FOXO1 regulates developmental lymphangiogenesis by upregulating CXCR4 in the mouse-tail dermis

The lymphatic system participates in fluid homeostasis, immune cell trafficking and surveillance, and absorption of fatty acids from the small intestine. Congenital deficits in the lymphatic vasculature can result in embryonic lethality with severe edema and impaired immune reactions (Tammela and Alitalo, 2010). Lymphangiogenesis (the growth of new lymphatic vessels from pre-existing lymphatic vasculature) is associated with normal embryonic development and diverse pathological conditions such as lymph node metastasis of malignant tumors, lymphedema distichiasis and inflammation (Kerjaschki, 2014; Tammela and Alitalo, 2010). However, despite its importance in various physiological and pathological conditions, the molecular mechanisms underlying lymphangiogenesis remain less understood than those involved in angiogenesis.

The first identified lymphatic-specific key regulators of developmental lymphangiogenesis were vascular endothelial growth factor (VEGF) C, VEGFD and their receptor VEGF receptor 3 (VEGFR3), which is expressed on the lymphatic endothelial cell (LEC) membrane (Achen et al., 1998; Enholm et al., 1998; Joukov et al., 1996; Yamada et al., 1997). Ligand binding to VEGFR3 activates downstream signaling molecules such as JNK, ERK and the PI3K/Akt pathway, resulting in cell proliferation, migration and LEC survival (Karkkainen and Petrova, 2000; Mäkinen et al., 2001). VEGFR3-deficient mice die at embryonic day (E) 9.5 because of cardiovascular failure before the emergence of lymphatic vessels (Dumont et al., 1998). Previous findings have also demonstrated that knockout (KO) of the Vegfc gene in mice impairs the formation of lymphatic vascular networks (Karkkainen et al., 2004). Moreover, it has previously been reported that fibroblast growth factor 2, angiopoietin 1, insulin-like growth factors 1 and 2, and platelet-derived growth factor BB contribute to lymphangiogenesis (Rinderknecht and Detmar, 2008). Recent data have highlighted several cases where established angiogenic signaling, such as Notch signaling (Krebs et al., 2000), ephrinB2/EphB4 signaling (Adams et al., 1999) and TGFβ signaling (Ferrari et al., 2009), also play pivotal roles in developmental lymphangiogenesis (Niessen et al., 2011, 2014; Wang et al., 2010). Regarding the transcriptional regulation of lymphangiogenesis, the PROX1 transcription factor in lymphatic endothelium is an essential regulator of developmental lymphangiogenesis (Wigle and Oliver, 1999). Other transcription factors, such as GATA2, NFATC1, COUP-TFII and FOXC2, are also reported to be regulators of developmental lymphangiogenesis, lymphatic valve development or LEC specification (Kazenwadel et al., 2015; Kulkarni et al., 2009; Norrmén et al., 2009; Srinivasan et al., 2010).

Forkhead box (FOX) transcription factor O (FOXO) family proteins, such as FOXO1, FOXO3a, FOXO4 and FOXO6, are mammalian orthologs of the Caenorhabditis elegans DAF-16 protein, and control the expression of genes involved in apoptosis (i.e. Bcl2l11 and Fasl), cell cycle progression (i.e. Cdkn1a and Cdkn1b), DNA damage repair (i.e. Gadd45a), cell metabolism (i.e. G6pd), oxidative stress resistance (i.e. Cat and Sod2) and other cellular functions (Accili and Arden, 2004; Eijkelenboom and Burgering, 2013; Huang and Tindall, 2007). The FOXO-family proteins are functionally rich downstream components of PI3K/Akt (Brunet et al., 1999). Akt phosphorylates FOXO proteins at conserved sites and induces their nuclear export, which suppresses their transcriptional activities (Accili and Arden, 2004; Brunet et al., 1999; Huang and Tindall, 2007). We and other groups have previously demonstrated that the FOXO1 is essential for developmental angiogenesis through histological studies in FOXO1 KO mice (Fukumoto et al., 2018; Furuyama et al., 2004; Wilhelm et al., 2016). Furthermore, through an in vitro study, FOXO1-deficient LECs failed to migrate toward exogenous ATP, owing to decreased purinergic receptor P2Y1 expression (Niimi et al., 2017). In the present study, we aimed to investigate the physiological function of FOXO1 in lymphangiogenesis in vivo; to this end, we produced endothelial cell-specific FOXO1-deficient mice and observed developmental lymphangiogenesis, focusing on cellular functions such as proliferation, apoptosis and migration. Moreover, we examined FOXO1-dependent cell migration and proliferation activity in vitro using primary lymphatic endothelial cells.

We used the mouse-tail dermis as a model with which to study developmental lymphangiogenesis. Developmental lymphangiogenesis occurs in the tail dermis during early postnatal development [postnatal day (P) 0-8], where it forms characteristic hexagonal networks of lymphatic vasculature. The mouse tail is, therefore, a good model for investigating abnormal developmental lymphangiogenesis (Niessen et al., 2011, 2014; Xu et al., 2010). In order to observe normal lymphangiogenic processes in the tail dermis, we performed immunohistochemical staining at various postnatal stages (P1-5) in wild-type mouse-tail dermis using a LYVE1-specific antibody, a marker of the lymphatic vasculature. Our results show that the quadrangular disconnected networks observed at P1 gradually developed into hexagonal connected networks at P5 (Fig. S1A). Data from a previous study showed that the lymphatic vasculature of the tail dermis forms characteristic multi-ring-shaped structures known as lymphatic ring complexes (LRCs), many of which are involved in developmental lymphangiogenesis (Xu et al., 2010). Our immunohistochemistry demonstrated that LRCs gradually become more robust during P1-5 (Fig. S1B), and that some of the filopodia-like structures extend toward other LRCs during P1-4 (Fig. S1B, arrows). The layer where these mesh-patterned lymphatic vessels exist is referred to as middle layer. Moreover, compared with the middle layer, LYVE1-positive ascending lymphatic vessels were observed in deeper layers (hereafter, referred to as the deep layer) (Fig. 1A). Very thin LYVE1-positive lymphatic vessels were observed in the layer above the middle layer – hereafter referred to as the superficial layer of the P3 and P5 tail dermis (Fig. 1A). However, superficial lymphatic vessels were not present in the P1 dermis (Fig. 1A). A schematic image detailing the development of the tail dermal lymphatic vessels is shown in Fig. 1B. It is considered that lymphatic vessels grow from the deep layer upwards towards the superficial layer. Next, we investigated the localization of FOXO1 within the vasculature of the wild-type developmental tail dermis. Immunohistochemistry experiments revealed that FOXO1 was strongly expressed in the LEC nuclei of each layer of P1, P3 and P5 dermis (Fig. 1C, Fig. S1C). In Pecam1⁺ LYVE1⁻ blood vessels, FOXO1 was strongly expressed in the blood endothelial nuclei of the superficial layer of each stage (Fig. S1C,E). However, we observed weaker nuclear FOXO1 expression in the P5 middle and deep blood vasculature of each stage compared with the lymphatic vasculature (Fig. 1C,D, Fig. S1C,D). This suggests that FOXO1 KO might have a more significant influence on the formation of lymphatic capillaries when compared with blood vessels, at least in the early postnatal tail dermis.

