The human AML cell lines NB4 (RRID: CVCL_0005), THP-1 (RRID: CVCL_0006), and OCI-AML3 (RRID: CVCL_1844) were sourced from the National Collection of Authenticated Cell Cultures (Shanghai, China) and cultured in RPMI-1640 medium (Gibco, USA) containing 10% heat-inactivated fetal bovine serum (FBS) (Corning, USA) and 1% penicillin-streptomycin (Gibco, USA). THP-1 cells used in this study were maintained in their undifferentiated state without phorbol 12-myristate 13-acetate (PMA) treatment and were selected as a representative AML cell model rather than a macrophage-like uptake model. HeLa cells (RRID: CVCL_0030) were obtained from Shanghai Genechem Co., Ltd. (Shanghai, China) and maintained in DMEM (Gibco, USA) supplemented with 10% heat-inactivated FBS (Corning, USA) and 1% penicillin-streptomycin (Gibco, USA). All cells were cultured at 37 °C in a 5% CO₂ incubator. The identities of all cell lines were verified by short tandem repeat analysis conducted within the past three years. NB4 cells were authenticated by VivaCell Biotechnology Co., Ltd. on July 17, 2023; THP-1 cells were authenticated by VivaCell Biotechnology Co., Ltd. on May 5, 2023; OCI-AML3 cells were authenticated by Procell Life Science & Technology Co., Ltd. on February 14, 2025; and HeLa cells were authenticated by the Cell Bank of the Chinese Academy of Sciences on April 9, 2025. Cells between passages 5 and 20 were used for all experiments. For infection, cells were treated with the following viruses provided by Shanghai Genechem Co., Ltd. (Shanghai, China): short hairpin negative control lentivirus (shNC), RCN1 knockdown lentivirus (shRCN1), TFAM knockdown lentivirus (shTFAM), RCN1 overexpression lentivirus, and TFAM overexpression lentivirus. The targeting sequences used for the lentiviral transduction are listed in Supplementary Table 1. Following infection, stably transduced cell populations were obtained through puromycin selection (1-2 μg·mL^-1; InvivoGen, USA). Throughout the study, all cells tested negative for mycoplasma contamination.

RNA was extracted using a Steady Pure Universal RNA Extraction Kit II (AG21022, Accurate Biology, China) according to the manufacturer’s instructions. A NanoDrop One (Thermo Fisher Scientific, USA) was used to determine RNA quality and concentration. Subsequently, 1 μg of total RNA was reverse-transcribed into cDNA using an Evo M-MLV Plus 1st Strand cDNA Synthesis Kit (AG11615; Accurate Biology, China). Quantitative polymerase chain reaction (qPCR) was performed using a SYBR Green Pro Taq HS Premix qPCR Kit III (AG11738; Accurate Biology, China) and target-specific primers, with glyceraldehyde-3-phosphate dehydrogenase (GAPDH) as the normalization control. The corresponding primer sequences are shown in Supplementary Table 2. All qPCR assays were performed at least three times on a 7300plus Real-Time PCR System (Applied Biosystems, USA).

Cells in the control and treated groups were equally divided into two fractions. The first fraction was exposed to 500 μL of 50 mM NaOH and heated for 30 min to ensure complete solubilization of cellular DNA. Following incubation, the alkaline solution was neutralized with 50 μL of 1 M Tris-HCl (pH 8.0). The resulting lysate was used as a normalization control for total mtDNA content, whereas the remaining fraction was resuspended in 500 µL of extraction buffer composed of 150 mM NaCl, 50 mM HEPES (pH 7.4), and 20 μg·mL^-1 digitonin, followed by gentle rotation at room temperature for 15 min to induce partial plasma membrane permeabilization. Samples were centrifuged at 17,000 × g for 10 min, and the cytosolic fraction was obtained from the supernatant. DNA was extracted from the cytosolic fraction using a QIAamp DNA Mini Kit (51304; Qiagen, Germany) following the manufacturer’s instructions. qPCR was performed on whole-cell lysates and cytosolic fractions with primers targeting nuclear DNA [nDNA; β2-microglobulin (β2M)] and mtDNA [cytochrome c oxidase subunit I (COXI)]. The mtDNA abundance in each fraction was quantified using cycle threshold (CT) values. A complete list of primer sequences is provided in Supplementary Table 2.

