For study purposes, all cell lines, including NK92, SKOV3, A2780, and SKOV3/diaminodichloroplatinum (DDP), were obtained from Procell (Wuhan, China) and cultured at 37°C in a humidified atmosphere containing 5% CO₂. NK92 cells and their derivatives were maintained in X-VIVO medium (Lonza, Basel, Switzerland) supplemented with 10% human platelet lysate and 300 IU/mL of recombinant human interleukin-2. SKOV3 and SKOV3/DDP cells were cultured in McCoy’s 5A medium (Sigma-Aldrich Chemie GmbH, Steinheim, Germany) containing 10% fetal bovine serum. A2780 cells were cultured in RPMI 1640 medium (Gibco, Carlsbad, CA, USA) supplemented with 10% fetal bovine serum (Gibco).

The lentiviral vector, designed and synthesized by Genepharm (Shanghai, China), was transfected into NK92 cells following the manufacturer’s lentiviral-transduction protocol. Stable cell lines were subsequently selected using puromycin (2 μg/mL).

Cells were incubated in conditioned medium for 48 h at 37°C. The conditioned medium was centrifuged at 500 ×g for 10 min at 4°C, followed by centrifugation at 16, 800 ×g for 30 min at 4°C. The supernatant was filtered through a 0.22-μm filter (Millipore, Burlington, MA, USA) and centrifuged at 120, 000 ×g for 70 min at 4°C. Pelleted exosomes were washed and resuspended in phosphate-buffered saline (PBS). Exosomes were characterized using nanoparticle-tracking analysis (NTA) with a Nicomp Z3000 system (Particle Sizing Systems, Santa Barbara, CA, USA). Morphology was assessed by transmission electron microscopy (TEM) using a JEM-1230 system (JEOL, Tokyo, Japan). For labeling, exosomes were incubated with PKH67 dye (Sigma-Aldrich, St. Louis, MO, USA) according to the manufacturer’s instructions.

NK92 cells and culture supernatants were collected for proteomic sequencing, with three samples per group. Samples were lysed in SDT buffer (4% SDS, 100 mM of Tris-HCl, pH 7.6), sonicated, and heated at 95°C for 15 min. After centrifugation (14, 000 ×g, 15 min), supernatants were quantified using a bovine serum albumin protein assay kit (P0012; Beyotime Biotechnology, Shanghai, China) and stored at −80°C.

For sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE), 20 μg of protein per sample was mixed with 6× loading buffer, heated at 95°C for 5 min, and separated on 12% polyacrylamide gels. Proteins were visualized using Coomassie Blue R-250 staining.

For mass spectrometry, 100 μg of protein per sample was reduced with 100 mM of DTT (95°C, 5 min). Detergents and low-molecular-weight components were removed via repeated ultrafiltration (Sartorius, 30 kDa, VN01H22) using Universal Assay Buffer buffer (8 M of urea, 150 mM of Tris-HCl, pH 8.5). Reduced cysteine residues were alkylated with 100 mM of iodoacetamide (100 μL in UA buffer; 30 min, dark). Filters were washed with UA buffer (3 × 100 μL) followed by 50 mM of NH₄HCO₃ (2 × 100 μL). Proteins were digested with 4 μg of trypsin (Promega Corp., Madison, WI, USA) in 40 μL of 50-mM NH₄HCO₃ (37°C, overnight). Peptides were collected as filtrate and desalted using C18 columns. Peptide concentration was determined by ultraviolet absorbance at 280 nm (extinction coefficient: 1.1 for 0.1% g/L of solution), accounting for tryptophan/tyrosine frequency in vertebrate proteins.

Samples were analyzed on a timsTOF Pro mass spectrometer (Bruker, Billerica, MA, USA) coupled to a nanoElute system (Bruker) via a CaptiveSpray source. Peptides were separated on a 25-cm × 75-μm analytical column (1.6-μm C18 beads; IonOpticks, Victoria, Australia) at 50 °C (column oven; Sonation GmbH, Biberach, Germany). After equilibration with four column volumes of buffer A (99.9% H₂O, 0.1% formic acid; 800 bars), separation was achieved at 300 Nl/min using a linear gradient. MS parameters were as follows: PASEF mode; *m/z*, 100-1700; 1/K₀, 0.75-1.4 V·s/cm² (100-ms ramp); capillary voltage, 1, 500 V; dry gas, 3 l/min (180 °C); 10 MS/MS scans/cycle (1.16 s); charge states, 0-5; active exclusion, 0.5 min; and collision energy, 20–59 eV.

