VCAM-1 mediates binding of tumor-associated macrophages to cancer cells

The extracellular region of VCAM-1 consists of 7 immunoglobulin (Ig) domains, one of which (domain 4) is missing in the VCAM-1-6D splice variant (Cybulsky et al., 1991). Alpha-4 integrins bind to separate sites on Ig domains 1 and 4 (Figure 2A), and the VCAM-1-6D variant therefore is of low-affinity (Vonderheide et al., 1994). By means of isoform-specific RT-PCR we determined that MDA231 and another human metastatic breast cancer cell model, CN34 (see below), predominantly express the long, high-affinity VCAM-1 variant (Figure S2A). To determine whether VCAM-1 on breast cancer cells can bind leukocytes, we incubated MDA231 monolayers with a suspension of U937 cells, a human monocyte cell line that expresses α4β1 integrin (Kalogeris et al., 1999). The MDA231-LM2 cells bound over 3-fold as many U937 cells as did the parental MDA231 cells (Figure 2B and C). shRNA-mediated depletion of VCAM-1 in MDA231-LM2 cells or addition of a blocking antibody against α4 integrin inhibited the binding of U937 cells (Figure 2D). We concluded that VCAM-1 on cancer cells can bind monocytes.

Open in a separate window

Figure 2

VCAM-1 on cancer cells binds tumor-associated macrophages

(A) Schematic illustration of VCAM-1 interactions with leukocyte α4 integrins.

(B,C) Adhesion assays of U937 cells on monolayers of parental MDA231 or MDA231-LM2 cells. (B) Representative photomicrographs of adherent TRITC-labeled U937 cells on the monolayer of indicated cancer cells. Scale bar, 100 μm. (C) quantification of the ratio of adherent U937 cells and cancer cells. Data are average ± SEM; n = 10 unit areas.

(D) U937 cells treated with anti-α4β1 integrin antibody (anti-α4) or Ig control (-) were tested for binding to monolayers of control or VCAM1-depleted MDA231-LM2. VCAM-1 knockdown was done using two independent shRNAs (sh1, sh2). Data are average ± SEM; n = 10 unit areas.

(E) Recombinant VCAM-1 (rVCAM-1) binging assay with different stromal cell populations from lung metastatic nodules, including F4/80⁺ macrophages (Macro), Gr-1⁺ neutrophils (Neutro) and CD31⁺ endothelial cells (Endoth) and GFP⁺ cancer cells. Cell surface binding of bovine serum albumin control (grey closed histogram) or rVCAM-1 (black open histogram) was determined by flow cytometry.

(F) Adhesion assays of stromal cells from lung metastatic nodules on monolayers of control or VCAM1-depleted (sh1, sh2) MDA231-LM2 cells. The ratio between adherent F4/80⁺ macrophages to GFP⁺ cancer cells was quantified by flow cytometry.

See also Figure S2.

Monocytic cells including macrophages enhance tumor progression (Joyce and Pollard, 2009; Qian and Pollard, 2010). Blockage of macrophage recruitment significantly decreases lung colonization by MDA231-LM2 cells in mice (Qian et al., 2009; Qian et al., 2011). However, the specific contribution of macrophages to metastatic colonization is not well understood. To identify metastasis stromal cell types that bind VCAM-1 we generated lung metastatic nodules by tail-vein inoculation of MDA231-LM2 cells expressing green fluorescent protein (GFP). Single-cell suspensions prepared from the resulting nodules were incubated with fluorochrome-conjugated recombinant VCAM-1 ectodomain (rVCAM-1) or bovine serum albumin as a control. Specific cell surface markers were used to identify and quantify different types of stromal cells (Figure S2B). VCAM-1 bound to metastasis-associated CD45⁺F4/80⁺ macrophages and CD45⁻ CD31⁺ vascular endothelial cells, but not to Gr1⁺ neutrophils or GFP⁺ cancer cells (Figure 2E). Next we prepared cell suspensions from lung metastasis nodules, depleted these suspensions of human EpCAM⁺ cancer cells, and panned the remaining cells over control or VCAM-1-depleted MDA231-LM2 monolayers. Analysis of the monolayer-bound fraction demonstrated VCAM-1-dependent binding of F4/80⁺ cells to MDA231-LM2 cells (Figure 2F).

