The Polybasic Insertion in Autotaxin α Confers Specific Binding to Heparin and Cell Surface Heparan Sulfate Proteoglycans*

Background: The autotaxin α splice variant (ATXα) contains a unique polybasic insertion of unknown function. Results: ATXα binds strongly to heparin and cell-associated heparan sulfate. Conclusion: The ATXα insertion confers specific binding to heparan sulfate proteoglycans thereby targeting LPA production to the plasma membrane. Significance: ATX isoforms use distinct mechanisms to ensure spatially restricted LPA production and signaling. Autotaxin (ATX) is a secreted lysophospholipase D that generates the lipid mediator lysophosphatidic acid (LPA), playing a key role in diverse physiological and pathological processes. ATX exists in distinct splice variants, but isoform-specific functions remain elusive. Here we characterize the ATXα isoform, which differs from the canonical form (ATXβ) in having a 52-residue polybasic insertion of unknown function in the catalytic domain. We find that the ATXα insertion is susceptible to cleavage by extracellular furin-like endoproteases, but cleaved ATXα remains structurally and functionally intact due to strong interactions within the catalytic domain. Through ELISA and surface plasmon resonance assays, we show that ATXα binds specifically to heparin with high affinity (Kd ∼10−8 m), whereas ATXβ does not; furthermore, heparin moderately enhanced the lysophospholipase D activity of ATXα. We further show that ATXα, but not ATXβ, binds abundantly to SKOV3 carcinoma cells. ATXα binding was abolished after treating the cells with heparinase III, but not after chondroitinase treatment. Thus, the ATXα insertion constitutes a cleavable heparin-binding domain that mediates interaction with heparan sulfate proteoglycans, thereby targeting LPA production to the plasma membrane.

Full-length ATX is synthesized as a pre-proenzyme. Upon removal of its N-terminal signal peptide followed by N-glycosylation, ATX is further processed by proprotein convertases and secreted via the classical secretory pathway as an active glycoprotein of 110 -120 kDa (16,17). Mature ATX consists of a central phosphodiesterase (PDE) domain flanked by two N-terminal somatomedin B-like (SMB) domains and a C-terminal nuclease-like (NUC) domain. Structural studies have revealed that the PDE domain contains a deep lipid-binding pocket at the active site and a nearby open tunnel that might function as a product release channel (18,19). The NUC domain serves to maintain the rigidity of the PDE domain, whereas the N-terminal SMB domains mediate binding of ATX to activated integrins (18,20). Integrin binding localizes ATX to the plasma membrane, thus providing a mechanism to generate LPA in the vicinity of its cognate receptors (20,21). Whether ATX uses additional or alternative mechanisms for spatially and temporally restricted LPA production is unknown.
The intricacy of the ATX-LPA receptor system is further enhanced by alternative splicing of ATX mRNA, giving rise to distinct isoforms (3,22,23). However, isoform-specific functions of ATX are still largely uncharacterized. The canonical isoform, termed ATX␤, was originally cloned from teratocarcinoma cells (3,24) and later found to be identical to plasma lysoPLD, accounting for LPA production in the circulation (1,2). Virtually all of our current understanding of ATX is derived from studies on ATX␤. A longer isoform, ATX␣, was originally isolated as an "autocrine motility factor" from melanoma cells (25) and is characterized by a unique 52-residue polybasic insertion of unknown function in the catalytic domain (3,26). Conceivably, the ATX␣ insertion could confer different catalytic activity, cellular localization, processing, or binding partner preference. On the other hand, it was reported that the insertion renders ATX␣ intrinsically unstable and prone to proteolyic degradation, which would be a mechanism to terminate ATX␣ activity rapidly in vivo (22,23).
