Chemotherapy elicits pro-metastatic extracellular vesicles in breast cancer models

Abstract

Cytotoxic chemotherapy is an effective treatment for invasive breast cancer. However, experimental studies in mice also suggest pro-metastatic effects of chemotherapy. Primary tumours release extracellular vesicles (EVs), including exosomes, that can facilitate the seeding and growth of metastatic cancer cells in distant organs, but the effects of chemotherapy on tumour-derived EVs remain unclear. Here we show that two classes of cytotoxic drugs broadly employed in pre-operative (neoadjuvant) breast cancer therapy, taxanes and anthracyclines, elicit tumour-derived EVs with enhanced pro-metastatic capacity. Chemotherapy-elicited EVs are enriched in annexin-A6 (ANXA6), a Ca²⁺-dependent protein that promotes NF-kB-dependent endothelial cell activation, Ccl2 induction, and Ly6C⁺CCR2⁺ monocyte expansion in the pulmonary pre-metastatic niche to facilitate the establishment of lung metastasis. Genetic inactivation of Anxa6 in cancer cells, or Ccr2 in host cells, blunts the pro-metastatic effects of chemotherapy-elicited EVs. ANXA6 is detected, and potentially enriched, in the circulating EVs of breast cancer patients undergoing neoadjuvant chemotherapy.

Introduction

Neoadjuvant chemotherapy may provide long-term clinical benefit in patients diagnosed with invasive breast cancer, especially when the primary tumour fully regresses before surgery ^1–6. However, the therapeutic benefits of neoadjuvant chemotherapy may be limited by tumour-promoting host responses that are induced by certain cytotoxic drugs ⁷. Several reports have documented pro-metastatic effects of cytotoxic agents in mouse mammary tumour models ^8–13. For example, paclitaxel (PTX), a microtubule-stabilizing drug often used in breast cancer therapy ^5,6, was reported to enhance expression of vascular-endothelial growth factor receptor-1 (VEGFR1) on pulmonary endothelial cells to facilitate cancer-cell adhesion and subsequent metastasis ¹³. Both PTX and doxorubicin (DOX) – an anthracycline also used in breast cancer therapy ^5,6 – increased the ability of perivascular TIE2⁺ macrophages ^14–16 to promote cancer-cell intravasation in primary mammary tumours, resulting in heightened pulmonary metastasis ^8,12. Collectively, pre-clinical data in mouse models suggest that the pro-metastatic capacity of certain chemotherapies may involve facilitation of both cancer cell intravasation in primary tumours and extravasation to secondary, metastatic sites.

Primary tumours release extracellular vesicles (EVs) that can modulate the biology of distant organ niches to enhance seeding and growth of metastatic cancer cells ^17–24. In this study, we examined the effects of PTX and DOX on the release, properties and pro-metastatic potential of tumour-derived EVs in mouse models of chemoresistant breast cancer.

Results

PTX enhances pulmonary metastasis in mouse mammary tumour models

We examined the effects of PTX on metastasis in two mouse breast cancer models: transgenic MMTV-PyMT mice (FVB/n background), which develop multifocal mammary tumours ^25–27, and immunodeficient Rag1^−/− (C57BL/6 background) or Swiss nu/nu mice challenged with 4T1 cancer cells ²⁸. In order to trace metastasis, the 4T1 cells were modified to express a fluorescent CD9-mCherry (mCh) fusion protein targeted to cellular membranes; in some experiments, 4T1 cells were further modified to express a human ERBB2 (HER2) transgene ²⁹. The 4T1 tumour studies used immunodeficient mice to avoid potential anti-tumour immune responses against mCh or HER2.

Tumour-bearing mice received 3 doses of PTX (10 mg/kg) or vehicle (cremophor, CREMO) before analysis (Fig. 1a). PTX had modest, if any, inhibitory activity on the growth of primary mammary tumours in both MMTV-PyMT and 4T1 models (Fig. 1b-e). However, it increased pulmonary tumour deposits in a fraction of the mice, in line with previous findings ^8–12. In particular, PTX increased the incidence (Fig. 1f, g) and mean size (Fig. 1f, h, i) of spontaneous metastases in lungs of MMTV-PyMT mice, as well as the seeding of mCh⁺ 4T1 cancer cells in the lungs of tumour-bearing Rag1^−/− mice (Fig. 1j-k), compared to vehicle. PTX also enhanced pulmonary seeding of mCh⁺HER2⁺ cancer cells in some tumour-bearing Swiss nu/nu mice (Fig. 1l), although it did not augment the frequency of mCh⁺HER2⁺ cancer cells in the systemic circulation (Fig. 1m). Together, these results indicate that PTX may augment, rather than limit, lung metastasis in mouse models of chemoresistant breast cancer.

Figure 1. PTX enhances pulmonary metastasis in mammary tumour-bearing mice.

a. Drug scheduling in tumour-bearing mice.

b. Cumulative weight of multifocal mammary tumours (mean ± s.e.m.) in MMTV-PyMT mice. CREMO, n=14 mice; PTX, n=16. Each dot represents one mouse carrying several tumours. Data show two independent experiments combined (EPFL cohort).

c. Volume of 4T1 or 4T1-mCh tumours (mean ± s.e.m.; n=7 mice/group) in untreated Rag1^−/− mice.

d. Weight of 4T1-mCh tumours (mean ± s.e.m.) in Rag1^−/− mice. CREMO, n=12 mice; PTX, n=14. Data show two independent experiments combined.

e. Weight of 4T1 (n=4) or 4T1-mCh/HER2 tumours (mean ± s.e.m.) in Swiss nu/nu mice. PBS, n=7 mice; CREMO, n=8; PTX, n=9. Statistical analysis by one-way ANOVA with Tukey’s multiple comparison test.

f. Representative hematoxylin/eosin (H&E) images of lung sections of MMTV-PyMT mice from the experiment shown in (b). Scale bars, 1 mm. Data are quantified in (g) and (h).

g-i. Number (g) and mean area (h, i) of pulmonary metastases (mean ± s.e.m.) in MMTV-PyMT mice. (g, h): CREMO, n=14 mice; PTX, n=16; two independent experiments combined (EPFL cohort). (i): PBS, n=21; PTX, n=49; five independent experiments combined (OHSU cohort). Statistical analysis in (g) and (i) by unpaired two-tailed Student’s t-test.

j. Fluorescence-activated cell sorting (FACS) analysis of mCh⁺CD31^– cancer cells (mean ± s.e.m.; relative to viable CD45⁻ lung-derived cells) in lungs of 4T1-mCh tumour-bearing mice. CREMO, n=8 mice; PTX, n=9. Statistical analysis as in (g). The FACS panels on the right show the gating strategy.

k. Representative confocal immunofluorescence images showing mCh⁺ (red) 4T1 cancer cells in lung sections of mice from the experiment in (j). Nuclei are stained with DAPI (blue). Scale bars, 200 μm (left and middle panel) and 50 μm (right panel).

l. FACS analysis of mCh⁺CD31⁻ cancer cells (mean ± s.e.m.) in lungs of 4T1 (n=4) or 4T1-mCh/HER2 tumour-bearing mice. PBS, n=7 mice; CREMO, n=8; PTX, n=9.

m. FACS analysis of mCh⁺HER2⁺CD45^– cancer cells (mean ± s.e.m.; absolute cell counts) in blood of 4T1-mCh/HER2 tumour-bearing mice. PBS, n=7 mice; CREMO, n=8; PTX, n=9. The FACS panels on the right show the gating strategy.

Source data are shown in Supplementary Table 5.

Chemotherapy-elicited EVs promote mammary tumour metastasis in mice

Tumour-derived EVs facilitate metastasis of primary tumours by altering the properties of pre-metastatic niche-associated host cells ^17–24. In order to gauge the participation of tumour-derived EVs in PTX-induced mammary tumour metastasis, we used sequential ultracentrifugation to isolate EVs ^30,31 from cell culture media of unmodified 4T1 cancer cells treated with either PTX or CREMO (PTX-EV and CREMO-EV, respectively). The resulting preparations were highly enriched in small EVs, as shown by both transmission electron microscopy (TEM; Fig. 2a) and nanoparticle tracking analysis (NTA; Supplementary Fig. 1a). Also, the EV-associated proteins CD9, CD81 and syntenin-1 were enriched in EVs compared to producer cells, as demonstrated by Western blotting analysis (Supplementary Fig. 1b). Although we cannot rule out co-purification of small apoptotic bodies, the physical and molecular properties of the preparations are consistent with bona fide EVs ^30–32 with mode size comprised between 100 and 150 nm.

Figure 2. Chemotherapy-elicited EVs are pro-metastatic in mouse and zebrafish tumour model.

a. TEM images of CREMO-EV (n=1 biological sample) and PTX-EV (n=2 independent biological samples) isolated from 4T1 cells. One representative image is shown for each EV type. Bottom panels show magnified fields of upper panels. Scale bars, 200 nm (upper panels) and 100 nm (bottom panels).

b. Schematics illustrating lung pre-conditioning and tumour colonization assays.

c. Number of 4T1 metastatic nodules (mean ± s.e.m.; n=8 mice/group) in lungs of pre-conditioned Balb/c mice. Statistical analysis by unpaired two-tailed Student’s t-test. Right panels show representative H&E images of lung sections (magnified fields below). Scale bars, 1 mm.

d. Procedure to isolate tumour-derived EVs from chemotherapy-treated MMTV-PyMT mice.

e, f. Number of MMTV-PyMT metastatic nodules (mean ± s.e.m.) in lungs of pre-conditioned FVB/n mice; CREMO-EV, n=13 mice; PTX-EV, n=14; PBS-EV, n=8; DOX-EV, n=9. Statistical analysis as in (c). Lower panels in (e) show representative H&E images of lung sections. Scale bars, 1 mm.

g. Number of MMTV-PyMT metastatic nodules (mean ± s.e.m.) in lungs of pre-conditioned FVB/n mice. PBS, n=9 mice; CREMO 1, n=9; CREMO 10, n=8; PTX 1 mg/kg, n=9; PTX 10 mg/kg, n=9; DOX, n=9. CREMO 1 and 10 are the vehicle controls for 1 and 10 mg/kg PTX, respectively. Statistical analysis by one-way ANOVA with Tukey’s multiple comparison test.

h. Concentration (mean ± s.e.m.; n=5 acquisitions of one sample/condition) and size distribution of EVs isolated from plasma of 4T1-mCh tumour-bearing mice treated as indicated, determined by NTA.

i. Schematics illustrating experiments in zebrafish embryos.

j. Volume of tumour deposits (mean ± s.e.m.) in embryos injected with CREMO-EVs (n=24), PTX-EVs (n=22), CREMO (n=12) or PTX (n=12), determined by confocal imaging analysis. Statistical analysis as in (g).

k. Confocal image of a representative zebrafish embryo injected with CFP⁺ MDA-MB-453 cells (blue). Blood vessels are GFP⁺ (green). The right panel shows the caudal haematopoietic area (CHA) with CFP⁺ tumour deposits. Scale bar, 0.5 mm.

l. Representative confocal images of the CHA of zebrafish embryos, imaged as in (k). Scale bar, 70 μm. Quantitative data are shown in (j).

