Autotaxin Stabilizes Blood Vessels and Is Required for Embryonic Vasculature by Producing Lysophosphatidic Acid*

Autotaxin (ATX) is a cancer-associated motogen that has multiple biological activities in vitro through the production of bioactive small lipids, lysophosphatidic acid (LPA). ATX and LPA are abundantly present in circulating blood. However, their roles in circulation remain to be solved. To uncover the physiological role of ATX we analyzed ATX knock-out mice. In ATX-null embryos, early blood vessels appeared to form properly, but they failed to develop into mature vessels. As a result ATX-null mice are lethal around embryonic day 10.5. The phenotype is much more severe than those of LPA receptor knock-out mice reported so far. In cultured allantois explants, neither ATX nor LPA was angiogenic. However, both of them helped to maintain preformed vessels by preventing disassembly of the vessels that was not antagonized by Ki16425, an LPA receptor antagonist. In serum from heterozygous mice both lysophospholipase D activity and LPA level were about half of those from wild-type mice, showing that ATX is responsible for the bulk of LPA production in serum. The present study revealed a previously unassigned role of ATX in stabilizing vessels through novel LPA signaling pathways.

Autotaxin (ATX) 2 is a motogen-like phosphodiesterase originally isolated from conditioned medium of human melanoma cells (1). Enforced expression of ATX in Ras-transformed NIH3T3 cells greatly enhances their invasive, tumorigenic, and metastatic potentials (2). In addition, enhanced expression of ATX has been demonstrated in various malignant tumor tissues (3). Thus, ATX is implicated in tumorigenic and metastatic potentials of cancer cells. ATX is also expressed in various tissues and is present at high concentration in various biological fluids including plasma, serum, and seminal plasma (4), implying specific roles of ATX in circulation.
Recently, ATX was shown to have lysophospholipase D (lysoPLD) activity, which converts lysophosphatidylcholine to a bioactive lysophospholipid, lysophosphatidic acid (LPA) (5,6). ATX also converts sphingosylphosphorylcholine into another bioactive lysophospholipid, sphingosine 1-phosphate (S1P) in vitro (7). Because LPA and S1P are regulators of cell motility and proliferation in various cell systems, they might be the effectors of the motogenic actions of ATX. LPA and S1P have been shown to have diverse roles in many biological processes that are mediated by G protein-coupled receptors (GPCRs) specific to LPA or S1P; there are five GPCRs for LPA (LPA [1][2][3][4][5] and five for S1P (S1P [1][2][3][4][5] ) with a number of putative GPCRs (8). Thus, ATX may exert its functions through these receptors. Indeed, ATX stimulates cell motility of tumor cells through one of the LPA receptors, LPA 1 (9), and ATX positively or negatively modulates cell motility depending on S1P receptor subtypes (7,10). To uncover the physiological role of ATX and to identify the endogenous product of ATX, we investigated ATX knock-out mice. In this study we show that ATX produces LPA, but not S1P, in circulating blood and that it contributes to blood vessel stability through novel LPA signaling pathways.

MATERIALS AND METHODS
ATX Knock-out Mice-ATX knock-out mice (atx Ϫ/Ϫ ) in the genetic background 129/SvEvBrd were produced by and obtained from Lexicon Genetics (The Woodlands, TX). The ATX gene was targeted with an ATX gene-targeting vector, which was designed to replace the initiation codon and first 45 amino acids encoded by exons 1 and 2 with a lacZ-neo cassette in the vector pKOS (Fig. 1A). The ATX knock-out mice were backcrossed with C57BL/6j mice at least six times before being used. Mice were genotyped by both Southern blotting and PCR of genomic DNA. The PCR primers for detection of wild-type alleles are 5Ј-ccgaatctctccgatcactac-3Ј and 5Ј-tccaacattaaagcgataacc-3Ј, and those for detection of mutant alleles are 5Ј-gcagcgcatcgccttctatc-3Ј and 5Ј-tccaacattaaagcgataacc-3Ј. Flk-1-laxZ knock-in mice (11) were kindly donated by Dr. Janet Rossant (The Hospital for Sick Children, Toronto, Canada). All mice used in this study were bred and maintained at the Animal Care Facility in the Graduate School of Pharmaceutical Sci-ences, the University of Tokyo, under specific pathogen-free conditions in accordance with institutional guidelines.