To prepare the protein samples, cells were incubated on ice for 30 min in radioimmunoprecipitation assay (RIPA) buffer (Thermo Fisher Scientific, USA) containing protease inhibitors (HY-K0010; MedChemExpress), while RBCEVs were lysed for 15 min. The resulting supernatants were denatured by boiling, and equal amounts of total protein (20 μg per lane) were loaded onto 4%-20% MeilunGel Precast Page Gel (MA04161-3; MeilunBio, China) alongside a PageRuler^TM Prestained Protein Ladder (26617; Thermo Fisher Scientific). Following electrophoresis, proteins were blotted onto polyvinylidene fluoride membranes using a standard wet-transfer apparatus. The membranes were blocked with 5% skim milk for 2 h at room temperature and washed twice with TBST (50 mM Tris, pH 8.0, 150 mM NaCl, and 0.1% Tween-20). The membranes were then incubated overnight at 4 °C with the following designated primary antibodies: anti-RCN1 (1:1,000, ab198996; Abcam), anti-β-actin (1:1,000, 4970; Cell Signaling Technology), anti-cGAS (1:1,000, 15102; Cell Signaling Technology), anti-STING (1:1,000, 13647; Cell Signaling Technology), anti-phospho-STING (1:200, 50907; Cell Signaling Technology), anti-TBK1 (1:1,000, 3504; Cell Signaling Technology), anti-phospho-TBK1 (1:1,000, 5483; Cell Signaling Technology), anti-OAS3 (1:1,000, 21915; Proteintech), anti-TFAM (1:1,000, ab272885; Abcam), and anti-ATAD3A (1:1,000, H00055210-D01P; Novus Biologicals). Following an additional TBST wash, the membranes were incubated with horseradish peroxidase-conjugated secondary antibodies diluted 1:5000 for 1 h at room temperature: anti-mouse (7076; Cell Signaling Technology), anti-rabbit (7074; Cell Signaling Technology), and mouse anti-rabbit immunoglobulin G (IgG) conformation-specific monoclonal antibody (5127; Cell Signaling Technology). Protein signals were detected using Immobilon® ECL HRP Substrate (WBULS0100; Millipore) and imaged with a Bio-Rad ChemiDoc (Bio-Rad, USA) gel documentation system. All Western blot experiments were performed using three independent biological replicates (n = 3). Representative immunoblots are shown in the figures, and densitometric quantification was performed using data from the three biological replicates.

After 24 h of shNC or shRCN1 lentiviral transfection, NB4 cells were subjected to puromycin selection (2 µg·mL^-1) under conditions with or without 0.5 µΜ STING inhibition (HY-112693; MedChemExpress). Flow cytometry was performed to detect cleaved caspase-1 using a FLICA® 660 Caspase-1 Assay Kit (9122; Immunochemistry)^[11].

Immunoprecipitation (IP) of RCN1 was performed in NB4 cells using a Pierce mass spectrometry (MS)-compatible magnetic IP kit (90409; Thermo Fisher Scientific). Fifty milligrams of the cellular pellet was lysed with 500 μL of immunoprecipitation combined with mass spectrometry (IP-MS) cell lysis buffer following the supplier’s protocol. The lysates were then mixed with 5 μg of IP antibody: anti-RCN1 (ab1989996; Abcam) or a rabbit IgG1 control (ab172730; Abcam) per sample at 4 °C overnight, followed by capture of the immune complexes using MS magnetic beads for 1 h at room temperature with continuous mixing on a rotating platform. Next, the immunoprecipitated beads were washed with IP-MS wash buffer, then eluted with IP-MS elution buffer. The eluted samples were subjected to MS analysis at Jingjie PTM BioLab (Suzhou, China).

