Placental labyrinth formation in mice requires endothelial FLRT2/UNC5B signaling

The placenta is an organ that is found primarily in mammals and that connects a developing fetus to the maternal uterine wall. The placental labyrinth is a structure located on the fetal side of the placenta, consisting of two major cellular components: trophoblasts and endothelial cells (Cross et al., 1994). Trophoblasts, which are the post-implantation derivatives of the trophectoderm, line the maternal blood spaces, whereas endothelial cells, which arise from the allantoic mesoderm of the embryo, border the fetal circulation (Downs, 2002). These two juxtaposed layers provide the interface for the bidirectional transfer of gases, nutrients, embryonic metabolic end-products, and other molecules between the embryo and mother. Hence, their proper formation is essential for the development of embryos (Rossant and Cross, 2001). Vascular development consists of three major processes: vasculogenesis, angiogenesis and vascular remodeling. In principle, all three of these processes depend on or are highly affected by signaling by vascular endothelial growth factor (VEGF) and its receptors (Potente et al., 2011). However, the detailed mechanisms surrounding VEGF vary greatly from organ to organ, probably contributing to organ-specific vascular patterning (Adams and Alitalo, 2007). The mechanisms that mediate vascularization of the labyrinth share many of the molecules that are required for embryonic vascular development, including hypoxia-inducible factors and VEGF expressed in trophoblasts (Adelman et al., 2000; Rossant and Cross, 2001; Cross et al., 2003; Hirashima et al., 2003). However, the endothelial machinery that specifically mediates establishment of the placental vasculature is largely unclear.

The nervous and vascular systems have been found to share a number of guidance cues for network formation, including members of the ephrin, netrin, semaphorin and slit families (Adams and Eichmann, 2010; Quaegebeur et al., 2011). Fibronectin leucine-rich transmembrane protein 2 (FLRT2) is a member of the fibronectin leucine-rich transmembrane (FLRT) family of proteins, which act as repulsive ligands of the UNC5 receptor family and as chemorepellents for neurons (Lacy et al., 1999; Yamagishi et al., 2011; Seiradake et al., 2014). Outside of the nervous system, FLRT2 is expressed in the epicardium, and its global knockout leads to reduced thickness of the ventricular myocardium, suggesting a role for FLRT2 in heart morphogenesis (Müller et al., 2011). However, the importance of FLRT2 as a neurovascular cue and the detailed function of FLRT2 in cardiovascular development are still unclear.

Here, we unexpectedly found that FLRT2 is expressed in endothelial cells specifically in the placental labyrinth. Mice lacking endothelial FLRT2 exhibited aberrant alignment of the placental endothelium, which disturbed feto-maternal circulation and might have been responsible for embryonic lethality at mid-gestation. This vascular deformity was caused by a lack of the endothelial repulsion that results from binding of FLRT2 to UNC5B. Netrin 1 has long been thought to be the functional ligand of UNC5B, with absence of netrin 1/UNC5B signaling being responsible for the vascular defects in UNC5B-deficient mice (Lu et al., 2004; Larrivée et al., 2007). Indeed, proteolytic fragments of netrin 1 bind UNC5B in diabetic endothelial cells and increase vascular permeability (Miloudi et al., 2016). However, in terms of developmental angiogenesis, the situation regarding the role of netrin 1/UNC5B signaling is complex and has been subject to debate; notably, the abnormalities reported for UNC5B-deficient mice are absent in netrin 1 mutants (Serafini et al., 1996; Wilson et al., 2006; Castets and Mehlen, 2010; Yung et al., 2015). Our results suggest that FLRT2 is the predominant ligand that is associated with the role of UNC5B in vascular development and indicate that fine-tuning of vascular morphogenesis by inter-endothelial repulsion contributes crucially to the development of the placental labyrinth.