In the present study, we explore the unique properties of ATX␣ compared with those of ATX␤, guided by the polybasic nature of the ATX␣ insertion and by the crystal structure of ATX␤. We show that the ATX␣ insertion constitutes a heparin-binding domain that mediates specific interaction with cell surface heparan sulfate (HS) proteoglycans. As such, the insertion functions to direct ATX␣ to the plasma membrane thereby ensuring localized LPA production and signaling. We also show that the insertion is susceptible to processing by (an) extracellular furin-like proprotein convertase(s), which might serve to fine tune the binding of ATX␣ to cell surface HS proteoglycans.
ATX Constructs and Recombinant Protein-For ATX overexpression studies, human ATX␤ was ligated in the pcDNA3 vector with a 3Ј Myc tag, as described previously (28). Human ATX␣, ligated in pcDNA3 containing a 3Ј Myc/His tag, was a generous gift from Dr. Andree Blaukat (Merck-Serono, Darmstadt, Germany). Mutant ATX␣(R340A) was generated using the Stratagene site-directed mutagenesis kit, with the following primers: p686-forward, GGCTAAGAGACCTAAGGCGAAA-GTTGCCCC and p687-reverse, GGGGCAACTTTCGCCTT-AGGTCTCTTAGCC. For production of recombinant protein, ATX␣ and ATX␤ were ligated in pcDNA5-FRT (Invitrogen) with a C-terminal Myc tag and a His 6 tag. HEK293 cells stably expressing ATX␣ or ATX␤ were generated using the FLPin system (Invitrogen). Recombinant His-tagged ATX was purified from conditioned HEK293 medium using POROS-20 MC columns preloaded with Cu 2ϩ , as described previously (29). The column was washed with 8 -10 column volumes of buffer A (20 mM Tris-HCl, pH 8.0, 150 mM NaCl). ATX protein was eluted with a linear gradient of buffer A containing 500 mM imidazole, and further purified using a Superose size exclusion column.
Expression Analysis-Cells were grown to confluence, and total RNA was extracted using the RNeasy mini kit and column DNase treatment (Qiagen). First-strand cDNA synthesis was performed using 2 g of total RNA, 0.5 g of oligo(dT) primers (Promega), 500 nM dNTPs (Roche Applied Science), 40 units of RNasin (Promega), 10 mM DTT (Invitrogen), and 200 units of Superscript II RT. Quantitative ATX expression was measured using the TaqMan gene expression probe Hs00196470_m1 in an ABI 7500 Fast Sequence Detection System (Applied Biosystems). Cycling parameters were 10 min at 95°C followed by 40 cycles of 15 s at 95°C and 1 min at 60°C. The relative product levels were quantified using the ddCt method and were normalized to GAPDH expression. Expression of the ATX␣ and ATX ␤ isoforms was measured by RT-PCR using primers (forward 5Ј-ATTACAGCCACCAAGCAAGG-3Ј and reverse 5Ј-TCCCT-CAGAGGATTTGTCAT-3Ј) located around the ATX␣-specific insertion, resulting in a 209-bp fragment for ATX␣ and a 366-bp fragment for ATX␤. Human plasmid DNA was used to confirm the specificity of the primers, which were species-independent.
Transfection and Western Blotting-HEK293 cells were transfected with ATX constructs using the calcium phosphate method. At 24 h after transfection, cells were incubated in serum-free DMEM for 48 h. Conditioned medium was centrifuged (5000 rpm for 30 min) to remove cell debris. Samples were analyzed by gel electrophoresis using NuPAGE 4 -12% Bis-Tris gels (Invitrogen) under either reducing or nonreducing conditions. For Western blot analysis, nitrocellulose filters were blocked using 5% nonfat powdered milk and probed with primary antibodies, followed by HRP-conjugated secondary antibodies (Dako, Glostrup, Denmark), and proteins were visualized using ECL detection (Amersham Biosciences).
Cell Migration-Chemotactic migration of human A2058 melanoma cells was measured in 48-well Boyden chambers (Neuroprobe) as described previously (30). Filters (8-m pores) were coated with gelatin, and cells were serum-starved overnight, trypsinized, and seeded in the upper chamber. ATX (2 ng ml Ϫ1 ) was added to the bottom chamber in the presence of 2 M LPC(18:1) and 1 mg/ml fatty acid-free BSA. After 4 h, cells at the top side were removed, and migrated cells at the bottom side of the filter were fixed, stained, and quantified.
