Crystal structure of Enpp1, an extracellular glycoprotein involved in bone mineralization and insulin signaling

Abstract

Enpp1 is a membrane-bound glycoprotein that regulates bone mineralization by hydrolyzing extracellular nucleotide triphosphates to produce pyrophosphate. Enpp1 dysfunction causes human diseases characterized by ectopic calcification. Enpp1 also inhibits insulin signaling, and an Enpp1 polymorphism is associated with insulin resistance. However, the precise mechanism by which Enpp1 functions in these cellular processes remains elusive. Here, we report the crystal structures of the extracellular region of mouse Enpp1 in complex with four different nucleotide monophosphates, at resolutions of 2.7–3.2 Å. The nucleotides are accommodated in a pocket formed by an insertion loop in the catalytic domain, explaining the preference of Enpp1 for an ATP substrate. Structural mapping of disease-associated mutations indicated the functional importance of the interdomain interactions. A structural comparison of Enpp1 with Enpp2, a lysophospholipase D, revealed marked differences in the domain arrangements and active-site architectures. Notably, the Enpp1 mutant lacking the insertion loop lost the nucleotide-hydrolyzing activity but instead gained the lysophospholipid-hydrolyzing activity of Enpp2. Our findings provide structural insights into how the Enpp family proteins evolved to exert their diverse cellular functions.

Enpp1 (also known as “PC-1”) is a type II transmembrane glycoprotein involved in the regulation of bone mineralization (1, 2). Enpp1 is expressed on the outer surfaces of mineralizing cells, such as osteoblasts and chondrocytes, and on the membranes of osteoblast- and chondrocyte-derived matrix vesicles. Physiological mineralization is regulated by the balance between the extracellular concentrations of inorganic phosphate (Pi), a substrate for mineralization, and inorganic pyrophosphate (PPi), an inhibitor of mineralization (3). Enpp1 negatively regulates bone mineralization by hydrolyzing extracellular nucleotide triphosphates (NTPs) to produce PPi, whereas tissue-nonspecific alkaline phosphatase positively regulates mineralization by hydrolyzing NTPs and PPi to produce Pi. The spontaneous ttw (tiptoe walking) mutant mouse, with a nonsense mutation in the Enpp1 gene, exhibits ectopic ossification of the spinal ligaments, a phenotype similar to ossification of the posterior longitudinal ligament, which is a common form of human myelopathy caused by ectopic ossification of spinal ligaments (4). Moreover, mutations in the Enpp1 gene are associated with generalized arterial calcification of infancy (GACI), a severe autosomal-recessive human disorder characterized by calcification of the internal elastic lamina of large- and medium-sized arteries and stenosis (5–7).

Enpp1 reportedly inhibits insulin signaling (8–17), although controversy remains (18–21). Enpp1 is overexpressed in fibroblastic cells from insulin-resistant individuals (8), and Enpp1 overexpression impaired insulin signaling in cultured cells and mice (12, 13). Enpp1 binds directly to the insulin receptor, thereby inhibiting its insulin-induced conformational changes (14). Moreover, the K173Q polymorphism of Enpp1 (often described as “K121Q,” assuming the use of the ATG start codon 156 bp downstream from the correct one) is associated with insulin resistance, type 2 diabetes, and obesity (15, 16).

Enpp1 is implicated in a variety of physiological and pathological conditions. However, the precise mechanisms by which Enpp1 participates in these cellular processes remain unclarified because of the lack of structural information. Although Enpp1 is essential for the regulation of physiological mineralization, its substrate specificity for different nucleotides and the molecular mechanism conferring its specificity remain unknown. It also is unclear why mutations of amino acid residues located outside the active site render the enzyme inactive and are associated with GACI. Moreover, the molecular mechanism by which Enpp1 inhibits insulin signaling has not been elucidated.

