Metastasis enables immunogenicity through migrasome-mediated antigen release

Abstract

Antigen release is a critical step in initiating antitumor immune responses, yet its regulation during metastasis is not well understood. Here we show that circulating tumor cells undergoing vascular migration produce migrasomes that serve as a metastasis-specific mechanism of antigen release. These migrasomes are enriched in tumor-associated antigens, including cancer-testis and mutated antigens, and are captured efficiently by antigen-presenting cells in secondary lymphoid organs, where they undergo cross-presentation to elicit CD8⁺ T cell-mediated immune responses that constrain metastatic progression. Genetic inhibition of migrasome formation enhances metastasis, while administration of purified cancer-derived migrasomes restores immune-mediated suppression of metastatic growth. These findings show paradoxically that metastasis can enhance tumor immunogenicity through migrasome-mediated antigen release, highlighting a link between cancer dissemination and immune activation and establishing migrasomes as a distinct and potent platform for endogenous tumor antigen delivery.

Here the authors show that circulating tumor cells shed migrasomes that carry tumor antigens, which in turn can activate CD8⁺ T cells through cross-presentation to restrict metastasis.

Main

Metastasis is the leading cause of cancer mortality. It marks a pivotal stage of tumor progression where malignant cells disseminate to distant sites and often escape immune elimination¹. Effective antitumor immunity hinges on the release and presentation of tumor antigens to the immune system. Such antigens are typically liberated when tumor cells die (for example, through immunogenic cell death)^2–4 or shed their contents in extracellular vesicles^5,6, allowing dendritic cells to capture these antigens and prime cytotoxic T lymphocytes against the cancer^6–11.

Tumor-derived extracellular vesicles have emerged as critical mediators of tumor–immune interactions, with a dualistic influence on immunity. On one hand, extracellular vesicles carry immunosuppressive cargo facilitating immune evasion—for example, tumor exosomes displaying programmed cell death ligand 1 (PD-L1) can inhibit cytotoxic T cells¹². On the other hand, extracellular vesicles can disseminate tumor antigens and proinflammatory signals: exosomes bearing major histocompatibility complex (MHC) class I-restricted tumor antigens can activate T cells and provoke antitumor immune responses^5,13. Thus, context-dependently, tumor-derived extracellular vesicles may serve either as agents of immune escape or as triggers of antitumor immunity^12,14–17.

Migrasomes are a recently identified organelle produced by migrating cells. As cells move, they leave behind long retraction fibers from which migrasomes bud off, carrying diverse cytosolic contents^18–20. Migrasomes have been implicated in modes of intercellular communication during development and disease²¹. In cancer, migrasomes have been observed accompanying metastasizing tumor cells and ferrying immune checkpoint proteins such as PD-L1, suggesting a role in facilitating immune evasion. However, it is unknown whether metastatic dissemination might also leverage migrasomes to release tumor antigens—a possibility that would represent a previously unrecognized mechanism of antitumor immune activation.

Here, we demonstrate that metastasis can paradoxically unmask tumor immunogenicity through migrasome-mediated antigen release. In mouse models, circulating tumor cells (CTCs) migrating through the vasculature shed abundant migrasomes enriched with tumor antigens. These migrasomes traffic to secondary lymphoid organs and are preferentially phagocytosed by antigen-presenting cells (APCs), which then cross-present migrasome-derived antigens to elicit robust, CD8⁺ T cell-dependent antitumor immune responses that constrain metastatic outgrowth. Our findings uncover a metastasis-dependent mechanism of antigen release that activates antitumor immunity. Thus, rather than serving only as a means of immune escape, the metastatic process can inadvertently engage immune surveillance and reveal tumor vulnerability.

Results

Tspan4 suppresses breast cancer metastasis

The 4T1 mouse mammary carcinoma cell line is highly invasive and can metastasize to multiple organs, including lymph nodes, blood, liver, lung, brain and bone²². The 4T1 line is widely regarded as a relevant model for human triple-negative breast cancer²³, as it lacks estrogen receptor, progesterone receptor and HER2 expression and exhibits aggressive metastasis. Tetraspanins play roles in regulating membrane signaling, adhesion and immune receptor dynamics, potentially impacting cancer progression and therapeutic responses²⁴. To investigate the role of Tspan4 in metastasis, we generated and intravenously (i.v.) injected wild-type (WT), Tspan4-knockout (Tspan4^−/−) and Tspan4-overexpressing (Tspan4^OE) 4T1 cells stably expressing luciferase into syngeneic female BALB/c mice. In vivo imaging was performed on day 14 following D-Luciferin injection using the IVIS system. Metastatic burden was significantly increased in mice injected with Tspan4^−/− 4T1 cells and markedly reduced with Tspan4^OE cells compared to WT (Fig. 1a,b). Tspan4 deficiency shortened, and overexpression extended, survival of tumor-bearing mice (Fig. 1c). Notably, Tspan4 knockout did not enhance primary tumor growth in subcutaneous models using immunocompetent BALB/c or immunodeficient BALB/c-nude mice (Extended Data Fig. 1a,b), indicating that Tspan4 selectively suppresses metastatic dissemination without affecting primary tumor growth.

Fig. 1. Tspan4 regulates the metastasis of 4T1 cells and the production of migrasomes during this process.

a, Cells (4T1) were injected i.v., and in vivo photo emission of 4T1 cells was detected by d-Luciferin injection and imaged using an IVIS spectrum imaging system on day 14. b, Statistics of the luminescence counts of 4T1 cells. n = 18 mice for each group. c, Survival plots for mice injected with WT, Tspan4^−/− and Tspan4^OE 4T1 cells. Tspan4^OE, n = 13 mice; WT, n = 12 mice; Tspan4^−/−, n = 9 mice. d, H&E-stained lung images of MMTV-PyMT;Tspan4^+/+and MMTV-PyMT;Tspan4^−/− mice. Scale bar, 2 mm. e, Quantification of the number of metastasis nodules in lung. T4^+/+, n = 21 mice, T4^−/−, n = 22 mice. f, Quantification of the incidence of lung metastasis. g, Confocal imaging of Tspan4-GFP expressing 4T1 cells after seeding on fibronectin (10 μg ml⁻¹) and laminin (1 μg ml⁻¹) for 24 h in vitro. Scale bar, 20 μm. h, Diagram of intravital imaging. i, Tspan4-GFP stably expressing 4T1 cells were injected into the mice for liver intravital imaging. Vessels were labeled with AF647-conjugated wheat germ agglutinin (AF647-WGA). Scale bar, 10 μm. Arrowheads indicate migrasomes. j, Confocal imaging of WT, Tspan4^−/− and Tspan4^OE 4T1 cells seeded in fibronectin (10 μg ml⁻¹) and laminin (1 μg ml⁻¹)-coated dishes and stained with WGA (1 μg ml⁻¹). Scale bar, 10 μm. k, Quantification of the number of migrasomes in WT, Tspan4^−/− and Tspan4^OE 4T1 cells. Tspan4^OE, n = 73 cells; WT, n = 80 cells; Tspan4^−/−, n = 56 cells. l, WT, Tspan4^−/− and Tspan4^OE 4T1 cells were transfected with PH-mCherry and injected into mice for intravital imaging. m,n, Quantification of the number of migrasomes produced by WT, Tspan4^−/− and Tspan4^OE 4T1 cells in vivo. Tspan4^OE, n = 20 cells; WT, n = 24 cells; Tspan4^−/−, n = 26 cells. Statistical data in b, e, k, m and n are presented as the mean ± s.e.m. P values were calculated using a two-tailed, unpaired t-test.

Source data

Extended Data Fig. 1. Migrasome characterization, comparison with other EV types, and orthogonal validation of migrasome-dependent metastasis suppression.

a, A total of 5 × 10⁵ WT and Tspan4^−/− 4T1 cells were subcutaneously injected (s.c.) into BALB/c mice, and solid tumor size was measured on the indicated days following the 4T1 cell injection. n = 10 mice for each group. b, A total of 5 × 10⁵ WT and Tspan4^−/− 4T1 cells were subcutaneously injected (s.c.) into BALB/c-nude mice, and solid tumor size was measured on the indicated days following the 4T1 cell injection. WT, n = 6 mice; Tspan4^−/−, n = 5 mice. c, Western blot analysis of 4T1 cells and purified 4T1 migrasomes using indicated antibodies. d, Representative bioluminescence images of mice immunized with equal protein amounts of migrasomes, sEVs, tumor lysates, or apoptotic bodies from Tspan4-GFP 4T1 cells, followed by intravenous challenge with WT 4T1 cells. e, Quantification of luminescence counts from each group. n = 8 mice for PBS, lysate and apoptotic body groups, n = 9 mice for migrasome group, n = 7 mice for sEV group. f, Survival curves of tumor-bearing mice from each group. n = 8 mice for apoptotic body group, n = 9 mice for other groups. g, Confocal imaging of 4T1 cells treated with Ctrl (DMSO) or UNC3230 and stained with WGA. Scale bar, 20 μm. h, Quantification of migrasome numbers per cell from Ctrl and UNC3230 groups. n = 119 cells for ctrl, n = 78 cells for UNC3230. i, Representative bioluminescence images of mice injected with DMSO-treated or UNC3230−pretreated 4T1 cells. j, Quantification of luminescence counts from each group. n = 13 mice per group. Statistical data are presented as mean ± s.e.m, P values were calculated using the two-tailed, unpaired t-test. Survival statistics were calculated using the log-rank test.

