The miRNA expression profiles of patients with AD and cognitively normal controls were retrieved from the Gene Expression Omnibus (GEO) database (datasets: GSE16759 (parietal lobe tissues) (22) and GSE48552 (prefrontal cortex tissues) (23) https://www.ncbi.nlm.nih.gov/geo/). In addition, blood and cerebrospinal fluid (CSF) expression data were obtained from the GSE46579 (serum tissues) (24) and GSE212623 (25) datasets in the GEO and the Alzheimer's Disease Neuroimaging Initiative (ADNI) databases (https://adni.loni.usc.edu/). Differentially expressed miRNAs were identified using the 'limma' package in R software (version 4.2.1) (https://www.r-project.org/), applying thresholds of P<0.05 and |log2-fold change (log2FC)|>1. The resulting candidate miRNAs were retained for subsequent analyses.

The microglial cell line BV2 was purchased from American Type Culture Collection. BV2 microglia were cultured in Dulbecco's modified Eagle's medium (DMEM; high glucose; Gibco; Thermo Fisher Scientific, Inc.) supplemented with 10% fetal bovine serum (FBS; Gibco; Thermo Fisher Scientific, Inc.) and 1% penicillin-streptomycin solution (100 U/ml penicillin and 100 μg/ml streptomycin). The cells were maintained at 37°C in a humidified atmosphere containing 5% CO₂.

To modulate gene and miRNA expression, multiple transfection strategies have been employed. Synthetic miR-223-3p mimics and inhibitors were purchased from Shanghai GeneChem Co., Ltd. To upregulate miR-223-3p expression, 2×10⁵ cells were transfected with synthetic miR-223-3p mimics (final concentration: 50 nM), while a scrambled sequence served as the negative control (NC, 50 nM). Conversely, miR-223-3p inhibitors were applied to suppress endogenous expression (final concentration: 100 nM). To overexpress SIRT1, the Notch1 intracellular domain (NICD1) and FBXW7, full-length cDNAs were PCR-amplified and subcloned and inserted into the pcDNA3.1 backbone to construct the corresponding overexpression plasmids [oe-Ubiquitination (Ub), oe-SIRT1, oe-NICD1 and oe-FBXW7)], with the empty pcDNA3.1 vector used as the NC (oe-NC). All plasmids were purchased from GEBERAL BIOL company. BV2 or 293T cells were transfected with 2 μg plasmid DNA per well in 6-well plates, with the empty pcDNA3.1 vector used as the negative control (oe-NC). For RNA interference, small interfering RNA (siRNA) transfection targeting mouse Hes1 was purchased from GEBERAL BIOL company and transfected at a final concentration of 50 nM using Lipofectamine siRNA Transfection Reagent, and transfection efficiency was assessed by measuring protein levels 72 h post-transfection. All plasmid transfections were carried out using X-tremeGENE HP DNA Transfection Reagent (Roche Diagnostics) according to the manufacturer's protocol.

For metabolomic analysis of metabolites using LC-MS/MS, data were acquired in both positive and negative ion modes to ensure comprehensive coverage of the metabolic network. For metabolomic profiling, chromatographic separation was performed on a Shimadzu Nexera X2 LC-30AD ultrahigh-performance liquid chromatography (UHPLC) system equipped with an XBridge BEH C18 column. The column was maintained at 40°C, with a mobile phase consisting of (A) water containing 5% acetonitrile and 10 mM ammonium acetate (pH 9.0) and (B) water containing 95% acetonitrile and 10 mM ammonium acetate (pH 9.0). The gradient program was set as follows: 0-2 min, 95% B; 2-9 min, linear decrease to 70% B; 9-10 min, further decrease to 30% B; 10-11 min, hold at 30% B; 11-11.5 min, return to 95% B; and 11.5-15 min, re-equilibration. The flow rate was 300 μl/min, the injection volume was 5 μl, and the autosampler was kept at 4°C. Mass spectrometric detection was performed on a triple quadrupole linear ion trap system operated in multiple reaction monitoring (MRM) mode, with optimized parent-product ion transitions for targeted metabolites. The collision energy and declustering potential were adjusted on the basis of preliminary optimization. Data acquisition and processing were carried out using Analyst TF 1.7.1 software (Shanghai AB SCIEX Analytical Instrument Trading Co.), with chromatographic peaks integrated via the Macleod algorithm and a minimum signal-to-noise ratio of 3:1 as the cutoff for detection. Intracellular lactate concentration was measured with a Lactate Content Assay kit (cat. no. AKAC001-1M-50S; Beijing Boxbio Science & Technology, Co., Ltd.), following the manufacturers' protocols.

