Inhibition of Autotaxin by Lysophosphatidic Acid and Sphingosine 1-Phosphate*

Autotaxin (ATX) or nucleotide pyrophosphatase/phosphodiesterase 2 (NPP2) is an NPP family member that promotes tumor cell motility, experimental metastasis, and angiogenesis. ATX primarily functions as a lysophospholipase D, generating the lipid mediator lysophosphatidic acid (LPA) from lysophosphatidylcholine. ATX uses a single catalytic site for the hydrolysis of both lipid and non-lipid phosphodiesters, but its regulation is not well understood. Using a new fluorescence resonance energy transfer-based phosphodiesterase sensor that reports ATX activity with high sensitivity, we show here that ATX is potently and specifically inhibited by LPA and sphingosine 1-phosphate (S1P) in a mixed-type manner (Ki ∼ 10–7 m). The homologous ecto-phosphodiesterase NPP1, which lacks lysophospholipase D activity, is insensitive to LPA and S1P. Our results suggest that, by repressing ATX activity, LPA can regulate its own biosynthesis in the extracellular environment, and they reveal a novel role for S1P as an inhibitor of ATX, in addition to its well established role as a receptor ligand.

The mode of action of ATX/NPP2 has long been elusive because the biological effects of ATX could not be explained by nucleotide hydrolysis. The surprise came when it was discovered that ATX is identical to plasma lysophospholipase D (lyso-PLD) and acts by hydrolyzing lysophospatidylcholine (LPC) into lysophosphatidic acid (LPA) (11,12), a lipid mediator that signals cell proliferation, migration, and survival via specific G protein-coupled receptors (13). It has now become clear that de novo production of LPA can fully account for the biological effects of ATX observed in cell culture. The lysophospholipid substrate range of ATX has recently been broadened by showing that the enzyme can also hydrolyze sphingosylphosphorylcholine (SPC) to yield sphingosine 1-phosphate (S1P) (14), a bioactive lipid with signaling properties very similar to those of LPA while acting on distinct receptors (15)(16)(17). The physiological significance of the SPC-to-S1P conversion is debatable, however, because the reported K m of ATX for SPC (14) is 3 orders of magnitude higher than the normal SPC levels in plasma and serum (18). Rather than through SPC hydrolysis, S1P is thought to originate largely from the phosphorylation of sphingosine by sphingosine kinases (19).
Mutational analysis has revealed that the lyso-PLD and nucleotide phosphodiesterase activities of ATX originate from the same catalytic site (20,21). Unexpectedly, the other two members of the NPP family (NPP1 and NPP3) lack intrinsic lyso-PLD activity despite their close homology to ATX (21). Given the differences in substrate specificity, it is not surprising that the NPPs appear to have largely unrelated physiological functions. The founding member, NPP1, hydrolyzes ATP into pyrophosphate, an inhibitor of calcification, and thereby regulates bone mineralization, whereas the third member, NPP3, promotes differentiation and invasion of glial cells by an unknown mechanism (3).
An unresolved question concerns the regulation of ATX activity. One puzzling observation is that LPA levels in plasma or freshly isolated blood are very low (22)(23)(24), yet plasma ATX is constitutively active and its substrate LPC abundantly present (Ͼ 100 M) (25). This suggests that ATX is negatively regulated in vivo, but physiological or pharmacological inhibitors of ATX have not been identified to date. In the present study we sought to examine how ATX activity is regulated in the extracellular milieu. To this end, we used a newly invented fluorescence resonance energy transfer (FRET)-based phosphodiesterase sensor (termed CPF4; see Ref. 26) that, as we show here, reports ATX activity in conditioned media with superior sensitivity. Using this assay system, we demonstrate that ATX, * This work was supported by the Dutch Cancer Society. The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact. □ S The on-line version of this article (available at http://www.jbc.org) contains supplemental Fig. 1.
cDNA Cloning-RNA extracted from human diploid foreskin fibroblasts was used to generate cDNA using Invitrogen reverse transcriptase. ATX cDNA was isolated using ATX-specific primers. The stop codon was removed and KpnI and NotI restriction sites were introduced at the 5Ј and 3Ј sites, respectively. After digestion, ATX was ligated in pcDNA3 vector with 3Ј Myc tag (ATX-Myc), 5Ј HA tag (HA-ATX), or 5Ј HA tag and 3Ј Myc tag (HA-ATX-Myc), and in pEGFP-N3 (ATX-GFP). Sequencing showed that the ATX inserts were identical to human ATX (GenBank TM accession number BC034961). The catalytically inactive mutant ATX(T210A) (27) was generated using the Stratagene sitedirected mutagenesis kit.
