Architecture of Eph receptor clusters

Abstract

Eph receptor tyrosine kinases and their ephrin ligands regulate cell navigation during normal and oncogenic development. Signaling of Ephs is initiated in a multistep process leading to the assembly of higher-order signaling clusters that set off bidirectional signaling in interacting cells. However, the structural and mechanistic details of this assembly remained undefined. Here we present high-resolution structures of the complete EphA2 ectodomain and complexes with ephrin-A1 and A5 as the base unit of an Eph cluster. The structures reveal an elongated architecture with novel Eph/Eph interactions, both within and outside of the Eph ligand-binding domain, that suggest the molecular mechanism underlying Eph/ephrin clustering. Structure-function analysis, by using site-directed mutagenesis and cell-based signaling assays, confirms the importance of the identified oligomerization interfaces for Eph clustering.

Eph receptors and their ephrin ligands control a diverse array of cell–cell interactions during patterning of the nervous, skeletal, and vascular systems (1, 2). Upon ephrin binding, the Eph kinase initiates “forward” signaling into receptor-expressing cells, and the ephrin cytoplasmic tail triggers “reverse” signaling into ligand-expressing cells. Ephs and ephrins are divided into two subclasses (A and B) on the basis of their affinities for each other. With some exceptions, EphA receptors (EphA1–A10) bind to A-class ephrins (ephrin-A1–A6), whereas EphB receptors (EphB1–B6) interact with the B-subclass ephrins (ephrin-B1–B3). Given that Ephs and ephrins are membrane-bound, their interaction occurs only at sites of cell–cell contact. In the absence of cell–cell interactions, they exist in loosely associated microdomains, which become more compact and well-ordered when Eph/ephrin complexes assemble to generate clearly defined signaling centers (2).

The extracellular Eph region contains a conserved N-terminal ligand-binding domain (LBD), an adjacent cysteine-rich domain (CRD) (3), followed by two fibronectin repeats (FN3) (Fig. S1). The cytoplasmic Eph region encompasses a regulatory juxtamembrane region connecting the kinase domain, a sterile α motif domain, and a postsynaptic density protein (PSD95), Drosophila disc large tumor suppressor (DlgA), and zonula occludens-1 protein (zo-1) binding motif. All ephrins possess a 20 KDa extracellular receptor-binding domain; B-type ephrins also contain a short cytoplasmic region.

Several crystal structures of complexes between the minimal Eph and ephrin-binding Eph domains have been reported (reviewed in ref. 4). The crystal structure of the complexed EphB2 and ephrin-B2 binding domains revealed two contact surfaces that are involved in the assembly of Eph/ephrin tetramers: an expansive high-affinity ephrin-binding channel, likely responsible for the initial interaction, and a smaller interface that mediates a lower affinity Eph-ephrin contact with an adjacent ephrin molecule (4).

In addition to these structurally defined ligand–receptor interactions, several observations revealed that additional protein interfaces are important for the generation and function of Eph/ephrin signaling centers at points of cell-cell contact: First, the EphB2-ephrinB2 crystal structure suggests a propensity of individual Eph/ephrin complexes to assemble, via direct Eph/Eph contacts, into heterooligomeric clusters. Furthermore, earlier findings indicated that Eph-Eph interactions via a C-terminal ectodomain region outside the LBD are critical for Eph function during development (3). Presumably such Eph-Eph interfaces act to recruit non-ligand-bound Eph receptors into Eph clusters (5). In addition, a random mutagenesis survey of the EphA3 ectodomain revealed that binding to ephrinA5 requires an interaction site located in the CRD. Although having only a modest contribution to ligand-binding affinity, mutation of this region severely effected receptor phosphorylation and recruitment of signaling molecules (6). Last, a recent report suggested that ephrin-A5 may also interact via the EphA3 fibronectin III repeats (7). To determine the structural basis of Eph clustering and explore underlying receptor–receptor and receptor–ligand interactions, we determined structures of the complete EphA2 ectodomain, alone and in complexes with its cognate ligands ephrin-A5 and ephrin-A1, the latter of which was recently used to elaborate spatiomechanical concepts related to EphA2 clustering (8). These crystal structures, supported by cell-based functional studies, show that the CRD mediates Eph/Eph interactions in the assembly of signaling-competent EphA2/ephrin clusters.