Enpp1 is a member of the ectonucleotide pyrophosphatase/phosphodiesterase (Enpp) family of proteins, which are conserved in vertebrates and hydrolyze pyrophosphate or phosphodiester bonds in various extracellular compounds, such as nucleotides and lysophospholipids (22, 23). The seven mammalian Enpp proteins, Enpp1–7, have distinct substrate specificities and tissue distributions and thus participate in different biological processes. Enpp2 (also known as “autotaxin”) is a secreted lysophospholipase D (lysoPLD) that hydrolyzes lysophosphatidylcholine (LPC) to produce lysophosphatidic acid (LPA), which in turn activates G protein-coupled receptors to evoke various cellular responses (24). The other Enpp family members are either membrane-bound or glycosylphosphatidylinositol-anchored proteins. Enpp1–3 are composed of two N-terminal somatomedin B (SMB)-like domains (SMB1 and SMB2), a catalytic domain, and a nuclease-like domain, whereas Enpp4–7 consist of a catalytic domain and lack the SMB-like and nuclease-like domains. The crystal structures of Enpp2 revealed that lipid substrates are accommodated within a hydrophobic pocket in the catalytic domain (25, 26), which is occluded by an insertion loop in a bacterial nucleotide pyrophosphatase/phosphodiesterase from Xanthomonas axonopodis (XaNPP). The Enpp family members (except for Enpp2) also have the corresponding insertion sequence. These observations explained why Enpp2 is the only family member that exhibits lysoPLD activity and suggested that the insertion loop contributes to defining the substrate specificities (27).

Here, we present the crystal structures of the extracellular region of mouse Enpp1 in complex with four different nucleotide monophosphates (NMPs), which explain the observed preference of Enpp1 for the ATP substrate. Unlike Enpp2, the SMB-like domains are disordered and do not interact with the catalytic domain in Enpp1, suggesting that the SMB-like domains in Enpp1 and Enpp2 have distinct roles. Structural mapping of disease-associated mutations indicated the functional significance of the interaction between the catalytic and nuclease-like domains in both Enpp1 and Enpp2.

Results

Substrate Specificity.

The extracellular region (residues 92–905) of mouse Enpp1 was overexpressed in HEK293S GnT1⁻ cells as a secreted protein and was purified by P20.1 antibody affinity and gel filtration chromatographies. When ATP was incubated with the purified protein, the production of AMP and PPi, but not ADP and Pi, was detected by mass spectrometry (Fig. 1A), indicating that Enpp1 hydrolyzes the phosphodiester bond between the α- and β-phosphate groups of ATP. A kinetic analysis showed that Enpp1 preferably hydrolyzes ATP (k_cat = 16 s⁻¹, K_m = 46 μM), compared with UTP (k_cat = 200 s⁻¹, K_m = 4.3 mM), GTP (k_cat = 820 s⁻¹, K_m = 4.2 mM) and CTP (k_cat = 8.7 s⁻¹, K_m = 1.2 mM) (Fig. S1). We also examined the substrate specificity by measuring the p-nitrophenyl thymidine 5′-monophosphate (pNP-TMP)–hydrolyzing activity in the presence of different NMPs. AMP inhibited the pNP-TMP–hydrolyzing activity more potently than TMP, GMP, and CMP (Fig. 1B). These results showed that Enpp1 preferably hydrolyzes ATP to produce AMP and PPi and confirmed that Enpp1 negatively regulates bone mineralization by hydrolyzing ATP, an abundant extracellular nucleotide.

Fig. 1.

Biochemical characterization. (A) Enpp1 hydrolyzes ATP to produce AMP and PPi. Purified Enpp1 was incubated with ATP, and then the reaction products were quantified by mass spectrometry. (B) Inhibition of Enpp1 activity by NMPs. The enzymatic activity was measured in the absence or presence of 0.5 mM NMPs, using 4 mM pNP-TMP as a substrate. (C) ATP– and pNP-TMP–hydrolyzing activities of Enpp1 mutants. (D) LPC– and pNP-TMP–hydrolyzing activities of the wild type and ΔIL mutant of Enpp1. Data are shown as mean ± SD (n = 3).

Overall Architecture.

We solved the crystal structures of the extracellular domain of Enpp1 in complex with four different NMPs (AMP, TMP, GMP, and CMP) at resolutions of 2.7–3.2 Å (Table S1). Because these four crystal structures are essentially identical (rmsd values less than 0.2 Å for aligned Cα atoms), we describe the AMP complex structure, unless otherwise stated. The structure consists of a catalytic domain (residues 190–578), a nuclease-like domain (residues 629–902), and two linker regions, L1 and L2 (residues 170–189 and 579–628, respectively) (Fig. 2 A and B). The catalytic domain interacts with the nuclease-like domain, and the L2 linker connects the two domains (Fig. 2B). The structure revealed that Enpp1 is N-glycosylated at Asn267, Asn323, and Asn567 and that the domain interaction is reinforced by the Asn567-linked glycan and the Cys462–Cys846 disulfide linkage, which correspond to the Asn524-linked glycan and the Cys413–Cys801 disulfide linkage, respectively, in mouse Enpp2 (Fig. 2 B and C) (25, 26). The spatial arrangement of the catalytic and nuclease-like domains is conserved in Enpp1 and Enpp2, suggesting that the interdomain interactions play similar roles in Enpp1 and Enpp2.

