Identification of targets of Prox1 during in vitro vascular differentiation from embryonic stem cells: functional roles of HoxD8 in lymphangiogenesis

The lymphatic system plays very important roles in the maintenance of tissue-fluid homeostasis and mediation of the afferent immune response (Karpanen and Alitalo, 2008). Insufficiency or obstruction of function of this system results in lymphedema, characterized by leaking of tissue fluid and swelling of affected tissue. Furthermore, in many types of cancer, the lymphatic vessels provide a major pathway for tumor metastasis to lymph nodes. Understanding of the molecular mechanisms that govern the formation of the lymphatic system is thus of crucial importance.

Embryonic lymphatic endothelial cells (LECs) arise by sprouting from the jugular veins and migrate towards mesenchymal cells expressing vascular endothelial growth factor (VEGF)-C, leading to the formation of the primary lymphatic plexus (Oliver, 2004). During these processes, the prospero-related homeobox-1 (Prox1) transcription factor marks the first LECs within embryonic cardinal veins. Importantly, in Prox1-deficient mice, the migration of endothelial cells expected to express Prox1 towards VEGF-C is arrested, resulting in complete lack of the lymphatic vasculature and embryonic lethality (Wigle and Oliver, 1999; Wigle et al., 2002). Transcription-profiling analyses have shown that Prox1 induces the expression of various LEC markers – including VEGFR3, a receptor for VEGF-C – in human dermal microvascular endothelial cells (HDMECs) (Hong et al., 2002; Petrova et al., 2002). Recently, we also showed that Prox1 inhibits sheet formation of mouse embryonic stem (ES)-cell-derived endothelial cells (MESECs) and induced various LEC markers, such as integrin α9 (Mishima et al., 2007). Furthermore, we confirmed that Prox1 induces integrin-α9 expression in human umbilical vein endothelial cells (HUVECs) and human dermal LECs (HDLECs), and found that integrin α9 is required for Prox1-induced inhibition of sheet formation and promotion of migration towards VEGF-C. Interestingly, Prox1 also downregulates the expression of blood vascular endothelial cell (BEC) markers such as VEGFR2 (Petrova et al., 2002; Mishima et al., 2007), suggesting that Prox1 regulates the differentiation of embryonic BECs to LECs by reprogramming gene expression profiles.

The primary lymphatic plexus, which is composed of thin lymphatic vessels, undergoes enlargement and maturation to form collecting lymphatic vessels that have characteristics different from those of primary lymphatic vessels, such as valve formation and attachment of smooth-muscle cells. FoxC2 transcription factor and angiopoietin-2 (Ang2), a ligand for Tie2 receptor tyrosine kinase, have been implicated in these maturation processes (Gale et al., 2002; Dellinger et al., 2008; Petrova et al., 2004). Prox1 expression is maintained in mature LECs, and is required for maintenance of the expression of LEC markers and characteristics of LECs (Mishima et al., 2007; Srinivasan et al., 2007; Johnson et al., 2008). However, the molecular mechanisms by which Prox1 regulates the maturation of lymphatic vessels remain to be elucidated.

In the present study, we identified novel target genes of Prox1 in MESECs. Expression of FoxC2 and Ang2 was induced by Prox1. Furthermore, we found that Prox1 induces the expression of HoxD8, which is capable of increasing the caliber of lymphatic vessels in an in vivo model of inflammatory lymphangiogenesis. These findings suggest that Prox1 induces the expression of a distinct group of genes to promote the maturation of lymphatic vessels.

We previously showed that Prox1 transgene expression in MESECs induces the expression of a group of genes that induce the differentiation of BECs to LECs (Mishima et al., 2007). To further identify the Prox1 target genes that are involved in lymphatic development, we performed cDNA microarray analysis using endothelial cells derived from mouse ES cells carrying a tetracycline (Tc)-inducible Prox1 transgene. Among approximately 400 genes whose levels of expression were regulated by Prox1 (supplementary material Tables S1 and S2), we found that expression of integrin α9, cyclin E1 and fibroblast growth factor receptor 3 (FGFR3), all of which have been reported to be Prox1 targets (Mishima et al., 2007; Petrova et al., 2002; Shin et al., 2006), was induced by Prox1 (Table 1). These findings suggest that cDNA microarray analysis identified an appropriate set of Prox1 targets.

