The crystal structure of a 250-kDa heterotetrameric particle explains inhibition of sheddase meprin β by endogenous fetuin-B

Significance

Membrane-bound metallopeptidases cut substrates as “sheddases” on the cell surface. Due to the irreversibility of peptide bond cleavage, they are regulated by specific, colocalizing protein inhibitors. We solved the crystal structure of the complex between the ectomoiety of the human integral membrane sheddase, meprin β (Mβ), and its only reported endogenous protein inhibitor, fetuin-B (FB), which reveals a large particle of ∼250-kDa and ∼160-Å maximal dimension. Our results unveiled a sophisticated “raised elephant trunk” mechanism as the structural basis for the very strong inhibition of Mβ by FB. Moreover, we derived a functional model for Mβ on the cell membrane, which foresees a transition between a membrane-proximal and a membrane-distal position for catalytic and inhibitory functions, respectively.

Abstract

Meprin β (Mβ) is a multidomain type-I membrane metallopeptidase that sheds membrane-anchored substrates, releasing their soluble forms. Fetuin-B (FB) is its only known endogenous protein inhibitor. Herein, we analyzed the interaction between the ectodomain of Mβ (MβΔC) and FB, which stabilizes the enzyme and inhibits it with subnanomolar affinity. The MβΔC:FB crystal structure reveals a ∼250-kDa, ∼160-Å polyglycosylated heterotetrameric particle with a remarkable glycan structure. Two FB moieties insert like wedges through a “CPDCP trunk” and two hairpins into the respective peptidase catalytic domains, blocking the catalytic zinc ions through an “aspartate switch” mechanism. Uniquely, the active site clefts are obstructed from subsites S₄ to S₁₀′, but S₁ and S₁′ are spared, which prevents cleavage. Modeling of full-length Mβ reveals an EGF-like domain between MβΔC and the transmembrane segment that likely serves as a hinge to transit between membrane-distal and membrane-proximal conformations for inhibition and catalysis, respectively.

Protein ectodomain shedding causes limited proteolysis on the cell membrane. It is an irreversible posttranslational modification that regulates the availability of the soluble circulating forms of hundreds of membrane-anchored proteins performing synaptic, paracrine, or endocrine functions (1). The executive peptidases, dubbed sheddases, are key regulators of biochemical processes ranging from cell signaling to cell adhesion. One such sheddase, meprin β (Mβ), is a widespread metallopeptidase (MP) that is highly abundant on the surface of various cell types (2). Mβ sheds surface proteins such as growth factors, cytokines, receptors, amyloid precursor protein (APP; at its β-secretase site), and other MPs (1, 3, 4). It is therefore essential for processes in which deregulation leads to neurodegenerative diseases, inflammatory bowel disease, fibrosis, nephritis, and cancer (2). This sheddase is a type-I integral membrane multidomain enzyme that includes a signal peptide for secretion (M¹−A²²; numbering according to UniProt [UP] entry Q16820), a propeptide (T²³−R⁶¹), a catalytic domain (CD; N⁶²−L²⁵⁹), a MAM domain (S²⁶⁰−C⁴²⁷), a TRAF domain (P⁴²⁸−S⁵⁹³), an EGF-like domain (D⁶⁰⁶−G⁶⁴⁷), a transmembrane segment (TM; I⁶⁵³−V⁶⁷³), and a cytosolic tail (CST; S⁶⁷⁴−F⁷⁰¹) (5, 6).

Owing to the irreversible character of peptide bond cleavage, sheddases must be carefully controlled to prevent aberrant proteolysis, which for Mβ occurs at the transcriptional level (2) and through secretion as a zymogen form that is activated by propeptide removal, as is the case for many other MPs (7). Moreover, Mβ is itself shed from the cell surface, which modifies its spectrum of substrates (8). Yet another regulatory mechanism is inhibition by fetuin-B (FB), the only known endogenous protein inhibitor (9). FB itself is essential for mammalian fertilization (10, 11) and a strong biomarker for Alzheimer’s disease (12).