To elucidate the function of FLRT2 in vascular development, we utilized Tie2-Cre mice, classically used to delete genes in endothelial cells. By crossing these mice with indicator mice (flox-CAT-EGFP), we observed Tie2-Cre expression in endothelial cells and hematopoietic cells at embryonic day (E) 12.5 (Fig. S1A-C), in agreement with results presented previously (Constien et al., 2001). When Flrt2-flox mice were crossed with Tie2-Cre mice, the resultant endothelial/hematopoietic-specific Flrt2 knockout mice (Tie2-Cre⁺Flrt2^flox/flox: herein referred to as Flrt2^ΔEHC) died between E13.5 and E14.5 (Fig. 1A,C). Contrary to this observation, in a previous study, some global Flrt2 knockout mice survived until weaning (Müller et al., 2011). This discrepancy could be explained by the difference in genetic backgrounds, as the previous study involved 129/C57Bl/6J mice, whereas our experiments were based on pure C57Bl/6J mice.

To examine the relevance of hematopoietic FLRT2, we crossed Flrt2-flox mice with Vav1-iCre mice (de Boer et al., 2003), in which expression of an optimized variant of Cre is expressed in all hematopoietic cells but not in endothelial cells (Fig. S1D-F). The hematopoietic Flrt2 knockout mice (Vav1-iCre⁺Flrt2^flox/flox: herein referred to as Flrt2^ΔHC) were viable with no detectable abnormalities (Fig. 1B,C).

To explore the cause of embryonic lethality in Flrt2^ΔEHC mice, they were examined at earlier stages of development. At E12.5, six out of 13 Flrt2^ΔEHC mice had no cardiac rhythm, and the remainder showed systemic congestion (Fig. 1D-F) without any apparent vascular deformities (Fig. 1G-N). However, at E11.5, none of the defects seen at E12.5 was evident (Fig. S1G-K). These data raised the possibility that the lack of FLRT2 in the endocardium of Flrt2^ΔEHC mice impaired cardiac morphogenesis and caused heart failure at E12.5 because Tie2-Cre is also expressed in the endocardium (Fig. S1L,M). However, morphological and histological examination of the hearts of Flrt2^ΔEHC mice showed no apparent structural defects at E12.5 (Fig. 1E,F,O,P). The specific role of FLRT2 in endocardium was examined by crosses with Nfatc1-Cre mice (Wu et al., 2012), in which Cre is expressed in all endocardial cells, some macrophages, and a few aortic endothelial cells, but not in peripheral endothelial cells (Fig. S1N-Q). The resultant Flrt2 knockout mice (Nfatc1-Cre⁺Flrt2^flox/flox) were viable and showed no detectable abnormalities (Fig. 1Q-T). Moreover, utilizing αMHC-Cre mice, we determined that FLRT2 expression in the myocardium is also dispensable for embryonic development (Fig. 1T). Together, these results indicated that endothelial FLRT2 was required for mid-gestation embryonic survival, and that a separate crucial abnormality might exist outside the embryos causing the lethality in Flrt2^ΔEHC mice.

The expression pattern of FLRT2 was examined through the generation of Flrt2^+/lacZ knock-in mice (Fig. S2A,B) and examination of β-galactosidase expression in these mice. In agreement with previously published results, we observed abundant β-galactosidase expression in a population of neurons and the epicardium (Yamagishi et al., 2011; Müller et al., 2011), but not in embryonic endothelial cells (Fig. 2A-C,G,H,K-M). Consistently, in the neonatal retina, we detected β-galactosidase expression in retinal ganglion cells, but not in endothelial cells (Fig. S2C,D), and deletion of Flrt2 in retinal ganglion cells by the use of Chx10-Cre mice did not impair retinal vascularization (Fig. S2E-G). It has previously been suggested (but not shown directly) that FLRT2 is expressed in the placental endothelium (Müller et al., 2011). Placental expression was therefore examined in our Flrt2^+/lacZ knock-in mice, which had a mosaic pattern of β-galactosidase expression in endothelial cells of the placental labyrinth (Fig. 2D-F,I,J,N,O). Immunohistochemistry with anti-FLRT2 antibodies detected patchy expression localized around endothelial cells in the placental labyrinth (Fig. S3A-C). This endothelial FLRT2 immunoreactivity was diminished in Flrt2^ΔEHC but not in Flrt2^ΔHC mice (Fig. S3D-I), confirming the specificity of the antibodies. Reverse transcription polymerase chain reaction (RT-PCR) analysis of cells sorted by fluorescence-activated cell sorting (FACS) (Fig. S4) confirmed that Flrt2 expression was enriched in CD31 (PECAM1)⁺ CD45 (PTPRC)⁻ endothelial cells in the placenta (Fig. 2P,Q). Moreover, Flrt2 expression was elevated in placental CD31⁺ endothelial cells at E12.5 (Fig. S5), suggesting that there is a requirement for endothelial FLRT2 around this developmental stage.

