Crystal Structure of CD26/Dipeptidyl-peptidase IV in Complex with Adenosine Deaminase Reveals a Highly Amphiphilic Interface*[boxs]

Dipeptidyl-peptidase IV (DPPIV or CD26) is a homodimeric type II membrane glycoprotein in which the two monomers are subdivided into a β-propeller domain and an α/β-hydrolase domain. As dipeptidase, DPPIV modulates the activity of various biologically important peptides and, in addition, DPPIV acts as a receptor for adenosine deaminase (ADA), thereby mediating co-stimulatory signals in T-lymphocytes. The 3.0-Å resolution crystal structure of the complex formed between human DPPIV and bovine ADA presented here shows that each β-propeller domain of the DPPIV dimer binds one ADA. At the binding interface, two hydrophobic loops protruding from the β-propeller domain of DPPIV interact with two hydrophilic and heavily charged α-helices of ADA, giving rise to the highest percentage of charged residues involved in a protein-protein contact reported thus far. Additionally, four glycosides linked to Asn229 of DPPIV bind to ADA. In the crystal structure of porcine DPPIV, the observed tetramer formation was suggested to mediate epithelial and lymphocyte cell-cell adhesion. ADA binding to DPPIV could regulate this adhesion, as it would abolish tetramerization.

CD26 or dipeptidyl-peptidase IV (DPPIV, 1 EC 3.4.14.5) is a ubiquitous, multifunctional integral type II membrane glycoprotein located on the surface of a variety of epithelial, endothelial, and lymphoid cells, and it also occurs in soluble form in serum. As serine exopeptidase, DPPIV cleaves N-terminal dipeptides from polypeptides with proline or alanine preferentially in the penultimate position, thereby regulating the activity of a variety of biologically important peptides (1). Besides its peptidase activity, one of the main functions of DPPIV is binding to adenosine deaminase (ADA, EC 3.5.4.4) and to the ex-tracellular matrix (2). All of these functions can influence T-cell proliferation (3,4). In addition, DPPIV interacts with human immunodeficiency virus-1 Tat protein (5), and the human immunodeficiency virus envelope protein gp120 inhibits binding of DPPIV to ADA (6).
Crystal structures of DPPIV free and in complex with different inhibitors show that DPPIV is a "U"-shaped homodimer, the two monomers being related by a pseudo-2-fold rotation axis coinciding with the symmetry axis of the "U" (3,(7)(8)(9). The six-residue-long N-terminal cytoplasmic tails of the DPPIV dimer are followed by 22-residue-long transmembrane ␣-helices. The subsequent extracellular ectodomains are divided into two domains each. Eight-bladed ␤-propeller domains (Arg 54 -Asn 497 ) form the arms of the "U" distal to the membrane plane, and ␣/␤-hydrolase domains (Gln 508 -Pro 766 ) proximal to the membrane form the bend of the "U" and harbor the catalytic triads that are required for peptidase activity.
Soluble DPPIV (without transmembrane ␣-helix) migrates as a dimer in gel filtration (3) but can also form higher molecular weight assemblies with apparent mass of 900 kDa (10). This could be due to a tetrameric assembly as observed in the crystal structure of porcine DPPIV (pDPPIV) that is associated with head-to-head binding between the ends of the arms of two "U"-shaped DPPIV dimers to form a "ʚ ʛ"-shaped complex. Engel et al. (9) suggested that pDPPIV tetramerization might be involved in cell-cell contacts, and this view would explain why DPPIV promotes adhesion between lymphocytes and epithelial cells that is inhibited by the addition of exogenous ADA (11).
ADA is a ubiquitous, soluble, and globular enzyme with a TIM barrel fold (eight parallel ␤-strands forming a barrel decorated by ␣-helices) (12). It is present in all mammalian tissues and involved in the development and function of lymphoid tissue. ADA binds specifically to the DPPIV of humans, cattle, and rabbits with dissociation constants of 3 to 20 nM depending on the organism (3). Binding of ADA to DPPIV is important in regulating the extracellular local concentration of adenosine (13,14). Inhibition of ADA reduces signals mediated by CD3 and T-cell receptors and suggests a correlation between ADA binding to DPPIV and T-cell activation (15), as elevated adenosine concentration inhibits the proliferation of T-lymphocytes. Although they share high sequence homology with their higher mammal analogs, DPPIV and ADA of rodents do not form a complex.
