Identification of HIF target genes in brain-metastatic BC cells by RNA-Seq and ChIP-Seq. MDA-MB-231-BrM2 (hereafter designated MDA231-BrM2) is a subclone of MDA-MB-231 (hereafter designated MDA231) human TNBC cells in which the capacity for brain metastasis was enriched by in vivo selection (16). Compared with the parental MDA231 cells, 1,957 RNAs were overexpressed in the MDA231-BrM2 subclone (16). To determine whether these overexpressed RNAs were enriched for HIF target genes, we stably transfected MDA231-BrM2 cells with a lentivirus vector encoding a non-targeting control (NTC) short hairpin RNA (shRNA) or vectors encoding shRNAs targeting HIF-1α and HIF-2α (HIF double knockdown [DKD]). These subclones were exposed to 20% or 1% O2 for 24 hours, RNA was isolated, and RNA sequencing (RNA-Seq) was performed on 3 biological replicates for each condition. In response to hypoxia, the expression of 2,234 RNAs was significantly increased (FDR < 0.05) in a HIF-dependent manner (Figure 1A). Gene Ontology (GO) analysis revealed that the most highly enriched category was “response to hypoxia” with –log10P value > 9 (Figure 1B). HIF-dependent RNAs were also enriched for hypoxia-related pathways (angiogenesis and canonical glycolysis), cancer-related terms (regulation of cell migration), and neural-related terms, suggesting that expression of these RNAs may contribute to brain metastasis (Figure 1B). Consistent with GO analysis, gene set enrichment analysis (GSEA) revealed that HIF-regulated RNAs were highly enriched for the hypoxia, glycolysis, and angiogenesis gene sets (Supplemental Figure 1, A–C; supplemental material available online with this article; https://doi.org/10.1172/JCI190470DS1), as well as a gene set associated with BC metastasis (Figure 1C).

Identification of HIF target genes in brain-metastatic BC cells by RNA-Seq and ChIP-Seq. (A and B) RNA-Seq analysis of MDA231-BrM2 cells exposed to 20% or 1% O2 for 24 hours was performed. Volcano plot (A) and Gene Ontology (GO) analysis (B) of HIF target genes are shown. (C) Gene set enrichment analysis (GSEA) revealed that expression of the “breast cancer metastasis” gene set was significantly correlated with the expression of HIF target genes in MDA231-BrM2 cells. (D) HIF-1α binding profiles at significantly called peak summits ± 1 kb in MDA231-BrM2 cells exposed to 20% or 1% O2, as determined by ChIP-Seq. (E) Venn analysis shows the overlap among 2,234 HIF-dependent hypoxia-induced genes as determined by RNA-Seq; 1,474 genes with HIF-1α binding sites by ChIP-Seq; and 1,957 genes overexpressed in MDA231-BrM2 cells, as compared with MDA231 cells (from ref. 16). (F) Heatmaps showing RNA expression of 44 shared genes in NTC versus HIF-double-knockdown cells at 20% or 1% O2 (left), and in MDA231 versus MDA231-BrM2 cells at 20% O2 (right).

To further investigate the role of HIF-1 in brain metastasis, we performed chromatin immunoprecipitation and DNA sequencing (ChIP-Seq) in MDA231-BrM2 cells exposed to 20% or 1% O2 for 16 hours with 3 biological replicates for each condition. Nuclear lysates were immunoprecipitated with HIF-1α antibody, followed by DNA sequencing. This identified 1,474 high-stringency genomic sites at which HIF-1α occupancy was induced by hypoxia (Figure 1D). We then identified RNAs that were (a) overexpressed in MDA231-BrM2 relative to MDA231 cells; (b) induced by hypoxia in MDA231-BrM2 cells in a HIF-dependent manner; and (c) encoded by a gene at which hypoxia-induced HIF-1α binding was identified by ChIP-Seq. We found that 44 RNAs fulfilled all 3 of these criteria (Figure 1, E and F, and Table 1). ITGB3, which encodes integrin subunit β3, was of interest because previous studies have implicated integrins α6β1 and α6β4 in lung metastasis and αvβ5 in liver metastasis; most importantly, integrin β3 was identified in EVs derived from brain-tropic MDA231-BrM2 cells (20), highlighting its potential role in brain metastasis and providing a strong rationale for its further investigation in this context.