Herein, we set out to investigate the structural determinants of Mβ inhibition by FB using recombinant protein constructs spanning CD-MAM-TRAF domains (MβΔC) and full-length FB, both produced in insect cells (SI Appendix). We were unable to crystallize the complex with human FB, but obtained a uniquely diffracting crystal using the mouse ortholog (SI Appendix, Fig. S1 A and B), which shares 62% sequence identity and is a valid model for the human variant. Before crystallization, we determined the optimal MβΔC:FB stoichiometry by native polyacrylamide gel electrophoresis (PAGE) (Fig. 1A) and found that complex formation was associated with cleavage of FB to yield N- and C-terminal fragments of ∼34 kDa and ∼23 kDa, respectively, which remained linked by disulphide bonds (Fig. 1 B and C). We determined the inhibition constant (K_i) of FB to be in the subnanomolar range (1.3E-10 M; Fig. 1D), consistent with the value reported for human FB (6.0E-11 M) (13), further supporting the equivalence of the murine ortholog. Moreover, the results of nano differential scanning fluorimetry showed that the melting temperature of mature MβΔC, but not its zymogen, rose by ∼10 °C when complexed with FB (Fig. 1E). This supports a strong stabilizing effect of the inhibitor on the peptidase, which was dependent on the buffer system chosen, but not the pH (Fig. 1 F and G).

Fig. 1.

In vitro analysis of the MβΔC:FB complex. (A) Native PAGE analysis depicting complex formation at three MβΔC:FB molar ratios. At 1:1.2, free MβΔC and the 2Mβ:1FB heterotrimer (arrow 1) coexist with the 2Mβ:2FB heterotetramer (arrow 2). At 1:2, all Mβ is complexed. (B) Reducing (Left) and nonreducing (Right) SDS-PAGE analysis of the complex showing inhibitor cleavage within the CTR, probably within segment E₂₆₇–E₂₇₆. The two cleavage products are denoted by * and ^#, respectively. (C) Western blotting analysis of the MβΔC:FB complex (after reducing SDS-PAGE) with antibodies (α) against full-length FB (red) or the histidine tag (green). Bands labeled with white * and ^# indicate the two FB cleavage products. (D) Fractional velocity plot of MβΔC inhibition by FB at 0.5 nM and 25 μM enzyme and substrate concentrations, respectively. The Inset depicts the derived K_i value. (E–G) Nano differential scanning fluorimetry assays as proxies of the stability of MβΔC variants and FB at different temperatures (E), in different buffers (F), and at various pH values (G). Isolated FB did not denature at 90 °C under the experimental conditions assayed. Buffers were as follows: 1, Hepes; 2, Hepes + SDS; 3, PBS; 4, Tris + NaCl; values for pH analysis: 1, pH 5; 2, pH 7.5; 3, pH 8.5 (SI Appendix, Experimental Procedures).

The crystal structure of the MβΔC:FB complex was solved by multistep molecular replacement, but the subsequent refinement with data to 2.8 Å was complicated due to anisotropy of the diffraction data, intrinsic flexibility of the complex, and the high glycosylation content of the crystals (SI Appendix, Experimental Procedures and Table S1). Overall, the structure reveals a ∼250-kDa heterotetrameric polyglycosylated particle with dimensions 160 × 115 × 110 Å, consisting of two peptidase moieties (chains A and C) and two inhibitor protomers (chains B and D) (Fig. 2 A and B). The MβΔC moieties contain a mixed α/β CD domain with an architecture typical of peptidases of the astacin family within the metzincin clan of MPs (6, 14, 15). A deep and narrow active site cleft accommodates the catalytic zinc ion at the midpoint and separates a lower C-terminal subdomain from an upper N-terminal one, in which the N terminus is buried in the protein structure.

Fig. 2.

Structure of the heterotetrameric MβΔC:FB complex. (A) Ribbon plot of the inhibitory particle in cross-eyed stereo representation. The view is down the twofold axis and the cross-linking disulphide between MAM domains is shown as yellow sticks surrounded by a red circle. Each domain is colored differently and labeled. The CPDCP trunk is shown as a red ribbon, the FB moieties are overlaid with their semitransparent Van der Waals surface, and the glycans are shown as stick models. Blue and magenta spheres denote sodium and zinc ions, respectively. (B) Same as A after vertical and in-plane 90° rotations. (C) Detailed view of a single MβΔC:FB complex in stereo. (D) Close-up view of C. Selected residues are shown and their side-chain carbons are colored pink (FB), green (CD of MβΔC), or blue (TRAF of MβΔC), and labeled red (MβΔC) or purple (FB). The three main elements of inhibition (the CPDCP trunk, hairpin I and hairpin II) are numbered 1 through 3.