In 4T1 lung metastatic nodules, CD45⁺CD4⁺ and CD45⁺CD8⁺ lymphocytes expressed low levels of cell surface α4 integrin compared to CD45⁺F4/80⁺ macrophages (Figure S2C). Macrophages comprised approximately 7% of the total cell population in the metastatic nodules (Figure S2C), whereas CD4+ and CD8+ lymphocytes collectively comprised 4% of the tumor cell population. Moreover, endothelial cells, which also bind to rVCAM-1 (Figure 2E), comprised <1% of the cell population from MDA231-LM2 lung nodules (Figure S2B). Overall, macrophages were the most abundant source of potential α4 integrin contacts in these lung nodules.

Cancer cell VCAM-1 is dispensable for extravasation and macrophage recruitment

Contacts with macrophages enhance cancer cell extravasation (Qian et al., 2009). Although U937 monocytes enhanced the ability of MDA231-LM2 cells to migrate through human pulmonary microvascular endothelial cell (HPMECs) monolayers, VCAM-1 depletion did not influence this process. This was observed when the monocytes were activated with PMA to induce adherence (Yamamoto et al., 2009) and were placed on the lower side of the transwell membrane (Figure 3A). Similar results were obtained when monocytes stimulated with cyclic-AMP were placed with cancer cells in the upper transwell chamber, with complement factor C5a as a monocyte chemoattractant (Rubin et al., 1991) in the bottom chamber (Figure S3A). U937 also had no effect when placed together with cancer cells in the upper transwell chamber, with CXCL12 as an MDA231-LM2 cell chemoattractant (Zhang et al., 2009) in the bottom chamber (Figure S3B). These results are consistent with our in vivo evidence that VCAM1 is not required for entry of breast cancer cells into the lung parenchyma.

Open in a separate window

Figure 3

Cancer cell VCAM-1 is dispensable for transendothelial migration or leukocyte recruitment but enhances tumor seeding

(A) MDA231-LM2 cell migration across monolayers of human pulmonary microvascular endothelial cells, with or without activated monocytes in the lower chamber of a transwell. Migrated cancer cells were quantified. Data are average ± SEM; n = 10 unit areas.

(B-E) Association of cancer cells with leukocytes in the lungs. (B) Schematic of the experimental procedure. (C) CMFDA-labeled 4T1 cells were intravenously inoculated into BALB/c mice and the association between cancer cells and CD4⁺ and CD8⁺ T cells and CD68⁺ macrophages was analyzed 24 h later. Shown are two representative micrographs per experimental group. Scale bar, 10 μm. (D) Percentage of cancer cells interacting with the indicated leukocytes cells. (E) Quantification of CD68⁺ cells surrounding cancer cells. Data are average ± SEM; n >30 cancer cells.

(F-H) Tumor self-seeding assay. (F) Schematic of experimental procedure. GFP/luciferase expressing cancer cells of inoculation of into the arterial circulation of mice bearing primary tumors formed by unlabeled cancer cells. (G) Representative images of mammary tumors seeded with indicated GFP⁺ MDA-231 LM2 cells. GFP⁺ cancer cells and tumor-associated macrophages were stained with anti-GFP and anti-CD68 antibodies in primary tumors (week 4). Scale bar, 10 μm. (H) Seeding of control (Ctrl) or VCAM-1 depleted (sh1 and sh2) MDA231-LM2 cells in mammary tumors after intracardiac inoculation was determined by BLI quantification. Data are average ± SEM; n = 10 mice.

See also Figure S3.

To determine if VCAM-1 on cancer cells influenced the recruitment of macrophages into the tumor, we labeled 4T1 or MDA231-LM2 cells with CellTracker™ Green CMFDA, and inoculated these cells into mice via the tail vein. After 24 h, we examined the association of CD4⁺ and CD8⁺ T lymphocytes (for 4T1 model) and CD68⁺ macrophages (for both 4T1 and MDA231-LM2 models) with CMFDA⁺ cancer cells in the lung parenchyma (Figure 3C, Figure S3C). Over 95% of cancer cells were surrounded by macrophages in close juxtaposition while fewer than 30% of cancer cells were associated with CD4⁺ or CD8⁺ T lymphocytes (Figure 3D). VCAM-1 depletion in cancer cells changed neither the percentage of cancer cells surrounded by immune cells nor the average number of interacting macrophages per cancer cell (Figure 3D,E). Furthermore, the lung nodules formed by the VCAM-1-depleted cells showed the same content of F4/80⁺, CD31⁺ and Gr1⁺ cells as did size-matched control nodules (Figure S2B). Thus, VCAM-1 expression in cancer cells is not required for their recruitment of, or exposure to macrophages.