The 44-gene signature identified through analysis of RNA-Seq and ChIP-Seq datasets from brain-metastatic MDA231-BrM2 cells

ITGB3 is a direct HIF-1 target gene in brain-metastatic BC cells. To further investigate whether ITGB3 expression is O2 regulated, we exposed 4T1 mouse TNBC cells and the in vivo–selected brain-metastatic 4T1-BR5 subclone (19) to 20% or 1% O2 for 24 or 48 hours, then harvested the cells for reverse transcription (RT) and quantitative real-time PCR (qPCR) (Figure 2A) or immunoblot (Figure 2B) assays, respectively. These results revealed that both ITGB3 mRNA and protein levels were induced by hypoxia. ITGB3 mRNA and protein expression was much higher in 4T1-BR5 than in 4T1 cells (Figure 2, A and B). To determine the HIF-dependence of ITGB3 expression, we established 4T1-BR5 and MDA231-BrM2 subclones stably transduced with a lentiviral expression vector encoding NTC shRNA or shRNA targeting HIF-1α, HIF-2α, or both (double knockdown [DKD]). RT-qPCR and immunoblot assays revealed that knockdown of HIF-1α, but not HIF-2α, abrogated the induction of ITGB3 mRNA and protein expression in both 4T1-BR5 (Figure 2, C and D) and MDA231-BrM2 (Figure 2, E and F) cells exposed to 1% O2. Flow cytometry revealed that hypoxia also induced ITGB3 protein expression in the plasma membrane in 4T1-BR5 (Figure 2G) and MDA231-BrM2 (Figure 2H) cells.

ITGB3 is a direct HIF-1 target gene. (A and B) ITGB3 mRNA and protein were analyzed by RT-qPCR (A) and immunoblot assays (B) in 4T1 and 4T1-BR5 cells exposed to 20% or 1% O2. ITGB3 mRNA was quantified relative to 18S rRNA and normalized to mean for 4T1 cells at 20% O2; mean ± SD (n = 3). ****P < 0.0001 vs. 4T1 at 20% O2; ####P < 0.0001 vs. 4T1 at 1% O2 (unpaired 2-tailed Student’s t test). (C–F) 4T1-BR5 (C and D) or MDA231-BrM2 (E and F) subclones were exposed to 20% or 1% O2, and ITGB3 mRNA or protein levels were analyzed by RT-qPCR (C and E) and immunoblot assays (D and F); mean ± SD (n = 3). ****P < 0.0001 vs. NTC at 20% O2; ####P < 0.0001 vs. NTC at 1% O2 (2-way ANOVA with Tukey’s multiple-comparison test). (G and H) Flow cytometry histograms showing anti-ITGB3 antibody binding to 4T1-BR5 (G) or MDA231-BrM2 (H) cells that were exposed to 20% or 1% O2; mean ± SD (n = 3). ****P < 0.0001 vs. 20% O2. (I) ChIP-Seq analysis revealed 3 matches to the HIF consensus binding site 5′-(A/G)CGTG-3′ or its complement (underlined) under the HIF-1α peak in the ITGB3 gene in MDA231-BrM2 cells using the Integrative Genomics Viewer (IGV) genome browser. (J) MDA231-BrM2 cells were exposed to 20% or 1% O2, and ChIP-qPCR was performed using antibodies against HIF-1α, HIF-1β, or HIF-2α. Primers flanking the nucleotide sequence shown in I were used for qPCR, and results were normalized to the mean result at 20% O2; mean ± SD (n = 3). ****P < 0.0001 vs. 20% O2; ns, not significant (unpaired 2-tailed Student’s t test).

Analysis of the ChIP-Seq data from hypoxic MDA231-BrM2 cells revealed a HIF-1 binding peak that was located 24 kb upstream of the human ITGB3 gene transcription start site, and inspection of the DNA sequence at this site revealed 3 matches to the HIF-1 binding sequence 5′-(A/G)CGTG-3′ (Figure 2I). To confirm that HIF-1 specifically binds at this position, MDA231-BrM2 cells exposed to 20% or 1% O2 for 16 hours were analyzed by ChIP-qPCR. Chromatin fragments were precipitated using antibodies against HIF-1α, HIF-1β, or HIF-2α, and fragments containing the putative binding site were quantified by qPCR. Binding of HIF-1α and HIF-1β, but not HIF-2α, was significantly enriched at the ITGB3 –24 kb site in hypoxic cells (Figure 2J), which is consistent with the finding that hypoxia-induced ITGB3 expression in MDA231-BrM2 cells required HIF-1α, but not HIF-2α (Figure 2, E and F). Taken together, the data presented in Figure 2 demonstrate that ITGB3 is a hypoxia-inducible HIF-1 target gene in human and mouse TNBC cells that are capable of metastasis to the brain.

