Engineering HLA-G-targeted extracellular vesicles nanoplatform for enhanced cancer therapy through precise cancer drug delivery Authors: Ming-You Shie, Shi-Wei Huang, Yeh Chen, Mei-Chih Chen, Chih-Ming Pan, Cheng-Yu Chen, Yen-Hong Lin, Min-Hua Yu, Kai-Wen Kan, Shao-Chih Chiu, Hui-Chun Ho, Yi-Wen Chen, Der-Yang Cho Published: 2025 ABSTRACT AbstractExtracellular vesicles (EVs) hold great potential as a therapeutic delivery system for cancer treatment. Here, we develop an innovative targeted drug delivery platform, human leukocyte antigen-G-VHH antibody-modified EV (α-HLA-G-EV), to enhance therapeutic efficacy. A genetically engineered HEK293T stable clone is utilized to produce α-HLA-G-EV, which are subsequently loaded with chemotherapeutic agents to create HLA-G-targeting drug-loaded EVs (drug@α-HLA-G-EV). The cytotoxicity of drug@α-HLA-G-EV is assessed in various cancer cell lines, demonstrating superior tumor targeting and therapeutic efficacy compared to standard chemotherapies. In vivo experiments using xenograft NPG mouse models, established with MDA-MB-231 and U87 cell lines, further confirm the enhanced antitumor activity of Doxorubicin@α-HLA-G-EV and Temozolomide@α-HLA-G-EV. These findings are consistent with results observed in patient-derived breast cancer and GBM cell models. Additionally, drug@α-HLA-G-EV causes far less damage to normal organs than Lipo-Dox. These findings highlight the potential of α-HLA-G-EV as a versatile platform for precise and efficient cancer treatment.Subject terms: Nanotechnology in cancer, Drug delivery Introduction Chemotherapy is a standard treatment for aggressive or metastatic cancers, especially those lacking specific targets amenable to targeted therapies1. Although chemotherapeutic agents effectively inhibit cancer cell proliferation and trigger cell death, they can also damage normal cells, leading to undesirable side effects2. And potentially leading to drug resistance through various mechanisms3. Appropriate formulation and delivery systems can mitigate the systemic toxicity of chemotherapeutic agents while increasing their effectiveness4–6. Nanoparticle-based drug carriers offer promising therapeutic platforms for cancer treatment by increasing intracellular drug uptake and minimizing side effects compared to conventional formulations7–9. Extracellular vesicles (EVs) endogenously released by cells are nanometer-sized particles and rich in various bioactive molecules including proteins, nucleic acids, and lipids10–12. EVs mediate intercellular communication between nearby or distant cells by delivering EV cargos, indicating the great potential of EVs as novel therapeutic vehicles13–16. The various studies have proven that dendritic cell-derived EVs loaded with tumor peptides increase the immune response to lung cancer17–19. Additionally, numerous studies demonstrated the successful utilization of EVs for drug delivery in cancer treatment with minimal risks of toxicity and immunogenicity20,21. Moreover, mesenchymal stem cell (MSC)-derived EVs have been explored as drug carriers for cancer treatment22–24. While the therapeutic potential of MSC-derived EVs is significant, concerns have arisen regarding their possible role in promoting cancer progression25–28. Several studies have revealed harmful contents in MSC-derived EVs29,30, which may aggravate cancer progression by enhancing tumor angiogenesis and modulating the tumor microenvironment24. Despite their potential as delivery platforms for various antitumor drugs, in vivo experiments show EVs tend to accumulate in the liver or spleen rather than tumor tissues31,32. To enhance therapeutic efficiency, numerous strategies have been employed to modify EVs with specific targeting properties for more precise delivery33–37. Human leukocyte antigen-G (HLA-G) is recognized as an immune checkpoint (ICP) and a tumor-associated antigen (TAA) prevalent in solid tumors38. The H scores of HLA-G are significantly higher in various tumors, including triple-negative breast cancer (TNBC), glioblastoma multiforme (GBM), pancreatic cancer, and ovarian cancer, than in their paired adjacent normal tissues39. Our previous studies have shown that targeting HLA-G presents several promising strategies for developing solid tumor treatments39–42. In this work, we show a stable clone of the HEK293T cells engineered with α-HLA-G VHH antibody-chimeric CD63 protein (called α-HLA-G Nb-chimeric CD63, Fig. 1). This engineering modification promotes the production of EVs featuring α-HLA-G Nb-chimeric CD63 protein on the surface, augmenting their targeting capability towards HLA-G-positive tumor cells through α-HLA-G Nb and enhancing the precision and efficacy of drug delivery. After loading with chemotherapeutic agents (e.g., doxorubicin, temozolomide, and carboplatin), the drug-loaded EVs (abbreviated as DOX@α-HLA-G EV, TMZ@α-HLA-G EV, and CBP@α-HLA-G EV, respectively) are used to treat TNBC, GBM, and ovarian cancer cells, respectively for in vitro and in vivo experiments to evaluate their potential as an advanced drug delivery platform. Finally, the precise delivery and antitumor efficacy of the chemotherapeutic-loaded α-HLA-G EV are further confirmed in patient-derived primary tumor cell models. Schematic diagram of HLA-G targeted-EV drug delivery platform. The diagram shows the engineering of HEK 293T cells to overexpress α-HLA-G Nb-chimeric CD63 fusion protein, resulting in the production of EVs enriched with the fusion protein. Modified EVs (α-HLA-G EV) were loaded with chemotherapeutic drugs to treat different types of cancer. Drugs encapsulated in α-HLA-G EV (drug@α-HLA-G EV) target HLA-G-positive cancer cells by recognizing HLA-G on the cancer cell membrane. The drug@α-HLA-G EV are then internalized by cancer cells, delivering chemotherapeutic drugs to kill cancer cells. Results Generation and characterization of α-HLA-G Nb-expressing EV To establish a genetically modified HEK293T stable clone, a plasmid encoding CD63 gene inserted with α-HLA-G Nb into its extracellular loop was constructed (Fig. 2A) and transfected into HEK293T cells as described in the Materials and Methods. After screening, a stable clone capable of producing EVs displaying α-HLA-G Nb on their surface was established. The engineered HEK293T cells were validated by measuring the expression levels of CD63 and VHH using western blot (Fig. 2B) and flow cytometry analysis (Fig. 2C, D). Targeting-specificity of the engineered EVs towards HLA-G was assessed by the surface plasmon resonance (SPR) binding assay. The consistently low KD value of 7.18 nM indicated a strong affinity of the engineered EVs for HLA-G (Fig. 2E). The results obtained from transmission electron microscopy (TEM) and nanoparticle tracking analysis (NTA) indicated that α-HLA-G EV exhibit spherical morphological characteristics (Fig. S1A) and size distribution of approximately 93 nm (Fig. 2F), which is comparable to unmodified HEK293T-derived EVs. Interestingly, the zeta potential measured via NTA demonstrated that α-HLA-G Nb-modified EVs possess a higher negative surface charge than unmodified EVs, suggesting distinct surface compositions between these groups (Fig. S1B). In addition, 86.3% of VHH-labeled α-HLA-G EV were detected using nanoflow cytometry (Fig. 2G). Western blot analysis indicated increased expression levels of CD63 and VHH in α-HLA-G EV compared with the control EVs, while other EV markers, including CD9, CD81, and Alix, were at identical levels (Fig. 2H). Characterization of α-HLA-G-engineered EV (αHLA-G EV). A Diagram of α-HLA-G Nb chimeric CD63 fusion protein construct and predicted structure, featuring transmembrane domains, an exposed α-HLA-G Nb inserted within the CD63 extracellular domain. B Western blot analysis for CD63 and VHH expression levels in HEK293T cells transfected with the αHLA-G-CD63 fusion construct. Representative Western blot from three independent experiments with similar results. Flow cytometry analysis for C CD63 and D VHH expression in HEK293T cells. E The binding affinity of α-HLA-G EV to HLA-G was determined via SPR binding affinity assay with αHLA-G EV (1 × 10¹¹ particles) coated on CM5 chips. F Nanoparticle tracking analysis (NTA) of EVs. G Flow cytometry analysis for the expression of α-HLA-G on EV surface by anti-VHH antibody. Data are representative of n = 6 independent experiments with similar results. H Western blot analysis for VHH levels and EV markers, including CD63, CD81, CD9, Alix, and VHH. Representative Western blot from three independent experiments with similar results. Data are representative of n = 6 independent experiments with similar results. Assessment of specificity of α-HLA-G EV against solid tumor cells To evaluate the specificity of α-HLA-G EV for targeting solid tumor cells, α-HLA-G EV and control EV were labeled by ExoGlow™ dye green fluorescence and subsequently used to treat