Results and Discussion

Overall Structures.

We crystallized and determined protein structures comprising the whole—or parts of the—EphA2 extracellular region, either in their apo forms (complete EphA2 ectodomain and LBD) or bound to their ephrin ligands ephrin-A1 or -A5 (LBD, LBD-CRD, and LBD-CRD-nFN3, Table 1). The structure of a full-length Ephrin receptor ectodomain (Fig. 1A) reveals an extended structure of tightly packed domains spanning 146 × 52 × 55 Å (Fig. S2).

Table 1.

Data collection and refinement statistics

Dataset	LBD	LBD + EphrinA1	LBD-CRD + EphrinA1	LBD-CRD-nFN3 + EphrinA5	LBD-CRD-nFN3-cFN3
EphA2 residues	23-202	23-202	23-326	27-435*	23-531
Expression host	Sf9	Sf9	Sf9	HEK293	Sf9
Crystallization buffer	25% PEG 3350, 0.1 M Ammonium Sulfate, 0.1 M Bistris pH 5.5	14.9% PEG 4000, 0.1 M Na Citrate pH 5.6, 20% Isopropanol	10% PEG 3350, 0.16 M Ammonium Phosphate	20% PEG 3350, 200 mM Ammonium Citrate, 0.1 M Na Citrate pH 5.8	3% PEG 4 K, 0.1 M NaAc, 0.1 M CaCo pH 5.5 with 0.5 M NDSB 256
PDB ID code	3C8X	3CZU	3MBW	3MX0	3FL7
Space group	P 31 2 1	P 65 2 2	C 2 2 21	P21	P 21 21 2
Unit cell (a, b, c, α, ß, γ)	92.518, 92.518, 41.291, 90, 90, 120	99.431, 99.431, 204.879, 90, 90, 120	58.265, 215.787, 107.263, 90, 90, 90	57.859, 89.049, 198.150, 90, 96.22, 90	59.358, 89.992, 136.476, 90, 90, 90
Beamline	RIGAKU FR-E	RIGAKU FR-E	APS SBC-CAT 19 ID	APS NE-CAT 24-ID-C	APS GMCA-CAT 23-ID-B
Wavelength	1.54178	1.54178	0.98792	0.98	0.97948
Resolution	35.0-1.95	25.0-2.65	39.0-2.80	45.0-3.50	41.0-2.50
Unique reflections	14,579	18,122	16,953	23,129	25,966
Data redundancy†	6.8 (5.0)	20.8 (21.2)	5.3 (5.3)	5.5 (5.2)	4.1 (4.0)
Completeness	99.9 (98.6)	100 (100)	99.7 (100)	91.4 (71.3)	99.7% (98.1%)
I/σI	15.44 (1.828)	35.74 (5.05)	23.65 (2.17)	21.6 (6.2)	22.43 (2.94)
Rsym	0.151 (0.849)	0.109 (0.714)	0.066 (0.668)	0.11 (0.22)	0.070 (0.375)
Rp.i.m. and Rmeas/Rr.i.m.	0.061 (0.415)	0.025 (0.158)	0.033 (0.333)
Refinement
Resolution	35.0-1.95	24.86-2.65	34.10-2.81	45.0-3.50	40.59-2.50
Reflections used	14,205	17,120	16,891	22,383	42,331
All atoms (hetero, solvent)	1,394 (0,114)	26.03 (39, 73)	3,305 (90,28)	8,380	3,741 (18, 46)
Rwork/Rfree‡	16.3/22.6	19.3/22.5	22.5/27.0	24.6/29.4 (30.2/39.7)	24.6/29.9
rmsd bond length	0.016	0.009	0.009	0.006	0.009
rmsd bond angle	1.52	1.23	1.15	1.10	1.24
Mean B factor	24.22	42.45	79.92	138.34	35.47
Ramachandran plot
Favored	96.8	95.7	96.5	79.4	93.6
Allowed	100	100	100	94.6	100
Disallowed	0	0	0	58	0

^*Ordered and modeled.