Source data are shown in Supplementary Table 5.

We pre-conditioned immunocompetent Balb/c mice with either PTX-EV or CREMO-EV (two doses, each corresponding to about 4x10⁹ EVs; Supplementary Fig. 1c), followed by intravenous injection of 4T1 cells (Fig. 2b). In this lung colonization assay, PTX-EV increased the number of metastatic lung nodules compared to CREMO-EV (Fig. 2c). We then asked if PTX could also affect the pro-metastatic capacity of EVs released from an intact tumour microenvironment. To this aim, we treated MMTV-PyMT mice with PTX or CREMO (see Fig. 1a above), excised mammary tumours three days after the last dose, and cultured tumour-derived cell suspensions for 48h in order to isolate EVs from conditioned media (Fig. 2d). We assessed the purity and properties of EV preparations by TEM (Supplementary Fig. 1d), NTA (Supplementary Fig. 1e-f), and Western blotting analysis (Supplementary Fig. 1g), and found no obvious differences between PTX-EV and CREMO-EV. We then pre-conditioned immunocompetent FVB/n mice with either EV preparation (Fig. 2b), followed by the intravenous injection of freshly isolated MMTV-PyMT tumour-derived cells. Similar to results obtained with 4T1 cells, PTX-EV increased the number of metastatic nodules in this MMTV-PyMT-based lung colonization assay (Fig. 2e).

Anthracyclines are broadly employed in neoadjuvant breast cancer therapy ^1,5,6. DOX, a lead anthracycline, was reported previously to modify the microenvironment of primary mammary tumours to potentially facilitate metastasis ^8,33. We treated MMTV-PyMT mice with DOX or vehicle (PBS) and purified tumour-derived DOX-EV and PBS-EV, respectively, according to the method shown in Fig. 2d. Similar to results obtained with PTX-EV, DOX-EV enhanced metastasis, compared to PBS-EV, in an MMTV-PyMT-based lung colonization assay (Fig. 2f). Of note, neither PTX nor DOX directly increased metastasis in this assay; rather, the free drugs limited, or even abated, lung colonization by MMTV-PyMT tumour-derived cells (Fig. 2g). These results indicate that chemotherapy-elicited, tumour-derived EVs mediate the pro-metastatic activity of the cytotoxic drugs.

PTX-elicited EVs facilitate tumour colonization in a zebrafish embryo model

Because tumour-derived EVs cannot be isolated from mouse plasma in amounts sufficient to perform EV transfer experiments in mice, we examined the pro-metastatic capacity of circulating PTX-EV in a zebrafish embryo model. In this system, cancer cells injected into the circulation extravasate to, and colonize the caudal haematopoietic area (CHA) of the tail of embryos; these events can be imaged by confocal microscopy, as embryos are translucent ^34,35. We found that PTX-EV isolated from plasma of 4T1-mCh tumour-bearing Rag1^−/− mice (Fig. 2h) enhanced CHA colonization by human MDA-MB-435 melanoma cells stably expressing cyan fluorescent protein (CFP), compared to CREMO-EV (Fig. 2i-l). Of note, free PTX did not facilitate CHA colonization by cancer cells in the same assay.

PTX modulates EV release from cancer cells

We then examined the effects of chemotherapy on EV release from cancer cells. PTX increased the release of EVs from both mouse and human mammary carcinoma cell lines (Supplementary Fig. 2a-c). It also increased mCh fluorescence in the blood of mice carrying 4T1-mCh tumours (Supplementary Fig. 2d-f). In these experiments, mCh fluorescence was associated with bona fide EVs (Supplementary Fig. 2g-i) and was resistant to proteolysis (Supplementary Fig. 2j-k), indicating intra-vesicular localization of the CD9-mCh fusion protein.

Cytotoxic agents, including PTX, promote cancer cell death in a dose-dependent manner (Supplementary Fig. 2a and Supplementary Fig. 3a-b). Although drug-induced cell stress may influence release of EVs and other microvesicles from pre-apoptotic cells ³², PTX-induced EV release from 4T1 cells was apoptosis-independent (Supplementary Fig. 3c).

RAB GTPases control EV biogenesis and release in mammalian cells ^36,37. Accordingly, both basal and PTX-induced EV release were reduced in Rab27a-deficient 4T1 cells (Supplementary Fig. 3d-g). PTX may enhance EV release from cancer cells by enforcing the trafficking of intracellular vesicles to plasma membranes (Supplementary Fig. 3h); this process may involve the association of RAB27A with microtubules, which are stabilized by taxanes ³⁸. We observed enhanced EV release also from 4T1 cells exposed to a different taxane, docetaxel (Supplementary Fig. 3i). Conversely, DOX did not promote EV release from mammary carcinoma cells (Supplementary Fig. 3j, k). Therefore, whereas taxanes may increase mammary tumour metastasis by enhancing both the release and pro-metastatic properties of tumour-derived EVs, anthracyclines may do so by specifically reinforcing the pro-metastatic features of the EVs.

Chemotherapy-elicited EVs are enriched in ANXA6

Enhanced pro-metastatic capacity of chemotherapy-elicited EVs may depend on their protein repertoire. We performed proteomic analysis of 4T1-derived PTX-EV and CREMO-EV using liquid chromatography-tandem mass spectrometry (LC-MS/MS). Unsupervised clustering analysis of the EV proteomes (Fig. 3a) revealed substantial differences (Supplementary Table 1). Over-represented proteins in PTX-EVs included annexin-A6 (ANXA6; Fig. 3b), a Ca²⁺-binding membrane-associated protein (Fig. 3c). ANXA6 controls membrane trafficking and cell signalling ³⁹, and has been previously implicated in both the positive and negative regulation of cancer cell invasion ^40–42. Validation experiments by Western blotting revealed that ANXA6 was more robustly detected in PTX-EV compared to the producer 4T1 cells or CREMO-EV (Fig. 3d). Notably, both PTX and DOX promoted ANXA6 loading into EVs of 4T1 cells and the MMTV-PyMT tumour-derived cell line PyMT-IK1 (Fig. 3e).

Figure 3. PTX enriches ANXA6 in EVs in a Ca²⁺-dependent manner.

a. Unsupervised clustering of proteins in CREMO-EV and PTX-EV (n=6 independent EV preparations/condition) from 4T1 cells, determined by LC-MS/MS analysis.

b. LC-MS/MS analysis of CREMO-EV and PTX-EV (n=6 independent EV preparations) showing total spectrum count (mean ± s.d.) of ANXA6, CD9 and CD81. Statistical analysis by two-way ANOVA with Sidak’s multiple comparison test.

c. Western blotting analysis of ANXA6, calnexin (CANX), GAPDH and RAB7 in 4T1 cells treated with CREMO or PTX for 24h before subcellular fractionation. The “membranes” fraction encompasses early and late endosomes, endoplasmic reticulum and mitochondria. The experiment was performed once.

d. Western blotting analysis of ANXA6 and CD9 in CREMO- or PTX-treated 4T1 cells, or matched CREMO-EV or PTX-EV. Additional experiments are shown in Fig. 3e, f, h; Fig. 4c; and Supplementary Fig. 1b.

e. Western blotting analysis of ANXA6 and CD81 in the indicated EV preparations isolated from either 4T1 or PyMT-IK1 cells.

f. Western blotting analysis of the indicated proteins in PyMT-IK1 cells and matched EVs 48h after treatment of the cells with PBS, CREMO, PTX, DMSO or DOX, with or without the calcium chelator BAPTA-AM. One representative experiment is shown of three performed for EVs and one for cells.

g. ANXA6 band intensity (mean ± s.d.; n=3 independent experiments, one of which is shown in (f) above) in the indicated EV preparations analysed by Western blotting. Statistical analysis as in (b).

h. Western blotting analysis of EVs from 4T1 cells treated for 48h with PBS, CREMO or PTX, with or without BAPTA-AM. One representative experiment is shown of three performed.

i. ANXA6 band intensity (normalized to CD81; mean ± s.d.; n=3 independent experiments, one of which is shown in (h) above) in the indicated EV preparations. Statistical analysis as in (b).

j. Protein content by BCA (left panel) and concentration by NTA (right panel) of the indicated EVs (mean ± s.d.; n=3 independent EV preparation/condition) obtained from 4T1 cells treated with or without BAPTA-AM. Statistical analysis as in (b).

Source data are shown in Supplementary Table 5. Unprocessed blots are shown in Supplementary Fig. 9.

Both taxanes ⁴³ and anthracyclines ⁴⁴ augment intracellular calcium ion levels. Interestingly, chemotherapy-induced ANXA6 loading into EVs was Ca²⁺-dependent (Fig. 3f-i), whereas EV release per se was Ca²⁺-independent (Fig. 3j). Western blotting and density gradient fractionation of EVs ³⁰ purified from medium conditioned by 4T1 or 4T1-mCh cells confirmed that ANXA6 was associated with bona fide EVs (Supplementary Fig. 1b above and Supplementary Fig. 4a-b) and, in particular, with the inner leaflet of EVs (Supplementary Fig. 4c-d). Neither PTX nor DOX increased ANXA6 in EVs released by non-transformed cells, such as mouse primary bone marrow dendritic cells and embryonic fibroblasts (Supplementary Fig. 4e-h).

EV-associated ANXA6 promotes mammary tumour metastasis

The aforementioned results suggested a potential role for ANXA6 in modulating chemotherapy-induced mammary tumour metastasis. To test this possibility, we generated Anxa6 knockout (KO) 4T1-mCh cells using CRISPR/Cas9 technology (Fig. 4a). We obtained two Anxa6-deficient clones (Fig. 4b), which were used in subsequent experiments. ANXA6 was exhaustively depleted from Anxa6-KO 4T1-mCh-derived EVs treated with either CREMO or PTX (Fig. 4c). ANXA6 deficiency in 4T1-mCh cells did not impair the enhancement of EV release by PTX (Fig. 4d); furthermore, Anxa6-proficient (WT) and KO EVs were indistinguishable based on TEM (Fig. 4e) and NTA (Fig. 4f-g).