Source data

To further validate this, we examined Tspan4 in the MMTV-PyMT transgenic model, which recapitulates key features of human breast cancer, including spontaneous lung metastasis^25,26. Tspan4^−/− mice were crossed with MMTV-PyMT mice to generate MMTV-PyMT;Tspan4^+/+ and MMTV-PyMT;Tspan4^−/− littermates. At 16 weeks, we observed a higher incidence of lung metastasis in the Tspan4-deficient cohort (Fig. 1d–f). Collectively, these results indicate that Tspan4 functions as a negative regulator of breast cancer metastasis, without altering primary tumor growth.

Tspan4 regulates migrasome formation in 4T1 cells

We previously showed Tspan4 is essential for migrasome formation in vitro and in chicken and zebrafish embryos^19,27,28. To determine whether migrasomes are formed by metastatic 4T1 cells, we examined their presence both in vitro and in vivo. We confirmed that 4T1 cells produce migrasomes in vitro (Fig. 1g). To investigate migrasome formation during tumor metastasis in vivo, we performed intravital imaging of livers from mice injected with 4T1 cells expressing Tspan4-GFP (Fig. 1h). Notably, circulating 4T1 cells generated retraction fibers and migrasomes within blood vessels (Fig. 1i and Supplementary Video 1).

To assess the role of Tspan4 in migrasome biogenesis, we examined migrasome formation in Tspan4^−/− 4T1 cells in vitro. Tspan4^−/− cells exhibited a significant reduction in migrasome number, whereas Tspan4^OE cells led to enhanced migrasome formation (Fig. 1j,k). We further examined migrasome formation in vivo by injecting Tspan4^−/− and Tspan4^OE cells and performing intravital imaging. Consistent with in vitro data, migrasome formation was significantly increased in Tspan4^OE cells and markedly decreased in Tspan4^−/− cells within blood vessels (Fig. 1l–n).

Injection of 4T1-derived migrasomes blocks metastasis

Next, we investigated whether the accelerated metastasis in Tspan4^−/− 4T1 cells is caused by impaired migrasome formation or by migrasome-independent functions of Tspan4. We reasoned that if impaired migrasome formation drives the enhanced metastasis, administering WT 4T1 migrasomes should rescue the phenotype. To efficiently obtain sufficient migrasomes, we used Tspan4^OE4T1 cells, which generate more migrasomes than WT cells. Purified migrasomes (Fig. 2a) showed the characteristic morphology—a round vesicle connected to a retraction fiber—by transmission electron microscopy (TEM) (Fig. 2b)¹⁸. Western blot analysis according to MISEV guidelines²⁹ confirmed the enrichment of established migrasome markers (INTEGRINS α5 and β1, NDST, PIGK, EOGT and CPQ) and the absence of exosome markers (CD63, TSG101 and SYNTENIN-1) or the ectosome-associated marker KIF23 (Fig. 2c,d and Extended Data Fig. 1c). Subsequently, we i.v. injected purified migrasomes into BALB/c mice, and 16 h later we injected WT and Tspan4^−/− 4T1 cells. Bioluminescence imaging 14 days later showed that migrasome injection significantly reduced the enhanced metastasis of Tspan4^−/− cells (Fig. 2e,f). As a control, ultraviolet (UV)-induced dead 4T1 cells did not reduce metastasis of Tspan4^−/− cells (Fig. 2e,f). We further compared migrasomes head-to-head with small extracellular vesicles (sEVs), tumor lysates and apoptotic bodies; equal protein amounts were injected before 4T1 challenge. Migrasomes significantly suppressed metastasis and prolonged survival, whereas sEVs, lysates and apoptotic bodies failed to inhibit metastatic growth; lysates even appeared to promote metastasis (Extended Data Fig. 1d–f). Thus, the enhanced metastasis in Tspan4^−/− cells is likely due to reduced migrasome formation, and migrasomes possess distinct immunostimulatory properties superior to other extracellular vesicle populations or tumor-derived materials.

Fig. 2. Injection of 4T1-derived migrasomes inhibits metastasis.

a, Purification of migrasomes (mig) from in vitro-cultured 4T1 cells stably expressing Tspan4-GFP. b, TEM image of purified 4T1 migrasomes. Scale bar, 1 μm. c, Western blot analysis of 4T1 cells and purified 4T1 migrasomes using indicated antibodies. d, Western blot analysis of sEV markers (CD63, TSG101, SYNTENIN-1), ectosome marker (KIF23), migrasome markers (EOGT, PIGK, CPQ), MHC class I and flotillin-1 in migrasome, sEV and ectosome fractions isolated from 4T1 cells. e, Migrasomes (from 4T1 cells), UV-irradiated 4T1 cells (dead cells) or PBS were i.v. injected into mice, and WT or Tspan4^−/− 4T1 cells were i.v. injected, at 16 h after injection of migrasomes. In vivo photo emission of 4T1 cells was detected by d-luciferin injection and imaged with an IVIS spectrum imaging system on day 14. f, Statistics of the luminescence counts of 4T1 cells. PBS + WT, n = 16 mice; PBS + Tspan4^−/−, n = 16 mice; migrasome + Tspan4^−/−, n = 17 mice; Dead-cell + Tspan4^−/−, n = 6 mice. g, Diagram of negative selection of in vivo 4T1-derived migrasomes. h, Confocal imaging of negatively selected 4T1-derived migrasomes. Scale bar, 20 μm. i, Western blot analysis of 4T1 cells, blood cells and IV-migs using the indicated antibodies. j, IV-migs or PBS were i.v. injected into mice, and 4T1 cells were i.v. injected at 16 h after injection of migrasomes. In vivo photo emission of 4T1 cells was detected by d-luciferin injection and photographed with an IVIS spectrum imaging system on day 14. k, Statistics of the luminescence counts of 4T1 cells. n = 10 mice for each group. l, Survival plots for mice from k. m, H&E-stained lung images from mice treated with PBS or 4T1 migrasomes (10 μg each). n, Quantification of the number of metastasis nodules per lung. PBS, n = 18; migrasome, n = 20. Statistical data in f, k and n are presented as the mean ± s.e.m. P values were calculated using a two-tailed, unpaired t-test. s.c., subcutaneous; sup, supernatant; ctrl, control.

Source data

To further confirm this effect is migrasome dependent rather than Tspan4 specific, we used an orthogonal approach targeting a distinct migrasome biogenesis pathway³⁰. Treatment of 4T1 cells with UNC3230, a selective inhibitor of the migrasome-essential kinase PIP5K1C, significantly reduced migrasome formation in vitro (Extended Data Fig. 1g,h). Moreover, mice injected with UNC3230-pretreated 4T1 cells exhibited markedly enhanced metastasis (Extended Data Fig. 1i,j), confirming that migrasome deficiency, whether genetic or pharmacological, promotes metastasis. These data further support that the enhanced metastasis in Tspan4^−/− cells results from reduced migrasome formation.

To investigate whether in vivo-produced migrasomes (IV-migs) have the same function, we isolated migrasomes from the blood of tumor-bearing mice. After i.v. injecting 4T1 cells, we used negative selection to purify migrasomes²⁰ (Fig. 2g). Confocal imaging revealed that the purified fraction are Tspan4-GFP-positive vesicles (Fig. 2h), and western blot confirmed enrichment of migrasome markers (TSPAN4-GFP, INTEGRIN α5, NDST and CPQ) without HISTONE H3 (Fig. 2i). Mice were i.v. injected with IV-migs (3 μg per mouse) and challenged 16 h later with 4T1 cells. IV-migs suppressed metastasis (Fig. 2j,k) and prolonged survival (Fig. 2l), supporting an essential role for migrasomes in limiting metastasis.

To further test the anti-metastatic effect, we used an orthotopic model: 2 × 10⁵ 4T1 cells were injected into the flank, and migrasomes were administered i.v. on days 8 and 15. Pulmonary metastasis was reduced at day 45 (Fig. 2m,n), confirming that 4T1-derived migrasomes can prevent breast cancer metastasis in vivo.

Migrasomes prime the adaptive immune response

Considering the antitumor potential of migrasomes, we hypothesized that migrasomes may inhibit the metastasis of 4T1 cells by activating the adaptive immunity. To test this, we used T cell‑deficient BALB/c‑nude mice. We i.v. injected luciferase-expressing WT, Tspan4^−/− and Tspan4^OE 4T1 cells into syngeneic female BALB/c-nude mice. Twelve days later, bioluminescence imaging revealed that, in contrast to the metastasis observed in WT BALB/c mice, where Tspan4 deficiency significantly enhanced metastasis, knockout of Tspan4 in the immune-deficient nude mice did not enhance metastasis. In fact, metastasis was reduced compared to the WT 4T1 cells. Conversely, Tspan4-GFP overexpression in 4T1 cells led to a slight enhancement in metastasis (Fig. 3a,b). Thus, Tspan4 promotes metastasis in nude mice but inhibits it in WT mice (Fig. 1a,b), indicating Tspan4‑mediated inhibition requires T lymphocytes. To test if migrasomes function in T cell‑deficient conditions, we injected migrasomes into nude mice bearing Tspan4^−/− cells and found no reduction in metastasis (Fig. 3a,b), confirming the anti‑metastatic effect is T cell dependent. To test CD8⁺ T cell dependence, we depleted CD8⁺ T cells (anti‑CD8α) in BALB/c mice before migrasome injection and 4T1 challenge. Migrasomes reduced metastasis in IgG controls but not after CD8⁺ T depletion (Fig. 3c,d), confirming a CD8⁺ T cell‑dependent mechanism.