^†Highest resolution shell shown in parenthesis.

^‡R_free calculated with 5% of the data.

Fig. 1.

EphA2/ephrin-A1 (5) structures, structural alignment, and ligand binding. (A) Left, backbone representation of superimposed structures. Right, all structures are shown from the same perspective in ribbon format. Each structure is labeled according to its PDB code and differentially colored. (B) The EphA2-EphrinA1 high-affinity heterodimer interface. Stereoscopic view of the interface with domains labeled and shown in ribbon format. Ligand residues that are within 4 Å of the LBD are shown as sticks; LBD residues as lines. Water molecules in the vicinity are shown as red spheres, hydrogen bonds as black dashed lines.

The fold and atomic details of the Eph CRD, encompassing ∼114 amino acids (residues 201–314) immediately C-terminal to the LBD, can be subdivided into two domains, an N-terminal domain similar to complement control module/short complement regulator domain, and a TNF receptor like CRD. The N-terminal half (201–260) includes five antiparallel β-strands arranged as a β-sandwich, whereby the first residue of the fold (Cys201) is disulfide-anchored to the end of the fourth β-strand (Fig. S3). A Dali search (9) reveals the top Protein Data Bank (PDB) hit with this domain is that of 2Z3R with a Z score of 4.5, an rmsd of 2.7 Å over 56 C^α atoms, with 13% sequence identity, including the cysteines involved in the disulphide bridges matching the Eph sequences 201–247 and 230–260. The C-terminal half of the Eph CRD comprises two β-strands and six tightly packed random coils, including four disulfide bridges, the first of which anchors the first residue of this half to the sixth CRD β-strand. It resembles the TNF receptor CRD, closely matching the Death Receptor 5 (PDB ID 2H9G) with a Z score of 4.9, an rmsd of 2.4 over 46 C^α atoms, and 17% sequence identity, including conservation of three disulphide bridges (262–273, 276–290, and 293–307). The two CRD halves are tightly packed against each other, with the N- and C-terminal residues of the CRD occurring on opposite sides of the long axis of the domain. Apart from some negatively charged patches, the CRD surface is predominantly neutral.

The N-terminal fibronectin-type-3 domain (nFN3) adopts a typical immunoglobulin-like fold (Fig. 1A and Fig. S4A), most closely homologous to Integrin Beta-4 (1QG3) and Plectin-1 (3F7P) FN3 (rmsd ∼2.0 Å over 91 C^α atoms, DALI search). Of note, lack of significant binding clefts at either ends of the domain suggests that nFN3 does not bind small-molecule ligands (Fig. S4B). Although cFN3 has the same topology as nFN3, it is structurally distinct with an rmsd of 7.7 Å extending over 68 C^α atoms (Fig. S4A): Its β3-β4 and β5-β6 loops are split open to reveal an aromatic-lined cleft that might represent a membrane-surface binding pocket (Fig. S4B). Structural homologs to the cFN3 include Neural Cell Adhesion Molecule 2, Fibronectin, and Tenascin, with rmsd values from 1.6 to 2.0 over 82 C^α atoms.

The elongated architecture of EphA2 is stabilized by extensive interdomain interactions. The first and last LBD domain residues (25–27 and 199–200) together with the β5-β6 loop form an interaction surface with the CRD, which buries 1,211 Å² surface area and is stabilized by six hydrogen or salt bonds (Fig. S5A). Likewise, a buried 701 Å² interface and a salt bridge stabilizes the CRD–nFN3 interaction (Fig. S5B). The association between nFN3 and cFN3 seems more flexible and, apart from a single hydrogen bond, not stabilized by buried protein surfaces.

High-Affinity Eph/Ephrin Heterodimer.

