Structure of a Tyrosine Phosphatase Adhesive Interaction Reveals a Spacer-Clamp Mechanism

See allHide authors and affiliations

Science  31 Aug 2007:
Vol. 317, Issue 5842, pp. 1217-1220
DOI: 10.1126/science.1144646


Cell-cell contacts are fundamental to multicellular organisms and are subject to exquisite levels of control. Human RPTPμ is a type IIB receptor protein tyrosine phosphatase that both forms an adhesive contact itself and is involved in regulating adhesion by dephosphorylating components of cadherin-catenin complexes. Here we describe a 3.1 angstrom crystal structure of the RPTPμ ectodomain that forms a homophilic trans (antiparallel) dimer with an extended and rigid architecture, matching the dimensions of adherens junctions. Cell surface expression of deletion constructs induces intercellular spacings that correlate with the ectodomain length. These data suggest that the RPTPμ ectodomain acts as a distance gauge and plays a key regulatory function, locking the phosphatase to its appropriate functional location.

The solid tissues of multicellular organisms are held together by interactions between cell adhesion molecules. These molecules can function as nucleation points for multiprotein assemblies that cluster at cell contacts and link to the cellular cytoskeleton. The opposing actions of protein tyrosine kinases and protein tyrosine phosphatases (PTPs) tune the level of tyrosine phosphorylation (13) to control the integrity of such assemblies. Type IIB receptor protein tyrosine phosphatases (RPTPs) combine cell adhesive and catalytic activities in one molecule (3) and hence are ideally equipped to act as initial sensors in phosphorylation-based signaling events. Homophilic (trans) interactions control the RPTP subcellular localization (4, 5) and are believed to modulate signaling, but a mechanistic understanding of this process has been lacking.

The type IIB RPTP family (6, 7) consists of four members: RPTPμ, RPTPρ, RPTPκ, and PCP2/RPTPλ. Their extracellular regions are predicted to share a common architecture, comprising six domains: one MAM (meprin/A5/μ) domain, one immunoglobulin (Ig)–like domain, and four fibronectin (FN) type III repeats (Fig. 1A). Our previously reported crystal structure of an N-terminal portion of RPTPμ revealed that the MAM and Ig domains form a structural unit (termed MIg) with a seamless interdomain interface (8) but could not explain the biological function of the molecule. Domains beyond MIg are required for cell adhesion (8), prompting us to attempt the structural solution of the full extracellular region.

Fig. 1.

Structure of the eRPTPμ monomer. (A) Ribbon diagram of eRPTPμ, in rainbow coloring (from N terminus in blue to C terminus in red). Sugars are depicted as slate-colored spheres. Modeled parts of the FN4 domain are colored in gray (but are not included in the structure refinement). Domains are indicated schematically at the left of the panel and, in RPTPμ, the ectodomain is followed by a transmembrane helix and two intracellular phosphatase domains. (B) Surface representation of eRPTPμ. Coloring is by residue conservation, based on an alignment of 23 sequences (shown in fig. S3A). (C to E) Close-up views of boxed interdomain junctions from (B). The Ig (magenta), FN1 (slate), FN2 (yellow), and FN3 (green) domains are shown as coils. Residues (35) involved in domain-domain interactions are drawn in stick representation (oxygen, red; nitrogen, blue; sulfur, lime). Potential hydrophilic interactions are marked as gray dotted lines.

The full-length human RPTPμ extracellular region (eRPTPμ) construct used for structural analysis consisted of residues Glu21 to Lys742 and contained 12 predicted N-glycosylation sites. To facilitate crystallization, we expressed the protein in glycosylation-impaired mammalian cells, and the resultant oligomannose-type glycans were further trimmed down to single N-acetylglucosamine moieties (9, 10). This protein crystallized in space group C2, with one molecule in the asymmetric unit. The structure was solved by molecular replacement and refined to a final Rwork of 25% (Rfree = 32%) with diffraction data between 20.0 and 3.1 Å resolution [see (11) and table S1]. The crystal structure reveals five well-ordered domains (Fig. 1A): all but the membrane-proximal FN repeat. The MIg structures solved in isolation (8) and in the context of the eRPTPμ are essentially identical [root mean square deviation (rmsd) = 1.0 Å for 255 Cα equivalents]; the only notable difference is a reorientation in the Ig domain of some 10° to optimize the interface with the first FN repeat (fig. S1, A and B). The three ordered FN repeats seen in the structure (termed FN1, FN2, and FN3; Fig. 1A) exhibit the classical FN type III topology. The electron density for the fourth FN domain, although sufficient to indicate its approximate position, remained largely disordered throughout refinement; therefore, this domain was not included in the final model (fig. S2, A and B). A pair of FN repeats in the cytoplasmic tail of integrin α6β4 shows particularly strong structural homology with the FN2-FN3 tandem of eRPTPμ [Protein Data Bank (PDB) entry 1QG3; rmsd = 2.1 Å for 170 Cα equivalents] (fig. S1, A and C). Although the sequence identity is only 24%, the residues in the α6β4 tandem FN domain interface and their equivalents in the eRPTPμ FN2-FN3 interface are highly conserved.