A Capturem Co-IP kit (635721, Takara Bio, USA) was used to prepare the samples and perform the IP according to the manufacturer’s protocol. Cell lysate (500 μL) was incubated with 5 μg of antibodies against RCN1, ATAD3A, or TFAM, or with rabbit IgG1 control (ab172730; Abcam). The mixture was applied to a pre-equilibrated Protein A column and centrifuged at 1,000 × g for 1 min at room temperature. After washing the column with 100 µL of wash buffer, bound complexes were eluted with 30 µL of elution buffer. The eluate was centrifuged at 1,000 × g for 1 min to collect the antibody-protein complexes, followed by a western blot analysis as described above. A sample corresponding to 2% of the input lysate was used as a control.

A proximity ligation assay (PLA) was performed using a Duolink In Situ Red Starter Kit (DUO92008; Sigma-Aldrich) in accordance with the manufacturer’s protocol. Cells overexpressing Myc-tagged ATAD3A (HeLa-ATAD3A) were first fixed using 4% paraformaldehyde (PFA) (Thermo Fisher Scientific), followed by permeabilization with 0.1% Triton X-100 and overnight incubation at 4 °C with primary antibodies against RCN1 (1:200, ab1989996; Abcam), ATAD3A (1:100, H00055210-D01; Abnova), Myc tag (1:50, 60003-2-Ig; Proteintech), and TFAM (1:100, sc-376672; Santa Cruz Biotechnology) prepared in blocking buffer. Duolink PLA probes (Anti-Mouse MINUS, DUO92004; Anti-Rabbit PLUS, DUO92002; Sigma-Aldrich) were then added at a 1:5 dilution and incubated at 37 °C for 1 h. After washing, a ligation was performed for 30 min, followed by signal amplification for 100 min using a red detection reagent. Samples were mounted using Duolink In Situ Mounting Medium containing 4’,6-diamidino-2-phenylindole (DAPI). Confocal images were acquired on an LSM800 microscope (Zeiss, Germany) and quantified manually using the “Threshold” and “Analyze Particles” tools in ImageJ/Fiji.

For cellular staining, HeLa and HeLa-ATAD3A cells were first seeded onto 12-mm poly-d-lysine-coated coverslips (Cytoglass, China) in 24-well plates at 5 × 10⁴ cells per well and cultured for 24 h. Cells were first fixed in 4% PFA at room temperature for 15 min, then treated with 0.1% Triton X-100 for permeabilization and incubated in phosphate-buffered saline (PBS) containing 5% FBS for blocking. Samples were incubated with primary antibodies, including anti-DNA (1:200, AC-30-10; Progene), anti-single-stranded DNA binding protein 1 (SSBP1, 1:1,000, 12212-1-AP; Proteintech), anti-RCN1 (1:1,000, ab1989996; Abcam), anti-ATAD3A (1:1,000, H00055210-D01; Abnova), anti-Myc tag (1:500, A190-104A; Thermo Fisher Scientific), and TFAM (1:1,000, sc-376672, Santa Cruz Biotechnology) overnight at 4 °C. After being washed, the cells were treated with Alexa Fluor 647-conjugated donkey anti-rabbit IgG (H+L) (711-605-152; Jackson ImmunoResearch) and Alexa Fluor 568-conjugated mouse anti-rabbit IgG (H+L) (715-575-150; Jackson ImmunoResearch) secondary antibodies for 1 h at room temperature. Alexa Fluor secondary antibodies were used at 1:500, and the coverslips were mounted with DAPI Fluoromount-G (36308ES; YEASEN, China). Confocal imaging was carried out on an LSM 800 microscope (Zeiss, Germany).