Immunofluorescence-HEK293 cells were transfected with C-terminally Myc-tagged ATX␣ or 〈⌻X␤. At 24 h after transfection, cells were fixed with ice-cold ethanol, and ATX was visualized by confocal microscopy using anti-Myc antibody (9E10) and anti-mouse Alexa Fluor 488.
ELISA-Binding of ATX (␣ or ␤) to heparin was evaluated by ELISA in two ways: first, application of ATX to heparin-coated wells of microtiter plates; and second, competition assays in which ATX was first incubated with heparin and then transferred to a heparin-coated plate. A 96-well flat-bottom plate was incubated for 16 h with 100 l of 10 g ml Ϫ1 heparin in coating buffer. As a control, wells were coated for 16 h with 1 g ml Ϫ1 ATX in 0.05 M carbonate-bicarbonate buffer (Sigma-Aldrich). After blocking with PBS containing 2% BSA (w/v) for 1 h, 1 g ml Ϫ1 ATX (in PBS containing 2% BSA) was added and incubated for 2 h. Plates were washed six times with PBS, and bound ATX was detected using goat anti-ATX IgG (1:200), fol-lowed by incubation with alkaline phosphatase-conjugated rabbit anti-goat IgG (1:2000; Sigma-Aldrich), both for 1 h. Plates were washed six times with PBS, and enzyme activity was detected using 100 l of a p-nitrophenyl phosphate solution (1 mg ml Ϫ1 ) containing 1 M diethanolamine and 0.5 mM MgCl 2 , pH 9.8. Absorbance was measured at 405 nm. For the competition assay, 0.1 g ml Ϫ1 ATX was incubated with an increasing content of heparin (0 -1 mg ml Ϫ1 ) for 1 h in 100 l and transferred to a 96-well plate previously coated with heparin. Plates were washed six times with PBS before measuring bound ATX, as described above.
Surface Plasmon Resonance Assays-ATX binding to heparin was measured by surface plasmon resonance spectroscopy using a Biacore T100 machine (Biacore, GE Healthcare). Recombinant ATX␣ and 〈⌻X␤ were covalently coupled to separate flow cells of a CM5 sensor chip at pH 4.0 at roughly equal amounts (about 10,000 response units). Increasing concentrations of unfractionated heparin in running buffer (150 mM NaCl, 20 mM Hepes, pH 7.6) were injected across the chip at flow rate of 50 l min Ϫ1 . Injection of 1 M NaCl was used to regenerate the chip surface at the end of each cycle. The response units for the binding of heparin were evaluated after subtraction of the signal in a flow cell of the same sensor chip which has been left uncoated. To determine the apparent dissociation constant (K d ), the background-subtracted data were fit using a model in which the total binding results from specific binding plus a linear nonspecific component and a concentration-independent background according to the equation where c is the protein concentration, R c the observed response for that specific concentration, B max the estimated maximal response, a the slope of the linear nonspecific components, and b the estimated background. The values for nonspecific binding and the background (a and b) were kept the same for both ATX␣ and 〈⌻X␤ binding, and only the K d value was fit individually for each protein. The model used was judged to be 99.9% more probable than a model assuming only specific binding, according to the Akaike information criterion. Analysis was performed with GraphPad software, using an average molecular mass value for heparin of 15 kDa. Flow Cytometry-For flow cytometry assays, SKOV3 cells were detached by enzyme-free cell dissociation solution (Millipore) and kept on ice for the remaining of the procedure. To establish ATX (␣ or ␤) binding capacity, cells were incubated with 3 g of ATX ml Ϫ1 . Bound ATX was detected by a goat anti-ATX IgG antibody (1:50) and an Alexa Fluor 488 fluorochrome-conjugated anti-goat IgG antibody (1:100; Invitrogen). To determine heparan sulfate or chondroitin/dermatan sulfate dependence, cells were predigested with 20 mIU of heparinase III or 100 mIU of chondroitinase ABC, respectively, ml Ϫ1 basal culture medium containing 1% FCS for 2 h at 4°C. Omission of the first antibody was used as a negative control. Flow cytometry analyses were performed using a BD Biosciences FACScan or FACSCalibur. Viable cells were gated using forward and side scatter, and the fluorescence of this population was measured.