Among the candidate molecules of Prox1 targets identified by the cDNA microarray analysis, we next searched for factors that have been implicated in lymphatic formation. Although Ang2 and FoxC2 have been implicated in the maturation of lymphatic vessels (Gale et al., 2002; Dellinger et al., 2008; Petrova et al., 2004), the molecular mechanisms by which their expression is regulated in LECs have not yet been elucidated. Because we found that their expression was induced by Prox1, we first confirmed the effects of Prox1 on the expression of Ang2 and FoxC2 in MESECs by quantitative reverse transcriptase (RT)-PCR analysis. As previously reported (Mishima et al., 2007), the removal of Tc from culture of VEGFR2-expressing (VEGFR2+) cells derived from Tc-Prox1 ES cells induced Prox1 transgene expression, whereas that from Tc-Empty ES cells did not (Fig. 1A,E). As shown in Fig. 1B,C, Prox1 transgene expression significantly promoted the expression of Ang2 and FoxC2.

We next examined whether these effects of Prox1 are observed in other types of endothelial cells. HUVECs were infected with adenovirus encoding Prox1 (Ad-Prox1) or non-coding adenoviruses (Ad-Null) (Fig. 2Aa,E). When Prox1 was transduced into HUVECs, Ang2 expression was significantly induced (Fig. 2Ba,E). By contrast, FoxC2 expression was only mildly induced (Fig. 2Ca,E).

Because LECs, but not BECs, express Prox1 endogenously (Fig. 2Ab,E), we expected that the levels of expression of endogenous Ang2 and FoxC2 might be higher in HDLECs than in HUVECs. As shown in Fig. 2Bb and E, Ang2 expression was higher in HDLECs than in HUVECs. However, the levels of expression of FoxC2 mRNA were comparable in HDLECs and HUVECs, and the expression of FoxC2 protein was only slightly higher in HDLECs than in HUVECs (Fig. 2Cb,E).

We previously showed that endogenous Prox1 expression in HDLECs is required for maintenance of the expression of LEC markers, including VEGFR3 and integrin α9 (Mishima et al., 2007). To examine the roles of endogenous Prox1 in HDLECs, HDLECs were transfected with siRNAs for Prox1 (siProx1) or negative control siRNA (siNC). When Prox1 expression was knocked-down (Fig. 2Ac,E), the expression levels of Ang2 and FoxC2 mRNAs were not significantly altered, and those of Ang2 and FoxC2 proteins were only weakly repressed (Fig. 2Bc,Cc,E). These results suggest that Prox1 is able to induce the expression of Ang2 in embryonic and mature endothelial cells, and that Prox1 differentially induces FoxC2 expression in embryonic and mature endothelial cells.

To identify novel targets of Prox1 involved in lymphangiogenesis, we focused on transcription factors whose expression was upregulated by Prox1 in MESECs. In addition to FoxC2, we identified homeobox transcription factors including HoxD8 and HoxD9 (Table 1). Homeobox (Hox) genes encode transcription factors, which play crucial roles in cell proliferation, migration and differentiation (Pearson et al., 2005). Hox proteins bind DNA weakly, but gain specificity and affinity by interaction with other proteins, termed modulators or cofactors. We therefore examined whether HoxD8 and/or HoxD9 form transcription complexes and function in a lymphatic-vessel-specific fashion. Some Hox-family members, such as HoxA9 and HoxD3, have been implicated in development of the cardiovascular system and endothelial-cell activation during neovascularization (Bruhl et al., 2004; Boudreau et al., 2004). However, the roles of HoxD8 in the vascular system have not yet been reported. Furthermore, no Hox-family members have yet been implicated in lymphangiogenesis.

Induction of HoxD8 expression by Prox1 in MESECs was confirmed by quantitative RT-PCR analysis (Fig. 1D). We also found that Prox1 induces HoxD8 expression in HUVECs (Fig. 2Da), suggesting that HoxD8 is induced by Prox1 in various types of endothelial cell.

We next examined the expression of HoxD8 in LECs. As shown in Fig. 2Db, the level of expression of HoxD8 in HDLECs was higher than that in HUVECs. Furthermore, knockdown of Prox1 expression in HDLECs led to a decrease in HoxD8 expression (Fig. 2Dc). These results suggest that the amount of HoxD8 transcripts is increased during Prox1-induced differentiation of BECs into LECs and that this increased expression is maintained by Prox1 in LECs, although these results were not confirmed at protein level because of the lack of antibodies that are able to detect endogenous HoxD8 proteins.

