Structural Basis of the Human Endoglin-BMP9 Interaction: Insights into BMP Signaling and HHT1

Abstract

Endoglin (ENG)/CD105 is an essential endothelial cell co-receptor of the transforming growth factor β (TGF-β) superfamily, mutated in hereditary hemorrhagic telangiectasia type 1 (HHT1) and involved in tumor angiogenesis and preeclampsia. Here, we present crystal structures of the ectodomain of human ENG and its complex with the ligand bone morphogenetic protein 9 (BMP9). BMP9 interacts with a hydrophobic surface of the N-terminal orphan domain of ENG, which adopts a new duplicated fold generated by circular permutation. The interface involves residues mutated in HHT1 and overlaps with the epitope of tumor-suppressing anti-ENG monoclonal TRC105. The structure of the C-terminal zona pellucida module suggests how two copies of ENG embrace homodimeric BMP9, whose binding is compatible with ligand recognition by type I but not type II receptors. These findings shed light on the molecular basis of the BMP signaling cascade, with implications for future therapeutic interventions in this fundamental pathway.

Keywords: TGF-β superfamily proteins; bone morphogenetic protein receptors; cell surface receptors; endoglin; growth differentiation factor 2; hereditary hemorrhagic telangiectasia; orphan domain; protein interaction domains and motifs; x-ray crystallography; zona pellucida domain.

Graphical abstract

Figure 1

ENG OR Consists of a Duplicated and Circularly Permutated Domain (A) Cartoon representation of the OR crystal structure, consisting of two domains (OR1/OR2) and rainbow-colored from blue (N terminus) to red (C terminus). Cys residues are shown in ball-and-stick representation. (B) OR topology with OR1 and OR2 β helices colored dark and light blue, respectively. β Strands (arrows) are numbered. (C) Superposition of OR1 and OR2, colored as in (B). (D) Suggested evolution of ENG OR by gene duplication and circular permutation. (E) Superposition of ENG OR2 and tubulin cofactor C (TBCC). See also Figures S1–S3 and S6 and Table S1.

Figure 2

A Hydrophobic Surface of ENG OR1 Binds to the Knuckle Region of BMP9 (A) Crystal structure of the OR-BMP9 complex, showing the dimer generated by crystal symmetry. OR is colored as in Figures 1B and 1C, and BMP9 chains are shown in yellow. (B) Non-reducing immunoblot and Coomassie (CBB) analysis of His pull-down of ^MOR/BMP9 co-expression experiments. Two copies of OR bind dimeric BMP9 but do not interact with each other. (C) Detail of the OR1-BMP9 interface, with the surface of OR1 colored by hydrophobicity (orange, most hydrophobic; white, least hydrophobic). (D) Alternative view of (C), with both proteins in cartoon representation and interface residues shown in ball-and-stick representation. (E) BMP9 is pulled down by non-mMBP fused wild-type OR but not by double or triple mutant ORs, indicating that the residues shown in (D) are essential for binding. See also Figures S4 and S8 and Table S1.

Figure 3

ENG Contains a Minimal ZP Module with Closely Interacting ZP-N and ZP-C Domains (A) Crystal structure of ENG ZP, colored according to secondary structure. Canonical disulfide bonds and the RGD motif are shown in ball-and-stick representation. The extra ZP-N disulfide (C350–C382) and the free ZP-C Cys involved in dimerization (C516) are also indicated. (B) Side-by-side comparison of the ENG and UMOD ZP modules. The secondary structure elements of ENG ZP-N and ZP-C are only spaced by five residues; thus, these domains are much closer to each other than the corresponding domains of UMOD, which are separated by a long linker (red). As a result, whereas the ZP-N homodimerization surface of UMOD is available to pair with another molecule (black/gray), the equivalent region of ENG interacts with ZP-C and maintains the protein in a monomeric state. See also Figures S1, S5, and S6 and Table S1.

Figure 4

Intermolecular Disulfide Bonding of ENG ZP C516 Generates a Homodimeric Clamp for the BMP9 Ligand (A) The disulfide bonds observed in the structures of OR (Figures 1A and 1B; Figure S2D) and ZP (Figure 3A; Figure S5D) are indicated in red. Because neither ENG E26-S337 (Castonguay et al., 2011) nor ^MOR (Figure 2B; Figure S1C) make intermolecular disulfide bonds, the pattern suggests that homodimerization of ENG is mediated by a C516-C516 disulfide in addition to C582-C582 (Guerrero-Esteo et al., 2002). (B) A non-reducing anti-His immunoblot of conditioned medium (M) shows secretion of monomeric and dimeric ENG ZP constructs (white and black arrowheads, respectively). Immunoblot of cell lysates (L) indicates that all constructs are expressed in equal amounts. Consistent with the ENG architecture proposed in (A), mutation of C516 abolishes homodimerization of ^MZP. Moreover, the additional C350–C382 disulfide found in ENG ZP-N (Figure 3A; Figure S5D) is essential for ^MZP secretion. (C) Disulfide connectivity and secretion state of the ^MZP constructs analyzed in (B). Lack of secretion is symbolized by a red cross. (D) Immunoblot and Coomassie analyses of co-expression experiments under non-reducing conditions show that two ENG molecules, preferentially a disulfide-bonded dimer, bind homodimeric BMP9. (E) SEC-MALS analysis of purified His-ECTO-BMP9 complex produced in HEK293S cells reveals two conformations whose common molecular mass (red line) is consistent with a 2:2 stoichiometry. (F) Theoretical model of the complete extracellular region of homodimeric ENG in complex with BMP9. The relative orientation of the ZP-C domains cross-linked by the C516-C516 intermolecular disulfide (dashed red ellipse) is compatible with the C582-C582 disulfide (Guerrero-Esteo et al., 2002) at the C terminus of the ectodomain (dashed black ellipse). See also Figures S1 and S6.

Figure 5

Theoretical Model of the ENG-BMP9-ALK1 Ternary Complex (A) Combination of the homodimeric ENG-BMP9 model shown in Figure 4F with structural information on how the type I receptor ALK1 binds BMP9 (Townson et al., 2012) suggests a possible architecture for the co-receptor-BMP ligand-type I receptor ternary complex on the plasma membrane. (B) Top view of the ENG-BMP9-ALK1 model. (C) SDS-PAGE analysis of ENG ECTO-FLAG, ALK1-His, and BMP9 co-expression followed by His pull-down confirms that the proteins form a triple complex in solution. Consistent with the model depicted in (A) and (B), ALK1 is only able to pull down ECTO when BMP9 is present. All samples were analyzed under non-reducing conditions, except for that marked with R (reducing conditions). (D) Top view of the crystal structure of the ALK1-BMP9-ActRIIB complex (Townson et al., 2012). Comparison with the ENG-BMP9-ALK1 model (B) suggests that ENG and ActRIIB compete for the same binding region on BMP9.

Figure 6

HHT1-Associated Mutations Disrupt the Folding of ENG (A and B) Overview and close-ups of ENG OR (A) and ZP (B), with patient mutation sites highlighted in magenta. Disulfide bonds and C516 are shown in ball-and-stick representation; coloring is according to Figures 2 and 3. Relative percentages of solvent accessibilities of residue side chains are indicated in parenthesis. See also Figure S7.