Main

Separase belongs to the clan CD family of cysteine proteases which includes caspases1. It contains a large amino (N)-terminal armadillo (ARM) repeat domain and a highly conserved carboxy (C)-terminal separase protease domain (SPD) that consists of a pseudo-protease domain (PPD) and an active protease domain (APD)14 (Fig. 1a and Extended Data Fig. 1). Cohesin forms an asymmetric ring to topologically entrap chromosomes (Fig. 1a)15,16. Separase cleaves the kleisin subunit to open the cohesin ring and trigger chromosome segregation. It also cleaves other substrates to regulate anaphase spindle elongation and centriole duplication17,18. High-resolution structures of separase have not been determined more than a decade since its discovery, hindering our understanding of its mechanism and regulation.

Figure 1: Structure of ctSPD.
figure 1

a, Domains and motifs of separase from C. thermophilum (top) and schematic drawing of cohesin (bottom). b, Sequence alignment of the cleavage sites of separase substrates; sc, Saccharomyces cerevisiae; sp, Schizosaccharomyces pombe; xl, Xenopus laevis; hs, Homo sapiens. c, Autoradiograph of the ctSPD cleavage assay with 35S-ctScc1 wild type (WT) or non-cleavable mutant (NC) as substrates. For gel source data, see Supplementary Fig. 1. d, Cartoon of the crystal structure of ctSPD. L4 is coloured magenta. Loops with no visible electron densities are indicated by dashed lines. e, Cartoon of caspase 9 (Protein Data Bank accession number 1JXQ), with the bound inhibitor shown as yellow sticks.

PowerPoint slide

Full size image

We found that SPD of C. thermophilum (ct) separase could be expressed in large quantities in bacteria without securin (Extended Data Fig. 2a). Recombinant ctSPD, but not the C2110S mutant, cleaved ctScc1 to produce two major fragments (Extended Data Fig. 2b). Separase is known to cleave after the EXXR (X, any residue) consensus motif2. Charge-reversal mutation of the 212EVGR215 motif in ctScc1 reduced cleavage by separase (Fig. 1b, c). An acyloxymethyl ketone (AMK)-containing peptide inhibitor derived from this cleavage site blocked ctScc1 cleavage in a dose-dependent manner (Extended Data Fig. 2c, d), and retarded the gel mobility of ctSPDWT, but not ctSPDC2110S, consistent with covalent inhibition (Extended Data Fig. 2e). Similar to separases from other species6, longer constructs of ctSPD containing an N-terminal extension underwent autocleavage at the 1643ELAR1646 site (Fig. 1b and Extended Data Fig. 2f). Thus, recombinant ctSPD was active.

We determined the crystal structure of ctSPD (Fig. 1d and Extended Data Table 1). It forms one globular domain with two sub-domains—the PPD and the APD—that pack against each other. APD has an overall fold similar to that of caspases (Fig. 1d, e and Extended Data Figs 3 and 4a). PPD also has a mixed α/β fold, but its central β-sheet has a topology different from that of caspases. One edge of this central sheet of PPD forms an edge-on interaction with that of APD, whereas the other edge is capped by a helical domain in PPD. A prominent helical insert of PPD forms a long coiled-coil and packs against APD.

The catalytic dyad H2083 and C2110 are located in loops L3 and L4 of APD (Fig. 1d and Extended Data Fig. 3a). An important mechanism of pro-caspase activation is the reorganization of L4, which can be achieved through homo-dimerization, cleavage of an internal linker, or both19,20,21,22. The geometry of the catalytic dyad and the extended conformation of L4 in ctSPD are similar to those in active caspase 9 (Fig. 1d, e), consistent with ctSPD being an active enzyme. Thus, separase activation does not require proteolytic cleavage of L4. Consistent with the importance of the L4 loop, mutations of two residues adjacent to C2110, M2108 and S2112, reduced the activity of ctSPD (Fig. 2a, b and Extended Data Fig. 4b). In contrast, mutations of L4 residues distal to C2110, including E2120 and F2121, enhanced the activity of ctSPD.

Figure 2: Contributions of the L4 loop and the helical insert to the activity of ctSPD.
figure 2

a, Cartoon of ctSPD with the catalytic dyad and L4 loop residues shown. b, d, Quantification of the protease activity of ctSPD WT and mutants (mean ± s.d., n = 3 independent experiments). Mutants with activities greater or less than twofold that of WT are in blue and red, respectively. c, Interactions between the N-terminal tag and a surface pocket of ctSPD. e, Autoradiograph of the 35S-ctScc1 cleavage assay by ctSPD WT or mutants. Bottom: Coomassie-stained gel of ctSPD proteins. f, A conserved basic pocket in ctSPD, with the 2Fo − Fc map of the bound citrate shown at 2.0σ.

PowerPoint slide

Full size image

A segment of the N-terminal tag of recombinant ctSPD binds to a conserved surface pocket in PPD adjacent to L4 (Figs 1d, 2c and Extended Data Fig. 4c). Although this tag is not required for the activity of ctSPD, mutations targeting residues in the tag-binding pocket altered the activity of ctSPD containing the tag (Fig. 2d and Extended Data Fig. 4d). Similar to mutations of the distal L4 residues, the D1698K and D1960K mutations enhanced the activity of ctSPD. We propose that securin or other regions of separase may bind to this tag-binding site, alter the conformation of L4, and affect the protease activity of separase. Even without bona fide ligands, binding of an artificial tag to this site can regulate the protease activity of ctSPD in a subtle way.