HoxD8 is one of the paralogous group-8 Hox genes; this group also includes HoxB8 and HoxC8. HoxD8-mutant mice exhibited two anterior homeotic transformations of thoracic vertebrae, although both of these phenotypes exhibited very low penetrance (van den Akker et al., 2001). Phenotypes of double and triple mutants revealed that HoxB8, HoxC8 and HoxD8 have redundant functions at the upper thoracic and sacral levels, including positioning of the hindlimbs. To examine whether other group-8 Hox genes are involved in lymphangiogenesis, we determined the expression of HoxB8, HoxC8 and HoxD8 in HUVECs and HDLECs by conventional RT-PCR analysis. Whereas all paralogous group-8 Hox genes were expressed in HUVECs, only HoxD8 was expressed in HDLECs (Fig. 3A). Furthermore, Prox1 failed to induce the expression of HoxB8 and HoxC8 in HUVECs (Fig. 3B), whereas it induced HoxD8 expression (Fig. 2Da), suggesting important functions for HoxD8 among the paralogous group-8 Hox genes in LECs.

Previous studies showed that HoxD8 is expressed along the anteroposterior axis in the nervous system, vertebral column, gut and kidney at the posterior end of the embryo, and that expression of it extends anteriorly (Izpisùa-Belmonte et al., 1990; van den Akker et al., 2001). However, expression of it in the developing lymphatic vessels has not been reported. To examine the in vivo significance of our finding that Prox1 induces HoxD8 expression in various cultured endothelial cells, we determined the levels of expression of HoxD8 in BECs and LECs derived from mouse embryos. BECs and LECs were isolated from embryonic day 14.5 (E14.5) mouse embryos by fluorescence-activated cell sorting (FACS) using antibodies for CD31 and LYVE-1, respectively, as described in Materials and Methods (Hirashima et al., 2008). The level of expression of HoxD8 in LECs was significantly higher than that in BECs (Fig. 4). This result suggests that the in vitro induction of HoxD8 expression by Prox1 demonstrated in the present study might mimic the process of embryonic lymphatic development.

To examine the roles of HoxD8 in lymphangiogenesis, we used a mouse model of chronic inflammation (Iwata et al., 2007). In this model, chronic aseptic peritonitis was induced by repeated intraperitoneal injection of thioglycollate medium, a proinflammatory agent, into immunocompetent BALB/c mice. As shown in Fig. 5A, outgrowths of lymphatic vessels were detected by anti-LYVE-1 staining in the inflammatory plaques formed on the peritoneal surface of the diaphragm. When adenoviruses encoding β-galactosidase (lacZ) were injected intraperitoneally in combination with thioglycollate medium, these lymphatic vessels were stained for anti-β-galactosidase (Fig. 5A), suggesting that newly generated lymphatic vessels are transduced by adenoviruses.

When adenoviruses encoding HoxD8 were injected, the diameters of lymphatic vessels increased significantly compared with those injected with adenoviruses encoding β-galactosidase (Fig. 5B,C). These findings suggest that HoxD8 regulates the caliber of lymphatic vessels.

The inflammatory plaques formed on the surface of the diaphragm consist of macrophages that express VEGF-A, VEGF-C and VEGF-D (Iwata et al., 2007). To examine whether HoxD8 that was transduced into macrophages alters the expression of angiogenic and/or lymphangiogenic factors, we determined the levels of expression of VEGF-A, VEGF-C and VEGF-D in the plaques that mainly consisted of macrophages. As shown in Fig. 5D, levels of expression were not statistically changed by adenovirally transduced HoxD8. These results suggest that the effects of HoxD8 on the lymphatic vessels are not indirectly mediated through VEGF-A, VEGF-C or VEGF-D secreted from macrophages.

Tie2 receptor tyrosine kinase is activated by its ligand, Ang1, and transduces signals to induce maturation of blood and lymphatic vessels. Ang2 is considered as an antagonist or agonist for Tie2, depending on endothelial-cell status. Furthermore, lymphatic phenotypes in knockout mice deficient for Ang2 were rescued by knocking-in the Ang1 gene in the Ang2 locus (Gale et al., 2002), suggesting that Ang2 might function as an agonist for Tie2 to induce maturation of lymphatic vessels. To examine the relationship between HoxD8 and Ang2, both of which are targets of Prox1, we examined the level of expression of Ang2 in HDLECs transfected with siRNAs for HoxD8 (siHoxD8) or siNC (Fig. 6A). A decrease in HoxD8 expression (Fig. 6A, top) resulted in partial decrease in the amounts of transcripts (Fig. 6A, bottom) and proteins (Fig. 6B) of Ang2 in HDLECs.