Embryonic death occurring from E10 to E13 is attributed to failures in cardiac function, hematopoiesis or placentation in many cases (Papaioannou and Behringer, 2012). Our observation of the expression of FLRT2 in placental endothelial cells suggested that Flrt2^ΔEHC mice might have placental defects, as Cre was expressed in placental endothelial cells in Tie2-Cre but not in Vav1-iCre mice (Fig. S6A,E). Placental Tie2-Cre expression was not observed in any trophoblastic-lineage cells marked by CD9 or the monocarboxylate transporters Mct1 (SLC16A1) or Mct4 (SLC16A3) (Fig. S6B-D), was limited to endothelial cells, and was not seen in hematopoietic cells (Fig. S6A), unlike in other tissues (Fig. S1A-C). Histological examination of the placenta revealed that the labyrinth was poorly formed and compact in Flrt2^ΔEHC mice at E12.5 (Fig. 3A-F), although the spongiotrophoblast layer was not affected (Fig. S7A-D). Detailed analysis revealed that numbers of nucleated (fetal) erythrocytes, which are normally contained within the endothelial compartment, were considerably lower in Flrt2^ΔEHC mice than in controls, whereas non-nucleated (maternal) erythrocytes successfully circulated in the epithelial compartment that was bordered by trophoblasts (Fig. 3G-N,Q). Further analysis of the endothelial structure of Flrt2^ΔEHC mice showed that endothelial alignment was aberrant and the endothelium was multi-layered, effectively obstructing the vascular lumen (Fig. 3K-N). Immunohistochemistry with laminin antibodies showed that the basement membrane was invaginating into the endothelial lumen of placentas in Flrt2^ΔEHC mice (Fig. 3O,P). The rate of endothelial proliferation did not differ between control and Flrt2^ΔEHC mice (Fig. S7E-G), suggesting that the observed endothelial multi-layering could not be attributed to aberrant proliferation.

The consequences of perturbed feto-maternal gas exchange were explored by examination of the embryos and placentas of Flrt2^ΔEHC mice. Bright-field microscopy images suggested that the branches of the umbilical arteries and veins in Flrt2^ΔEHC mice were not occluded, but instead contained dark-colored blood (Fig. 4A-C), possibly as a result of impaired gas exchange in the labyrinth. Accumulation of a hypoxia marker, pimonidazole, was significantly higher in Flrt2^ΔEHC embryos than in controls, notably in the head region, where oxygen consumption is probably very high (Smith et al., 2013) (Fig. 4D-G).

The precise action of FLRT2 on endothelial cells was elucidated with an in vitro assay involving primary cultures of endothelial cells. Previously, X-ray crystallography (Seiradake et al., 2014) revealed that FLRT2 binds to FLRT2 (homophilic binding) as well as to UNC5B, UNC5C and UNC5D (heterophilic binding) via different extracellular regions of FLRT2 (Fig. 5A). FLRT2 is known to act as a repulsive cue on neurons through binding to UNC5B and UNC5D (Yamagishi et al., 2011). Among the UNC5 proteins, only UNC5B is indispensable for embryonic development, and Unc5b knockouts are lethal at approximately E12.5, similar to the Flrt2^ΔEHC genotype (Lu et al., 2004; Navankasattusas et al., 2008). It was, therefore, important to select an endothelial line that expressed both FLRT2 and UNC5B. Human umbilical vein endothelial cells (HUVECs) express these two genes at high levels, making this cell line appropriate for further in vitro studies (Fig. 5B). To evaluate whether the repulsive mechanism of interaction between FLRT2 and UNC5B acts in endothelial cells in the same way as in neuronal cells (Yamagishi et al., 2011; Seiradake et al., 2014), a Transwell assay and a stripe assay were performed in HUVECs with recombinant FLRT2 and UNC5B proteins. Compared with vehicle only, FLRT2 significantly decreased migration of HUVECs in the Transwell assay and resulted in repulsion of the cells away from stripes containing the recombinant protein (Fig. 5C-L; Movie 1). These effects were almost completely abolished by neutralization of FLRT2 by the addition of UNC5B, suggesting that the repulsive action of extracellular FLRT2 was mediated by heterophilic interaction with UNC5B, and not by homophilic binding to FLRT2 (Fig. 5C-L). Similar effects were seen with human aortic endothelial cells (HUAECs) (Fig. 5M,N), which express UNC5B but have little or no endogenous expression of FLRT2 (Fig. 5B). Human dermal lymphatic endothelial cells (HDLECs) have little or no UNC5B expression (Fig. 5B), and they did not show repulsion from FLRT2 in stripe assays (Fig. S8A), confirming that the repulsive action of FLRT2 was mediated by UNC5B. To evaluate the role of endogenous UNC5B expression in seeded cells, knockdown of Unc5b by small interfering RNA (siRNA) was performed in HUVECs. Treatment with si-Unc5b significantly reduced the repulsive action of FLRT2 on HUVECs (Fig. S8B).