cancer cell lines, including MDA-MB231 (TNBC), while the evaluations for U87 (GBM) and SKOV3 (ovarian cancer) are presented in Figs. S3 and S4, respectively, along with their corresponding non-tumorigenic cells (breast epithelial MCF10A cells, astrocytes, and ovarian fibroblast). The higher levels of HLA-G in tumor cell lines than in their corresponding non-tumorigenic cells were identified (Fig. 3A). EV-uptake by tumor cells and non-tumorigenic cells was visualized at indicated time points through a confocal microscope, which revealed a significantly greater uptake of α-HLA-G EV at the specified time points in tumor cells (Fig. 3B). We further confirmed the uptake of EVs through flow cytometry analysis (Fig. 3C), indicating a significantly higher uptake of α-HLA-G EV in MDA-MB-231 (uptake by 28.5% of cells at 6 h and 94.5% at 24 h). Conversely, MCF10A showed minimal responses to α-HLA-G EV, with an uptake capacities of 1.9% at 24 h after administration. The uptake analysis of U87 and SKOV3 is presented separately in Fig. S3 and Fig. S4, showing a time-dependent increase in α-HLA-G EV internalization, with U87 cells reaching 44.3% uptake at 6 h and 76.6% at 24 h, and SKOV3 cells reaching 70.7% at 6 h and 99.5% at 24 h. In contrast, their respective non-tumorigenic counterparts, astrocytes and ovarian fibroblasts, and exhibited minimal uptake, with rates of 0.9% and 2.1%, respectively, at 24 h. The uptake and targeting of α-HLA-G EV across different cell types. A HLA-G expression levels on MDA-MB-231 and MCF10A. B Confocal microscopy images showing EV and α-HLA-G EV uptake by MDA-MB-231 and MCF10A at 6, 12, and 24 h. EVs were labeled with ExoGlow (green), the cytoskeleton with phalloidin (red), and nuclei with DAPI (blue). Representative images shown from three independent experiments. The scale bar is 100 µm. C Flow cytometry analysis showing EV and α-HLA-G EV uptake by the respective cell types over time. D Flow cytometry quantification and E Confocal microscopy images showing α-HLA-G EV targeting efficiency in co-culture systems of MDA-MB-231 and MCF10A. Representative images shown from three independent experiments. The scale bar is 100 µm. Data are representative of n = 6 independent experiments with similar results. Furthermore, a co-culture system consisting of HLA-G positive tumor cells and their corresponding normal cells was established to evaluate the specific targeting capacity of α-HLA-G EV through flow cytometry analysis (Fig. 3D). After incubation with α-HLA-G EV for 24 h, the results showed that more than 90% of tumor cells had internalized α-HLA-G EV, showing an absolute difference from their corresponding normal cells (less than 2%). In addition, α-HLA-G EV (green) uptake was significantly greater in tumor cells (red) compared to that of their corresponding normal cells (purple) (Fig. 3E). These results suggest that α-HLA-G EV demonstrate specific targeting functionality towards HLA-G positive cells, leading to their uptake by these tumor cells while minimally interacting with normal cells. In order to analyze the potential regulatory role of soluble HLA-G (sHLA-G) in EV binding, was examined by quantifying sHLA-G secretion and evaluating its impact on EV targeting efficiency. The results revealed a positive correlation between sHLA-G secretion and cell density (Fig. S2A). Then, α-HLA-G EVs were pre-incubated with sHLA-G before co-culturing with MDA-MB-231 cells. The results indicated a marginal but statistically insignificant reduction in EV uptake, from 94.9% in α-HLA-G EV compared to 91.3% in sHLA-G pre-treated α-HLA-G EV (Fig. S2B). These findings suggest that while sHLA-G may partially compete with membrane-bound HLA-G for EV interaction, its effect is minimal and does not significantly impair targeted delivery. Characterization of drug-loaded α-HLA-G EV First-line chemotherapy drugs for various cancer types, including DOX (free or as the pegylated liposomal formulation Lipo-Dox, which is designed to enhance drug stability and reduce systemic toxicity) for TNBC, TMZ for GBM, and CBP for ovarian cancer, were used for drug loading. Lipo-Dox was specifically included as a control to evaluate and compare the drug delivery efficiency and safety of α-HLA-G EV. After loading with DOX, the physicochemical attributes and intracellular trafficking of the DOX-loaded EVs were assessed. The TEM nano-architecture images revealed that both α-HLA-G and control EVs exhibited spherical profiles after loading with DOX (Fig. 4A). As depicted in Fig. 4B, quantification of the particle concentration on the size spectrum measured via NTA confirmed that the size of α-HLA-G EV was consistent with that of the control EV, and there was no significant difference with the loading of DOX (the modal diameter was uniform at 108 nm). The characteristic peaks of DOX were discernible, confirming that DOX was successfully encapsulated in α-HLA-G EV (Fig. 4C). The scatter plots from nano flow cytometry showed that approximately 75% of both α-HLA-G EV and control EV were positive for DOX-loading, indicating successful encapsulation of DOX within the EVs (Fig. 4D). The pH of typical physiological environment is about 7.4, while the tumor microenvironment is generally acidic. Therefore, we compared the release amount of DOX from EVs and α-HLA-G EV in PBS at pH 7.4 and pH 6.4 over time (Fig. 4E). Temporal dynamics of the DOX release profile showed the prolonged and stable release of DOX from both EVs at pH 6.4, with no significant difference between the control EV and α-HLA-G EV. However, the amount of DOX released under pH 7.4 is relatively small. This suggests that DOX@EV may release more drugs faster in the tumor microenvironment than in the normal physiological environment. In addition, fluorescence images highlighted the internalization of DOX, and the colocalization of DOX with α-HLA-G EV in MDA-MB-231 cells in the DOX@α-HLA-G EV treatment group for 12 h (Fig. 4F). Physicochemical properties and intracellular trafficking of drug-loaded EVs. A TEM images (The scale bar is 100 nm), B NTA depicting EV size distribution, and C UV-Vis spectroscopy confirmed DOX encapsulation within EVs. D Analyzing the proportion of EVs carrying Dox using nanoflow cytometry. E Release kinetics of DOX from EVs over time in PBS with pH 6.4 and pH 7.4 conditions. Data are representative of n = 4 independent experiments with similar results. Data are presented as mean ± SD. P values were calculated using one-way ANOVA with Tukey’s post hoc test. F Confocal microscopy images of MDA-MB-231 after co-culture with DOX@EV and DOX@α-HLA-G EV for 12 h, respectively. F-actin (purple), nuclei (blue), EV (green), and DOX (red). The scale bar is 100 µm. The impact of α-HLA-G EV uptake on immune modulatory marker expression and tumor-immune interactions was assessed through in vitro assays (Fig. S5). After treated with α-HLA-G EV, the expression of HLA-G and PD-L1 of cancer cells was slightly changed, but there was no statistical significance. Meanwhile, co-culture experiments with PBMCs showed no significant changes in PBMC-mediated tumor cytotoxicity. These findings indicate that α-HLA-G EV may not affect immune marker expression or modulate tumor-immune interactions under the tested conditions. Moreover, to investigate the role of HLA-G in α-HLA-G EV-mediated drug delivery, both HLA-G overexpression, knockdown, and knockout models were generated in MDA-MB-231 cells. The knockdown model, established through transfection, resulted in partial overexpression or suppression of HLA-G expression, while CRISPR/Cas9-mediated knockout led to a complete loss of HLA-G (Fig. S6A-B). The expression of HLA-G in cancer cells does not affect cell activity (Fig. S6C). Comparative cytotoxicity analysis revealed that DOX@α-HLA-G EV induced markedly lower cell death in HLA-G knockout cells than in wild-type MDA-MB-231 cells, underscoring the dependence of EV uptake and drug efficacy on HLA-G expression (Fig. S6D). Notably, the reduction in cytotoxicity in the knockout model was comparable to that observed in the DOX@EV-treated group in Fig. 5A, reinforcing the essential role of HLA-G in facilitating targeted drug delivery via α-HLA-G EV. In addition to genetic manipulation, we further employed pharmacological approaches to modulate HLA-G expression levels. Treatment with interferon-γ (IFN-γ) or azacytidine significantly upregulated HLA-G expression in MDA-MB-231 cells (Fig. S6E). These treatments enhanced the surface presentation of HLA-G and consequently promoted the uptake efficiency of α-HLA-G EV, confirming the dependence of EV internalization on HLA-G levels (Fig. S6F). Evaluation of cytotoxic effects of DOX-loaded EV on MDA-MB-231 cells. A Cell viability assay of MDA-MB-231 cells treated with free-DOX, liposomal Dox (Lipo-Dox), EV, α-HLA-G EV, DOX@EV, and DOX@α-HLA-G EV. Data are representative of n = 6 independent