To study the influence of FOXO1 on developmental lymphangiogenesis in the tail dermis, we first used a tamoxifen-inducible Tie2 (tyrosine kinase with immunoglobulin-like and EGF-like domain 2) -positive endothelial cell-specific Foxo1 conditional KO mice: Tie2-CreER^T2-Foxo1^flox/flox. A Tie2-CreER^T2-Foxo1^flox/flox mouse injected with tamoxifen is hereafter referred to as Foxo1^Tie2KO. Tie2-CreER^T2-Foxo1^flox/+ mice were used as littermate controls. Tie2 expression in developmental lymphatic vessels has previously been reported (Morisada et al., 2005). The schedule of tamoxifen injection is shown in Fig. S2A. Immunohistochemical analysis of tail dermis and western blot analysis of lung lysate from KO pups revealed that FOXO1 was successfully knocked out (Fig. S2B,C). The body weight of KO pups significantly decreased at P5 (Fig. S2D). All KO pups died by P11 (Fig. S2E), although we could not find any macroscopic findings, such as hemorrhage and edema. Anti-LYVE1 immunohistochemical staining showed that there was a remarkable reduction in the lymphatic vasculature in the middle layer of the dermis of KO mice compared with the control dermis at P5 (Fig. S2F). Further study revealed that tamoxifen-induced Cre-mediation results in reduced lymphatic vessel length and in fewer branch points per LRC in Foxo1^Tie2KO mice compared with those of control mice at P5 (Fig. S2G-J). Moreover, lymphatic vessels were dilated in Foxo1^Tie2KO mice compared with control mice at P5 (Fig. S2G,K). Next, cell proliferation was evaluated by intraperitoneal BrdU injection. The results show that the number of BrdU-positive LECs per length was higher in Foxo1^Tie2KO dermis compared with the control dermis (Fig. S2L,M). Subsequently, we evaluated apoptosis by immunostaining using antibody against cleaved caspase 3. Results show that the number of apoptotic LECs was lower in Foxo1^Tie2KO dermis compared with the control dermis (Fig. S2N,O). Furthermore, we studied the blood capillaries present in the dermis of Foxo1^Tie2KO mice using anti-Pecam1 and anti-LYVE1 immunohistochemical staining (Fig. S3). Unlike the remarkable phenotype in the lymphatic vasculature, we observed only a slight increase in the length of blood capillaries in the superficial layer of Foxo1^Tie2KO dermis (Fig. S3A-G).

As both endothelial and hematopoietic cells express Tie2, Foxo1 was deleted in the endothelium and also in hematopoietic cells in the Foxo1^Tie2KO mice. Consequently, inactivation of Foxo1 in hematopoietic cells may play a role in lymphangiogenesis. To assess this hypothesis, we investigated dermal lymphatic development in Cdh5-CreER^T2-Foxo1^floxflox mice with tamoxifen-induced endothelial Foxo1 KO. Tamoxifen-treated Cdh5-CreER^T2-Foxo1^floxflox mice are hereafter referred to as Foxo1^Cdh5KO. The littermates of Cdh5-CreER^T2-Foxo1^flox/+ mice were used as controls. The schedule of tamoxifen injection was the same as that explained for Tie2 Cre-mediated KO (Fig. 2A). Successful Cdh5 Cre-mediated Foxo1 KO was confirmed by immunohistochemical analysis of tail dermis and western blot analysis of lung lysates (Fig. 2B,C). The body weight of P5 Foxo1^Cdh5KO pups was significantly reduced compared with that of control pups (Fig. 2D). Interestingly, unlike Tie2 Cre-induced KO, ∼70% of Foxo1^Cdh5KO pups survived until at least P21 (Fig. 2E). The dermal lymphatic phenotypes of P5 Foxo1^Cdh5KO pups were similar to Foxo1^Tie2KO pups, with reduced vessel length and branch points, increased vessel width, increased LEC proliferation, and decreased LEC apoptosis (Fig. 2F-O). We next performed anti-VEGFR3 immunostaining in P5 pups to assess the phenotype of collecting or pre-collecting lymphatic vessels. The length of collecting and pre-collecting lymphatic vessels was not altered in KO pups, while reduced branch points and increased width of collecting and pre-collecting lymphatic vessels were observed in KO pups compared with control pups (Fig. S4A-D). We also assessed mesenteric lymphatic vessels in P5 KO pups to determine whether the phenotypes are dermis specific or not. Anti-PROX1 immunostaining revealed that mesenteric lymphatic vessels were not affected in terms of branch points and width in KO pups (Fig. S5A-C). We also investigated the blood vessels of P5 Foxo1^Cdh5KO pups. Immunohistochemical studies showed that Pecam1⁺ LYVE1⁻ blood vessels were not different in terms of their length, branch points and diameter in middle layer (Fig. S6A,F-H). Interestingly, we found that blood vessel width in the superficial layer was dilated in Foxo1^Cdh5KO pups compared with control pups (Fig. S6A,K).