Fig. 2.

Overall architecture. (A) Domain organization of mouse Enpp1. (B) Crystal structure of the extracellular domain of Enpp1 in complex with AMP. Catalytic domain, cyan; nuclease-like domain, magenta; L1, wheat; L2, yellow-green; EF hand-like motif, pink; insertion loop, gold. AMP and N-glycans are shown as green and yellow sticks, respectively. The bound zinc and calcium ions are shown as gray and yellow-green spheres, respectively. Disulfide linkages are shown as sticks. The two SMB-like domains, which were disordered in the crystal structure, are indicated by circles. (C) Crystal structure of Enpp2 in complex with 14:0-LPA (PDB ID 3NKN); color code as in B. The SMB1 and SMB2 domains are colored orange and brown, respectively.

Catalytic Domain.

The catalytic domain of Enpp1 is structurally similar to those of Enpp2 (25, 26) (PDB ID 3NKM, 48% sequence identity, rmsd = 1.1 Å for 341 Cα atoms) and XaNPP (28) (PDB ID 2GSU, rmsd = 1.6 Å for 339 Cα atoms) (Fig. 2 B and C and Fig. S2). As in Enpp2 and XaNPP, two zinc ions are bound within the active site of Enpp1. One zinc ion is coordinated by Asp358, His362, and His517, and the other is coordinated by Asp200, Thr238, Asp405, and His406 (Fig. 3). A previous mutational analysis confirmed the functional significance of these zinc-coordinating residues (29). The α-phosphate group of AMP is bound between the two zinc ions, consistent with our functional data showing that Enpp1 hydrolyzes ATP to produce AMP and PPi (Fig. 1A).

Fig. 3.

Nucleotide recognition. Active site of Enpp1 in complex with AMP (A), TMP (B), GMP (C), and CMP (D). The bound NMPs are shown as green sticks. F_O–F_C omit electron density maps, contoured at 3.5 σ, are shown as blue meshes. The bound zinc ions and water molecules are shown as gray and red spheres, respectively. Hydrogen bonds and coordinate bonds are shown as dashed gray and yellow lines, respectively.

Nuclease-Like Domain.

The nuclease-like domain of Enpp1 is structurally similar to that of Enpp2 (25, 26) (PDB ID 3NKM, 42% sequence identity, rmsd = 1.4 Å for 248 Cα atoms) (Fig. 2 B and C and Fig. S3). As in Enpp2, a calcium ion is coordinated by the side chains of Asp780, Asp782, Asp784, and Asp788 and the main-chain carbonyl group of Arg786, forming an EF hand-like motif. In Enpp2, Asp735 (corresponding to Asp780 in Enpp1) interacts with Lys430 (corresponding to Lys479 in Enpp1), and the K430A mutation impaired the protein stability (26). In Enpp1, Asp780 interacts with Lys479 in the catalytic domain, suggesting the importance of the EF hand-like motif for the interdomain interactions in both Enpp1 and Enpp2.

SMB-Like Domains.