A cryo-electron microscopy study of the DPPIV⅐ADA complex at 22 Å resolution confirmed the location of the ADA binding site on the ␤-propeller domain of DPPIV (16). Because details of intermolecular interactions remained elusive at this low resolution, ADA was oriented relative to DPPIV on the basis of mutagenesis studies, which suggested that helix ␣2 of ADA binds to segments of ␤-propeller blades IV and V of DPPIV (17)(18)(19).
We crystallized the complex of human DPPIV (hDPPIV) ectodomain with bovine ADA (bADA), which shares 91% amino acid sequence homology with human ADA (hADA). The crystal structure determined at 3.0 Å resolution shows the intermolecular contacts in a highly amphiphilic interface that contribute to and stabilize hDPPIV⅐bADA complex formation.

EXPERIMENTAL PROCEDURES
Production, Purification, and Crystallization of hDPPIV⅐bADA-Full-length, enzymatically active hDPPIV was expressed in Sf9 cells and purified (20). N-terminal sequencing showed that residues 1-29 were uniformly truncated during purification. The hDPPIV⅐bADA complex was prepared by mixing hDPPIV and bADA (Sigma type VIII from calf intestinal mucosa) with a molar excess of bADA in 20 mM Tris, 500 mM NaCl, pH 8.0, and incubating at 4°C overnight. The formed complex was analyzed by gel electrophoresis under nondenaturing conditions (20), purified by gel filtration on Superdex 200 10/30 (Amersham Biosciences) with 20 mM Tris-HCl, 150 mM NaCl, pH 8.0 (16), and concentrated to 7 mg/ml. For crystallization, a reduced factorial screen (Hampton Research) was set up using the vapor diffusion method. Crystals were obtained with 20 -22% polyethylene glycol 3350, 200 mM NaCl, and 100 mM Tris-HCl, pH 8.0, and optimized by microseeding. Prior to data collection, the crystals were soaked in 25% polyethylene glycol 3350, 200 mM NaCl, 100 mM Tris, pH 8.0, supplemented with 20% glycerol, and cryo-cooled. Diffraction quality improved by annealing the crystal to room temperature twice for 3 s.
Structure Determination-X-ray data were collected at BESSY II, Berlin, Germany (Beamline BL1 of Free University, Berlin) and processed with DENZO and SCALEPACK (21); see Table I. The structure of hDPPIV⅐bADA was solved by a molecular replacement method with the program MOLREP (22) using glycoside-depleted hDPPIV as a partial search model (Protein Data Bank (PDB) code 1N1M), yielding two DPPIV dimers per asymmetric unit. Difference electron density clearly showed the presence of ADA molecules. Although an additional search using bADA (PDB code 1KRM) located four bADA molecules, the asymmetric unit is occupied by two (hDPPIV⅐bADA) 2 complexes. Refinement was done using the REFMAC5 program (23) by applying four-fold noncrystallographic symmetry restraints to the protein chains but omitting the interface regions (24). Manual model building was done with the program O (25). Glycoside residues could be fitted to the electron density at eight of the expected nine N-glycosylation sites except for Asn 685 . For statistics, see Table I. The figures of molecules were prepared with MolScript (26) and RASTER3D (27).

RESULTS AND DISCUSSION
Overall Structure-hDPPIV was expressed and purified as full-length protein, but N-terminal sequencing revealed that it was cleaved during purification after residue 29, explaining its solubility in detergent-free buffer. The final electron density map of hDPPIV⅐bADA featured no density for the N-terminal residues 30 -38 presumably because of high flexibility and/or disorder of this segment. Ser 39 is the first residue with appropriate electron density of the hDPPIV ectodomain.