For more precise analysis of the Foxo1 KO phenotype on lymphatic vasculature, we next used the Prox1-CreER^T2-Foxo1^flox/flox mouse, in which a Foxo1 deficit was specifically induced in lymphatic vessels but not in blood vessels following tamoxifen treatment. Tamoxifen-treated Prox1-CreER^T2-Foxo1^floxflox mice are hereafter referred to as Foxo1^Prox1KO. Littermates of Prox1-CreER^T2-Foxo1^flox/+ mice were used as experimental controls. The schedule of tamoxifen injection was the same as that used in case of Tie2 and Cdh5 Cre-mediated KO (Fig. S7A). Immunohistochemical analysis of tail dermis and western blot analysis of lung lysates showed successful Foxo1 KO in the lymphatic vasculature of Foxo1^Prox1KO pups (Fig. S7B,C). The fitness of Foxo1^Prox1KO pups was good at P5. The body weight of Foxo1^Prox1KO pups was not different from control pups (Fig. S7D). All Foxo1^Prox1KO mice survived until P21 (Fig. S7E). The dermal lymphatic abnormalities in P5 Foxo1^Prox1KO pups were similar to those observed in Foxo1^Tie2KO and Foxo1^Cdh5KO pups; these mice had reduced vessel length and branch points, dilated vessels, increased numbers of BrdU-positive LECs and decreased cleaved caspase 3-positive LECs (Fig. S7F-O).

The following analysis was performed with Foxo1^Cdh5KO mice because they allow for long-term analysis and have the most severe lymphatic vasculature phenotypes. It is possible that FOXO1-independent pathways correct the lymphatic abnormalities observed in Foxo1-deficient vasculature in the long term. To investigate this possibility in Foxo1^Cdh5KO mice, we treated mice with tamoxifen at P1, P2 and P3, and subsequently harvested tail dermis at P8 or P21 (Fig. 3A). Efficient Foxo1 KO in the lymphatic vasculature was confirmed by dermal immunohistochemistry and western blot analysis of lung lysates (Fig. S8A-D). Similar to P5, we found that at P8 there was a reduction in middle lymphatic vessel length and branch points compared with control mice (Fig. 3B,D,E). We also observed dilated lymphatic vessels in the middle layer of Foxo1^Cdh5KO mice (Fig. 3B,F), but total lymphatic vessel length was not altered in the superficial layer of Foxo1^Cdh5KO mice compared with control mice (Fig. 3B,C). At P21, dermal lymphatic capillaries were properly connected and they covered the whole tail dermis in both control and Foxo1^Cdh5KO mice (data not shown). Total lymphatic vessel length in the middle layer was not altered (Fig. 3G,I). However, at P21, we observed increased diameter and a reduction in the number of branch points per LRC in the middle layer of Foxo1^Cdh5KO mice compared with control mice as with P5 and P8 (Fig. 3G,J,K). Interestingly, unlike at P8, the total length of lymphatic capillaries in the superficial layer at P21 decreased in Foxo1^Cdh5KO mice compared with control mice (Fig. 3G,H).

To investigate whether the time points for tamoxifen treatment affected lymphatic development in Foxo1^Cdh5KO mice, we injected mice with tamoxifen at P4 and P5 or P6 and P7, and subsequently sacrificed and harvested tail dermis at P8 (Fig. 4A). Successful Foxo1 KO in lymphatic vasculature was confirmed by dermal immunohistochemistry and western blot using lung lysate (Fig. S8E-H). When we injected tamoxifen at P4 and P5, we found that the total lymphatic vessel length, number of branch points per LRC and lymphatic vessel diameter in the middle layer were not altered in Foxo1^Cdh5KO mice compared with control mice (Fig. 4B,D-F). However, the total vessel length of superficial lymphatic capillaries was decreased in Foxo1^Cdh5KO mice compared with that of control mice (Fig. 4B,C). When tamoxifen was injected at P6 and P7, total lymphatic vessel length in the middle layer was also not altered in Foxo1^Cdh5KO mice compared with control mice (Fig. 4G,I). However, the branch point count per LRC was reduced and the lymphatic vessel in the middle layer was slightly dilated in Foxo1^Cdh5KO mice compared with control mice (Fig. 4G,J,K). Interestingly, the total vessel length of superficial lymphatic capillaries was not altered in Foxo1^Cdh5KO mice compared with control mice (Fig. 4G,H). The results of long-term analysis and injection-point analysis are summarized in Fig. 4L.