Recombinant ATX-Two recombinant mouse ATX proteins (wild-type ATX and catalytically inactive T209A mutant ATX) with a His tag at the N terminus were expressed and purified using a baculovirus system and nickel column chromatography (HisTrap HP; GE Healthcare), respectively.
Preparation of ATX-depleted Serum-To establish antimouse ATX monoclonal antibody, the recombinant mouse ATX protein (50 g) was used to immunize rats (WKY/Izm strain) with Freund's complete adjuvant into the hind footpads. Cells from the enlarged medial iliac and inguinal lymph nodes were fused with mouse myeloma (PAI) cells. The antibodysecreting hybridoma cells were selected by screening with an enzyme-linked immunosorbent assay and immunoprecipitation. One clone (5E5 (rat IgG1)) was found to have activity to immunoprecipitate ATX in mouse serum. 5E5 reacted with mouse, human, and bovine ATX but does not react with rat ATX. The monoclonal antibody was purified from culture supernatant of hybridoma cells and was coupled to Sepharose 4B beads (2 mg/ml) (GE Healthcare). To deplete ATX from mouse serum, mouse serum (1 ml) was incubated with the 5E5-Sepahrose 4B (40 l) for 2 h at 4°C, and the resulting supernatant was used for measuring lysoPLD activity and LPA production. ATX bound to the 5E5-Sepahrose 4B was eluted with 100 mM glycine, pH 2.5. This gave rise to a single 100-kDa band on sodium dodecyl sulfate polyacrylamide gel electrophoresis, showing that 5E5 is specific to ATX.
Immunohistochemistry, Whole-mount Immunostaining, and Whole-mount LacZ Staining-Tissue sections (5 m) were dehydrated, embedded in paraffin, incubated in 3% (v/v) H 2 O 2 for 20 min, incubated overnight at 4°C with first antibodies (EPOS anti-a-SMA/HRP; DAKO), stained with diaminobenzidine according to the manufacturer's protocol, and counterstained with hematoxylin. For whole-mount embryo immunostaining, embryos were fixed in 4% paraformaldehyde, dehydrated, incubated with 5% hydrogen peroxide in methanol, rehydrated through a methanol series to phosphate-buffered saline, incubated in phosphate-buffered saline containing 4% bovine serum albumin and 0.1% Triton X-100, incubated with rat anti-PECAM monoclonal antibody (BD Biosciences), incubated with peroxidase-conjugated goat anti-rat IgG (American Qualex), and stained with diaminobenzidine as a peroxidase substrate. LacZ staining was performed as described (11). B, genotyping of atx ϩ/ϩ , atx ϩ/Ϫ , and atx Ϫ/Ϫ E8.5 embryos by PCR using genomic DNA and two sets of primers (fw and rev for detection of wild-type allele and Neo and rev for detection of mutant allele, see in panel A). C, quantitative real-time RT-PCR analysis of atx ϩ/ϩ , atx ϩ/Ϫ , and atx Ϫ/Ϫ embryos at E8.5. The level of ATX mRNA was measured using quantitative real-time RT-PCR and is expressed as a relative value to glyceraldehyde-3-phosphate dehydrogenase mRNA.
Allantois Culture-Allantoides were dissected from embryos at E8.5 and cultured in Dulbecco's modified Eagle's medium containing glutamine and antibiotics (18). To perform the vessel formation, the allantois explants were cultured for 24 h (37°C; 5% CO 2 ) in the presence or absence of the factors tested. In some cases, to examine the effect of factors on stabilization of preformed vessels the allantoides were first cultured for 24 h in the presence of amniotic fluids (isolated from E12.5 embryos) to allow formation of stable vascular networks and then cultured for an additional 18 h in the presence or absence of the factors tested. Cells were fixed with 4% paraformaldehyde and then immunostained with anti-PECAM monoclonal antibody.