We next examined the effects of gain-of-function of HoxD8 on Ang2 expression in HUVECs. When HoxD8 as well as Prox1 was transduced into HUVECs, the level of Ang2 transcripts was increased (Fig. 6C), which was confirmed at protein level (Fig. 6D). These results suggest that both Prox1 and HoxD8, one of the targets of Prox1, induce Ang2 expression.

Although Prox1 expression is maintained in mature LECs, the molecular mechanisms by which it is maintained have not yet been elucidated. We therefore examined whether HoxD8, the expression of which is induced by Prox1, plays any roles in the regulation of Prox1 expression. When HoxD8 expression in HDLECs was knocked-down by siRNA (Fig. 6A, top), endogenous expression of Prox1 decreased (Fig. 7A), suggesting that HoxD8 is required for the maintenance of endogenous Prox1 expression in LECs. Moreover, we found that the expression of a HoxD8 transgene (Fig. 7B, top) induced endogenous Prox1 expression in HUVECs (Fig. 7B, bottom). These results were confirmed at the protein level by western blot analysis (Fig. 7C). Taken together, these findings suggest that HoxD8 plays an important role in positive feedback of Prox1 expression in LECs.

In the present study, we identified a novel group of Prox1 targets using cDNA microarray analysis of MESECs. We found that Prox1 induces the expression of Ang2 and FoxC2, both of which are involved in the maturation of lymphatic vessels (Gale et al., 2002; Dellinger et al., 2008; Petrova et al., 2004). We also found that Prox1 induces the expression of HoxD8, which increased the diameter of lymphatic vessels in an in vivo model of inflammatory lymphangiogenesis. Furthermore, HoxD8 maintains endogenous Prox1 expression in LECs.

Although previous studies (Petrova et al., 2002; Hong et al., 2002; Shin et al., 2006) identified various LEC markers as Prox1 targets using microarray analyses of mature endothelial cells (HDMECs), they did not find that Ang2 or FoxC2 expression was upregulated by Prox1. As shown in Fig. 2Ca, induction of FoxC2 by Prox1 in HUVECs was not as potent as that in MESECs (Fig. 1C), which might explain why previous studies did not identify these genes as targets of Prox1. By contrast, VEGFR3 expression was not upregulated by Prox1 in our analysis. VEGFR3 expression has been reported to be high in embryonic endothelial cells (Tammela et al., 2008), which might have led to the failure of VEGFR3 induction by Prox1 in MESECs. These results suggest that MESECs might represent a different status of endothelial cells from HDMECs and HUVECs, and confirm the usefulness of the cDNA microarray analysis performed in the present study.

Ablation of Ang2 or FoxC2 does not result in defects in the formation of the primary lymphatic plexus, which is severely affected in Prox1-knockout mice (Wigle and Oliver, 1999). However, remodeling and maturation of lymphatic vessels is defective in mice deficient for Ang2 or FoxC2 (Gale et al., 2002; Dellinger et al., 2008; Petrova et al., 2004). In Ang2-deficient mice, lymphatic vessels do not mature or they exhibit a collecting vessel phenotype, without proper recruitment of smooth-muscle cells or postnatal remodeling (Dellinger et al., 2008). FoxC2-knockout mice exhibit abnormal lymphatic vascular patterning, increased pericyte recruitment of lymphatic vessels and defective valve formation (Petrova et al., 2004), which lead to the phenotype of lymphedema-distichiasis syndrome (LD). The present finding that Prox1 induces the expression of Ang2 and FoxC2 suggests that Prox1 indirectly induces the maturation of embryonic lymphatic vessels through induction of a group of maturation-inducing factors.

We identified HoxD8 as a novel target of Prox1. Hox genes play very important roles during embryonic development. We found that HoxD8 increases the caliber of lymphatic vessels in a model of inflammatory lymphangiogenesis. However, this effect of HoxD8 on the caliber size did not mimic the functions of Prox1. We found that adenoviruses encoding Prox1 significantly decreased the diameters of lymphatic vessels (supplementary material Fig. S1A,B) and that Prox1 seemed to increase the number of lymphatic vessels as compared with the control (supplementary material Fig. S1A). Both Prox1 and HoxD8 induced the expression of Ang2, which is involved in the remodeling and maturation of lymphatic vessels. Similar to Prox1, however, adenoviruses encoding Ang2 significantly decreased the diameters of lymphatic vessels (supplementary material Fig. S1C,D), suggesting that Ang2 is not involved in the HoxD8-mediated increase of the caliber of lymphatic vessels. Several lines of evidence have suggested that Ang2 that is present in smooth-muscle cells plays important roles in the formation of lymphatic systems (Gale et al., 2002; Dellinger et al., 2008). The effects of adenovirally introduced Ang2 on the lymphatic formation (supplementary material Fig. S1C,D) might mimic the effects caused by Ang2 secreted by smooth-muscle cells.