Fig. 3. Migrasomes induce CD8⁺ T cell priming.

a, PBS or purified migrasomes were i.v. injected into BALB/c-nude mice on day 0. After 16 h, 4T1 cells were i.v. injected, and bioluminescence imaging (IVIS) was performed on day 12. b, Quantification of bioluminescence signals. PBS + Tspan4^OE, n = 11 mice; PBS + WT, n = 11 mice; PBS + Tspan4^−/−, n = 12 mice; Mig + Tspan4^−/−, n = 12 mice. c, Anti-CD8a or IgG (160 μg) was injected intraperitoneally (i.p.) as the time bar indicates. After 24 h, migrasomes or PBS were i.v. injected, and 16 h later WT 4T1 cells were i.v. injected. Bioluminescence imaging was performed on day 14. d, Quantification of bioluminescence. n = 10 mice for each group. e, Schematic diagram of the experimental timeline for bone marrow transplantation and migrasome treatment. Eight‑week‑old C57BL/6J mice were irradiated (10 Gy) and reconstituted with WT or B2m^−/− bone marrow. After 8 weeks, all mice received OT‑I CD8⁺ T cells, followed by migrasomes or vehicle, and LLC‑OVA challenge. Lungs were collected at the endpoint. f, Representative H&E staining of lung tissue sections from each group, revealing metastatic tumor foci. Scale bar, 2.5 mm. g, Quantification of metastatic tumor foci per lung section. n = 12 mice for WT BM + Ctrl, n = 13 mice for WT BM + Mig, n = 11 mice for B2m^−/− BM + Ctrl and B2m^−/− BM + Mig groups. h, Schematic diagram of the experimental timeline. i, Representative H&E-stained lung sections showing metastatic foci. Scale bar, 2.5 mm. j, Quantification of metastatic foci. n = 6 mice per group. k, In vitro killing assay. Mice were i.v. injected with migrasomes or PBS on days 0 and 7. On day 19, splenic CD8⁺T cells were co-cultured with 4T1 cells, stained with annexin V, and imaged. Scale bar, 20 μm. l, Quantification of annexin V-positive cells. n = 15 fields per group. Statistical data are presented as the mean ± s.e.m. P values were calculated using a two-tailed, unpaired t-test.

Source data

We next investigated how migrasomes activate CD8⁺ T cells. Proteomic analysis and western blotting revealed that migrasomes are enriched with MHC class I molecules but lack co-stimulatory molecules such as CD80 and CD86 (Extended Data Fig. 2a–c), suggesting that migrasomes are unlikely to directly prime naive T cells and instead require professional APCs. To identify which APCs capture migrasomes, we incubated purified Tspan4-GFP⁺ migrasomes with mouse splenocytes in vitro. Confocal imaging showed that migrasomes were predominantly engulfed by F4/80⁺ macrophages and F4/80⁻CD11c⁺ dendritic cells, whereas no uptake was observed in CD19⁺ B cells (Extended Data Fig. 2d–f). In vivo, i.v. injected migrasomes trafficked to the spleen, lymph nodes and lungs, where they were similarly taken up by macrophages and dendritic cells (Extended Data Fig. 3a,b).

Extended Data Fig. 2. Migrasomes are enriched in MHC class I and are preferentially phagocytosed by macrophages and dendritic cells.

a, Heatmap plot showing distribution of MHC class I in 4T1 cells and migrasomes. b, Western blot analysis of MHC class I expression in migrasome and cell lysate fractions from 4T1 cells. c, Confocal imaging of migrasomes generated by Tspan4-GFP expressing 4T1 cells stained with anti–MHC class I antibody. Scale bar, 20 μm. 10 μm (enlarged). d, Time-lapse confocal imaging of purified migrasomes (Tspan4-GFP) incubated with mouse splenocytes and stained with APC anti-F4/80. Arrowheads indicate migrasomes engulfed by F4/80⁺ macrophages. Scale bar, 10 μm. e, Confocal imaging of migrasomes incubated with mouse splenocytes and stained with PE anti-CD11c and APC anti-F4/80. Arrowheads indicate migrasomes engulfed by F4/80⁻CD11c⁺ dendritic cells. Scale bar, 10 μm. f, Confocal imaging of migrasomes incubated with mouse splenocytes and stained with PE anti-CD19 and APC anti-F4/80. Arrowheads indicate migrasomes; note absence of uptake by CD19⁺ B cells. Scale bar, 10 μm.

Source data

Extended Data Fig. 3. Biodistribution of migrasomes to secondary lymphoid organs following intravenous injection.

a, Confocal imaging of lung, spleen, inguinal lymph node, and axillary lymph node harvested at 6 h post-intravenous injection of purified migrasomes (Tspan4-GFP). Tissues were stained with APC anti-F4/80 and PE anti-CD11c. Scale bar, 20 μm. b, Confocal imaging of lung, spleen, inguinal lymph node, and axillary lymph node harvested at 24 h post-intravenous injection of purified migrasomes (Tspan4-GFP). Tissues were stained with APC anti-F4/80 and PE anti-CD11c. Scale bar, 20 μm.

Source data

To directly test whether APCs are required for migrasome-mediated T cell activation, we performed in vitro co-culture experiments. Naive CD8⁺ T cells isolated from WT mice failed to upregulate activation markers (CD69, CD25, CD44) when cultured with migrasomes alone. However, the addition of bone marrow-derived macrophages (BMDMs) enabled robust T cell activation (Extended Data Fig. 4a–d). Using the OT-I T cell antigen receptor transgenic system, we further demonstrated that migrasomes derived from OVA-expressing cancer cells activated OT-I CD8⁺ T cells only in the presence of bone marrow-derived dendritic cells (BMDCs), BMDMs or total splenocytes (Extended Data Fig. 4e–g). To confirm the requirement for APC-mediated cross-presentation in vivo, we performed bone marrow transplantation experiments using B2m^−/− mice, whose APCs lack surface MHC class I. Because B2m^−/− mice are on a C57BL/6J background incompatible with BALB/c-derived 4T1 cells, we used the C57BL/6J-derived Lewis lung carcinoma (LLC) model and confirmed that LLC-derived migrasomes similarly suppress metastasis in a T cell-dependent manner. In the LLC lung carcinoma model, migrasome injection significantly suppressed lung metastasis in recipients of WT bone marrow but failed to do so in recipients of B2m^−/− bone marrow (Fig. 3e–g and Extended Data Fig. 5a). To further define the relative contributions of dendritic cells and macrophages in migrasome-mediated tumor suppression, we utilized Batf3^−/− mice, which lack CD8α⁺ and CD103⁺ cross-presenting dendritic cells³¹. In the LLC model, the metastasis-suppressive effect of migrasomes was significantly diminished in Batf3^−/− mice compared to WT controls (Fig. 3h–j and Extended Data Fig. 5b). Moreover, combined Batf3 deficiency and macrophage depletion via anti-CSF1R (α-CSF1R) showed a trend toward further reducing the antitumor efficacy of migrasomes (Fig. 3h–j and Extended Data Fig. 5b), indicating that both dendritic cells and macrophages contribute to migrasome-induced immune protection. Taken together, these data suggest that migrasomes can promote T cell-dependent metastasis inhibition through a mechanism requiring capture and cross-presentation by professional APCs, particularly macrophages and dendritic cells.

Extended Data Fig. 4. Antigen-presenting cells are required for migrasome-mediated CD8⁺ T cell activation.

a, Flow cytometry plots showing CD25 expression on naive CD8⁺ T cells after 5 days of co-culture under indicated conditions: negative control, positive control (anti-CD3/anti-CD28), migrasomes alone, or migrasomes plus BMDMs. b, Flow cytometry plots showing CD69 expression under the same conditions. c, Flow cytometry plots showing CD44 expression. d, Flow cytometry plots showing CD44 and CD62L co-staining. Naive CD8⁺ T cells were purified from mouse spleen and lymph nodes using a mouse CD8⁺ T cell negative isolation kit (STEMCELL Technologies). e, Flow cytometry plots showing CD69 expression on OT-I CD8⁺ T cells after 5 days of co-culture under indicated conditions: negative control, positive control (anti-CD3/anti-CD28), migrasomes alone, migrasomes plus BMDCs, migrasomes plus BMDMs, or migrasomes plus splenocytes. f, Flow cytometry plots showing CD25 expression under the same conditions (5 days). g, Flow cytometry plots showing Granzyme B-FITC expression. OT-I CD8⁺ T cells were purified from OT-I transgenic mice (7 days).

Source data

Extended Data Fig. 5. APC‑dependent suppression of lung metastasis by migrasomes.

a. Representative photographs of whole lungs harvested from each experimental group at the end of the experiment (WT bone marrow + Ctrl; WT bone marrow + Migrasomes; B2m^−/− bone marrow + Ctrl; B2m^−/− bone marrow + Migrasomes), showing visible metastatic nodules. Scale bar, 1 cm. b. Representative photographs of whole lungs harvested from each experimental group at the end of the experiment (WT + Ctrl; WT + Mig; Batf3^−/− + Mig; Batf3^−/− + anti-CSF1R + Mig), showing visible metastatic nodules. Scale bar, 1 cm.

Source data

Cytotoxic CD8⁺ T cells play a major role in antitumor immunity. To test whether migrasome injection induces the generation of cytotoxic CD8⁺ T cells, we performed a cytotoxic CD8⁺ T cell killing assay in vitro. We purified migrasomes from Tspan4-GFP-expressing 4T1 cells and i.v. injected them into female BALB/c mice on days 0 and 7. On day 19, we isolated CD8⁺ T cells from the spleen and lymph nodes of both migrasome-immunized and control mice, and co-cultured them with 4T1 cells in vitro. We then stained the cells with annexin V and conducted live-cell imaging using confocal microscopy for 12 h. Confocal imaging revealed that CD8⁺ T cells isolated from migrasome-immunized mice attacked 4T1 cancer cells and induced the externalization of phosphatidylserine, a hallmark of apoptosis. In contrast, CD8⁺ T cells isolated from control mice had minimal effects on 4T1 cells in the co-culture system (Fig. 3k,l). These data indicate that migrasomes can induce the generation of specific cytotoxic CD8⁺ T cells capable of killing 4T1 cells.