Data were processed in MaxQuant (v1.6.17.0; Max Planck Institute of Biochemistry, Munich, Germany) against a reference database. Notable parameters include: precursor mass tolerance, 6 ppm; fragment ion tolerance, 20 ppm; enzyme, trypsin/P (maximum of two missed cleavages); fixed modification, carbamidomethylation; and variable modifications, N-terminal acetylation and methionine oxidation. Peptide and protein false-discovery rates were set to 1%. Label-free quantification determined protein abundance. Differentially expressed proteins were defined as those exhibiting a fold change of >2 or <0.5 (*P* < 0.05, Student’s *t* test).

Total RNA from cells and exosomes was isolated using TRIzol reagent (Invitrogen, Carlsbad, CA, USA) according to the manufacturer’s protocol, with a sample size of three per group. Purified total RNA served as the starting material for library preparation. Briefly, 3' and 5' RNA adapters were ligated to respective ends of small RNAs. First-strand cDNA was synthesized using a reverse transcription primer. Double-stranded cDNA libraries were generated by polymerase chain reaction (PCR) amplification; then, purification and size selection were completed to retain inserts of 18–40 bp.

Library quality was assessed by quantifying insert fragments and determining effective library concentrations. Qualified libraries were pooled in equimolar ratios based on effective concentration and sequenced on an Illumina (San Diego, CA, USA) platform to meet target data requirements.

Raw sequencing data were generated by base-calling Illumina fluorescence images, yielding short reads in FASTQ format (Cock et al., 2010), which encode nucleotide sequences and corresponding quality scores. Raw read quality was evaluated using Fastp (v0.23.1; Chen et al., 2018).

Statistically analyze the expression levels of known and novel miRNAs in each sample, and normalize the expression levels using TPM (Zhou et al., 2010). TPM= (read count * 1, 000, 000)/libsize (libsize: total miRNA read count). For samples with biological replicates, use DESeq2 (Love MI et al., 2014) for differential expression analysis between two comparison groups. DESeq2 provides statistical routines to determine differential expression in digital gene expression data using a model based on the negative binomial distribution. For samples without biological replicates, use the edgeR (Robinson MD et al., 2010) TMM algorithm to normalize read count data for analysis.

Total RNA was isolated from cells and exosomes using TRIzol reagent (Invitrogen) according to the manufacturer’s protocol, with a sample size of three per group. RNA concentration and purity were quantified using a NanoDrop ND-1000 spectrophotometer (NanoDrop Technologies, Wilmington, DE, USA), and integrity was assessed with an Agilent 2100 bioanalyzer (Agilent Technologies, Santa Clara, CA, USA). Samples with an RNA integrity number of <7.0 were excluded. Approximately 5 μg of total RNA underwent ribosomal RNA depletion using the Ribo-Zero rRNA Removal Kit (Illumina). Depleted RNA was fragmented via divalent cations at elevated temperatures. Fragmented RNA was reverse-transcribed into first-strand cDNA; second-strand cDNA synthesis was then performed using Escherichia coli DNA polymerase I, RNase H, and dUTP (yielding U-labelled second-strand DNA). Double-stranded cDNA fragments underwent end-repair and 3′ adenylation, followed by ligation to indexed adapters. Adapter-ligated fragments were size-selected using AMPure XP beads (Beckman Coulter, Brea, CA, USA). After USER enzyme (NEB, Ipswich, MA, USA) treatment, ligated products were PCR-amplified (95°C for 3 min; 8 cycles of 98°C for 15 s, 60°C for 15 s, 72°C for 30 s; 72°C for 5 min). The resulting cDNA libraries exhibited an average insert size of 300 ± 50 bp. Finally, paired-end sequencing (2 × 150 bp) was conducted on an Illumina HiSeq 4000 platform, following the manufacturer’s guidelines.

