The patient population is described in details in the main section. Human tissue samples were processed and stored at the pathology laboratory of the Centre Oscar Lambret cancer clinic (Lille, France). Following surgery, tissues were fixed in 4% paraformaldehyde and embedded in paraffin. For each tumor, analyses were performed on sections from the paraffin block which had been used for estrogen receptor, progesterone and HER2 expression analyses and for diagnosis, considered by the pathologists to be the most representative of the lesion.

For staining human tissues, a primary monoclonal rabbit anti-human ICAM-1 antibody (Abcam, Cambridge, UK; ab53013), a monoclonal mouse anti-human CD31 (DAKO #M823, clone JC/70A), a monoclonal mouse anti-CD34 (DAKO #M7165, clone QBend-10), and a goat biotinylated anti-rabbit IgG (BA-1000, Vector laboratories) were used. For staining mouse tissues, an anti-mouse ICAM-1 antibody (Abcam ab25375), an anti-rat-Alexa 488 antibody (Life Technologies A21208), and a biotinylated rat anti-mouse-CD31 (Pharmingen 553371) were used. Antibodies were reconstituted and stored according to the manufacturers' recommendations.

Four successive paraffin sections (4-µm) were prepared from each sample paraffin block and laid on superfrost + slides, dried overnight at 56°C, and rehydrated before staining. For ICAM-1 staining, slides were placed in methanol, 0.3% H₂O₂ for 20 min at room temperature, rinsed in TRIS-HCl 10 mM, pH 7.5, NaCl 0.15 M (TRIS buffer saline, TBS) for 5 min, incubated in Antigen retrieval, solution citrate (Vector Laboratories, Peterborough, UK) for 20 min at 90°C, then for 20 min at room temperature, and rinsed twice in TBS. Slides were incubated in TRIS-HCl 0.1 M, pH 7.5, NaCl 0.15 M, 20% goat serum, 0.5% blocking buffer from TSA Biosystem (Perkin Elmer, Beaconsfield, UK) for 2 h and incubated in rabbit anti-human ICAM-1 (1:200) overnight at 4°C in a humidified atmosphere. Slides were then washed in TBS for 5 min twice and incubated in goat biotinylated anti-rabbit IgG (1:250) for 45 min. Slides were then washed three times in TBS for 5 min, incubated in Streptavidin-A-HRP (TSA Biosystem kit, 1:100) in blocking buffer for 30 min, washed 3 times in TRIS-HCl 0.1 M, pH 7.5, NaCl 0.15 M, Tween-20 0.25% (TNT) for 5 min, then incubated in biotinylated tyramine (1:50) in amplification buffer (TSA Biosystem, Perkin Elmer), washed 3 times in TNT for 5 min, incubated in streptavidin-A-HRP (TSA Biosystem kit, Perkin Elmer, 1:100) in blocking buffer for 30 min, washed twice in TNT for 5 min and the immune complexes revealed using DAB kit (Vector Laboratories) under microscope observation. Staining was stopped with tap water and slides counterstained with hemalun, dehydrated and mounted.

CD31 and CD34 immunostainings were performed using a BenchMark ULTRA automat and dedicated reagents (Ventana Medical Systems, Roche Diagnostics, Basel, Switzerland). For CD31 staining, slides were deparaffinized, incubated at 95°C for 36 min in cell conditioning buffer 1, washed in Reaction Buffer. Endogenous peroxydases were inactivated using UV INHIBITOR for 4 min and slides further washed in Reaction Buffer. Slides were then incubated in anti-CD31 (Dako, Agilent Technologies, Inc., Santa Clara, CA, USA, #M0823, 1:20) for 28 min, rinsed in Reaction Buffer and incubated in HRP UNIV MULT for 8 min, washed in Reaction Buffer, developed using UV DAB and counterstained in Hematoxylin for 4 min, then in Bluing Reagent, washed in Reaction Buffer before dehydration and mounted. For CD34 staining, slides were processed similarly except that the cell conditioning buffer 1 treatment was performed for 8 min and the slides were incubated with mouse anti-CD34 (DAKO, clone QBend-10, 1:50) for 20 min.