Notably, the effects of FLRT2 on proliferation in HUVECs and HUAECs were negligible (data not shown), suggesting that, in vitro, FLRT2 is involved primarily in endothelial repulsion and not in proliferation, which is in agreement with the in vivo data in placentas of Flrt2^ΔEHC mice (Fig. S7E-G).

Our results suggested that FLRT2 signals through UNC5B in the placental labyrinth. Endothelial cells in the labyrinth express UNC5B at high levels (Navankasattusas et al., 2008). In Unc5b-deficient mice, the results of a previous study with Doppler-flow analysis showed that blood flow is reversed in diastole within the umbilical arteries, suggesting uteroplacental insufficiency (Navankasattusas et al., 2008). Immunohistochemical analysis of histological abnormalities in placentas of Unc5b^−/− mice showed that the area of the placental labyrinth was significantly reduced at E12.5 compared with Unc5b^+/+ mice (Fig. 6A,B,I), although the spongiotrophoblast layer was not affected (Fig. S9). Similar to our observations in Flrt2^ΔEHC mice, placental endothelial alignment was perturbed and fetal circulation was disrupted in Unc5b^−/− mice (Fig. 6C-H). Taken together, these data suggest that the vascular abnormalities in Flrt2^ΔEHC placentas are associated with the repulsive action of FLRT2, acting on the UNC5B receptor on endothelial cells.

VEGF/VEGF receptor signaling is required for vascular growth during development throughout the body. However, the cellular and molecular mechanisms associated with vascular morphogenesis in each organ vary, and organ-specific angiogenesis involves various VEGF-dependent and VEGF-independent processes. Our results have identified a cellular and molecular mechanism that is specifically active in placental angiogenesis. We have demonstrated that formation of the placental labyrinth requires coordinated inter-endothelial repulsion mediated by FLRT2, which is known to be a repulsive guidance cue for axons, acting via binding to the UNC5B receptor. Our data suggested that, in both Flrt2^ΔEHC mice and Unc5b^−/− mice, fetal blood circulation in the placental labyrinth was blocked because of disorganization of vascular structures, impairing feto-maternal gas exchange (Fig. 6J).

The requirement for FLRT2 in development was first demonstrated in whole-body Flrt2 knockout mice (Müller et al., 2011), in which the thickness of the ventricular walls was severely reduced at E11.5, which is earlier than our observation of placental defects in Flrt2^ΔEHC mice. The phenotype of the whole-body Flrt2 knockout mice was primarily caused by impaired epicardial integrity, without any endocardial abnormality. In Flrt2^ΔEHC mice, we observed thinning of ventricular walls at E14.5 (data not shown) but not at E12.5, suggesting that placental defects precede cardiac defects in these mice. Overall, the phenotype we observed in Flrt2^ΔEHC mice seems to be independent of the cardiac defects previously observed in global Flrt2 knockout mice (Müller et al., 2011).