Unexpectedly, electron densities were not observed for the two SMB-like domains, and there is sufficient room to accommodate the two SMB-like domains in the crystal lattice. An SDS/PAGE analysis of the dissolved crystals revealed a single band (∼100 kDa) similar to the purified protein (Fig. S4A), indicating that the crystallized proteins contain the SMB-like domains. The reported Enpp1 SMB1 domain (PDB ID 2YS0) is structurally similar to the Enpp2 SMB1 domain (25, 26) (PDB ID 3NKM, 53% sequence identity, rmsd = 1.2 Å for 34 Cα atoms) (Fig. S4B). The SMB2 domains of Enpp1 and Enpp2 share 48% sequence identity, indicating that the Enpp1 SMB2 domain also adopts a rigid structure. A structural comparison of Enpp1 with Enpp2 indicated that the SMB1 domain of Enpp1 cannot interact with the catalytic domain in the way observed in Enpp2 because of steric clashes with the insertion loop (Fig. S4C). In addition, Arg283, Gln290, and Gln344 in the catalytic domain of Enpp2, which participate in the interaction with the SMB2 domain (25), are replaced with Glu330, Glu337, and Asp391, respectively, in Enpp1 (Fig. S4C). These observations suggested that, unlike Enpp2, the spatial arrangement of the SMB-like domains of Enpp1 is not fixed by the interaction with the catalytic domain. To confirm this idea, we prepared an Enpp1 mutant bearing the Turbo3C protease recognition sequence between SMB2 (Lys169) and L1 (Lys170), incubated the purified mutant protein with the protease, and then performed a pulldown assay using P20.1-Sepharose. We found that the SMB-like domains were not pulled down together with the rest of the protein, suggesting that the SMB-like domains do not interact with the catalytic domain (Fig. S4D). In contrast, the equivalent Enpp2 mutant with the protease recognition sequence between SMB2 (Glu140) and L1 (Ser141) was not expressed in HEK293T cells, suggesting differences in the flexibility between SMB2 and L1 in Enpp1 and Enpp2. Notably, the protease treatment had almost no effect on the enzymatic activity (Fig. S4E), indicating that the SMB-like domains are dispensable for the enzymatic activity of Enpp1. Given that Enpp1 is a type II transmembrane protein, the mobile SMB-like domains of Enpp1 may act as a molecular anchor that connects the transmembrane region and the catalytic domain.

Nucleotide Recognition.

The crystal structures in complex with the four different NMPs revealed that the phosphate groups and ribose moieties are recognized by the protein in a similar manner, and the nucleobase moieties are sandwiched between the side chains of Phe239 and Tyr322 (Fig. 3). The F239A and Y322A mutants showed reduced hydrolytic activities for ATP and pNP-TMP (Fig. 1C). The AMP N6 atom is recognized by Trp304 and Asp308 through a water-mediated hydrogen bond network (Fig. 3A) in the AMP complex, and the D308A mutant exhibited decreased ATP–hydrolyzing activity (Fig. 1C). In contrast, the nucleobase moieties of TMP, GMP, and CMP are not recognized by Enpp1 through hydrogen-bonding interactions (Fig. 3 B–D). These observations can explain the preference of Enpp1 for ATP as a substrate, as described above (Fig. 1A). Asp308 of Enpp1 is replaced with Glu160 in XaNPP (Fig. S2C) (28), suggesting differences in the substrate preferences of Enpp1 and XaNPP (although the substrate preference of XaNPP remains unclear).

Molecular Determinants of Substrate Specificity.

The present structure revealed that the insertion loop (residues 304–323) participates in the formation of the substrate-binding pocket, with Trp304 in the WPG motif forming a hydrophobic core with Leu196, Ser198, His242, Ile245, Val246, Trp289, and Thr351 (Fig. 4 A and B). The main-chain amide groups of Trp304 and Tyr322 hydrogen bond with the side chains of Ser307 and Asp308, respectively. His242 in Enpp1 corresponds to Leu213 in Enpp2, which participates in the formation of the lipid-binding hydrophobic pocket (Fig. 4 C and D). As previously reported (30), the H242L mutant showed reduced phosphodiesterase activity (Fig. 1C), indicating the importance of the interaction between His242 and Trp304 in the formation of the nucleotide-binding pocket. To further examine whether the insertion loop is the major determinant of the substrate specificity, we prepared the Enpp1 mutant lacking residues 304–323 (ΔIL mutant) and measured its pNP-TMP– and LPC–hydrolyzing activities. The ΔIL mutant showed drastically reduced pNP-TMP–hydrolyzing activity compared with the wild type (Fig. 1D), indicating the importance of the insertion loop for the nucleotide recognition. Notably, the ΔIL mutant displayed lysoPLD activity (Fig. 1D), suggesting that a lipid-binding pocket is generated by the deletion of the insertion loop. However, the lysoPLD activity of the ΔIL mutant (0.025 nmol μg⁻¹ h⁻¹) was much lower than that of Enpp2 (38 nmol μg⁻¹ h⁻¹) (25). In Enpp2, the hydrophobic pocket is formed by conserved hydrophobic residues, such as Ile167, Leu213, Leu216, Ala217, Leu259, Trp260, Phe273, Val302, Ala304, and Met512 (25), which correspond to Leu196, His242, Ile245, Val246, Ile288, Trp289, Tyr302, Phe349, Thr351, and Met555, respectively, in Enpp1 (Fig. 4 B and D). Thus, the replacements of His242, Val246, Phe349, and Thr351 in Enpp1 with smaller hydrophobic residues may be required for the formation of a hydrophobic pocket optimized for accommodating lipid substrates. In addition, Tyr302 of Enpp1 is located at a different position from the corresponding Phe273 of Enpp2, and Tyr302 interacts with Phe303, Arg331, Ala334, and Trp338 (Fig. 4B). Taken together, these observations suggested that, in addition to the loop deletion, amino acid replacements may have been necessary for the molecular evolution of Enpp1 to Enpp2.