Figure 4. EV-associated ANXA6 promotes mammary tumour metastasis.

a. Schematics of the lentiviral vectors used to disrupt the expression of Anxa6, Rab27a or Rela in cells.

b, c. Western blotting analysis of ANXA6, GAPDH and CD81 in 4T1-mCh cells (b) or secreted EVs (c). Anxa6-WT (parental line) and two independent Anxa6-KO clones are shown, either untreated (b) or treated as indicated (c). The experiments were performed once.

d. mCh mean fluorescence intensity (MFI; left y axis) and concentration by NTA (right y axis) of EVs (mean ± s.d.; n=3 independent cell cultures/condition) released by Anxa6-KO 4T1-mCh cells (clone #1) treated for 72h with CREMO or PTX. Statistical analysis by unpaired two-tailed Student’s t-test.

e. Representative wide-field TEM images of the indicated EVs isolated from Anxa6-WT (top panels) or Anxa6-KO (clone #1, bottom panels) 4T1-mCh cells. Scale bars, 200 nm. The experiment was performed once.

f. Mode size (mean values ± s.d.) of the indicated EVs isolated from Anxa6-WT (top panel) or Anxa6-KO (clone #1, bottom panel) 4T1-mCh cells, determined by NTA. Data show results from n=5 (Anxa6-WT) and n=3 (Anxa6-KO) independent EV preparations.

g. Concentration (mean ± s.e.m.; n=3 acquisitions of one sample/condition) and size distribution of EVs isolated from medium conditioned by CREMO- or PTX-treated Anxa6-KO 4T1-mCh cells (clone #1), determined by NTA.

h. mCh⁺ metastatic area fraction per lung section (mean ± s.e.m.) in Rag1^-/- mice pre-conditioned with the indicated EVs. CREMO-EV/Anxa6-WT, n=7 mice; PTX-EV/Anxa6-WT, n=7; CREMO-EV/Anxa6-KO (clone #1), n=7; PTX-EV/Anxa6-KO (clone #1), n=7. Statistical analysis by two-way ANOVA with Tukey’s multiple comparison test. Representative images of lung sections stained with DAPI (white) are shown on the right; the mCh signal (red) was acquired as direct fluorescence. Scale bars, 1 mm.

i. mCh⁺ metastatic area fraction per lung section (mean ± s.e.m.) in Rag1^-/- mice pre-conditioned with EVs. DMSO-EV/Anxa6-WT, n=7 mice; DOX-EV/Anxa6-WT, n=8; DMSO-EV/Anxa6-KO (clone #1), n=8; DOX-EV/Anxa6-KO (clone #1), n=8. Statistical analysis as in (h).

Source data are shown in Supplementary Table 5. Unprocessed blots are shown in Supplementary Fig. 9.

We then isolated Anxa6-WT and KO EVs from 4T1-mCh cells treated in vitro with either PTX or DOX (or the appropriate vehicle) to pre-condition Rag1^−/− mice before the intravenous injection of 4T1-mCh cells, according to a lung colonization assay (see Fig. 2b above). Remarkably, ANXA6 deficiency in PTX-EV or DOX-EV disrupted their capacity to facilitate lung colonization by 4T1-mCh cells (Fig. 4h-i). The depletion of ANXA6 from 4T1 cells (Supplementary Fig. 4i) did not alter loading of PTX or DOX into EVs (Supplementary Fig. 4j), arguing against the possibility that ANXA6-deficient EVs had compromised pro-metastatic capacity owing to changes in intravesicular drug levels, which were negligible irrespective of ANXA6 status (Supplementary Fig. 4k).

PTX induces CCL2 expression and Ly6C⁺ monocyte expansion in lungs of mammary tumour-bearing mice

We analysed expression of cytokines known to facilitate mammary tumour metastasis, such as CCL2, CXCL12, CSF1 and ANGPT2 ⁴⁵, in lungs of tumour-free and 4T1-mCh tumour-bearing Rag1^−/− mice treated with either PTX or CREMO. PTX significantly increased Ccl2 expression both at the mRNA (Fig. 5a) and protein (Fig. 5b) level in lungs of 4T1-mCh tumour-bearing but not tumour-free mice.

Figure 5. PTX induces CCL2 expression and Ly6C⁺ monocyte expansion in lungs of mammary tumour-bearing mice.

a. Quantitative polymerase chain reaction (qPCR) analysis of the indicated genes (mean ± s.e.m.) in lungs of either tumour-free (TF) or 4T1 tumour-bearing (TB) mice treated as indicated. TF + CREMO, n=5 mice; TF + PTX, n=5; TB + CREMO, n=8; TB + PTX, n=9. Statistical analysis by two-way ANOVA with Tukey’s multiple comparison test.

b. ELISA-based CCL2 protein (mean ± s.e.m.) in lungs of mice treated as in (a). Tumour-free + CREMO, n=5 mice; Tumour-free + PTX, n=5; Tumour-bearing + CREMO, n=6; Tumour-bearing + PTX, n=6. Statistical analysis as in (a).

c. FACS analysis of CD45⁺CD11b⁺Ly6C⁺Ly6G^–F4/80⁺ monocytes (Ly6C⁺ Mo) in lungs of tumour-bearing mice. A representative sample is shown to illustrate the gating strategy. FMO, fluorescence minus one.

d-g. FACS analysis of Ly6C⁺ Mo (mean ± s.e.m.) in tumour-bearing mice treated as indicated. Data show the frequency of Ly6C⁺ monocytes in the CD45⁺ population, relative to control (CREMO or PBS). (d): CREMO, n=7 mice; PTX, n=8. (e): PBS, n=5; CREMO, n=5; PTX, n=5. (f): PBS, n=7; CREMO, n=8; PTX, n=9. (g): CREMO, n=13; PTX, n=14. Statistical analysis by unpaired two-tailed Student’s t-test (d, g) or one-way ANOVA with Tukey’s multiple comparison test (e, f). Data in (g) show three independent experiments combined.

h, i. FACS analysis of Ly6C⁺ Mo (mean ± s.e.m.) in lungs of tumour-free FVB/n (h) or Rag1^-/- (i) mice treated as indicated and analysed two days after treatment. (h): PBS, n=12 mice; CREMO, n=6; PTX 1 mg/kg, n=8; PTX 10 mg/kg, n=6; DOX, n=6. (i): CREMO, n=5; PTX, n=5. Statistical analysis by one-way ANOVA with Tukey’s multiple comparison test (h) or unpaired two-tailed Student’s t-test (i).

Source data are shown in Supplementary Table 5.

CCL2 was reported previously to enhance Ly6C⁺CCR2⁺ monocyte-assisted metastasis of mammary tumours ^46–49. Consistent with upregulation of CCL2, PTX increased the relative abundance of Ly6C⁺ monocytes in lungs of 4T1-mCh tumour-bearing Rag1^−/− mice, 4T1 tumour-bearing Balb/c mice, 4T1-mCh/HER2 tumour-bearing Swiss nu/nu mice, and transgenic MMTV-PyMT mice, compared to vehicle (Fig. 5c-g). Conversely, PTX did not alter the relative abundance of Ly6C⁺ monocytes in the lungs of tumour-free mice (Fig. 5h-i). These results indicate that signals emanating from primary tumours are required for PTX-induced upregulation of CCL2 and expansion of Ly6C⁺ monocytes in lungs.

Chemotherapy-elicited EVs induce pulmonary CCL2 expression and Ly6C⁺ monocyte expansion

We examined whether the effects of PTX on CCL2 and Ly6C⁺ monocytes were dependent on EVs. To this aim, we performed pre-conditioning studies with EVs in tumour-free mice (Fig. 6a). PTX-EV isolated from 4T1 or 4T1-mCh cells increased lung Ccl2 expression in Balb/c and Rag1^−/− tumour-free mice, respectively (Fig. 6b, c). We obtained similar results when we pre-conditioned tumour-free FVB/n mice with PTX-EV isolated ex vivo from MMTV-PyMT tumour-derived cells (see Fig. 2d above, and Fig. 6d). Paralleling Ccl2 expression, the relative abundance of lung Ly6C⁺CCR2⁺ monocytes increased in response to PTX-EV (Fig. 6e-i). Notably, both Ccl2 expression and Ly6C⁺ monocytes were also increased in response to DOX-EV isolated ex vivo from MMTV-PyMT tumour-derived cells (Fig. 6j-k).

Figure 6. Ly6C⁺ monocytes mediate the pro-metastatic activity of chemotherapy-elicited EVs.

a. Schematics of EV pre-conditioning studies in tumour-free mice.

b-d. qPCR analysis of Ccl2 (mean ± s.e.m.) in lungs of tumour-free mice that received EVs. (b): CREMO-EV, n=8; PTX-EV, n=9. (c): CREMO-EV, n=8; PTX-EV, n=7. (d): CREMO-EV, n=10; PTX-EV, n=9. Statistical analysis by unpaired two-tailed Student’s t-test.

e-g. FACS analysis of Ly6C⁺ Mo (mean ± s.e.m.) in lungs of tumour-free mice that received EVs. (e): CREMO-EV, n=8; PTX-EV, n=9. (f): CREMO-EV, n=6; PTX-EV, n=7. (g): PBS (no EVs), n=11; CREMO-EV, n=14; PTX-EV, n=15. Statistical analysis by unpaired two-tailed Student’s t-test (e, f) or one-way ANOVA with Tukey’s multiple comparison test (g).

h. qPCR analysis of Ccr2 (mean ± s.e.m.) in lungs of mice that received MMTV-PyMT tumour-derived CREMO-EV (n=10) or PTX-EV (n=9). Statistical analysis as in (b).

i. Correlation between Ccl2 and Ccr2 transcript levels in lungs of FVB/n mice shown in (d) and (h). The Pearson correlation coefficient (r) is indicated.

j, k. qPCR analysis of Ccl2 (j) and FACS analysis of Ly6C⁺ Mo (k) in the lungs of tumour-free mice (mean ± s.e.m.) that received PBS-EV (n=5) or DOX-EV (n=5). Statistical analysis as in (b).

l. Bioluminescence (BL) analysis (total photon flux; mean ± s.e.m.) of Ccr2 WT or KO mice pre-conditioned with EVs, analysed at day 10 post-cell injection. The right panel shows representative mice. CREMO-EV/Ccr2-WT, n=10; PTX-EV/Ccr2-WT, n=11; CREMO-EV/Ccr2-KO, n=7; PTX-EV/Ccr2-KO, n=8. Statistical analysis by two-way ANOVA with Tukey’s multiple comparison test. Data show three independent experiments combined.

m. qPCR analysis of Ccr2 (mean ± s.e.m.) in myeloid cells (Ly6C⁺ Mo; Ly6C^low Mo; and non-alveolar macrophages, Mac) FACS-sorted from lungs of tumour-free mice that received MMTV-PyMT tumour-derived CREMO-EV (n=5; except for Mac, n=4) or PTX-EV (n=6). Statistical analysis as in (l).

n, o. qPCR analysis of Ccl2 and Ccr2 (n) and FACS analysis of Ly6C⁺ Mo (o) in lungs of mice (mean ± s.e.m; n=6 mice) that received CREMO-EV or PTX-EV from either Anxa6-WT or Anxa6-KO 4T1-mCh cells. Statistical analysis by two-way ANOVA with Sidak’s multiple comparison test.

Source data are shown in Supplementary Table 5.