RESULTS AND DISCUSSION
ATX␣: Structure and Expression-The mammalian ATX gene is organized into 27 exons. The three best known splice variants of ATX, termed ␣, ␤, and ␥, differ by the presence or absence of sequences encoded by exons 12 and 21 (Fig. 1A). ATX␣ contains an exon 12-encoded region of 52 residues in the PDE domain. The canonical ATX␤ isoform lacks exons 12 and 21, whereas the "brain-specific" ATX␥ isoform (formerly PD-1␣) contains a 25-amin acid insertion encoded by exon 21 in the NUC domain (32,33). A recently identified fourth isoform, termed ATX␦, is identical to ATX␤ except for a fourresidue deletion in the region that links the NUC and PDE domain (23).
Here we explore the properties of ATX␣ relative to those of ATX␤. The 52-amino acid insertion in ATX␣ is highly basic with a calculated pIϳ11.5. Sequence inspection revealed a consensus cleavage site for furin and furin-like endoproteases (target sequence RXK/RR; Ref. 34) between residues Arg-340 and Lys-341 (Fig. 1B). Cleavage of ATX␣ at Arg-340 by (an) unidentified protease(s) was observed in a previous study (22). Fur-thermore, we noticed that the insertion contains putative heparin-binding motifs, which are commonly characterized by clusters of basic residues (Fig. 1B, indicated in blue). The crystal structure of ATX␤ (18,19) allowed us to understand the localization of the insertion in ATX␣. The insertion is predicted to be unstructured and solvent-exposed, starting right before an ␣-helix in the PDE domain (␣-helix no. 10; Ref. 18). In Fig. 1C, it is depicted as a flexible, hydrophilic loop that protrudes out of the PDE domain, pointing away from the catalytic site and the substrate-binding pocket. However, we cannot rule out the possibility that the insertion adopts a more ordered conformation and might interact with the ATX surface. Given its location and hydrophilic nature, the insertion loop is unlikely to affect the tertiary structure of the ATX␣ catalytic domain. We examined ATX expression in various tissues and cell lines. In agreement with previous studies (3,8,22) and publicly available expression data (35), highest ATX mRNA expression (all isoforms) was found in brain, kidney, spleen, and lung, and somewhat lower expression in liver, skin, and adipose tissue ( Fig. 2A). ATX␣ mRNA has been detected in kidney and liver (22) as well as in melanoma (25) and non-small-cell lung cancer cells (36), but otherwise ATX␣ is much less widely expressed than ATX␤. We detected ATX␣ mRNA in human skin fibroblasts (HF cells) and glioblastoma cells (SNB-78 and U-138) as well as in mouse skin (Fig. 2B), presumably reflecting expression in dermal fibroblasts because skin-derived keratinocytes lacked detectable ATX␣ (data not shown). ATX␣-specific antibodies raised against sequences in the insertion recognized overexpressed ATX␣ in HEK293 cells, but unfortunately they failed to detect endogenous ATX␣ in selected cell lines. 5 It is of note that ATX␣ was consistently found co-expressed with ATX␤ in the cell lines and tissues examined (Fig. 2B). We therefore examined whether both isoforms might interact, but co-immunoprecipitation studies using transfected HEK293 cells did not support this possibility (data not shown). In transfected cells, ATX␣ and ATX␤ showed a similar localization pattern in intracellular vesicles, consistent with both isoforms following the classical trafficking route prior to secretion (Fig.  2C).
