(A) The indicated epithelial cell lines were incubated for 3 h with TGFβ and then total RNA was subjected to microarray analysis. The heat map represents the change in expression levels of the 153 genes within the TBRS.
(B) TBRS status was assessed in a MSK/EMC cohort of 368 primary breast cancer tumors with known lung or bone metastatic outcomes. Red denotes a strong correlation between individual tumor gene expression profiles and the TBRS while blue indicates no correlation. Estrogen receptor α(ER) expression status is also indicated. Blue and red marks above the heat map indicate tumors that at any time developed bone or lung metastases, respectively.
(C) Kaplan-Meier curves representing the probability of cumulative lung (left panel) or bone (right panel) metastasis-free survival for this cohort. Tumors are categorized according to their TBRS and ER status. The P values for the ER-negative tumor comparisons are shown.
(D) Hierarchical clustering was performed on the MSK/EMC cohort with the indicated pathological and genomic markers including the TBRS, the lung metastasis signature (LMS), the wound response signature (Wound), the 70-gene prognosis signature (70-gene), size (Size >2cm), the basal molecular subtype (Basal), and the ER status. Red marks above the map indicate tumors that developed lung metastasis.
(E) Lung metastasis-free survival restricted to patients with ER-negative tumors. Patients were categorized according to their TBRS and LMS status. P value shown for the LMS+ tumor comparisons.

(A) Schematic of lung metastasis assay from an orthotopic breast cancer inoculation.
(B) Immunoblots using indicated antibodies were performed on whole-cell extracts from control, Smad4 knockdown, and Smad4-Rescue LM2 cells.
(C) Mice injected with 5×10⁵ cells into the fourth mammary fat pads were measured for tumor size at day 28. n=14; error bars indicated s.e.m
(D) Blood from tumor-bearing mice was isolated and red blood cells lysed. RNA from the remaining cells was extracted for qRT-PCR. The presence of circulating tumor cells was assessed as a function of human-specific GAPDH expression relative to murine β2-microglobulin, in 3 mL of mouse blood perfusate. n=8; error bars indicate s.e.m.
(E) Bioluminescent quantification of lung seeding elicited from orthotopically implanted breast tumors. Orthotopic tumors were grown to approximately 300mm³, mastectomies were performed and lung seeding was quantified using bioluminescence imaging seven days later. n=7–15; error bars indicated s.e.m; p-values calculated using the one-tailed unpaired t-test.
(F) Representative bioluminescent images (inset) and lung histology of mice with the median value of bioluminescence in each experimental group in (D). Breast cancer cells were stained with human vimentin.

(A) LM2 cells and a clinically-derived pleural effusion sample (CN34.2A) were pretreated with TGFβ for 6 h. LM2 (2×10⁵) and CN34.2A (4×10⁵) cells were injected into the lateral tail vein and lung colonization was analyzed by in vivo bioluminescence imaging.
(B) Bar graph represents 24h time point measurements of the normalized photon flux from animals injected with either LM2 or CN34.2A cells. In vivo bioluminescent mouse images shown are from representative animals. n=7 for the LM2 experiment while n=6 for the CN34.2A experiment; error bars indicate s.e.m; p-values calculated using the one-tailed unpaired t-test.
(C) Bone colonization assays were performed by intracardiac injection of LM2 or BoM-1833 cells (3×10⁴). Samples were pretreated with 100 pM of TGFβ for 6 h and compared to an untreated control. Plot represents normalized photon flux from mouse hindlimbs
(D) Bar graphs represent seven-day time point analysis of the normalized photon flux from the mouse hind limbs. n=8; error bars indicate s.e.m.

(A) Microarray and qRT-PCR analysis for the four epithelial cell lines treated with TGFβ. Fold change values for the TGFβ induction of ANGPTL4 are indicated.
(B) Box-and-whisker plot comparing ANGPTL4 and NEDD9 TBRS-negative and -positive ER-negative tumors from the MSK/EMC cohorts. P value was calculated using the Wilcoxon rank sum test.
(C) TGFβ-induced changes in the mRNA expression of LMS genes in a panel of clinically derived pleural effusion samples and LM2 cells. Cells were treated with 100 pM of TGFβ for 3 h and analyzed by qRT-PCR using primers for the indicated genes. ER status for each breast cancer patient is designated.
(D) LM2 breast cancer cells were treated with 100pM of TGFβ for the indicated lengths of time and ANGPTL4 mRNA levels were analyzed using qRT-PCR.
(E) Treatment of LM2 (left panel) and pleural effusion-derived CN37 sample (right panel) with TGFβ and the TGFβ-receptor kinase inhibitor, SB431542. qRT-PCR expression levels are shown relative to the untreated control sample.
(F) MDA-231, LM2 control, LM2-Smad4-depleted, and LM2-Smad4-Rescue cell lines were treated with 100 pM TGFβ for 3 h. TGFβ-induced fold changes of ANGPTL4 were analyzed by qRT-PCR analysis, n=3; error bars indicate standard deviation.

