein dysregulation—align with and extend the growing body of evidence implicating ferroptosis in AD pathogenesis.

Unlike prior studies that have predominantly focused on downstream ferroptosis effectors such as GPX4 or ACSL4, our multi-modal omics integration unbiasedly identifies SLC38A2 as a central metabolic vulnerability in AD, tracing the pathology to an upstream failure in glutamine transport. By spatially mapping the loss of SLC38A2 to specific hippocampal niches and correlating it with downregulated glutamine catabolic processes, we demonstrate that ferroptosis in AD is not merely a secondary outcome of oxidative stress, but rather stems from a fundamental inability of neurons to import sufficient glutamine for GSH synthesis. Mechanistically, SLC38A2 acts as a requisite gatekeeper for redox homeostasis; its depletion compromises the glutamine-GSH axis, impairs GPX4-mediated lipid peroxide detoxification, and heightens susceptibility to ferroptotic cell death. This model is corroborated by our in vitro findings, where SLC38A2 silencing in HT22 cells depleted intracellular GSH, exacerbated erastin-induced ferroptosis, and attenuated the rescue effects of ferrostatin-1 and melatonin. Although future brain-specific knockdown models are required to fully delineate these dynamics in vivo, the established abundance of SLC38A2 in hippocampal neurons [45, 46] underscores the physiological significance of this transport deficit, suggesting that restoring SLC38A2 function may offer a novel strategy to bolster metabolic resilience against neurodegeneration.

While the neuroprotective efficacy of melatonin in AD is well-established, the precise molecular events governing its suppression of ferroptosis have remained elusive [47, 48]. Our data delineate the SLC38A2 axis as a pivotal target in this regulatory landscape. Given melatonin’s pleiotropic nature, the robust protection observed herein likely reflects a synergistic interplay between this newly defined SLC38A2-mediated anti-ferroptotic mechanism and canonical pathways, including direct Aβ neutralization, membrane stabilization, receptor-mediated signaling, and free radical scavenging [49–52].

Mechanistically, we propose that melatonin may sustains SLC38A2 expression through a dual-regulatory model encompassing both transcriptional activation and post-transcriptional stabilization. First, through MT1/MT2-dependent pro-survival signaling (e.g., PI3K-Akt/CREB), melatonin may promote transcriptional programs that support SLC38A2 expression [53, 54]. Second, because SLC38A2 is an adaptive stress-response gene, its transcription can be induced by the amino-acid response pathway, where ATF4 is a central driver [55–57]. ATF4 availability is governed by the PERK-eIF2α-ATF4, and melatonin has been reported to modulate this axis and increased ATF4 mRNA [58–60]. Moreover, acting post-transcriptionally, we previously demonstrated that melatonin induces lncRNA NEAT1, which is requisite for stabilizing Slc38a2 mRNA/protein in HT22 cells [59]. Thus, melatonin may maintain SLC38A2 by enhancing ATF4-linked transcription and reinforcing NEAT1-dependent mRNA stability. Future studies combining MT1/MT2 blockade with ATF4 ChIP assays will delineate the precise contributions.

Choice of disease model influences which biological questions can be addressed most cleanly. Transgenic AD mice (e.g., 5xFAD or APP/PS1) are invaluable for capturing chronic and progressive amyloidosis that resembles aspects of clinical AD [61, 62]. In contrast, the Aβ₁₋₄₂ ICV injection model was selected here to interrogate acute Aβ-induced neurotoxicity, oxidative stress, and ferroptosis within a defined temporal window [32]. This model provides precise control over the onset of pathology and the conformation of Aβ oligomers, ensuring synchronized disease severity across experimental groups. Such controllability makes it a highly efficient and well-established paradigm for screening the underlying molecular mechanisms of neuroprotective agents, such as melatonin, particularly in the context of early-stage intervention [32, 33]. Indeed, in the current study, ICV Aβ₁₋₄₂ oligomers administration produced cognitive impairment, neuronal loss, and molecular shifts consistent with ferroptotic stress, in agreement with prior studies using Aβ₁₋₄₂ oligomer-induced AD models [33, 63].