The catalytic metal is bound by three histidine residues imbedded in a long zinc-binding consensus sequence spanning H¹⁵²EXXHXXGXXH¹⁶², which is characteristic of metzincins. The downstream MAM domain has a “jelly-roll” β-sandwich architecture with a front and back five-stranded antiparallel β-sheet and a “dimerization loop” grafted into two vicinal edge strands of the back sheet. Moreover, a crossover loop between the sheets frames a sodium-binding site. Finally, the TRAF domain is a β-sandwich with a five-stranded antiparallel front sheet and a four-stranded antiparallel back sheet. This domain is the agglutinating core of the MβΔC monomer (Fig. 2C) and interacts with the MAM and CD domains through protein surfaces of 739/684 Ų and 943/949 Ų for each of the two complexes, respectively, while no significant contacts occur between the CD and MAM domains. Remarkably, the TRAF and CD domains are attached laterally, so that the active site cleft of the enzyme is extended on its primed side along the TRAF surface to include subsites S₆′−S₁₀′ (nomenclature according to ref. 16). The two MβΔC moieties form a compact ellipsoidal particle of ∼115 Å in the largest dimension, which overall is very similar to the unbound MβΔC structure previously published (6), with C2 twofold symmetry and a buried intermolecular surface area of 1,235 Ų. The dimer is covalently linked by an intermolecular disulphide bond between residues C³⁰⁵ from either dimerization loop (Fig. 2A).

The FB moieties comprise cystatin-like domain CY1 (P₃₀–S₁₄₆; for numbering, see UP Q9QXC1), a linker hereafter dubbed “CPDCP-trunk” (K₁₄₇–P₁₅₈), cystatin-like domain CY2 (S₁₅₉–E₂₆₇), and most of the C-terminal region (CTR; S₂₆₈–P₃₈₈) from residue P₃₀₃ onwards, which forms a disulphide-linked structural unit with CY1. CY1 and CY2 adopt the cystatin fold (17) and consist of a curled and twisted five-stranded antiparallel β-sheet of simple up-and-down connectivity, with an α-helix inserted between the first two strands that nestles in the concave face of the sheet, perpendicular to the strands. The second and third strands, as well as the fourth and fifth, are connected by loops and give rise to “hairpin I” and “hairpin II,” respectively. The two cystatin domains are connected by the CPDCP-trunk region, which protrudes from the FB surface and features a short α-helix and segment C₁₅₄PDCP₁₅₈ arranged in a rigid disulphide-linked 1,4-type tight turn. After CY2, the CTR is folded irregularly, partially disordered, and cleaved. After P₃₀₃, the chain is parallel to the fifth strand of CY1, and the downstream part of the domain is very flexible and runs unevenly along the convex surface of the CY1 β-sheet, as described previously for the unbound structure (18).

The complex has a remarkable glycan structure; ∼16 kDa of the ∼65-kDa sugars could be modeled as 93 N-linked monosaccharides attached to seven and two asparagine residues in each MβΔC and FB moiety, respectively. The glycan chains conform to two patterns (Fig. 3A), present in recombinant proteins from insect cells, which are similar to those of mammalian glycosylation pathways. Remarkably, the glycan chain attached to N⁵⁴⁷ spans all 12 monosaccharides (Fig. 3B). Notably, the sugar moieties attached to N³⁷⁰, N⁴³⁶, and N⁴⁴⁵ from MAM, and N⁵⁴⁷ and N⁵⁹² from TRAF, are all oriented toward the interdomain space, and create a “sugar channel,” which may have a role in binding of particularly hydrophilic or glycosylated substrates more efficiently. This would be consistent with the reported function of MAM and TRAF domains in protein–protein interactions.

Fig. 3.