Fig. 4.

Substrate binding pocket. (A) Molecular surface of Enpp1. The insertion loop is shown as a tube. (B) Nucleotide-binding pocket of Enpp1 (stereo view). The insertion loop is shown in gold in A and B. (C) Molecular surface of Enpp2 (PDB ID 3NKN). (D) Lipid-binding hydrophobic pocket of Enpp2 (PDB ID 3NKN) (stereo view). The SMB-like domains are omitted for clarity in D. The bound zinc ions are shown as gray spheres in B and D.

Discussion

Although Enpp1 hydrolyzes various nucleotide substrates in vitro (31), its substrate preference was unknown. Our functional analysis revealed that Enpp1 preferentially hydrolyzes ATP to produce AMP and PPi. Moreover, the present structures provide a molecular basis for PPi production by Enpp1, through ATP hydrolysis. In Enpp2, lipid substrates are accommodated in a deep, hydrophobic pocket (25, 26). In contrast, in Enpp1, the nucleotide substrates are accommodated in the pocket formed by the insertion sequence, which occludes the hydrophobic pocket. These structural differences clearly explain why Enpp1 hydrolyzes nucleotides but not lipids. Moreover, the deletion of the insertion sequence of Enpp1 resulted in the generation of lysoPLD activity (albeit lower than that of Enpp2). A structural comparison of Enpp1 and Enpp2 suggested that Enpp2 gained the hydrophobic pocket during the course of evolution through the deletion of the insertion loop, followed by amino acid replacements. These observations reinforced our previous proposal that the insertion sequence participates in defining the substrate specificity of the Enpp family proteins.

A number of genetic mutations in Enpp1 are associated with GACI, a human disorder with a hypermineralization phenotype (5–7). Most of these mutations mapped to the catalytic and nuclease-like domains (Fig. 5 A and B). Among these, the R456Q, L579F, L611V, C726R, N792S, E893X, and Y901S mutations abolished the enzymatic activity in human Enpp1 (5, 32). The structure of mouse Enpp1 revealed that most of these residues participate in intradomain or interdomain interactions (Fig. 5 B–G). In the catalytic domain, Arg438 (Arg456; equivalent residues of human Enpp1 are indicated in parentheses) hydrogen bonds with Asp258 and Glu490 (Fig. 5C), and Leu561 (Leu579) forms a hydrophobic core with Thr291, Ala292, Gln295, Val297, and Leu559 (Fig. 5D). The D276N mutation in Aps276 (equivalent to Asp258 of mouse Enpp1) also is associated with GACI. In the nuclease-like domain, Glu873 (Glu893) interacts with Lys895 and Phe880 (Fig. 5E). Asn772 (Asn792) and Tyr881 (Tyr901) hydrogen bond with Thr676 and with Thr816 and Ser833, respectively (Fig. 5E). These observations suggested that these inactivating mutations result in either the loss of interactions or steric clashes within the individual domains, highlighting the functional significance of the catalytic and nuclease-like domains. The ttw mouse carries the G568stop nonsense mutation, which results in the deletion of the nuclease-like domain (4), and the Enpp1 mutant protein with a truncated nuclease-like domain (Δ804–905) is catalytically inactive (18), further highlighting the functional significance of the nuclease-like domain. Leu593 (Leu611) on the L2 linker interacts with Arg641 and Leu653 in the nuclease-like domain, and Arg641 interacts with Glu589 on the L2 linker (Fig. 5E). Cys706 (Cys726) in the nuclease-like domain forms a disulfide bond with Cys607 on the L2 linker (Fig. 5F). These observations indicated that the interaction between the nuclease-like domain and the L2 linker is important for the enzymatic activity. In addition, disease-causing mutations are mapped at the domain interface (H500P in the catalytic domain and D804H and R888W in the nuclease-like domain) (Fig. 5G). His482 (His500) and Arg868 (Arg888) hydrogen bond with Asp871 and with Tyr250 and Glu547, respectively. Asp784 (Asp804) is located in the EF hand-like motif, and Asp780 within this motif interacts with Lys479 in the catalytic domain. These observations indicated the functional significance of the interaction between the catalytic and nuclease-like domains. The structural mapping of the disease-causing mutations revealed that the integrity not only of the individual domains but also of the interdomain interactions is important for the enzymatic activity of Enpp1.