The five N-terminal domains of eRPTPμ form a rigid structure, with very short linkers and extensive interdomain interactions (Fig. 1, B to E). As previously reported, the MAM-Ig interface is spanned by a continuous β sheet (8). The Ig-FN1 interface is stabilized by a network of interactions involving primarily polar residues (Fig. 1C), whereas the FN1-FN2 and FN2-FN3 interfaces contain, in addition to multiple potential hydrogen bonds, a disulfide bond and hydrophobic interactions, respectively (Fig. 1, D and E). Typically, all residues involved in these interdomain interactions are strictly conserved within the vertebrate type IIB RPTPs (Fig. 1B and fig. S3, A and B). Thus, this rigid rodlike architecture appears to be a defining characteristic of type IIB RPTP ectodomains. However, specific points of flexion occur in other cell adhesion molecules and may be necessary to allow for limited adjustments in orientation to facilitate cell-cell interactions (12). The FN4 domains of RPTPμ and RPTPκ are proteolytically processed in the trans-Golgi network by subtilisin-like protein convertases (13, 14). This cleavage does not lead to release of the molecule from the cell surface (5, 13) but could introduce some flexibility in the membrane-proximal region of the ectodomain. A further point of flexion may be provided by residues linking the FN4 domain to the transmembrane region.

Type II RPTPs are homophilic cell adhesion molecules that form high-affinity dimers in solution (8, 15) and cause cell aggregation when expressed on the surface of normally nonadhesive cells (1620). The crystal packing of eRPTPμ reveals an extensive interaction surface between two molecules related by a crystallographic two-fold axis (Fig. 2A; buried surface per molecule equals 1630 Å2 for a probe radius of 1.4 Å). Residues from four domains (MAM, Ig, FN1, and FN2) contribute to this dimer interface (Fig. 2A), which is mainly hydrophilic, involving 14 potential hydrogen bonds, and which is further stabilized by 24 residues and a glycan per monomer, forming van der Waals contacts (Fig. 2, B to D). The main interaction surface is formed between the MAM and Ig domains from one molecule and the FN1 and FN2 domains from the other. To verify the biological importance of this interface, we performed site-directed mutagenesis of residues (marked by asterisks in Fig. 2, C and D) that appear to play a crucial role for its integrity and are highly conserved within the RPTP family. The oligomeric state of mutant proteins was assessed by size-exclusion chromatography (soluble ectodomain constructs; fig. S4) and in vivo cell adhesion assays (transmembrane constructs; Fig. 2E). Mutations of Arg239 and Arg240 to glutamate or Tyr297 and Trp299 to alanine abolished dimerization in both assays. In addition, a FN2 domain mutation (Arg409 to glutamate) impaired dimerization in the chromatography assay and completely abolished cell adhesion. This finding concurs with previously published serial deletion data that defined MAM-Ig-FN1-FN2 as the minimal unit required for cell adhesion (8).

Fig. 2.

eRPTPμ dimerization. (A) Ribbon diagram of the eRPTPμ dimer. The solvent-accessible surface is drawn in light gray, and the domains appear in blue (MAM), magenta (Ig), slate (FN1), yellow (FN2), green (FN3), and gray (FN4). The asterisk marks the crystallographic twofold axis. (B) Electrostatic properties. One monomer is shown as a solvent-accessible surface colored by electrostatic potential contoured at ±10 kT (red, acidic; blue, basic), and the other monomer is shown as a black ribbon. (C) The dimer interface. MAM and Ig domains of one molecule interact with FN1 and FN2 domains of another molecule. Domains are colored as in (A). Residues involved in dimer interactions are drawn in stick representation (oxygen, red; nitrogen, blue). Potential hydrophilic interactions are marked as gray dotted lines. Asterisks mark residues targeted for mutagenesis. (D) Hydrophobic interactions. Color coding is as in (C), and the N92-linked sugar is colored in green and forms stacking interactions with the indole ring of W151. (E) Cell adhesion assays. Non adherent insect Sf9 cells were infected with baculovirus constructs expressing either enhanced green fluorescent protein (EGFP) alone or RPTPμ-EGFP fusion constructs, wild type and mutant, and observed by phase contrast (top row) and fluorescence (bottom row) microscopy. Formation of aggregates indicates RPTPμ ectodomain adhesive function (8).