Raw data (raw reads) of fastq format were firstly processed through in-house perl scripts. In this step, clean data (clean reads) were obtained by removing reads containing adapter, reads containing ploy-N and low quality reads from raw data. At the same time, Q20, Q30 and GC content the clean data were calculated. All the downstream analyses were based on the clean data with high quality.

Differential expression analysis of two conditions/groups (three biological replicates per condition) was performed using the DESeq2 R package (1.16.1). DESeq2 provide statistical routines for determining differential expression in digital gene expression data using a model based on the negative binomial distribution. The resulting P-values were adjusted using the Benjamini and Hochberg’s approach for controlling the false discovery rate. Genes with an adjusted P-value.

Total protein from cell lines or exosomes was extracted using radioimmunoprecipitation assay buffer supplemented with PMSF (Solarbio, Beijing, China) and quantified with a bovine serum albumin kit (Beyotime Technologies). Equal amounts of protein were separated by SDS-PAGE and transferred to polyvinylidene fluoride membranes (Millipore, Burlington, MA, USA). After blocking with 5% bovine serum albumin in TBST, membranes were incubated overnight at 4°C with primary antibodies, followed by 1-h incubation with horseradish peroxidase (HRP)-conjugated secondary antibodies at 37°C. Protein bands were visualized using a chemiluminescence kit (Millipore) and quantified by densitometry.

Primary antibodies included granzyme (13588-1-AP, rabbit polyclonal; Proteintech, Chicago, IL, USA), perforin (14580-1-AP, rabbit polyclonal; Proteintech), CD9 (60232-1-Ig, mouse monoclonal; Proteintech), CD81 (66866-1-Ig, mouse monoclonal; Proteintech), TSG101 (28283-1-AP, rabbit polyclonal; Proteintech), calnexin (10427-2-AP, rabbit polyclonal; Proteintech), and Lamp2b (ab18529, rabbit polyclonal; Abcam, Cambridge, UK).

Secondary antibodies included goat anti-rabbit IgG-HRP (ab6728; Abcam) and goat anti-mouse IgG-HRP (ab205719; Abcam, Cambridge, UK).

Total RNA was extracted with RNAiso Plus (TaKaRa Bio, Shiga, Japan), following the manufacturer’s protocol. RNA integrity and purity were assessed using a NanoDrop One spectrophotometer (NanoDrop Technologies). Reverse transcription was performed using the PrimeScript RT Reagent Kit with gDNA Eraser (TaKaRa Bio). Genomic DNA was eliminated by incubation at 42°C for 2 min. Tissue RNA was reverse-transcribed under the following conditions: 37°C for 15 min and 85°C for 5 s. cDNA was stored at −80°C immediately after synthesis. qRT-PCR was conducted on a QuantStudio 5 real-time PCR system (Applied Biosystems, Foster City, CA, USA) using Hieff qPCR SYBR Green Master Mix (Yeasen Biotechnology, Shanghai, China). Reaction conditions included the following: initial denaturation at 95°C for 5 min; 40 cycles of 95°C for 10 s, and annealing/extension at 60°C for 30 s (with primer-specific annealing temperatures where required). Relative RNA quantification was calculated via the 2 −ΔΔCt method, using GAPDH as the internal reference gene. The primer sequences used were as follows (F, forward; R, reverse): Lamp2b -F: GAAAATGCCACTTGCCTTTATGC; Lamp2b -R: AGGAAAAGCCAGGTCCGAAC; GAPDH-F: AACGGATTTGGTCGTATTGG; GAPDH-R: TTGATTTTGGAGGGATCTCG; ABCB1-F: TTGCTGCTTACATTCAGGTTTCA; ABCB1-R: AGCCTATCTCCTGTCGCATTA; and miR-31-5p-F: AGGCAAGATGCTGGCATAGCT.

Based on the predicted miR-31-5p binding site, we amplified the 3′UTR sequences of wild-type GPRC5A (GPRC5A-wt) and a site-directed mutant (GPRC5A-mut), then cloned these into the dual-luciferase reporter vector psiCHECK2 (GeneChem, Shanghai, China). HEK293T cells, selected for their high transfection efficiency, were used for all assays. Cells were co-transfected with either the GPRC5A-wt or GPRC5A-mut reporter plasmid, along with miR-31-5p mimic or negative control mimic, using Lipofectamine 3000 (Thermo Fisher Scientific, Waltham, MA, USA). At 48 h post-transfection, cells were harvested, and luciferase activity was quantified using a dual-luciferase reporter assay kit (Beyotime Biotechnology). Firefly luciferase activity served as the internal control for normalizing Renilla luciferase activity.