Transfection and Western Blotting-HEK293T cells were transfected with ATX constructs using the calcium phosphate method. At 24 h after transfection, cells were exposed to serum-free Dulbecco's modified Eagle's medium for 30 h. Conditioned medium was centrifuged (5000 rpm for 30 min) to remove cell debris. Medium was used without further purification and analyzed for the presence of ATX by Western blotting and activity assays. There was very little variation in ATX activity between different batches of conditioned medium. For Western blot analysis, nitrocellulose filters were blocked using 5% nonfat powdered milk and probed with primary antibodies (9E10 anti-Myc and 3F10 anti-HA; Roche Applied Science) and horseradish peroxidase-conjugated secondary antibodies (Dako, Glostrup, Denmark). Proteins were visualized using the ECL detection system (Amersham Biosciences).
Recombinant ATX-The Bac-to-Bac baculovirus expression system (Invitrogen) was used for ATX production. ATX cDNA fused to the honeybee melittin signal sequence at the 5Ј end was cloned into the pFastBac I vector (Invitrogen). The resulting plasmid was then used for generating recombinant baculovirus to infect Sf9 insect cells, which were grown in SF-900 II medium (Invitrogen); the multiplicity of infection was ϳ5. After 60 h of infection, medium containing secreted ATX was collected by low speed centrifugation and applied onto a Q-Sepharose column (Amersham Biosciences). The flow-through was applied onto a hydroxyapatite column (Bio-Rad), and the bound proteins were eluted with a linear potassium phosphate gradient. The ATX-containing fractions were applied onto an isopropyl column (Amersham Biosciences), and bound proteins were eluted with a decreasing linear gradient of ammonium sulfate. Purity of ATX-containing fractions was Ͼ 95% as shown by SDS-PAGE and Coomassie Blue staining. The approximate yield was about 3 mg of pure protein from 10 liters of culture supernatant.
Recombinant NPP1-A soluble secreted form of human NPP1 was made by in-frame ligation of its extracellular domain to the BamHI site FIG. 1. ATX processing and secretion. A, human full-length ATX was expressed in HEK293T cells as a fusion protein with an N-terminal HA tag and/or a C-terminal Myc tag, as indicated. ATX expression in cell lysates and conditioned medium was analyzed by Western blotting using antibodies against the HA and Myc epitopes. Equal amounts of cell lysate and medium were applied in each experiment. HA-ATX is undetectable in the medium, consistent with N-terminal cleavage prior to secretion. IB, immunoblot; TM, transmembrane domain; SO, somatomedin-like domain; PDE, phosphodiesterase domain; NUC, nucleaselike domain. threonine residue (T210) is essential for catalytic activity. B, subcellular localization of ATX and the LPA 1 receptor (both Cterminally fused to GFP) in HEK293T cells. Unlike LPA 1 , ATX is not detected at the cell surface but, instead, localizes to intracellular vesicles and reticular structures. Cells transfected with ATX-GFP or LPA 1 -GFP (28) were fixed with ice-cold methanol at 24 h after transfection.
FIG. 2. lyso-PLD activity of ATX. A, lyso-PLD activity. Hydrolysis of 1-[1-14 C]palmitoyl-LPC into LPA by ATX-Myc in conditioned medium was analyzed by thin layer chromatography. The concentration of unlabeled LPC was 1 M. The reaction was terminated after 2 h. LPC and LPA standards were run separately (not shown). LPA, S1P, and phosphatidic acid (PA) were added at 1 M. B, graphical representation of data from two independent thin layer chromatography assays. C, time course of LPC hydrolysis into LPA and inhibition by LPA, added at 1 M.