The pro-metastatic capacity of chemotherapy-elicited EVs is dependent on Ly6C⁺CCR2⁺ monocytes and ANXA6

To explore the functional involvement of Ly6C⁺CCR2⁺ monocytes in PTX-EV-mediated pulmonary metastasis, we performed pre-conditioning experiments with EVs in Ccr2 KO mice (C57Bl/6 background), which have impaired monocyte-assisted mammary cancer metastasis ⁴⁹. Whereas PTX-EV derived from E0771-LG mouse mammary carcinoma cells enhanced pulmonary seeding of firefly luciferase-expressing E0771-LG:Fl cells in Ccr2 WT mice, they failed to do so in Ccr2 KO mice (Fig. 6l). Interestingly, Ccr2 was upregulated in Ly6C⁺ monocytes, but not other myeloid cells, that were sorted from the lungs of tumour-free FVB/n mice pre-conditioned with PTX-EV (Fig. 6m). These data strongly argue that Ly6C⁺CCR2⁺ monocytes mediate EV-induced, pulmonary mammary cancer metastasis.

Because Anxa6 KO PTX-EV failed to enhance pulmonary colonization by 4T1-mCh cells (see Fig. 4h-i above), we asked whether PTX-EV-mediated effects on pulmonary CCL2 expression and Ly6C⁺ monocyte expansion were also dependent on ANXA6 in EVs. ANXA6 deficiency in PTX-EV isolated from 4T1-mCh cells prevented Ccl2 and Ccr2 upregulation (Fig. 6n) and the increase of Ly6C⁺ monocytes (Fig. 6o) in the lungs of Rag1^−/− mice. Together, these results indicate that chemotherapy-induced enrichment of ANXA6 in mammary tumour-derived EVs supports a pro-metastatic cascade in the lung that is dependent on local Ly6C⁺ monocyte expansion.

Chemotherapy-elicited EVs promote inflammatory EC activation through ANXA6 transfer

The internalization of breast cancer-derived EVs into lung endothelial cells (ECs) may influence the initial steps of pulmonary metastasis by altering the biology of those cells in the pre-metastatic niche ^24,50. We observed increased proportions of lung ECs displaying mCh fluorescence after treating 4T1-mCh tumour-bearing Rag1^−/− or 4T1-mCh/HER2 tumour-bearing Swiss nu/nu mice with PTX, compared to CREMO (Fig. 7a-c).

Figure 7. Chemotherapy-elicited EVs promote inflammatory EC activation through ANXA6 transfer.

a. FACS of mCh⁺ ECs (mean ± s.e.m.) in lungs of 4T1 (n=6) or 4T1-mCh (CREMO, n=8; PTX, n=9) tumour-bearing Rag1^−/− mice. Statistical analysis by one-way ANOVA with Tukey’s multiple comparison test. Right panels show gating strategy.

b. Representative confocal images of anti-CD31 endothelial (green) and anti-mCh (magenta) immunostaining of lung sections from 4T1-mCh tumour-bearing mice treated as in (a); nuclei are stained with DAPI (blue). Scale bars, 10 μm.

c. FACS of mCh⁺ ECs (mean ± s.e.m.) in lungs of 4T1 (n=4) or 4T1-mCh/HER2 (PBS, n=7; CREMO, n=8; PTX, n=9) tumour-bearing Swiss nu/nu mice. Statistical analysis as in (a).

d. Western blotting analysis of the indicated proteins in Anxa6-KO bEnd.3 cells. The experiment was performed twice; Supplementary Fig. 6b shows a replicate experiment.

e. Duolink staining of Anxa6-KO bEnd.3 cells showing ANXA6/p65 proximity (number of white dots/cell; mean ± s.e.m.; n=8 randomly selected images, each containing at least 12 cells). Statistical analysis as in (a). Right panels show representative images; nuclei are stained with DAPI (blue). Scale bars, 30 μm. Data show one experiment of two performed.

f. Western blotting analysis of p65 (left), NF-kB activity (middle), and qPCR of Ccl2 (right) in bEnd.3 cells (mean ± s.d.; n=3 independent cell cultures/condition). Statistical analysis by unpaired two-tailed Student’s t-test.

g, h. NF-kB activity (g) and qPCR of Ccl2 (h) in bEnd.3 cells (mean ± s.d.; n=3 independent cell cultures/condition). Statistical analysis by two-way ANOVA with Sidak’s multiple comparison test. Data show one experiment of two (g) or three (h) performed.

i, j. qPCR analysis of Ccl2 (i) and NF-kB activity (j) in bEnd.3 cells (mean ± s.d.; n=3 independent cell cultures/condition). Statistical analysis by two-way ANOVA with Tukey’s multiple comparison test (i, left panel) or unpaired two-tailed Student’s t-test (i, right panel; and j).

k. qPCR of Ccl2 in mCh⁺CD31⁺CD45^– ECs sorted from lungs of FVB/n mice (mean ± s.e.m.; n=5 mice). Statistical analysis by unpaired one-tailed Student’s t-test. Right panel shows the purity of the sorted ECs.

Source data are shown in Supplementary Table 5. Unprocessed blots are shown in Supplementary Fig. 9.

Increased mCh fluorescence in lung ECs may denote internalization of tumour-derived EVs, a process that can occur through EV-cell fusion ⁵¹. We then used a murine EC line, bEnd.3, to study effects of EVs on ECs. bEnd.3 cells internalized mCh⁺ EVs in a dose-dependent manner (Supplementary Fig. 5a, b); interestingly, bEnd.3 cells exposed to medium conditioned by PTX-treated 4T1-mCh or PyMT-IK1-mCh cancer cells displayed higher mCh fluorescence than bEnd.3 cells exposed to medium conditioned by CREMO-treated cells (Supplementary Fig. 5c). Moreover, PTX-EV from either 4T1-mCh or PyMT-IK1-mCh cells were internalized by bEnd.3 cells more efficiently than matched amounts of CREMO-EV (Supplementary Fig. 5d). Finally, uptake of EVs by bEnd.3 cells was independent of ANXA6 (Supplementary Fig. 5e), suggesting that other molecules (Supplementary Table 1), such as integrins ³⁸, could mediate the preferential uptake of PTX-EV.

PTX-EV isolated from either 4T1-mCh or PyMT-IK1-mCh cells transferred ANXA6 (and mCh) to bEnd.3 cells that were made Anxa6-deficient using CRISPR/Cas9 (Fig. 7d and Supplementary Fig. 6a-b). Duolink proximity ligation assay, which enables imaging of protein pairs that co-localize within 40 nm, indicated that PTX-EV-mediated transfer of ANXA6 to Anxa6 KO bEnd.3 cells was conducive to ANXA6 co-localization with the NF-kB subunit p65 (Fig. 7e and Supplementary Fig. 6c-d), consistent with the ability of ANXA6 to interact with p65 (Ref ⁵²). Of note, matched doses of CREMO-EV did not reveal detectable ANXA6/p65 proximity in the same assay, possibly because PTX-EV fuse more efficiently than CREMO-EV with bEnd.3 cells (see Supplementary Fig. 5c-e above) and contain higher amounts of ANXA6 than CREMO-EV (see Fig. 3d-j above).

NF-kB (p65/Rela) activates Ccl2 transcription ⁵³. In order to gain insight into a potential mechanistic link between ANXA6 transfer to ECs and Ccl2 upregulation, we generated Rela (p65) KO bEnd.3 cells, which had defective NF-kB activation and Ccl2 expression (Fig. 7f and Supplementary Fig. 6e). PTX-EV isolated from 4T1-mCh cells failed to induce NF-kB activation (Fig. 7g) and Ccl2 upregulation (Fig. 7h) in Rela KO bEnd.3 cells. Conversely, PTX-EV isolated from 4T1-mCh or PyMT-IK1 cells increased NF-kB activity and Ccl2 expression in Rela-proficient bEnd.3 cells; notably, these responses required ANXA6 in EVs. Accordingly, Anxa6 KO bEnd.3 cells had defective NF-kB activation and Ccl2 transcription in response to tumour-necrosis factor (TNF) compared to Anxa6 WT cells; of note, neither Csf1 nor Cxcl12 were regulated by ANXA6 (Supplementary Fig. 6f-i). Similar to PTX-EV, DOX-EV from 4T1 cells also increased Ccl2 transcript levels (Fig. 7i) and NF-kB activity (Fig. 7j) in bEnd.3 cells. Therefore, EV-associated ANXA6 triggers an NF-kB-dependent pro-inflammatory response in ECs.

We then examined whether PTX-EV could activate Ccl2 transcription in lung ECs, which are a source of CCL2 in mammary tumour-bearing mice ⁵⁴. We used fluorescence-activated cell sorting (FACS) to isolate lung ECs of tumour-free FVB/n mice pre-conditioned with MMTV-PyMT tumour-derived PTX-EV or CREMO-EV (see Fig. 2d above) and found enhanced Ccl2 expression in lung ECs exposed to PTX-EV (Fig. 7k). Taken together, these findings suggest that chemotherapy-elicited tumour EVs promote pro-inflammatory EC activation, CCL2 upregulation, Ly6C⁺ monocyte accumulation, and tumour colonization at metastatic sites through a mechanism involving horizontal transfer of EV-associated ANXA6 to the pulmonary endothelium.

Chemotherapy-elicited EVs display broad cellular tropism in tumour-bearing mice

Besides ECs, other lung-resident cells may mediate metastasis in response to chemotherapy-induced EVs. Indeed, CD11b⁺Gr1^– lung macrophages/monocytes also internalized mCh⁺ EVs in PTX-treated, 4T1-mCh Rag1^−/− or 4T1-mCh/HER2 Swiss nu/nu tumour-bearing mice (Supplementary Fig. 7a-b). Moreover, PTX-EV isolated from MMTV-PyMT tumour-derived cells upregulated interleukin-6 (Il6) expression in the lung (Supplementary Fig. 7c) and, specifically, lung-associated non-alveolar macrophages (Supplementary Fig. 7d) of tumour-free FVB/n mice; of note, IL-6 was previously implicated in chemotherapy resistance ⁵⁵ and macrophage-assisted mammary tumour metastasis ^47,56.

Finally, PTX-EV transferred mCh to liver sinusoidal ECs more efficiently than CREMO-EV (Supplementary Fig. 7e, f), suggesting that chemotherapy-elicited EVs might also promote breast cancer metastasis to the liver. However, the rapid growth kinetics of both primary tumours and pulmonary metastasis in the breast cancer models employed in our study complicate the assessment of slow-growing liver metastases.

ANXA6 is detected in circulating EVs of breast cancer patients undergoing neoadjuvant chemotherapy

We finally investigated effects of chemotherapy on EV-associated ANXA6 in human breast cancer cells and patients with breast cancer. Both PTX and DOX increased ANXA6 protein levels in EVs released by human MDA-MB-231 breast cancer cells (Fig. 8a). Furthermore, EVs purified from the plasma of breast cancer patients undergoing neoadjuvant chemotherapy (Supplementary Table 2 and Fig. 8b) had increased ANXA6 content by LC-MS/MS in five out of six cases, compared to pre-treatment levels (Fig. 8c, Supplementary Fig. 8 and Supplementary Table 3). Because we could isolate limited amounts of EVs from each sample, we were only able to verify the LC-MS/MS data by Western blotting in one case (patient #56; Fig. 8d). Notably, ANXA6 levels in the EVs decreased at the end of neoadjuvant therapy in the patients who achieved a partial or complete response (five out of six), probably reflecting shrinkage of the tumour in response to chemotherapy. Also, the finding that EV-associated ANXA6 levels increased on-therapy in one patient with progressive disease (#52) strongly suggested that EV-associated ANXA6 was, in fact, of cancer cell origin. While limited, these data support the notion that chemotherapy augments ANXA6 levels in circulating EVs of patients with breast cancer.