ITGB3 expression is required for hypoxia-induced migration and invasion of brain-metastatic BC cells. ITGB3 forms an αvβ3 heterodimer with ITGAV, which has been implicated in cancer cell adhesion, migration, and invasion (41). To explore the biological function of ITGB3 and ITGAV in brain-metastatic BC cells, we established knockdown subclones of 4T1-BR5 and MDA231-BrM2 by transducing them with vectors encoding shRNA targeting ITGB3 or ITGAV. Knockdown efficiency in 4T1-BR5 (Figure 3, A–D) and MDA231-BrM2 cells (Supplemental Figure 2, A–D) was determined by RT-qPCR and immunoblot assays. The effect of ITGB3 or ITGAV shRNA on cell migration and invasion was analyzed using Boyden chamber assays, in which cells were seeded onto a porous membrane that was either uncoated or coated with Matrigel, a sarcoma-derived basement membrane preparation (42). Hypoxia increased the migration and invasion of 4T1-BR5 (Figure 3, E–H) and MDA231-BrM2 (Supplemental Figure 2, E–H) NTC subclones. However, hypoxia-induced migration and invasion were markedly impaired by expression of shRNAs targeting ITGB3, ITGAV, or HIF-1α (Figure 3, E–H, and Supplemental Figure 2, E–H).

ITGB3 expression is required for hypoxia-induced migration and invasion of brain-metastatic BC cells. (A–D) 4T1-BR5 subclones that were stably transduced with a lentivirus encoding an NTC shRNA or shRNA targeting ITGB3 or ITGAV were subjected to RT-qPCR (A and C) or immunoblot assays (B and D). Data are shown as mean ± SD (n = 3). **P < 0.01, ***P < 0.001, ****P < 0.0001 vs. shNTC; ns, not significant vs. shNTC (unpaired 2-tailed Student’s t test). (E–H) 4T1-BR5 subclones were seeded on top of uncoated (E and F) or Matrigel-coated (G and H) Boyden chamber inserts and incubated at 20% or 1% O2 for 16 (E and F) or 24 (G and H) hours. Cells on the underside of the insert were stained with crystal violet and imaged by light microscopy (E and G; scale bars: 100 μm). The stained area was quantified using ImageJ (NIH) and expressed as mean ± SD (n = 3). **P < 0.01, ***P < 0.001, ****P < 0.0001 vs. shNTC at 20% O2; ####P < 0.0001 vs. shNTC at 1% O2 (2-way ANOVA with Tukey’s multiple-comparison test).

Digoxin is a HIF inhibitor that blocks HIF-1α protein accumulation and inhibits primary tumor growth, as well as lymph node and lung metastasis, of MDA231 cells implanted in the mammary fat pad of immunodeficient mice (43, 44). MDA231-BrM2 cells were treated with digoxin at 100 or 200 nM during exposure to 20% or 1% O2 for 24 hours. RT-qPCR assays revealed that digoxin treatment blocked hypoxia-induced expression of CA9 mRNA, an established HIF-1 target gene product (45), and ITGB3 mRNA, but had no effect on the expression of RPL13A mRNA, which is neither hypoxia-induced nor HIF-regulated (Figure 4A). Cilengitide is an RGD pentapeptide that specifically blocks the binding of integrin αvβ3 to its ligands (46). To further investigate the effect of HIF-1–mediated ITGB3 expression on the migration and invasion of brain-metastatic BC cells, we pharmacologically inhibited HIF-1 or ITGB3 by treating cells with digoxin or cilengitide, respectively. Boyden chamber assays revealed that treatment with digoxin or cilengitide significantly decreased hypoxia-induced migration (Figure 4, B and C) and invasion (Figure 4, D and E) of MDA231-BrM2 cells. Both genetic and pharmacological inhibition of ITGB3 decreased the migration and invasion of MDA231-BrM2 cells at 20% O2 (Supplemental Figure 2, E–H, and Figure 4, B–E), indicating that ITGB3 expression is required for basal as well as hypoxia-induced migration and invasion of MDA231-BrM2 cells.