To investigate the cause of lymphatic phenotypes in Foxo1 KO dermis, we performed complementary DNA (cDNA) microarray analysis using cDNA samples from primary human dermal LECs (HDLECs) transfected with FOXO1-specific small interfering RNA (siRNA; siFOXO1) or a negative-control siRNA (siNC). FOXO1 knockdown using siFOXO1 was successful (Fig. 5B). The cDNA microarray results are shown in Tables S1 and S2 as the log ratios of the target-gene expression levels in siFOXO1 versus siNC transfectants. Among these genes, expression of CXCR4 was especially decreased in siFOXO1-transfected HDLECs when compared with that in siNC-transfected HDLECs (Table S1), as confirmed by quantitative RT-PCR (qRT-PCR; Fig. 5A). Furthermore, using western blotting and immunocytochemical analysis, we found that CXCR4 levels were downregulated in FOXO1 knock-down HDLECs (Fig. 5B,C). CXCR4 and its ligand, C-X-C chemokine ligand 12 (CXCL12), form a signaling axis that is involved in normal development of the hematopoietic, cardiovascular, renal and nervous systems (Mithal et al., 2013; Nagasawa et al., 1996; Sierro et al., 2007; Takabatake et al., 2009). In fact, previous findings have revealed that the CXCR4/CXCL12 axis contributes to physiological and pathological angiogenesis through its chemotactic activity, both in vitro and in vivo (Mirshahi et al., 2000; Ping et al., 2011; Salcedo et al., 1999; Strasser et al., 2010). Based on these findings, we performed immunohistochemical staining with CXCR4-specific antibodies using tail dermis from P1, P3 and P5 wild-type mice. We used an anti-CXCR4 antibody (clone UMB2) that specifically recognizes the non-phosphorylated CXCR4 protein. Therefore, absence of a signal using this antibody suggests that either CXCR4 is absent, or that all of the expressed CXCR4 is phosphorylated. For clarification, we performed pretreatment of whole tail dermis with lambda protein phosphatase (PP) before immunohistochemical staining. No signal post-PP treatment suggests that CXCR4 is absent. In the tail dermis of P1 wild-type mice, the CXCR4 protein was expressed on lymphatic capillaries in the middle and deep layers (Fig. 5D,H). However, in the P3 and P5 dermis of wild-type mice, the CXCR4 protein was only expressed in the lymphatic capillaries of the superficial layer, but not in the middle and deep layers (Fig. 5F-H). In the wild-type tail dermis that was not incubated with PP, the CXCR4 protein was not detected in lymphatic vessels at P1 (Fig. 5E), P3 or P5 stages (data not shown), indicating that all of the expressed CXCR4 protein in the lymphatic vessels of postnatal developing tail dermis is phosphorylated. To test CXCR4 reduction in Foxo1-KO dermis, we performed immunohistochemistry using anti-CXCR4 and anti-LYVE1 antibodies in the KO dermis and control dermis treated with PP at P3 and P5. However, it was difficult to show the CXCR4 reduction in P3 and P5 lymphatic capillaries because of the absence of superficial lymphatic vessels in the dermis of KO mice (Fig. S9A,B). As a result, we performed short-term KO analysis in the P1-2 dermal middle lymphatic vasculature that expressed CXCR4. Tamoxifen was injected 12 h after birth (P0.5) and tail dermis was harvested 36 h after birth (P1.5) (Fig. 6A). Efficient Foxo1 KO was confirmed by dermal immunohistochemistry and western blot using lung lysate (Fig. 6B,C). Anti-CXCR4 and anti-LYVE1 immunohistochemical studies in the dermis of P1.5 mice treated with PP displayed a remarkable reduction in CXCR4 levels in the lymphatic vasculature of Foxo1^Cdh5KO dermis compared with control dermis (Fig. 6D,E). Furthermore, to identify the source of CXCL12, we performed immunohistochemistry with a CXCL12-specific antibody in control and Foxo1^Cdh5KO mice tail dermis. Results from these experiments revealed that CXCL12 was strongly expressed in the middle layer of the lymphatic vasculature but downregulated in superficial lymphatic capillaries (Fig. S10). The localization and signal intensity of CXCL12 were not altered by Foxo1 KO (Fig. S10).

In order to determine whether the FOXO1 protein drives CXCR4 expression, we performed luciferase reporter assays using luciferase reporter vectors containing variable-length segments of the mouse genomic Cxcr4 region (497 bp, 224 bp or 75 bp upstream of the start codon). Among these, the 75 bp region of genomic Cxcr4 corresponds to the 5′-UTR. The Cxcr4 promoter region proximal to the transcription start site (TSS) is highly conserved between humans and mice (Fig. 7A). The strongest luciferase activity was observed following co-transfection with a reporter vector containing 224 bp or 497 bp of the Cxcr4 genomic region, and a plasmid encoding a nuclear-localized constitutively active Foxo1 mutant [FOXO1(CA)] (Fig. 7B). The 5′-UTR had no luciferase activity (Fig. 7B). The results suggest that the 149 bp region upstream of the TSS could be a potential target region of the FOXO1 protein. Furthermore, ChIP qPCR shows that anti-FOXO1 antibodies precipitated this region of the CXCR4 promoter (Fig. 7C). Exon 2 of CXCR4 (negative control) was not precipitated by anti-FOXO1 antibody (Fig. 7C). However, this region overlaps with a TATA box (Fig. 7A) and is not a typical FOXO1-binding element (FBE). Consequently, we next investigated the possibility that FOXO1 acts as a transcriptional factor in the upstream enhancer elements of the CXCR4 gene. To identify the enhancer of this gene, we used the ENCODE database (the database of DNA functional elements; www.encodeproject.org/) and found H3K27ac binding (in HUVECs), FOXO1 binding (in B-lymphocyte) and a DNaseI hypersensitivity region ∼120 kb upstream from the TSS (Fig. 7D). Within that region, we found a ∼1.3 kb region with three FBEs (Fig. 7E), one of which is fully conserved between humans and mice (Fig. 7E). We performed anti-FOXO1 ChIP qPCR to amplify three different regions (regions 1-3) each containing one FBE (Fig. 7E). Region 3 contains the fully conserved FBE (Fig. 7E). The results show that region 3 was precipitated most abundantly by FOXO1 (Fig. 7F). This indicates that FOXO1 binds directly to the putative enhancer of the CXCR4 gene in LECs.

CXCL12 is reported to chemoattract CXCR4-expressing cells, and to mediate lymphocyte trafficking and immune functions (Bleul et al., 1996, 1997); however, its effects on LECs are still poorly understood. Therefore, we analyzed HDLEC migration towards CXCL12 gradients in collagen-coated, transwell membranes by counting Giemsa-stained migratory cells. HDLEC migration increased when the lower chamber contained 0.1-100 ng ml⁻¹ CXCL12; however, chemoattraction towards 1000 ng ml⁻¹ CXCL12 was marginal (Fig. 8A,B). Moreover, migration was inhibited by pre-incubation with 5 μM of the CXCR4 antagonist AMD3100 (Fig. 8A,B). These results indicate that HDLECs migrate towards extracellular CXCL12 via CXCR4. As mentioned above, FOXO1 knockdown led to downregulation of CXCR4 expression in HDLECs. It is therefore possible that FOXO1 knockdown affected HDLEC migration towards CXCL12. To investigate this possibility, we examined the migration of siNC or siFOXO1 transfected HDLECs, transduced with adeno-associated viral EGFP vector (AAV-EGFP) or AAV-CXCR4, in collagen-coated transwell membranes containing 100 ng ml⁻¹ CXCL12. We found that AAV-CXCR4 could rescue the Cxcr4 mRNA downregulation in siFOXO1 transfectants (Fig. 8C). The migration assay showed that HDLEC migration of siFOXO1 transfectants was lower than that observed in siNC transfectants (Fig. 8D,E), and that CXCR4 overexpression rescued the decreased migration observed in siFOXO1 transfectants (Fig. 8D,E).