RBCEVs were purified following a previously described method^[31]. Peripheral blood (10 mL per donor) was collected from 10 healthy adult volunteers (6 males and 4 females, aged 24-55 years) with no history of chronic diseases or active infections. Collection of peripheral blood samples was approved by the Institutional Review Board of Shenzhen Second People’s Hospital (Approval No. 2023-194-02PJ). Written informed consent was obtained from all participants in accordance with the Declaration of Helsinki. Briefly, red blood cells (RBCs) were prepared from whole blood using a leukodepletion filter (Nigale, China). Vesiculation was induced by incubating RBCs in PBS supplemented with 0.1 mg·mL^-1 CaCl₂ and 10 μM calcimycin (HY-N6687; MedChemExpress), followed by overnight incubation at 37 °C in 5% CO₂. Following vesiculation, cell debris and residual RBCs were removed by sequential centrifugation. The supernatant was subsequently subjected to ultracentrifugation to pellet RBCEVs. The EV pellet was resuspended in PBS and further purified by size-exclusion chromatography (SEC). Purified RBCEVs were collected, suspended in PBS supplemented with 4% D-(+)-Trehalose (HY-N1132; MedChemExpress), and maintained at -80 °C until use.

RBCEVs were processed for transmission electron microscopy (TEM) analysis by fixation in 2% PFA for 10 min at room temperature. Next, 20 μL of the fixed sample was applied dropwise onto 200-mesh grids and allowed to adsorb for 10 min at room temperature. Excess liquid was gently blotted away with filter paper, and the grids were subsequently treated with 2% uranyl acetate for 5 min to achieve negative staining, after which the remaining stain was removed. After air-drying for 10 min, samples were imaged using a JEM1400 transmission electron microscope (JEOL, Tokyo, Japan). RBCEV size distribution, particle concentration, and zeta potential were analyzed using a Zetaview PMX120-Z instrument (Particle Metrix, Meerbusch, Germany). Purified RBCEVs were diluted in 10 mM HEPES buffer to achieve an optimal particle concentration range (1 × 10⁷-1 × 10⁸ particles·mL^-1) and loaded into a pre-calibrated sample chamber. Measurements were performed at 11 distinct positions per sample, and each sample was analyzed in triplicate. The instrument was calibrated using 100-nm polystyrene standard particles prior to analysis. All measurements were conducted at a controlled temperature of 23-25 °C.

The siRNA, including siNC, siRCN1, and Cy5-labeled siRCN1, was commercially synthesized by BGI (Shenzhen, China) and subsequently loaded into RBCEVs via transfection. Following the manufacturer’s protocol, a total of 1 μg of siRNA was transfected into 50 μg of RBCEVs using Exo-Fect^TM Exosome Transfection Kit (System Biosciences, USA). Supplementary Table 3 summarizes the siRNA sequences used in this study. To remove the unencapsulated siRNA and transfection reagents, siRNA-loaded RBCEVs underwent three washes with PBS, followed by centrifugation (21,000 × g) for 30 min after each wash. To determine the efficiency of siRNA loading, the transfected RBCEV pellets were dissolved in 10 μL of nuclease-free water supplemented with 1% Triton X-100. Electrophoretic separation was performed on a 2% agarose gel after the lysate was treated with 6× DNA loading dye (New England Biolabs). A gradient dilution of free siRNA was used as a control. Gels were imaged using a Bio-Rad ChemiDoc gel documentation system.

For a cellular uptake analysis of the siRNA-loaded RBCEVs, A total of 1 × 10⁵ THP-1 cells were seeded into each well and maintained overnight. The cells were treated for 2 h with 20 μg unlabeled or Cy5-labeled siRCN1-loaded RBCEVs. Cells were collected following treatment and subjected to two washes with washing buffer (PBS containing 1% FBS), and the proportion of Cy5-positive cells was analyzed by flow cytometry (Novios; Beckman Coulter, USA).