Following a 48-h incubation period, cell culture supernatant was collected and exosomes were isolated by ultracentrifugation. The exosome pellet was washed in PBS via ultracentrifugation and resuspended. All centrifugation steps were performed at 4°C. A 30-μg aliquot of exosomes was incubated with 100 μmol of Cy3-labeled cholesterol-modified ABCB1 siRNA in 100 μL of solution for 3 h at 37°C. Unbound siRNA was subsequently removed by washing the exosomes with PBS via ultracentrifugation. The pellet was resuspended in PBS, and fluorescence was measured to quantify siRNA loading. Loading efficiency (%) was determined by comparing sample fluorescence intensity (measured at Cy3’s excitation/emission maxima: 548 nm/570 nm using a Victor3–1420 Multilabel Plate Reader [PerkinElmer, Waltham, MA, USA]) to a standard curve of free Cy3-labeled ABCB1 siRNA.

Cell apoptosis was detected using an annexin V-FITC/propidium iodide apoptosis detection kit (Beyotime). The 1 × 10⁵ cells were harvested and double-stained with FITC and propidium iodide after transfection, followed by analysis on a flow cytometer (FACSCalibur; Becton Dickinson, Frankin Lakes, NJ, USA). The acquired data were analyzed and fitted using the FlowJo™ software (Becton Dickinson).

For the CCK-8 assay, 1500 treated cells were seeded in 96-well plates. After the cells were cultured for 0, 12, 24, 48, 72 and 96 h, the cells were added to a CCK-8 (Dojindo Laboratories, Kumamoto, Japan) and incubated at 37 °C for 2 h. The OD450 value was measured using a microplate detector.

For cell-migration assays, 1.5 × 10⁵ treated cells were seeded in Transwell chambers with an 8-μm well membrane (Corning, Inc., Corning, NY, USA). After 4 h of culture, the lower-layer cells were fixed, while the upper-layer cells were removed by wiping. Migrating cells were stained with Giemsa (Solarbio). Images of the cells were obtained using light microscopy in four fields.

All mouse procedures were approved by the Animal Care and Use Committee of the National Research Institute for Family Planning, National Institutes of Health Commissioners. Four-week-old BALB/c nude mice were purchased from Vital River Laboratories (Beijing, China). Following 1-week acclimatization, mice received intraperitoneal or subcutaneous injections of stabilized SKOV3 or SKOV3/DDP cells (2 × 10₆ cells in 100 μL of PBS), as required by the experimental design. Animals were randomly assigned to experimental groups using a computer-generated random number sequence after baseline assessments. Group allocation was concealed until interventions commenced. Throughout the study, investigators performing surgeries, administering treatments, and assessing outcomes were blinded to group allocation. Coded labels were used for all samples/data, with the key retained by an independent researcher until final analysis. Body weight and activity levels were monitored regularly. Experiments were terminated if mice exhibited severe lethargy, >20% weight loss, or overt cachexia. The mice were anesthetized by intraperitoneal injection with tribromoethanol at a dose of 240 mg/kg. For intraperitoneal tumor models, bioluminescence was quantified using the IVIS Lumina system (Caliper Life Sciences; Hopkinton, MA, USA) after administration of D-luciferin (150 mg/kg; Yeasen Bio, Shanghai, China). At experimental endpoints, mice were euthanized by CO₂ inhalation (30% flow rate of chamber volume per minute for CO2). Major organs and tumors were harvested, fixed in paraffin, and processed for hematoxylin and eosin (HE) staining or immunohistochemistry (IHC).

For IHC, paraffin-embedded sections were heated at 60 °C for 2 h, deparaffinized, and subjected to antigen retrieval. Sections were incubated with the following primary antibodies (Proteintech) at 37 °C for 60 min: anti-HE4 (1:500), anti-Ki67 (1:200), anti-vimentin (1:8000), and anti-ABCB1 (1:500). Subsequently, sections were incubated with secondary antibodies at 37 °C for 15 min, followed by HRP-conjugated streptavidin for 10 min. Chromogenic development was performed using 3, 3′-diaminobenzidine, and cell nuclei were counterstained with hematoxylin. Stained sections were visualized and imaged using an Olympus microscope (Olympus Corp., Tokyo, Japan).