Furin-mediated Cleavage of the Insertion at R340-It was reported previously that the insertion renders ATX␣ unstable and prone to proteolyic degradation in vivo (22,23). We examined the expression of both ATX␣ and ATX␤ in HEK293 cells under reducing and nonreducing conditions. When analyzed by reducing SDS-PAGE, secreted ATX␣ was consistently detected not only as a ϳ120-kDa protein but also as two cleavage fragments of ϳ70 kDa and ϳ40 kDa, as shown by using antibodies against the C-and N-terminal parts, respectively (Fig. 3A). In contrast, ATX␤ was always detected as the fulllength protein (Fig. 3A). This result is consistent with secreted ATX␣ being cleaved at the furin consensus site (Arg-340).   , and seventh lanes). Cleaved ATX␣ was detected in the medium (upper two panels), but not in cell lysates (lower panel). Cells were treated with the furin inhibitor Dec-RVKR-CMK overnight (10 M) (fourth and fifth lanes). Conditioned medium was treated with recombinant furin for 1 h at 37°C (sixth and seventh lanes). Antibodies used were anti-ATX C-term (polyclonal; Cayman)) and anti-ATX N-term (monoclonal 4F1). The residual faint bands observed in the third, fourth, fifth, and seventh lanes may be attributable to ATX␣ cleavage at nonoptimal cleavage sites present in the insertion. B, detection of HEK293 cell-conditioned media containing ATX␣ or ATX␤ under reducing and nonreducing SDS-PAGE conditions. ATX␣ cleavage is not observed under nonreducing conditions. Antibody was polyclonal anti-ATX C-term. C, crystal structure of ATX␤ with the ATX␣ insertion superimposed. The N-terminal cleavage fragment is depicted in light gray, the C-terminal fragment in dark gray. The lipid-binding pocket is composed of residues in both halves. LPA is depicted in the binding pocket. The disulfide bridge Cys-157-Cys-403 connects the two fragments. The model was tilted 90°C on the x axis compared with Fig. 1B to emphasize the extended interface between the two cleavage fragments. The polybasic insertion of ATX␣ protrudes into the direction of the viewing plane, depicted by its first and last amino acids (325 and 376, respectively). CMK (37)) prevented proteolysis. Furthermore, addition of recombinant furin caused a nearly complete cleavage of fulllength ATX␣ but left ATX␣(R340A) and ATX␤ intact (Fig. 3A). Importantly, ATX␣ cleavage was detected only in the medium but not in cell lysates (Fig. 3A), indicating that proteolytic processing occurs extracellularly rather than along the intracellular trafficking route. Although furin and furin-like proprotein convertases predominantly reside in the trans-Golgi network, they can also traffic to the cell surface and be shed or secreted into the extracellular milieu (38,39). In particular, furin and its relatives PC5 and PACE4 are known to be secreted and to process diverse extracellular substrates, including growth factors, receptors, and metalloproteinases (38 -41). Which of the secreted furin family members may preferentially attack the ATX␣ insertion remains to be determined.
As shown in Fig. 3B, ATX␣ fragmentation was not observed under nonreducing conditions (Fig. 3B). Thus, furin-mediated cleavage at Arg-340 leaves ATX␣ fully intact. As can be inferred from the structure (Fig. 3C), cleavage at Arg-340 leaves the Nand C-terminal parts tightly together due to an extended interaction surface area between both parts (10,980 Å 2 ; as calculated by PISA (42)). The stability of the interface is primarily due to van der Waals interactions (Ϫ76 kcal mol Ϫ1 ) and numerous hydrogen bonds (Ϫ69 kcal mol Ϫ1 ), as revealed by FoldX analysis (43). Stability is further reinforced by ionic interactions (Ϫ12 kcal mol Ϫ1 ) and a disulfide linkage between Cys-157 and Cys-403 (Ϫ3 kcal mol Ϫ1 ) (Fig. 3C). It thus appear that the N-an C-terminal parts of cleaved ATX␣ form a very stable structure (calculated free energy of dissociation Ͼ100 kcal mol Ϫ1 ) that can only be separated under reducing SDS-PAGE conditions. These results refute previous notions that the ATX␣ isoform is unstable and rapidly degraded in vivo (22,23).