of the cleavable signal sequence from influenza hemagglutinin, and cloned into a mammalian expression vector driven by the SR-␣ promoter. After linearization, the DNA was transfected into Chinese hamster ovary cells, and a highly expressing stable transfectant was isolated by single cell cloning. The soluble enzyme was purified from culture supernatants by sequential chromatography consisting of large scale anion exchange chromatography, Blue-Sepharose, AMP-Sepharose, and final polishing on a Uno-Q column (Bio-Rad). Purity was confirmed by SDS-PAGE and silver staining.
lyso-PLD Assay-To measure lyso-PLD activity, radiolabeled LPC (1-[1-14 C]palmitoyl, Amersham Biosciences) and unlabeled LPC (1 M) were dried under nitrogen, and the mixture was reconstituted in Trisbuffered saline (140 mM NaCl, 5 mM KCl, 1 mM CaCl 2 , 1 mM MgCl 2 , adn 50 mM Tris, pH 8.0) and sonicated, and BSA (2 mg/ml) was added. The reaction was started by the addition of ATX-containing conditioned medium. Lipids were extracted with butan-1-ol. All of the solutions were made 0.02 M in acetic acid and extracted with 0.5 volume of butan-1-ol as described previously (22). In brief, after thorough mixing and centrifugation, the butan-1-ol phase was removed, and the water phase was extracted once again. Butanol fractions were washed with 1 volume of butan-1-ol-saturated water and dried under nitrogen. Phospholipids were separated by thin layer chromatography on silica gel-60 plates in chloroform/methanol/acetic acid/water (50:30:8:4). Lipids were detected by autoradiography.
Phosphodiesterase Assays-Phosphodiesterase activity toward pNP-TMP and bis-pNPP was measured by light absorbance. 40 l of HEK293T-conditioned medium was added to 160 l of Tris-buffered saline (140 mM NaCl, 5 mM KCl, 1 mM CaCl 2 , 1 mM MgCl 2 , and 50 mM Tris, pH 8.0) containing pNP-TMP or bis-pNPP in 96-well plates. After incubation for 3 h at 37°C, the amount of liberated para-nitrophenolate (pNP) was determined by reading the absorbance at 405 nm in a Victor Wallac plate reader.
CPF4 FRET Assay-CPF4 was synthesized as described previously (26) and maintained as a 10-mM stock solution in Me 2 SO. Recombinant ATX in Tris-buffered saline (140 mM NaCl, 5 mM KCl, 1 mM CaCl 2 , 1 mM MgCl 2 , 50 mM Tris, pH 8.0) or ATX-Myc-containing conditioned Dulbecco's modified Eagle's medium (buffered with 50 mM Tris, pH 8.0) was incubated with or without the indicated reagents, and CPF4 was added at a concentration of 2 M unless indicated otherwise. CPF4 fluorescence was monitored (at 37°C) in a BMG Fluorstar 96-well plate reader (excitation at 355 nm, emission at 460 and 520 nm). Curve fitting was carried out using GraphPad software.
Fluorescence Microscopy-Cells transfected with ATX-GFP or LPA 1 -GFP (28) were fixed with ice-cold methanol at 24 h after transfection. ATX and LPA 1 were detected by anti-GFP antibody using a Leica confocal microscope.

RESULTS AND DISCUSSION
ATX Processing in HEK293 Cells-ATX is synthesized as a type II transmembrane glycoprotein of ϳ125 kDa, consisting of a very short N-terminal region, a single transmembrane domain, and a large catalytic ectodomain. ATX undergoes membrane-proximal cleavage to yield a soluble enzyme (29), yet little is known about ATX biosynthesis and proteolytic processing. In particular, it remains unclear whether the transmembrane form of ATX is expressed on the cell surface.
As a starting point for the present studies, we expressed full-length ATX in HEK293T cells using three differently HAand Myc-tagged constructs and analyzed ATX expression in cell lysates as well as in the culture medium. As shown in Fig.  1A, full-length ATX is detected only in cell lysates, whereas ATX in the medium lacks the N-terminal HA tag, consistent with secreted ATX being produced by cleavage of the N terminus. The Golgi-disturbing agents brefeldin A and monensin blocked ATX secretion, indicating involvement of the classical export route (supplemental Fig. 1A). When analyzed by confocal microscopy, ATX is detected in intracellular vesicles but not at the plasma membrane; in contrast, LPA 1 receptors localize primarily to the cell surface under the same experimental conditions (Fig. 1B). From these results we conclude that ATX follows the classical secretory pathway and that proteolytic removal of its N terminus, including the transmembrane region, occurs intracellularly rather than at the plasma membrane.