The ECAR was assessed using the Seahorse XF Glycolysis Stress Test (Agilent Technologies, Inc.). Briefly, cells (4×10⁴ per well) were seeded into XF24 microplates and cultured overnight. On the day of measurement, the cells were equilibrated in Seahorse assay medium, and the ECAR was monitored at baseline, followed by sequential injections of 10 mM glucose, 1 μM oligomycin (cat. no. HY-N6782; MedChemExpress) and 50 mM 2-deoxy-D-glucose (2-DG; cat. no. HY-13966; MedChemExpress). Basal glycolysis and glycolytic capacity were determined according to the manufacturer's protocol.

For the OCR assessment, the culture medium was replaced 1 h prior to the assay with mitochondrial stress test buffer consisting of Seahorse XF Base Medium supplemented with 10 mM glucose, 2 mM glutamine and 1 mM pyruvate. The microplates were subsequently incubated at 37°C in a CO₂-free chamber. In accordance with the manufacturer's instructions, an XF sensor cartridge sequentially delivered respiratory modulators to reach final concentrations of oligomycin (1.5 μM), FCCP (0.5 μM) and rotenone/antimycin A (0.5 μM).

Total RNA was isolated using SteadyPure Universal RNA Extraction Kit II (cat. no. AG21022; Hunan Accurate Bio-Medical Co., Ltd.), and complementary DNA was synthesized with Evo M-MLV RT Mix Kit with gDNA Clean for qPCR Ver. 2 (cat. no. AG1728; Hunan Accurate Bio-Medical Co., Ltd.) following the manufacturer's protocol. RT-qPCR was subsequently conducted with SYBR Green Premix Pro Taq HS qPCR Kit (cat. no. AG11701; Hunan Accurate Bio-Medical Co., Ltd.) on Light Cycler@480 System (Roche Diagnostics). The thermocycling conditions were as follows: Initial denaturation at 95°C for 30 sec; followed by 40 cycles of denaturation at 95°C for 5 sec and annealing/extension at 60°C for 30 sec. β-actin was used as an internal control to standardize relative mRNA levels. miR-223-3p expression was quantified using stem-loop reverse transcription followed by qPCR. Briefly, a miRNA-specific stem-loop RT primer was used to selectively reverse transcribe mature miRNA, and RT-qPCR was performed using a miRNA-specific forward primer and a universal reverse primer. The U6 gene was used as an internal control. All PCR primers were designed and synthesized by GENERAL BIOL Biotech and are listed in Table SI.

For cellular lysate, cells were washed with 1X PBS and then lysed in RIPA buffer (cat. no. G2002; Wuhan Servicebio Technology Co., Ltd.) supplemented with Halt protease inhibitor (cat. no. G2008; Wuhan Servicebio Technology Co., Ltd.) for 20 min at 4°C. The lysate was subsequently centrifuged at 14,000 × g for 5 min at 4°C to pellet insoluble material. Protein concentrations in the extracts were determined using the bicinchoninic acid (BCA) assay (cat. no. P0012; Beyotime Institute of Biotechnology) against a bovine serum albumin standard curve. Protein samples (10 μl per lane) were loaded into Bolt 10-12% Tris-Glycine Plus gels and subsequently transferred onto polyvinylidene difluoride membranes. After blocking with an appropriate reagent, the membranes were incubated with the indicated primary antibodies overnight at 4°C, followed by incubation with horse-radish peroxidase-conjugated secondary antibodies for 1 h at room temperature. The membranes were rinsed three times with TBST (0.1% Tween, 10 min each), and protein signals were detected using an enhanced chemiluminescence (ECL) system (cat. no. KF8001; Affinity Biosciences). Detailed information on the primary and secondary antibodies employed is summarized in Table SII. The band intensities were quantified using ImageJ software (version 1.54g; National Institutes of Health) and normalized to the β-actin.