We next examined the proliferative effect of CXCL12 on HDLECs in BrdU assays. Immunocytochemical staining showed that CXCL12 increased BrdU-positive HDLECs in a dose-dependent manner. This dose-dependent effect was lost and BrdU-positive cells increased in low CXCL12 levels in siFOXO1 transfected HDLECs (Fig. S11A,B). AMD3100 inhibited CXCL12-dependent LEC proliferation in siNC transfectants, but did not inhibit the increased proliferation observed in siFOXO1 transfectants (Fig. S11A,B). The increased cell proliferation observed in FOXO1-deficient cells is generally thought to be due to a reduction of FOXO-mediated cell cycle inhibitory genes, such as CDKN1A, CDKN1B, CDKN2A, CDKN2B, RBL2 and GADD45A (Bouchard et al., 2007; Furukawa-Hibi et al., 2002; Katayama et al., 2008; Kops et al., 2002; Medema et al., 2000; Nakamura et al., 2000; Seoane et al., 2004; Tran et al., 2002). To investigate their expression, qRT-PCR analysis targeting these genes was performed. Interestingly, among the cell-cycle inhibitory genes we analyzed, only CDKN2B was downregulated in siFOXO1 transfected HDLECs (Fig. S11C). We next investigated the reduction of CDKN2B in Foxo1-deficient dermal lymphatic vessels. Anti-CDKN2B immunohistochemical studies in control and Foxo1^Cdh5KO dermis shows that CDKN2B is downregulated in Foxo1^Cdh5KO lymphatic vasculature compared with control mice (Fig. S11D,E). These results agreed with the increased cell proliferation in the dermal lymphatic vasculature in KO pups. We next investigated the mechanism of reduced apoptotic LECs observed in Foxo1-deficient lymphatic vasculature. In general, reduced apoptosis in FOXO1-deficient cells is thought to be caused by reduction of pro-apoptotic proteins, such as Bim (encoded by BCL2L11) or FasL (encoded by FASLG), which are directly transcribed by FOXO (Brunet et al., 1999; Dijkers et al., 2000; Gilley et al., 2003; Stahl et al., 2002). We investigated the mRNA expression of BCL2L11 and FASLG in siFOXO1 transfected HDLECs compared with siNC transfectants. We observed no FASLG expression in HDLECs (data not shown). BCL2L11 expression was decreased in siFOXO1 transfected HDLECs compared with siNC transfectants (Fig. S11F). We next performed immunohistochemistry with anti-Bim antibody on the tail dermis of control and Foxo1^Cdh5KO mice. We found that Bim was expressed in lymphatic capillaries in P5 control dermis, but drastically downregulated in P5 Foxo1^Cdh5KO dermis (Fig. S11G,H).

We next examined the effect of CXCR4 blockage in the developing tail dermis. Scratched wild-type tails were immersed in 2 mM AMD3100 solution daily from P1 to P4 and tail dermises were subsequently harvested at P5 and used for immunohistochemistry (Fig. 9A). Tails were scratched to allow for efficient permeation of AMD3100. To check the permeation of AMD3100 into the dermis, we performed anti-CXCR4 immunohistochemical staining in tail dermises that were not treated with PP. AMD3100 treatment allowed us to detect CXCR4 without PP treatment (Fig. 9B), which meant efficient CXCR4 inactivation. Immunohistochemical studies using anti-LYVE1 antibodies showed decreased lymphatic vessel length in superficial and middle layers, but not in the deep layer in AMD3100-treated tail dermis compared with vehicle control-treated dermis (Fig. 9C-F). In the middle layer of AMD3100-treated dermis, LRC and its complexities were decreased compared with the vehicle control (Fig. 9C,G,H). However, lymphatic vessel width in the middle layer of AMD3100-treated dermis was not altered compared with vehicle control (Fig. 9C,I).

Failure to form normal connected lymphatic vasculature, as presently occurs in the Foxo1-deficient pups, is likely caused by impaired LEC sprouting and migration, which is essential for normal lymphatic network formation (James et al., 2013). Similar developmental lymphangiogenesis abnormalities in the tail dermis have been previously reported in mice with blocked signaling of Notch1-Dll4, ALK1 or VEGFR3 (Niessen et al., 2011, 2014). We identified downregulation of Bim and CDKN2B in Foxo1-deficient LECs as possible causes of dilation of lymphatic vessels in lymphatic Foxo1-deficient pups. However, Wilhelm et al. (2016) recently reported that endothelial FOXO1 works as a gatekeeper of endothelial quiescence to suppress excess growth and proliferation of endothelial cells by inhibiting MYC protein. Our qPCR analysis detected a slight increase of the MYC in FOXO1-depleted HDLECs (data not shown). Therefore, Myc upregulation might be one cause of the excess proliferation observed in Foxo1-deficient lymphatic vessels. Alternatively, the data concerning BrdU incorporation in HDLECs (Fig. S11A,B) suggest that FOXO1-deficient LECs lose control of cell proliferation due to addition of exogenous CXCL12. This might be partly attributable to the downregulation of CDKN2B, which induces G1 arrest. Dilation of lymphatic vessels was reported in the dorsal skin of Foxc1^−/−, Foxc2^−/−, Tgfbr1^−/− and Tgfbr2^−/− mice (Fatima et al., 2016; James et al., 2013). However, Foxc1^−/− and Foxc2^−/− mice do not have decreased numbers of lymphatic branch points, and Tgfbr1^−/− and Tgfbr2^−/− mice do not have decreased numbers of apoptotic cells in their lymphatic vessels. In this respect, the phenotypes observed in Foxo1-deficient lymphatic vasculature are likely due to primary effects. The same phenotypes observed presently between the three different Cre-mediated deletions of FOXO1 and minor blood vessel abnormalities in these KO pups also strongly suggest that lymphatic malformation in endothelial Foxo1-deficient mice primarily arise from LEC abnormalities. Altered blood vessel formation was evident only in the superficial layer of Foxo1^Tie2KO and Foxo1^Cdh5KO dermis, consistent with strong FOXO1 localization in superficial blood vessels, but not in middle and deep blood vessels of P1 and P3 dermis (Fig. 1C,D, Fig. S1C,D).