Fig. 5.

Structural mapping of disease-causing mutations. (A) Mapping of mutations associated with GACI on the primary structure of human Enpp1. (B) Mapping of the disease-causing mutations on the crystal structure of mouse Enpp1. (C–G) Close-up views of boxed areas in B. The residues of mouse Enpp1 corresponding to the disease-associated residues of human Enpp1 are shown as gray sticks in B–G.

Enpp1 interacts directly with the insulin receptor and prevents insulin-induced conformational changes in the receptor, thereby inhibiting insulin signaling (14). The Enpp1 K173Q polymorphism is associated with obesity and type 2 diabetes (15), and the K173Q mutant protein inhibited insulin signaling more efficiently, by interacting with the insulin receptor more strongly than the wild type (17). Lys173 is replaced by a histidine residue in mouse Enpp1, suggesting that the K173Q polymorphism is specific to human Enpp1. Lys173 of human Enpp1 is located within the SMB2 domain, and thus this domain may participate in the interaction with the insulin receptor. The SMB2 domain of Enpp2 binds to integrins (26), and the SMB domain of vitronectin binds to plasminogen activator inhibitor-1 (PAI-1) (33) and the urokinase receptor (uPAR) (34). Lys173 of human Enpp1 corresponds to Arg126 of the mouse Enpp2 SMB2 domain and Tyr28 of the vitronectin SMB domain (Fig. S4B). Arg126 of Enpp2 is exposed to the solvent (25), and Tyr28 of vitronectin participates in the interaction with PAI-1 (33) and uPAR (34). These observations support the notion that Lys173 of human Enpp1 participates in the interaction with the insulin receptor.

We previously hypothesized that the interaction between the catalytic and nuclease-like domains contributes to maintaining the structural integrity of the hydrophobic pocket in Enpp2 (25). However, the present results revealed that the interdomain interaction is conserved in Enpp1 and is important for the enzymatic activity, although Enpp1 lacks a hydrophobic pocket. Recent molecular dynamics simulations revealed that the interdomain interaction contributes mainly to the correct positioning of the catalytic threonine residue in the catalytic domain, explaining the requirement of the conserved interdomain interaction in Enpp1 and Enpp2 (35). We also found that the SMB-like domains of Enpp1 do not interact with the catalytic domain (Fig. 2B) and are dispensable for the enzymatic activity (Fig. S4E), consistent with the mapping of most of the disease-causing mutations on the catalytic and nuclease-like domains (Fig. 5A) (5–7). These observations suggested that the SMB-like domains of Enpp1 act as a flexible molecular anchor, consistent with the notion that the SMB2 domain of human Enpp1 interacts with the insulin receptor. In Enpp2, the SMB-like domains interact with the catalytic domain and form a hydrophobic channel, which may serve as an exit for lipid products to specific G protein-coupled receptors (25, 26). Thus, the distinct arrangements of the SMB-like domains are likely to reflect their functional differences in Enpp1 and Enpp2.

In summary, our findings suggest that Enpp1 participates in different biological processes through distinct sets of domains: the catalytic and nuclease-like domains for bone mineralization and the SMB-like domains for insulin signaling. Moreover, our findings indicate that Enpp1 and Enpp2 exert diverse cellular functions because of their distinct domain arrangements and active-site architectures, although they share similar primary structures.

Materials and Methods

The extracellular region (residues 92–905) of mouse Enpp1 was expressed, purified, and crystallized as described previously (36). X-ray diffraction data were collected at 100 K on beamline BL32XU at SPring-8 (Hyogo, Japan). The crystal structure in complex with AMP was determined by the single-wavelength anomalous dispersion method, and the crystal structures in complex with the other NMPs were determined by molecular replacement. Data collection and refinement statistics are provided in Table S1. Molecular graphics were prepared using CueMol (http://www.cuemol.org). Detailed methods are described in SI Materials and Methods.