Figure 8. ANXA6 is detected in circulating EVs of breast cancer patients undergoing neoadjuvant chemotherapy.

a. Western blotting analysis of EVs isolated from MDA-MB-231 cells treated as indicated. The experiment was performed three times for PTX and once for DOX.

b. Schematic of the treatment timeline and time-points of blood collection in breast cancer patients (n=6). AC, anthracycline (DOX) plus cyclophosphamide.

c. LC-MS/MS-based quantification of ANXA6 in EVs isolated from plasma of breast cancer patients (n=6) before chemotherapy (pre-treatment), after AC, and after PTX. The data show quantitative values of ANXA6 presented as fold-change versus pre-treatment level. Note that the amount of EVs that could be isolated from patient #38 after PTX was insufficient to perform LC-MS/MS analysis. Tumour response was assessed at the time of surgery.

d. Western blotting analysis of ANXA6 in plasma EVs of one patient (#56), analysed at the indicated time-points. The experiment was performed once.

Source data are shown in Supplementary Table 5. Unprocessed blots are shown in Supplementary Fig. 9.

Discussion

Neoadjuvant chemotherapy improves the management of invasive breast cancer by inducing pathological complete responses associated with significantly reduced risk of recurrence in a fraction of the patients ^1–6. However, therapy-induced host responses ⁷, such as those described in our study, may limit the benefits of pre-operatory treatments in some patients. It is tempting to speculate that human primary breast tumours that fail to readily regress on neoadjuvant chemotherapy (a condition of partial or complete chemoresistance) may also release pro-metastatic EVs that, in turn, facilitate seeding, survival or early colonization of metastatic niches by chemoresistant cancer cells. However, it should be cautioned that we did not study mouse survival in association with the various treatments, so we currently ignore whether increased metastatic seeding and outgrowth in response to chemotherapy-elicited EVs would translate into shorter survival in our experimental cancer models.

The potential participation of tumour-derived EVs in the process of human breast cancer metastasis is currently poorly understood and may be challenging to explore, given the lack of validated breast-cancer-specific EV markers ⁵⁷. In the future, chemotherapy-elicited EVs might provide biomarkers for predicting metastasis risk associated with neoadjuvant chemotherapy in patients who do not achieve a complete response. Interestingly, ANXA6 has been detected in EVs isolated from the MDA-MB-231-derived clone 4175, which has lung tropism ²⁰. Furthermore, EV-associated ANXA6 has been implicated in the progression of pancreatic cancer in a mouse model ⁴². The significance of EV-associated ANXA6 for breast cancer metastasis may, therefore, merit further investigation.

Online Methods

Mice

FVB/n, Balb/c, C57BL/6 and Swiss nu/nu mice were purchased from Charles River Laboratories. Ccr2-KO mice (C57BL/6 background) were obtained from The Jackson Laboratory (https://www.jax.org) and bred under standard conditions at the University of Edinburgh (United Kingdom). Rag1^−/− mice (C57BL/6 background) were available in the EPFL mouse facility (Switzerland). Transgenic MMTV-PyMT mice (FVB/n background) were available in both EPFL mouse facility and Department of Comparative Medicine barrier facility (Oregon). Male mice heterozygous for the Polyoma virus middle TAg (PyMT) were bred with wild-type females; pups were genotyped by Transnetyx (http://www.transnetyx.com). All mice employed in this study were maintained in pathogen-free barrier animal facilities in accord with State regulations for the care and use of mice in experimental research.

The studies involving mice were compliant with all relevant ethical regulations regarding animal research. Most of the experiments were performed at EPFL; some experiments were performed at the University of Edinburgh (Ccr2-KO mice) and at Oregon Health & Science University (OHSU), Portland (MMTV-PyMT mice, partim). All procedures were performed according to protocols approved by the Veterinary Authorities of the Canton Vaud according to the Swiss Law (licenses 2577, 2577.1 and 2916); the UK Animal Scientific Procedures Act (1986) license (PPL 70/8065); and the Public Health Service of USA (PHS), the Office for the Protection from Research Risks (OPRR) A-3304-01, the United States Department of Agriculture (USDA) 92-R-001, and the Association for Assessment and Accreditation of Laboratory Animal Care (AAALAC) International. OHSU maintains AAALAC International accreditation and OLAW assurance (A3304-01), and all procedures with small animals at OHSU were IACUC approved (IS00001484).

Cell lines

The mouse cell lines 4T1 (mammary adenocarcinoma; from ATCC), bEnd.3 (brain endothelial cells; provided by F. Bussolino, University of Turin, Italy), 293T (provided by L. Naldini, San Raffaele Scientific Institute, Milan, Italy), and PyMT-IK1 (mammary adenocarcinoma; isolated in the De Palma’s lab) were maintained in Iscove's modified Dulbecco's medium (IMDM, Sigma) supplemented with 10% fetal bovine serum (FBS; Gibco), L-glutamine (Amimed) and penicillin/streptomycin (Gibco). The PyMT-IK1 cell line was isolated from a primary mammary tumour of a 14-week old, transgenic MMTV-PyMT female mouse. Mouse E0771-LG cells (mammary adenocarcinoma; derived from the parental cell line E0771), either unmodified or stably transfected with a firefly luciferase cDNA for in vivo bioluminescence imaging (E0771-LG:Fl; both provided by T. Kitamura, University of Edinburgh, United Kingdom) ^49,58, were maintained in Dulbecco’s medium (DMEM, Gibco) with 10% FBS, L-glutamine and penicillin/streptomycin. The human cell line MDA-MB-231 (breast adenocarcinoma; from ATCC) was cultured in RPMI 1640 (Gibco) with 10% FBS and penicillin/streptomycin. The human melanoma cell line MDA-MB-435 stably transfected to express cyan fluorescent protein (CFP; provided by R. Klemke, University of Sheffield, United Kingdom) was maintained in DMEM with 10% fetal calf serum (FCS; BioWhittaker, United Kingdom) and 50 µg/ml G418 (Sigma) to maintain CFP expression. All cell lines used tested negative for Mycoplasma contamination.

Primary cell cultures

Bone marrow dendritic cells (BMDCs) were obtained by flushing long bones of 6-week old C57BL/6 mice, as described previously ²⁹, and cultured for 6 days in RPMI 1640 medium with 10% FBS, L-glutamine, penicillin/streptomycin, and GM-CSF (100 ng/ml, Peprotech).

Mouse embryonic fibroblasts (MEFs) were obtained from C57BL/6 mice on day E14.5, as described previously ⁵⁹, and cultured in DMEM with 10% FBS, L-glutamine, penicillin/streptomycin. We verified the purity of MEFs (CD45^–CD31^–EpCAM^–CD140a⁺) by fluorescence-activated cell sorting (FACS).

In order to obtain MMTV-PyMT tumour-derived cells, primary mammary tumours of 14-week old female MMTV-PyMT mice were dissected, minced and digested with collagenase IV (0.2 mg/ml, Worthington), dispase (2 mg/ml, Life Technologies) and DNaseI (0.002 mg/ml, Life Technologies) in IMDM medium for 30 min at 37°C. The cell suspensions were filtered using a cell strainer (70 μm) before plating. Tumour-derived cells were then cultured for 48h in IMDM supplemented with 5% EV-depleted FBS, L-glutamine, and penicillin/streptomycin before intraveneous injection in tumour-free mice.

Treatment of primary cells with chemotherapeutic drugs was conducted in complete medium supplemented with EV-depleted, 5% FBS for 3 days. At the end of the treatment, the viability of BMDCs and MEFs was assessed by Trypan Blue staining and WST-1 assay, respectively.

Generation, production and use of lentiviral vectors (LVs)

Lentiviral vectors (LVs) expressing a CD9.mCherry fusion protein or human HER2 were described previously ^29,60. In order to disrupt Rab27a, Rela or Anxa6 in bEnd.3 or 4T1 cells, we generated self-inducible CRISPR-Cas9-based LVs, as described previously ²⁹. Briefly, we annealed and ligated the DNA sequences shown below using the BsmBI restriction site in the U6-(sgRNA).TetO-CAS9.hPGK-PURO/2A/rtTA LV (Ref ²⁹):

Rab27a sense: ACCGTATCCAGTTTCGGACATTG

Rab27a antisense: AAACCAATGTCCGAAACTGGATA

Rela sense: ACCGGGCTGGGGTCGGCGTACGG

Rela antisense: AAACCCGTACGCCGACCCCAGCC

Anxa6 sense: ACCGGATTCCAGGTCTCGCTCAT

Anxa6 antisense: AAACATGAGCGAGACCTGGAATC.

To generate the NF-kB reporter LV, we isolated the minimal mouse Fos promoter element and three copies of the major histocompatibility complex (MHC) class I KB element (TGGGGATTCCCCA) by digesting a plasmid obtained from Addgene (https://www.addgene.org/26699/) with HincII and PvuII restriction enzymes. The sequence was then inserted downstream to the minimal CMV promoter (mCMV) of a bidirectional hPGK.GFP/mCMV.mCh LV, after digestion with EcoRV.

Vesicular stomatitis virus (VSV)-pseudotyped, third-generation LVs were produced by transient four-plasmid co-transfection into 293T cells and concentrated by ultracentrifugation, as described previously ⁶¹. Expression titres of LVs encoding fluorescent proteins were determined in 293T cells by limiting dilution. LV stocks were stored at –80°C.

PyMT-IK1 and 4T1 cells were transduced with LVs expressing CD9.mCh; 4T1-mCh cells were further transduced with a HER2-expressing LV for use in some experiments. Cells expressing HER2 and/or mCh were sorted by FACS (Aria II apparatus, BD Biosciences) to establish pure cell populations expressing the transgene(s) of interest.

bEnd.3 and 4T1-mCh cells were transduced with CRISPR-Cas9 LVs to disrupt Rab27a, Rela or Anxa6. After transduction, the cells were cultured in complete IMDM medium with puromycin (5-10 μg/ml, Sigma) and doxycycline (10 μg/ml, Sigma) for 4 days to select transduced cells. The cells were then plated in normal medium to isolate cell clones. Cell colonies showing no expression of RAB27A, RELA or ANXA6 by Western blotting were expanded and employed in further experiments.

Cytotoxic drugs

Paclitaxel (PTX) dissolved in cremophor EL (Labatec) was obtained at stock concentration of 6 mg/ml. PTX was used at 100 ng/ml (cell culture experiments), 10 mg/kg (mouse models) or 100 nM (zebrafish models), unless indicated otherwise. Cremophor EL (CREMO; Sigma) was used as control vehicle for PTX, both in vitro and in vivo. Oregon Green-488-conjugated PTX (Thermo Fisher Scientific) was reconstituted in DMSO and used at a concentration of 100 ng/ml. Doxorubicin and docetaxel (Sigma) were reconstituted in DMSO, while gemcitabine (Sigma) was reconstituted in water. Unless indicated otherwise, doxorubicin, docetaxel and gemcitabine were used in cell culture experiments at 200, 100 and 10 ng/ml, respectively. For mouse studies, doxorubicin (DOX; adriblastine; Pfizer) was dissolved in PBS and used at 8 mg/kg.