Cancer-testis antigens are presented in migrasomes

The fact that migrasomes can induce an adaptive immune response involving cytotoxic CD8⁺ T cells led us to speculate that migrasomes generated by 4T1 cells may prime the adaptive immune response in vivo by presenting non-self-antigens from cancer cells. To test whether tumor-associated antigens are presented in migrasomes, we purified 4T1-derived migrasomes and cell bodies from approximately 200 15-cm tissue culture dishes and performed quantitative mass spectrometry analysis. The samples were labeled with tandem mass tags (TMTs) of different masses, combined and then analyzed (Fig. 4a). The quality of the purified migrasomes was assessed by the enrichment of migrasome marker proteins, including TSPAN4, TSPAN9, NDST, PIGK and INTEGRINS α5, αV and β1 (refs. ^18,32–34; Fig. 4b). Notably, multiple MHC class I subunits were also enriched in the migrasome fraction (Fig. 2d and Extended Data Fig. 2a). We found that more than 1,000 proteins were enriched in migrasomes (migrasome/cell abundance ratio ≥ 2), and more than 400 proteins were depleted in migrasomes (migrasome/cell abundance ratio < 0.5; Fig. 4a). Migrasomes were highly enriched for two testis-specific proteins, BRDT and MSH4 (Fig. 4c), normally restricted to male germ cells. However, these proteins are frequently re-expressed in cancer and can serve as tumor-specific antigens due to their high immunogenicity and biased expression. To test whether the testis-specific proteins enriched in migrasomes can act as cancer-testis antigens, we synthesized the detected BRDT and MSH4 peptides and conducted in vivo immunization experiments. We immunized syngeneic female BALB/c mice with BRDT or MSH4 peptides mixed with adjuvant, followed by i.v. injection of 4T1 cells to assess metastasis. In vivo bioluminescence imaging revealed a significant reduction in 4T1 cell metastasis in BRDT peptide-immunized mice compared to OVA peptide-immunized mice (Fig. 4d,e). In contrast, immunization with MSH4 peptides did not reduce metastasis. This suggests that migrasome-enriched BRDT could elicit antitumor immunity. To visualize how BRDT is delivered to migrasomes, we generated 4T1 cells stably coexpressing Tspan4-GFP and mCherry-tagged BRDT. Live-cell time-lapse imaging revealed that BRDT signal was detectable within premature migrasomes at the earliest stages of formation and continuously trafficked along retraction fibers into the growing migrasome (Extended Data Fig. 6a), providing direct evidence for active enrichment of this cancer-testis antigen during migrasome biogenesis.

Fig. 4. Tumor-associated antigens are enriched in migrasomes.

a, Volcano plot showing mass spectrometry-based quantification of TMT-labeled proteins. Upper‑right dots represent a migrasome/cell abundance ratio ≥ 2, P < 0.05; upper‑left dots represent a migrasome/cell abundance ratio < 0.5, P < 0.05. n = 3 independent experiments. P values were calculated using a two-tailed, two-sample unequal variance t-test. b, Heat map plot showing the distribution of migrasome marker proteins in 4T1 cells and migrasomes. c, Heat map plot showing testis-specific proteins and a top 10 list of mutated proteins that are enriched in migrasomes. d, BRDT, DDHD1(D318H) and SDC2(A18S) peptides (100 μg) were mixed with Poly I:C (50 μg) and injected (i.v.) into mice on day 0 and day 12. Seven days later, 4T1 cells were injected (i.v.) and in vivo photo emission of 4T1 cells was detected by d-luciferin injection and imaged using an IVIS spectrum imaging system on day 33. e, Statistics of the luminescence counts of 4T1 cells. n = 25 mice for each group. f, Genome sequencing of 4T1 cells to identify single-nucleotide mutations. The reference genome sequence of BALB/c was from the UCSC database. Statistical data in e are presented as the mean ± s.e.m. P values were calculated using a two-tailed, unpaired t-test.

Source data

Extended Data Fig. 6. Live-cell imaging of tumor antigen trafficking during migrasome formation.

a, Time-lapse confocal images of Tspan4-GFP 4T1 cells expressing mCherry-tagged Brdt. Scale bar, 20 μm, 3 μm (enlarged). b, Time-lapse confocal images of Tspan4-GFP 4T1 cells expressing mCherry-tagged Sdc2^A18S. Scale bar, 20 μm, 3 μm (enlarged). c, Time-lapse confocal images of Tspan4-GFP 4T1 cells expressing mCherry-tagged Ddhd1^D318H. Scale bar, 20 μm, 3 μm (enlarged).

Source data

Mutated antigens are localized in migrasomes

In addition to cancer-testis antigens, it is well known that cancer cells accumulate genetic alterations, which give rise to mutated peptides known as neoantigens. These neoantigens are foreign in nature and can activate an adaptive immune response with exquisite specificity. To identify these mutated proteins, we isolated genomic DNA from 4T1 cancer cells and performed genome sequencing to identify all the mutations in 4T1 cells, as previous described^35–37. We found that 745 genes with single-nucleotide mutations and 69 insertion or deletion mutations were detected in the 4T1 cell genome compared to the BALB/c genome (Fig. 4f).

Next, we cross-referenced the list of mutated genes in 4T1 cells with the list of proteins enriched in 4T1-derived migrasomes, resulting in a list of mutated proteins that were enriched in migrasomes (Fig. 4c). To test whether these enriched and mutated proteins could serve as potential tumor-associated antigens in 4T1 cells, we synthesized point-mutated peptides from the top ten proteins and performed in vivo immunization experiments. Female BALB/c mice were immunized twice with the synthesized peptides mixed with an adjuvant, followed by i.v. injection of 4T1 cells to evaluate the antitumor effect of these peptides. In vivo bioluminescence imaging revealed a significant decrease in the metastasis of 4T1 cells in mice immunized with the DDHD1(D318H) and SDC2(A18S) mutated peptides, compared to the controls (Fig. 4d,e and Extended Data Fig. 7a). These data provide evidence that DDHD1(D318H) and SDC2(A18S) have the ability to elicit antitumor immunity. Live-cell imaging of 4T1 cells coexpressing Tspan4-GFP and mCherry-tagged SDC2(A18S) or DDHD1(D318H) showed that SDC2(A18S), which exhibits a membrane-localized pattern, accumulated on the migrasome membrane during formation, whereas DDHD1(D318H) trafficked into the interior of migrasomes, similar to BRDT (Extended Data Fig. 6b,c). These observations suggest that both membrane-associated and soluble mutated antigens are actively transported into migrasomes.

Extended Data Fig. 7. Additional mutated peptide immunization and migrasome production by primary PyMT tumor cells.

a, Peptides (100 μg) were mixed with adjuvant and injected (i.v.) into mice respectively at day 0 and day 12. Seven days later, 4T1 cells were injected (i.v.) and in vivo photo emission of 4T1 cells were detected by D-Luciferin injection and photographed with IVIS spectrum image system at day 33. b, Confocal image of primary PyMT tumor cells which were seeded in fibronectin (10 μg/mL) and laminin (1 μg/mL) coated dish and stained with WGA for imaging. Scale bar, 50 μm.

Source data

In summary, we identified several potential tumor-associated antigens, including the cancer-testis antigen BRDT and the mutated antigens DDHD1(D318H) and SDC2(A18S), which are presented in migrasomes and may serve as cancer-specific targets for antitumor immune responses.

Tumor-associated antigen depletion promotes 4T1 metastasis

We identified DDHD1(D318H), SDC2(A18S) and BRDT as non-self-antigens that can trigger a specific antitumor response against 4T1 cancer cells. To test their roles, we generated Ddhd1-, Sdc2- and Brdt-knockout 4T1 cells and performed i.v. transplantation in female BALB/c mice. We found that knockout of Ddhd1, Sdc2 or Brdt promotes the metastasis of 4T1 cells (Fig. 5a–d). These data suggest that DDHD1, SDC2 and BRDT can potentially serve as non-self-antigens in the antitumor immune response.

Fig. 5. Tumor-associated antigen depletion aggravates 4T1 metastasis.

a, WT, Ddhd1^−/− and Sdc2^−/− 4T1 cells were injected (i.v.), and in vivo photo emission of 4T1 cells was detected by d-luciferin injection and imaged using an IVIS spectrum imaging system on day 14. b, Statistics of the luminescence counts of 4T1 cells. n = 15 mice for each group. c, WT and Brdt^−/− 4T1 cells were injected (i.v.), and in vivo photo emission of 4T1 cells was detected by d-luciferin injection and imaged using an IVIS spectrum imaging system on day 14. d, Statistics of the luminescence counts of 4T1 cells. n = 15 mice for each group. e, WT, Ddhd1^−/−, Sdc2^−/− and Brdt^−/− 4T1 cells were injected (i.v.) into BALB/c-nude mice and in vivo photo emission of 4T1 cells was detected by d-luciferin injection and imaged using an IVIS spectrum imaging system on day 12. f, Statistics of the luminescence counts of 4T1 cells in BALB/c-nude mice. n = 10 mice for each group. g, Migrasomes were purified from Tspan4-GFP stably expressing WT and Brdt^−/− 4T1 cell lines and injected (i.v.) into BALB/c mice. PBS served as a negative control. 4T1 cells were injected (i.v.) on the next day after migrasome injection. In vivo photo emission of 4T1 cells was detected by d-luciferin injection and imaged using an IVIS spectrum imaging system on day 14. h, Statistics of the luminescence counts of 4T1 cells. n = 15 mice for each group. i, Mice were i.v. injected with PBS control or Sdc2-/Ddhd1-deficient migrasomes, followed by i.v. injection with 4T1 cells the next day. In vivo photo emission of 4T1 cells was detected by d-luciferin injection and imaged using an IVIS spectrum imaging system on day 14. j, Statistics of the luminescence counts of 4T1 cells. n = 10 mice for PBS and Sdc2^−/− groups, n = 9 mice for Ddhd1^−/− group. Statistical data in b, d, f, h and j are presented as the mean ± s.e.m. P values were calculated using a two-tailed, unpaired t-test.