Female NOD scid gamma (NSG) mice (aged approximately 7 weeks) were obtained from the Shanghai Model Organisms Center (Shanghai, China). All animal procedures followed the National Institute of Health guidelines and were approved by the Institutional Animal Care and Use Committee of Shenzhen Following Precision Medical Research Institute (license no. AP-SZZX-2020-12-013). To establish an AML xenograft model, we subcutaneously injected 1 × 10⁶ THP-1 cells into the right flank of NSG mice. Tumor size was assessed every 3 days using a digital caliper. The corresponding volumes were determined through the following equation: V = ¹/₂ × length × width² (mm³). Upon reaching a tumor volume of approximately 50 mm³, the tumor-bearing mice were randomly assigned to different treatment groups by simple randomization (n = 6 per group) and intratumorally injected with either siRCN1-loaded RBCEVs (siRCN1-EV) or scrambled negative control siRNA-loaded RBCEVs (siNC-EV) every 3 days for a total of four injections. Each injection contained 50 μg RBCEVs, resulting in a cumulative dose of 200 μg EVs per mouse. The siNC-EV group served as the negative control to account for any effects associated with the RBCEV carrier and nonspecific siRNA delivery. Humane endpoints included a tumor volume exceeding 1,500 mm³, ulceration or necrosis of the tumor, body weight loss greater than 20%, or signs of severe distress, impaired mobility, or inability to access food or water. Animals reaching these criteria were euthanized immediately. On day 12 after the first intratumoral injection, all mice were euthanized, and tumors, blood, and organs (including kidney, heart, spleen, and liver) were collected for subsequent analysis.

Throughout the animal experiments, the mice were routinely monitored for signs of toxicity or stress. Upon completion of the treatment, blood samples were collected from mice under isoflurane anesthesia via retro-orbital bleeding. To evaluate potential treatment-related toxicity, biochemical analysis was performed using a Mindray BS-420 automatic biochemistry analyzer (Shenzhen, China), according to the manufacturer’s instructions. The following parameters were measured: alkaline phosphatase (ALP), aspartate aminotransferase (AST), alanine aminotransferase (ALT), creatinine, total bilirubin, and creatine kinase (CK).

At the end of the treatment, AML mouse organs, including the heart, liver, spleen, and kidney, were harvested and fixed overnight in 10% buffered formalin (Thermo Fisher Scientific). The tissues were dehydrated through a graded ethanol series (70%, 95%, and 100%) and embedded in paraffin (Thermo Fisher Scientific) following three baths in molten paraffin at 62 °C. Sections (5 μm thick) were prepared using a microtome (RM2135; Leica). The sections were dewaxed in xylene and rehydrated using a descending ethanol gradient (95%, 90%, 80%, and 70%). Subsequently, the sections were stained with a hematoxylin solution (H8070; Solarbio, China), rinsed with tap water, differentiated in 1% hydrochloric acid alcohol for several seconds, rinsed once more, and blued in 0.6% ammonia water. The sections were counterstained with eosin (E8080; Solarbio, China) for 3-5 min. After dehydration using an ascending ethanol series (70%, 95%, and 100%) and clearing in xylene, the sections were mounted with neutral balsam (G8590; Solarbio, China). Finally, the slides were scanned using a Panoramic 1000 scanner (3DHISTECH Ltd., Budapest, Hungary).

GraphPad Prism 9 was used for statistical analysis. An unpaired two-tailed Student’s t-test was employed to compare data between the control and treated groups. For datasets involving more than two groups, one-way ANOVA followed by Tukey’s multiple comparisons test was used. Two-way analysis of variance (ANOVA) was used for intergroup comparisons. All experiments were performed in at least three independent replicates (n = 6 mice per group for the animal experiments). Statistical significance was set at P < 0.05. Data are shown as mean (standard deviation).