Functional Eph/ephrin signaling clusters assemble from high-affinity Eph/ephrin heterodimers, which aggregate into heterotetramers and higher-order oligomers (1). The two EphA2/ephrin-A1 and one EphA2/ephrin-A5 complexes elucidated in our study show strikingly similar structural arrangements (Fig. 1A, alignment): The two EphA2/ephrin-A5 heterodimers in the asymmetric unit of the corresponding crystals differ only by an rmsd of 0.3 Å between equivalent C^α positions. As expected, high-affinity EphA2/ephrin-A1/5 interactions involve only the N-terminal globular LBD (Fig. 1B). Overall, the EphA2/ephrin-A1/5 heterodimers are very similar to known Eph/ephrin structures involving only the Eph LBD (4, 9). The high-affinity ligand/receptor interface centers around the G-H loop of ephrin-A1 or -A5, which is inserted in a channel on the surface of EphA2 (Fig. 1B and Fig. S6). Four antiparallel β-strands define the two sides of the channel and two strands line its back. The ligand binds by attaching the side of its β-sandwich to the outside surface of the channel and inserting its long G-H loop into the channel, which then becomes buttressed by a receptor loop closing in from the top. The binding is dominated by van der Waals contacts between two predominantly hydrophobic surfaces, because the ligand buries Gln109, Phe111, Thr112, Pro113, Phe114, Thr115, Leu116, and Gly117 (Fig. S6). Gln109 interacts not only with the sides of the channel but also with Phe100 and Pro101 from the long EphA2 loop at the top of the interface. Pro113 is in direct contact with the Cys70-Cys188 disulfide bridge in EphA2. Adjacent to the channel/G-H-loop interactions, a second, structurally separate, contact area encompasses the ephrin-A1/5 docking site along the upper surface of the receptor. Here the ephrin β-sandwich interacts via a network of hydrogen bonds and salt bridges (Eph-ephrin: Arg103-Glu119; Arg159-Asp86; Asp53-Lys107) (Fig. S6).

Comparisons of Bound and Unbound EphA2-Ectodomain (ECD) and EphrinA5.

Interestingly, the overall structure of the ephrin-bound EphA2-ECD is very similar to that of the unbound protein, with an rmsd between equivalent C^α positions of 0.9 Å (Fig. 1A). Indeed, the most significant conformational changes involve loops within the ephrin-binding interface. The fact that there is little conformational differences in the various crystal lattices implies a very rigid rod-like architecture of the Eph ectodomain, at least in the region encompassing the LBD, CRD, and nFN3, which is not modulated by ephrin binding. Likewise, ephrin-A5/1 does not undergo significant structural rearrangements upon EphA2 binding and can be superimposed onto the structure of the unbound molecule with rmsd between equivalent C^α positions of ∼0.4 Å. The only significant conformational changes upon complex formation involve the rearrangement of the Eph-binding (G-H) loop, which becomes structurally complementary to the ephrin-binding channel on the Eph-LBD surface.

Eph/Ephrin Heterotetramers.

The full EphA2/ephrin-A5 ectodomain complex forms a heterotetramer in solution (Fig. S7), and crystal packing reveals two potential Eph/ephrin heterotetrameric assemblies (Fig. 2). These two heterotetramers are generated via two distinct EphA2/EphA2 interfaces and, when combined, would generate a continuous Eph/ephrin assembly (Fig. 2A).

Fig. 2.

Eph/ephrin assemblies. Left, ribbon diagram of four molecules each of receptor and ligand are shown with receptors colored differently for clarity. Right, two assemblies are shown; the first (“heterotetramerization assembly”) is mediated only by the LBD, the second (“clustering assembly”) is mediated by both LBD and CRD. This figure was generated from the EphA2/ephrin A5 complex but also applies to the other, ephrin-A1, complex structures.

The first of these heterotetramers is generated by Eph-Eph and Eph-ephrin interactions that encompass only Eph residues within the LBD. Indeed this Eph-LBD/ephrin heterotetramer is also observed in all our complex structures (Figs. 2B and 3). Architecturally, these EphA2-LBD/ephrinA1/5 complexes are similar to the circular heterotetramers observed in the EphB2-LBD/ephrin-B2 structure (10), although the precise interfaces and interactions are quite distinct. Indeed, whereas Eph-Eph contacts are not observed in the EphB2-LBD/ephrin-B2 tetramer, both Eph/ephrin and Eph/Eph interactions facilitate EphA2/ephrinA1/5 tetramers (Fig. 3). Moreover, ephrin glycosylation units may also contribute to this interaction (Fig. 3C and PDB ID code 3MBW). The total surface area buried in the EphA2/ephrinA1/5 heterotetramers (2,088 Å²) is somewhat smaller than the surface area buried in the EphB2/ephrin-B2 heterotetramers (2,532 Å²). Because the presence of Eph/ephrin contacts suggests that the formation of this heterotetrameric complex is dependent on ephrin binding, and because the architecturally similar B-class assembly has been referred to as “low affinity Eph/ephrin heterotetramers” (1), we will also refer to these EphA2/ephrinA1/5 and EphA2/EphA2 interfaces simply as the “heterodimerization” interfaces. This assembly is seen in all three of our EphA2/ephrin structures, even in the absence of the CRD domain.