Interestingly, we observed different survival ratios of Foxo1^Tie2KO, Foxo1^Cdh5KO and Foxo1^Prox1KO mice (Fig. 2, Figs S2 and S7). Conventional and constitutive Tie2-Cre mediated deletion of Foxo1 are reportedly lethal in the middle embryonic period (Furuyama et al., 2004; Wilhelm et al., 2016). Wilhelm et al. also reported that mice with Pdgfb-Cre mediated conditional deletion of Foxo1 survived until at least P21. These data suggest that the death of the Foxo1^Tie2KO pups was due to non-endothelial abnormalities, such as hematopoietic or endocardial malformation. Considering that Foxo1^Prox1KO pups did not die, the death of Foxo1^Cdh5KO pups was likely due to blood vessel abnormalities, except for skin tissue. The detailed cause of death requires further study.

Long-term data of Foxo1^Cdh5KO mice at P8 or P21 (Fig. 3) suggest that FOXO1 is necessary for sprouting and migration of LECs from the middle layer to the superficial layer, and suggest that other FOXO1-independent pathways might compensate for the aforementioned sprouting and migration of LECs between P5 and P8. Reduced superficial lymphatic vessels in P21 KO dermis indicates that FOXO1 is required to maintain superficial capillaries. Moreover, the results of our later inactivation analysis in Foxo1^Cdh5KO mice (Fig. 4) suggest that lymphatic vessels in the middle layer undergo development until P3, and that superficial lymphatic vessels undergo FOXO1-dependent development from P4 to P5. These results also indicate the completion of sprouting and migration of LECs from the middle to the superficial layer by P5. However, we cannot explain the different phenotypes in the middle lymphatic capillaries between the P4 to P5 treatment and the P6 to P7 treatment. As one possibility, the different duration of Foxo1 deficit (4 days in the P4, P5-P8 regimen and 2 days in the P6, P7-P8 regimen) might affect the downstream gene profile of FOXO1.

Our findings potentially implicate CXCR4 downregulation in impaired developmental lymphangiogenesis in Foxo1-deficient mice. CXCR4 expression in the endothelial cells of normal mammalian tissues may be limited in blood vessels, especially in the arteries. CXCR4 protein is reportedly expressed in blood capillaries, but not in lymphatic capillaries during inflammation-related vasculogenesis in the mouse ear dermis (Zgraggen et al., 2014). However, CXCR4 was expressed in the trunk lymphatic vessel of developing zebrafish, with limited localization and time-frame for expression (Cha et al., 2012), and in lymphatic vessels growing into a Matrigel plug containing VEGFC and in the tumor-associated lymph node and melanoma tissue (Zhuo et al., 2012). These data suggest that CXCR4 is expressed in lymphatic vessels under tightly regulated conditions, such as peripheral cytokine concentrations. These findings are consistent with our findings that CXCR4 expression in developing dermal lymphatic capillaries was very transient and localized, indicating that CXCR4 is expressed in lymphatic capillaries that are still developing, and not in fully formed lymphatic vessels. The lack of alteration in mesenteric lymphatic vessels in Foxo1^Cdh5KO pups (Fig. S5) is evidence that lymphangiogenic regulation by FOXO1-CXCR4 is tissue and time specific. We also found abnormalities in (pre-) collecting lymphatic vessels in Foxo1^Cdh5KO pups (Fig. S4). Because we did not detect CXCR4 in collecting lymphatics (data not shown), this phenotype seems not to be attributable to the downregulation of CXCR4. This is consistent with the lack of reduction in collecting vascular length (Fig. S4B).

CXCL12 is expressed in the epidermis, papillary dermis, blood vessels, nerve tissues, sweat glands and hair follicles of adult rat skin (Avniel et al., 2006). CXCL12 was also localized in LYVE1-positive dermal lymphatic endothelium in a 2,4-dinitro-1-fluorobenzene-induced contact dermatitis mice model (Kabashima et al., 2007). However, it is unlikely that LEC-derived autocrine CXCL12 induces the migration of LECs because of the non-concomitant expression of CXCR4 and CXCL12 in the LECs of the same layer. Moreover, chemoattraction between the LECs of the two different layers is also unlikely because the pattern of CXCR4/CXCL12 expression (CXCL12 in the middle LECs and CXCR4 in the superficial LECs) does not match the direction of chemoattraction. Another source of CXCL12 for LEC migration likely exists. As CXCL12 is a secreted protein, in situ hybridization could be suitable to detect this elusive CXCL12 source.

Among other FOX transcription factors, FOXC1 and FOXC2 increase Cxcr4 expression in mouse embryonic endothelial cells (Hayashi and Kume, 2008). FOXC1 and FOXC2 are regulate developmental lymphangiogenesis in embryonic mice (Fatima et al., 2016). Moreover, FOXO3a binds to the promoter region of the CXCR4 gene in prostate cancer cells (Dubrovska et al., 2012). In this study, qRT-PCR demonstrated that FOXO1 knockdown did not induce FOXC1/2 or FOXO3A downregulation (data not shown), suggesting that CXCR4 downregulation caused by FOXO1 knockdown is a primary effect.

Consistent with the migration assay results (Fig. 8), a previous study has reported that murine primary LECs migrate toward CXCL12 supplied at 1 to 100 ng ml⁻¹ (Zhuo et al., 2012). Other studies have demonstrated that various CXCR4-positive cells, including lymphocytes, hematopoietic cells, malignant tumor cells, endothelial cells and epithelial cells, migrate toward CXCL12 at approximately the same concentrations (Fernandis et al., 2003, 2004; Mirshahi et al., 2000; Wuchter et al., 2014; Yokohama-Tamaki et al., 2015). Furthermore, the impaired lymphangiogenesis in postnatal tail dermis treated with CXCR4 antagonist strongly suggests that impaired lymphangiogenesis in the tail dermis of Foxo1-deficient mice is attributable, in part, to the downregulation of the CXCR4 protein in LECs, followed by inhibition of cell migration towards the surrounding CXCL12 gradient. Interestingly, CXCR4 blockage in the developing tail dermis did not alter lymphatic vessel width, unlike the Foxo1-deficient tail dermis. This indicates that dilated lymphatic vessels in Foxo1-deficient tail are not completely due to attenuated CXCL12-CXCR4 signaling, but instead to decreased FOXO1 and its other target genes.