Stained slides were analyzed by two independent oncologists, including a breast cancer pathologist, using an Axioplan 2 microscope (Zeiss) and compared to the corresponding hematoxylin/phloxin safran-stained slides used for identification of the tumor sub-regions for diagnosis. Image acquisition was done using the ZEN2012 blue edition software (Carl Zeiss AG, Oberkochen, Germany).

Paraffin sections (4-µm) were laid on superfrost + slides and dried overnight at 56°C. Digoxigenin-labelled sens and anti-sens egfl7 probes were prepared from a pCMV-sport 6 plasmid containing the full length human egfl7 cDNA using the Riboprobe Gemini system II kit (Promega Corporation, Inc., Madison, WI, USA). Slides were processed using a Ventana Discovery automat as described previously (8).

Primary human umbilical vein endothelial cells (Lonza Biologics PLC, Slough, UK) were cultured in EGM-2 medium and passaged according to the manufacturers' recommendations. Cells (4×10⁴/well) between passage 2 and 5 were plated in 0.33 cm² wells and transfected with 50 nM non-targeting small interfering RNA (siRNA, siCtrl, D-001810-01, Dharmacon, Inc., Lafayette, CO, USA), or with siRNA targeting egfl7 [siEgfl7, J-015668-10, Dharmacon, Inc., (15,16)] or ICAM-1 (siICAM-1, L-003502-00 Dharmacon, Inc.) using Lipofectamine RNAiMax (Life Technologies, ThermoFisher Scientific, Inc., Waltham, MA, USA). Three days later, the cells were lysed and total RNA isolated using the Nucleospin RNA kit and reagents (Macherey-Nagel, Düren, Germany). RNA were quantified using a Nanodrop (ThermoFisher Scientific, Inc.) and RT was performed using the High Capacity Reverse Transcription kit (Life Technologies, ThermoFisher Scientific, Inc.). qPCR was performed in duplex reactions, mixing TaqMan FAM-labelled probes for human egfl7 (Hs00211952_m1) or for ICAM-1 (Hs00164932.m1) with the normalizing β2-microglobulin (B2M) VIC-labelled probe in the same tube and processed in a StepOne machine. Data were expressed as 2^−ΔΔCq where ΔCq=Cq of gene-Cq of B2M, and ΔΔCq=ΔCq sample-ΔCq control.

Sample storage, handling and analysis were performed according to the European regulations and the Helsinki Declaration. Patient consent and legal authorizations were obtained for all the analyses performed and for the processing of patients personal data. The protocol was approved by the ‘Comité de Protection des Personnes Nord-Ouest IV’ on January 12, 2010.

All statistical analysis was performed using STATA 11.2 statistical software (StataCorp, College Station, TX, USA). Analyses used Chi2 and Fisher's exact tests for categorical data and Kruskall-Wallis test for continuous variables and Cuzick's test for trend across ordered groups. P<0.05 was considered to indicate a statistically significant difference.

Patients included in the present study underwent surgery for breast cancer at the Oscar Lambret Center cancer clinic between January 1st, 2005 and July 31st, 2005, for histologically-proven invasive ductal carcinoma (IDC), invasive lobular carcinoma (ILC), or ductal carcinoma in situ (DCIS). Patients were at least 18 years-old, had not been previously treated with chemotherapy, and had not received neo-adjuvant chemotherapy. Patients whose tumors exhibited a histology-proven lobular carcinoma in situ were excluded from the study. The cohort therefore included: 30 DCIS and 174 invasive carcinomas, among which 131 IDC (64.2%, Table I), 27 ILC (13.2%), and 16 others types (tubular carcinoma, apocrine, or neuroendocrine carcinoma, 7.8%). The population was composed of 39 stage 1 (22.4%), 98 stage 2 (56.3%), and 37 stage 3 carcinomas (21.2%). Among invasive cancers, 20 (11.5%) exhibited HER2 gene amplification as defined by a 3+ score using immunohistochemistry or >=2+ score when using chromogenic in situ hybridization. Hormonal receptors were positive as defined by estrogen receptor >=10% and/or progesterone receptor >=10% in 85.8% of cases. A total of 13 triple negative breast cancer (7.5%), as defined by estrogen and progesterone receptor equal to zero associated with Her2 negative expression, were identified within this cohort.