experiments with similar results. Data are presented as mean ± SD. P values were calculated using one-way ANOVA with Tukey’s post hoc test. B Flow cytometric analysis of MDA-MB-231 cells stained with propidium iodide (PI) and Annexin V, indicating apoptosis progression post-treatment. C Live/Dead staining of MDA-MB-231 cells demonstrating cell death upon treatment with DOX encapsulated by different formulations for 24 h. The scale bar is 400 µm. D Assessment of ROS generation in MDA-MB-231 cells after various formulated DOX treatments by flow cytometry. E Western blot bands and F quantification of apoptotic markers including PARP, cleaved-PARP, Bcl-2, Bax, Caspase 3, and cleaved-Caspase 3 in MDA-MB-231 cells following various treatments, with β-actin served as loading control. Data are representative of n = 3 independent experiments with similar results. Data are presented as mean ± SD. P values were calculated using one-way ANOVA with Tukey’s post hoc test. G Assessment of ROS generation in AC16 cardiomyocytes after various formulated DOX treatments by flow cytometry. Additionally, the spherical morphology maintenance was observed in TMZ- and CBP-loaded α-HLA-G EV (Figs. S7A and S7E, respectively). The size distributions of TMZ@α-HLA-G EV and CBP@α-HLA-G EV were consistent with their corresponding control groups (TMZ@EV vs. TMZ@α-HLA-G EV and CBP@EV vs. CBP@α-HLA-G EV) with no significant difference (Fig. S7B and S7F, respectively). UV-Vis spectroscopic analysis confirmed the encapsulation of TMZ and CBP, with characteristic absorbance peaks visible in TMZ@α-HLA-G EV (Fig. S7C) and CBP@α-HLA-G EV (Fig. S7G), respectively. The release kinetics of TMZ from both TMZ@EV and TMZ@α-HLA-G EV, as well as CBP from CBP@EV and CBP@α-HLA-G EV, exhibited sustained and gradual release profiles under acidic conditions (pH 6.4) over 48 hours, respectively (Fig. S7D and S7H). In vitro antitumor activity of α-HLA-G EV-encapsulated chemotherapy drugs The in vitro antitumor efficiency of DOX encapsulated in various formulations against MDA-MB-231 cells was assessed at 2 µM of DOX treatment in each group. Compared with other formulations, DOX encapsulated in α-HLA-G EV showed the greatest anticancer potency against MDA-MB-231 cells (Fig. 5A). One day after treatment with DOX@α-HLA-G EV, approximately 60% of cells were killed, with cell death exceeding 85% three days post-treatment. Notably, the cytotoxicity of DOX@α-HLA-G EV was significantly greater than unformulated-DOX and Lipo-DOX. Furthermore, the results of apoptosis analysis with Annexin V/PI staining demonstrated that after treatment with different formulations of DOX for 3 days, DOX@αHLA-G EV presented a most effective anticancer activity, leading to apoptosis in more than 80% of MDA-MB-231 cells (Fig. 5B). As shown in Fig. 5C, the red fluorescence intensity, indicating the distribution of DOX in the nuclei, was higher in the DOX@α-HLA-G EV group than in other groups. The dead cell proportion (green) was higher in the DOX@α-HLA-G EV treatment group. These indicate that α-HLA-G EV could effectively target HLA-G positive MDA-MB-231 cells and deliver DOX into the cells, inducing cell death. Moreover, the intracellular ROS generation in MDA-MB-231 cells was significantly enhanced after treatment of DOX@α-HLA-G EV for 24 h (Fig. 5D). The observed decrease in Bcl-2 levels, coupled with the concurrent increase in apoptotic markers, including PARP1, cleaved PARP1, Bax, and cleaved Caspase-3 (Fig. 5E, F), aligns with the intensity of apoptosis induced by various formulations of DOX, as illustrated in Fig. 5B. Following this, ROS analysis was performed in the AC16 human cardiomyocyte cell line after treatment with different DOX formulations to evaluate the cardiotoxicity associated with these treatments. The results showed that free-DOX enhanced ROS production in AC16 cells (80.3%), while the ROS levels induced by DOX@αHLA-G EV were relatively low (5.4%) (Fig. 5G). The release kinetics of DOX from α-HLA-G EV were assessed under normoxic and hypoxic conditions to simulate the tumor microenvironment. As shown in Fig. S8, DOX release exhibited a steady increase over 48 h under hypoxia, whereas drug release remained minimal under normoxia. These results indicate that the hypoxic environment enhances the release of DOX from α-HLA-G EV, suggesting a potential mechanism for improved drug delivery efficiency in solid tumors. To further assess the cytotoxic efficacy of α-HLA-G EV as a platform for chemotherapy drugs to kill cancer cells, U87 and SKOV3 cell lines were treated with TMZ@α-HLA-G EV and CBP@α-HLA-G, respectively. The cell viability assay demonstrated a significant reduction in viability over time. The cytotoxicity of HLA-G-EV-formulated-chemotherapy drugs was greater than non-formulated drugs, with the strongest cytotoxic effects observed after 3 days of TMZ@α-HLA-G EV treatment in U87 cells (Fig. S9A) and CBP@HLA-G EV treatment in SKOV3 cells (Fig. S9B), respectively. Flow cytometry analysis further confirmed a higher level of apoptosis in the α-HLA-G EV-encapsulated chemotherapy groups compared to the control and non-encapsulated chemotherapy groups, as indicated by the increased number of Annexin V/PI double-positive cells, indicating late-stage apoptosis. In addition, Live/Dead staining corroborated these findings, showing that TMZ@α-HLA-G EV and CBP@α-HLA-G EV induced higher proportions of cancer cell death compared to their control and naked drugs, respectively, 48 h post-treatment. To further assess the efficacy of HLA-G EV as a precision delivery platform, Western blot analysis confirmed HLA-G expression across multiple solid tumor cell lines, including bladder cancer cells (T24), pancreas cancer cells (MIA PaCa), oral cancer cells (OECM), lung cancer cells (A549), cervical cancer cells (HeLa), and colorectal cancer cells (Caco-2), supporting their classification as HLA-G-positive cancer cells (Fig. S10). Different α-HLA-G EV-encapsulated chemotherapeutic drugs were then used to treat these tumor cells (Fig. S11). The greater uptake of α-HLA-G EV by HLA-G positive cancer cells, coupled with the significant rise in cytotoxicity induced by α-HLA-G EV-encapsulated chemotherapy drugs, underscores their enhancement of targeting and therapeutic efficacy, highlighting the potential of α-HLA-G EV as a precision drug delivery platform for cancer therapy. Tumor-targeted delivery efficacy of α-HLA-G EV An in-situ TNBC xenograft NPG mice model was established to evaluate the delivery specificity of α-HLA-G EV as described in the Materials and Methods. The IVIS imaging revealed a progressive accumulation of signal intensity in the tumor region in DOX@αHLA-G EV-treated mice (Fig. 6A), while both DOX@αHLA-G EV and DOX@EV also accumulated in the liver, lungs, and spleen (Fig. 6B). Quantitative analysis confirmed that the fluorescence intensity in the tumor tissue of the DOX@αHLA-G EV group was seven times that of DOX@EV, further showing that DOX@αHLA-G EV still have excellent tumor-targeting capability after administration for 24 h. Afterward, vital organs, including the brain, heart, lungs, liver, kidneys, spleen, and muscles, were harvested and subjected to TUNEL staining to consider the effect of DOX-loaded EVs on these organs. The results shown in Fig. 6C revealed that the structural integrity of most organs was very similar among the three groups, and no changes in organ damage were found. In addition, a GBM xenograft mice model was established to evaluate the anti-tumor efficiency of TMZ-loaded α-HLA-G EV. A progressive accumulation of signal intensity in the tumor region was observed within 1 h after administration of TMZ@HLA-G EV (Fig. S12A and S12B), with signals also visible in the liver, lungs, and spleen. The significantly higher intensity in the tumor tissues implies the precise tumor-targeting ability of the TMZ@α-HLA-G EV. The results in Fig. S12C indicated no detectable damage or abnormalities in the examined organs of the TMZ@α-HLA-G EV-treated mice compared to the untreated mice. In vitro delivery specificity and biodistribution of DOX@α-HLA-G EV. A in vivo fluorescence images of orthotopic MDA-MB-231 tumor-bearing mice treated with DOX@EV and DOX@α-HLA-G EV. B Representative ex vivo fluorescent images of DiR-labeled EVs in inoculated tumors and vital organs. The quantification of total fluorescence intensity is shown below. Data are representative of n = 3 independent experiments with similar results. Data are presented as mean ± SD. P values were calculated using one-way ANOVA with Tukey’s post hoc test. C Representative photographs of different organ tissue sections with TUNEL staining. Representative images shown from three independent experiments. n = 6 mice per group. The scale bar is 5 mm and 200 µm. In vivo therapeutic efficacy of α-HLA-G EV-encapsulated chemotherapy drugs To compare the in vivo anticancer