The present data suggest that FOXO1 KO in LECs decreases migration toward CXCL12 by downregulating CXCR4 (Fig. 10). FOXO1 KO induces excess LEC proliferation and decreased LEC apoptosis (Fig. 10). This results in a disconnected and dilated structure of the lymphatic vasculature in endothelial Foxo1-deficient mouse-tail dermis (Fig. 10). Increased proliferation and decreased apoptosis in FOXO1-deficient LECs might be caused by the reduction of CDKN2B and Bim (Fig. 10). However, CDKN2B and Bim are two of various possibilities. Furthermore, the roles that FOXO1 serve in physiological lymphangiogenesis during early embryonic development and in pathological lymphangiogenesis during inflammation or tumor cell recruitment, as that observed in the tail dermis, remain unknown. Inhibition of CXCR4/CXCL12 signaling suppresses corneal inflammatory lymphangiogenesis (Du and Liu, 2016). Future studies to elucidate the involvement of FOXO1 in pathological lymphangiogenesis would clarify the utility of this protein as a potent therapeutic target.

The production of C57BL/6-Foxo1^flox/flox and C57BL/6-Foxo1^flox/+ mice was as previously described (Miyazaki et al., 2012). Tie2-CreER^T2 C57BL/6 mice showing endothelial cell-specific tamoxifen-inducible Cre recombinase expression (Forde et al., 2002) were kindly provided by Dr Bernd Arnold (German Cancer Research Center, Heidelberg, Germany). Tie2-Cre ER^T2-Foxo1^flox/flox mice and Tie2-Cre ER^T2-Foxo1^flox/+ mice were used for experiments. The Cre genotype was determined by PCR using the primer set shown in Table S4 and flox phenotypes were determined as previously described (Miyazaki et al., 2012). Tamoxifen (Sigma-Aldrich) dissolved in peanut oil (Sigma-Aldrich) was intraperitoneally injected (50 μg/g body weight) into Tie2-Cre ER^T2-Foxo1^flox/flox and Tie2-Cre ER^T2-Foxo1^flox/+ mice at the indicated time point to specifically delete Foxo1 in endothelial cells. The production of Cdh5-CreER^T2 mice is as previously reported (Okabe et al., 2014). Cdh5-Cre ER^T2-Foxo1^flox/flox mice and Cdh5-Cre ER^T2-Foxo1^flox/+ mice were used for experiments. The production of Prox1-CreER^T2 mice is as previously reported (Bazigou et al., 2011). Prox1-Cre ER^T2-Foxo1^flox/flox mice and Prox1-Cre ER^T2-Foxo1^flox/+ mice were used for experiments. All the animal experiments were approved by The Osaka University Graduate School of Medicine Intramural Animal Care and Use Committee or The Kagawa Prefectural University of Health Sciences Institutional Animal Care and Use Committee. We used as few animals as possible and took steps to minimize their suffering.

Mouse-tail dermis was harvested as described previously (Niessen et al., 2014), cut to 3 mm in width and fixed in 100% methanol at −20°C.

Saline solution containing BrdU (Wako, Japan) was intraperitoneally injected into mice (50 μg/g body weight) at P5 1 h before they were sacrificed.

The tails of wild-type C57BL/6 mice were scratched with a needle at P1 and immersed in PBS containing 50% DMSO (vehicle) or 2 mM AMD3100 from P1 to P4 daily and harvested at P5. The purpose of this scratching was to allow for efficient permeation of AMD3100 and not to induce pathological lymphangiogenesis. The efficient CXCR4 antagonism in this method was confirmed (Fig. 9B). Only one streaky scratch was made on the dorsal side of tail down to the depth of the epidermis. The duration of exposure was 5 min. Immunohistochemical studies were performed only on the ventral side of tail dermis to avoid the influence of pathological conditions.

Tissues were pre-incubated with 1 M hydrochloric acid (Sigma-Aldrich) for 1 h at 37°C to detect BrdU. The whole tail dermis was washed, blocked with blocking reagent (Perkin Elmer) and incubated with the primary antibodies shown in Table S3. To detect total CXCR4 protein using the UMB2 antibody (Abcam), the whole-tail dermis was incubated with 800 U Lambda Protein Phosphatase (New England Biolabs) for 1.5 h at 32°C before blocking. After incubation with the primary antibodies, the tissues were washed and incubated with the secondary antibodies (Table S3) and subsequently washed and mounted. All tissues were observed by confocal laser scanning microscopy (Olympus).

HDLECs were purchased from PromoCell (Germany) and cultured in a 5% CO₂ atmosphere at 37°C, using medium from the Endothelial Basal Medium MV2 Kit (PromoCell) supplemented with 5% FCS. HDLECs from passage 5 were used for all experiments. The HDLEC we purchased originated from the juvenile foreskins of individual donors. All in vitro analysis was performed using HDLECs from two different donors and confirmed only minor variation between these. HeLa cells were cultured in DMEM containing 10% FBS in a 5% CO₂ atmosphere.

Specific siRNAs against human FOXO1 (siGENOME human FOXO1 Smart Pool 2308) and a random siRNA as a negative control (non-targeting siRNA#3) were purchased from Thermo Fisher Scientific. The final additive concentration of siRNA was 0.2 nM, which is low enough to limit off-target effects of siRNA. All siRNAs were transfected using the Lipofectamine RNAiMAX reagent (Invitrogen). AAV-EGFP and AAV-CXCR4 vector plasmid were produced using AAVpro Helper Free System (AAV6) (Takara Bio, Japan), transfected in HEK293T (AAVpro 293T cell line, Takara Bio, Japan) and purified with AAVpro purification kit (Takara Bio, Japan). We used the AAV vector at 1:1000 dilution.