Pharmacological inhibition of HIF-1 or ITGB3 impairs migration and invasion of brain-metastatic BC cells. (A) MDA231-BrM2 cells were exposed to 20% or 1% O2 for 24 hours in the presence of vehicle (DMSO) or digoxin, and expression of CA9, ITGB3, and RPL13A mRNA was assayed by RT-qPCR. Data are shown as mean ± SD (n = 3). ****P < 0.0001 vs. DMSO at 20% O2; ####P < 0.0001 vs. DMSO at 1% O2; ns, not significant (2-way ANOVA with Tukey’s multiple-comparison test). (B–E) MDA231-BrM2 cells were seeded on top of uncoated (B and C) or Matrigel-coated (D and E) Boyden chamber inserts and incubated at 20% or 1% O2 in the presence of DMSO, digoxin (200 nM), or cilengitide (5 μM). Cells on the underside of the insert were stained with crystal violet and imaged by light microscopy (B and D; scale bars: 100 μm). The stained area was quantified using ImageJ and expressed as mean ± SD (n = 3). *P < 0.05, **P < 0.01, ****P < 0.0001 vs. 20% O2 DMSO; ####P < 0.0001 vs. 1% O2 DMSO; ns, not significant (2-way ANOVA with Tukey’s multiple-comparison test).

Next, MDA231-BrM2 cells were stably transfected with an ITGB3 expression vector, and ITGB3 protein overexpression was confirmed by immunoblot assays (Supplemental Figure 3A). ITGB3 overexpression significantly increased migration and invasion of MDA231-BrM2 cells under non-hypoxic conditions (Supplemental Figure 3, B–E). Collectively, these results indicate that HIF-1–mediated ITGB3 expression is necessary for increased migration and invasion under hypoxic conditions, and that increased ITGB3 expression is sufficient to increase migration and invasion under non-hypoxic conditions.

HIF-1–induced ITGB3 expression is required for brain colonization by BC cells. When hypoxic BC cells enter the circulation, they are no longer subjected to hypoxia. In order to investigate how long hypoxic BC cells maintain increased ITGB3 expression after reoxygenation, we exposed 4T1-BR5 cells to 20% or 1% O2 for 48 hours and reoxygenated them for different periods of time. Immunoblot assays revealed that hypoxia-induced ITGB3 expression was maintained elevated for 24 hours and slowly decayed between 48 and 96 hours of reoxygenation (Supplemental Figure 4A). BC cells subjected to hypoxia for 48 hours and reoxygenated for 48 hours maintained an increased capacity for migration and invasion compared with cells maintained at 20% O2 (Supplemental Figure 4, B and C). These results suggest that there is a significant window of time in which post-hypoxic BC cells are endowed with increased capacity for migration and invasion.

To determine whether the hypoxia-induced increases in migration and invasion in vitro were associated with enhanced brain colonization in vivo, we exposed 4T1-BR5 cells to 20% or 1% O2 for 48 hours, injected 50,000 cells into the cardiac left ventricle (LV) of syngeneic BALB/c mice, and harvested the brains 14 days later for histological analysis using hematoxylin and eosin (H&E) staining. Prior exposure of 4T1-BR5 cells to hypoxia significantly increased the area of brain colonization (Figure 5, A and B). To further validate the role of HIF-1 and ITGB3 in promoting brain metastasis, we injected the shNTC, shHIF1A, or shITGB3 subclone of 4T1-BR5 (50,000 cells) or MDA231-BrM2 (250,000 cells) into the LV and analyzed brain colonization 14 or 40 days later, respectively. Knockdown of either HIF-1α or ITGB3 markedly decreased brain colonization following LV injection of 4T1-BR5 (Figure 5, C and D) or MDA231-BrM2 (Figure 5, E and F) cells.