Cytotoxic drugs (or control vehicle) were added once to the cell culture medium and the experiments terminated after 24 to 72h. Mice received three doses of either PTX or DOX, as described below.

Treatment of transgenic MMTV-PyMT mice

Female transgenic mice were treated with 10 mg/kg PTX or CREMO in PBS intraperitoneally (EPFL cohort) or intraveneously (OHSU cohort), or 8 mg/kg DOX or PBS intraperitoneally. All mice received 3 drug/vehicle doses at 5-day (PTX) or 7-day (DOX) intervals, starting at 11-12 weeks of age, when multifocal mammary tumours are evident in this model ⁴⁶. The mice were euthanized 3 days after the last dose. Tumour weight measurements were performed at necropsy by dissecting the ten mammary glands.

Treatment of mice carrying subcutaneous tumours

Subcutaneous 4T1 tumours were established by implanting 10⁶ 4T1, 4T1-mCh or 4T1-mCh/HER2 cells in the right flank of 6 to 8-week old Balb/c, Rag1^−/− or Swiss nu/nu mice. Both female and male Rag1^−/− mice were employed; only female Balb/c and Swiss nu/nu mice were employed. Treatments were initiated when the tumour volume reached 100-200 mm³; the mice were euthanized 3 days after the last drug dose. Tumour growth was monitored biweekly and the tumour size determined by caliper measurements. The tumour volume was calculated using the formula: tumour volume [mm³] = (length [mm]) × (width [mm]) × (height [mm]) × 0.5236. Tumour weight measurements were performed at necropsy.

Isolation of EVs from media conditioned by continuous and primary cell cultures

In order to isolate EVs, cells were cultured and expanded in the appropriate medium. Cells at about 40-50% confluence were then moved to medium containing 5% EV-depleted FBS obtained by ultracentrifugation of standard FBS at 134000 x g for 16h at 4°C, followed by filtering through a 0.1 µm vacuum filtration bottle. EVs were isolated from the conditioned medium using sequential ultracentrifugation, as described ^29,30. Briefly, conditioned medium was centrifuged at 500 x g for 5 min; 2000 x g for 10 min; and 4600 x g for 20 min, at 4°C to remove dead cells and debris. The medium was then ultracentrifuged at 134000 x g for 70 min at 4°C using a Beckman ultracentrifuge and SW32Ti rotor. The pellet was washed in 35 ml of PBS and ultracentrifuged again at 134000 x g for 70 min at 4°C. The resulting EV preparation was dissolved in PBS or RIPA buffer (Sigma), depending on the application, and either used immediately or stored at –80°C. The average EV yield obtained from 4T1, PyMT-IK1, MDA-MD-231, E0771-LG, BMDCs and MEFs was, respectively, 7.3, 0.8, 0.8, 1.8, 0.1 and 1 µg of EVs (determined by BCA) per ml of cell culture medium.

Isolation of MMTV-PyMT tumour-derived EVs

After treatment of the mice with cytotoxic drugs, primary MMTV-PyMT tumours were made into single cell suspensions as described above. The tumour-derived cells were cultured for 48h in IMDM supplemented with 5% EV-depleted FBS, L-glutamine and penicillin/streptomycin before EV isolation by ultracentrifugation. The average yield of purified EVs was 50 µg per gram of tumour.

Isolation of EVs from mouse or human plasma

We obtained blood of 4T1-mCh tumour-bearing mice treated with PTX or CREMO by cardiac puncture after deep sedation (150 mg/kg pentobarbital, intraperitoneally). To obtain plasma for EV isolation or mCh quantification, blood was centrifuged at 500 x g for 5 min; 2000 x g for 15 min; and 10000 x g for 20 min at 4°C to remove cells, debris and large vesicles. Human plasma was obtained as described below.

To isolate EVs, plasma was ultracentrifuged at 134000 x g for 70 min at 4°C using a Beckman ultracentrifuge and SW55Ti rotor. The pellet was washed in 4.5 ml of PBS and ultracentrifuged again at 134000 x g for 70 min at 4°C. The resulting EV preparation was dissolved in PBS and either used immediately or stored at –80°C.

Quantification of EVs

EV preparations were analysed for protein content using the BCA Protein Assay Reagent Kit (Thermo Fisher Scientific).

The size and concentration of EVs were determined by Nanoparticle Tracking Analysis (NTA) using a Nanosight NS300 device (Malvern Instruments, United Kingdom). We tested different sample dilutions (1:50 to 1:2000, in PBS) whereby particle concentration was within the optimal range of detection (5 × 10⁷ - 1 × 10⁹ particles/ml). All camera settings were set at the beginning of the session and kept fixed during all acquisitions in a given experiment: camera level, 7-9; camera gain, 10-12; detection threshold, 2-4. For each sample, at least three videos of 60 sec for at least one EV dilution were taken and analysed using the NTA software 3.0 with default settings.

Transmission electron microscopy (TEM) of purified EVs

Purified EVs (5 μg in 15 μl of PBS) were applied to carbon-coated 400 mesh grids (Electron Microscopy Sciences) for 5 min, then washed with PBS and stained with 2% uranyl acetate for 30 sec, as described previously ⁶². Images were obtained using a transmission electron microscope device (Tecnai Spirit, FEI Company).

Sucrose gradient fractionation of EVs

Purified EVs were subjected to continuous sucrose gradient fractionation using established procedures ^31,63. Briefly, twelve stock solutions with decreasing sucrose concentration were layered on top of purified EVs (800 μg, in 100 μl PBS) in a total volume of 12 ml, followed by ultracentrifugation at 100000 x g for 16h at 4°C with a Beckman ultracentrifuge and a SW40Ti rotor. Six fractions of 2 ml each were then collected and transferred to new tubes; the density of each sucrose fraction was measured using a refractometer. Each fraction was washed with 10 ml of PBS and ultracentrifuged at 110000 x g for 70 min at 4°C as above. EVs were resuspended in 100 μl of PBS prior to subsequent analyses.

Quantification of EV-associated mCh fluorescence in cell culture media and mouse plasma

mCh fluorescence was measured in 100 µl of PBS containing purified EVs, or 100 µl of mouse plasma, using a spectrophotometer (TECAN infinite2000 plate-reader, TECAN). The following parameters were used: excitation wavelength, 586 nm; emission wavelength, 625 nm; number of reads, 4.

To degrade free (non-EV-associated) mCh protein, a proteinase K reaction was performed using 40% of plasma and 0.25 mg/ml proteinase K (Promega) in 1 mM guanidinium hydroxide (GuOH)/30mM Tris/5mM CaCl₂, for 2h at 55°C.

Determination of PTX and DOX concentration in EVs

The content of PTX and DOX in purified EVs was assessed by measuring the fluorescence intensity of Oregon Green-488-conjugated PTX (Thermo Fisher Scientific) and the spontaneous fluorescence of DOX (Sigma), respectively, using a spectrophotometer (TECAN infinite2000 plate-reader). We measured several dilutions of the EVs in PBS, as well as equal amounts of drug-free EVs from DMSO-treated cells for background subtraction. The following parameters were utilized for measuring fluorescence of Oregon Green-488-conjugated PTX and DOX, respectively: excitation wavelengths, 495 and 470 nm; emission wavelengths, 525 and 585 nm; number of reads: 4 per EV dilution. Standard curves of Oregon Green-488-PTX and DOX were employed to extrapolate drug concentration in EVs from fluorescence intensity values.

EV pre-conditioning of mice

Cell- or tumour-derived EVs (15 μg in 100 µl of PBS) were centrifugated at 4600 x g for 1 min at 4°C to remove sedimentable aggregates before intravenous injection into 6 to 8-week old, tumour-free mice (FVB/N, C57BL/6/Rag1^-/-, Balb/c, C57BL/6/Ccr2-KO or C57BL/6). Two EV doses were administered two days apart. For gene expression and FACS analyses, tumour-free mice were euthanized 2 days after the last EV dose. For experimental metastasis (lung colonization) studies, cancer cells were infused intravenously one day after the last EV administration and euthanized at the time points described below.

Lung colonization studies

4T1-mCh cells (5 x 10⁵) were injected intravenously in 6-8 week-old, male or female Rag1^-/- mice one day after EV pre-conditioning. The mice were euthanized 3 weeks after cancer cell injection and lungs fixed, sectioned, and analysed for direct mCh fluorescence to quantify metastatic tumour burden.

4T1 cells (1 x 10⁵) were injected intravenously in 6 week-old female Balb/c mice one day after EV pre-conditioning. The mice were euthanized 10 days after cancer cell injection and lungs fixed, sectioned, and stained with hematoxylin/eosin (H&E) to quantify metastatic tumour burden.

MMTV-PyMT tumour-derived cells (5 x 10⁵) were injected intravenously into 6-7 week-old female FVB/n mice one day after EV pre-conditioning. The mice were euthanized three weeks after tumour cell injection and lungs fixed, sectioned, and stained with H&E to quantify metastatic tumour burden.

E0771-LG:Fl cells (1x10⁶) were intravenously injected into 6-8 week-old, male or female C57BL/6 or C57BL/6/Ccr2-KO mice one day after EV pre-conditioning. Bioluminescence was determined 10 days after cancer cell injection.

In vivo bioluminescence imaging

C57BL/6 or Ccr2-KO mice were anesthetised with isofluorane (3-4 % induction, 1-2% maintenance) and administered 150 μg of luciferin (GoldBio) intraperitoneally per gram of body weight. Mice were imaged under Photon Imager Optima (Biospace Lab) for 15 min until the bioluminescence signal reached a plateau. Imaging was done 30 min and 10 days after E0771-LG:Fl cell injection. Photon counts (photon/second/cm2/sr) in the lung area were analysed using the M3 Vision software (Biospace Lab).

Analysis of pulmonary metastasis in mice

To analyse lungs for the presence of mCh-negative tumour nodules, formalin or paraformaldehyde-fixed lungs were sectioned into 5- or 8 μm-thick sections, respectively, at 100 μm intervals. Three to four large lung sections were stained with H&E. Sections were scanned with an Olympus slide scanner (VS120-L100) endowed with a 10x objective and fully automated scanning capabilities. Tumour nodules (>10-20 cancer cells) were counted and their area measured using the OlyVIA (Olympus) and ImageJ softwares, respectively. The number and mean area of the metastatic nodules were calculated by averaging data from individual sections.