293T and BV2 cells were lysed in 500 μl of IP lysis buffer (cat. no. P70100; New Cell & Molecular Biotech) containing phenyl-methyl sulfonyl fluoride. The lysates were centrifuged at 12,000 × g for 10 min at 4°C, and the clarified supernatants were incubated overnight at 4°C with the indicated antibody and Protein A/G agarose beads (20 μl; cat. no. sc-2003; Santa Cruz Biotechnology, Inc.) under gentle rotation. After six washes with the designated buffer, the bound proteins were eluted and subjected to western blot analysis as aforementioned.

The functional characterization of differentially expressed genes (DEGs) was conducted through Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis. KEGG enrichment was implemented using the R package clusterProfiler to identify biologically relevant signaling pathways (26). In addition, gene set enrichment analysis (GSEA) was performed with hallmark gene sets retrieved from the Molecular Signatures Database (MSigDB; http://www.gsea-msigdb.org/gsea/downloads.jsp) (27,28).

ROC curve analysis was performed to evaluate the diagnostic performance of miR-223-3p in patients with AD. ROC curves were constructed by plotting sensitivity (true positive rate) against 1-specificity (false positive rate) across a range of cutoff values. The area under the ROC curve (AUC) was calculated as a quantitative measure of diagnostic accuracy. An AUC of 0.5 indicates no discriminative ability, whereas an AUC of 1.0 represents perfect diagnostic performance; therefore, values closer to 1 indicate improved diagnostic accuracy. Sensitivity, specificity, and the corresponding 95% confidence intervals (CIs) were calculated to further assess diagnostic reliability.

BV2 microglia in the logarithmic growth phase were plated in six-well plates at a density of 3×10⁵ cells per well. For detection, cells were incubated with 5 μM C11-BODIPY (Thermo Fisher Scientific, Inc.) at 37°C for 30 min, followed by washing and flow cytometric analysis using a Beckman Coulter CytoFLEX S (Beckman Coulter, Inc.) according to the manufacturer's instructions. ROS levels were quantified in the FITC channel.

To inhibit glycolytic activity, BV2 microglia were incubated for 24 h with oxamate (10 mM; cat. no. HY-W013032A; MedChemExpress) and 2-DG (10 mM; cat. no. HY-13966; MedChemExpress). The SIRT1 inhibitor EX527 (10 μM; 49843-98-3; TargetMol) and the SIRT1 activator SRT1720 (5 μM; 1001645-58-4; TargetMol) were added separately and incubated for 24 h. The γ-secretase inhibitor DAPT (20 nM; cat. no. HY-13027; MedChemExpress) was administered in parallel experiments, while DMSO served as the vehicle control.

To determine the half-maximal inhibitory concentration (IC₅₀) values, BV2 cells were seeded at 10,000 cells per well into 96-well plates and treated with increasing concentrations of oxamate or 2-DG for 24 h. The medium was then replaced, 10 μl of CCK-8 solution (Beijing LABLEAD, Inc.) was added to each well, and the absorbance at 450 nm was measured using a microplate reader (Multiskan FC; Thermo Fisher Scientific, Inc.). Cell viability was assessed according to the manufacturer's instructions, and the IC₅₀ values were calculated based on dose-response curves.

BV2 cells were plated in 12-well culture plates at a density of 3×10⁵ cells per well. After attachment, they were exposed to 2 μM CHX (66-81-9; TargetMol) for the designated time periods, followed by cell lysis and protein extraction for western blotting.