Fig. 3.

Eph/ephrin heterotetramerization assembly. (A) Ribbon and (B) surface representations of the LBD-ephrin-A5 heterotetramer with each domain colored differently. (C) Detailed stereoscopic view of the 2∶2 assembly.

The second heterotetrameric assembly (Fig. 2B) is generated only via Eph-Eph interactions, suggesting that its formation is not dependent on ephrin binding. In fact, these interactions are conserved in all CRD-containing structures. We will therefore refer to these Eph/Eph interactions and interfaces as “clustering.” The EphA2/ephrin clustering interactions involve two distinct Eph/Eph interfaces—one in the LBD (Fig. 4 A and B) and one in the CRD (Fig. 4C).

Fig. 4.

Clustering assembly. (A) Ribbon display of the clustering assembly with each EphA2 molecule colored differently; (B) detailed stereoscopic view of the LBD-mediated clustering interaction. Middle, ribbon diagram of two molecules of the unbound EphA2—dimer in light orange and pale green. (C) Detailed view of CRD-mediated clustering. Residues that mediate clustering are shown in stick format and labeled. Dashes represent van der Waals or hydrogen bond interactions.

The LBD-mediated clustering interface is almost entirely polar, involving several salt bridges and hydrogen bonds (Fig. 4B). At the center of this Eph-Eph interface Lys116 from one of the Eph molecules makes a salt bridge with Glu117 and Asp104 from the other, whereas the side chain of Thr144 hydrogen-bonds with the main chain carbonyl of Pro147. Because the interface is twofold symmetric, the reverse salt bridge and hydrogen bond are also present (Fig. 4). The CRD-mediated clustering interface involves a leucine-zipper-like assembly with a large number of van der Waals interactions. The hydrophobic zipper is formed by residues Pro221, Leu223, Leu254, Val255, and Ile257 (Fig. 4C). Upon formation of the clustering interface, approximately 850 Å² of surface area is buried in each Eph molecule—large enough for this interface to be considered biologically relevant. Moreover, the interaction surfaces are composed of conserved residues across human Eph receptors, further underlying its potential functional significance (Fig. S8).

Interestingly, both the Eph-Eph heterodimerization and clustering interfaces are present in the free Eph-ECD crystals and in all complex structures (Eph-LBD-CRD-nFN3/ephrin-A5 and Eph-LBD-CRD/ephrin-A1). This observation suggests high affinity and that the continuous Eph-ECD/Eph-ECD assemblies are formed independent of ephrin binding at high enough Eph concentrations.

Role of the Eph/Eph Interfaces for Stability and Function of Eph/Ephrin Signaling Clusters.

To compare the relative contributions of receptor–receptor interactions within the LBD and CRD of EphA2, we designed GFP-tagged EphA2 deletion mutants lacking either domain: ΔLBD is truncated between residues 28–198, and in ΔCRD residues 201–325 are replaced by a Gly-Ser-Gly-Ser linker. We tested these mutants functionally, by transfecting mutant or WT EphA2-GFP cDNAs into HEK293 cells and analyzed their capacity to support ligand-independent Eph kinase activation that is induced upon transient overexpression. Immunoblot analysis of anti-GFP immunoprecipitated receptors demonstrated markedly reduced relative phosphorylation levels of both deletion mutants compared to WT EphA2, confirming the involvement of both domains in Eph-Eph clustering (Fig. 5A). Importantly, ligand-independent activation of the ΔCRD mutant was most strongly affected, confirming its critical role in Eph signaling initiation as was suggested previously for EphA3 (3, 5).