VCAM-1 in cancer cells enhances the seeding of breast tumors

We further tested the role of VCAM-1 under conditions that challenged cancer cells to seed new or established tumors. VCAM-1 depletion significantly diminished the rate of tumor outgrowth when only 5×10⁴ cells were implanted in mammary fat pad tumor assays (Figure S3D-F, compare to tumor formation by 5×10⁵ implanted cells in Figure 1C). Recently we showed in experimental models that established breast tumors are powerful magnets for CTCs (Kim et al., 2009). In this process, called “self-seeding”, CTCs that pass through a breast tumor can effectively reinvade the tumor by virtue of their metastatic abilities, the leaky tumor vasculature, and the attraction of inflammatory cytokines IL6 and IL8 in the tumor microenvironment. Once in that environment, aggressive seeder clones can attract myeloid cells through the release of CXCL1 and expand in that myeloid-rich microenvironment (Kim et al., 2009). In a tumor self-seeding assay (Figure 3F), MDA231-LM2 seeder cells were indeed surrounded by CD68⁺ macrophages (Figure 3G). MDA231 seeder cells that were depleted of VCAM-1 showed a marked decrease in the ability to populate the recipient tumors (Figure 3H). Collectively these results suggest that VCAM-1 supports the outgrowth of cancer cells that infiltrate the lungs and other restrictive sites. Moreover, these results raised the possibility that VCAM-1–leukocyte interactions are a source of this advantage.

VCAM-1 protects breast cancer cells from apoptosis in the lung microenvironment

To determine how VCAM-1 expression in breast cancer cells contributes to tumor outgrowth in the lungs, we transduced a doxycycline-driven VCAM1 shRNA expression system in MDA231-LM2 cells. Doxycycline addition to the culture media decreased VCAM1 expression within 48h, whereas a full recovery of VCAM-1 expression after doxycycline removal took 5 days (Figure 4A,B). Cells cultured without doxycycline were inoculated into tail vein, and one day after the inoculation doxycycline was administered to the mice for the remainder of the experiment (Figure 4C). This treatment caused a sustained inhibition of lung metastatic colonization. In a converse experiment, cancer cells were incubated with doxycycline for 3 days in order to suppress VCAM-1 expression prior to inoculation of these cells into mice. The recipient mice were not treated with doxycyclin. Metastatic colonization of the lungs under these conditions was as extensive as with control MDA231-LM2 cells (Figure 4D). Therefore, VCAM-1 was dispensable for cancer cell entry into the lungs but was required for their subsequent outgrowth.

Open in a separate window

Figure 4

VCAM-1 protects cancer cells from apoptosis in the lung microenvironment

(A, B) MDA231-LM2 cells with a doxycycline inducible shVCAM-1 vector were cultured in the presence of doxycycline (A) or were placed in doxycycline-free medium after culturing with doxycycline (B), in order to determine the kinetics of VCAM1 depletion and recovery, respectively. (C) Control or doxycycline-driven VCAM1 shRNA cells were inoculated into mice, and all mice were treated with doxycycline. Lung colonization was assessed by quantitative BLI. (D) BLI quantification of lung colonization after the indicated cancer cells pretreated with doxycycline were injected into mice that were not given doxycycline. The mice were not administrated with drug. Data are average ± SEM; n = 8 mice.

(E) TUNEL staining and quantification in the lungs harvested 6 weeks after intravenous injection of cancer cells as indicated in Figure 1E. Representative photomicrographs, and quantification of TUNEL⁺ cells. Data are average ± SEM; n = 10 unit area/mouse, 3 mice/group. Scale bar, 50 μm.

(F) The indicated cancer cells were injected into the fourth mammary fat pad of mice and the lungs were harvested as indicated in Figure 1A. Immunostaining for human vimentin and cleaved caspase 3 were performed on consecutive lung tissue sections. Scale bar, 100 μm.

