We used culture media obtained from Invitrogen (CA, USA) and Cambrex (MD, USA). Fetal bovine serum (FBS) was obtained from Biowest (Nuaillé, France) and was heat-inactivated before use. Growth Factor Reduced Matrigel was obtained from Becton Dickinson Biosciences (NJ, USA). Bovine serum albumine (BSA), cell-culture grade gelatin, fibronectin, DTT, protamine sulphate, aminoguanidine and amifostine (WR-2721) were purchased from the Sigma Chemical Co (MO, USA). The metabolically active form of amifostine (WR-1065) was synthesized by Dr M. Zanda (CNR, Milan, Italy). The VEGF-A neutralizing polyclonal antibody was obtained from R&D Systems (Minneapolis, USA). Primary antibodies against human HIF-2α and β-actin were obtained from Santa Cruz Biotechnology (CA, USA); anti-human Ku80 from Serotec (Oxford, UK); anti- human HIF-1α from BD Transduction Laboratories (Le Pont de Claix, France); anti-VEGFR-2 from ReliaTech (Braunschweig, Germany); anti- phospho-eIF2α (Ser51), anti-eIF2α and antibodies to p42/44 and their phosphorylated counterparts from Cell Signaling Technology (MA, USA). Primers for ATF4 were as described by Namba et al. 37 . Primers for GADD34, CHOP, EDEM and BIP were as previously described 21 . Other primer used for mRNA expression analysis, were reported in Additional File 1 (Table S1). All primers were obtained from Proligo (Paris, France).

Human mammary carcinoma cells MCF7 and colon carcinoma HCT116 cells were a gift from Dr P Hainaut (International Agency for Research on Cancer, Lyon, France). MCF7 cells and U87 human glioma cells (ATCC, HTB-14) were grown in Dulbecco's modified eagle medium (DMEM) supplemented with 10% FBS. HCT116 cells were grown in McCoy medium, 10% FBS. Human umbilical EC (HUVEC; Clonetics, CA, USA) were propagated up to eight passages on a 0.2% gelatine matrix in endothelial growth medium (EGM-2; Bulletkit, Walkersville, USA). Treatments of tumour cells with amifostine were performed in the presence of 4 mM aminoguanidine (AG), as previously described 10 . AG prevents the catabolization of amifostine by FBS Cu-dependant amine-oxydases into cytotoxic compounds 38 . Amifostine treatments on HUVEC were performed without AG. Hypoxic conditions were obtained at 3% O₂ in a Heraeus incubator BB-6060.

MCF7 were transduced using lentivectors expressing the fluorescent marker E-GFP and containing short hairpin RNA sequences against HIF-1α [HIF-1α.small hairpin (sh)RNA] or red fluorescent protein (RFP.shRNA) as described 39 . For transduction, MCF7 cells (5.10⁴ cells per well in a 24-multiwell plate) were incubated for 24 h in complete medium. Cells were then incubated with viral supernatants from 293T cells for 24 h at 37°C in the presence of protamine sulphate. Transduced cells were sorted out 5 days post-transduction by cytofluorimetry. Enhanced green fluorescent protein (E-GFP) positive cells were used for the experiments.

Small interfering RNAs (siRNAs) were purchased from Eurogentec (Liège, Belgium). The sequence of the ATF4-targeting SiRNA (SiATF4) was as previously described 40 . A SiATF4 mutated in three nucleotides served as control (SiMUT, 51). Cells at a 50% density were transfected with 250 nM of SiRNA in OptiMEM using Lipofectamine Plus (Invitrogen, CA, USA). After 24 h, cells were either treated or not treated with amifostine; cells and supernatants were then collected for RNA isolation and protein analysis.

Cells at 60% confluence were grown in 5-cm diameter dishes for the indicated period of time. VEGF-A concentration was measured in cell-conditioned media using a commercial VEGF-A ELISA kit (R&D Systems, Minneapolis, USA). Assays were performed in triplicate and calibration curves were obtained using human recombinant VEGF-A. The results were analysed using the Softmax Pro4.0 software (Molecular Devices Corporation, CA, USA). Cells were counted using a cell counter (Coulter, Becton Dickinson, NJ, USA).

The paracrine effect of amifostine was assayed as follows. MCF7 cells were first incubated for 48 h, either in the presence of AG alone or with both AG- and amifostine. Conditioned media (CM) in these conditions were dialysed at 4°C in order to remove the two chemicals. HUVEC grown in 96-multiwell plates to 80% confluence were then starved for 4 h in DMEM without fetal calf serum (FCS) and incubated for 16 h in the presence of CM. Wst-1 assays (Roche Applied Science, IN, USA) were then performed as follow: Wst-1 reagent was added to the cell medium after a 12 h incubation with CM and absorbance was read 4 h later at 440 nm using a spectrophotometer (Molecular Devices Corporation). Results were analysed using the Softmax Pro 4.0 software (Molecular Devices Corporation). The assay was performed in triplicates. In similar experiments, CM were pre-incubated for 45 min at 37°C with a VEGF-A neutralizing antibody or with an irrelevant anti-ß-actin antibody, prior to addition to HUVEC.