HIF-1–induced ITGB3 expression is required for colonization of brain by BC cells. (A and B) 4T1-BR5 cells were exposed to 20% or 1% O2 for 48 hours and injected into left ventricle of BALB/c mice. On day 14, brains were harvested, and sections were analyzed by H&E staining. Representative images (A; scale bars: 200 and 50 μm in top and bottom panels, respectively) and quantification of metastatic area (B; mean ± SD; n = 5 mice, 3 sections per brain) are shown. **P < 0.01 vs. 20% O2 (unpaired 2-tailed Student’s t test). (C and D) Representative images of H&E-stained brain sections (C; scale bars: 200 and 50 μm in top and bottom panels, respectively) and quantification of metastatic area (D; mean ± SD; n = 5 mice, 3 sections per brain) 14 days after intracardiac injection of 4T1-BR5 subclones expressing the indicated shRNA are shown. ***P < 0.001, ****P < 0.0001 vs. shNTC (1-way ANOVA with Tukey’s multiple-comparison test). (E and F) Representative images of H&E-stained brain sections (E; scale bars: 200 and 50 μm in top and bottom panels, respectively) and quantification of metastatic area (F; mean ± SD; n = 8 mice, 3 sections per brain) 40 days after intracardiac injection of MDA231-BrM2 subclones expressing the indicated shRNA are shown. ****P < 0.0001 vs. shNTC (1-way ANOVA with Tukey’s multiple-comparison test).

ITGB3 is exported from BC cells via EVs. BC-derived EVs have been shown to remodel the microenvironment of distant organs to promote metastatic niche formation (20) and brain metastasis (21). ITGB3 was identified not only in EVs isolated from brain-tropic MDA231-BrM2 cells, but also in EVs from liver-tropic and lung-tropic BC cells, as well as EVs from liver-tropic and lung-tropic colorectal, gastric, and pancreatic cancer cells (20). To confirm the expression of ITGB3 in EVs and the effect of hypoxia on EV biogenesis, MDA231-BrM2 and 4T1-BR5 cells were exposed to 20% or 1% O2 for 48 hours. The conditioned medium was collected, and EVs were isolated by ultracentrifugation. The number and size distribution of EVs were characterized by nano–flow cytometry (NFCM) (Figure 6A and Supplemental Figure 5A) and nanoparticle tracking analysis (NTA) (Figure 6B and Supplemental Figure 5B). Consistent with previous findings (47, 48), both MDA231-BrM2 and 4T1-BR5 cells generated more EVs at 1% O2 than at 20% O2, but with similar size distribution. Transmission electron microscopy revealed that the EVs had a typical cup-shaped morphology with a size range of 50–150 nm (Figure 6C and Supplemental Figure 5C), which was consistent with the data from NFCM (Figure 6A and Supplemental Figure 5A) and NTA (Figure 6B and Supplemental Figure 5B).

ITGB3 is exported from BC cells via EVs. (A and B) MDA231-BrM2 cells were exposed to 20% or 1% O2 for 48 hours, and EVs were isolated and characterized by nano–flow cytometry (A) and nanoparticle tracking analysis (B). (C) Representative transmission electron microscopy images of EVs derived from MDA231-BrM2 cells are shown. Scale bars: 100 nm. (D) EVs and corresponding whole-cell lysates were characterized by immunoblot assays. (E and F) EVs derived from MDA231-BrM2 cells were stained with antibody against ITGB3 (E) or ITGB4 (F) and analyzed by nano–flow cytometry. Blue, unstained; red, antibody-stained; black, total particle population.

Immunoblot assays of MDA231-BrM2 EVs, as compared with whole-cell lysates, revealed enrichment of ITGB3 and ITGAV, along with the well-established EV markers CD9, CD63, CD81, and TSG101, whereas the EVs did not contain calnexin, a protein localized to endoplasmic reticulum (Figure 6D). Knockdown of ITGAV reduced the incorporation of ITGB3, but not other proteins, in the EVs, and knockdown of ITGB3 reduced ITGAV incorporation into EVs (Supplemental Figure 5G). These results indicate that the observed effects of ITGB3 are likely to occur in the context of αvβ3 heterodimers. Knockdown of both HIF-1α and ITGB3 had no additional effects on EV composition (Supplemental Figure 5, D–G). Hypoxia increased the percentage of EVs containing ITGB3 on their surface from 25.8% to 43.4% in MDA231-BrM2 (Figure 6E) and from 13.8% to 18.6% in 4T1-BR5 (Supplemental Figure 5H), as determined by NFCM. In contrast, there was no effect of hypoxia on the percentage of EVs that displayed ITGB4 (Figure 6F), which was previously identified as a lung metastasis–specific integrin (20). These data indicate that hypoxia-induced ITGB3 is selectively expressed on the surface of EVs produced by brain-tropic MDA231-BrM2 and 4T1-BR5 BC cells.