293T cells were transfected with Ub, Myc-FBXW7 and His-NICD1 expression plasmids for 48 h, followed by exposure to MG132 (10 μM; cat. no. 133407-82-6, TargetMol) for 4 h. After treatment, the cells were rinsed with PBS and lysed, and immunoprecipitation was performed using 20 μl of proteinA/G beads (cat. no. sc-2003; Santa Cruz Biotechnology, Inc.). The precipitated proteins were then subjected to western blotting, and Notch1 ubiquitination was detected with an anti-Ub antibody.

Complementary oligonucleotide pairs containing either the wild-type (WT) or the mutant (MUT) sequence of the FBXW7 target region were designed and synthesized by Shanghai GenePharma Co., Ltd. The annealed fragments were subsequently cloned and inserted into the pmirGLO Dual-Luciferase miRNA Target Expression Vector (Promega Corporation). For the reporter assay, cells were seeded in 24-well plates and co-transfected with 100 ng of the FBXW7 reporter plasmid (WT or MUT) and 50 nM miR-223-3p mimics or negative control using Lipofectamine 3000 (Thermo Fisher Scientific, Inc.). A total of 48 h after transfection, luciferase activity was determined using a dual-luciferase assay system (cat. no. E1910; Promega Corporation) and normalized to Renilla luciferase activity.

Data distribution was first examined with normality tests. Comparisons between two groups with normally distributed variables were performed with unpaired t-tests, whereas one-way ANOVA was applied for analyses involving three groups, followed by Bonferroni's post hoc test to adjust for multiple comparisons. When the data deviated from normality, Wilcoxon rank-sum and Kruskal-Wallis tests were used for two- and three-group comparisons, respectively. The results are expressed as the mean ± standard deviation (SD), and all analyses were two-tailed, with statistically significant difference set at P<0.05. All the statistical analyses and data visualization were performed using GraphPad Prism (version 8.0; Dotmatics) and R software (version 4.2.1).

miR-223-3p expression was significantly altered in patients with AD across multiple brain regions and biofluids. In the GEO datasets, miR-223-3p levels were significantly increased in the parietal lobe tissues of patients with AD compared with controls (GSE16759, P=0.032; Fig. 1A). A similar upregulation was observed in the prefrontal cortex (GSE48552, P=0.002; Fig. 1B). Consistently, peripheral samples also showed elevated miR-223-3p levels in AD. In the GEO dataset GSE46579, both serum and CSF from patients with AD exhibited significantly higher miR-223-3p levels than those from controls (both P<0.001; Fig. 1C and D). These findings were further validated using the ADNI cohort. Compared with controls, patients with AD demonstrated significantly increased miR-223-3p levels in both serum and CSF samples (P<0.001 for both comparisons; Fig. 1E and F), confirming the consistent dysregulation of miR-223-3p across independent datasets and sample types.

To evaluate the diagnostic performance of miR-223-3p for AD, ROC curve analyses were performed. In brain tissues from the GEO dataset, miR-223-3p showed excellent discriminative ability, with an AUC of 0.938 (95% CI: 0.764-1.000) in the parietal lobe (Fig. 1G) and 0.972 (95% CI: 0.895-1.000) in the prefrontal cortex (Fig. 1H). In peripheral samples, the AUC was 0.820 (95% CI: 0.705-0.935) for serum (Fig. 1I) and 0.943 (95% CI: 0.879-1.000) for CSF in the GEO dataset (Fig. 1J). Similar diagnostic performance was observed in the ADNI cohort, with AUC values of 0.860 (95% CI: 0.797-0.924) for serum (Fig. 1K) and 0.825 (95% CI: 0.758-0.892) for CSF (Fig. 1L). Taken together, these results indicate that miR-223-3p is consistently upregulated in both brain tissues and biofluids of AD patients and exhibits promising diagnostic value for distinguishing AD from controls.