Glycan structure of the MβΔC:FB complex. (A) Scheme depicting the two types of N-linked glycan chains, I (Left) and II (Right), which are consistent with glycan/codes H/010.F and E1/M9.2 found in glycoproteins produced in Trichoplusia ni insect cells (24). Type-I chains were identified for N²¹⁸, N³⁷⁰, N⁴³⁶, N⁴⁴⁵, and N⁵⁹² of MβΔC, and N₁₃₉ of FB. Type-II chains were identified for N²⁵⁴ and N⁵⁴⁷ of MβΔC, and N₄₀ of FB. The most complete chains encompassing all seven and 12 saccharides were found linked to N²¹⁸/N³⁷⁰ and N⁵⁴⁷, respectively. Blue squares and circles represent β-N-acetyl-D-glucosamine and α-D-glucose, respectively; green squares and circles represent β-D-mannose and α-D-mannose, respectively; red triangles represent α-l-fucose. The type of glycosidic bond is indicated. (B) Detailed view of the MβΔC:FB complex in stereo around the glycan chain attached to N⁵⁴⁷ superposed with the final σ_A-weighted (2mF_obs-DF_calc)-type Fourier map contoured a 0.8 σ.

Inhibition of dimeric MβΔC is caused by insertion of the FB moieties into the respective active site clefts, which bury 1,535 Ų (complex A·B; ΔⁱG = −11.4 kcal/mol) and 1,506 Ų (complex C·D; ΔⁱG = −9.2 kcal/mol) and form 32 and 30 hydrogen bonds or salt bridges, respectively, including two protein–metal interactions each (SI Appendix, Table S2). Participating segments include E⁸⁴–N⁸⁶, F¹¹⁹–E¹³⁷, A¹⁴²–R¹⁴⁶, T¹⁴⁹–H¹⁵⁶, W¹⁶¹–H¹⁶², L¹⁸¹–I¹⁹⁵, and Y²¹¹–Q³¹⁷ from the peptidases, and M₁₁₁–Y₁₁₃, K₁₄₇–S₁₆₇, M₁₉₇–Y₂₀₇, V₂₄₂–T₂₆₀, and E₃₁₂–D₃₁₅ from the inhibitors.

The CY1/CTR moiety of FB does not participate directly in inhibition, but contributes to the structural integrity of the FB molecule. This is consistent with cleaved FB evincing a 12-fold higher K_i value against Mβ (13). By contrast, hairpins I and II plus the CPDCP trunk of CY2 form a tripartite wedge that blocks the entire cleft. It uniquely binds subsites S₄ through S₁₀′ of the enzyme, apart from S₁ and S₁′ that flank the scissile bond, and S₅′ (Figs. 2 C and D and 4A). This explains why the inhibitor is not cleaved even though the three wedge components bind to the cleft in the direction of an authentic substrate; no peptide bond is placed directly on top of the catalytic zinc (Fig. 2 C and D). Instead, the latter is bound by the side chain of D₁₅₆ of the CDPCP trunk. This is reminiscent of the latency mechanism of the MβΔC zymogen in which D⁵² from the propeptide blocks the zinc in an “aspartate-switch” mechanism that is typical of astacin MPs (19). However, in the latter case, the polypeptide runs across the active site cleft in the opposite direction of a substrate. Moreover, hairpin I includes a conserved QWVXGP motif that, together with the CPDCP trunk, contributes to inhibition of MβΔC and crayfish astacin, which only spans the CD domain (18). By contrast, hairpin II uniquely contacts the subsites located on the TRAF surface of MβΔC (S₆′–S₁₀′; Fig. 2D). This element was undefined in the isolated FB structure, and in the astacin complex, where it protruded nonproductively from the molecular surface (18). These CD-distal cleft subsites are mainly shaped by the second β-strand of the TRAF front sheet (G⁵⁵⁴–A⁵⁸¹), which interacts with the extended propeptide in the Mβ zymogenic structure through antiparallel interactions (6). Overall, these findings indicate that hairpin II is specific for Mβ inhibition and contributes to the very low K_i value (see above). It is key to a specific “raised elephant trunk” inhibition mechanism for this MP (Fig. 4A), which is distantly reminiscent of single-domain cystatins that target cysteine peptidases.

Fig. 4.