Furthermore, studies using in-vitro-cultured HDLECs revealed that the decrease in HoxD8 expression failed to affect the number of HDLECs (supplementary material Fig. S2A). We also found that the decreased HoxD8 expression did not alter the migration of HDLECs towards VEGF-C (supplementary material Fig. S2B). Nonetheless, adenovirally introduced HoxD8 might have caused inflammatory cells to secrete a different profile of lymphangiogenic factors. Although we showed that HoxD8 does not significantly alter the expression of VEGF-A, VEGF-C or VEGF-D in macrophages (Fig. 5D), it might have induced the expression of other cytokines that regulate the caliber size of lymphatic vessels. The molecular mechanisms by which HoxD8 regulates the diameter of lymphatic vessels remain to be elucidated in the future.

We found that both Prox1 and HoxD8 induce Ang2 expression. During angiogenesis, FoxC2 controls Ang2 expression by direct activation of the Ang2 promoter and promotes maturation of blood vessels (Xue et al., 2008). Taken together with the present finding that Prox1 induces the expression of FoxC2 and HoxD8, these results suggest that the transcriptional networks that include Prox1, HoxD8 and FoxC2 play important roles in Ang2 expression during lymphatic development.

Recently, Sox18 was reported to play important roles in the initial induction of Prox1 expression in venous endothelial cells by direct binding to the Prox1 promoter (François et al., 2008). Sox18 is expressed in a subset of venous endothelial cells prior to Prox1 expression and is coexpressed with Prox1 during the formation of the primary lymphatic plexus. However, Sox18 expression in LECs ceases during the maturation of lymphatic vessels. Endogenous Prox1 expression in LECs thus requires transcriptional inducers that are expressed in LECs. In the present study, we found that HoxD8 maintains endogenous Prox1 expression in LECs. These results suggest a novel positive-feedback-loop mechanism in which Prox1 expression is maintained by HoxD8, the expression of which is induced by Prox1. Our findings also indicate that HoxD8 plays crucial roles in the maturation and maintenance of lymphatic vessels, and suggest the possibility that HoxD8 can serve as a new therapeutic target in the treatment of inflammatory diseases, lymphedema, and lymphatic metastasis of tumors.

Maintenance, differentiation, culture and cell sorting of Tc-inducible MGZ5TcH2 ES cell lines were performed as described (Yamashita et al., 2000; Masui et al., 2005; Mishima et al., 2007). Briefly, VEGFR2-expressing endothelial progenitor cells derived from Tc-Empty or Tc-Prox1 ES cells were cultured in the absence or presence of Tc for 3 days. HUVECs and HDLECs were purchased from Sanko Junyaku and TaKaRa, and cultured in endothelial basal medium (EBM) containing 2% and 5% fetal bovine serum (FBS), respectively, supplemented with endothelial-cell growth supplement (TaKaRa). Recombinant adenoviruses encoding mouse Prox1 and HoxD8 were generated and used as described (Yamazaki et al., 2009). The HoxD8 construct in the pSG5 vector was kindly provided by Vincenzo Zappavigna (University of Modena and Reggio Emilia, Italy). Purification of adenoviruses was performed using Virakit for adenovirus 5 (Virapur).

We obtained LECs and BECs from mouse embryos as described previously (Hirashima et al., 2008). E14.5 mouse embryos, after removing their liver and spleen, were dissected and digested with 1.2 U/ml Dispase (Gibco), 50 μg/ml DNase1 (Roche) and 0.05% collagenase S-1 (Nitta Gelatin) to obtain single-cell suspension. After blocking Fc-receptors with an anti-mouse CD16/CD32 Fc receptor (FcR; PharMingen), all cells were stained with phycoerythrin (PE)-conjugated CD45 antibody (PharMingen) to sort CD45– non-hematopoietic cells using AutoMACS (Miltenyi Biotec). The cells were also stained with biotinylated anti-LYVE-1 antibody (ALY7; eBioscience) followed by allophycocyanin-conjugated streptoavidin (PharMingen) to visualize LYVE-1+ cells (LECs). The cells were also co-stained with a fluorescein isothiocyanate (FITC)-conjugated anti-PECAM-1/CD31 antibody (PharMingen) to visualize CD31+ cells (BECs). We sorted CD31+ LYVE-1– cells as BECs and CD31+ LYVE-1+ cells as LECs using FACS Vantage (Beckton Dickinson).