Reverse transcription (RT) were performed as previously described 39 . A semi-quantitative analysis of VEGF-A was performed by co-amplifying VEGF-A and β-actin in a Thermal Cycler (Eppendorf AG, Hamburg, Germany) at 94°C for 40 s, 59°C for 30 s, 72°C for 50 s, throughout 35 cycles, with a final elongation step of 3 min at 72°C. Real time quantitative PCR (Q-PCR) analyses were performed using the MX3000p thermocycler (Stratagene, CA, USA) and the SYBRgreen dye (ABgene, Epsom, UK) methodology. The relative abundance of transcripts was calculated by using α-tubulin or β-actin as standards.

MCF7 cells and HUVEC were grown up to the subconfluence in 10 cm diameter dishes. HUVEC were starved for 24 h in EBM-2 medium, 2% FCS prior to treatments. Treatments were then performed in fresh medium (EBM-2 for HUVEC, complete culture medium for MCF7) and total protein extracts were collected on ice at different time points with an electrophoretic mobility shift assay buffer-B 11 for HIF and mitogen activated protein kinase analysis or as described by Drogat et al. for eIF-2α analysis 22 . Forty to fifty micrograms of proteins were separated in SDS-PAGE gels and transferred to a 0.4 μm nitrocellulose membrane (Schleicher & Schuell, Dassel, Germany). Proteins of interest were detected using specific primary antibodies (see figure legends) and secondary antibodies were coupled to horseradish peroxidase (DAKO SA, Glostrup, Denmark). Blots were revealed using the enhanced chemiluminescence +reagent (Amersham Pharma Biotech, NJ, USA) followed by radioautography or direct quantification of chemiluminescence (Fujifilm LAS-3000).

VEGF-A₁₆₅ was radio-labelled as previously described 41 to a specific activity ranging from 20,000 to 100,000 cpm/ng. HUVEC were seeded at 7000 cells/cm2 in six-well plates, cultured for 2 days in EGM-2 and starved for 24 h in 2% FBS-containing EBM-2 prior to a 12 h-incubation in the presence, or not, of 1 mM WR-1065. The plates were washed once with phosphate buffered saline (PBS) and the cells were incubated for 2 h at 4°C with 125I-VEGF-A in DMEM, 20 mM HEPES, 0.2% gelatine. They were rinsed three times in PBS and finally solubilized at room temperature in a solution containing 2% Triton X-100, 10% glycerol and 1 mg/mL BSA. Total cell lysates were counted for radioactivity using a Kontron MR 250 gamma-counter. Each value reported in the graph was the mean of duplicate determinations. Unspecific binding was defined as the amount of radioactivity bound in the presence of 20 μg/mL protamine sulphate, which gave results similar to the use of a 100-fold excess of VEGF-A. The experiments were repeated three times with similar results. Binding data were analysed using the GraphPad Prism 4.0 software.

HUVEC cell migration was measured using 24-well-Transwell plates (Becton Dickinson Labware, Le Pont de Claix, France). Prior to cell seeding, inserts (8-μm pore size) were precoated overnight at 4°C on the lower side using 10 μg/mL fibronectin in PBS. Membranes were then blocked 1 h at room temperature with 1% heat-inactivated BSA. HUVEC (10⁵ cells per well) were seeded in the upper compartment and allowed to migrate for 7 h throughout the membranes in DMEM containing 0.5% FBS and 0.1% BSA and in the presence of different concentrations of WR-1065. VEGF-A (10 ng/mL) was added in the lower compartment as chemo-attractant. After migration, cells that had migrated were fixed and stained in a solution containing 30% methanol, 10% acetic acid and 0.1% Coomassie Blue. The extent of cell migration was analysed by counting cells in at least six random fields per Transwell filter using the LUCIA image analysis software (Laboratory Imaging, Praha, Czech Republic). Results are from three independent experiments.

Amifostine was tested for its ability to modulate endothelial cell differentiation into tube-like structures using a Matrigel surface assay. Cells (2 × 10⁵ cells/mL of Growth Factor Reduced Matrigel, BD Biosciences, MA, USA) were seeded in EBM-2 medium supplemented with 0.2% FBS and 1% L-glutamine and allowed to adhere for 1 h. Media were then replaced by EBM-2 enriched with 0.5% FBS, containing increasing concentrations of WR-1065 with or without 10 mg/mL of VEGF-A. Cells were then incubated for 3 h at 37°C. Total length of tube-like structures was assessed using the LUCIA image analysis software. Results shown are representative of three independent experiments. Each data point is the mean of six randomly chosen fields.