Source data

To test whether the enhanced metastasis in Ddhd1-, Sdc2- and Brdt-knockout 4T1 cells depends on the adaptive immune response, we performed a 4T1 transplantation experiment in T lymphocyte-deficient nude mice using these knockout 4T1 cells. In contrast to metastasis in healthy normal mice, where knockout of Ddhd1, Sdc2 and Brdt significantly enhances metastasis, in vivo bioluminescence imaging revealed no significant difference in metastasis between WT and Ddhd1-, Sdc2- and Brdt-knockout cells (Fig. 5e,f). These data imply that the enhanced metastasis in Ddhd1-, Sdc2- and Brdt-knockout cells depends on T lymphocytes, further supporting the idea that DDHD1(D318H), SDC2(A18S) and BRDT could be non-self-antigens.

To test whether DDHD1, SDC2 and BRDT are key antigens in migrasomes, we constructed Tspan4-GFP constitutively expressing cell lines capable of generating more migrasomes based on Ddhd1-, Sdc2- and Brdt-knockout 4T1 cell lines. We then purified migrasomes from these cells, which lack these antigens, and injected them into female BALB/c mice to examine whether these migrasomes could induce a specific immune response to 4T1 cancer cells. We next conducted a 4T1 transplantation and migrasome rescue experiment. The data show that Brdt-deficient migrasomes only slightly reduced the metastasis of 4T1, which was significantly less effective than WT 4T1-derived migrasomes (Fig. 5g,h), indicating that the cancer-testis antigen BRDT represents a major immunogenic component of migrasomes. In contrast, Ddhd1-deficient or Sdc2-deficient migrasomes still significantly reduced metastasis (Fig. 5i,j), suggesting that although these mutated antigens are enriched in migrasomes and can independently elicit T cell responses, their individual contribution to the overall antitumor activity of intact migrasomes may be less pronounced than that of BRDT, likely reflecting differences in antigen immunogenicity, abundance or the efficiency of cross-presentation by APCs. Collectively, these results suggest that tumor-derived migrasomes carry multiple non-self-antigens that collectively contribute to their immunostimulatory capacity, with the cancer-testis antigen BRDT playing a particularly prominent role.

Migrasomes have potential as therapeutic vaccines

To explore the potential significance of migrasomes in cancer immunotherapy, we tested whether the injection of tumor-derived migrasomes can inhibit cancer development in tumor-bearing mice. First, we injected WT 4T1 cells into female BALB/c mice via i.v. injection to generate 4T1-bearing mice. Subsequently, we purified migrasomes from Tspan4-GFP overexpressing 4T1 cells and injected them into tumor-bearing mice on days 5 and 7 after 4T1 cell injection. PBS and dead cells served as controls for therapeutic effect evaluation. In vivo bioluminescence imaging revealed that metastasis of 4T1 cells was significantly reduced in the migrasome-treated group compared with the PBS group, whereas the UV-induced dead cells did not show this effect (Fig. 6a,b). We also observed that migrasomes combined with anti-PD-1 further reduced metastasis and provided better treatment outcomes (Fig. 6c,d). These data suggest that tumor cell-derived migrasomes hold potential as therapeutic vaccines.

Fig. 6. Migrasomes have therapeutic vaccine potential for cancer immunotherapy.

a, Cells (4T1) were injected (i.v.) at day 0. Migrasomes were purified from Tspan4-GFP expressing 4T1 cells and injected (i.v.) into mice on day 5 and day 7. PBS and UV-irradiated 4T1 cells (Dead-cell) served as controls. In vivo photo emission of 4T1 cells was detected by D-luciferin injection and imaged using an IVIS spectrum imaging system on day 14. b, Statistics of the luminescence counts of 4T1 cells. PBS, n = 20 mice; migrasome, n = 19 mice; Dead-cell, n = 10 mice. c, 4T1 cells were injected (i.v.) on day 0. Migrasomes were purified from Tspan4-GFP-expressing 4T1 cells and injected (i.v.) into the mice on day 5 and day 7. Anti-PD-1 was also injected (i.p.) on day 5 (2.5 mg per kg body weight). PBS and IgG served as controls. In vivo photo emission of 4T1 cells was detected by d-luciferin injection and imaged using an IVIS spectrum imaging system on day 14. d, Statistics of the luminescence counts of 4T1 cells. n = 12 mice for each group. e, Anatomical lung images of MMTV-PyMT mice treated with PBS or PyMT tumor cell-derived migrasomes (10 μg per mouse, each time) from 10 weeks (day 0) for three times at the indicated time points. Arrowheads indicate the metastasis nodules. f, Representative H&E-stained anatomical and histological lung images from e. Insets: higher-magnification views. g, Quantification of the metastasis nodule number per lung image. PBS, n = 15 mice; migrasome, n = 17 mice. Statistical data in b, d and g are presented as the mean ± s.e.m. P values were calculated using a two-tailed, unpaired t-test.

Source data

We next investigated the therapeutic potential of de novo mammary cancer-derived migrasomes in the MMTV-PyMT mouse model. We isolated and prepared a single-cell suspension from the primary tumors of MMTV-PyMT mice. Subsequently, we seeded these primary tumor cells onto fibronectin-coated and laminin-coated dishes to generate migrasomes in vitro. Confocal imaging revealed that the primary PyMT⁺ cancer cells produced migrasomes in vitro (Extended Data Fig. 7b). We then purified the primary PyMT tumor cell-derived migrasomes and injected them into MMTV-PyMT mice at 10 weeks of age, administering the treatment three times to assess the effect on metastasis. Anatomical and histological observations, along with hematoxylin and eosin (H&E) staining, revealed that injection of PyMT tumor cell-derived migrasomes reduced the incidence of lung metastasis (Fig. 6e–g). These results further support our hypothesis that tumor cell-derived migrasomes may serve as a therapeutic tumor vaccine.

OSCC-derived migrasomes are enriched with tumor-specific antigens

To test whether human primary tumor cell-derived migrasomes are also enriched with tumor-specific antigens, we collected oral squamous cell carcinoma (OSCC) samples from three individuals who had not received chemotherapy, radiation or other treatments. We prepared a single-cell suspension from the OSCC samples and seeded the cells onto collagen-coated dishes to generate migrasomes. We found that primary OSCC cells produced a substantial number of migrasomes in vitro (Fig. 7a and Supplementary Video 2). We then purified the migrasomes from the cultured primary OSCC cells and characterized them using scanning electron microscopy (SEM). The SEM images showed that these purified structures were vesicles connected to retraction fibers (Fig. 7b), which is a morphological hallmark of migrasomes. This suggests that the structures we purified from primary OSCC cells are indeed migrasomes.

Fig. 7. Migrasomes have clinical significance in human cancers.

a, Confocal images of OSCC primary tumor cells seeded in collagen-coated dishes and stained with WGA for imaging. Scale bars, 50 μm and 10 μm (enlarged). b, SEM image of purified OSCC-derived migrasomes. Scale bar, 1 μm. c, Heat map plots show migrasome marker proteins enriched in OSCC-derived migrasomes. d, Heat map plots show mutated proteins enriched in OSCC-derived migrasomes. e, Heat map plots show testis-specific proteins enriched in OSCC-derived migrasomes. The values of the mass spectrometry results in e were normalized according to the amount in the cell and then plotted. f, Tspan4 RNA expression in transcripts per million (TPM) in breast invasive carcinoma tumors and normal tissues from the UCSC Xena Toil RNA-sequencing dataset (TCGA + GTEx). In the box plots, the center line denotes the median, the box represents the interquartile range (25th–75th percentiles), whiskers extend to 1.5 times the interquartile range (excluding outliers), and jitter points represent individual samples. Data were analyzed at https://gepia.cancer-pku.cn/. g, The survival curves of Tspan4-low and Tspan4-high groups dichotomously divided by Tspan4 median in breast invasive carcinoma using TCGA dataset. Data were analyzed at https://gepia.cancer-pku.cn/. HR, hazard ratio. h, Diagram illustrating migrasome-mediated antigen release during metastasis to unlock immunogenicity. Illustration in h created in BioRender; Lam, R. https://biorender.com/r8v3lz6 (2026).

Source data

To identify tumor antigens, we used a strategy combining quantitative mass spectrometry analysis of OSCC-derived migrasomes with genome sequencing of primary OSCC cells. We found that migrasome marker proteins, including TSPAN4, TSPAN9, integrins α5, β1 and αV, NDST, PIGK and EOGT, were enriched in OSCC-derived migrasomes (Fig. 7c). Additionally, we identified numerous mutated proteins enriched in migrasomes derived from all OSCC human samples (Fig. 7d), which is consistent with our findings in mouse breast cancer models.

In addition to mutated proteins, we also identified several testis-associated proteins, including CCDC83, C3ORF20, GK2, TEX35, APH1B, CCDC42 and PGK2, which were enriched in OSCC-derived migrasomes (Fig. 7e). These data suggest that tumor-specific antigens are enriched in OSCC-derived migrasomes.