Type IIB RPTP ectodomains are highly conserved across species and between family members (typically 50 to 60% amino acid identity; fig. S3). Despite this level of sequence conservation, the homophilic interactions of RPTPμ and RPTPκ show strict specificity in cellular assays (19). Such behavior suggests an in vivo sorting function for these cell adhesion molecules, as described for cadherins (2): a conclusion reinforced by the generally distinct and complementary expression patterns observed during embryonic development (7, 21). The eRPTPμ structure, when coupled with a sequence alignment between RPTP ectodomains, reveals a strictly conserved scaffold of core interactions (Fig. 2, C to E, and fig. S3A), whereas peripheral areas of the dimer interface show substantial variability; these differences are likely to have evolved to provide the specificity for homophilic segregation (fig. S5A). The high level of sequence conservation also allows us to map a series of RPTPρ mutations, found for an analysis of PTPs in human colorectal cancers (22), onto the eRPTPμ structure. Although the role of these mutations in tumorigenesis requires investigation, it is clear that one group of them can be predicted to disrupt the MAM or Ig domain folds, whereas a second set may weaken the interdomain junctions and hence perturb the rigidity of the ectodomain (fig. S5B).

Type IIB RPTP substrates are components of the cadherin-catenin complexes at cell contacts (3, 2325), and RPTPμ is known to associate with multiple cadherins (N, E, R, VE) (26, 27), modulating their adhesive properties. The ectodomains of the cadherins also form trans dimers, and our structural analysis reveals a match between the dimensions of the RPTPμ trans dimer and cadherin-mediated cell junctions (2831). Cadherins mediate intercellular contacts in organisms ranging from ascidians to humans (2). In vertebrates, cadherin-based interactions need to vary in stability from the relatively static contacts in epithelia or vascular endothelia to the more dynamic ones required by neuronal growth cones. Adhesion-sensor molecules, which selectively localize to cadherin-mediated cell contacts, would provide a mechanism by which to modulate the stability of these regions. If the dimensions of the RPTPμ trans dimers are rigidly set to correspond to a particular cell-cell spacing, this adhesion interaction has the potential to act as a spacer clamp, locking the phosphatase activity at the adherens junction.

To test the ability of the RPTPμ trans dimer to function as a spacer clamp, we expressed a series of transmembrane constructs in which the length of the ectodomain was serially reduced (by deletion of FN4 and FN3-FN4) but the adhesive interaction was conserved (11). Electron microscopy (EM) of immunolabeled cryosections confirmed that, in this system, the RPTPμ molecules preferentially located to cell-cell interfaces (Fig. 3, A and B). Measurement of the cell-cell interfaces revealed that expression of each construct resulted in a distinct spacing (Fig. 3, C to E, and fig. S6). Furthermore, these distinctive cell-cell spacings correlated with the number of domains in the extracellular region construct and hence with the length of the trans dimer.

Fig. 3.

Ectodomain length of RPTPμ controls intermembrane spacing at homotypic adhesive interfaces. (A) RPTPμ accumulation at adhesive interfaces revealed by immuno-EM analysis of cell aggregate cryosections (original magnification, ×23,000). (B) Close-up view of an interface region from (A), showing the membrane bilayers (original magnification, ×49,000). (C to E) Transmembrane RPTPμ constructs induce extensive cell contact regions with aligned plasma membranes (original magnification, ×49,000). The intermembrane distances become progressively shorter as RPTPμ domains are deleted. Measurements (taken at regions of the interfaces where membranes were parallel and clearly defined) reveal the following distances between the extracellular borders of the outer leaflets: 23.7 ± 3.0 nm [n = 115 measurements, from 26 interfaces, one shown in panel (C)], 17.2 ± 1.9 nm [n = 159 measurements, from 32 interfaces, one shown in panel (D)], and 13.1 ± 3.1 nm [n = 114 measurements, from 29 interfaces, one shown in panel (E)]. Intermembrane-distance frequency histograms are provided in fig. S6. Thus, a decrease in cell-cell spacing of ∼6 nm between panels (C) and (D) corresponds to the deletion of one FN domain plus the tail region, and a decrease of ∼4 nm between panels (D) and (E) corresponds to the deletion of one FN domain, per RPTPμ monomer. The observed full-length RPTPμ-induced intermembrane distance is in agreement with published data on similarly treated samples of cadherin-mediated cell junctions (30, 31).