efficacy of DOX@α-HLA-G EV with various formulations of DOX, TNBC xenograft NPG mice were established by orthotopic implantation of MBA-MD-231 cells, and the inoculated tumors were collected on day 42 after the first dose of treatment. The treatment regimens are listed in Fig. 7A. The significantly reduced tumor volume in mice treated with DOX@α-HLA-G EV verified its suppression effect on tumor growth (Fig. 7B). Tumor growth monitored by the IVIS system indicated that compared with other dosage forms, DOX@α-HLA-G EV most effectively suppressed tumor growth and prolonged the survival of mice (Fig. 7C–E). After the last treatment (day 35), tumor progression and survival rate worsened sharply in most treatment groups. In contrast, the tumor progression was consistently slow, and the survival was extended up to day 98 in the DOX@α-HLA-G EV group. Throughout the treatment period, no significant differences in body weight were observed among the groups (Fig. S13) In vivo evaluation of anticancer effects of DOX-loaded EV. A Schematic representation of in vivo experimental design indicating schedule for treatment and subsequent assessments in CDX-TNBC NPG mice model. B Image showcasing the differences between tumors collected from each treatment group, illustrating size and shape differences. C Bioluminescence imaging using IVIS at various time points post-treatment, showing effects of EV, α-HLA-G EV, DOX, Lipo-Dox, DOX@Exo, and DOX@α-HLA-G EV on tumor progression. Group sizes: n = 6 mice per group; DOX@α HLA G EV, n = 8 mice. D Graphical representation of tumor growth tracking for 56 days from first treatment across different groups. Data are representative of n = 6 independent experiments with similar results. Data are presented as mean ± SD. P values were calculated using one-way ANOVA with Tukey’s post hoc test. E Kaplan–Meier plots to depict the survival of mice under various treatments. The collected tumors were then subjected to TUNEL assay and IHC staining, along with their quantification, to assess Bax, Bcl-2, and Ki67 expression. The results presented in Fig. 8A suggest that DOX@α-HLA-G EV treatment led to increased apoptosis and reduced proliferation within tumor nodules. In addition, mouse hearts were collected and analyzed using TUNEL assays and their quantification to assess the potential cardiac cytotoxicity induced by the different formulations of DOX. The higher levels of TUNEL-positive cells in the heart after DOX (36.2%) and Lipo-Dox (27.3%) treatments indicated increased cardiac apoptosis compared to the control (2.0%) and α-HLA-G EV (2.1%) groups. Conversely, the occurrence of TUNEL-positive heart cells in the mice treated with DOX@α-HLA-G EV (2.1%) was rare and comparable to the control and α-HLA-G EV groups, suggesting minimal cardiac cytotoxicity (Fig. 8B). Correspondingly, the expression levels of HLA-G in various types of human primary cells were evaluated, and cytotoxicity assessment after treatment with different formulations of DOX was performed. The results showed that free-DOX and Lipo-DOX treatments exhibited significant cytotoxicity on normal cells, whereas α-HLA-G EV had minimal cytotoxic effects on normal cells (Fig. S14). In vivo assessment of antitumor efficacy and cardiac safety of DOX-loaded EVs. A Histological and quantitative analysis revealed apoptosis and cellular proliferation levels in tumor tissues through TUNEL, Bcl-2, Bax, and Ki67 staining. The scale bar is 75 μm. Data are representative of n = 6 independent experiments with similar results. Data are presented as mean ± SD. P values were calculated using one-way ANOVA with Tukey’s post hoc test. B TUNEL assays and quantitative analysis evaluating apoptosis induction in hearts collected from CDX-TNBC NPG mice treated with various formulated DOX and EVs. The scale bar is 75 μm. Data are representative of n = 6 independent experiments with similar results. Data are presented as mean ± SD. P values were calculated using one-way ANOVA with Tukey’s post hoc test. Subsequently, a GBM orthotopic mouse model was utilized to further evaluate the therapeutic effect of TMZ@α-HLA-G EV. The treatment schedule is presented in Fig. S15A and S15F. Tumor progression was monitored using the IVIS imaging system at different time points (Fig. S15B). The quantitative results demonstrated that TMZ@α-HLA-G EV significantly inhibited tumor growth over time (Fig. S15D) and extended the survival of mice (Fig. S15E) compared to control, α-HLA-G EV, and TMZ treatment groups, while body weight remained stable across all treatment groups (Fig. S15C). After completion of the experiment, mice were sacrificed, and their brain tissues were harvested for histological evaluation through H&E, TUNEL, and Ki67 stains (Fig. S16B), and the quantification of these staining is shown in Fig. S16C, S16D, and S16E. The results of H&E staining showed a significant reduction in tumor area and preservation of brain architecture after TMZ@α-HLA-G EV (2.4 cm2) treatment compared to Ctl (27.4 cm2), α-HLA-G EV (21.5 cm2), and TMZ (14.7 cm2) groups. Additionally, TUNEL staining indicated a higher level (86.3%) of apoptosis in the brain tissues of the TMZ@α-HLA-G EV-treated mice, with Ki67 staining (0.1%) demonstrating a marked decrease in cell proliferation in the tumor region. These results suggest that TMZ@α-HLA-G EV effectively promote apoptosis and suppress tumor progression in a GMB mouse model. In vitro response of patients-derived cancer cells to α-HLA-G EV-encapsulated chemotherapy drugs To further explore the therapeutic potential of chemotherapy drug-loaded α-HLA-G EV, primary cells isolated from breast cancer and GBM tumor tissues were treated with DOX@α-HLA-G EV and TMZ@α-HLA-G EV, respectively. The response of primary tumor cells was assessed. Primary breast cancer cells from two patients exhibited high proportions of HLA-G-positive cells (76.6 and 85.7%) (Fig. 9A, B). In these HLA-G-positive breast cancer cells, uptake of DOX@α-HLA-G EV was markedly higher than that of DOX@EV, as demonstrated by immunofluorescence imaging, which showed colocalization of F-actin and EVs. This targeting capability of α-HLA-G EV was further confirmed by flow cytometry, revealing that a large proportion of the primary cancer cells internalized the α-HLA-G EV, significantly surpassing the uptake of control EV. Furthermore, the viability of primary breast cancer cells after treatment with DOX@α-HLA-G EV for 24 and 48 h was evaluated, significantly dropping to below 50% in both patient samples, highlighting the therapeutic efficacy of the α-HLA-G EV-encapsulated DOX. Analysis of the cytotoxic effects of DOX@α-HLA-G EV on patient-derived primary breast cancer cells. The primary breast cancer cells isolated from (A) patient 1 and B patient 2 were analysed HLA-G expression by flow cytometry. Comparison of EV and α-HLA-G EV internalization by cells through fluorescence image. and flow cytometry analysis for uptake by cancer cells. The scale bar is 100 μm. Cell viability assay measuring the effect of the indicated treatments on patient-derived cancer cells at 24 h and 48 h post-treatment. n = 2 patients. Data are representative of n = 6 independent experiments with similar results. Data are presented as mean ± SD. P values were calculated using one-way ANOVA with Tukey’s post hoc test. Primary GBM cells, derived from two different patients, with 81.2% and 64.6% positivity of HLA-G expression, were performed to identify the efficacy of TMZ@α-HLA-G EV (Fig. S17A and S17B). Consistent with the observations in the primary breast cancer cells, the GBM cells exhibited substantial internalization of α-HLA-G EV. The high uptake rates of HLA-G EV (88.6% and 61.2%) indicated effective targeting of the α-HLA-G EV to these HLA-G-positive GBM cells. The cytotoxicity of the TMZ-loaded α-HLA-G EV was pronounced, with significantly greater reductions than naked-TMZ in cell viability observed after 24 and 48 h of treatment. These results demonstrate the effective tumor-targeting and cytotoxicity of drug-loaded α-HLA-G EV in both primary breast cancer and GBM cells, supporting the potential clinical applications of this precision-targeted drug delivery platform. Discussion Nanotechnology has contributed substantially to the development of novel carriers for precise drug delivery in recent decades43–45. Lipid-based carriers provide a versatile platform for drug encapsulation, facilitating the clinical translation of diverse formulations, such as Lipo-Dox. In addition to synthetic carriers, cell-derived EV-based carrier systems have also attracted considerable interest46–48. Anchoring targeting moieties specific to ligands or receptors exclusively present on tumor cells to the surface of therapeutics is a conventional strategy to enhance drug delivery and accumulation in cancerous tissues49. However, the success of clinical translation is usually