Furthermore, using a negative-selection approach combined with electron microscopy, we detected tumor migrasomes in the plasma of individuals with breast or lung cancer. We note that this negative selection method, similar to its application in CTC enrichment, provides enrichment rather than absolute purification of tumor‑derived structures; the resulting fraction may still contain residual normal cell-derived structures. We therefore refer to these preparations as the tumor‑migrasome‑enriched fraction. These circulating structures exhibited the hallmark vesicle–retraction fiber morphology and were enriched in the migrasome markers INTEGRIN α5, NDST, PIGK and EOGT, as well as MHC class I molecules. This marker profile was consistent with that of 4T1-derived migrasomes, while lacking the sEV marker CD63 (Extended Data Fig. 8a–e). These findings indicate that the enrichment of tumor antigens in migrasomes is conserved across multiple human cancer types and further support the translational potential of migrasomes as a platform for antigen identification and immunotherapy.

Extended Data Fig. 8. Detection and isolation of tumor-derived migrasomes from cancer patient plasma.

a, Schematic workflow for isolating tumor migrasomes from patient plasma using differential centrifugation and magnetic bead–based negative selection. b, Scanning electron microscopy (SEM) image of the non-tumor–derived fraction showing vesicle-fiber structures decorated with magnetic beads. Scale bar, 0.5 μm. c, SEM image of the Tumor-migrasomes-enriched fraction showing vesicle-fiber structures without magnetic bead binding. Scale bar, 0.5 μm. d, Western blot analysis of migrasome markers (integrin α5, NDST, PIGK, EOGT), MHC class I, and sEV marker CD63 in tumor-migrasomes-enriched fraction (Tmig) and sEV fractions from lung cancer patient plasma. e, Western blot analysis of same markers in Tmig and sEV fractions from breast cancer patient plasma.

Reduced Tspan4 in breast invasive carcinoma is predictive of poor survival

To determine the clinical significance of our findings, we explored publicly available datasets on human cancer from The Cancer Genome Atlas (TCGA). We found that Tspan4 is not frequently mutated in the majority of cancer types. However, Tspan4 was downregulated in breast invasive carcinoma compared to normal tissue (Fig. 7f). This observation is consistent with our findings that Tspan4 knockout enhanced the metastasis of 4T1 and PyMT de novo cancers (Fig. 1a–f).

We then examined whether the expression level of Tspan4 in breast invasive carcinoma correlates with clinical outcomes. We divided 1,070 individuals with breast invasive carcinoma into two groups based on the median expression of Tspan4 and compared overall survival between the Tspan4-high and Tspan4-low groups. Our analysis revealed no notable difference in overall survival between the Tspan4-high and Tspan4-low groups during the first 10 years. However, after 10 years, individuals in the Tspan4-high group had notably longer overall survival than those in the Tspan4-low group (Fig. 7g), indicating that Tspan4 is a positive prognostic factor for breast invasive carcinoma.

These results are consistent with our findings that overexpression of Tspan4 enhanced migrasome formation and inhibited the metastasis of 4T1 breast cancer in mice, and they align with the broader observation that Tspan4 downregulation is a frequent event in multiple cancer types. Together, these clinical correlations underscore the essential significance of migrasomes in cancer immunotherapy and suggest that Tspan4 expression may serve as a biomarker for migrasome-mediated immune surveillance.

Discussion

In this study, we found that CTCs produce migrasomes during migration through blood vessels, and these migrasomes effectively stimulate CD8⁺ T cell-dependent antitumor immunity. Reducing migrasome formation by knocking out Tspan4 enhances metastasis, a phenotype rescued by administration of purified migrasomes. Proteomic and genomic analyses identified tumor-associated antigens including cancer-testis and mutated antigens enriched in migrasomes. We further show that migrasomes require uptake and cross-presentation by macrophages and dendritic cells in secondary lymphoid organs. Human OSCC-derived migrasomes similarly carry tumor antigens, and preliminary evidence shows that tumor patient-derived migrasomes are detectable in plasma. TCGA analysis indicates that Tspan4 is downregulated in breast invasive carcinoma and correlates with prolonged survival. In summary, our study reveals migrasome-mediated antigen release as a metastasis-dependent mechanism of immune activation (Fig. 7h).

Our findings expand the understanding of extracellular vesicle-mediated tumor–immune interactions. While exosomes can either promote immune evasion¹² or stimulate immunity^5,13, migrasomes are distinct organelles intrinsically linked to cell migration³⁸. Migrasome biogenesis is migration dependent, rendering antigen release a metastasis-enriched event. Compared with sEVs, tumor lysates and apoptotic bodies, migrasomes exhibited superior metastasis suppression (Extended Data Fig. 1d–f), likely owing to their enrichment of MHC class I molecules, efficient trafficking to lymph nodes and spleen, and uptake by APCs (Extended Data Figs. 2d–f and 3a,b). Additionally, migrasomes can package damaged mitochondria and mitochondrial DNA³³, providing endogenous danger signals that enhance cross-presentation³⁹—a property distinct from other EV subtypes.

The functional consequence of migrasome release likely depends on tumor immunogenicity. In immunocompetent mice, Tspan4 overexpression suppressed metastasis of 4T1 tumors, whereas in immunodeficient nude mice it promoted metastasis (Fig. 3a,b). Pan-cancer TCGA analysis showed Tspan4 downregulation in some cancers (for example, breast and lung) and upregulation in others (for example, glioblastoma and melanoma). Thus, in highly immunogenic tumors, Tspan4 expression may be selected against, thereby evading migrasome-mediated immune activation; in poorly immunogenic contexts, migrasomes might facilitate metastasis through immune-independent mechanisms.

A long-standing question is why so few metastatic cells survive in distant organs⁴⁰. Our study suggests that migrasome-mediated antigen release contributes to metastatic control. Once shed into the circulation, migrasomes traffic to lymphoid organs and are phagocytosed by APCs, effectively functioning as an in situ-generated vaccine against dissemination. Given that migrasomes are produced specifically during cell migration, this mechanism couples metastasis to immune activation, explaining the inefficiency of metastatic colonization and revealing a natural immune surveillance process that could be therapeutically amplified through migrasome-based vaccination strategies.

These findings have translational implications. We propose generating migrasomes from resected tumor tissue ex vivo for autologous reinfusion as a personalized vaccine. This concept is supported by the efficacy of PyMT-derived migrasomes in the autochthonous PyMT model (Fig. 6e–g). Compared with whole tumor cell vaccines⁴¹, migrasomes offer antigen content, intrinsic adjuvant properties and efficient lymphatic trafficking. Furthermore, combining migrasome proteomics with tumor genomics may facilitate rapid identification of patient-specific antigens for personalized immunotherapy.

The superior immunogenicity of migrasomes may stem from their biogenesis. Migrasomes assemble via tetraspanin-enriched macrodomains and are enriched in MHC class I molecules (Extended Data Fig. 2a–c)²⁷. They also contain secreted cytokines^19,42,43 and, under stress, damaged mitochondria³³, which may activate innate sensing pathways^39,44. These features may collectively enhance APC activation and cross-presentation efficiency.

We observed that the cancer-testis antigen BRDT exhibited stronger immunogenicity than mutated antigens, and only BRDT-deficient migrasomes showed substantially reduced activity (Fig. 5g,h). The restricted expression of cancer-testis antigens to testes and tumors may reduce self-tolerance⁴⁵, enabling more robust T cell responses. Nonetheless, the immunodominance hierarchy is influenced by multiple factors⁴⁶, warranting further investigation.

Given the inherent diversity, heterogeneity and complexity of tumors, the role of migrasomes in tumor immunity and metastasis likely varies across tumor types. In ‘cold tumors’, with poor immune infiltration and diminished antitumor immunity⁴⁷, migrasome-induced T cell responses may be insufficient to overcome T cell exclusion and immunosuppression within the tumor microenvironment⁴⁸. In such immunologically ‘cold’ contexts, migrasomes might fail to elicit productive immunity, or could even exert pro-metastatic functions independent of adaptive immunity, as suggested by our nude mouse data (Fig. 3a,b) and the TCGA expression heterogeneity. Thus, tumor-specific studies are critical to systematically investigate how migrasomes modulate tumor progression, immune evasion and immune response. Moreover, understanding whether migrasome production and cargo loading can be therapeutically manipulated—by enhancing Tspan4 expression or promoting immunogenic cargo sorting—may open new avenues for boosting antitumor immunity in patients with metastatic disease. These studies will be essential for understanding the context-dependent roles of migrasomes and for the development of targeted therapeutic strategies that exploit migrasomes to enhance antitumor immunity in specific tumor types.

Methods

Mice

All animal experiments were approved by the Institutional Animal Care and Use Committee (IACUC) and conducted in accordance with governmental regulations and the guidelines of Tsinghua University and Zhejiang University for animal welfare (approval nos. 18-YL2, 22-YL2 and ZJU20250528). Mice used in the experiments were randomly selected from age-matched and sex-matched groups, where applicable, and assigned to experimental or control groups. WT C57BL/6J mice were purchased from the Laboratory Animal Resources Center of Tsinghua University, while BALB/c and nude mice were purchased from Charles River and housed at the Tsinghua University and/or Zhejiang University Animal Center. Mice were kept in ventilated cages within a specific pathogen-free facility under a 12-h light–12-h dark cycle, with a temperature range of 22–26 °C and humidity maintained at 30–70%, and provided with sterile pellet food (Xietong Shengwu, XTI01JX‑002) and water ad libitum. All tumors remained within the ethics-approved maximum diameter of 2 cm.

Generation of Tspan4^−/− mice

Tspan4^−/− mice were generated using CRISPR–Cas9 technology. The sgRNA sequence used for CRISPR–Cas9 targeting was 5′-ATGCAGCCCACGAAGCCGA-3′. Genotyping was performed by PCR using the following primers: Tspan4-knockout-F: GTCAACCTCACCCTGTGTGT and Tspan4-knockout-R: GTACTCCAGACCCAGCAACC.

Generation of MMTV-PyMT; Tspan4^−/− mice

MMTV-PyMT mice were kindly provided by the laboratory of H. Zheng at Tsinghua University. To generate MMTV-PyMT⁺; Tspan4^−/− mice, Tspan4^−/− mice were crossed with MMTV-PyMT mice.