Inhibitory mechanism and model for Mβ function on the membrane. (A) Scheme depicting the raised elephant trunk mechanism specifically for Mβ, with the view flipped vertically relative to that in Fig. 2 B and C. The CPDCP trunk (red) and the subjacent stabilizing segment R₁₁₀–S₁₁₅ (purple) mimic a raised trunk and the tusk (1 and Top Right Inset). The strands of hairpins I (2) and II (3), colored in pink, represent the forelegs and hindlegs, respectively. Hydrogen bonds between main-chain atoms are shown as blue dashed lines, C₁₅₄ and C₁₅₇ of the CDPCP trunk are linked via a disulphide bond (yellow line), and D₁₅₇ binds directly to the catalytic zinc cation (green dashed line). Overall, cleft subsites from S₄ to S₁₀′, which are provided by CD and TRAF, are contacted (except for S₁, S₁′ and S₅′). A sulfate anion in S₁′ matches the specificity of Μβ for negatively charged residues (25). (B) (Left) Working model of the full-length Mβ dimer anchored to the cytoplasmic membrane in a “membrane-distal” conformation when inhibited by two FB moieties, namely, the inhibitor complex. The model is based on the experimental structure and a homology model for the EGF-like, TM, and CST regions of the sheddase. Each MβΔC domain is colored as in Fig. 2 and labeled. FBs are magenta and purple, glycans are gray. The homology models of EGF-like, TM, and CST of Mβ are colored white, pale yellow, and white, respectively, to underpin that they are not experimental structures. (Right) Working model of the sheddase in its predicted “membrane-proximal” substrate complex, further featuring segment 624 to 723 of APP as a red ribbon (see also ref. 6). A magenta arrow indicates the APP cleavage site within the CD active-site cleft of the right Mβ protomer. Note that the orientation of the CD-MAM-TRAF core is the same in both complexes, which are just separated by a pure translation of ∼40 Å (green arrow). Red asterisks indicate the hypothetical hinge points in the upstream and downstream linkers of the EGF-like domains for such a transition to happen, and green arrows denote the symmetric rotations that the EGF domains would undergo. In addition, the TM-CST moieties would also move apart by ∼15 Å within the membrane (short green arrows).

Finally, we analyzed the architecture and function of Mβ anchored at the cytoplasmic membrane using a homology model of the full-length enzyme (Fig. 4B). The model indicates that the EGF-like domains probably function as spacers between the dimeric particle core and the TM segments. These spacers would enable rotation around hinge points within the flanking linkers of this domain in an example of “molecular athletics,” allowing the core to adopt disparate positions with respect to the membrane, from distal to proximal, by a simple translation of ∼40 Å. The inhibitory complex with FB would adopt a distal arrangement, whereas a proximal arrangement would be required during shedding of membrane-anchored substrates, which would need to adopt an “N-like trajectory” (6) to penetrate the active site cleft adequately for catalysis (Fig. 4B).

In conclusion, we here reported the structure of a ∼250-kDa heterotetrameric, polyglycosylated inhibitory particle, which sheds light on the regulation of the activity of Mβ by its endogenous FB inhibitor through a “raised elephant trunk” inhibition mechanism. Geometric considerations about a modeled substrate complex of Mβ suggest that the membrane-bound enzyme may switch between membrane-proximal and membrane-distal locations by simple transition of the dimeric core facilitated by the EGF-like domains acting as hinges.

Materials and Methods

The full experimental procedures are provided in SI Appendix. Briefly, mouse FB and human Mβ were produced in insect cells by recombinant baculovirus-induced overexpression and further prepared as previously reported (9, 13, 18, 20). The complex was crystallized by sitting-drop vapor diffusion and solved by molecular replacement. Inhibition studies were carried out as described (9, 13) and nano differential scanning fluorimetry was performed using a Prometheus NT.48 apparatus according to the manufacturer’s instructions. Native PAGE and sodium dodecyl sulfate (SDS)-PAGE were carried out as described (21–23). A homology model of full-length Mβ was built linking a comparative model of the EGF-like domain, the transmembrane helix, and the cytosolic tail with the experimental structure spanning the CD, MAM, and TRAF domains and subjecting it to geometric refinement to obtain reasonable geometry and eliminate clashes.

Data Availability

Structure coordinates data have been deposited in the Protein Data Bank (7AUW). All study data are included in the article and/or supporting information.