(A) Secreted Angptl4 protein levels in the control, ANGPTL4 knockdown, or ANGPTL4-rescue LM2 cells were analyzed by ELISA. n=3; error bars indicate standard deviation
(B) Tumors size measurements were taken from mice inoculated into the fourth mammary fat pads with 5×10⁵ control, ANGPTL4 knockdown, or ANGPTL4-rescue LM2 cells. Tumor size was measured at day 28. n=14; error bars indicated s.e.m
(C) The presence of circulating tumor cells was assessed by qRT-PCR as a function of human-specific GAPDH expression relative to murine β2-microglobulin, in 3 mL of mouse blood perfusate. n=10; error bars indicate s.e.m.
(D) Peri-aortic lymph node metastasis from indicated tumors was analyzed by ex-vivo detection of luciferase activity. Positive bioluminescent signal in extracted lymph nodes indicated presence of metastasized tumor cells.
(E) Photon flux measurements of breast cancer cells seeding the lung from orthotopically injected tumors as indicated. n=13–15 error bars indicated s.e.m; p-values calculated using the one-tailed unpaired t-test.
(F) ANGPTL4 mRNA levels were determined by qRT-PCR analysis in cells that were incubated for 6 h with or without TGFβ. n=3; error bars indicate standard deviation.
(G) Lung colonization analysis was performed by injecting 2×10⁵ cells into the lateral tail vein. Prior to injection, cells were treated as indicated with 100 pM TGFβ for 6 h. Bar graphs represent 24 h time point measurements of the normalized photon flux. n=14–21; error bars indicate s.e.m; p-values calculated using the one-tailed unpaired t-test.
(H) Normalized photon flux measurements from tail vein injected animals. Lung colonization measurements were taken from animals injected with control or Angptl4 overexpressing LM2 cells. n=6; error bars indicate s.e.m; p-values calculated using the one-tailed unpaired t-test.

(A) HUVEC monolayers were grown to confluence on fibronectin coated slides and then treated for 24 h with rhAngptl4. Slides were subsequently fluorescently stained with anti-ZO-1 antibody, phalloidin, and anti-β-catenin antibody.
(B) HUVEC monolayers were treated for 24 h with media conditioned by control LM2 cells or LM2 cells that overexpress Angptl4. Samples were stained for ZO-1 and phalloidin.
(C) GFP-labeled MDA-231 cells were injected via the tail vein and allowed to lodge in the lungs. One day post injection, a rhodamine-dextran dye was injected into circulation. Three hours after dye injection, lungs were extracted and frozen sections were obtained. Representative confocal images are shown here of cells with and without accumulation of dye in the lung parenchyma.
(D) Images were obtained as described in (C) with control or Angptl4 overexpressing MDA-MB-231 cells. A region of interest was drawn around the GFP-labeled cells and the amount of dextran dye was quantified based on rhodamine emissions. n=40 cells; error bars indicate s.e.m; p-values calculated using the one-tailed unpaired t-test.
(E) Indicated cell lines were seeded into trans-well inserts that were previously covered with a HUVEC monolayer. Cells that migrated cross the endothelial layer into the bottom side of the transwell membrane were quantified with Volocity software. n=15, error bars indicate s.e.m; p-values calculated using the one-tailed unpaired t-test.
(F) Schematic model of the cytokine relay set up by TGFβ activity in the primary tumor. ER− primary tumor cells that are exposed to TGFβ respond with ANGPTL4 induction via the Smad pathway. As they enter the circulation and reach the lung capillaries, these cells secrete Angptl4 which disrupts endothelial cell junctions thereby enabling the cancer cells to more efficiently enter the lung parenchyma.