See also Figure S4.

VCAM-1 depletion caused a marked increase in the proportion of apoptotic cells in the lung metastatic colonies (Figure 4E). We also observed a high incidence of apoptosis in lung nodules seeded from mammary tumors (Figure 4; refer to protocol in Figure 1B). We saw no effect of VCAM-1 depletion on the expression of the cell proliferation marker Ki67 or the endothelial cell marker CD31 in lung nodules that were either size-matched or time-matched relative to controls (Figure S4A-C). The lesions formed by VCAM-1-depleted cells remained low in VCAM-1 expression compared to size-matched control lesions, as determined by qRT-PCR analysis of lung nodules (Figure S4D). Collectively the results suggest that VCAM-1 specifically enhances the survival of breast cancer cells that infiltrate the lung parenchyma.

The VCAM-1 intracellular domain mediates metastatic cell survival

We investigated the mechanism underlying this unprecedented role of VCAM-1. We generated a truncated form, VCAM-1(Δcyt), lacking the cytoplasmic region (Figure 5A) and compared this construct with the wild-type VCAM-1 for the ability to restore lung metastatic activity in VCAM-1-depleted cancer cells. In addition to the MDA231-LM2 model system we also used the CN34-LM2 cell line. This line is a lung metastatic derivative from the pleural fluid of a patient with stage IV metastatic breast cancer (Tavazoie et al., 2008). The VCAM-1 and VCAM-1(Δcyt) constructs were expressed at similar levels in these two cell lines, as determined by immunoblotting analysis (Figure 5B, and data not shown) and flow cytometry (Figure 5C). Both constructs similarly increased the ability of MDA231-LM2 and CN34-LM2 cells to bind U937 monocytes (Figure 5D). However, whereas the expression of wild-type VCAM-1 increased the ability of both cell lines to colonize the lungs (Figure 5E) and to avert apoptosis (Figure 5F), VCAM-1(Δcyt) was defective in these functions in both cell lines (Figure 5E, F).

Open in a separate window

Figure 5

The VCAM-1 intracellular domain mediates survival of cancer cells in the lungs

(A) Domain structure of the wild type VCAM-1 with extracellular (EC), transmembrane (TM) and intracellular (IC) domains and VCAM-1 (ΔCyt) construct lacking the IC. WT or ΔCyt VCAM-1 were overexpressed in VCAM1-depleted (by shRNA 1) MDA231-LM2 cells (MDA-LM2-sh1) and CN34-LM2 cells.

(B,C) Expression of the wild type VCAM-1, VCAM-1(ΔCyt) or empty vector (Ctrl) in MDA231-LM2 cells that were stably depleted of endogenous VCAM-1 by the expression of shRNA1 and CN34-LM2 cells. (B) VCAM-1 was detected by western immunoblotting in MDA231-LM2 cells. (C) The expression of VCAM-1 on the cell surface was detected by flow cytometry.

(C) U937 cell adhesion assays on monolayers of the indicated MDA231 or CN34 cell lines. Data are average ± SEM; n = 10 unit areas.

(D, E) The indicated cell lines were intravenously injected into mice. 6 weeks after tumor cell injection, lung BLI was measured (D), and TUNEL⁺ cells were quantified in lung tissue sections (E). Data are average ± SEM; n = 5 mice.

Engagement of VCAM-1 triggers survival signals

The interaction of leukocyte α4 integrin with VCAM-1 induces cytskeletal rearrangements in endothelial cells, and this effect can be mimicked with the use of anti-VCAM-1 crosslinking antibodies (van Wetering et al., 2003). We adopted this approach to test the effect of VCAM-1 clustering on the apoptotic responsiveness of cancer cells to TRAIL (tumor necrosis factor-related apoptosis-inducing ligand), a pro-apoptotic cytokine expressed in human lung metastatic nodules (Zhang et al., 2009). In both MDA231-LM2 and CN34-LM2 cell cultures, addition of TRAIL induced the activation of effector caspase-3, as determined by the accumulation of cleaved caspase-3 (Figure 6A) (Nicholson et al., 1995). Incubation with anti-VCAM-1 antibody and secondary F(ab’)₂ fragment markedly blunted this effect in the cells expressing full-length VCAM-1 but, importantly, not in cells expressing VCAM-1(Δcyt) (Figure 6A).