B2m-knockout mice (strain S-KO-01190) were purchased from Cyagen. Batf3-knockout mice (C57BL/6J background) were a generous gift from the L. Wu laboratory at Tsinghua University. OT-I mice (C57BL/6J background) were a generous gift from the H. Qi laboratory at Tsinghua University.

Human OSCC experiments

The experiments involving human OSCC were approved by the Ethics Committee of the Institutional Review Board at Beijing Stomatological Hospital, Capital Medical University (approval no. CMUSH-IRB-KJ-PJ-2024-40). Informed consent was obtained from the participants. No compensation was offered to any participant.

Isolation of tumor cell-derived migrasomes from human blood

The use of human blood samples was approved by the Ethics Committee on Human Research at the West China Hospital of Sichuan University (permission number 2023 review; no. 1546). Informed consent was obtained from all participants. No compensation was offered to any participant. Plasma samples (5–10 ml) were collected from individuals with cancer (EDTA-anticoagulated) after obtaining informed consent. Plasma was centrifuged at 1,000g for 5 min at room temperature, and the supernatant was transferred to a new tube. The supernatant was further centrifuged at 1,500g for 5 min to remove residual cells and debris, and the resulting supernatant was ultracentrifuged at 20,000g for 60 min at 4 °C to pellet crude migrasomes. The crude migrasome pellet was resuspended in PBS and subjected to negative selection using the EasySep CTC Enrichment Kit (19657, STEMCELL Technologies) following the manufacturer’s instructions, with an additional depletion step to minimize platelet and erythrocyte contamination. Specifically, 5–10 μg of biotinylated anti-human CD41 and anti-human CD235ab antibodies were used, and 20 μl of streptavidin microbeads were added to the sample, followed by incubation before magnetic separation.

Intravital imaging

AF647-WGA (1–3 μg, Thermo Fisher, W32466) was i.v. administered to BALB/c or nude mice (female, 8–12 weeks old) for blood vessel labeling. Anesthesia was induced with an i.p. injection of avertin (375 mg per kg body weight). Following anesthesia, the mice were dissected to expose the spleen, where 1 million PH-mCherry-expressing 4T1 cells were injected. Afterward, the liver was exposed for intravital imaging using a Dragonfly spinning-disk microscopy system.

Cell culture

The cell lines 4T1 and LLC, as well as primary cancer cells, were cultured in RPMI 1640 medium (Gibco) or DMEM (Gibco) supplemented with 10% fetal bovine serum (Vivicum), 2 mM GlutaMAX and 100 U ml⁻¹ penicillin–streptomycin. The cultures were maintained in a humidified incubator at 37 °C with 5% CO₂. The 4T1-luc (stable luciferase) and parental 4T1 cell lines were kindly provided by the laboratories of X. Lin (Tsinghua University) and L. Deng (Shanghai Jiao Tong University), respectively, and LLC cells were obtained from the American Type Culture Collection (CRL-1642).

Purification of migrasomes from cultured cells

Migrasomes were purified by iodixanol-sucrose density gradient centrifugation with an Optiprep kit (LYSISO1, Sigma-Aldrich). For 4T1/PyMT migrasomes, cells were cultured on dishes coated with fibronectin (1 μg ml⁻¹, 1 h) and laminin-521 (0.1 μg ml⁻¹, 1 h) for 36 h. For OSCC migrasomes, cells were cultured on dishes coated with collagen (20 μg ml⁻¹, 3 h) for 24 h. After medium removal, cells were trypsinized, and the resulting suspension was centrifuged at 1,000g for 10 min at 4 °C to remove cell bodies, followed by centrifugation at 2,000g for 10 min at 4 °C to further eliminate residual cell debris. The supernatant was centrifuged at 20,000g for 60 min at 4 °C to collect crude migrasome.

The pellet was resuspended in extraction buffer (Sigma-Aldrich) and subjected to density gradient centrifugation at 150,000g for 4 h at 4 °C using a multi-step Optiprep gradient (3%, 5%, 8%, 12%, 16%, 19% (sample), 22.5%, 25%). Gradient volumes were 300 μl (3%, 5%, 8%), 900 μl (12%), 1,000 μl (16%), 800 μl (19%), 500 μl (22.5%) and 300 μl (25%). Fractions 3–5 (500 μl each) were collected and diluted with 500 μl PBS, followed by centrifugation at 20,000g for 20 min at 4 °C. The resulting pellet was washed with PBS and centrifuged at 1,000g for 10 min at 4 °C. The supernatant was collected and centrifuged at 20,000g for 20 min at 4 °C to yield purified migrasomes.

Isolation of migrasomes in vivo

BALB/c mice were intraperitoneally injected with 1 μg per gram of body weight of anti-CD41 in 200 μl PBS. After 8–10 h, 2 million Tspan4-GFP-expressing 4T1 cells were i.v. injected into the mice. Five to six hours after injection, anesthesia was induced with avertin (375 mg per kg body weight, i.p.). Blood was collected from the ocular venous plexus into tubes containing blood collection buffer (PBS with 40 mM EDTA; 1 ml per mouse). The blood was processed in a series of centrifugations: first at 800g for 5 min at 4 °C, then at 1,000g for 15 min at 4 °C to remove blood cells, followed by a final centrifugation at 20,000g for 60 min at 4 °C to isolate the crude extracellular structures fraction.

For negative selection of 4T1-derived migrasomes, 50 ml of blood was collected per batch. The crude extracellular structures from platelet-depleted mice were resuspended in 20 μl PBS and incubated with 10 μl of each biotinylated antibody: anti-mouse CD41, anti-mouse Ter119, anti-mouse CD31 and anti-mouse CD45.2 (eBioscience) for 30 min at 4 °C, with mixing 2–3 times. After incubation, 40 μl streptavidin-coated microbeads (BEAVER, 22308) were added and the mixture was incubated for 60 min at 4 °C, with gentle mixing 2–3 times. Following bead binding, the tube was placed in a DynaMag-Spin magnet (Invitrogen, 12320D) for 10–20 min at 4 °C. The supernatant was transferred to a new tube and centrifuged at 20,000g for 30–60 min at 4 °C to obtain the in vivo 4T1-derived migrasomes.

Isolation of primary PyMT tumor cells

Primary tumor tissues were collected from 10- to 12-week-old MMTV-PyMT mice, then minced into small pieces. The tissue was digested for 30 min at 37 °C in a digestion mixture containing 1 mg ml⁻¹ collagenase I (C8140, Solarbio) and collagenase IV (C8160, Solarbio). After digestion, single-cell suspensions were obtained, and tumor-associated cells, including lymphocytes and fibroblasts, were depleted using a Mouse Tumor Cell Isolation Kit (130-110-187, Miltenyi Biotec), to enrich for tumor cells.

In vitro cancer cell killing assay

Migrasomes (4T1 derived; 20 μg) were i.v. injected into BALB/c mice on days 0 and 7 for immunization. On day 19, CD8⁺ T cells were isolated from the spleen and lymph nodes. The isolated CD8⁺ T cells were then co-cultured with 4T1 cells at a ratio of 10:1 (CD8⁺ T cells:4T1 cells) for 19 h. Subsequently, live-cell imaging was performed.

Quantitative proteomics

For 4T1, equal amounts of protein were reduced, alkylated, trypsin-digested, desalted and labeled with TMT reagents (Thermo, Pierce). The TMT-labeled peptides were mixed, desalted (Sep-Pak tC18), and fractionated into 24 fractions by high-pH RPLC (XBridge BEH300 C18 column, 5 μm, 300 Å, 250 × 4.6 mm; Waters) using mobile phases A (2% acetonitrile, pH 10) and B (98% acetonitrile, pH 10). For liquid chromatography coupled with tandem mass spectrometry (LC–MS/MS), peptides were separated on a homemade C18 column (100 μm × 50 cm, 1.9 μm, 120 Å; Dr. Maisch) using an Ultimate 3000 system coupled to a Q-Exactive HF-X mass spectrometer (Thermo). A linear gradient was applied at 300 nl min⁻¹. The mass spectrometer operated in data-dependent acquisition mode, scanning at 300–1,800 m/z at a resolution of 60,000; the top 40 most intense peaks (charge ≥2) were selected for fragmentation (normalized collision energy (NCE), 32%; 1-Da isolation, 15-s dynamic exclusion). MS/MS spectra were acquired at a resolution of 15,000 (automatic gain control (AGC) target, 1 × 10⁵; maximum injection time, 50 ms). MS/MS data were searched against the UniProt mouse database using SEQUEST/Proteome Discoverer 2.2 (full tryptic, up to two missed cleavages; fixed modifications: carbamidomethyl (C), TMT (K, N-term); variable: oxidation (M); precursor mass tolerance, 20 ppm; fragment, 20 mmu). PSMs were validated by Percolator at a 1% false discovery rate. All experiments were performed in triplicate.

For OSCC, protein concentrations (BCA assay) were determined, and 6 μg protein per sample was used, adjusted to 20 μl with 1% sodium deoxycholate. Samples were reduced (10 mM TCEP), alkylated (40 mM chloroacetamide) and digested overnight with trypsin/LysC (100:1) at 37 °C. Peptides were desalted with SDB-RPS stage tips, concentrated and resuspended in 0.1% formic acid. For LC–MS/MS, using an Orbitrap Ascend mass spectrometer, peptides were loaded onto a trap column (75 μm × 20 mm, 3 μm C18) using mobile phase A (0.1% formic acid) under 620 bar, and separated on an analytical column (100-μm inner diameter, ReproSil-Pur C18-AQ 1.9 μm) with a 240-min gradient of 6–55% B (80% acetonitrile, 0.08% formic acid) at 250 nl min⁻¹. Mass spectrometry data were acquired in data-independent acquisition mode: full-scan MS (400–1,200 m/z; resolution of 240,000; AGC target, 4 × 10⁵) and MS/MS (120 variable windows; NCE, 30%; resolution of 30,000; AGC target, 5 × 10⁵; maximum injection time, 59 ms). Data were analyzed with Spectronaut 15.6 (default settings) against the UniProt human database, using MS2 quantification and q-value filtering.