Fig. 5.

Cellular studies. (A) Immunoprecipitates from HEK293 cells transfected with increasing amounts of GFP-tagged WT EphA2, ΔLBD-EphA2, or ΔCRD-EphA2 were immunoblotted with α-EphA2 and α-phosphotyrosine antibodies. Densitometry quantified EphA2 phosphorylation relative to EphA2 expression, by using samples with most similar EphA2 levels (lanes 2, 5, and 8). (B) Activation of WT and mutated EphA2 in HEK293 cells by preclustered ephrin-A5. Single = L223R, double = L223R,L254R, and triple = L223R,L254R,V255R. (C) HEK293 cells stably expressing WT or point mutated EphA2, or control cells, were transfected with ΔLBD-EphA2-GFP (ΔLBD) and stimulated with clustered ephrinA5-Fc. Protein A sepharose pull-downs of ephrinA5-Fc associated receptors were Western blotted with α-EphA2 and α-GFP antibodies. The graph shows the amount of ΔLBD pulled down (anti-GFP blot) via association with full-length EphA2, relative to full-length EphA2, quantified by densitometry. (D) Parental HEK293 cells or derived clones stably expressing WT EphA2 or EphA2 point mutants as indicated were transfected with ΔLBD-EphA2-GFP. Alexa594-ephrinA5 conjugated Dynabeads were added to the cells for 5 min before cultures were rinsed and fixed for microscopy. Panels show representative images of EphA2-GFP and Alexa594-ephrinA5 fluorescence, white arrows indicating beads with recruited EphA2-GFP, yellow arrows indicating beads in contact with cells lacking recruitment. Insets show higher magnification images of boxed regions. The average proportion of beads in contact with cells that recruited ΔLBD-EphA2-GFP was determined for each cell line (WT EphA2, triple or double mutant or control cells) and is shown in the graph (± SEM).

To evaluate the contribution of ligand-independent clustering to EphA2 activation, we interrogated EphA2 point mutants in a cell-based EphA2 phosphorylation assay. The mutants were designed to substitute hydrophobic residues in the leucine-zipper-like clustering interface predicted from the crystal structure with positively charged ones. They included Arg substitutions at positions Leu223 (single mutation), Leu223 and Leu254 (double mutation), or Leu223, Leu254, and Val255 (triple mutation). Analysis of HEK293T cell clones stably expressing WT or mutant EphA2 by flow cytometry confirmed that all of these exogenous receptors were expressed on the cell surface at similar levels and were capable of ephrin-A5 binding (Fig. S9). Anti-phospho-tyrosine Western blot analysis of ephrin-A5-Fc stimulated cells revealed significantly reduced activation of EphA2 mutant compared to WT-EphA2: As expected, double and triple substitutions affected activation stronger than single substitutions (Fig. 5B). In contrast, substitutions of charged residues within the LBD region of the clustering interface did not affect ephrin-induced EphA2 phosphorylation, suggesting that increased hydrophobicity of the interface may compensate for loss of the salt bridge caused by these mutations (Fig. 5B). Together, these findings in live cells confirm the relevance of the Eph–Eph interactions observed in the crystal structures for the formation of functional signaling clusters.

To directly assess the effect of EphA2 point mutations on clustering via the CRD but in the absence of LBD-mediated interactions, we transfected HEK293T cell clones, expressing WT or L-R-mutated EphA2 with GFP-tagged EphA2 lacking the ligand-binding domain (ΔLBD-EphA2-GFP). This “reporter” allowed us to monitor coclustering via its CRD with ephrin-A5-bound WT EphA2 or with the L-R-substitution mutants. Anti-GFP Western blot analysis of ephrin-bound EphA2 demonstrated that the relative level of coprecipitated GFP-tagged reporter was notably reduced in cells expressing L-R-substituted EphA2 receptors, indicating that CRD interface point mutations, most prominently the triple Arg substitution, disrupt the ability for CRD-mediated clustering (Fig. 5C).