limited due to the multiple mechanisms orchestrated by complex tumor microenvironments, which may disrupt targeting efficiency50. Based on their natural characteristics, EVs are a potential drug delivery platform with minimal safety concerns13,51. Although natural EVs have been proposed for delivering antitumor drugs, they are mostly retained in the liver or spleen rather than in tumor tissue31,32. Therefore, improving the specificity and efficiency of EV cargo delivery is crucial to enhance therapeutic efficacy, reducing off-target consumption by normal cells and minimizing toxicity52. Utilizing EVs with external or internal modifications is a promising approach for precisely delivering therapeutic cargo within the tumor region37,53. Genetic engineering of source cells has been demonstrated to produce EVs decorated with specific antibodies for directing them toward cancer cells37,54. In addition to the decoration techniques, selecting specific tumor targets is a primary factor determining the targeting efficacy of engineered EVs. This study selected HLA-G as the target molecule of solid tumors due to its prevalence across a broad spectrum of solid tumors41. Our previous studies have featured the potential of targeting HLA-G for developing therapeutic strategies that precisely target solid tumors39,55. Subsequently, inspired by the tetraspanin nature of CD63, which is enriched on the membrane of cell-derived EVs, we implemented a targeting strategy by inserting α-HLA-G-Nb into the CD63 construction. The plasmid encoding α-HLA-G-Nb-chimeric CD63 construct is transfected into HEK293T cells, as these cells are known for reduced immunogenicity and toxicity risks, making them suitable for clinical applications56. After transfection, a genetically engineered HEK293T stable clone is selected to produce modified EVs displaying α-HLA-G-Nb on their surface. This approach may hold the promise of α-HLA-G EV as a universal drug carrier platform for precise delivery, overcoming the challenges of cancer therapy. The In vitro experiments demonstrated that α-HLA-G-EV accurately identified HLA-G-positive cancer cells (Fig. 3). Additionally, the chemotherapy drugs encapsulated by α-HLA-G-EV showed improved intratumorally biodistribution compared to those encapsulated by unmodified EVs (Fig. 6). Further supporting this observation, DOX release is significantly enhanced under hypoxia, with a steady increase over 48 h, whereas release remained minimal under normoxia (Fig. S8). These findings suggest that the hypoxic tumor microenvironment may facilitate drug release from α-HLA-G EV, further supporting their applicability in tumor-targeted drug delivery. This observation aligns with previous reports demonstrating that hypoxia enhances the degradation of EVs and accelerates the release of encapsulated therapeutics57. The stable expression of HLA-G and PD-L1 following α-HLA-G EV uptake indicates that these engineered EVs primarily serve as drug carriers rather than immune modulators, maintaining the immune phenotype of tumor cells (Fig. S5). Additionally, the lack of significant changes in PBMC-mediated tumor cytotoxicity further supports their specificity for HLA-G-positive tumor cells without altering tumor-immune interactions. Notably, a minor increase in PD-L1 levels was observed following α-HLA-G EV treatment, which aligns with previous reports suggesting that HLA-G may upregulate PD-L1 expression39. While this effect was not substantial, it warrants further investigation to fully elucidate potential immunological implications. Collectively, these findings reinforce the feasibility of α-HLA-G EV as a precise tumor-targeted drug delivery system with minimal off-target immune effects. These indicate that our approach to genetically engineering EVs with targeted capabilities has proven successful, and that accurately targeting tumor cells has the advantage of reducing vital organ damages caused by chemotherapeutics58,59. Maintaining the native size distribution of EVs is essential for their biological functions and cellular uptake22. Following drug loading, α-HLA-G-EV did not change their morphology and size distribution (Fig. 4), implying a beneficial chemotherapeutic drug carrier of α-HLA-G-EV. In addition, Wei et al. have established DOX-loaded MSC-derived EVs and suggested a greater cellular uptake and antitumor efficiency, while low cytotoxicity to myocardial cells compared to the free DOX22. This highlights the potential of EVs as a drug delivery platform to treat cancer. The present study corroborates previous findings and advances α-HLA-G EV-encapsulated DOX therapy. Enhanced internalization of DOX and colocalization with α-HLA-G EV in MDA-MB-231 cells confirm that α-HLA-G EV could improve their targeting specificity, thus enhancing the precise delivery of chemotherapeutics (Fig. 4F). Moreover, the in vivo experiment indicates minimal toxicity of DOX@α-HLA-G EV to cardiomyocytes (Fig. 5F) and heart (Fig. 7G) compared with free-DOX and Lipo-DOX. Engineered EVs exhibit increased binding affinity and their uptake by target cells expressing the corresponding receptors/ligands60. This increased internalization of EVs promotes a “Trojan horse” effect, whereby EVs modified with specific molecules facilitate the crossing of cellular barriers and enable more efficient delivery61. The findings in our study validate this concept. Specifically, in vitro experiments revealed that the cytotoxicity of DOX encapsulated in α-HLA-G EV was greater than twice that of free-DOX (Fig. 5A). The same trend was also observed in TMZ and CBP encapsulated in α-HLA-G EV (Fig. S9). Moreover, compared with the findings in other literature, the efficacy of DOX@α-HLA-G EV in killing TNBC in this study was much higher than that of free-DOX62–64. This remarkable cytotoxicity of α-HLA-G EV-encapsulated chemotherapy drugs against different cancer cells underscores the enhanced potency achievable with lower concentrations of these drugs, further accentuating the potential of α-HLA-G EV as a precise delivery system for chemotherapeutic agents. The in vivo experiments demonstrate significant antitumor efficacy of DOX@α-HLA-G EV and TMZ@α-HLA-G EV, the remarkable internalization of α-HLA-G EV and growth inhibition of α-HLA-G EV-encapsulated DOX and TMZ were observed in patient-derived primary cancer cells (Fig. S16). The off-target and on-target off-tumor toxicities are critical for developing innovative antitumor therapies. These observations suggest minimal toxicity of DOX@α-HLA-G EV treatment to vital organs and highlight the potential of α-HLA-G EV as a promising chemotherapy delivery platform for future clinical applications. In addition to the superior targeting and therapeutic efficiency of α-HLA-G EV-encapsulated chemotherapy drugs for the treatment of breast cancer, GBM, and ovarian cancer, the efficacy of the α-HLA-G EV platform to precisely deliver different chemotherapy drugs against various HLA-G-positive cancer cell lines was further confirmed (Fig. S9). The results suggest that the robust targeting capability of α-HLA-G EV can be harnessed to load and deliver a variety of chemotherapeutic agents, and even be expended to load other agents, such as nucleic acids and immune-modulating compounds. Future developments may leverage α-HLA-G EV to create tailored treatments for different cancer types, addressing their unique molecular profiles and therapeutic challenges. In summary, we introduced an innovative approach that utilizes modified EVs derived from genetically engineering HEK293T cells to overexpress α-HLA-G Nb chimeric CD63 for the targeted delivery of chemotherapy drugs. The α-HLA-G EV precisely target HLA-G-positive cancer cells, enhancing drug internalization into target cells while minimizing consumption by surrounding tissues. Compared with conventional free drug treatment, drug@α-HLA-G EV exhibit superior antitumor efficacy in vitro and in vivo. The minimal toxicity observed in vital organs implies the safety of this targeted drug delivery system, providing an alternative to conventional approaches that are often compromised by systemic toxicity and potential cardiac risks. Taken together, we suggest that α-HLA-G EV as a targeted drug delivery platform holds promise for revolutionizing cancer treatment by offering a precise, efficient, and minimally toxic method for administering anticancer drugs. Methods Cell line and culture conditions MDA-MB-231 (Cat No: HTB-26) was purchased from ATCC and maintained in DMEM (Gibco) with 10% Fetal Bovine Serum (FBS, Gibco) and 1% penicillin/streptomycin (Gibco), and 1% NEAA (Gibco) and cultured at 37 °C with 5% CO2 in air. The medium was switched every three days to encourage optimal growth, and subculturing was done as required based on cell confluency. U87 (Cat No: HTB-14), SKOV3 (Cat No: HTB-77), T24 (Cat No: HTB-4), A549 (Cat No: CRM-CCL-185), HeLa (Cat No: CCL-2), CaCo-2 (Cat No: HTB-37), MIAPaCa (Cat No: CRL-1420) were purchased from ATCC and maintained in DMEM, (Gibco) with 10% Fetal Bovine Serum (FBS, Gibco) and 1% penicillin/streptomycin (Gibco). They were cultivated at 37 °C with 5% CO2 in the air. Subculturing was done as needed based on cell confluency, and the medium was replaced every three days to encourage optimal growth. OECM was purchased from MERK and maintained in RPMI, (Gibco) with 10% FBS and 1% penicillin/streptomycin. Every three days, the medium was switched to encourage optimal growth, and subculturing was done as necessary based on cell confluency. Human mammary epithelial cells (HMEpC, Cat No: #7610), human pulmonary fibroblasts (HPF, Cat No: # 3300), human bronchial epithelial cells (HBEpC, Cat No: #3210), human pulmonary microvascular endothelial cell (HPMEC, Cat No: #3000), human pulmonary artery endothelial cell (HPAEC, Cat No: #3100), human umbilical vein endothelial cells (HUVEC, Cat No: #8000), human renal cortical epithelial cell (HRCEpC, Cat No: #4110), human hepatocytes (HH, Car No: #5200), human astrocytes (HA, Cat No: #1800), human corneal epithelial cells (HCEpC, Cat No: #6510), human dermal fibroblasts (HDF, Cat No: #2300), human thymic epithelial cells (HTyE, Cat No: #3910) were purchased from ScienCell (Carlsbard, CA, USA). Human bone marrow-derived hematopoietic stem cells (HHHSC, SER-CD34-1F) were purchased from Zen-Bio. Cardiomyocytes (Cat No: SCC109) was purchased from MERK. All cells were cultured in their respective recommended media and cultured at 37 °C with 5% CO2 in air. Construction and expression of CD63-anti-HLA-G VHH fusion proteins A synthetic gene encoding the anti-HLA-G VHH (α-HLA-G Nb) was identified via screening the HuSdL® Human Single Domain Antibody Library from Creative Biolabs. This gene was then constructed for insertion into the human CD63 gene and subsequently cloned into the pcDNA3.4 vector (Thermo, Waltham, MA, USA), using KpnI and EcoRI as restriction sites (Thermo). This cloning facilitated the expression of α-HLA-G Nb chimeric CD63 fusion proteins on EVs. Endo-Free Plasmid Maxiprep Kits (Geneaid, Taipei, Taiwan) were used to prepare plasmids for transfection, ensuring endotoxin removal to maintain cell health and transfection efficiency. HEK293T cells were transfected with the engineered plasmids using Lipofectamine 3000 (Thermo), following the manufacturer’s protocol to optimize outcomes. To establish a stable clone expressing the fusion proteins, the transfected HEK293T cells were treated with G418 antibiotics (Thermo) for two weeks. The successfully selected cells (HEK293T-α-HLA-G cells) were expanded and cryopreserved for long-term storage and further analysis. This methodological approach ensured the efficient production and isolation of EVs bearing the α-HLA-G VHH chimeric CD63 fusion proteins for subsequent studies. Flow cytometry (Accuri 6 Plus, BD Biosciences, Franklin Lakes, NJ, USA) and western blot were used to characterize the expression profiles of CD63 and α-HLA-G Nb on the cell surface, with unmodified EVs as a control. Data were analyzed using FlowJo software to generate detailed images. Mycoplasma detection Using the MycoSEQ™ Plus Mycoplasma Detection Kit (Thermo Fisher Scientific), mycoplasma contamination in cell cultures was evaluated in accordance with the manufacturer’s instructions. Using the provided sample preparation reagents, 1 mL of culture supernatant was extracted from each cell line under sterile conditions and then subjected to nucleic acid extraction. Primers, TaqMan® probes specific for more than 90 mycoplasma species, and the supplied lyophilized master mix were used to quantitatively real-time amplify the purified DNA. The cycle threshold (Ct) values of test samples were compared to the internal positive control and negative template control of the kit in order to ascertain if mycoplasma DNA was present or not. According to the kit’s validation criteria, samples were deemed mycoplasma-positive if the Ct value was ≤38. It was verified that none of the samples used in the subsequent tests were contaminated with Mycoplasma. EVs isolation, concentration, and purification HEK293T-α-HLA-G cells (2 × 108) were seeded and evenly distributed onto the carrier surface of the CelCradle® benchtop bioreactor (ESCO Aster, Singapore) with 500 mL DMEM culture medium (Gibco, Grand Island, NY, USA) supplemented with exosome-depleted fetal bovine serum (Gibco) and 1% of penicillin/streptomycin/amphotericin B mixture (Gibco). Three to four days post-supplementation, the medium was collected, centrifuged at 2000 x g for 15 min to remove cell debris, and filtered through a 0.22 μm filter. The collected supernatant was subjected to tangential flow filtration (MAP.03-plus TFF System, Lefo Science, Taipei, Taiwan) using 300 kDa cut-off membranes. The final EVs-enriched sample was collected from the inner tubing and stored at 4 °C. Supernatants from α-HLA-G Nb-chimeric CD63 fusion protein-expressing HEK293T stable clone were purified using a VHH-capturing resin (GenScript, Piscataway, NJ, USA). The α-HLA-G EV were then eluted using chromatography systems (ÄKTA™, Cytiva, Marlborough, MA, USA), concentrated through a 30 kDa cut-off membrane, and then resuspended in phosphate-buffered saline (Gibco). Surface plasmon resonance (SPR) analysis for the binding affinity of α-HLA-G EV CM5 chips were used for SPR analysis by BIAcore T200 (GE Healthcare, Piscataway, NJ, USA). α-HLA-G EV (3 × 1010) were suspended in PBS for immobilization on the chips with maximum surface retention, followed by the SURFACE PREPARATION process and selecting the condition with a higher surface concentration of recombinant HLA-G protein (125, 62.50, 31.25, 15.63, 7.81, and 3.91 nM) on the chip. The REGENERATION METHOD mode was then selected for the regeneration scouting and surface performance test, followed by BINDING ANALYSIS and DIRECT BINDING assessments to determine protein binding. Finally, KINETIC ANALYSIS and MASS TRANSFER were selected for the kinetic assay concomitantly with the binding experiments. The data were analyzed using kinetic constants to determine the KD values. TEM preparation and staining For transmission electron microscopy (TEM, JEM-1400; JEOL Ltd., Tokyo, Japan) analysis, a phosphotungstic acid (Alfa Aesar, Haverhill, MA, USA) solution was prepared by dissolving the acid in deionized water to achieve a concentration of 5%. This solution was then filtered through a 0.22 µm filter to ensure purity. Samples were prepared by applying 5.5 µL of the specimen onto a carbon-coated copper grid and allowing it to settle for 10–15 min. Subsequently, the grid was fixed with 4% formaldehyde for 10 min, and any residual formaldehyde. The grid was then immersed horizontally in the prepared dye for 30 s and rinsed thrice with ddH2O to remove excess dye. After rinsing, the grid was placed on filter paper to dry overnight at 37 °C before proceeding with TEM imaging. Nanoparticle tracking analysis NTA measurements were conducted using the ZetaView instrument (Particle Metrix, Microtrac, Meerbusch, Germany) to assess the size distribution, concentration, and zeta potential of EVs and drug-loaded EVs. Calibration was performed using 100 nm polystyrene beads according to the NTA manufacturer’s guidelines. Samples were filtered through a 0.22 µm membrane filter and diluted in PBS to achieve a final volume of 1 mL, ensuring a 50–200 particle count per field of view to obtain accurate size distribution curves and concentration data. The zeta potential measurements were performed at two positions for each sample to ensure consistency. EV marker analysis To analyze the expression profiles of α-HLA-G Nb on the EV surface, EVs were counted using a CytoFLEX nano (Beckman Colter, Brea, CA, USA), with unmodified EVs as a background control. The purity of α-HLA-G Nb-expressing EVs was stained with iFluor 647-conjugated anti-VHH antibody (clone 96A3F5, Cat No: A01861, GenScript, Piscataway, NJ, USA), using unmodified EVs as a background control. To analyze the proteins extracted from EVs, EV samples were lysed and followed by centrifugation at 12,000 × g for 15 min at 4 °C to clarify the lysate. The supernatant was mixed with protein sample buffer and denatured in a 95 °C dry bath for 10 min. Each sample, containing 20 μg of protein, was subjected to electrophoresis on 10% SDS-PAGE gel and transferred onto a PVDF membrane, followed by blocking with 5% bovine serum albumin for 1 h at room temperature and incubated with the primary antibodies including anti-CD63 (clone ARC51703, Cat No: A19023, Abclonal, Woburn, MA, USA), anti-CD81 (Cat No: A5270, Abclonal), anti-CD9 (clone ARC0330, Cat No: A19027, Abclonal), anti-Alix (Cat No: A2215, Abclonal), anti-VHH (clone 96A3F5, Cat No: A01861, Abclonal), and anti-β-actin (clone ARC5115-01, Cat No: AC026, Abclonal) overnight