An important aspect of our findings is that they suggest that FLRT2 is the predominant ligand of UNC5B, which explains the severe vascular defects observed in UNC5B-deficient mice (Lu et al., 2004; Navankasattusas et al., 2008). Netrin 1 has long been thought to be the ligand of UNC5B that is involved in angiogenesis. Treatment of endothelial cells with netrin 1 results in tip-cell-filopodia retraction, and this effect is abolished in UNC5B-deficient mice. Morpholino knockdown of ntn1 (the gene encoding netrin 1) in zebrafish has demonstrated the repulsive guidance activity of netrin 1 for endothelial cells through UNC5B signaling (Lu et al., 2004). However, Ntn1 knockout mice survive until birth and have no vascular abnormalities (Yung et al., 2015), indicating that netrin 1 is not necessarily the only ligand for UNC5B, and an absence of netrin 1 might be compensated for by other ligands in vivo, specifically roundabout homolog 4 or netrin 4 (Koch et al., 2011; Lejmi et al., 2008; Li et al., 2012). Conflicting reports exist regarding the role of netrin 1 as either a pro-angiogenic or anti-angiogenic factor (Lu et al., 2004; Larrivée et al., 2007; Wilson et al., 2006; Castets and Mehlen, 2010). One argument that appears to resolve this dilemma is the ability of netrin 1 to block apoptosis that would otherwise be induced by unbound UNC5B receptors, thus promoting the survival of endothelial cells during angiogenesis (Castets et al., 2009).

Vascular repulsive cues are usually provided by non-endothelial cells around blood vessels. For example, semaphorin 3E, which is secreted by retinal neurons, binds plexin D1-expressing endothelial tip cells and represses their ability to project filopodia (Kim et al., 2011). Notably, consequences of the loss of semaphorin 3E or plexin D1 are unevenness of the growing front and reduction in branching of the vascular network. Our results also provide evidence that a lack of proper endothelial repulsion results in an overall reduction in vascular growth. Endothelial cell-derived semaphorin 3A has been shown to repel filopodia projection by endothelial tip cells through plexin A1 and neuropilin 1 receptors (Ochsenbein et al., 2016). In agreement with this result, in our model endothelial FLRT2 acts on neighboring UNC5B-expressing endothelial cells to fine-tune alignment and prevent aberrant layering. Upon disruption of FLRT2/UNC5B signaling, endothelial clustering without a change in cell number might lead to failure to expand the total area of vascular plexus.

In contrast to the mosaic pattern of expression of FLRT2 that we observed in endothelial cells, UNC5B has previously been ubiquitously detected in all placental endothelial cells (Navankasattusas et al., 2008). Transmembrane FLRT2 proteins are proteolytically cleaved to release soluble ectodomains (Seiradake et al., 2014; Yamagishi et al., 2011) that diffuse into the extracellular spaces. Thus, it is likely that one FLRT2-expressing endothelial cell acts repulsively on multiple UNC5B-expressing cells, both in paracrine and autocrine manners. Our results indicated that the luminal/abluminal polarity was somehow disrupted in the Flrt2^ΔEHC mice, as the basement membrane was invaginating into the endothelial lumen. At the same time, endothelial cells were abnormally clustered, suggesting their impaired alignment or deregulated migration. To evaluate these kinetic components (polarity, alignment and migration) individually will require in vivo live imaging, which could not be applied to the mouse placenta with technology available at the time of this study. Although we could not clearly separate the defects in alignment, polarity and migration, the results of our in vitro studies suggest that the defects in Flrt2^ΔEHC mice involve a mixture of these factors. Endothelial collapse, which is another potential cause of this kind of endothelial clustering, was not apparent, as indicated by the presence of the apoptotic marker cleaved caspase 3 in very few endothelial cells; also, the number of such apoptotic endothelial cells was equal in placentas of control and Flrt2^ΔEHC mice (data not shown).

Finally, our data have demonstrated that a FLRT2-UNC5B interaction, resulting in inter-endothelial repulsion, is required for the proper organization of endothelial structures in the placental labyrinth. It will be of interest to explore the roles of similar repellent mechanisms directed by other guidance molecules in the development of other vascular beds. Furthermore, it will be essential to determine whether FLRT2 contributes to adult angiogenesis, in particular angiogenesis associated with certain neoplasms, as tumor angiogenesis often utilizes developmentally important angiogenic molecules (Kubota, 2012).