At the same time, the acute nature of ICV Aβ₁₋₄₂ oligomers delivery imposes clear boundaries on interpretation. It models a rapid Aβ burden rather than the gradual, age-dependent accumulation and systems-level remodeling seen in human AD, and some aspects of chronic pathology (including long-term network adaptation and complex microenvironmental interactions) are not fully recapitulated [62]. Complementing the in vivo work, we employed HT22-based oxidative injury paradigms in which H₂O₂ reliably induces ROS-driven neuronal damage and inflammatory/apoptotic features that overlap with AD-relevant injury patterns [64], offering experimental repeatability and interpretability [65–67]. The consistency of ferroptotic signatures and melatonin’s protective profile across these complementary models strengthens the central inference.

Despite these insights, our study has limitations. First, mechanistic necessity for SLC38A2 has been demonstrated most directly in neuronal cells, whereas in vivo confirmation using neuron-specific SLC38A2 perturbation (e.g., AAV-mediated knockdown/overexpression or conditional approaches) remains to be completed. Second, while the MT1/MT2-ATF4/NEAT1 framework provides a coherent explanatory model, direct pathway dissection (receptor antagonism, transcription-factor binding validation, and NEAT1 loss-of-function) is needed to establish causal ordering in vivo. Third, extending evaluation to transgenic models will be important to determine whether targeting this axis modifies long-term neurodegeneration beyond the acute Aβ-driven window captured here. Finally, sample size represents a practical constraint. Although the current design used n = 5/group and post-hoc power analyses suggested adequate statistical power for primary cognitive endpoints (1 − β > 0.80), and key findings remained robust under conservative statistical criteria, replication in larger cohorts will further strengthen generalizability.

In conclusion, our study supports a model in which melatonin preserves SLC38A2-dependent glutamine supply to sustain the GSH-GPX4 defense, thereby restraining lipid peroxidation and ferroptotic neuronal injury in AD (Fig. 8I). This mechanistic framing highlights SLC38A2 as a tractable metabolic target for anti-ferroptotic intervention and provides a focused rationale for future validation in chronic AD models and translational settings.

Abbreviations

AD

Alzheimer’s Disease

SLC38A2

glutamine transporter solute carrier family 38 member 2

Aβ

Beta-amyloid

GSH

Glutathione

MDA

Malondialdehyde

ROS

Reactive Oxygen Species

OFT

Open field test

MWM

Morris water maze

ICV

Intracerebroventricular

IP

Intraperitoneal

GPX4

Glutathione peroxidase 4

ACSL4

Acyl-CoA synthetase long chain family member 4

ALOX15

Arachidonate 15-lipoxygenase

SA

Superoxide anion

MADs

Median absolute deviations

UMAP

Uniform manifold approximation and projection

GCLC

Glutamate-cysteine ligase catalytic subunit

GCLM

Glutamate-cysteine ligase modifier subunit

OHSCs

Organotypic hippocampal slice cultures

Authors’ contributions

L.L, Y.J, and C.Z contributed equally as joint first authors. Q.G, L.L, Y.Z, J.L, and H.Z designed experiments, supervised studies, and wrote the manuscript. Y.J, C.Z, Q.G, H.H, J.L, Z.C, and M.L performed the experiments. L.L, J.L, and H.Z edited and revised the manuscript. All authors approved the final version of the manuscript.

Funding

This work was supported by grants from the Scientific Research Project of the Provincial Scientific Research Institute of Heilongjiang Province (H.Z, CZKYF2023-1-A011), the National Natural Science Foundation of China (L.L, 82370279), the Natural Science Foundation of Heilongjiang Province (Q.G, LH2024H019), and the Scientific Research Project of Basic Scientific Research Business of Heilongjiang Provincial Colleges and Universities in 2023 (Q.G, 2023-KYYWF-0251).

Data availability

Data are available from the corresponding author on reasonable request.

Declarations

Ethics approval and consent to participate

The animal investigations were carried out in compliance with the Guidelines for the Care and Use of Laboratory Animals established by the US National Institutes of Health (NIH Publication No. 85 − 23, revised 1996). All animal experimentation protocols were pre-approved by the Experimental Animal Ethics Committee of Harbin Medical University, China (Animal Experimental Ethical Inspection protocol number HMUIRB2023018).

Consent for publication

All authors reviewed the results and contributed to the final manuscript. All authors approved this manuscript for publication.

Competing interests

The authors declare no competing interests.