Our previous work suggested that RCN1 downregulation promotes pyroptosis via the type I interferon pathway, and that STING inhibition partially attenuates this effect, implicating engagement of the cGAS-STING pathway^[11]; however, the detailed mechanism has not yet been fully elucidated. Compared with control cells, NB4 cells transfected with shRCN1 lentivirus exhibited increased pyroptosis, while treatment with a 0.05 μM STING-I attenuated the RCN1 downregulation-induced pyroptosis [Figure 1A]. STING-I also partially reversed the RCN1 deficiency-reduced NB4 cellularity [Figure 1B]. RCN1 deficiency markedly reduced the NB4 cellularity [Figure 1B], and similar effects were observed in THP-1 and OCI-AML3 cells [Figure 1C and D]. To determine whether RCN1 knockdown influences innate immune signaling, we assessed the cGAS-STING-type I interferon pathway in AML cell lines. Western blotting revealed increased expression of phosphorylated STING, phosphorylated TBK1, and OAS3 in RCN1-deficient NB4, THP-1, and OCI-AML3 cells [Figure 1E-G], indicating pathway activation. Densitometric analyses of the western blot results further confirmed these findings [Supplementary Figure 1A-C]. Furthermore, qPCR results showed significantly elevated levels of interferon-stimulated genes (ISGs), including IFIT1, IFI27, RSAD2, OAS1, and ISG15, in RCN1-knockdown cells compared to control cells [Figure 1H-J]. In summary, these results indicate that RCN1 downregulation activates the cGAS-STING-type I interferon pathway in AML cells.

Disruption of mitochondrial integrity can lead to mtDNA escape into the cytosol. Given that cytoplasmic mtDNA activates cGAS-STING signaling, we initially measured cytosolic mtDNA levels using qPCR. The cytosolic mtDNA/nDNA ratio increased by approximately 1.63-fold in RCN1-deficient NB4 cells and approximately 1.7-fold in RCN1-deficient THP-1 cells [Figure 2A and B]. To determine whether mtDNA release contributes causally to cGAS-STING activation following RCN1 depletion, we next depleted mtDNA using ethidium bromide (EB) treatment and examined the resulting effects on downstream signaling. Low concentrations of EB reportedly inhibit mtDNA replication and transcription, leading to mitochondrial dysfunction and eventual mtDNA depletion while having little effect on nDNA. Compared to the controls, EB treatment markedly reduced mtDNA content by approximately 77% in RCN1-knockdown NB4 cells and approximately 82% in THP-1 cells [Figure 2C and D].

To directly visualize mtDNA redistribution following RCN1 depletion, we performed confocal microscopy in HeLa cells. RCN1 downregulation in HeLa cells led to an increased accumulation of cytosolic DNA foci that did not co-localize with mitochondrial signals as detected by confocal microscopy following dsDNA and mitochondrial staining, consistent with mtDNA release [Figure 2E]. Confocal microscopy consistently revealed a reduction in mtDNA foci, as indicated by decreased co-localization between DNA and mitochondrial signals, in EB-treated HeLa cells following dsDNA and mitochondrial staining [Figure 2F]. Moreover, EB treatment partially attenuated the cGAS-STING-type I interferon response in RCN1-knockdown NB4 and THP-1 cells [Figure 2G and H] and reduced ISG expression [Figure 2I and J]. Overall, these results suggest that cytosolic mtDNA contributes to cGAS-STING-dependent type I interferon signaling induced by RCN1 downregulation.