Animal care was performed in accordance with the Guidelines of Keio University for Animal and Recombinant DNA experiments. The Tie2-Cre (Arita et al., 2014), Chx10-Cre (Muranishi et al., 2011; Yoshikawa et al., 2016), Unc5b^−/− (Wilson et al., 2006), Nfatc1-Cre (Wu et al., 2012), αMHC-Cre (Agah et al., 1997) and flox-CAT-EGFP (Kawamoto et al., 2000) mice have been described previously. Vav1-iCre mice (de Boer et al., 2003) were obtained from the Jackson Laboratory. An embryonic stem cell clone [Flrt2^{tm1a(EUCOMM)Wtsi}] obtained from the European Conditional Mouse Mutagenesis Program (EUCOMM) was used to generate Flrt2-flox mice. To generate Flrt2^+/lacZ mice, the Flrt2 knock-in allele was generated by homologous recombination of exon 2, replaced by a lacZ gene followed by a loxP-flanked PGK-promoter-driven neomycin gene. Flrt2^+/lacZ mice were maintained on a CD1 background, and the rest of the mutant mice were crossed with C57BL/6J mice at least ten times. For Vav1-iCre-specific deletion, a Flrt2^flox/flox male was mated with Vav1-iCre^hetFlrt2^flox/+ females. Otherwise, a Cre^hetFlrt2^flox/+ male was mated with Flrt2^flox/flox females. For analysis, embryos and neonates of both sexes were used without exact determination.

Surgically dissected tissues were fixed overnight in 4% paraformaldehyde (PFA) in PBS, and then snap-frozen in optimal cutting temperature (OCT) compound (Sakura Finetechnical, Tokyo, Japan). All placental specimens were sectioned (10-15 µm thick) at the plane where umbilical arteries and veins were attached.

Enucleated eyes were fixed for 20 min in 4% PFA in PBS and then dissected as described previously (Okabe et al., 2014). Retinal cups were post-fixed for 20 min. Other tissues were dissected out from embryos and fixed overnight in 4% PFA in PBS. All whole-mount tissues were stained as described below.

IHC of whole-mount samples or tissue sections was performed as described previously (Kubota et al., 2011). The primary monoclonal antibodies used were hamster anti-CD31 (Chemicon, MAB1398Z; 1:1000), Cy3-conjugated mouse anti-α-smooth muscle actin (Sigma-Aldrich, C6198; 1:500), rat anti-Ter119 (also known as LY76) (R&D Systems, MAB1125; 1:1000), rabbit anti-CD9 (Abcam, ab92726; 1:500) and rat anti-F4/80 (also known as ADGRE1) (Serotech, MCA497R; 1:500). The primary polyclonal antibodies used were Alexa Fluor 488-conjugated rabbit anti-GFP (Molecular Probes, A21311; 1:500), goat anti-GFP (Rockland, 600-101-215; 1:2000), rabbit anti-laminin (Sigma-Aldrich, L9393; 1:500), rabbit anti-ERG (Abcam, ab92513; 1:2000), rabbit anti-β-galactosidase (Millipore, AB986; 1:500), chicken anti-Mct1 (Millipore, AB1286-I; 1:200), rabbit anti-Mct4 (Millipore, AB3314P; 1:200), goat anti-FLRT2 (R&D Systems, AF2877; 1:1000), and rabbit anti-trophoblast specific protein alpha (Tspα; also known as TPBPA) (Abcam, ab104401; 1:1000). Secondary antibodies used were Alexa Fluor 488-conjugated IgGs (Molecular Probes, A11034, A11006, A11055; 1:500) or Cy3/Cy5- DyLight549/DyeLight649-conjugated IgGs (Jackson ImmunoResearch, 711-165-152, 112-165-167, 127-165-160, 711-605-152, 112-605-167, 127-605-160; 1:500). For the 5-ethynyl-2'-deoxyuridine (EdU) incorporation assay, the Click-iT EdU Alexa Fluor 488 Imaging Kit (Invitrogen, C10337) was used. In brief, pregnant mice were injected intraperitoneally 2 h before euthanasia with 0.5 ml EdU solution (0.5 mg ml⁻¹) and stained according to the manufacturer's protocol. For the detection of hypoxic cells, the Hypoxyprobe-1 Plus Kit (Chemicon, HP2-100) was used according to the manufacturer's instructions. In brief, 60 mg kg⁻¹ pimonidazole was injected intraperitoneally into pregnant mice 30 min before euthanasia, and sections were stained with Hypoxyprobe Mab1-FITC. For nuclear staining, specimens were treated with 4′,6-diamidino-2-phenylindole (DAPI) (Molecular Probes, D-1306).