Total RNAs were extracted using the RNeasy Mini Kit (QIAGEN) to perform microarray and RT-PCR analyses. Oligonucleotide microarray analysis was performed using GeneChip Mouse Genome 430 2.0 Array (Affymetrix) according to the manufacturer's instructions. FileMaker Pro software (Filemaker) was used for statistical analysis. RNAs were reverse transcribed by random priming using Superscript III Reverse Transcriptase (Invitrogen). Expression of various Hox-family members was compared by RT-PCR analysis. PCR products were separated by electrophoresis in 1% agarose gel and visualized with ethidium bromide. Quantitative RT-PCR analysis was performed using the ABI PRISM 7500 Fast Real-Time PCR System (Applied Biosystems) and Power SYBR Green (Applied Biosystems). The primer sequences are shown in supplementary material Table S3.

siRNAs were introduced into cells using HiPerFect reagent (QIAGEN) according to the manufacturer's instructions. The target sequence for human HoxD8 was 5′-UUUGUCUUCCUCUUCGUCUACCAGG-3′ (Stealth RNAi; Invitrogen), whereas that for human Prox1 was 5′-CACCUUAUUCGGGAAGUGCAA-3′. Negative control was obtained from Invitrogen (Stealth RNAi Negative Control Med GC).

Whole mounts of diaphragms were fixed with 10% neutral-buffered formalin for 1 hour at 4°C and washed overnight in phosphate-buffered saline (PBS) containing 10%, 15% and 20% sucrose at 4°C, followed by immunostaining. Immunostaining was carried out with anti-β-galactosidase (1:1000 dilution; Cappel) and anti-LYVE-1 (1:200 dilution; Abcam) antibodies. Stained specimens were examined using an LSM 510 META confocal microscope (Carl Zeiss). All images were imported into Adobe Photoshop as JPEGs or TIFFs for contrast manipulation and figure assembly. Antibodies to FLAG epitope and α-tubulin for western blot analysis were obtained from Sigma. Anti-Prox1 and -Ang-2 antibodies were from Chemicon and Abcam, respectively. Anti-human FoxC2 was described previously (Mani et al., 2007). Western blot analyses were performed as described (Watabe et al., 2003). The bound antibody was detected using a chemiluminescent substrate (ECL; Amersham) and a LAS-4000 Luminescent image analyzer (Fuji Photo Film).

BALB/c mice at 5 weeks of age, obtained from Charles River Laboratories, were used. The model of chronic aseptic peritonitis was described previously (Iwata et al., 2007). We intraperitoneally injected 2 ml of 3% thioglycollate medium (BBL thioglycollate medium, BD Biosciences) into BALB/c mice every 2 days for 2 weeks to induce peritonitis. Adenovirus encoding β-galactosidase or HoxD8 was also intraperitoneally injected twice per week during the same period. The mice were then sacrificed, and their diaphragms excised and prepared for immunostaining as described above. Statistical analysis was performed using GraphPad Prism5 (GraphPad Software). Results were expressed as individual values and mean values ± s.e. Differences were evaluated by Mann-Whitney test and considered statistically significant at P<0.05. Plaques consisting of macrophages were obtained from the peritoneal surface of the diaphragm and treated with RNAlater (Ambion), followed by RNA isolation and quantitative RT-PCR analyses for VEGF-A, VEGF-C and VEGF-D.

Recombinant adenoviruses encoding mouse Prox1 and HoxD8 were generated and used as described (Yamazaki et al., 2009). Ang2 construct was kindly provided by Gou Young Koh (Korea Advanced Institute of Science and Technology, Republic of Korea).

Cells were seeded at a density of 5×10⁴ cells/well in 12-well plates and transfected with siRNAs using HiPerfect (QIAGEN). Cells were trypsinized and counted by a Coulter counter on day 2 after siRNA transfection. The experiments were performed in triplicate.

Migration assay was performed as described previously (Mishima et al., 2007). As a chemoattractant, 100 ng/ml of recombinant VEGF-C (Calbiochem) were used.