Parallel tissue sections were prepared for the positive identification of blood vessels, for the quantitation of activated vessels among the identified blood vessels, and for the quantitation of egfl7 expression levels in the same tumor areas. The analysis was concentrated on peritumoral vessels and the tumor sample block which had been used for the initial diagnosis of the patient's lesion was selected, thus being the most representative and the closest tissue to the tumor.

In order to quantify the activated blood vessels in peritumoral areas, the initial step was to establish a staining procedure for positively identifying blood vessels in the human tissue sections. For this purpose, immunostaining using anti-CD34 and anti-CD31 antibodies were compared, as both are widely used markers of endothelial cells (17). Staining human breast cancer samples with either antibody clearly revealed the blood vessels (Fig. 1A) but staining for CD31 provided a more specific signal and gave rise to a much lower background than staining for CD34, which also stained non-vascular structures. Therefore all breast cancer tissue sections were subsequently identified with CD31 staining and blood vessels were positively identified as CD31⁺ structures.

Activated blood vessel endothelial cells in breast cancer samples were identified following ICAM-1 immunostaining. ICAM-1 staining was membranous in endothelial cells. Its expression in vascular structures was confirmed by checking the CD31 staining in parallel sections of the same area (Fig. 1B). In case of an absence of ICAM-1⁺ vessels and in order to dismiss any technical artifact on a specific sample, ICAM-1⁺ staining was verified on lobular islets or galactophoric channels of the same sample, as these structures also stain for this marker. The percentage of ICAM-1⁺ activated vessels compared with ICAM-1⁻ non-activated vessels, was evaluated and scored. Score 0 was defined as no ICAM-1⁺ vessels, scores 1+, 2+, and 3+ were defined by 1–29%, 30–60% and >60% of ICAM-1⁺ vessels, respectively. Activation of blood vessels was heterogeneous within any given tumor, as vessels displayed different states of activation in the same sample depending on their location and on their size (Figs. 1B and 2). Among the 175 cases analyzed for ICAM-1, 68 lesions (38.9%) were scored 0 for ICAM-1, 62 (35.4%) were scored 1+, 28 (16.0%) were scored 2+ and 17 (9.7%) were scored 3+. A significant correlation was not identified between ICAM-1 expression in peritumoral blood vessels and clinical data such as cancer type, grade, metastases, hormone receptors, and HER2 status, or with the triple-negative type, using the Chi2 test if theoretical size ≥5 and Fisher exact test otherwise.

Regarding egfl7 transcript detection using in situ hybridization, positive cases were defined based on the nuclear staining of endothelial cells and scoring was made using the same criteria as for ICAM-1 staining. Among the 148 cases analyzed for egfl7 expression, 63 (42.6%) were scored 0, 30 (20.3%) were scored 1+, 27 (18.2%) were scored 2+ and 28 (18.9%) were scored 3+. A significant correlation was not identified between egfl7 expression in peritumoral area and clinical data including cancer type, grade, metastases, hormone receptors, nor HER2 status using the Chi2 test if theoretical size ≥5 and Fisher exact test otherwise.

However, a strong and significant correlation was apparent between the ICAM-1 scores and the egfl7 scores in the analyzed lesions (P=0.015, Fisher exact test). Cuzick's tests for trend demonstrated that when the ICAM-1 score increased, the egfl7 score decreased (Fig. 3, P=0.004, Table II), and vice-versa (P=0.001, Table II). These observations directly implied that, regardless of the type and grade of breast cancer lesions, the expression levels of ICAM-1 and those of egfl7 were somehow linked.

In order to assess which gene between ICAM-1 and egfl7 regulated the expression of the other one, we deregulated either egfl7 or ICAM-1 in human primary endothelial cells by RNA interference and checked for the expression of the other gene. Targeting egfl7, which lowered its expression by 69.4%, induced a 4.9-fold increase in ICAM-1 expression levels in the same cells. On the other hand, the downregulation of ICAM-1 expression levels by 89.3% did not significantly affect the expression levels of egfl7 in the same cells (Fig. 4). Thus, these results show that egfl7 constitutively represses the expression of ICAM-1 in endothelial cells but that ICAM-1 does not regulate egfl7 expression.