Embryos, retinas and placentas were fixed for 30 min in 1% glutaraldehyde in PBS and stained as whole-mount samples or snap-frozen in OCT compound before sectioning. The samples were then incubated for 2 h for whole mounts or 1 h for sections at 37°C with 1 mg ml⁻¹ 5-bromo-4-chloro-3-indolyl-β-D-galactopyranoside (X-gal) in a standard X-gal reaction buffer (35 mM potassium ferrocyanide, 35 mmol l⁻¹ potassium ferricyanide, 2 mmol l⁻¹ MgCl₂, 0.02% NP-40 and 0.01% sodium deoxycholate in PBS). For whole-mount embryos and placentas, stained tissues were clarified by immersion in a solution of benzyl alcohol and benzyl benzoate (1:2). For sectioned specimens, Nuclear Fast Red (Thermo Fisher Scientific) was used for counter-staining.

Fluorescent images were obtained using a confocal laser scanning microscope (FV1000; Olympus, Tokyo, Japan). Quantification of cells or parameters of interest was conducted in a 600×600 µm field of view (for all in vitro assays), a 200×200 µm field of view (for counting EdU⁺ cells) or the entire section (for measuring radial growth and the labyrinth area) in scanned images. ImageJ software (NIH, Bethesda, MD, USA) was used for the quantification of indicated areas in scanned images.

Placental tissues were incubated for 20 min at 37°C in Dulbecco's Modified Eagle's Medium (DMEM) containing 1 mg ml⁻¹ collagenase IV from Clostridium histolyticum (Sigma-Aldrich), 1 U ml⁻¹ Dispase (Thermo Fisher Scientific) and 1 U ml⁻¹ DNase (Takara-Bio), before cells were dissociated by gentle trituration. Dissociated single cells were preincubated with Fc block (Becton Dickinson) to avoid nonspecific binding of antibodies, and then incubated with allophycocyanin (APC)-conjugated anti-CD31 (Biolegend, 102510; 1:500) and APC-Cy7-conjugated anti-CD45 (Biolegend, 103116; 1:500). Stained samples were analyzed and sorted by SORP FACSAria (Becton Dickinson) with FlowJo software (TreeStar, Ashland, OR, USA), as described previously (Kubota et al., 2011). Debris and dead cells were excluded by forward and side scatter and a negative gate for propidium iodide staining (Fig. S4). FACS studies routinely included unstained control samples and compensation tubes for APC and APC-Cy7. With these negative and positive control tubes, fluorescence voltages and the compensation matrix were set according to the manufacturer's instructions and applied to all the samples that were analyzed.

Total RNA was prepared from placental tissues or cultured cells, and reverse transcription was performed using Superscript II (Invitrogen). cDNA samples were then subjected to PCR amplification with the following primers: (1) mouse Flrt2, 5′-AGACAAGGCTGCCAGATTACA-3′ and 5′-TAAAATGCAACGTGATGGGG-3′; (2) mouse Cd31, 5′-GCTGAGGAAAACTCCTTCAC-3′ and 5′-TTCTGGATGGTGAAGTTGGC-3′; (3) mouse Mct1, 5′-ATTGGAGGGCCAGTGGGATA-3′ and 5′-TCAGACAGGGCTCTCCTCCT-3′; (4) mouse Mct4, 5′-GCTGTGGTGGACGAGGGT-3′ and 5′-TGGTTTCCGGGGTATGAA-3′; (5) mouse β-actin, 5′-ATGGATGACGATATCGCTGC-3′ and 5′-AGCACTGTGTTGGCATAGAG-3′; (6) human FLRT2, 5′-GGAGAAGACGCAGCCGTCAG-3′ and 5′-TTGAAGTCCAGAACTCCAGC-3′; (7) human UNC5B, 5′-ACATTTGGAGTGGCAGCCAG-3′ and 5′-TCGCTTTGGTGGCAAAGTAA-3′; (8) human β-actin, 5′-ATGGATGATGATATCGCCGC-3′ and 5′-TACAGGTCTTTGCGGATGTC-3′.