Since our findings suggested that RCN1 deficiency promotes mtDNA release and subsequent cGAS-STING activation, we next sought to identify the molecular mechanism linking RCN1 to mtDNA homeostasis. To this end, we performed IP-MS to identify proteins that may interact with RCN1 in wild-type NB4 cells. A bioinformatics analysis revealed that three mitochondrial proteins (ATAD3A, SSBP1, and TFAM) were among the top 10 differentially enriched proteins [Figure 3A and B]. We found that RCN1 deficiency was associated with reduced TFAM expression, with no apparent effect on ATAD3A or SSBP1 levels [Figure 3C]. Because TFAM plays a central role in maintaining mitochondrial nucleoid integrity and preventing mtDNA release, we focused our subsequent analyses on the potential relationship between RCN1 and TFAM. Given previous reports that ATAD3A serves as a key mitochondrial scaffold protein, TFAM is essential for maintaining mtDNA stability, and the ATPase domain of ATAD3A binds to TFAM^[33], we hypothesized that RCN1 downregulation induces mtDNA release by perturbing TFAM function via the ATAD3A scaffold. Co-IP assays using NB4 cells overexpressing RCN1 confirmed that ATAD3A bridges the interaction between RCN1 and TFAM [Figure 3D]. To further assess the intracellular association among these proteins, we performed a PLA to examine the interactions between endoplasmic reticulum (ER)-associated RCN1, mitochondrial ATAD3A, and TFAM. The PLA results further supported this finding. Red punctate signals indicated prominent intracellular interactions between ATAD3A and TFAM as well as between RCN1 and ATAD3A, whereas interactions between RCN1 and TFAM were reduced [Figure 3E]. Samples processed without primary antibodies showed no detectable signals and were used as negative controls (NC). Confocal microscopy further validated these findings. Co-localization signals (shown in yellow) were strong between ATAD3A and TFAM, detectable between RCN1 and ATAD3A, and reduced between RCN1 and TFAM [Figure 3F].

Given that RCN1 deficiency reduced TFAM expression and that TFAM is a key regulator of mtDNA stability, we next investigated whether TFAM mediates the activation of cGAS-STING signaling induced by RCN1 downregulation. Compared with control cells, NB4 and OCI-AML3 cells infected with shTFAM showed decreased cell viability to approximately 63% and 51%, respectively [Figure 4A and B]. The cytosolic mtDNA levels increased by approximately 2-fold in TFAM-deficient NB4 cells and approximately 1.7-fold in TFAM-deficient OCI-AML3 cells [Figure 4C and D]. A western blot analysis further revealed enhanced cGAS-STING-type I interferon signaling in TFAM-knockdown cells, as reflected by elevated levels of phosphorylated STING, TBK1, and OAS3 [Figure 4E and F]. Notably, when TFAM was concurrently overexpressed with RCN1 knockdown in NB4 cells, TFAM upregulation reversed the phenotypic changes induced by RCN1 downregulation, including mtDNA leakage, reduced cell viability, and type I interferon pathway activation [Figure 4G-J]. Taken together, these results demonstrate that RCN1 downregulation triggers cGAS-STING-type I interferon signaling through TFAM dysregulation.

Having identified RCN1 as a potential therapeutic target in AML, we next explored whether RCN1 silencing could be achieved using an RBCEV-mediated siRNA delivery strategy. Given their favorable biocompatibility, high nucleic acid-loading capacity, and biodistribution profile that overlaps with major sites of AML involvement, RBCEVs were selected as delivery vehicles. RBCEVs were isolated from peripheral blood samples donated by healthy volunteers after informed consent using a series of centrifugation steps, followed by SEC [Figure 5A]. Western blot verified the expression of vesicle-specific markers, including tumor susceptibility gene 101, stomatin, glycophorin A, and ALG-2-interacting protein X, whereas the ER marker calnexin was absent. Hemoglobin A and GAPDH, which were highly expressed in both RBCEVs and their parent blood cells, were also detected [Figure 5B]. A ZetaView analysis showed that the average RBCEV diameter was approximately 160 nm [Figure 5C]. Zeta potential measurements using ZetaView indicated a negative surface charge averaging -21 mV [Figure 5D]. TEM images revealed intact RBCEVs with a typical vesicle-like morphology and bilayer membrane structure [Figure 5E]. Similar results were obtained in two additional independent RBCEV preparations [Supplementary Figure 2A-C], indicating low batch-to-batch variability and good reproducibility of RBCEV production. To enable functional studies of RCN1 silencing, RBCEVs were subsequently loaded with siRNA using the Exo-Fect reagent. Agarose gel electrophoresis confirmed a loading efficiency of approximately 80% and was calibrated against a standard curve of free siRNA [Figure 5F].