We validated the functional importance of these findings by analyzing with confocal fluorescence microscopy in HEK293T cell clones the recruitment of ephrin-binding-compromised ΔLBD-EphA2-GFP to full-length WT or mutant receptors. Localized recruitment and clustering of ΔLBD-EphA2-GFP to Alexa⁵⁹⁴ephrinA5-coated beads added to the cells was discernible from GFP fluorescence marking the outline of the beads: Thus, in cells expressing full-length WT EphA2, but not in control (293) cells, bead-associated GFP fluorescence confirmed a robust Eph/Eph interaction via the intact CRD (Fig. 5D). By comparison, cells expressing triple L-R EphA2 point mutants and, to a lesser extent, double Arg substitutions revealed significantly reduced GFP fluorescence around the ephrin-coated beads (Fig. 5D), confirming that residues 223, 254, and 255 of EphA2 mediate receptor–receptor interactions and facilitate ephrin-independent receptor clustering in intact cells.

Implications for Eph Signaling Initiation

Ligand-induced activation of receptor tyrosine kinases (RTKs) is a tightly regulated process that is not yet well understood at the molecular level. It seems likely that the different RTK families utilize different molecular mechanisms for coupling ligand binding with activation of the catalytic kinase domain. The receptor structures reported here reveal that the extracellular region of Eph RTKs are composed of four individual structural domains that together fold into a unique rod-like structure. The Eph ectodomain is rigid, and ligand binding does not cause significant conformational changes in any of the individual domains or significant structural rearrangements between them. Thus, ligand-induced conformational changes in the receptor extracellular domain do not seem to be the molecular mechanism driving Eph signal transduction.

Our structures reveal that the Eph CRD is a unique protein-interaction/dimerization module, which cooperates with ligand-dependent clustering to mediate the assembly of continuous oligomers, a process that could also happen independently of ligand binding at high receptor concentrations. We confirmed this concept by structure-based mutagenesis in combination with cell-based signaling and receptor-visualization assays, also explaining the previously observed recruitment of non-ligand-bound receptors into signaling clusters (5). The presence at the cell surface of highly ordered receptor assemblies is a unique feature of the Eph receptors and has not been observed in any other receptor kinase family.

Another unique characteristic of Eph/ephrin signaling is the dependence of Eph activation and downstream signalling on membrane-attached and preculstered ephrin ligands. Continuous Eph/Eph and Eph/ephrin assemblies in our crystals therefore suggest that the function of ephrin ligands might be to increase local receptor concentration so that ordered Eph/ephrin assemblies can be formed on the cell surface. Indeed, it has been shown that treatment of cells with antibodies recognizing the Eph ectodomain can also induce Eph receptor activation and initiation of downstream signaling (11).

Finally, EphA2 clustering has been associated with tissue invasion by cancer cells. Indeed, nearly half of human breast cancers overexpressed the receptor (12). Our studies confirm that, at high concentrations, Eph receptors could cluster independent of ligand, potentially leading to transforming phenotypes.

Summary

Ephrin-dependent Eph receptor clustering and subsequent downstream signaling cause cytoskeleton reorganization that leads to the contact-dependent cell–cell attraction or repulsion that is involved in tissue patterning (13). Our study reveals the structure of the functional Eph/ephrin assemblies at the cell surface and suggests a mechanism for Eph receptor clustering and activation that involves bivalent homotypic interactions between the LBD and CRD domains in neighboring receptors. Previously, we demonstrated that the CRD plays a critical role in the formation of Eph/ephrin “signalosomes” (3, 5) and now reveal the specific molecular regions involved, as well as the underlying structural mechanisms. Focusing on EphA2, we determined a series of EphA2 ectodomain structures containing the CRD, including that of the full EphA2 extracellular region, both alone and in complex with A-class ephrin ligands. The same CRD-mediated EphA2 assemblies are observed under all different crystallization conditions and space groups, in both the presence and absence of bound ligand. Importantly, we document the physiological relevance of the proposed activation mechanism by using structure-based mutagenesis in a variety of cell-based signaling systems, including EphA2 phosphorylation and cell-surface localization and clustering.

Plasmid construction, host-cell growth, protein purification, crystallization, structure determination, and cell-culture studies are described in SI Methods.

Note Added in Proof.

While this manuscript was under consideration, Y. Jones and colleagues published similar structure findings (23).