Data were analyzed using GraphPad Prism 7.0 (GraphPad Software, San Diego, CA, USA) and expressed as mean ± standard deviation values. Differences between two independent groups were compared using unpaired two-sided Student’s t tests. Statistical significance was defined as P < 0.05. Figures were assembled using Adobe Illustrator 2020 (Adobe Inc., San Jose, CA, USA).

To investigate the role of NK92 cell-derived exosomes in immunotherapy, we isolated and characterized exosomes from NK92 cell supernatants using ultracentrifugation (Figure 1A). TEM revealed that the exosomes exhibited a spherical morphology (Figure 1B). NTA indicated an average exosome size of 139 nm (Figure 1C). Western blot analysis confirmed the presence of exosomal markers (CD9, CD81, TSG101) and the absence of the negative control calnexin in isolated exosomes compared to whole NK92 cell lysates (Figure 1D). Furthermore, we detected expression of the lysosomal membrane protein Lamp2b, a potential carrier for targeting peptides in NK92-derived exosomes by western blot analysis (Figure 1D).

Current research indicates that NK92 cells mediate tumor cytotoxicity primarily through secreted proteins. To determine whether NK92 cell-derived exosomes retain this functional capacity, we performed comparative proteomic profiling of NK92 cells and their exosomes. We identified 776 differentially abundant proteins, with 231 significantly enriched in exosomes (Figure 2A). Subcellular localization analysis revealed 12.3% of differential proteins are secreted extracellular effectors (Figure 2B). Gene Ontology (GO) analysis of exosome-enriched proteins demonstrated their predominant involvement in immune system activation and regulation, particularly complement response pathways (Figure 2C). Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analysis further indicated these proteins operate primarily via the complement and coagulation cascades (Figure 2D). Notably, we confirmed granzyme and perforin expression within NK92 cell-derived exosomes (Figure 2E). Given the established regulatory roles of exosomal RNAs in tumor progression, we characterized miRNA expression profiles using small RNA-seq. Hierarchical clustering revealed distinct miRNA signatures between NK92 cells and their exosomes (Figure 2F). We detected 1286 differentially expressed miRNAs, with 885 enriched in exosomes. Upon applying a threshold (|log2FC| ≥ 1.0, P < 0.05), 247 miRNAs exhibited significant differential expression (152 upregulated, 95 downregulated; Figure 2F). Analysis of the 152 exosome-enriched miRNAs using the dbDEMC database indicated their reduced expression across multiple cancer types (Supplementary Figure S1A). This finding, consistent with known functions of non-coding RNAs, suggests that NK92-derived exosomes possess tumor-suppressive potential. Prediction and functional analysis of miRNA target genes revealed that these differentially expressed targets modulate cell proliferation and cytotoxicity through diverse pathways. GO analysis and KEGG pathway analyses substantiated the significant regulatory role of NK92 cells and their exosomes in oncogenesis (Figures 2G, H).

Exosomes can be genetically engineered to enhance their tumor-targeting specificity, thereby improving therapeutic efficacy. In this study, we generated NK92 cell-derived exosomes displaying the tLyP-1 targeting peptide (CGNKRTR) on their surface by fusing it with the exosomal membrane protein Lamp2b (Figure 3A). Sanger sequencing of genomic DNA extracted from engineered cells confirmed successful integration of the target gene (Figure 3B). Engineered NK92 cell-derived exosomes (referred to as exo^tLyP-1) were isolated via ultracentrifugation. Characterization by TEM, NTA, and western blot confirmed that their physicochemical properties (size, morphology, and presence of exosomal markers) were comparable to those of unmodified exosomes (called exo^ctrl) (Figures 3C–E). Furthermore, the engineered exosomes retained their cytotoxic cargo, including perforin and granzyme (Figure 3F).