To analyse lungs for the presence of mCh⁺ tumour nodules or disseminated cancer cells, paraformaldehyde-fixed lungs were sectioned into 8 μm-thick sections at 100 μm intervals. Four large sections were counterstained with DAPI and scanned with an Olympus slide scanner (VS120-L100) to acquire both mCh and DAPI fluorescence. Images were analysed using the ImageJ software and the VSI Biop tool. The total metastatic area was calculated by normalizing the mCh⁺ area to the total lung (DAPI⁺) area and by averaging data from individual sections. In some experiments, mCh⁺ cancer cells were also quantified by FACS analysis of homogenised lung tissue.

Zebrafish studies

The studies involving zebrafish (Danio rerio) were compliant with all relevant ethical regulations regarding animal research. Zebrafish were maintained according to standard protocols. All tumour studies were authorised by the UK Home Office under licence 40/3690 and performed at the zebrafish facility of the University of Sheffield, United Kingdom. The transgenic zebrafish line fms:UNM ⁶⁴ was crossed with the Tg(tie2:GFP) line ⁶⁵ to obtain the double transgenic fms:UNM; Tg(tie2:GFP) line, which enables the direct visualization of blood and lymphatic vessels.

Zebrafish embryos were maintained in E3 medium at 28°C. Before injection of EVs (or free drug) and cancer cells, 56h post-fertilization embryos were anesthetized by immersion in 0.04 mg/ml of tricaine (MS222; Sigma) and immobilized in 1% low melting point agarose (Sigma). Each embryo received 200-300 CFP⁺ MDA-MB-435 cells together with purified EVs (0.326 μg/ml) or drugs (PTX or CREMO; 100 μM) in 4 nl of saline, via the duct of Cuvier with a microinjector (World Precision Instruments, United Kingdom).

The embryos were analysed 18 h post-injection. After anaesthesia/immobilization, we used a spinning disc confocal microscope (Olympus IX81, PerkinElmer, UK) to image the tail of the zebrafish and capture Z-stacks at 3 µm increments. The full tail of the embryo was covered by two images at 10x magnification. Image acquisition and processing were performed with Volocity 6.3 software (Improvision, Coventry, United Kingdom). Cancer cell aggregates were visualized by capturing the CFP fluorescence of the cells, and the total metastatic volume was calculated from the XYZ analysis of CFP fluorescence in each of the tumour Z-stacks.

Preparation of tumours, lungs and livers for FACS analysis and cell sorting

Before sacrifice, deeply anaesthetised mice were perfused with PBS to remove intravascular blood cells. Tumours, lungs and livers were then harvested, washed extensively in PBS, and then minced before digestion with collagenase IV (0.2 mg/ml), dispase (2 mg/ml) and DNase I (0.002 mg/ml) in IMDM medium for 20 min at 37°C. The cell suspensions were filtered using a cell strainer (70 μm) and washed in PBS containing 2 mM EDTA and 2% FBS. Red blood cells were lysed using a Red blood cell Lysis buffer (Sigma).

Cell suspensions were incubated with rat anti-mouse FcγIII/II receptor (CD16/CD32) blocking antibodies (4 μg/ml) for 15 min at 4°C prior to extracellular staining with conjugated antibodies (Supplementary Table 4) for 20 min at 4°C. Before analysis, cells were resuspended in DAPI-containing buffer to identify non-viable cells. All samples were analysed using an LSRII apparatus (BD Biosciences). Compensation was performed using fluorescence-minus-one (FMO) controls.

To sort cells, we used a FACS Aria II apparatus (BD Biosciences). Each cell population was isolated from a single organ/mouse. Typically, 2x10³ to 2x10⁵ cells were obtained from each sorting session and all populations were verified by performing post-sorting analysis. Cell types were identified using the following combinations of cell markers:

Endothelial cells (ECs): CD45^–CD31⁺

Ly6C⁺ classical monocytes: CD45⁺CD11b⁺Ly6C⁺Ly6G^–F4/80⁺

Ly6C^low monocytes: CD45⁺CD11b⁺Ly6C^lowLy6G^–F4/80^–

Monocytes/macrophages: CD45⁺CD11b⁺Gr1^– or CD45⁺CD11b⁺Ly6G^–Ly6C^low

Alveolar lung macrophages: CD45⁺CD11c⁺SiglecF⁺F4/80⁺

Non-alveolar lung macrophages: CD45⁺CD11c^–SiglecF^–CD11b⁺Ly6G^–Ly6C^lowF4/80⁺

Neutrophils: CD45⁺CD11b⁺Ly6G⁺

Disseminated cancer cells: CD45^–CD31^–mCh⁺

Quantification of circulating cancer cells by FACS

Blood was collected by cardiac puncture of deeply anesthetised mice and collected in EDTA-containing tubes. To enumerate circulating cancer cells, we used 250 μl of erythrocyte-depleted blood. The cell suspension was processed as described above, and circulating cancer cells identified as CD45^–mCh⁺HER2⁺ cells using an LSRII apparatus.

Immunofluorescence staining of tumour and organ sections

Tumours and organs were fixed in 4% paraformaldehyde for 24h at 4°C, equilibrated in sucrose solutions for 24h, and then embedded in O.C.T compound (Cryomatrix, Thermo Fisher Scientific) on dry ice with isopentane (VWR Chemicals) before storage at -80°C. Tissue sections of 8 μm were obtained using a Leica cryostat CM1950 (Leica Biosystems). Before staining, the sections were rehydrated and then incubated with blocking solution (1% BSA, 10% FBS, 0.3% Triton X-100, and rat anti-mouse FcγII/III receptor 1:100, in PBS) for 1h at room temperature. The sections were incubated overnight at 4°C in 100-200 μl of blocking solution containing the primary antibodies listed in Supplementary Table 4. Non-conjugated antibodies were revealed with secondary antibodies. After staining, nuclei were labeled with DAPI (1 μg/ml) and sections mounted in Dako Fluorescence Mounting Medium, covered with cover glass (Heathrow Scientific), and stored at 4°C.

Immunofluorescence staining of cultured cells

Cells were seeded on glass chamber slides (Lab-Tek) or glass coverslips (Electron Microscopy Sciences) coated with 50 μg/ml poly-L-lysine (Sigma). After 16-24h, the cells were treated with drugs or EVs for an additional 24h. At the time of analysis, the cells were washed with PBS, fixed with 4% paraformaldehyde for 20 min, permeabilised with a solution containing 0.1% Triton X-100 in PBS for 10 min, and blocked with 1% BSA in PBS for 1h at room temperature. Primary antibodies (Supplementary Table 4) were then applied overnight at 4°C. In some experiments, Alexa Fluor-488 Phalloidin (Life Technologies, 1:40) was added for labelling the actin cytoskeleton. Slides were washed and secondary antibodies applied for 1h in blocking solution at the final concentration of 1:200. After washing, cell nuclei were labelled with DAPI (1 μg/ml) or DRAQ5 (20 µM, Thermo Fisher Scientific) and the slides mounted.

Duolink staining

bEnd.3 cells were seeded on poly-L-lysine-coated glass coverslips in a 48-well plate and incubated for 24h with PBS, CREMO- or PTX-EVs (50 μg/ml) from 4T1 cells. Cells were then washed with PBS and fixed with 4% paraformaldehyde for 20 min. After several washes with PBS, cells were permeabilised with 0.3% Triton for 20 min. The Duolink PLA technology (Sigma) was then applied to examine physical proximity between ANXA6 and p65. Specifically, the Red Starter Kit Mouse/Rabbit was used according to the manufacturer’s instructions. The primary rabbit anti-ANXA6 (Life Technologies, 720161) and mouse anti-p65 (Santa Cruz, clone F-6, sc-8008) antibodies were incubated overnight at 1:400 and 1:150, respectively, together with Alexa Fluor-488 Phalloidin (Life Technologies, 1:40) for labelling the actin cytoskeleton.

Image acquisition by confocal microscopy

In order to image immunostained tissue sections and cultured cells, we used a Zeiss LSM700 confocal microscope coupled to a high sensitivity Axiocam MRm (B/W) camera. Fluorescent signals from individual fluorophores were acquired sequentially from single optical sections, analysed and pseudocoloured using the ZEN software (Zeiss) or ImageJ.

Subcellular localization of mCh by immunofluorescence staining and confocal microscopy

To determine the subcellular localization of mCh in 4T1-mCh cultures exposed to drugs, confocal images were analysed using the ImageJ software. For each cell, the lining of the plasma membrane (+/- 0.5 μm) was determined manually by visualizing beta-tubulin immunostaining and saved as a region of interest (ROI), while the lining of the nucleus was determined according to DAPI staining by applying a Yen threshold with the Analyze particle tool of ImageJ. The cytoplasmic area was defined as the area between the plasma membrane and the nucleus. mCh signal intensity was calculated in both plasma membrane and cytoplasm compartments, and expressed as a ratio.

Quantification of mCh and ANXA6 transfer to ECs

To study mCh transfer, bEnd.3 cells were seeded in 24-well plates either directly on the plastic (for FACS and Western blotting) or on poly-L-lysine-coated glass coverslips (for immunofluorescence staining). Cells were then treated with purified EVs (1-200 μg/ml of total protein) or medium conditioned by cultured cancer cells (CM). Before use, CM was centrifuged at 500 x g for 5 min; 2000 x g for 5 min; and 4600 x g for 20xmin, at 4°C to remove dead cells, debris and large vesicles. bEnd.3 cells were then allowed to grow for additional 24h before analysis. Before analysis, the cells were extensively washed with PBS and treated with trypsin/EDTA to detach them (FACS), or lysed with RIPA buffer, or fixed with paraformaldehyde (immunofluorescence staining).

Chemicals and cytokines

Cas-BIND Pro Pan Caspase Pro-VAD-FMK (Verget Bioscience) or control DMSO were added to 40-50% confluent cells at a final concentration of 10 μM in EV-depleted medium. After 2h, PTX (100 ng/ml) or CREMO were added for another 72h. The cell permeant chelator BAPTA-AM (Life Technologies) or control DMSO were added to 40-50% confluent cells to a final concentration of 3 μM in EV-depleted medium. After 2h, PBS, PTX (100 ng/ml), DOX (200 ng/ml) or the appropriate vehicle were applied for another 48h. Recombinant human TNFα was purchased from Peprotech and used at 20 ng/ml in complete growth medium.

Cell viability and apoptosis assays

Cell viability was assessed by a WST-1 assay (Roche) according to the manufacturer’s instructions. Briefly, 5000 cells were seeded in 96-well plates and treated with drugs as described above. After 72h, 10 μl of WST-1 reagent was added to each well and absorbance at 450 nm was measured using a TECAN plate-reader (TECAN).

Cell apoptosis was assessed by FACS analysis of biotin-conjugated AnnexinV (BD Biosciences) and 7AAD (Sigma), according to the manufacturer’s instructions. Briefly, 200000 cells were seeded in 6-well plates and treated with drugs. After 48h, the cells were trypsinized and transferred to a tube along with the conditioned medium. Cells were centrifuged at 2000 rpm, washed with ice-cold PBS, and resuspended in 1x binding buffer at a concentration of 10⁶ cells/ml. The cell suspension was then stained with biotin-conjugated AnnexinV for 15 min at room temperature, followed by streptavidin-eFluor 450 (eBioscience; 1:100) for 15 min on ice. After washing, cells were resuspended in 7AAD (1:200) and immediately analysed by FACS.