Open in a separate window

Figure 6

VCAM-1 engagement by α4 integrins triggers anti-apoptotic signals via the cytoplasmic domain

(A) VCAM-1 clustering antibodies (Ab X-linking) were added to the indicated cell lines in the presence or absence of 10 ng/ml TRAIL for 3 h. Caspase-3 activation was detected by western immunoblotting.

(B-D) The indicated cancer cells were co-cultured with U937 cells in the presence of TRAIL and α4 integrin antibody (anti-α4) or comparable Ig control (-). (B) After 24 h, TUNEL⁺ (red) and green CMFDA-labeled cancer cells were scored under fluorescent microscope. Arrow indicates a TUNEL⁺/CMFDA⁺ cancer cell and asterisk indicates a (CMFDA-negative) U937 cell. (C) Percentage of TUNEL⁺/CMFDA⁺ cancer cells. Data are average ± SEM; n = 10 unit areas; p value calculated using t test. Scale bar, 20 μm. (D) Cancer cells were quantified after 48 h of co-culture. Data are average ± SEM; n = 4 samples.

To determine the effect of engaging VCAM-1 with its natural α4 integrin counter-receptors we incubated cancer cells with U937 monocytes. Cancer cells expressing VCAM-1, VCAM-1(Δcyt), or no exogenous VCAM-1 were pre-labeled with CMFDA, and apoptosis was assessed by TUNEL staining (Figure 6B). Expression of VCAM-1 reduced to half the incidence of apoptosis in cancer cells that were exposed to TRAIL, whereas VCAM-1(Δcyt) provided no protection (Figure 6C). Furthermore, addition of anti-α4 integrin blocking antibody prevented VCAM-1 from exerting this anti-apoptotic effect (Figure 6C). Cancer cell counts corroborated these findings (Figure 6D). Thus, upon engagement by α4 integrin on leukocytes, VCAM-1 in breast cancer cells delivers anti-apoptotic signals via its cytoplasmic tail.

VCAM-1 clustering activates Akt signaling

To identify signaling pathways that are activated by VCAM-1 engagement, we performed antibody-mediated clustering of VCAM-1 in MDA231 and CN34 cells under low concentration of serum growth factors. Cell lysates were probed with phosphopeptide-directed antibodies that detect the activation state of 46 protein kinases and their protein substrates (Phospho-Kinase Array Kit) as well as Smad2/3 and NF-κB (Table S1). Among these signal transducers, only Akt phoshorylation at S473 showed a consistent increase in response to VCAM-1 clustering (Table S1). S473 phosphorylation is an Akt activating event triggered by PI3K stimulation and catalyzed by the TORC2 complex (Alessi et al., 1996; Collins et al., 2003; Sarbassov et al., 2005). S473 phosphorylation by TORC2 is thought to prime Akt for further phosphorylation and activation by 3-phosphoinositide dependent protein kinase 1 (PDK1) at T308 (Zoncu et al., 2011). Using western immunoblotting we confirmed that VCAM-1 clustering induces Akt phosphorylation at S473 in human MDA231, CN34 and mouse 4T1 breast cancer cells (Figure 7A,B). S473 phosphorylation peaked 10-15 min after VCAM-1 clustering. No Akt phosphorylation was detected in MDA231 or CN34 cells expressing VCAM-1(Δcyt) (Figure 7A). Addition of the PI3K inhibitor LY-294002 abolished the induction of Akt phosphorylation by VCAM-1 clustering (Figure 7B). Furthermore, in the presence of LY-294002, VCAM-1 clustering no longer suppressed caspase 3 activation by TRAIL (Figure 7C). Collectively, these results suggest that engagement of VCAM-1 on breast cancer cells triggers activation of the PI3K/Akt pathway and protects cancer cells from death signals.

Open in a separate window

Figure 7

VCAM-1 clustering activates PI3K/Akt signaling with the involvement of Ezrin

(A) Time course of Akt phosphorylation at the TORC2 site S473 after addition of VCAM-1 clustering antibodies. The indicated human MDA-231 and CN34 and mouse 4T1 breast cancer cells were used in low-serum media. The ratio of phospho-Akt to total Akt is shown below each lane.