To assess whether tLyP-1 modification enhanced targeting of ovarian cancer cells, we conducted in vitro and in vivo experiments. Ultimately, we determined that PKH67-labeled exo^tLyP-1 exhibited significantly enhanced cellular uptake by tumor cells compared to PBS or exo^ctrl (Figure 3G). Subsequently, in a subcutaneous tumor mouse model, DIR-labeled exo^tLyP-1 was systemically administered via tail vein injection. In vivo imaging performed 24 h post-injection revealed significantly greater accumulations of exo^tLyP-1 at the tumor site compared to control mice (Figure 3H). These results demonstrate that tLyP-1 modification confers superior tumor-targeting capability.

We next evaluated the functional consequences of exo^tLyP-1 uptake on ovarian cancer cells. CCK-8 and Transwell assays indicated that co-incubation with exo^tLyP-1 significantly reduced tumor cell proliferation and migration (Figures 3I–K). Apoptosis assays further demonstrated that exo^tLyP-1 significantly promoted tumor cell death compared to PBS or exo^ctrl treatments (Figures 3L, M). In summary, exo^ctrl exhibit enhanced tumor-targeting specificity and showed potent anti-tumor efficacy against ovarian cancer cells.

To investigate the potential anti-tumor efficacy of NK92 cell-derived exosomes in vivo, SKOV3 cells were transplanted into mice via intraperitoneal or subcutaneous injection. Mice were randomized into groups on postoperative day 7 and subsequently treated via tail vein injection every 7 days for a total of five administrations (Figures 4A, B).

In the intraperitoneal tumor-bearing model, tumor progression was monitored by bioluminescence imaging on postoperative day 42. PBS-treated mice exhibited rapid tumor growth and intense bioluminescence. In contrast, significantly reduced bioluminescence intensity was observed in both the exo^ctrl and exo^tLyP-1-treated groups compared to the PBS group, indicating a therapeutic response against ovarian cancer in the former (Figures 4C, D). Notably, the exo^tLyP-1 group displayed a significantly smaller bioluminescent area and weaker intensity than the exo^ctrl group, suggesting exo^tLyP-1 confers superior anti-tumor activity.

Given the propensity of ovarian cancer cells to metastasize to the liver and lungs, we assessed the effects of exo^tLyP-1 on metastatic potential through pathological analysis of these organs. HE staining revealed metastatic lesions characteristic of ovarian cancer in mouse livers and lungs (Figures 4E–G). Subsequent detection of the ovarian cancer-specific marker HE4 in these tissues corroborated the staining findings (Figure 4H; Supplementary S2A, B). Furthermore, expression of the mesenchymal marker vimentin, indicative of migratory potential, was significantly lower in the exo^tLyP-1 group compared to both the PBS and exo^ctrl groups (Figure 4I; Supplementary S2C, D). This reduction confirms that exo^tLyP-1 inhibits ovarian cancer cell migration.

In the subcutaneous tumor model, tumor volume was significantly reduced in both the exo^ctrl- and exo^tLyP-1-treated groups relative to the PBS group, demonstrating inhibition of malignant proliferation (Figure 4J). Histopathological analysis of tumor tissues via HE staining confirmed characteristic ovarian cancer malignancy (Figure 4K). Immunohistochemical analysis of the ovarian cancer marker HE4 and the proliferation marker Ki67 further demonstrated that both exo^ctrl and exo^tLyP-1 inhibit ovarian cancer cell proliferation, with the inhibitory effect of exo^tLyP-1 being more pronounced (Figures 4L, M; Supplementary S2E, F).

To investigate the anti-tumor mechanism of NK92 cell-derived exosomes in ovarian cancer, we performed small RNA-seq on SKOV3 cells following co-incubation with exosomes (Figure 5A). Compared to using the PBS control, incubation with NK92 cell-derived exosomes upregulated 52 miRNAs (Figure 5B). GO and KEGG analyses revealed that these miRNAs modulate multiple signaling pathways critical for cell survival and metastasis (Figures 5C, D). Intersection of miRNAs enriched in NK92 cell-derived exosomes with those upregulated in SKOV3 cells identified the following six candidates: hsa-miR-31-5p, hsa-miR-200a-3p, hsa-miR-29b-3p, hsa-miR-365a-3p, hsa-miR-30b-5p, and hsa-miR-486-5p (Figure 5E). Cross-referencing with the dbDEMC cancer miRNA database confirmed three of these to be suppressors of ovarian cancer, with miR-31-5p exhibiting the greatest abundance (Figure 5F).