Determination of LPA and S1P Concentration, lysoPLD Activity, and Western Blotting-LPA and S1P concentrations in plasma and serum were determined as described previously (12)(13)(14). lysoPLD activity was determined as described using 14:0 lysophosphatidylcholine as substrate (5). Western blotting of ATX was performed as described using ATX-specific monoclonal antibody (15).
In Situ Hybridization-Expression of ATX and LPA 1 was detected in paraffin-embedded or frozen sections by in situ hybridization using digoxigenin-labeled RNA probes for ATX and LPA 1 as previously described (16). Specimens were treated with proteinase K at 2 g/ml in phosphate-buffered saline for 5 min and then refixed in 4% paraformaldehyde/phosphate-buffered saline for 20 min, followed by acetylation. Finally, digoxigenin was immunodetected with 5-bromo-4chloro-3-indolyl phosphate (Roche Applied Science).
Quantitative RT-PCR-From embryos and yolk sacs at E8.5-E10.5, total RNA was isolated using ISOGEN (Nippongene, Toyama, Japan) and reverse transcribed using the SuperScript first-strand synthesis system for RT-PCR (Invitrogen). Primers used to detect ATX, LPA receptors, glyceraldehyde-3-phosphate dehydrogenase, and ␤-actin were described previously (17). Quantitative RT-PCR was performed as described (9) using ABI PRISM7000 sequence detection system (Applied Biosystems). Each sample was normalized to the number of ␤-actin or glyceraldehyde-3-phosphate dehydrogenase transcripts.  . ATX is responsible for bulk LPA production in plasma. A, LPA levels in plasma from atx ϩ/ϩ and atx ϩ/Ϫ mice determined by a coloriometric method. Plasma LPA levels were reduced by about one half in atx ϩ/Ϫ mice. B, lysoPLD activity of ATX-depleted mouse serum. lysoPLD activity was determined by liberation of choline from lysophosphatidylcholine (LPC) using 14:0-LPC as a substrate. C, production of LPA in ATXdepleted mouse serum during incubation at 37°C. Open circles represent ATX-depleted serum, and closed circles represent control serum. D, S1P level in plasma from atx ϩ/ϩ and atx ϩ/Ϫ mice. Plasma S1P levels were not significantly different. Statistical significance was analyzed using Student's t test. Statistically significant differences (p Ͻ 0.05) are indicated by asterisks in panels A and B.

RESULTS
Vascular Defects in ATX-null Embryos-To understand the physiological role of ATX, we investigated atx knock-out mice. The ATX gene was targeted with an ATX gene-targeting vector that was designed to replace the initiation codon and first 45 amino acids encoded by exons 1 and 2 with a lacZ-neo cassette in the vector pKOS (Fig. 1A). The generation of mutant mice was confirmed by Southern blotting (not shown) and polymerase chain reaction (PCR) analyses of tail genomic DNA (Fig.  1B). RT-PCR analysis of embryos carrying the knock-out allele confirmed that ATX mRNA was not produced from the disrupted locus (Fig. 1C). The plasma lysoPLD activity and plasma ATX protein level in atx ϩ/Ϫ mice were about half the values found in atx ϩ/ϩ mice (Fig. 2, A and B), showing that ATX is responsible for most of the lysoPLD activity in plasma. The plasma LPA concentrations in atx ϩ/Ϫ mice were about half of those in atx ϩ/ϩ mice (Fig. 3A). In addition, in ATX-depleted mouse serum, which is prepared by mixing the mouse serum with Sepharose 4B beads coupled with anti-ATX monoclonal antibody (5E5), both lysoPLD activity and LPA production during incubation at 37°C were hardly detected (Fig. 3, B and C). These results clearly indicate that ATX is a major enzyme that produces LPA in serum. By contrast, the S1P levels in plasma (Fig. 3D) and serum (not shown) from atx ϩ/ϩ mice were unchanged. Despite the reduced LPA level, heterozygous mice appeared phenotypically normal. Viability and fecundity were similar to those in wild-type mice.