Western blot analysis of embryonic tissues was performed as described elsewhere (Okabe et al., 2014). The primary antibodies used were goat anti-FLRT2 (R&D Systems, AF2877; 1:2000), rabbit anti-β-galactosidase (Cappel Laboratories, 08559761; 1:500) and mouse anti-tubulin (Sigma-Aldrich, T5168; 1:2000).

HUVECs, HUAECs or HDLECs (Takara-Bio) were cultured in EGM-2 medium (Takara-Bio) at 37°C in 20% O₂ following a previously described protocol (Okuno et al., 2012). Chemotaxis was assessed using Transwell polyester-membrane cell-culture inserts (3.0 µm pore, Sigma-Aldrich) according to the method described previously (Kubota et al., 2011). Both the upper and lower surfaces of the membranes were coated with 0.1% gelatin, and 600 µl EGM-2 medium with or without 5 µg ml⁻¹ recombinant FLRT2 (Abcam) and/or 5 µg ml⁻¹ recombinant human UNC5B (R&D Systems) was added to the lower wells of 24-well plates. Cells were re-suspended in EGM-2, added to the upper wells (1×10⁵ cells per well), and incubated for 12 h. Cells were then fixed by 4% PFA in PBS and stained with DAPI. The number of cells attached to the lower surfaces of the membranes were counted by confocal microscopy as described above.

Stripe assays were performed as described previously (Yamagishi et al., 2011). Briefly, 50 µg ml⁻¹ recombinant control and FLRT2 proteins (Abcam) were pre-clustered with FITC-conjugated anti-His antibodies (Abcam; ab1270; 1:20) in PBS for 30 min. For competition assays, pre-clustered FLRT2 proteins were preincubated with 50 µg ml⁻¹ recombinant UNC5B proteins (R&D Systems) for 45 min. These protein solutions were injected into matrices (90 µm width) placed on 60 mm dishes (Knöll et al., 2007). After incubation at 37°C for 30 min, the dishes were washed with PBS and the matrices were carefully removed, resulting in green fluorescent stripes. The stripes were coated with 50 µg ml⁻¹ bovine serum albumin as black stripes for 30 min at 37°C and washed three times with PBS. The stripes were further incubated with 20 µg ml⁻¹ laminin (Invitrogen) at 37°C for 1 h, and 10,000 HUVECs or HUAECs were cultured at 37°C for 24 h on the stripes. After incubation, cells were fixed with 4% PFA for 20 min and permeated with 0.3% Triton X-100 in PBS for 15 min at room temperature. DAPI was used to visualize the nuclei, and the total numbers of cells on green and black stripes were quantified. To create movies, phase-contrast pictures were obtained every 5 min from 0.5 h to 16 h after plating the cells and merged with images in the fluorescent channel using Biostation (Nikon, Japan). For RNA interference, HUVECs were washed once with OptiMEM (Thermo Fisher Scientific) and transfected with 40 nmol l⁻¹ siRNA Duplex (Qiagen) using 6 µl ml⁻¹ Lipofectamine RNAiMAX (Invitrogen) in OptiMEM according to the manufacturer's instructions. After 4 h, the transfection medium was removed and complete culture medium was added. Cells were cultured for a further 48 h before applying to the stripe assay. To knock down Unc5b, FlexiTube siRNA GS219699 for UNC5B (Qiagen) was used. As a control, a siRNA-negative control duplex oligonucleotide (5′-CUUACGCUGAGUACUUCGATT-3′, 5′-UCGAAGUACUCAGCGUAAGTT-3′; Sigma) was used.

Results are expressed as mean±s.d. The comparisons between the averages of the two groups were evaluated using a two-tailed Student's t-test. P-values of <0.05 were considered statistically significant.