at 4 °C. Subsequently, after hybridization with a secondary antibody, bolts were visualized using an ECL kit and quantified using ImageJ software (v1.53, National Institutes of Health, Baltimore, MD, USA). Visualization of EV uptake α-HLA-G EV were labeled using ExoGlow™-Protein EV Labeling Kit (System Biosciences, Palo Alto, CA, USA) according to the manufacturer’s instructions. Briefly, EVs were incubated with ExoGlow™, which specifically binds to EVs proteins, for 30 min at 37 °C65. The free dye was removed through 3 K concentrator columns at 12,000 g (Vivaspin 20, Cytiva, Marlborough, MA, USA), and the labeled EVs were then resuspended in DMEM. For imaging, 104 cells were seeded in µ-slide (ibidi GmBH, Martinsried, Germany) and incubated with fluorescently labeled EVs for the indicated periods. Afterward, cells were fixed with 4% paraformaldehyde, the nuclei were stained with DAPI, and the cytoskeleton was stained with Alexa Fluor 594 phalloidin (Invitrogen, Grand Island, NY, USA). Imaging was performed using a Leica TCS SP8 confocal microscope (Leica Microsystems, Wetzlar, Germany). Flow cytometry analysis For flow cytometric analysis, ExoGlow™ staining was performed as described above for imaging analysis. After staining, cancer cells and corresponding non-tumorigenic cells (105 cells) were co-cultured with the fluorescently labeled α-HLA-G EV for 12 h. Cells were trypsinized at various time points, washed with PBS, and resuspended in 200 μL of PBS for flow cytometry (Accuri 6 Plus, BD Biosciences, Franklin Lakes, NJ, USA) analysis to quantify EV uptake. Soluble HLA-G Quantification and EV Uptake Assay Cell uptake assay was used to accurately address the role of soluble HLA-G (sHLA-G). First, α-HLA-G EV (2 × 108) was incubated with sHLA-G (5 ng) for 3 h on ice before co-cultured with MDA-MB231. After being cultured for 24 h, the cells were collected and quantified EV uptake by flow cytometry (Accuri 6 Plus, BD Biosciences, Franklin Lakes, NJ, USA). HLA-G knockdown and knockout model establishment CRISPR-Cas9-mediated HLA-G knockout was performed using the PX459 (SpCas9-Puro) vector, which was designed to express a single guide RNA (sgRNA) targeting the HLA-G gene. MB231 cells were transfected with the CRISPR-Cas9 plasmid using Lipofectamine 3000 (Thermo Fisher Scientific), following the manufacturer’s instructions, with 2.5 µg of plasmid DNA per transfection. After 24 hours, the transfection medium was replaced with fresh complete medium to support cell recovery. To select successfully transfected cells, 1 µg/mL puromycin was applied for 7 days. Total RNA was then extracted using the PureLink™ RNA Mini Kit (Thermo Fisher Scientific) according to the manufacturer’s protocol. Reverse transcription was performed with SuperScript™ IV VILO™ Master Mix, using 500 ng of RNA per reaction. Quantitative PCR (qPCR) was conducted with TaqMan™ Fast Advanced Master Mix (Thermo Fisher Scientific) on an Azure Cielo™ Real-time PCR system, with HLA-G as the target gene and GAPDH as the internal control. Gene expression levels were analyzed using the ΔΔCt method. Pharmacological induction of HLA-G expression To evaluate the effect of pharmacological stimulation on HLA-G expression and subsequent EV uptake, MDA-MB-231 cells were treated with recombinant human interferon-γ (IFN-γ, 1000 IU/mL, ProSpec, Rehovot, Israel) or with 5-azacytidine (5-AZA, 5 μM, Sigma-Aldrich, St. Louis, MO, USA) for 24 hours. Following treatment, HLA-G expression was analyzed by flow cytometry. To determine EV uptake efficiency, ExoGlow™-labeled α-HLA-G EVs were incubated with pre-treated or untreated MDA-MB-231 cells for 12 h. Flow cytometry was used to collect and analyze cells to quantify internalization. Co-culture analysis for cellular interaction Cancer cells and corresponding non-tumorigenic cells were incubated with CellMask™ Red dye (Invitrogen) and CellMask™ Deep Red dye (Invitrogen) at a density of 5 × 104 cells at 37 °C for 30 min. Following incubation, the cells were washed three times with PBS to remove excess dye. The differentially stained cancer cells and their corresponding normal cells were mixed equally and co-cultured in a 3.5 cm glass dish. The next day, ExoGlow™-stained EVs were introduced into the co-culture. Imaging was performed at various intervals using a confocal microscope to observe the interaction and uptake of EVs by cancer and normal cells in the co-culture system. For flow cytometry analysis, cancer cells (MDA-MB-231, U87, or SKOV3 cells) were incubated with CellMask™ Red dye, while their corresponding non-tumorigenic cells (MCF10A, HA, or OF) were incubated with CellMask™ Deep Red dye at 37 °C for 30 min. After thoroughly washing with PBS, an equal amount (5 × 105 cells) of the stained cancer cells and their corresponding non-tumorigenic cells were mixed and co-cultured in a 6 cm dish. On the next day, ExoGlow™-labeled EVs were added to the culture. The samples were collected and analyzed at the designated time points using a BD Accuri 6 Plus. Flow cytometry data were then processed and analyzed using FlowJo software, facilitating detailed image plotting and quantitative analysis of EV–cell interactions within the co-culture. Preparation and analysis of drug-loaded α-HLA-G EV EVs (1 × 10¹¹) were encapsulated with DOX (2 mg), TMZ (5 mg), and CBP (5 mg), respectively, followed by a 1-h incubation at 37 °C. The mixture was then centrifuged with a 3 K concentrator at 12,000 × g and 4 °C for 20 min to remove free drugs. The supernatant collected from the concentrator’s upper layer was further processed by vortexing and centrifugation at 12,000 × g, 4 °C for 30 min. UV-visible spectrophotometry was performed to analyze the samples for drug absorption. The absorbance values of samples were measured across a wavelength range of 300–700 nm to assess the encapsulation efficiency and stability of these α-HLA-G EV-encapsulated chemotherapy drugs. In addition, nanoflow cytometry was used to analyze DOX loading efficiency into α-HLA-G EV. α-HLA-G EV were labeled with the ExoGlow™-Protein EV described above for imaging analysis. The release of chemotherapy drugs from EVs was evaluated in PBS at pH 7.4 and 6.4 to simulate physiological and tumor microenvironment conditions, respectively. In addition, we also evaluated the release status in a hypoxic environment as a basis for simulating the internal environment of tumors. Samples were incubated in a 96-well plate at 37 °C, then centrifuged at 12,000 × g and 4 °C for 30 min at predetermined intervals. The supernatant was collected and analyzed using a UV-visible spectrophotometer to quantify drug release profile. In addition, Assessment of cytotoxic effects To evaluate the cytotoxic effects of various treatments, 5000 cancer cells (MDA-MB-231, U87, and SKOV3) per well were plated in a 96-well plate and allowed to adhere for 1 day. Treatments were administered as follows: EVs, α-HLA-G EV, free drug, drug@EV, and drug@α-HLA-G EV, each added to designated wells. Cytotoxicity was measured at different times after treatment with PrestoBlue HS (Invitrogen) reagent to quantify viable cell numbers. Cells without treatment served as the control group and provided a baseline for comparison. The ReadyProbes™ Viability Kit (Invitrogen) was used to differentiate between live and dead cells, with each population stained distinctively. This approach facilitated qualitative and quantitative cell viability analyses by imaging with the M7000 system (Invitrogen). For apoptosis analysis, cells were seeded at 105 cells/well in a 6-well plate and cultured for 1 day. Following this, cells were treated with EV, α-HLA-G EV, free-drug, drug@α-HLA-G EV, and other formulated drugs, respectively. After 24 h of treatment, cells were harvested and stained with an Annexin V-FITC/PI Apoptosis Kit (Elabscience). The stained samples were then analyzed using a flow cytometer to determine the rate of apoptosis induced by each treatment. Finally, western blot was also used to analyze apoptosis-related proteins with specific antibodies, including PARP1 (clone ARC0091, Cat No: A19612, Abclonal), Bcl-2 (Cat No: A0208, Abclonal), Bax (clone E63, Cat No: ab32503, Abcam, Cambridge, MA, USA), Caspase-3 (clone ARC0133, Cat No: A19654, Abclonal), and β-actin. Reactive oxygen species (ROS) analysis MDA-MB-231 or AC16 cells were cultured in 6-well plates (10,000 cells/well) for 1 day and then labeled using ROS assay stain kit (Invitrogen). Subsequently, cells were treated for 24 h with EV, α-HLA-G EV, DOX, Lipo-DOX, DOX@EV, and DOX@α-HLA-G EV, respectively, and then collected for evaluation by flow cytometry. Establishment of the CDX-TNBC mouse model All in vivo experiments were performed using 8-week-old NPG (female, 20–22 g) mice purchased from BioLASCO Taiwan Co., Ltd. The experimental protocols were approved by the Committee for Animal Experiments at China