(B) Cancer cells overexpressing VCAM-1 were pretreated with PI3K inhibitor LY294002 (LY) or vehicle DMSO (-) for 30 min. VCAM-1 antibody clustering was applied for 10 min and activated Akt (S473) was detected by western blotting. The ratio of phospho-Akt to Akt is shown below each lane.

(C) Cancer cells overexpressing VCAM-1 were pretreated with LY294002 (LY) or vehicle (-) for 30 min and VCAM-1 clustering antibodies and TRAIL (10 ng/ml) were added for 3 h as indicated. Caspase-3 activation was detected by western immunoblotting.

(D) Time course of Ezrin phosphorylation at Y353 and Ezrin-Akt interaction after addition of VCAM-1 clustering antibodies to the indicated cells. The indicated proteins were analyzed by western immunoblotting of whole cell lysates (input) or anti-Ezrin immunoprecipitates (IP: Ezrin).

(E) Time course of Akt association with the cell membrane fraction after addition of VCAM-1 clustering antibodies to the indicated cells. Immunoblotting with antibodies against EGFR and β-actin served as markers of membrane and cytosolic fractions, respectively.

(F) Model of VCAM-1 mediated Akt activation. VCAM-1 clustering triggers Ezrin phosphorylation at Y353, which is a known docking site for PI3K. This event is accompanied by Akt binding to Ezrin, Akt association with the cell membrane, and Akt phosphorylation at the TORC2 site S473 for propagation of survival signals.

See also Figure S5 and Table S1.

VCAM-1 clustering triggers Akt binding to Ezrin

The short intracellular tail of VCAM-1 and of intercellular cell adhesion molecules 1~3 (ICAM-1~3) interact with Ezrin, a member of ERM (Ezrin/Radixin/Moesin) protein family (Fehon et al., 2010). ERM are cytoplasmic proteins that function as a link between certain transmembrane proteins and the actin cytoskeleton (Fehon et al., 2010). In endothelial cells, the interaction of VCAM-1 with Ezrin triggers cytoskeletal rearrangements for the translocation of VCAM-1-bound leukocytes from blood into tissue (Barreiro et al., 2002; van Wetering et al., 2003). In addition to this role, Ezrin can signal cell survival through the PI3K pathway (Gautreau et al., 1999). Antibody-mediated crosslinking of ICAM-2 triggers phosphorylation of Ezrin at Y353, which results in binding and activation of PI3K (Perez et al., 2002). Y353 is lacated within the alpha-helical domain of Ezrin, and Y353 phosphorylation is thought indirectly involve Src kinases (Chuan et al., 2006; Krieg and Hunter, 1992). Based on these clues, we hypothesized that VCAM-1 engagement may trigger Akt activation though Ezrin. Indeed, antibody-mediated clustering of VCAM-1 in MDA231-LM2 or CN34 cells in low-serum conditions led to a rapid phosphorylation of Ezrin at Y353 (Figure 7D). VCAM-1(Δcyt) failed to mediate this effect (Figure 7D). Moreover, this effect was accompanied by the formation of a complex between Ezrin and Akt (Figure 7D) and an enhanced recruitment of Akt to the membrane (Figure 7E, Figure S5). The results suggest that VCAM-1 signaling activates Ezrin Y353 phosphorylation to help PI3K recruit Akt to the membrane, leading to TORC2 activation of Akt signaling (summarized in Figure 7F).

Cell-cell binding assays

2×10⁶ prelabeled U937 cells were allowed to adhere to the monolayer of cancer cells. Adherent cells (red) and the nucleus of total cells (DAPI staining in blue) were scored by fluorescence microscopy. The ratio of adherent U937 cells to cancer cells was calculated as red counts/(blue counts – red counts). In the indicated experiments, U937 cells were pre-incubated with α4 integrin antibody (clone PS/2 from AbD Serotec) or Ig control for 30 min. To identify the cell types that adhere to cancer cells, human EpCAM-depleted stromal cell suspensions from lung metastatic nodules were allowed to adhere to GFP⁺ cancer cell monolayers. Single-cell suspensions of cancer cells and adherent cells were incubated with fluorochrome-conjugated F4/80 antibody. F4/80⁺ macropahges and GFP⁺ cancer cells were quantified using flow cytometry.