Statistical analysis of data was performed using the paired Student's t test P value (*P < 0.05; **P < 0.01).

The potential effects of the cytoprotective drug amifostine on the vascular bed still remains unclear and controversial. In order to better characterize the angiogenesis-related effects of this molecule, we analysed the effects of its dephosphorylated active form (WR-1065) on the mRNA expression of VEGF-A, a growth factor that plays a major role in tumour angiogenesis.

Three human tumour cell lines from different tissue origins (U87 glioma cells, MCF7 breast carcinoma cells and HCT116 colon carcinoma cells) were treated with increasing amounts of WR-1065, surrounding 1 mM, which is a dose corresponding to the measured concentration of the drug in tissues of animal models 8 42 . Total VEGF-A mRNA level was determined by quantitative PCR (Q-PCR) and its different isoforms were distinguished by semi-QPCR. As shown in Figure 1A, WR-1065 treatment induces the accumulation of VEGF-A mRNA in a dose-dependent manner in all cell types. This effect was mostly observed for transcripts that correspond to the two diffusible isoforms (VEGF_{121, 165}) of the growth factor (Figure 1B). As determined for MCF7 cells, VEGF-A mRNA increase is accompanied by a significant release of VEGF-A protein in the culture medium (Figure 1C). These results suggest a potential pro-angiogenic effect of amifostine.

HIF-1α and HIF-2α are involved in the transactivation of VEGF-A gene in response to hypoxia 43 and amifostine has been described as a potent hypoxia-mimetic compound that could activate HIF-1α both in vitro and in vivo 13 . Therefore, we analysed by Western Blot the accumulation of the hypoxia responsive transcription factors HIF-1α and HIF-2α in MCF7 cells treated with 1 mM WR-1065. As shown in Figure 2A, WR-1065 treatment did not lead to the accumulation of either HIF-1α or HIF-2α in treated cells, whereas a strong increase in HIF-1α was observed in cells subjected to low oxygen concentration (3% instead of 20%). Consistently, 1 or 2 mM WR-1065 failed to induce any nuclear translocation of HIF-1α whereas lowering oxygen led to a significant increase of HIF-1α in the nucleus (Additional File 2, Figure S1).

We then measured the mRNA level of two representative HIF-1α-dependent genes, namely HK2, GLUT1. As expected, expression of these genes was strongly increased under low oxygen conditions (3% O₂ Figure 2B). In comparison, their expression was significantly repressed (≈ 50%) in response to WR-1065 treatment, which suggest an HIF-1α-independent effect of WR-1065 on VEGF-A. In order to definitively exclude an HIF-1-dependent activation of VEGF-A by amifostine, we abolished HIF-1α protein expression by stably expressing an shRNA directed against the mRNA of HIF-1α 39 . MCF7 cells were then treated with WR-1065 or subjected to low oxygen conditions (3%). Under low level of oxygen (3%), shRNA.HIF but not shRNA.RFP (Red Fluorescent Protein), led to a strong inhibition of both HIF-1α protein (90%, Figure 2C) and VEGF-A mRNA accumulation (50%, Figure 2D). In response to WR1065 treatment, knock down of HIF-1α did not significantly modify the increase of VEGF-A mRNA expression. Altogether, these results indicate that the accumulation of VEGF-A mRNA in WR-1065-treated cells does not depend on HIF-1α activation.

VEGF-A mRNA induction through HIF-1-independent signalling pathways has already been described in response to a variety of cellular stresses, including the cell response to the accumulation of mis/unfolded proteins (UPR). Since WR-1065 is a thiol reducing agent, we questioned its effect on the UPR triggering pathways. We first analysed the expression of the UPR-target genes BiP, EDEM, GADD34 and CHOP in cells treated with WR-1065. As shown in Figure 3A, the expression of the IRE1-dependent genes EDEM and BIP were slightly (twofold) increased under these conditions. Consistently, the splicing of the XBP1 transcript, another hallmark of IRE1 activation, was neither observed in MCF7 nor in U87 cells treated with WR-1065 (Additional File 3, Figure S2). This confirms that IRE1 was not significantly activated by amifostine. Besides, GADD34 and CHOP genes were highly up-regulated (10- to 13-fold) in response to WR-1065 treatment, which evokes an activation of the eIF2α/ATF4 dependent pathway. Indeed, eIF2α was transiently phosphorylated after 2 h of treatment with WR-1065 (Figure 3B). Consistently, blockade of ATF4 expression using a siRNA strategy (Figure 3C) led to the inhibition of WR-1065-induced VEGF-A expression, both at the mRNA (80 to 90%, Figure 3D) and at the protein (50%, Additional File 3, Figure S3) levels. Overall, these results indicate that ATF4 is required for amifostine-induced VEGF-A up-regulation.