Upon confirming the loading of siRCN1 into RBCEVs, we first evaluated whether RBCEVs could efficiently deliver siRCN1 into AML cells. Cellular uptake of siRNA-loaded RBCEVs was assessed by incubating THP-1 cells with unlabeled (unlabeled-siRCN1-loaded EV) or Cy5-labeled siRCN1-loaded RBCEVs (Cy5-siRCN1-loaded EV), followed by flow cytometry. We detected up to approximately 95% Cy5-positive cells, indicating high cellular uptake efficiency [Figure 6A]. Notably, THP-1 cells used in this study were maintained in an undifferentiated state and therefore did not exhibit a macrophage-like phenotype. Undifferentiated THP-1 cells showed only modestly higher RBCEV uptake than NB4 and OCI-AML3 cells; PMA-induced macrophage-like THP-1 cells displayed markedly greater RBCEV uptake [Supplementary Figure 3]. Importantly, Cy5-siRNA loading did not noticeably affect the size distribution or morphology of RBCEVs, indicating that EV integrity was preserved [Supplementary Figure 4]. Moreover, free Cy5 dye and free Cy5-siRNA controls confirmed that the detected intracellular fluorescence was primarily attributable to RBCEV-mediated delivery of encapsulated siRNA rather than nonspecific uptake [Supplementary Figure 5]. Having confirmed efficient cellular uptake, we next examined whether RBCEV-mediated delivery resulted in effective RCN1 silencing and anti-leukemic activity. THP-1 cells were then treated with siRCN1-loaded RBCEVs (siRCN1-EV), which significantly reduced RCN1 expression by approximately 67% and cell viability by approximately 54% compared with siNC-loaded RBCEVs (siNC-EV) [Figure 6B-D]. To explore the in vivo therapeutic effects of RBCEVs delivering siRCN1, an AML xenograft model was established by subcutaneously inoculating NSG mice with 1 × 10⁶ THP-1 cells. Mice received intratumoral injections every 3 days. Each injection contained 50 μg RBCEVs (total EV protein quantified by BCA assay), corresponding to an estimated 5.1 × 10¹⁰ particles based on the experimentally determined relationship between EV protein content and particle number [Supplementary Figure 6]. According to the experimentally determined siRNA loading efficiency (~80%), each RBCEV dose contained approximately 0.8 μg encapsulated siNC or siRCN1. The tumor size was measured with a caliper at 3-day intervals [Figure 6E]. Mice treated with siRCN1-EV exhibited significantly smaller tumors and slower leukemia growth than control mice [Figure 6F and G]. No statistically significant differences in body weight were detected during the treatment period [Figure 6H].

At the end of treatment, whole blood, along with the liver, heart, spleen, and kidney, was obtained from mice to evaluate the potential toxicity of siRCN1-EV. Compared with the siNC-EV group, mice treated with siRCN1-EV showed no significant differences in kidney function markers (urea and creatinine) or the muscle damage marker CK [Figure 7A]. Because RBCEVs are prone to accumulation within the liver, potential hepatotoxicity was also assessed. No significant differences were observed in liver toxicity parameters, including ALP, AST, ALT, and total bilirubin levels, between the siRCN1-EV treatment group and the siNC-EV control group [Figure 7B]. Furthermore, H&E-stained sections of liver, heart, spleen, and kidney revealed no significant pathologic changes in siNC-EV- or siRCN1-EV treated mice compared with tumor-free mice, indicating that RBCEV-mediated delivery of siRCN1 is well-tolerated and non-toxic [Figure 7C-F].