The crystal asymmetric unit contains two complexes, each with (hDPPIV⅐bADA) 2 stoichiometry and a molar mass of 275 kDa (Fig. 1A). Each hDPPIV binds one bADA at the periphery of the ␤-propeller domain near the ends of the arms of the "U" (Fig. 1B), thereby preserving the pseudo-two-fold symmetry of the complex (Fig. 1A). All crystal contacts of (hDPPIV:bADA) 2 are distant to the interface region, suggesting that they will not structurally influence the binding of hDPPIV to bADA. Super-position of main chain structures of hDPPIV (PDB code 1N1M) and bADA (PDB code 1KRM) with their equivalents in (hDPPIV⅐bADA) 2 yielded average root mean square deviations of 0.8 and 0.9 Å, respectively, indicating that their structures are nearly identical. In particular, the interface regions of hDPPIV and bADA do not rearrange significantly upon complex formation and show root mean square deviations of 1.0 and 0.6 Å, respectively, for main chain atoms. In the (hDPPIV⅐bADA) 2 complex, bADA does not block the possible path of substrate through the central tunnel of the ␤-propeller or through the side opening to the active site of hDPPIV (Fig.  1), and conversely, binding of hDPPIV (Fig. 1B) does not block the active site of bADA. This agrees with studies indicating that upon complex formation, DPPIV and ADA remain catalytically active (28).
The hDPPIV⅐bADA Interface-The hDPPIV:bADA interface buries 930-Å 2 solvent-accessible surface area/polypeptide corresponding to 3 and 7% of the total accessible surface of individual hDPPIV and bADA molecules, respectively. Four glycoside residues linked to the conserved Asn 229 of hDPPIV are well defined in the electron density (Fig. 2). They contact 33 Arg-Arg-Gly-Ile 36 of bADA and form one intermolecular hydrogen bond, and the protein-sugar interactions increase the buried surface by 170 Å 2 . This contact, however, is not a prerequisite for ADA binding as shown by site-directed mutagenesis of Asn 229 3 Ala or deletion of any other N-glycosylation site (29).
hDPPIV binds bADA with two adjacent loops. Loop A (Ile 287 -Asp 297 ) connects blades IV and V, and loop B (Asp 331 -Gln 344 ) links ␤-strands ␤3 and ␤4 of blade V (Fig. 3). The loops protrude from the propeller blades and form a cleft accommodating helix ␣2 (Pro 126 -Asp 143 ) of bADA (Fig. 4A). Interactions between helix ␣1 (Arg 76 -Ala 91 ) of bADA and loop A of hDPPIV complete the interface (Fig. 4). As indicated in Fig. 3 by triangles, 14 and 13 residues in hDPPIV and bADA, respectively, are engaged in intermolecular contacts at the binding interface (see Fig. 4B). The supplementary information provides all interfacial contacts shorter than 3.9 Å.
A total of 11 intermolecular hydrogen bonds are formed between hDPPIV and bADA (Fig. 4). bADA-Glu 139 on helix ␣2 appears to be important, as the carboxylate oxygen O⑀1 forms hydrogen bonds to loop A of hDPPIV involving hDPPIV-Ala 291 NH, hDPPIV-Ser 292 NH, and hDPPIV-Ser 292 O␥. In addition, bADA-Glu 139 O⑀2 is hydrogen-bonded to hDPPIV- Gln 344 N⑀ on loop B, and hDPPIV-Gln 344 O⑀ is in contact with bADA-Arg 142 N on helix ␣2. The array of 11 hydrogen bonds with one hydrogen bond per 169 Å 2 of buried surface is in close agreement to the average value of 10 hydrogen bonds per protein-protein interface and one hydrogen bond per 170 Å 2 of buried interface area (30).