The increased release of VEGF-A by WR-1065-treated tumour cells suggests a possible paracrine activity mediated by the chemoprotectant amifostine on EC. In order to confirm this, MCF7 cells were stimulated by WR-1065 for 48 h and cell-conditioned media were then tested for their ability to stimulate starved EC using the MTT metabolic assay (Figure 4). WR-1065 indeed stimulated the release by tumour cells of a diffusible EC growth factor that is able to significantly increase the metabolic activity of EC cells (34%). This stimulatory effect was partly inhibited (20%) by 1 μg of a VEGF-A neutralizing antibody, indicating that VEGF-A produced by tumour cells plays a significant role in this endothelial cell stimulation. Increasing the concentrations of the neutralizing VEGF-A antibody (from 0.03 to 1 μg) failed to produce a higher inhibitory effect, whereas it completely abolished the effect of recombinant VEGF-A. These results suggest that additional EC growth factors may also contribute to the tumour-mediated EC stimulation.

The stimulatory paracrine effect of VEGF-A on EC should predict stimulation of the neovascularization process. However, since EC cells are the first cellular target of amifostine in vivo, we also considered the effect of amifostine directly on EC proliferation, migration and differentiation into tubular-like structures. Thus, HUVEC were grown in the presence of VEGF-A with various concentrations of WR-2721. This allows a progressive release of WR-1065 throughout the duration of treatment. Compared to untreated cells, WR-2721-treated cells have a significantly reduced cell growth and the effect was dose-dependent (Figure 5A, ED₅₀ ~ 0.5 mM after a 2-day incubation, see insert). This growth-inhibition was also observed using bovine EC of aortic origin (BAE; Additional File 4, Figure S4A). The inhibitory activity may result from a p53-dependent cell cycle arrest in G₁ as previously described ( 10 , see also Additional File 4, Figure S5). We then determined whether amifostine may influence VEGF-A-induced EC migration. HUVEC were treated, or not, with VEGF-A in the presence of increasing concentrations of WR-1065 and allowed to migrate for 7 h. Figure 5B shows that WR-1065 inhibits the VEGF-A-induced migration of EC in a dose dependent manner. Again, similar result was obtained using BAE cells (Additional File 4, Figure S4B). Finally, we analysed the effects of amifostine treatment on the in vitro formation of capillary-like structures by EC. For this, HUVEC were cultured on Matrigel and treated or not with VEGF-A and/or WR-1065. Figure 5C shows that treatment with WR-1065 completely abolished the VEGF-A induced capillary formation. WR-1065 alone had no significant effect, neither on migration, nor on tubule formation on serum-starved EC (data not shown). Taken together these results indicate that amifostine strongly impairs the activity of VEGF-A on EC proliferation, migration and differentiation, making them unresponsive to a potential paracrine effect mediated by this growth factor.

In order to understand how amifostine does inhibit EC response to VEGF-A, HUVEC were treated with WR-1065 and the level of VEGF receptor 2 (VEGFR-2) was measured by Western blot. Figure 6A shows that from 2 h to 24 h of incubation with WR-1065, VEGFR-2 protein expression is decreased by ~50% in treated cells as compared to untreated cells. This down-regulation occurs most likely at different levels including the transcriptional level, since VEGFR-2 mRNA is transiently down-regulated in these conditions (Additional File 5, Figure S6). Scatchard analysis of 125I-VEGF-A binding to its receptors on HUVEC was also performed. In agreement with the decrease in VEGFR-2 protein expression, the binding data of three independent experiment consistently showed a ~25% to 30% decrease in the number of VEGF-A molecules associated to high affinity receptors after a 12 h pre-treatment with 1 mM of WR1065 (Figure 6B). The apparent dissociation constant was not significantly modified in these conditions, which indicates that the decrease in VEGF-A binding is due in part to the down regulation of its receptors. In addition to this long-term effect of amifostine on VEGFR2 expression and availability at the cell membrane, immediate phosphorylation of p42/p44 kinase in HUVEC in response to VEGF-A was strongly reduced (~64%) when applied in the presence of WR1065 (Figure 6C). This was also observed to a greater extent in BAE cells (Additional File 4, Figure S4C). Together, these results indicate that WR-1065 has both immediate and long-term effects on the EC responses by impairing VEGF-A binding and signalling.