VCAM-1 clustering assays

VCAM-1 in human cancer cells were crosslinked with 10 μg/ml of mouse anti-human VCAM-1 monoclonal antibody (clone 1G11, Beckman Coulter) and 50 μg/ml goat anti-mouse IgG₁ F(ab’)₂ (Jackson ImmunoResearch) for indicated time frames while 10 μg/ml of mouse anti-human VCAM-1 monoclonal antibody (Clone MVCAM.A429, AbD Serotec) and 50 μg/ml goat anti-rat IgG₁ F(ab’)₂ (Jackson ImmunoResearch) were used in 4T1 mouse breast cancer cells. Mouse or rat IgG was used as control. For apoptosis analysis, cancer cells were incubated with 100 ng/ml TRAIL (Peprotech) for 3 h. Caspase 3 was detected by western immunoblotting. To identify signaling mediators, VCAM-1 clustering was performed in serum-starved human breast cancer cells for the indicated time periods. Where indicated, 50 μM LY294002 or DMSO control was added during the incubation with anti-VCAM-1 and F(ab’)₂. Cell lysates were analyzed using Human Phospho-Kinase Array Kit (R&D Systems). In the indicated experiments, Western blot was performed using antibodies against phospho-Akt (S473), Akt, phospho-NF-κB p65 (S468) or phospho-Smad2 (S465/467) and phospho-Smad3 (S423/425)(all from Cell Signaling). The ratio of phospho-Akt to total Akt was quantified by Image J software (Abramoff et al., 2004). For co-immunoprecipitation assays, cell pellets were lysed with IP buffer (Perice). Equal amounts of protein were incubated with anti-ezrin (clone 3C12, Abcam) and protein G (GE Healthcare). Ezrin, phospho-ezrin (Y353, BD Pharmingen) and Akt were detected in the immunoprecipitates. For cell fractionation, membrane and cytosolic fractions were harvested using Subcellular Protein Fractionation kit (Peirce). Akt was detected in each fraction while EGFR or β-actin (Cell Signaling) were used as the loading control for membrane or cytosolic fraction, respectively.

To clustering VCAM-1 in tumor cells by physiologically interacting with α4 integrin in macrophages, confluent cancer cells were co-cultured with U937 cells (2.5×10⁵ U937 cells/cm²) in the presence of 20 ng/ml TRAIL. To detect the percentage of apoptotic cancer cells, cancer cells were pre-labeled with green CMFDA before the co-culture. TUNEL staining was performed 24 h after incubation and observed under the fluorescent microscope. To quantify survived tumor cell, cells were lysed by Passive Lysis Buffer (Promega) and the luciferase activity was detected by Luciferase Assay System (Promega) using Glomax 96 Microplate Luminometer (Promega). Standard curve was generated by detecting the luciferase activity of a series numbers of cancer cells and the absolute numbers of survived cancer cells were calculated. In the indicated experiments, functional blocking antibody of α4 integrin (10 μg/ml) was applied in the co-culture system.

Analysis of human metastatic tissue samples

VCAM1 transcript level was analyzed in 18 lung, 19 brain and 17 bone metastatic samples from breast cancer patients (GSE14020) (Zhang et al., 2009). A correlation coefficient was calculated between VCAM1 and CD31 expression in each dataset. A leukocyte gene expression signature, including 182 genes, was based on the SAGE data from CD45⁺CD10⁻ leukocytes isolated from breast tumors (Allinen et al., 2004). After mapping these genes on the Affymetrix U133A platform, the probe sets of these genes were queried in a dataset from 67 metastatic samples from breast cancer patients (GSE14020)(Zhang et al., 2009). We clustered all these samples using the heatmap.2 function in the R statistical package. The most robust subcluster was defined as the leukocyte signature high group. VCAM1 expressions were compared between these two groups.

We thank J. Joyce and members of the Massagué lab for insightful discussion and technical suggestions, and the MSKCC Molecular Cytology Core Facility for histological sample staining and analysis. This work was supported by NIH grants CA126518 and CA94060, and the Alan and Sandra Gerry Metastasis Research Initiative. QC is supported by Life Sciences Research Foundation Fellowship. JM is an investigator of the Howard Hughes Medical Institute.