HIF-1α expression and ITGB3 expression promote interaction of EVs with brain ECs. The BBB is a unique semipermeable structure of the microvasculature in the central nervous system that is mainly composed of ECs, pericytes, and astrocytes and can tightly control the translocation of molecules, proteins, and particles to the brain (49). Tumor-derived EVs have been shown to cross the BBB (50). To determine whether EVs produced by brain-tropic BC cells can interact with brain ECs, we generated MDA231-BrM2 subclones expressing the bioluminescence resonance energy transfer reporter protein palmitoylated-EGFP-nanoluciferase (PalmGRET), which can robustly and specifically label the inner membrane of EVs (51). EVs labeled with PalmGRET were isolated from shNTC, shHIF1A, and shITGB3 subclones of MDA231-BrM2 cells that were exposed to 20% or 1% O2. Human hCMEC/D3 cells, which are derived from brain ECs and manifest BBB characteristics (52), were incubated with these labeled EVs for 24 hours, followed by flow cytometry to identify ECs with bound PalmGRET+ EVs. EVs derived from hypoxic NTC cells showed a significant increase in association with brain ECs; moreover, the association of brain ECs with EVs from shHIF1A and shITGB3 cells was significantly decreased in comparison with EVs from shNTC cells (Figure 7, A and B). Similar results were obtained using HBEC-5i cells (Supplemental Figure 6, A and B), which are also microvascular ECs derived from human brain (53).

HIF-1α expression and ITGB3 expression promote the interaction of EVs with brain ECs. (A and B) GFP+ PalmGRET-EVs (2 μg; derived from MDA231-BrM2 cells exposed to 20% or 1% O2 for 48 hours) were incubated with hCMEC/D3 ECs for 24 hours, and then the cells were analyzed by flow cytometry (A) and quantified (B; mean ± SD, n = 3). ****P < 0.0001 vs. NTC at 20% O2; ####P < 0.0001 vs. NTC at 1% O2 (2-way ANOVA with Tukey’s multiple-comparison test). (C–E) PalmGRET-EVs (10 μg; derived from MDA231-BrM2 cells exposed to 20% or 1% O2 for 48 hours) were injected into the left ventricle of BALB/c mice, and then the brains were analyzed by fluorescence microscopy and nanoluciferase assay. Representative fluorescence microscopy images (C; scale bars: 100 and 20 μm in top and bottom panels, respectively), quantification of fluorescent area (D), and nanoluciferase activity (E) are shown; mean ± SD (n = 5 mice, 3 sections per brain). **P < 0.01, ****P < 0.0001 vs. 20% O2 (unpaired 2-tailed Student’s t test). (F–H) PalmGRET-EVs (10 μg; derived from MDA231-BrM2 subclones exposed to 1% O2 for 48 hours) were injected into the left ventricle of BALB/c mice, and then the brains were analyzed by fluorescence microscopy and nanoluciferase assay. Representative fluorescence microscopy images (F; scale bars: 100 and 20 μm in top and bottom panels, respectively), quantification of fluorescent area (G), and nanoluciferase activity (H) are shown; mean ± SD (n = 5 mice, 3 sections per brain). ***P < 0.001, ****P < 0.0001 vs. shNTC (1-way ANOVA with Tukey’s multiple-comparison test).