SEM

Purified migrasomes were carefully placed onto silicon wafers coated with poly-L-lysine (Sigma, P4707), which had been incubated for 1 h at room temperature. The samples were left to adhere for at least 4 h, followed by fixation in a mixture of 2.5% glutaraldehyde and 2% paraformaldehyde for over 1 h. After fixation, the samples were washed with 0.1 M phosphate buffer consisting of 36 ml of 0.2 M Na₂HPO₄•12H₂O, 14 ml of 0.2 M NaH₂PO₄•2H₂O, and 50 ml of distilled water, adjusted to pH 7.2. The samples were then rinsed in phosphate buffer for 10 min before being incubated with a solution of 1% osmium tetroxide and 1.5% potassium ferricyanide for 30 min. After additional washing with distilled water, the samples were dehydrated through a graded ethanol series (50%, 70%, 80%, 90%, 100%, 100%, 100%; 2 min per concentration). Following dehydration, the samples were subjected to critical-point drying using a Leica EM CPD300. Finally, a 10-nm gold layer was sputter-coated onto the samples, which were subsequently examined using a FEI Helios NanoLab G3 UC SEM.

Peptide immunization

BALB/c mice were immunized by i.v. injection of a mixture containing 50 μg of poly(I:C) and 100 μg of synthesized peptide on days 0 and 12. Following the immunization, 4T1 cells were i.v. injected into the mice. On day 33, in vivo luminescence of the 4T1 cells was detected by administering D-luciferin and imaging using the IVIS Spectrum system.

In vivo UNC3230 treatment

Cells (4T1) stably expressing firefly luciferase were cultured to approximately 80% confluence and treated with vehicle (PBS containing dimethylsulfoxide) or 80 μM UNC3230 (16401, Cayman Chemical; dissolved in dimethylsulfoxide) for 24 h. Cells were then harvested, washed and resuspended in PBS. Female BALB/c mice (6–8 weeks old) were i.v. injected with 7 × 10⁵ treated 4T1 cells per mouse. Two hours after tumor cell inoculation, mice received an intraperitoneal injection of UNC3230 at 1 μg per gram of body weight or an equivalent volume of vehicle. A second identical dose of UNC3230 or vehicle was administered on day 4. On day 14, mice were injected intraperitoneally with D-luciferin, anesthetized with isoflurane, and imaged using an IVIS Spectrum in vivo imaging system to evaluate metastatic burden.

In vitro T cell activation using BMDMs and BMDCs

Bone marrow cells were collected from C57BL/6 mice under sterile conditions. To generate BMDCs, cells were cultured in RPMI 1640 medium supplemented with 10% (vol/vol) heat-inactivated CELLMAX serum (inactivated at 56 °C), 1× penicillin–streptomycin, 55 μM β-mercaptoethanol, 1 mM sodium pyruvate, 2 mM L-glutamine, 10 mM HEPES and 200 ng ml⁻¹ recombinant mouse FLT3L (250-31L, PeproTech). To generate BMDMs, bone marrow cells were cultured in RPMI 1640 medium with the same basal supplements plus 20 ng ml⁻¹ recombinant mouse M-CSF (315-02, PeproTech). Cultures were maintained for 8 days with half-medium changes every 2–3 days. CD8⁺ T cells were isolated from mouse spleens by negative selection using a commercial kit (19853, STEMCELL Technologies) according to the manufacturer’s protocol. BMDCs or BMDMs (5 × 10⁵ cells per well) were seeded in 24-well plates and co-cultured with 5 μg of migrasomes purified from LLC cells together with freshly isolated CD8⁺ T cells (1 × 10⁶ cells per well) in the corresponding BMDC or BMDM medium further supplemented with 10 ng ml⁻¹ recombinant human IL-2. After co-incubation for the indicated times, cells were collected, blocked with Fc receptor blocker and surface-stained with fluorophore-conjugated antibodies: FITC anti-mouse CD69 (104505, BioLegend), Brilliant Violet 605 anti-mouse CD25 (102036, BioLegend), Alexa Fluor 700 anti-mouse CD62L (104426, BioLegend), Brilliant Violet 421 anti-mouse CD8a (100738, BioLegend) and PE anti-mouse/human CD44 (103007, BioLegend). For intracellular granzyme B staining, cells were subsequently fixed and permeabilized using the BD Perm/Wash buffer and then stained with FITC anti-mouse granzyme B (11-8898-82, Invitrogen). Stained cells were acquired on a flow cytometer and analyzed with appropriate software.

Bone marrow transfer and adoptive T cell transfer metastasis model

Eight-week-old C57BL/6J recipient mice were exposed to 10 Gy of total body irradiation. Twenty-four hours after irradiation, mice were reconstituted via tail vein injection with bone marrow cells freshly isolated from sex-matched and age-matched WT C57BL/6J donors or B2m-knockout donors. Bone marrow cells were collected, filtered and depleted of red blood cells using ACK lysis buffer before transplantation. After engraftment, mice were maintained on antibiotic-containing drinking water for 8 weeks to allow full immune reconstitution. OT-I T cell antigen receptor transgenic CD8⁺ T cells were purified from pooled spleens and peripheral lymph nodes of OT-I mice by negative selection using a mouse CD8⁺ T cell isolation kit (STEMCELL Technologies) according to the manufacturer’s protocol. Each reconstituted mouse received 1 × 10⁶ OT-I CD8⁺ T cells via i.v. injection. Twenty-four hours after T cell transfer, mice were i.v. injected with either PBS (control group) or LLC-OVA-derived migrasomes (experimental group) isolated as described above. Sixteen hours after migrasome administration, all mice were i.v. challenged with 1 × 10⁶ LLC-OVA tumor cells. On day 25 after tumor cell injection, mice were euthanized for phenotyping.

Isolation of OSCC cells

Tumor samples were obtained from individuals diagnosed with OSCC after resection of the primary tumor. Approximately 1–2 cm³ of tumor tissue was collected, ensuring that necrotic areas were avoided. To remove blood contamination, the tissue was washed with PBS. The tumor tissue was then finely minced into 1-mm³ pieces using scissors and subsequently digested with 1 mg ml⁻¹ of type I collagenase and 1 mg ml⁻¹ of dispase in a 37 °C water bath for 30 min. After digestion, the enzyme solution was discarded, and the isolated OSCC cells were collected. The resulting cells were cultured, and the P2–P6 generation cells were used for subsequent experiments.

Statistics and reproducibility

Data analysis, statistical testing and visualization were performed using GraphPad Prism 9 and GraphPad Prism 10. Results are represented as the mean ± s.e.m. when it is not indicated specifically. A two-tailed, unpaired Student’s t-test was used to compare two groups. Statistical differences are indicated as exact P values in figures. The data distribution was assumed to be normal, but this was not formally tested. All statistical tests were conducted at a 95% confidence interval. No statistical methods were used to predetermine sample sizes. Rather, sample size was determined based on similar studies in the field. Data collection and analysis were not performed blind to the conditions of the experiments. Each biological experiment was obtained from different and independent samples. Quantification and statistics were derived from n = 3 independent experiments unless otherwise specified. All other experiments were performed independently at least twice but usually three to four times. Specifically, the experiments in Fig. 7a and Extended Data Fig. 2b were repeated three times with similar results, and those in Figs. 2b–d,i and 7b and Extended Data Figs. 1c, 2c–f, 3a,b, 7b and 8b–e were repeated twice with similar results. For in vivo studies, age- and sex-matched mice were randomly assigned to experimental and control groups for all experiments, and the animal numbers used in the experiments are indicated in the corresponding figure legends. No data were excluded from the analyses. The specific statistical methods applied to each analysis and the results are detailed in the figure legends.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Online content

Any methods, additional references, Nature Portfolio reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/s41590-026-02641-0.

Author contributions

L.Y. and D.J. conceived the experiments, supervised the project and wrote the paper. D.J. and J.H. carried out the experiments. R.X., M.S., S.L., H.J., Y.X., Y.C., B.Y., X.R. and X.T. helped to perform the experiments. X.F. helped to collect the OSCC samples. L.J. helped to collect the human blood samples. All authors discussed the paper, commented on the project and contributed to preparing the paper.

Peer review

Peer review information

Nature Immunology thanks Graça Raposo, Michael Sixt and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. Primary Handling Editor: Nick Bernard, in collaboration with the Nature Immunology team.

Funding

This research was supported by the National Natural Science Foundation of China (grant nos. 32522027, 32330025 and U25A20630), the National Key R&D Program of China (grant nos. 2025YFA1804303, 2024YFA1307301 and 2024YFF1502900), the New Cornerstone Science Foundation (NCI202529) and Scientific and Technological Innovation Project of China Academy of Chinese Medical Sciences (grant no. CI2023C024YL).

Data availability

Sequencing data and proteomics data from this study have been deposited at the China National Center for Bioinformation (CNCB) under accession codes CRA042049, HRA018134, OMIX016539 and OMIX016540. Source data are provided with this paper. All other data are present in the paper and its Supplementary Information.

Competing interests

L.Y. is the scientific founder of Migrasome Therapeutics. The other authors declare no competing interests.

Extended data

is available for this paper at https://doi.org/10.1038/s41590-026-02641-0.

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s41590-026-02641-0.