ATX Catalysis-We examined the kinetic parameters of ATX␣ versus ATX␤, using LPA(18:1) and the nucleotide deriv-  ative pNP-TMP as substrates. As shown in Fig. 4A, the catalytic efficiency (V max /K m ) of ATX␣ toward LPC was approximately 70% of that of ATX␤, due to a lower V max value (Fig. 4A). Treatment of ATX␣ with recombinant furin (to fully cleave the insertion) did not affect lysoPLD activity (Fig. 4B). Moreover, ATX␣ and ATX␤ (at 2 nM) were equally effective in stimulating the transwell migration of A2058 melanoma cells in the presence of 2 M LPC during a 4-h time period (Fig. 4C). From these results, we conclude that ATX␣ and ATX␤ show roughly similar LPChydrolyzing efficiency, regardless of the cleavage state of ATX␣.
However, the catalytic efficiency of ATX␣ toward pNP-TMP was approximately 5-fold lower than that of ATX␤ due to a much lower V max (Fig. 5A). We currently have no structural data that can explain this observation. LPA inhibited pNP-TMP hydrolysis by ATX␣ and ATX␤ with K i values of 50 and 110 nM, respectively, suggesting that the insertion may exert a moderately positive effect on LPA binding affinity ( Fig. 5B; see also Ref. 28). Finally, both isoforms were equally sensitive to the ATX inhibitor HA130 (Ref. 31 and data not shown).
ATX␣ Binds Specifically to Heparin-The positively charged Lys/Arg clusters in the ATX␣ insertion (Fig. 1B) could confer binding to heparin, possibly in a cooperative manner (44,45). Heparin is structurally very similar to HS, which is ubiquitously expressed as proteoglycans on cell surfaces and in extracellular matrices, where it recruits growth factors, chemokines, and FIGURE 6. ATX binding to immobilized heparin measured by ELISA. A, ATX␣ and ATX␤ were applied to heparin-coated wells as detailed under "Experimental Procedures." Left columns, ATX␣ binds strongly to heparin, whereas ATX␤ does not. Right columns, control experiments in which ATX␣ and ␤ were directly immobilized to the wells (without heparin), gave similar readings for both isoforms. B, competition ELISA is shown. ATX␣ and ATX␤ were preincubated with various amounts of heparin prior to application to heparin-coated wells. Binding of ATX␣ to immobilized heparin was inhibited by preincubation with heparin in a dose-dependent manner (IC 50 ϳ 300 nM, based on a molecular mass of unfractionated heparin of 15 kDa). other molecules to fine tune signaling events under both physiological and pathophysiological conditions (46 -48). High affinity binding of ATX to heparin in vitro could point to an HS-mediated mechanism for localized LPA production in vivo.
We first determined the binding of ATX␣ and ATX␤ to unfractionated heparin (from intestinal mucosa) using ELISA, in which recombinant ATX was added to heparin-coated wells. We found that ATX␣ binds strongly to immobilized heparin, whereas ATX␤ hardly did (Fig. 6A). As a control for specificity, both isoforms were immobilized directly to the wells (without prior heparin coating), which gave similar readings for both isoforms (Fig. 6A). In addition, we used a competition ELISA, in which ATX was preincubated with various amounts of soluble heparin prior to addition to heparin-coated wells. We observed a dose-dependent inhibition of ATX␣ binding to immobilized heparin (IC 50 ϳ4.5 g ml Ϫ1 or 300 nM), confirming that the observed ATX␣-heparin interaction is specific (Fig. 6B).