Intercrossing of heterozygous animals produced no homozygous pups, indicating that ATX mutations are recessive embryonic lethal. To determine when homozygous embryos were dying, embryos were isolated at various stages of gestation. Homozygous embryos survived at least to embryonic day 10.5 (E10.5) ( Table 1). From E9.5, homozygous embryos were clearly abnormal, and at E11.5 almost all of them were dead in utero. Several defects were evident in the homozygous embryo proper. These included growth retardation, open and kinky neural tubes, pericardial effusion, reduced number of somites, incomplete turning, pallor, and fragility (Fig. 4). The hearts beat until E10.5, but contractions were weak and irregular. In addition, yolk sacs were apparently abnormal in homozygous (atx Ϫ/Ϫ ) embryos. The yolk sac of atx Ϫ/Ϫ embryos completely lacked large vitelline vessels at E9.5 and E10.5, whereas wild-type (atx ϩ/ϩ ) yolk sacs had well developed vitelline vessels (Fig. 5A). Histological analysis also showed a clear difference in morphology between atx ϩ/ϩ and atx Ϫ/Ϫ tissues. Individual vessels were clearly visible in atx ϩ/ϩ tissue, with numerous regularly spaced attachments between the visceral endoderm and mesoderm layers (Fig. 5A). Visceral endoderm and mesoderm were more widely separated in atx Ϫ/Ϫ tissue, with occasional attachments forming numerous large spaces that were sparsely populated with blood cells (Fig. 5A). Despite the vascular defects, embryonic erythrocytes were present (Fig. 5A, arrows), indicating that blood cells differentiated. To specifically visualize the endothelial vasculature in the developing yolk sac and embryo, we utilized flk-1 heterozygous mice, in which expression of lacZ is regulated by the promoter of the flk-1 gene and is restricted to endothelial cells (11). We generated atx ϩ/ϩ flk ϩ/Ϫ and atx Ϫ/Ϫ flk ϩ/Ϫ embryos with flk-1 promoter-regulated lacZ expression. To do this we first crossed atx ϩ/Ϫ mice with flk ϩ/Ϫ (flk.LacZ) mice and generated atx ϩ/Ϫ flk ϩ/Ϫ . Then atx ϩ/ϩ flk ϩ/Ϫ and atx Ϫ/Ϫ flk ϩ/Ϫ embryos were generated by crossing atx ϩ/Ϫ flk ϩ/Ϫ mice. At E8.5, atx Ϫ/Ϫ flk ϩ/Ϫ yolk sacs had slightly rough but almost comparable tubular structures to atx ϩ/ϩ flk ϩ/Ϫ yolk sacs (Fig. 5B). In contrast, at E9.5, atx Ϫ/Ϫ flk ϩ/Ϫ yolk sacs were found to  have severe vascular defects, i.e. they had no well developed fine blood vessels (Fig. 5B). Consistent with this observation, both normal and abnormal vessels were visible in atx Ϫ/Ϫ flk ϩ/Ϫ yolk sacs at E9.0 (Fig. 5B). Embryos at E8.5 atx Ϫ/Ϫ had vascular defects in the head (arrows) and cardiac (sinus venosus) regions but not in the dorsal aortae (Fig. 5B). Staining with antibodies against PECAM, another endothelial cell-specific marker, confirmed the vascular defects of the yolk sac in homozygous embryos (not shown). Incomplete vascular remodeling was also observed in the embryo proper. Whole-mount anti-PECAM staining revealed well developed thin and fine tubule-like structures in atx ϩ/ϩ embryos (Fig.  5C). In sharp contrast, the structures were thick and irregularshaped in atx Ϫ/Ϫ embryos (Fig. 5C). At E9.5, the atx ϩ/ϩ placenta had both embryonic and maternal blood vessels (distinguishable by the size of their erythrocytes), whereas the atx Ϫ/Ϫ placenta had only maternal blood vessels (Fig. 5D). At E9.5, chorioallantoic attachment and fusion were observed both in atx ϩ/ϩ and atx Ϫ/Ϫ embryos (Fig. 5E). However, it was evident that atx Ϫ/Ϫ allantois lacked the numerous vessel lumen visible in the atx ϩ/ϩ allantois (Fig. 5E). Immunostaining of the yolk sac and embryo proper at E9.5 with anti-␣-smooth muscle actin, a specific marker of vascular smooth muscle cells, showed that