ATX Activity toward Lipid and Non-lipid Substrates-Having established that ATX is not detectable as a plasma membrane-anchored ecto-enzyme, we set out to examine the catalytic activity of soluble ATX. To this end, we used conditioned medium from ATX-Myc-transfected HEK293T cells without further purification or, in some experiments, ATX purified from Sf9 cell supernatant (supplemental Fig. 1B). The catalytically inactive mutant ATX(T210A) served as a negative control (27).
The lyso-PLD activity of ATX was measured by the conversion of LPC(16:0) to LPA using thin layer chromatograpy ( Fig.  2A). LPC hydrolysis by ATX proceeded at a constant rate for at least 90 min (Fig. 2C, left panel). The K m for LPC was estimated at ϳ150 M, in agreement with previously reported values (100 -260 M) (11,12,14) and in the range of normal LPC levels in plasma (25). When screening multiple lipids as potential modulators of lyso-PLD activity, we observed that ATX-catalyzed LPC hydrolysis was significantly inhibited by LPA and S1P (1 M, complexed to albumin). Other phospholipids tested, including phosphatidic acid, did not show such an effect ( Fig. 2 and results not shown). This suggests that ATX is subject to product inhibition by LPA. Detailed analysis of lyso-PLD inhibition is obscured, however, by limitations with the standard LPC hydrolysis assay, which generates concentrations of LPA in excess of its inhibition constant (see below). As a result, formation of LPA during the course of the assay causes the apparent inhibition of ATX by added LPA to be less pronounced than the true inhibition.
To avoid the complications of lipid product inhibition and the limitations of end-point LPC hydrolysis assays, we took advantage of the fact that ATX uses a common reaction mechanism for the hydrolysis of lipid and non-lipid substrates (20, 21). As non-lipid substrates, we tested pNP-TMP, a standard NPP substrate, and the symmetric phosphodiester bis-pNPP, a sub- strate for type I phosphodiesterases and certain bacterial PLDs (30 -32). The liberated product, pNP, was quantified colorimetrically. As shown in Fig. 3A, hydrolysis of pNP-TMP followed Michaelis-Menten kinetics with a K m of 0.9 mM, somewhat lower than the value reported by others (12). ATX was also capable of hydrolyzing bis-pNPP (Fig. 3C), in keeping with a previous report (21). bis-pNPP hydrolysis rates were halfmaximal at ϳ0.5 mM but did not obey Michaelis-Menten kinetics, apparently because of substrate inhibition at higher concentrations (Ͼ1 mM; results not shown).
Similar to what is observed with LPC as a substrate, the hydrolysis of pNP-TMP and bis-pNPP by ATX was significantly inhibited by LPA and S1P (1 M) (Fig. 3, B and C). It thus appears that, importantly, LPA and S1P exert their inhibitory effect regardless of the nature of the substrate.
Effects of BSA-Using pNP-TMP as a substrate, we observed that ATX activity was strongly inhibited by normal serum albumin (BSA) at concentrations of Ͼ50 g/ml (0.75 M; ϳ0.1% of the concentration in fetal calf serum). In marked contrast, delipidated BSA (Fig. 3B, faf-BSA) had no inhibitory effect, even when added at high concentrations (5 mg/ml; 75 M). The observation that albumin is the major binding protein for LPA and S1P strongly suggests that ATX inhibition by "normal" BSA is largely due to these lysophospholipids. When LPA was complexed to delipidated albumin (at a 1:1 molar ratio), we found that its ability to inhibit ATX was not affected. As can be inferred from Fig. 3D, the IC 50 value for the LPA⅐BSA (1:1) complex to inhibit ATX was close to 0.1 M, using a substrate concentration of 1 mM (i.e. equal to the K m ).