To explore the role of miR-223-3p in microglial functions, in vitro microglial polarization experiments were designed. BV2 microglia were used to establish three experimental conditions: NC, miR-223-3p mimic and miR-223-3p inhibitor. RT-qPCR verified the efficient modulation of miR-223-3p expression among the groups (Fig. 2A). Functionally, miR-223-3p overexpression increased the transcription of M1-associated markers [iNOS (inducible nitric oxide synthase), IL-1β and TNF-α] but suppressed M2-associated markers (Arg-1, YM-1 and FIZZ1) (Fig. 2B). Western blotting further demonstrated elevated iNOS protein levels and reduced Arg-1 levels in miR-223-3p-overexpressing cells (Fig. 2C). In addition, flow cytometry revealed a pronounced increase in ROS levels following miR-223-3p overexpression (Fig. 2D). Collectively, these findings suggest that miR-223-3p drives microglial polarization toward the proinflammatory M1 phenotype.

To elucidate the molecular basis of this phenotype, two groups of microglia were subjected to transcriptome sequencing (Fig. 2E). KEGG pathway enrichment revealed that glycolysis/gluconeogenesis and HIF-1 signaling were significantly altered (Fig. 2F). GSEA further confirmed the enrichment of glycolysis in the miR-223-3p mimic group (Fig. 2G). These results illustrated that the M1 proinflammatory phenotype may be associated with increased aerobic glycolysis.

To assess whether miR-223-3p reshapes microglial glucose metabolism, LC-MS/MS-based metabolomics were performed on BV2 cells transduced with NC, a miR-223-3p mimic, or a miR-223-3p inhibitor. A heatmap revealed robust, group-dependent shifts in metabolite profiles (Fig. 3A). Targeted quantification revealed that miR-223-3p overexpression increased the levels of lactate (Fig. 3B) and pyruvate (Fig. 3C), reduced the level of NADP⁺ (Fig. 3D), and increased the level of NADPH (Fig. 3E). Seahorse assays further demonstrated that miR-223-3p increased the ECAR, increasing both basal and compensatory glycolysis (Fig. 3F and G), and increased the OCR, increasing basal, maximal and spare respiratory capacity (Fig. 3H and I). Consistent with the metabolomics data, lactate content was greater in the miR-223-3p mimic group (Fig. 3J). Western blotting confirmed a concomitant increase in global protein lactylation (Fig. 3K).

To evaluate the potential cytotoxic effects of glycolytic inhibition, BV2 cells were treated with increasing concentrations of oxamate or 2-DG, and cell viability was assessed using a CCK-8 assay. The calculated IC₅₀ values were 31.70 mM for oxamate and 26.56 mM for 2-DG (Fig. S1A and B). Based on these results, 10 mM oxamate and 10 mM 2-DG, which are well below their respective IC₅₀ values, were selected for subsequent experiments to suppress lactylation (Fig. 3L). Inhibiting protein lactylation attenuated the miR-223-3p-induced transcription of M1 markers (iNOS, IL-1β and TNF-α) while restoring the expression of M2 markers (Arg-1, YM-1 and FIZZ1) (Fig. 3M). Lactylation inhibition also reduced intracellular ROS levels (Fig. 3N and O). These results indicate that miR-223-3p promotes microglial M1 polarization through glycolysis-associated lysine lactylation.

SIRT1 is a known delactylase that regulates protein lactylation. RT-qPCR and western blotting revealed that miR-223-3p suppressed both the transcription and translation of SIRT1 (Fig. 4A and B). To explore the potential mechanism underlying miR-223-3p-mediated regulation of lactylation, it was examined whether SIRT1 influences protein lactylation levels. The western blot results showed that treatment with the SIRT1 activator SRT1720 significantly decreased the overall Pan-Kla signal compared with the control group (Fig. 4C). By contrast, inhibition of SIRT1 with EX-527 resulted in a clear increase in Pan-Kla levels (Fig. 4D). To confirm causality, SIRT1 was reintroduced into wild-type and miR-223-3p mimic-transfected BV2 cells (Fig. S2A; Fig. 4E and F). Western blotting revealed that SIRT1 overexpression attenuated the increase in lactylation induced by miR-223-3p (Fig. 4G). Consistently, both RT-qPCR and western blotting demonstrated that restoring SIRT1 expression reduced the expression of M1 polarization markers (iNOS, IL-1β and TNF-α) (Fig. 4H and I). Furthermore, while miR-223-3p elevated ROS levels in microglia, SIRT1 overexpression diminished ROS accumulation (Fig. 4J and K).