smooth muscle cells were present and surrounded the endothelial cells in both the yolk sac ( Fig. 5F) and embryo proper (not shown). This indicates that mural cell investment is not the cause of vascular defects in atx Ϫ/Ϫ embryos. These analyses suggest that initial blood vessel formation occurs properly in atx Ϫ/Ϫ embryos but the newly formed blood vessels fail to develop into mature vessels in the absence of ATX. RT-PCR experiments detected modest expression of ATX mRNA in both the yolk sac and embryo proper until E10.5 (Fig.  6A). In situ hybridization experiments detected weak expression in endodermal cells that surround the yolk sac at E8.5 (Fig.  6B). By contrast, significant ATX protein expression was observed in amniotic fluids in all embryonic stages tested, and  its level was nearly equivalent to the plasma level as judged by both Western blotting and lysoPLD activity (Fig. 2, A and C). As was observed in plasma, the lysoPLD activity and ATX protein in amniotic fluids from atx ϩ/Ϫ embryos at E11.5, 12.5, and 13.5 were almost half the value found in atx ϩ/ϩ embryos (Fig. 2, A  and C). Western blot analysis also showed that ATX expression was not detectable in either yolk sac or embryo proper at E9.5, although ATX protein was highly expressed in amniotic fluids (Fig. 2, D and E). These analyses indicate that most of the ATX proteins produced in embryos are released and concentrated in amniotic fluids.
ATX and LPA Prevent Disassembly of Vessels in Allatois Culture System-To determine the role of ATX and to assess its enzymatic activity in vascular remodeling, we utilized the allantois explant culture system (18). As reported by Argraves et al. (18), when E8.5 allantois explants were cultured in the presence of 10% fetal calf serum, a highly branched network of PECAMpositive vessels was visible after 24 h of culture (Fig. 7A). The amniotic fluids isolated from E12.5 wild-type embryos were found to have a similar but an even more pronounced effect on vessel formation than fetal calf serum (Fig. 7A). Allantois explants from atx Ϫ/Ϫ embryos also showed a highly branched network of PECAM-positive vessels when they were cultured in the presence of amniotic fluids isolated from E12.5 wild-type embryos (Fig. 7A) or fetal calf serum (not shown). Adding LPA or recombinant ATX to this system had little effect on vessel formation (Fig. 7A), showing that ATX and LPA themselves are not angiogenic. In the presence of amniotic fluids, the vessels that formed within the first 24 h of culture kept their tubular structure during the next 18 h of culture (Fig. 7B). However, in the absence of amniotic fluid, the preformed nascent vessels were unstable and disassembled (Fig. 7B). The recombinant ATX, but not catalytically inactive ATX, prevented the disassembly (Fig. 7B). LPA also prevented the disassembly of preformed vessels, whereas S1P had no effect (Fig. 7B). These results indicate that ATX in the amniotic fluids has a role in stabilizing vessels.
To know the molecular mechanisms of how LPA stabilizes blood vessels, we examined the expression of LPA receptors. RT-PCR and in situ hybridization experiments showed that among the four LPA receptors LPA 1 is predominantly expressed in the yolk sac and embryo proper at E8.5, E9.5, and E10.5 (Fig. 6C), where it is expressed by cells in the endodermal layer other than endothelial cells (Fig. 6D). Expression levels of  (19). Ki16425 did not show antagonistic activity toward LPA 4 (19). Adding Ki16425 at 5 M in the allantois culture system did not affect the vessel-stabilizing effects of ATX and LPA (Fig. 8). Thus, it is unlikely that LPA 1 , LPA 2 , and LPA 3 are involved in the vessel-stabilizing effect of ATX and LPA.