A cDNA microarray study was performed using cDNA from HDLECs transfected with siFOXO1 or siNC RNA for 72 h. The array was examined by Takara Bio (Japan) using the Agilent Expression Array service.

Human recombinant SDF1 α (CXCL12) was purchased from Peprotech and dissolved in 0.1% BSA to a final concentration of 100 ng μl⁻¹. AMD3100, a specific antagonist of CXCR4, was purchased from Sigma-Aldrich and dissolved in distilled water to a final concentration of 20 mM. The final additive concentration of AMD3100 was 5 μM.

After various treatments, HDLECs were fixed with 4% paraformaldehyde (PFA). To detect BrdU, cells were incubated in 1 M hydrochloric acid for 20 min at 37°C. Cells were then treated with 0.5% (v/v) Triton X-100, blocked with BSA (Sigma-Aldrich), and subsequently incubated with the appropriate primary antibody (Table S3), washed and incubated with the appropriate secondary antibody (Table S3). After washing, cells were treated with DAPI (Sigma-Aldrich) and mounted. Slides were observed by fluorescence microscopy (BX53-34-FL-2, Olympus). To detect CXCR4 expression, cells were incubated with primary antibody before fixation with 4% PFA.

HDLECs were lysed in the buffer containing 2% SDS and protease/phosphatase inhibitor cocktail (Nacalai Tesque, Japan). Cell lysates were subjected to SDS-PAGE and transferred to a PVDF membrane (Millipore, USA). The membrane was blocked with 3% skim milk and incubated with the appropriate primary antibody (Table S3). After washing, the membrane was incubated with the appropriate secondary antibody (Table S3). The membrane was developed using the ECL detection system (GE Health Care) or X-ray film.

Total RNA was harvested from HDLECs with RNA ISO (Takara Bio, Japan). Subsequently, cDNA was synthesized with reverse transcriptase (ReverTraAce with gDNA remover, TOYOBO, Japan) and qPCR was performed using the THUNDERBIRD qPCR Mix (TOYOBO, Japan) in the CFX Connect Real-Time System (Bio-Rad) or StepOnePlus Real-Time System (Applied Biosystems). Expression levels were normalized to ACTB expression. The primer sets used for qRT-PCR are shown in Table S4.

The 497 bp, 224 bp and 75 bp fragments from the murine genomic CXCR4 region, upstream from the start codon, were cloned by PCR using genomic DNA from a C57BL/6J mouse. Briefly, flanking 5′ XhoI and 3′ HindIII sites were added during PCR, and the PCR product was digested with these restriction enzymes. Subsequently, the PCR products were inserted in the pGL4.14 vector (Promega), which was pre-digested with XhoI and HindIII. Construction of the murine FOXO1-expression vector (pcDNA3.1-FOXO1-myc-His) has been previously described (Furuyama et al., 2000). Next, 1.5×10⁴ HeLa cells were transfected with 100 ng each of the reporter vector and the FOXO1-expression vector (or empty pcDNA3.1 vector), as well as 6 ng of the pGL 4.74 TK promoter-driven Renilla luciferase vector as a transfection control, in a 96-well plate using the Lipofectamine 2000 reagent (Invitrogen). At 24 h post-transfection, luciferase activity was measured using the Dual-Luciferase Reporter Assay System (Promega) in a chemiluminescence-scanning plate reader (Berthold Technologies).

Eighty percent confluent HDLECs in three 10 cm dishes were fixed in 1% formalin and used for ChIP assays. ChIP assays were performed using the ChIP-IT Express kit (Active Motif) and an anti-FOXO1 antibody (1:67, Abcam, ab39670). Primers used are shown in Table S4. qPCR was performed using the THUNDERBIRD qPCR Mix (Toyobo) using StepOnePlus Real-Time System (Applied Biosystems). The data were calculated as fold enrichment compared with the isotype IgG control.

The apical chambers of transwell plates (Falcon Cell Culture Inserts, 24-well, 8 μm; Corning) were pre-coated with type-I collagen from rat tail (BD Biosciences). HDLECs (5.0×10⁴) in serum-free medium (if necessary containing 5 μM AMD3100) were added to the upper chamber. The lower chamber was filled with serum-free medium containing various concentrations of CXCL12 or BSA (as a control). Transwell membranes were inserted and incubated for 5.5 h at 37°C. After incubation, non-migrated cells in the apical chamber were removed, and the basolateral side of each filter was fixed with 4% PFA and subsequently Giemsa stained. Migratory cells were counted by microscopy.

All image analyses were performed with the analysis software for FV10i (Olympus) or ImageJ software (Abràmofff et al., 2004; Schneider et al., 2012). The detailed method for measurement of lymphatic vessels are shown as Fig. S12 and its legend. Blood vessels were analyzed with the same method used for lymphatic measurement. Total blood vessel length was measured and then divided by the area of the microscopic field. We analyzed three control pups and three KO pups, with three microscopic fields per pup. BrdU-positive cells and cleaved caspase 3-positive cells were counted and divided by vessel length.

Nuclear FOXO1 fluorescence intensity was measured with ImageJ software. Each LEC nucleus was encircled after DAPI staining and fluorescence intensity inside the surrounded regions were calculated. All LEC nuclei in a ×1600 microscopic field were measured. The microscopic field with the best anti-FOXO1 immunostaining property was chosen for each sample and measured. Three wild-type pups at each stage were analyzed.

All experiments were repeated at least three times. For large sample sizes (n≧5), statistical analyses were performed with Student's t-test or Welch's t-test. P<0.05 was considered to indicate a statistically significant difference. The P values were corrected using Bonferroni's method for multiple comparisons. For small sample sizes (n<5), no statistical analysis was performed and describing whether it is a ‘significant’ difference or not has been avoided. Instead, an individual data plot for small sample size comparison is shown.

We thank Dr Bernd Arnold for providing Tie2CreER mice and Dr Taija Mäkinen for providing Prox1 CreER mice.