The interface in the hDPPIV⅐bADA complex is strongly amphiphilic. Of the 14 residues of hDPPIV forming intermolecular contacts at the interface, seven are apolar, six neutral polar, and only one (Arg 336 ) is charged (Figs. 3, 4B, and 5). This agrees grossly with the average distribution of residues in protein-protein interfaces, which comprise 57% apolar, 24% neutral polar, and 19% charged residues (30). By contrast, all of the 13 residues forming the interface region of bADA are polar, and among these, 9 (69%) are charged (see Fig. 5). The charged residues of bADA form two intramolecular and one intermolecular salt bridges (bADA-Arg 76 ⅐⅐⅐bADA-Glu 128 , bADA⅐Arg 142 ⅐⅐⅐bADA-Asp 143 ), and (bADA-Asp 127 ⅐⅐⅐hDPPIV-Arg 336 , respectively), and the other charged residues of bADA form hydrogen bonds to hDPPIV (Fig. 4). The highly charged interface region of bADA is unusual. Of all 27 residues participating in complex formation at the hDPPIV⅐bADA interface, 10 (37%) are charged, which is the highest percentage ever observed in protein⅐protein complexes, with the next highest value being 27% (30). As the sequences of human and bovine ADA are identical at the binding interface (Fig. 3), except for the conservative substitution hADA-Glu 77 /bADA-Asp 77 that forms a hydrogen bond to hDPPIV-Gln 286 N⑀ (Fig. 4), structures of hDPPIV in complex with bovine and human ADA are supposed to be nearly identical.
Binding Studies Are Consistent with the (hDPPIV⅐bADA) 2 Structure-ADA from mouse and rat does not bind to endogenous DPPIV but binds weakly (in the M range) to the DPPIV of primates, cattle, and rabbits. Although this lack of complex formation in rodents might be associated with amino acid substitutions in loops A and B of DPPIV and in helix ␣2 of ADA (Fig. 3), it has inspired mutational, binding, and kinetic studies on complex formation between DPPIV (18) and ADA (31).
For wild type (WT) hADA nearly all residues on helix ␣2 were substituted, and the binding of the variants to WT rabbit DPPIV (rDPPIV) was studied in terms of dissociation constants (K D ) and binding kinetics (k on , k off ) (19). The affinity of the complex between WT hADA and rDPPIV (K D ϭ 17 nM) is ϳ300ϫ stronger than between WT murine ADA (mADA) and rDPPIV (K D ϭ 5,400 nM). Compared with WT hADA, the most significant effects of single point mutations performed on helix ␣2 of hADA were found with Glu 139 3 Ala, Arg 142 3 Ala, and Asp 143 3 Ala, which showed an ϳ10to 8-fold decrease in binding affinity to rDPPIV, with K D between 160 and 112 nM (31). In all cases, the kinetic constants for complex formation (k on ) were comparable (k on ϳ2 ϫ 10 4 M Ϫ1 s Ϫ1 ) and similar to that for WT hADA and rDPPIV, whereas those for complex dissociation differed by one power (k off ϳ2-4 ϫ 10 Ϫ3 s Ϫ1 ) compared with WT hADA, (k off ϳ4 ϫ 10 Ϫ4 s Ϫ1 . Such behavior (nearly constant k on but differing k off in point-mutated protein complexes) suggests important contributions by hydrophobic interactions (32), which, besides the salt bridges and hydrogen bonds detailed above, must add considerably to hDPPIV⅐bADA binding. This agrees with the finding that the (hDPPIV⅐bADA) 2 complex dissociates at low ionic strength, suggesting that hydrophobic amino acids are essential for binding (33).
As interfacial residues of mADA and hADA differ solely in positions 142 and 143 on helix ␣2 (Fig. 3), a chimeric mADA construct containing the human ␣2 helix (residues 126 -143) would be expected to bind to rDPPIV in a similar manner as WT hADA. The expected complex, however, showed a ϳ35-fold reduced K D of 591 nM compared with ϳ17 nM for rDPPIV⅐hADA. This is not due to conformational changes of ␣2, as superimposed backbone atoms of helices ␣1 and ␣2 of free mADA (PDB code 2ADA) with (hDPPIV⅐bADA) 2 yielded a root mean square deviation of 0.53 Å. Although this is not significant at the 3.0-Å resolution level, the differences in dissociation constants may be due to other substitutions in mADA compared with hADA. The (hDPPIV:bADA) 2 model shows that two nonconserved residues in the binding region of bADA, Gln 175 and Ser 131 , are located at ϳ5 Å to Asn 338 of hDPPIV and exchanged for charged Lys 175 and Asp 131 in mADA. This change could lead to far reaching electrostatic effects, thereby reducing the binding affinity in the complex formed by rDPPIV and chimeric mADA.