We next examined ATX-heparin binding using surface plasmon resonance, in which recombinant ATX was immobilized and heparin was used as the analyte. Again, we found that ATX␣ binds heparin much stronger than does ATX␤ (Fig. 7, A-C). As shown in Fig. 7A, heparin binds to ATX␣ rapidly and dissociates slowly, consistent with specific binding. In contrast, ATX␤ showed very fast heparin binding and dissociation kinet-ics (Fig. 7B), suggestive of nonspecific binding dominated by electrostatic interactions. On the basis of an average molecular mass of heparin of 15 kDa, we determined the apparent K d of ATX␣ for heparin to be approximately 10 nM. The K d of ATX␤ could not be determined reliably because heparin binding did not reach a plateau; however, it is at least 10-fold lower than that of ATX␣ (Fig. 7C). We conclude that the insertion of ATX␣ confers saturable, high affinity binding to heparin with a slow dissociation rate.
An electrostatic surface charge representation of ATX sheds some light on how and where heparin may bind (Fig. 7D). The model shows several positively charged patches common to both isoforms. It is seen that the basic insertion of ATX␣ is located close to one of those patches in the catalytic domain (Fig. 7D), suggesting that the insertion and the positively charged surface patch might act in a cooperative manner to enhance binding to heparin/HS. Co-crystallization studies using defined heparin/HS fragments should help to test this scenario and, furthermore, uncover to what extent proteolytic cleavage at Arg-340 may affect heparin binding.
We also tested whether heparin may affect the catalytic activity of ATX␣. As illustrated in Fig. 8, heparin enhanced the lyso-PLD activity of ATX␣ toward LPC(18:1) up to 2-fold in a concentration-dependent manner (EC 50 ϳ100 nM). In contrast, heparin had little or no effect on the lysoPLD activity of ATX␤.
ATX␣ Binds to Cell Surface HS Proteoglycans-The observed high affinity binding of ATX␣ to heparin strongly suggests that this isoform interacts preferentially with HS proteoglycans, as a means to target LPA production to the cell surface. Although heparin and HS are structurally similar, HS is less negatively charged and shows more structural heterogeneity compared with heparin. We measured the binding of both ATX␣ and ATX␤ to SKOV3 ovarian carcinoma cells, which have a well characterized glycosaminoglycan expression pattern (49 -51). Flow cytometry analysis shows that ATX␣ binds abundantly to the cell surface of SKOV3 cells, whereas ATX␤ does not (Fig. 9). Cell binding was strictly dependent on HS because pretreatment of the cells with the HS-degrading enzyme heparinase III almost completely abolished ATX␣ binding. Binding was HSspecific in that digestion of cell-associated chondroitin/dermatan sulfate, another class of highly sulfated glycosaminoglycans, did not affect ATX␣ binding (Fig. 9). Because heparin stimu-  . ATX binding to human carcinoma cells. ATX␣ binds abundantly to SKOV3 ovarian carcinoma cells (A), whereas ATX␤ does not (B), as measured by flow cytometry. ATX␣ binding was strongly reduced by predigestion of cell-associated heparan sulfates using heparinase III (Hep3), but not after digestion of chondroitin/dermatan sulfates using chondroitinase ABC (ChABC). For technical details see "Experimental Procedures." lates ATX␣ activity in vitro (Fig. 8), binding of ATX␣ to HS proteoglycans may serve not only to target LPA production to the plasma membrane, but also to increase the catalytic efficiency of this particular isoform locally.
In conclusion, our study is the first to define an isoformspecific function for ATX. The polybasic insertion of ATX␣ constitutes a heparin-binding region that targets ATX activity to the plasma membrane via binding to cell surface HS proteoglycans. Previous studies have established that ATX binds to activated integrins via its N-terminal SMB domains, which provides one mechanism for localized production of LPA close to its cognate receptors (18,20). Binding of ATX␣ to cell surface HS proteoglycans via its unique polybasic insertion represents an additional, isoform-specific mechanism for spatially and temporally restricted LPA production (21). Future studies should explore to what extent furin-mediated cleavage of the insertion may modulate its heparin/HS-binding properties, either positively or negatively, and assess whether ATX␣, HS, and a furin-like protease can be detected in a ternary complex at the cell surface.