DISCUSSION
ATX was originally identified as a cell motility-stimulating factor for cancer cells (1). Later, ATX was found to have enzymatic activity to produce a bioactive phospholipid, LPA (5,6,20). LPA has been defined as a growth factor-like lipid with numerous biological activities, including stimulation of cell motility (3). In this study to elucidate the physiological role of ATX we analyzed ATX knock-out mice. To our surprise ATX knock-out mice were embryonic lethal, showing that ATX is indispensable for embryonic development. Blood vessel formation was severely affected in ATX-null embryos, suggesting novel roles of ATX and LPA in the formation of blood vessels in the embryonic stage. In addition, the present study confirmed that ATX is responsible for bulk LPA production in blood, but not for S1P production. The phenotype of ATX-null embryos was quite different from those of LPA receptors reported so far (21)(22)(23). These results raise a possibility that ATX and LPA contribute to embryonic blood vessel formation through as yet unidentified LPA receptors.
Nam et al. (24) reported that injection of Matrigel mixed with purified ATX into athymic nude mice resulted in new blood vessel formation within the plug and that ATX stimulated human umbilical vein endothelial cells grown on Matrigel to form tubules. From these results they suggested that ATX was an angiogenic factor. Our present data suggest that ATX itself is not angiogenic but it stabilizes preformed vessels through an unknown mechanism. Because Matrigel contains several factors that promote vessel formation, ATX may contribute to the vessel formation possibly in cooperation with such factors.
The formation of vasculature by vasculogenesis and angiogenesis is essential not only for embryonic development but also for the unrestrained growth of tumors (25). ATX stimulates both cell proliferation and cell motility of cancer cells through LPA production (5). In addition, overexpression of ATX is frequently associated with malignant tumors such as small cell lung cancer (26), renal cell cancer (27), hepatocellular carcinoma (28,29), breast cancer (30,31), Hodgkin lymphoma (32), thyroid cancer (33), and glioblastoma (17). Thus, ATX has been implicated in the progression of malignant tumors. In this study we showed that ATX has an additional role in blood vessel formation, possibly by stabilizing preformed blood vessels. The present study raises the possibility that ATX, in addition to stimulating proliferation and motility of tumor cells (5), contributes to the progression of tumors by stabilizing blood vessels in the vicinity of tumors.
ATX is capable of producing S1P because ATX also catalyzes a reaction to convert sphingosylphosphocholine to S1P (7). We showed in this study that ATX is a major producing enzyme for LPA, but not for S1P, in blood. It is accepted in recent reports that S1P is produced intracellularly from sphingosine through phosphorylation reaction mediated by two isozymes of sphingosine kinase (sphK) (sphingosine kinase 1 and sphingosine kinase 2) (34,35). Interestingly, vascular defects were also observed in sphingosine kinase-null and S1P receptor-null embryos. Both sphk1 Ϫ/Ϫ ,sphk2 Ϫ/Ϫ double mutant embryos (36) and s1p 1 Ϫ/Ϫ ,s1p 2 Ϫ/Ϫ ,s1p 3 Ϫ/Ϫ triple mutant embryos (37) died around E12.5, showing that the phenotypes are milder than that of atx Ϫ/Ϫ embryos. Thus, despite their similar structures, LPA and S1P have distinct synthetic pathways, targets, and functions.
Among the four LPA receptors reported so far, LPA 1 is predominantly expressed in embryos around E9.5, whereas LPA 2 , LPA 3 , and LPA 4 are weakly expressed (Fig. 6, C and D). We showed that Ki16425, an antagonist for LPA 1 , LPA 2 , and LPA 3 , had little effect on vessel-stabilizing activity of ATX and LPA (Fig. 8). Thus, it is unlikely that LPA 1 , LPA 2 , and LPA 3 are involved in this process. Consistent with this, phenotypes of LPA receptor knock-out mice are quite different from that of ATX knock-out mice. LPA 1 , LPA 2 , and LPA 3 single and LPA 1 and LPA 2 double knock-out mice are reported to be viable (21)(22)(23). LPA also exerts its role through other targets such as GPR23/LPA 4 (38). Very recently a novel G protein-coupled receptor, GPR92/LPA 5 , was identified as a fifth cellular receptor for LPA (39,40), which showed the highest expression in gut. LPA 4 and LPA 5 knock-out mice have not been reported so far. It is possible that ATX exerts its role in blood vessel stabilization through these newly identified LPA receptors via signaling pathways independent of classical LPA receptors (LPA 1 , LPA 2 , and LPA 3 ).