As miRNAs typically function through mRNA degradation, we conducted transcriptomic and proteomic sequencing of exosome-treated SKOV3 cells to identify miR-31-5p targets (Figures 5G, H). Integration of TargetScan predictions with downregulated transcripts and proteins identified two candidates: GPRC5A and SPARC (Figure 5I). Notably, high GPRC5A expression correlated with poor overall survival in ovarian cancer patients, while SPARC expression showed no significant association, implicating GPRC5A in disease progression (Figure 5J, S3ASupplementary S3).

To validate targeting, we quantified miR-31-5p and GPRC5A expression in SKOV3 cells incubated with exo^ctrl or exo^tLyP-1. Exosomal incubation upregulated miR-31-5p and downregulated GPRC5A (Figures 5K, L), suggesting a regulatory relationship. Protein-level validation using miR-31-5p mimics/inhibitors and GPRC5A siRNA confirmed inverse correlation, while miR-31-5p overexpression did not alter SPARC (Figure 5M, Supplementary S3B). Dual-luciferase reporter assays in 293T cells demonstrated direct binding of miR-31-5p to the GPRC5A 3′UTR (Figure 5S).

Functional assays revealed that miR-31-5p promotes apoptosis and suppresses proliferation/metastasis in ovarian cancer cells by degrading GPRC5A (Figures 5N–R). Consistent with this, analysis of in vivo mouse models (subcutaneous/intraperitoneal) showed that exosome treatment reduced GPRC5A expression, inhibiting tumor growth and metastasis (Supplementary Figure S3C).

Rescue experiments using miR-31-5p inhibitor and GPRC5A siRNA confirmed that GPRC5A knockdown reverses the pro-tumorigenic effects of miR-31-5p suppression (Supplementary Figures S3D–G). Also, quantitative PCR indicated functional delivery of 5%-15% of exosomal miR-31-5p to recipient cells (Supplementary Figure S4A). Collectively, these data demonstrate that GPRC5A is a direct target of miR-31-5p and mediates its anti-tumor effects.

To preserve the integrity of NK92 cell-derived exosomes and facilitate ABCB1 siRNA delivery into recipient cells, we employed the cholesterol-modification strategy established by Kang et al., wherein siRNA adsorption to exosomal membranes is enabled via cholesterol conjugation. Cholesterol-modified ABCB1 siRNA was synthesized and co-incubated with exosomes. Laser confocal microscopy revealed co-localization of CY3-labeled ABCB1 cholesterol-modified siRNAs and PKH67-labeled exosomes, confirming successful siRNA loading onto exosomes (Figure 6A). Fluorescence intensity quantification indicated ~70% loading efficiency. Furthermore, ~40% of the loaded siRNA was internalized and functionally delivered to recipient cells (Supplementary Figures S4B, C).

Following 24-h co-incubation of ovarian cancer cells with exosomes (exo^ctrl or exo^tLyP-1) loaded with cholesterol-modified ABCB1 siRNA, qRT-PCR detected significant gene silencing (Figure 6B). To assess whether exo^{tLyP-1/SiABCB1} sensitizes chemoresistant ovarian cancer cells to therapeutics, we generated a SKOV3/DDP cell line (Figure 6C). Combining exo^{tLyP-1/SiABCB1} with DDP significantly inhibited SKOV3/DDP proliferation (CCK-8 assay; Figure 6D), enhanced apoptosis (flow cytometry; Figures 6E, F), and reduced migration (Transwell assay; Figures 6G, H) compared to controls.

For in vivo evaluation, mice bearing subcutaneous or peritoneal xenografts received tail vein injections of exo^{tLyP-1/SiABCB1}. ABCB1 expression was markedly reduced in tumors, livers, and lungs (Figures 6I, J; Supplementary S4D, E), demonstrating functional siRNA delivery via exosomes. Moreover, exo^{tLyP-1/SiABCB1} combined with DDP robustly suppressed tumor growth and metastasis in vivo relative to PBS or monotherapy (Figures 6K–M). These results establish that NK92 cell-derived exosome-mediated siABCB1 delivery synergizes with DDP to inhibit ovarian cancer progression, revealing a promising translational strategy.