CPF4, a FRET-based Sensor of ATX Activity-Because the pNP-TMP colorimetric assay showed only moderate sensitivity (detection limit, ϳ0.15 g of ATX/ml), we explored a newly developed FRET-based phosphodiesterase sensor termed CPF4 (26). CPF4 is a bis-pNPP-derived probe, in which both phenyl moieties are linked to coumarin and fluorescein, respectively, resulting in FRET with high efficiency (Fig. 4A). Cleavage of the phosphodiester group by a nonspecific phosphodiesterase from snake venom causes loss of FRET, providing a convenient ratiometric readout of enzyme activity (26). Notably, the CPF4 fluorescence ratio is insensitive to pH in the physiological range (pH 7.0 -8.0) (26). A major advantage of fluorescencebased sensors like CPF4 is their high sensitivity, allowing detection of very low concentrations of enzyme.
We examined whether CPF4 serves as a substrate for ATX. As shown in Fig. 4B, ATX-Myc causes a prominent loss of FRET (increased donor fluorescence and a concomitant decrease in acceptor fluorescence), indicative of substrate hydrolysis, which can be monitored in real time (Fig. 4C). No CFP4 signal change was detected with the inactive mutant ATX(T210A). A plot of the initial hydrolysis rates versus CPF4 concentration revealed Michaelis-Menten kinetics with an average K m as low as ϳ4 M (range, 2.5-6 M) for ATX-Myc in HEK293 cell-conditioned medium (Fig. 4D). About the same K m value was found with highly purified ATX from Sf9 cell supernatant (see below and Fig. 6), suggesting that conditioned medium does not contain significant amounts of competitive inhibitors of ATX (which would increase the apparent K m ).
It is noteworthy that the apparent affinity of ATX for CPF4 is 2-3 orders of magnitude higher than that for lysophospholipids and nucleotides (K m ϭ 0.1-1.0 mM; one notable exception is diadenosine triphosphate with an apparent K m of 8 M (33)). Also with bis-pNPP, a "non-Michaelis-Menten" substrate, halfmaximal hydrolysis rates are observed at concentrations as high as ϳ0.5 mM. Thus attachment of the coumarin-fluorescein tandem to bis-pNPP, which increases hydrophobicity and disturbs structural symmetry, converts bis-pNPP into a highaffinity substrate for ATX, suggesting that the coumarin-fluorescein tandem fits into a hydrophobic pocket involved in substrate binding. Although the overall catalytic efficiency (V max /K m ) of CPF4 hydrolysis was about 3-fold lower than that observed with the other substrates, the FRET-based assay showed superior sensitivity; even at a 5000-fold dilution of HEK293T cell-conditioned medium, ATX activity was still detectable. The estimated detection limit of the CFP4 assay was 3 ng of ATX/ml, exceeding that of the standard colorimetric assay by at least 50-fold. This makes CPF4 the preferred probe for measuring ATX activity in conditioned media and body fluids, although it remains to be seen to what extent the presence of binding proteins and/or nonspecific phosphodiesterases may limit the usefulness of CPF4 in assessing ATX activity in complex biological fluids.
We also examined whether CPF4 is a substrate for nonmammalian secreted lyso-PLDs, notably the broad specificity PLD from S. chromofuscus and the sphingomyelin-and LPC-specific PLDs from C. pseudotuberculosis and Loxosceles laeta (28). When assayed under the conditions used for ATX, none of these exogenous PLDs (10 nM) was able to cleave CPF4 (results not shown). It thus appears that among the known secreted PLDs, only ATX can hydrolyze CPF4.
Inhibition of ATX by LPA and S1P-Because of its superior sensitivity and convenient readout, the CPF4 assay was used in our further analysis of ATX inhibition. We determined the concentration dependence of LPA and S1P for inhibiting ATX activity using a substrate concentration close to the K m . Under those conditions, the IC 50 value for 1-oleoyl-LPA is ϳ0.1 M (Fig. 5A), very similar to the value found with pNP-TMP or LPC as substrate ( Fig. 3D and results not shown). S1P inhibited ATX activity with the same dose dependence as observed for LPA (Fig. 5A). Other natural lipids, including dioleoylphosphatidic acid, 1-oleoyl-glycerol, sphingosine (10 mM), glycerol 3-phosphate (1 mM), and free fatty acids, did not affect ATX activity. Although AMP has been reported to inhibit the NPP reaction (1, 2), we did not observe any effect of either AMP or ATP (1 mM) on ATX activity (results not shown).