The Notch1/Hes1 signaling axis has been reported to repress SIRT1 transcription by directly binding to its promoter (29). However, whether miR-223-3p utilizes this pathway to regulate SIRT1 remains unclear. The present RT-qPCR analysis revealed that miR-223-3p significantly upregulated Hes1 mRNA levels but had no effect on Notch1 transcription (Fig. 5A). However, western blotting revealed that miR-223-3p increased the protein expression of both Hes1 and NICD1 (Fig. 5B), suggesting that miR-223-3p may promote NICD1 protein stability or translation through indirect mechanisms. To further investigate the role of Hes1 in regulating microglial lactylation and inflammatory activation, siRNA-mediated knockdown of Hes1 was performed in BV2 cells. Western blot analysis confirmed that Hes1 silencing significantly increased SIRT1 expression compared with the control group (Fig. 5C). Consistently, the global level of Pan-Kla was significantly reduced following Hes1 knockdown (Fig. 5D). In addition, the expression of M1-associated inflammatory markers was significantly decreased (Fig. 5E). To test functional involvement, the Notch1 inhibitor DAPT was applied. It was found that Notch1 inhibition also restored SIRT1 expression at both the transcript and protein levels (Fig. 5F and G). The inhibition of Notch1 attenuated the ability of miR-223-3p to increase lactylation (Fig. 5H) and drive microglial M1 polarization (Fig. 5I and J). Flow cytometry further confirmed that blocking Notch1 reduced the increase in ROS levels induced by miR-223-3p (Fig. 5K). These findings indicate that miR-223-3p activates the Notch1/Hes1 pathway to suppress SIRT1 transcription, thereby promoting lactylation and skewing microglia toward the proinflammatory M1 phenotype.

To elucidate how miR-223-3p regulates the Notch1/Hes1 signaling pathway, target prediction databases were first used and FBXW7 was identified as a potential downstream target of miR-223-3p (Fig. 6A). RT-qPCR and western blot analyses confirmed that miR-223-3p suppressed FBXW7 expression in microglia (Fig. 6B and C). The predicted binding site of miR-223-3p within the FBXW7 3' untranslated region is shown in Fig. 6D, and a dual-luciferase reporter assay confirmed the direct binding of miR-223-3p to FBXW7 (Fig. 6E).

Next, it was explored whether miR-223-3p affects NICD1 degradation pathways. When proteasomal degradation was blocked, inhibition of miR-223-3p modestly increased NICD1 levels, whereas blocking lysosomal degradation reduced NICD1 expression (Fig. 6F). These results suggested that miR-223-3p promotes proteasome-dependent degradation of NICD1 via FBXW7 while stabilizing NICD1. Indeed, inhibiting miR-223-3p increased FBXW7 expression, increased NICD1 degradation, and accelerated NICD1 turnover following CHX treatment (Fig. 6G). Interactions between FBXW7 and NICD1 were verified by co-IP assays (Fig. 6H and I). Moreover, FBXW7 overexpression promoted the ubiquitination and degradation of NICD1 in both 293T and BV2 cells (Fig. S2B-D; Fig. 6J and K). These findings demonstrate that miR-223-3p directly targets and represses FBXW7, thereby reducing NICD1 ubiquitination and proteasomal degradation, which stabilizes NICD1 and facilitates downstream Notch1/Hes1 signaling.