To further confirm the interaction of EVs with brain ECs, MDA231-BrM2 cells were exposed to 20% or 1% O2 for 48 hours, and PalmGRET+ EVs were isolated and injected into the LV. Four hours after injection, we perfused the mice with 4% paraformaldehyde and harvested the brains. Fluorescence microscopy of brain sections revealed a significant increase in the number of PalmGRET+ EVs from cells exposed to 1% as compared with 20% O2 (Figure 7, C and D). Similarly, nanoluciferase assays of brain lysates showed significantly increased signal from mice injected with EVs from hypoxic MDA231-BrM2 cells (Figure 7E). Next, the shNTC, shHIF1A, and shITGB3 subclones of MDA231-BrM2 cells were exposed to 1% O2, PalmGRET+ EVs were collected and injected into the LV, and brains were harvested for analysis. Fluorescence microscopy identified a significantly decreased number of EVs from shHIF1A and shITGB3 as compared with shNTC cells (Figure 7, F and G). Similar results were obtained from nanoluciferase assays of brain lysates (Figure 7H). Taken together, the results presented in Figures 6 and 7 demonstrate that hypoxia induces HIF-1–dependent ITGB3 expression in EVs, which is required for the increased retention of circulating EVs in the brain.

ITGB3+ EVs promote interaction of BC cells with ECs and increase endothelial permeability. Next, a monolayer of hCMEC/D3 brain ECs was exposed for 24 hours to EVs collected from BC cells cultured at 20% or 1% O2. The ECs were then cocultured with GFP-expressing MDA231-BrM2 cells for 1 hour, after which non-adherent cells were removed by gentle rinsing (Figure 8A). Significantly more GFP+ MDA231-BrM2 cells interacted with hCMEC/D3 cells when the latter had been pretreated with EVs from BC cells cultured at 1% as compared with 20% O2, as determined by immunofluorescence (Figure 8, B and C) or flow cytometry (Supplemental Figure 7, A and B). In addition, EVs from shNTC cells promoted greater interaction of GFP+ BC cells with ECs than EVs from shHIF1A or shITGB3 cells, as determined by immunofluorescence (Figure 8, D and E) and flow cytometry (Supplemental Figure 7, C and D).

ITGB3+ EVs promote interaction of BC cells with ECs and increase endothelial permeability. (A–E) As shown in the schematic created with BioRender (biorender.com) (A), hCMEC/D3 ECs were seeded on 6-well plates, cultured to confluence, and treated for 24 hours with EVs from MDA231-BrM2 cells that were exposed to 20% or 1% O2 (B and C) or subclones that were exposed to 1% O2 (D and E). GFP+ MDA231-BrM2 cells were then added onto the hCMEC/D3 monolayer and incubated for 1 hour, and non-adherent cells were removed by washing with PBS. Adherent BC cells were imaged by fluorescence microscopy (B and D; scale bars: 10 μm) and quantified (C and E; mean ± SD, n = 3). *P < 0.05 or ***P < 0.001 vs. no EVs (C) or shNTC EVs (E); ##P < 0.01 vs. 20% O2 EVs (C) (by 1-way ANOVA with Tukey’s multiple-comparison test). (F) hCMEC/D3 ECs were seeded on Boyden chamber filters, cultured to confluence, and incubated with EVs from MDA231-BrM2 subclones for 24 hours. FITC-dextran was added to the upper chamber, and fluorescence in the lower chamber was measured 20 minutes later using a plate reader. Data are shown as mean ± SD (n = 3). ****P < 0.0001 vs. no EVs; ####P < 0.0001 vs. 20% O2 shNTC-EVs; &&&&P < 0.0001 vs. 1% O2 shNTC-EVs (2-way ANOVA followed by Tukey’s multiple-comparison test).

To investigate the effect of EVs on the permeability of brain ECs, we treated hCMEC/D3 monolayers in Boyden chambers with EVs from shNTC, shHIF1A, and shITGB3 subclones of MDA231-BrM2 cells exposed to 20% or 1% O2 for 48 hours. FITC-dextran was added to the upper chamber, and fluorescence in the bottom chamber was measured using a plate reader. EVs from shNTC cells increased EC permeability; EVs from hypoxic shNTC cells had the greatest effect (Figure 8F). The increased permeability induced by EVs from hypoxic shNTC cells was not observed when ECs were treated with EVs from hypoxic shHIF1A or shITGB3 cells (Figure 8F). Taken together, the data in Figure 8 indicate that ITGB3+ EVs promote interaction of BC cells with brain ECs and increase endothelial permeability.