Importantly, the inhibitory effects of LPA and S1P were specific for ATX in that the activity of the structurally related ecto-phosphodiesterase NPP1 was insensitive to either lipid (Fig. 5A). Inhibition of ATX by LPA was dependent on the length of the acyl chain; maximal inhibition was induced by 1-oleoyl-LPA, 1-palmitoyl-LPA, and 1-myristoyl-LPA, whereas short-chain LPA(6:0) had no measurable effect (Fig. 5B). The apparent requirement of a long acyl chain suggests that LPA and S1P directly interact with a hydrophobic pocket on ATX. The inhibitory effects of LPA and S1P were nonadditive (Fig.  5B), suggesting that both lipids act on the same regulatory site. We conclude that, in marked contrast to the G protein-coupled receptors (17), ATX does not discriminate between LPA and S1P as ligands.
We next examined the mechanism of inhibition by LPA and S1P. Substrate titration studies revealed that LPA and S1P are mixed-type inhibitors, producing a reduction in V max and an increase in K m (Fig. 6). In other words, inhibition of ATX by LPA/S1P has both a noncompetitive and a competitive compo-FIG. 6. Kinetic analysis of ATX inhibition by LPA and S1P. Inhibition of ATX-Myc by 1-oleoyl-LPA (A) and S1P (B) added at the indicated concentrations. Inhibition by LPA and S1P produces a decrease in V max and an increase in K m , indicative of "mixed-type" inhibition. Similar results were obtained with purified recombinant ATX (not shown). nent, resulting from a combination of a decreased turnover number and decreased affinity of the active site for its substrates. Double-reciprocal plot analysis (Fig. 6) yielded an inhibition constant (K i ) of 110 nM for LPA and 50 nM for S1P, values well within the biologically active range of LPA and S1P. We note that these K i values are ϳ1000-fold lower than the reported K m for the LPC substrate (100 -200 M) (11,12,14), indicating that ATX binds LPA and S1P much more strongly than it binds its physiological substrate(s).
Conclusion -We have employed a novel and highly sensitive FRET-based phosphodiesterase sensor to show that the catalytic activity of ATX, secreted by the classical export pathway, is potently and specifically inhibited by LPA and S1P. Kinetic analysis revealed that LPA and S1P act as mixed-type inhibitors with an inhibition constant (K i ) of ϳ0.1 M. The inhibition of ATX by LPA and S1P is specific in that (i) short-chain LPA(6:0) and other lipids have no effect, and (ii) the closely related ecto-phosphodiesterase NPP1, which lacks lyso-PLD activity, is insensitive to LPA and S1P. Because ATX primarily functions as an LPA-generating lyso-PLD, our findings imply that LPA is capable of controlling its own biosynthesis in the cellular environment. Product inhibition of ATX will prevent excessive accumulation of bioactive LPA in the extracellular milieu and may explain why steady-state plasma levels of LPA are very low despite the abundance of its precursor LPC. That serum LPA levels rapidly increase following platelet activation (24) suggests that plasma ATX may also be positively regulated by as yet unknown factors, an important issue that needs to be further explored.
Our findings point to a novel role for S1P as an inhibitor of ATX, in addition to its well documented role as a receptor agonist. The important implication is that changes in extracellular S1P levels may greatly influence LPA production and signaling. Because the S1P concentrations in blood are significant (16,34), close to the inhibition constant for ATX reported here (ϳ 0.1 M), ATX activity in plasma could be permanently suppressed by S1P under basal conditions. Our results further imply that ATX has a unique binding site for LPA and S1P, not present in NPP1. Interestingly, a photo-affinity labeling study has shown that, in addition to its catalytic site, ATX contains an isoform-specific sequence that can be cross-linked to ATP (35). Mutagenesis experiments, in combination with structural studies, should reveal whether this second nucleotide-binding region is in fact an allosteric site for LPA and S1P. Identification and characterization of the LPA/S1P binding site(s) is a major challenge for future studies and will likely facilitate the development of selective ATX inhibitors that could be of value in anticancer therapy.