hDPPIV binds with a hydrophobic surface to ADA (Fig. 5). Compared with rodent DPPIV, which does not bind to endoge-nous ADA, four residues in the ADA binding site of hDPPIV are not conserved in any of the two rodent sequences, as depicted in Fig. 3. These are the bulky and hydrophobic residues Leu 294 , Ile 295 , and Val 341 in hDPPIV that are all replaced by Thr in rat and by Ala 294 , Arg 295 , and Thr 341 in mouse DPPIV, respectively. The positively charged Arg 336 in hDPPIV is replaced by Val 336 and Thr 336 in rat and mouse, respectively. The significance of Leu 294 or Val 341 for ADA binding affinity was shown by mutational and subsequent binding studies, which indicated that mutations Leu 294 3 Ala or Val 341 3 Ala in hDPPIV reduced binding affinity to bADA, and drastic replacements Leu 294 3 Arg and Val 341 3 Lys even abolished bADA binding. The mutation Ile 295 3 Ser showed no detectable effect (18), probably because Ile 295 is located at the outer interface region (Figs. 4 and 5). The introduction of these three bulky hydrophobic residues in hDPPIV increases the hydrophobicity of the interface (Fig. 5). This may indicate a directional shift to the evolution of strong ADA binding affinity to DPPIV in higher mammals contrasted to the loss of binding in rodents as proposed by Abbott et al. (18).
The fourth nonconserved residue in the ADA binding site of hDPPIV is Arg 336 , which is replaced for Val and Thr in rat and mouse DPPIV, respectively (Fig. 3). Arg 336 , the only charged residue in the hDPPIV interface, forms a salt bridge to bADA-Asp 127 (see Figs. 4 and 5) and might therefore improve binding affinity.
ADA Binding Would Interfere with Tetramer Formation of DPPIV-Because the (hDPPIV⅐bADA) 2 complex does not show significant structural changes compared with the isolated components hDPPIV and bADA, and their enzymatic activities are not affected (28), any modulation of a DPPIV-mediated signal transduction across the membrane that is induced by ADA binding may consequently be associated with yet unidentified sterical interference.
In this context, porcine DPPIV is of interest because it shares 88% amino acid sequence identity with hDPPIV and crystallized as a tetramer forming a "ʚ ʛ"-shaped complex (9). This is associated with the four ␤-strands Asn 279 -Gln 286 that form two antiparallel ␤-sheets, thus extending the propeller blades IV to two eight-stranded ␤-sheets, and the dimer-dimer interactions are augmented by the outer ␤-strands of blades V (9). These segments (indicated by violet arrows in Fig. 1) share a homologous sequence in hDPPIV (Fig. 3), and although Gln 286 is engaged in bADA binding (Fig. 4B), tetramer formation of DPPIV and binding of ADA are mutually exclusive. With this view, a recent study is explained that shows that hDPPIV promotes adhesion of lymphocytes to human epithelial cells, and that adhesion is significantly reduced by incubation with  (35). Contacting residues (distance cutoff, 3.9 Å) are assigned in one-letter code followed by sequence number. Positively and negatively charged residues are blue and red, unpolar and polar but uncharged residues are yellow and orange, respectively. The bold arrow indicates the access to the active site through the ␤-propeller opening of DPPIV and the thin curved double arrow indicates the salt bridge between hD-DPIV-Arg 336 and bADA-Asp 127 . The two parallel arrows indicate the location and orientation of ␣-helices ␣1 and ␣2 of bADA as depicted in Figs. 1 and 3. exogenous ADA (11). This suggests that ADA could regulate cell-cell adhesion mediated by DPPIV, although no evidence has yet been provided that DPPIV directly participates in cellcell adhesion.
The proposed key role of the glycosylation state of Asn 279 in preventing or enabling ADA binding (9), however, can be ruled out. This is because (hDPPIV⅐bADA) 2 shows electron density for two N-acetylglucosamine residues attached to each Asn 281 (corresponding to Asn 279 in the porcine sequence) at an ϳ15 Å distance to the bADA surface (indicated by an asterisk in Fig.  1B), which is too far to directly interfere with ADA binding.