Gene expression analyses by quantitative polymerase chain reaction (qPCR)

Small tissue fragments were obtained from the lungs of perfused tumour-free or bearing mice, washed in PBS, frozen on dry ice, and stored at –80°C. Total RNA was purified using miRNeasy Mini kit guidelines (Qiagen). RNA was quantified with a NanoDrop ND-2000 and retrotranscribed using SuperScript®Vilo cDNA synthesis kit (Invitrogen).

All qPCR reactions were performed using the following TaqMan gene assays: Hprt Mm 01545399-m1, B2m Mm 00437762_m1, Gapdh Mm99999915_g1 Angpt2 Mm00545822_m1, Csf1 Mm00432686_m1, Il6 Mm00446190_m1, Cxcl12 Mm 00445553-m1, Ccr2 Mm04207877_m1, and Ccl2 Mm00441242, all from Applied Biosystems/Life Technologies. Each assay was performed in three technical replicates using 10 ng of total cDNA per reaction and 2x TaqMAN Universal master mix (Life Technologies). qPCR reactions were performed for 40 cycles in standard mode using an ABI7900HT apparatus (Applied Biosystems). Raw data were extracted and evaluated using the SDS software v2.4. For the analysis of the results, we calculated the average of the median threshold cycle (C_t) of one or two reference genes (Hprt, Gapdh and/or B2m) for each sample (reference C_t), as described previously ⁶⁶.

Western blotting

Cells or purified EVs were lysed in ice-cold RIPA buffer containing Halt protease and phosphatase inhibitors (Thermo Fisher Scientific). Conditioned medium (CM) was concentrated using Vivaspin® 20 columns (5000K MW cutoff, Sartorius). Protein concentration was determined by BCA Protein Assay Reagent Kit (Thermo Fisher Scientific). Proteins were denatured at 95°C for 5 minutes with 6x Laemmli buffer (Bioworld), or at 90°C for 10 min with LDS Sample buffer 4x and Reducing agent 10x (both from Life Technologies). Unless stated otherwise, matched protein amounts for each sample were analysed in each experiment.

Proteins (10 to 20 µg) were separated by 12% or 4-12% (gradient) SDS-PAGE (Life Technologies) and blotted onto PVDF membranes (Amersham Hybond 0.45 µm, GE Healthcare). Membranes were blocked with 5% milk in TBS-T for 24h, incubated with primary antibodies (Supplementary Table 4) for 12 h at 4°C, and then with HRP-conjugated secondary antibodies for 1h at room temperature. Proteins of interest were visualized using ECL Western Blotting substrate (Pierce, Rockford, IL, USA) and a Fusion FX7 device (Peqlab). In some experiments, PVDF membranes were instead blocked with Odyssey Blocking Buffer (LI-COR Biosciences), incubated with primary antibodies for 12 h at 4°C, followed by IRDye 800CW or 680RD-conjugated secondary antibodies (LI-COR Biosciences) in 0.1% Tween and 0.01% SDS for 1h at room temperature. Blots were imaged on the LI-COR Odyssey scanner.

ELISA

After dissection, lungs were immediately frozen on dry ice and stored at -80°C. For protein extraction, lungs were washed with ice-cold PBS and minced into small fragments before mechanical homogenization in PBS containing Halt phosphatase and protease inhibitors (Thermo Fisher Scientific). After three freeze-thaw cycles to break cellular membranes, homogenates were centrifuged at 13000 rpm for 15 min at 4°C. To measure CCL2, we used the Quantikine Mouse/Rat CCL2/MCP-1 Immunoassay from R&D (MJE00) and 50 mg of total protein for each reaction. ELISA was performed according to the manufacturer’s instructions. Absorbance was measured using a TECAN Safire2 plate-reader (TECAN).

To examine if ANXA6 localizes on the EV surface, we used a homemade ELISA and a polyclonal anti-ANXA6 antibody (720161, Thermo Fisher Scientific). Antibodies against the endoplasmic reticulum protein GP96 (clone 9G10, Enzo) or the extracellular domain of CD81 (clone B-11, Santa Cruz) were used as negative and positive control for staining the outer EV surface, respectively. The antibodies were diluted (5 μg/ml) in ELISA coating solution (Thermo Fisher Scientific) and used for coating a Nunc MaxiSorb 96-well plate (Thermo Fisher Scientific). The plate was then blocked with ELISA blocking solution (Thermo Fisher Scientific) and washed (two washes with PBS with 0.04% Tween, followed by two with PBS only) before adding the EVs (150 μg/ml, in PBS) for an overnight incubation at 4°C. Acquisition of EV-associated mCh fluorescence, before and after four washes with PBS to remove unadsorbed EVs, used a TECAN Safire2 plate-reader. A final acquisition was performed after the lysis of the EVs with 0.3% Triton in PBS for 2h at 4°C and three PBS washes.

EDTA treatment of EVs

To examine the localization of ANXA6 in EVs, EVs were treated with PBS or EDTA in PBS (1 mM) for 30 min at 37°C. After washing three times in PBS, the EVs were loaded onto Vivaspin® 500 columns (300000 MW cutoff, Sartorius) and centrifuged at 4000 x g for 15 min to retain EVs and discharge free macromolecules, before Western blotting analysis.

Late endosome enrichment

Cells were treated with drugs for 24h. After cell homogenization with a 25-gauge needle, we enriched late endosomes by continuous density gradient centrifugation ⁶⁷. The 25%/HB interface, enriched in late endosomes, and the 35%/40.6% interface, enriched in heavier membranes (e.g., endoplasmic reticulum and mitochondria), were collected and concentrated by ultracentrifugation (100000 x g for 1h) using a Beckman ultracentrifuge. The 40.6% fraction enriched in cytosolic proteins was also collected to concentrate proteins with Vivaspin® 20 columns (3000K MW cutoff, Sartorius).

NF-kB reporter assay

Cells were transduced with the NF-kB reporter LV and treated with 10 μg/ml EVs several weeks after transduction. Two days post-EV treatment, the cells were trypsinized, washed in EDTA and analysed by FACS. To quantitate NF-kB activity, we calculated the fluorescence intensity of both GFP and mCh after subtracting the background fluorescence of untransduced cells. In studies involving PTX-EVs, NF-kB activity was determined as the ratio between mCh and GFP fluorescence. In studies involving DOX-EVs, NF-kB activity was determined from GFP fluorescence, as DOX is fluorescent in the red channel.

Liquid chromatography-tandem mass spectrometry (LC-MS/MS)

EVs (17 μg) of PTX- or CREMO-treated 4T1 cells were directly analysed by LC-MS/MS. EV-derived proteins of MDA-MB-231 cells (1 μg) or human plasma (1 μg) were first resolved on a denaturing 4-12% polyacrylamide gradient gel. Gel slices were excised in the area predicted to contain ANXA6 (ranging from 60 to 85 kDa) to elute proteins for LC-MS/MS.

To perform LC-MS/MS, we applied a previously published procedure ⁶⁸. Briefly, peptides were desalted using StageTips, dried in a vacuum concentrator, and separated by reverse phase chromatography using a Dionex Ultimate 3000 RSLC nano UPLC system connected in-line with an Orbitrap Elite (Thermo Fisher Scientific). Database search was performed using Mascot 2.5 (Matrix Science) and SEQUEST in Proteome Discoverer v.1.4. against a murine or human Uniprot protein database. Data were further processed and inspected in ScaffoldTM 4.8.4 (Proteome Software); quantitative values normalized to total spectra were used to compare the proteomic profiles.

Plasma isolation from breast cancer patients

Using an IRB-approved sample collection protocol at the Massachusetts General Hospital Cancer Center, subjects were consented for serial venous blood collections. The study was compliant with all relevant ethical regulations regarding research involving human participants. Plasma samples were obtained from six women diagnosed with invasive breast cancer with histopathologically confirmed tumour stage ranging from IA to IIIA; immunohistochemical subtyping and lymph node status of each patient are reported in Supplementary Table 2. All subjects had 20 ml of blood collected into Streck tubes prior to initiation of neoadjuvant chemotherapy, which consisted of four cycles of every two week dose-dense DOX at 60 mg/m2 and cyclophosphamide at 600 mg/m² (AC), followed by four cycles of every two week PTX at 175 mg/m². A second 10 ml blood sample was collected prior to starting PTX (on the day of the first PTX infusion), and a third 10 ml blood sample on the last or second-to-last cycle of PTX.

All blood samples were centrifuged at 1600 x g at room temperature for 10 min. The supernatant (plasma) was then transferred to one fresh 10 ml centrifuge tube without disturbing the cellular layer, and centrifuged at 3000 x g at room temperature for 10 min. The supernatant from the second centrifugation was stored into 2 ml pre-labelled cryogenic vials at -70°C. Before EV isolation, plasma was centrifuged at 10000 x g for 20 min at 4°C; one ml of plasma was processed for EV isolation.

Statistics and reproducibility

Information on the study outline, sample size and statistical analysis is presented in the main text, figures, and figure legends. Studies involving independent cohorts of mice were typically performed once, with the exceptions stated in the figure legends. Each study was designed to use the minimum number of mice required to obtain informative results (i.e., quantitative data suited to statistical analysis) and sufficient material for further studies. No specific statistical tests were used to predetermine the sample size.

The investigators were not blinded when assessing the results or analysing data. On rare occasions, outliers at end-point were excluded by using the ROUT method (GraphPad Prism) to identify outliers. In rare cases, selected samples were lost or excluded from specific analyses because of technical flaws during sample collection, processing or data acquisition.

The number of biological (non-technical) replicates for each experiment is indicated in the figure legends. All experiments were repeated independently with similar results. When less than three biological replicates were available, statistics were not computed. Data are shown as mean values ± standard error of the mean (s.e.m.) for most analyses, with the exception of in vitro studies with a sample size of n=3, in which mean values ± standard deviation (s.d.) are reported. In experiments involving animals, each dot in the chart represents one mouse or zebrafish embryo. Source data for all experiments are provided in Supplementary Table 5.

Statistical analysis of the data was performed using the GraphPad Prism software. Statistical analysis of differentially expressed proteins in the LC-MS/MS 4T1-EV study was performed using an unpaired two-tailed t-test implemented in R; P values were adjusted after correcting for false discovery rate using a Benjamini-Hochberg method. Statistical analysis of experiments involving more than two experimental groups used one-way ANOVA with Tukey’s correction for multiple comparisons, unless indicated otherwise. When the influence of two nominal variables on one measurement variable was examined, a two-way ANOVA followed by Tukey’s or Sidak's correction was applied. Experiments involving two experimental groups were analysed by unpaired two-tailed student’s t-test with 95% confidence interval, unless indicated otherwise.

Statistical significance of the data is summarized as follows: *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001. Differences among experimental groups were considered significant at P values of less than 0.05. Exact P values are provided in all instances, except when P > 0.05 or P < 0.0001.