ITGB3+ EVs increase permeability of brain ECs by activating VEGFR2 signaling. Vascular endothelial growth factor (VEGF) receptor 2 (VEGFR2) signaling is the dominant pathway that regulates the permeability of ECs (54), and signaling via integrin αvβ3 has been reported to augment VEGFR2 signaling (55–57). BC cell expression of αvβ3 was also shown to support colonization after intracranial implantation (58). We hypothesized that ITGB3+ EVs increase permeability by stimulating VEGFR2 signaling. To test this hypothesis, we exposed hCMEC/D3 brain ECs to EVs isolated from MDA231-BrM2 cells exposed to 20% or 1% O2 for 48 hours. After 24 hours, the ECs were treated with recombinant human VEGFA165 protein, and cell lysates were prepared. Immunoblot assays revealed that BC cell–derived EVs enhanced VEGFA-induced VEGFR2 phosphorylation/activation, and the effect was augmented when the EVs were derived from hypoxic BC cells (Figure 9A). Furthermore, the augmented VEGFR2 activation by EVs from hypoxic shNTC cells was attenuated when the EVs were isolated from shHIF1A or shITGB3 cells (Figure 9B).

ITGB3+ EVs increase permeability of brain ECs by augmenting VEGFR2 signaling. (A) hCMEC/D3 cells were incubated for 24 hours with or without EVs (isolated from MDA231-BrM2 cells exposed to 20% or 1% O2 for 48 hours), followed by stimulation with VEGFA165 for 30 minutes. Whole-cell lysates were prepared for immunoblot assays. (B) hCMEC/D3 cells were incubated for 24 hours with or without EVs (isolated from MDA231-BrM2 subclones that were exposed to 1% O2 for 48 hours), followed by stimulation with VEGFA165 for 30 minutes, and immunoblot assays were performed. (C and D) hCMEC/D3 cells were seeded into Boyden chamber inserts, cultured to confluence, and incubated with or without EVs from empty vector (Control) or ITGB3-overexpressing (ITGB3-OE) MDA231-BrM2 subclones in the presence of vehicle or 5 μM sunitinib for 24 hours. FITC-dextran was added to the upper chamber, and fluorescence in the lower chamber was measured 20 minutes later and presented as mean ± SD (n = 3). ns, not significant vs No EVs without sunitinib; **P < 0.01 vs. control EVs without sunitinib; ****P < 0.0001 vs. ITGB3-OE EVs without sunitinib (2-way ANOVA followed by Tukey’s multiple-comparison test). (E and F) A monolayer of hCMEC/D3 cells was incubated for 24 hours with EVs (isolated from MDA231-BrM2 subclones that were exposed to 20% or 1% O2 for 48 hours), and then MDA231-BrM2 cells were added. BC cells that transmigrated through the EC monolayer to the lower chamber were stained with crystal violet and imaged by light microscopy (E; scale bars: 100 μm). The area of stained cells was quantified using ImageJ (F; mean ± SD, n = 3). ***P < 0.001 vs. 20% O2/shNTC-EVs; ####P < 0.0001 vs. 1% O2/shNTC-EVs (2-way ANOVA followed by Tukey’s multiple-comparison test).

To investigate whether VEGFR2 signaling is implicated in the increased endothelial permeability induced by ITGB3+ EVs (Figure 8F), we treated hCMEC/D3 ECs with EVs isolated from control (empty vector) MDA231-BrM2 cells or from MDA231-BrM2 cells overexpressing ITGB3 (ITGB3-OE cells) in the presence or absence of the VEGFR2 inhibitor sunitinib (Figure 9C). Compared with EVs from control cells, EVs from ITGB3-OE cells provoked significantly increased permeability (Figure 9D). Furthermore, coadministration of sunitinib (59) markedly blunted the effect of EVs from ITGB3-OE cells, whereas the effect of sunitinib when coadministered with EVs from control cells was more modest (Figure 9D). These results are consistent with the hypothesis that ITGB3+ EVs stimulate augmented VEGFR2 signaling, leading to increased endothelial permeability. This increased permeability should facilitate the transmigration of BC cells through the endothelium. To test this hypothesis, an endothelial monolayer was treated with EVs from shNTC, shHIF1A, or shITGB3 cells exposed to 20% or 1% O2, followed by the addition of BC cells to the Boyden chamber. EVs from hypoxic shNTC cells augmented BC cell transmigration across the endothelial monolayer, and this effect was abolished when EVs from hypoxic shHIF1A or shITGB3 cells were used (Figure 9, E and F).