Alongside the above observations, several lines of evidence converge to support the importance of size and homophilic adhesive properties for the ectodomain-regulated localization and function of RPTPμ. At low cell density, RPTPμ has an even distribution over the cell surface, whereas in confluent cultures, surface expression increases substantially (threefold) and localization is restricted to cell-cell contacts (4, 5). Because this increase is not due to the up-regulation of gene expression, the protein appears to be “trapped” at cell contacts through homophilic binding (5). Del Vecchio and Tonks (4) have shown that, in confluent bovine aortic endothelial and Madin-Darby canine kidney cell cultures, RPTPμ is strictly co-localized with cadherins at intercellular contacts and excluded from the narrow spacings of the tight junctions. Moreover, a RPTPμ construct lacking the Ig domain, and therefore unable to establish homophilic interactions, has a diffuse surface expression pattern even in confluent cultures (4). This demonstrates that the intracellular region, despite its ability to interact with cadherins, is not sufficient to colocalize these proteins at the cell junctions. The PTPs generally have little substrate specificity, and they rely on noncatalytic domains to control their subcellular distribution and therefore indirectly regulate their activity by restricting access to particular substrates at defined locations (6, 32). RPTPs are known to be constitutively active, and ligand-induced inactivation has been reported for type I and IV subfamilies. Such a mechanism is unlikely to apply to type IIB RPTPs, where an active enzyme would be required to maintain cadherin-catenin complexes in a dephosphorylated state and thus contribute to the stability of cell contacts (2). In this context, for type IIB RPTPs, the ectodomain-mediated trans homophilic interactions appear to represent the driving force for correct localization and function.

Our results on RPTPμ suggest how the type IIB RPTPs modulate the stability of adherens junctions (Fig. 4). The ectodomain trans interaction is switchedoff at acidpH(8, 18) (i.e., until RPTPμ reaches the cell surface). The rigid, rulerlike ectodomain then acts as a sensor of intercellular distances, matching cadherin-mediated cell contacts, at which point the trans interaction serves as a spacer clamp, locking the phosphatase activity into proximity with the target substrates. The spacer-clamp action of RPTPμ represents the inverse strategy to the size-exclusion mechanism proposed to regulate the cell surface location of another RPTP, CD45; in that case, the mismatch between the RPTP ectodomain and the intercellular spacing is thought to contribute to T cell signaling by expelling the phosphatase activity from local zones of cell-cell contact (33, 34). Unlike CD45, RPTPμ is maintained at cell contacts, potentially increasing the local level of phosphatase activity. Because of the high affinity of the trans interaction, the balance between cell adhesion versus mobility can only be shifted by the action of the ADAM 10 protease (14). In both CD45 and RPTPμ, however, ectodomain size and rigidity appear to provide a mechanism to allow cell-cell spacings to regulate intercellular multi-molecular assemblies.

Fig. 4.

Model of adhesion-regulated RPTPμ signaling. Cadherins [ectodomains shown in orange, PDB entry 1L3W (29)] establish intercellular contacts via trans interactions, as well as cis interactions (black arrow) (2, 29). RPTPμ (shown in blue) trans interactions are pH sensitive (8, 18), which is consistent with the polar nature of the interface, and therefore cannot form at the low pH of the secretory pathway. Cell surface RPTPμ molecules rapidly recirculate, unless there is an appropriate recognition match (5). Trans RPTPμ dimerization may be complemented by weak interactions in cis (black arrow and question mark) (8, 15). RPTPμ can stabilize the cadherin-catenin complex [drawn schematically: α-catenin (yellow circles), β-catenin (light green ovals), and p120-catenin (dark green ovals)] by dephosphorylation (3)(red arrows). Type IIB RPTPs are processed in multiple proteolytic steps (5, 13, 14). Protein convertases (in the trans-Golgi network) nick the FN4 domain (13, 14), potentially contributing flexibility. ADAM 10 cleaves close to the membrane (thick gray lines), causing the shedding of RPTPμ (5, 14) and cadherin (36) ectodomains. Subsequent γ-secretase–dependent intramembrane cleavage releases the RPTPμ intracellular region (blue ovals) (14). The cadherin and RPTPμ ectodomains (crystal structures drawn to the same scale) are shown perpendicular to the cell surface to simplify the figure. EM analysis of adherens junctions and desmosomes has revealed the possibility of non-orthogonal orientations with respect to the membrane surface [with variable tilt angles (28, 31)], but it is not clear to what extent this is caused by sample preparation procedures or flexibility of the juxtamembrane regions.

Supporting Online Material

Materials and Methods

Figs. S1 to S6

Table S1


References and Notes

View Abstract

Stay Connected to Science

Navigate This Article