Medical University (CMUIACUC-2023-092). Mice were housed in a specific pathogen-free facility per committee guidelines. To establish an orthotopic breast cancer model, 106 MDA-MB-231-luc cells in 100 µL of PBS were subcutaneously injected into the left fourth mammary gland of mice, mimicking human disease conditions. After two weeks of tumor establishment, the mice received weekly intravenous tail vein injections of different EVs encapsulated with or without DOX (4 mg/kg of DOX/100 µL PBS for each mouse) for six weeks. Tumor progression was monitored using an IVIS Spectrum in vivo imaging system (PerkinElmer, Waltham, MA, USA), which provides bioluminescence imaging to visualize and quantify tumor growth. Survival rates were assessed based on mortality or when tumor volume exceeded 1000 mm3, requiring euthanasia per ethical guidelines. For each treatment group, six mice were used. Establishment of the CDX-GBM mouse model A total of 20 6-week-old female NPG mice to establish an orthotopic GBM model, 106 U87-luc cells in 10 µL of PBS were injected into the right brain of each mouse. After two weeks of tumor establishment, mice received weekly intravenous tail vein injections of different EVs encapsulated with or without TMZ (5 mg/kg of TMZ in 100 µL PBS) for six weeks. Tumor progression was monitored using an IVIS Spectrum in vivo imaging system, which provides bioluminescence imaging to visualize and quantify tumor growth. Survival rates were assessed based on mortality or when tumor volume exceeded 1000 mm3, requiring euthanasia per ethical guidelines. Biodistribution study The orthotopic CDX-TNBC or CDX-GBM mice model was established for biodistribution analysis of therapeutic agents. Two weeks post-inoculation of cancer-luc cells, 1010 EVs were labeled with XenoLight DiR (PerkinElmer) in 100 µL of PBS and administered intravenously to the mice through tail vein injection. The IVIS imaging system, equipped with fluorescence filters set to 745 nm excitation and 800 nm emission, was used at predetermined intervals to track the in vivo distribution of injected EVs. Simultaneously, the bioluminescence channel facilitated continuous monitoring of tumor volume, allowing for a comprehensive evaluation of treatment efficacy and EV biodistribution relative to tumor location and size. After 1 day, the animals were euthanized via CO2 asphyxiation for subsequent analyses. Immunohistochemical analysis Following the treatment protocols, mice were euthanized, and tumors and organs were excised and fixed in 4% paraformaldehyde overnight. The tissues were dehydrated with graded alcohol and embedded in paraffin. Thin sections measuring 3 μm were prepared for immunohistochemical analysis. Sections were treated with 3% hydrogen peroxide for 15 minutes to quench endogenous peroxidase, followed by blocking to reduce non-specific binding. After the sections incubated with the primary antibodies Ki67 (Cat No: ab15580, Abcam), Bcl-2, and Bax at 37 °C for 2 h, the sections were exposed to biotinylated secondary antibodies and amplified using a polymer system at room temperature. Color development was achieved with a DAB substrate, and counterstaining was performed with H&E for detailed nuclear and cytoplasmic visualization. Apoptotic cells within the tissue sections were detected using the TUNEL assay with the In Situ Cell Death Detection Kit POD (Roche, Basel, Switzerland). Tumor sections were incubated with the TUNEL reaction mixture at 37 °C, followed by the Converter-POD for 30 min. After PBS washing, the sections were developed with DAB, counterstained with hematoxylin, and then mounted with neutral resin. Additionally, the heart, liver, spleen, lung, kidney, and tumor underwent comprehensive histological examination using H&E staining. Clinical outcome evaluation TNBC tumor and GBM specimens were obtained from China Medical University Hospital, with protocols approval by the ethics committees of each institute (CMUH110-REC1-025). Fresh tissues were cut into 5 mm³ pieces and digested in dispase (2 U/mL) and collagenase (1 mg/mL) at 37 °C on an orbital shaker. The resulting cell suspension was filtered through a glass pipette and a 40 µm sieve to create a single-cell suspension and resuspended in a 6 cm dish containing DMEM. After several days of culture, a proportion of the cells were identified HLA-G-positive cells and analyzed EV uptake. Then, the cytotoxicity with various drug combinations was analyzed. Statistics and reproducibility All quantitative data are presented as mean ± standard deviation (SD). Statistical analyses were performed using one-way analysis of variance (ANOVA) followed by Tukey’s post-hoc test for multiple comparisons. A p value of <0.05 was considered statistically significant. No statistical method was used to predetermine sample size. No data were excluded from the analyses. The experiments were not randomized. The Investigators were not blinded to allocation during experiments and outcome assessment. Ethics statement All animal experiments were conducted in compliance with institutional guidelines and approved by the Institutional Animal Care and Use Committee (IACUC) of China Medical University, Taiwan (Approval No. CMUIACUC-2023-092). The maximum allowable tumor burden was set at 1000 mm³ in accordance with ethical regulations, and this threshold was not exceeded in any experimental group. Human patient-derived tumor tissues were collected in accordance with the protocols approved by the Research Ethics Committee of China Medical University and Hospital (CMUH), Taiwan (CMUH110-REC1-025). In total, tumor specimens were obtained from four patients, including two breast cancer patients (all female, average age 55 years) and two glioblastoma patients (one male, one female; average age 74 years). Written informed consent was obtained from all participants prior to sample collection. No personally identifiable information was disclosed. Reporting summary Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article. Supplementary information Supplementary Information Reporting Summary Transparent Peer Review file Source data Source Data Acknowledgements The authors acknowledge receipt of grants from the National Science and Technology Council of Taiwan (NSTC 112-2628-E-039-001-MY3, 1114-2314-B-039-003, 113-2314-B-039-005, 114-2811-E-039-001, and 114-2327-B-039-001) and China Medical University (CMU113-MF-27). Experiments and data analyses were performed in part by the Medical Research Core Facilities, Office of Research & Development, China Medical University, Taichung, Taiwan. Author contributions M.Y.S. designed and carried out the experiments, acquired and analyzed the data, and wrote the manuscript. S.W.H., Y.C., M.C.C., C.M.P., C.Y.C., Y.H.L., and M.H.Y. carried out the experiments. K.W.K., S.C.C., and H.C.H. interpreted the data. Y.W.C. and D.Y.C. supervised the entire project. Peer review Peer review information Nature Communications thanks Jurgen Kuball, Thierry Vandamme, and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. A peer review file is available. Data availability The authors declare that the data supporting the findings of this study are available within the article, source data, and its Supplementary Information. Source data are provided with this paper. All images and graphical elements presented in Figures, and Supplementary Figs. were entirely created by the authors using original content. No previously published or third-party graphical elements were used. Accordingly, no permission or license is required for the publication of these figures. Source data are provided with this paper. Competing interests Dr. Hui-Chun Ho is employed by Shine-On BioMedical Co.,Ltd., Taichung, Taiwan. All other authors confirm they have no conflicts of interest. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. These authors contributed equally: Ming-You Shie, Shi-Wei Huang, Yeh Chen. Contributor Information Ming-You Shie, Email: eric@mail.cmu.edu.tw. Yi-Wen Chen, Email: evinchen@mail.cmu.edu.tw. Der-Yang Cho, Email: dycho1212@gmail.com. Supplementary information The online version contains supplementary material available at 10.1038/s41467-025-66451-y. References Associated Data Supplementary Materials Supplementary Information Reporting Summary Transparent Peer Review file Source Data Data Availability Statement The authors declare that the data supporting the findings of this study are available within the article, source data, and its Supplementary Information. Source data are provided with this paper. All images and graphical elements presented in Figures, and Supplementary Figs. were entirely created by the authors using original content. No previously published or third-party graphical elements were used. Accordingly, no permission or license is required for the publication of these figures. Source data are provided with this paper.