Unlike active caspase 9, which forms a homodimer19, separase contains an internal PPD in the same polypeptide chain that packs against and stabilizes its APD. In particular, the helical insert of PPD makes extensive contacts with APD and bridges the two sub-domains (Extended Data Fig. 5a, b). Deletion of the helical insert or mutations of key residues at the helical insert–APD interface, including D1805 and W2143, abolished the expression of soluble ctSPD in bacteria (Fig. 2e and Extended Data Fig. 5a–c). Several helical-insert residues, including C1782 and H1783, are located close to the active site (Extended Data Fig. 5a). Mutations of these residues did not affect the solubility of ctSPD, but reduced the protease activity (Extended Data Fig. 5d). Moreover, residues from the tip of the helical insert, along with residues from APD, form a basic pocket that binds a citrate molecule (Fig. 2f). Mutations of these conserved residues, with the exception of R1794E, diminished separase activity (Fig. 2e and Extended Data Fig. 1). Therefore, the helical insert is critical for both the structure integrity and activity of separase.

Phosphorylation of Scc1 by Plk1 enhances Scc1 cleavage by separase10,11. This cleavage-enhancing phosphorylation is opposed by the shugoshin–PP2A complex bound to cohesin23,24,25,26,27. Incubation of ctScc1, but not ctScc1 S210A, with human Plk1 (hsPlk1) enhanced the cleavage of ctScc1 by ctSPD (Extended Data Fig. 6a). Addition of the hsPlk1 inhibitor BI2536 blocked this enhancement. The phospho-mimicking S210E mutation stimulated ctScc1 cleavage by separase (Extended Data Fig. 6b). Thus, Plk1-dependent phosphorylation of ctScc1 at S210 enhances Scc1 cleavage by separase.

We next determined the crystal structures of ctSPD bound to unphosphorylated (AMK) or phospho-S210-containing (pAMK) inhibitors (Extended Data Table 1). The overall structure of ctSPD–AMK and ctSPD–pAMK complexes is virtually identical to that of free ctSPD, indicating that substrate binding does not induce notable conformational changes. Only the C-terminal 212EVGR215 segment of the unphosphorylated AMK inhibitor was visible (Extended Data Fig. 6c), whereas all residues of the pAMK inhibitor had clearly defined electron density (Fig. 3a). In both structures, the active-site cysteine C2110 is covalently linked to ctScc1 R215 at the P1 position. Consistent with its role in stabilizing the oxyanion during catalysis, ctSPD H2083 of the catalytic dyad is located close to carbonyl group of ctScc1 R215. R215 forms a salt bridge with ctSPD D2151 at the base of a deep, acidic S1 pocket (Fig. 3b, c). ctScc1 E212 at P4 inserts into the aforementioned citrate-binding pocket, forming favourable electrostatic and hydrogen bonding interactions (Fig. 3b and Extended Data Fig. 6d). ctScc1 V213 and G214 form minimal contacts with ctSPD. Mutations of residues lining the S1 and S4 pockets in ctSPD greatly diminished separase activity (Fig. 3d).

Figure 3: Structural basis of Scc1 cleavage by separase.
figure 3

a, Cartoon of ctSPD bound to pAMK (shown as sticks overlaid with the 2Fo − Fc map at 1.0σ). b, Cross-sectional view of the surface drawing of ctSPD–ctScc1 coloured with its electrostatic potential (blue, positive; red, negative; white, neutral). The bound Scc1 peptide is shown as sticks. c, The S1 pocket of ctSPD that recognizes the P1 arginine. Dashed lines indicate hydrogen bonds or favourable electrostatic interactions. d, Autoradiograph of the cleavage reaction of ctSPD WT and mutants with 35S-ctScc1 as substrate. Active-site/S1 mutants are labelled red; S4 mutants are labelled blue. Bottom: Coomassie-stained gel of the ctSPD proteins. Asterisk marks an aberrant cleavage product of D2151A. e, The pS210-binding site. f, Autoradiograph of the cleavage reactions of ctSPD WT and mutants with 35S-ctScc1 as substrate, with or without a prior incubation with hsPlk1.

PowerPoint slide

Full size image

The S1 pocket mutant D2151A of ctSPD cleaved ctScc1 at a different site (Fig. 3d). The ctScc1 E180K mutation abolished this aberrant cleavage (Extended Data Fig. 6e), indicating that the mutant separase cleaved the 180ELGM183 site. Thus, D2151 not only selects for basic residues but also discriminates against hydrophobic residues at P1. ctSPD charge reversal mutants D2151R and R2152E did not efficiently cleave the complementary charge reversal mutants of ctScc1 (Extended Data Fig. 6f), indicating that other residues in the S1 and S4 pockets contribute to substrate recognition. Because most residues lining the S1 and S4 pockets are conserved among separases in all species (Extended Data Figs 1 and 7), our analyses establish the basis for the EXXR substrate specificity of separase.

Phosphorylation of the substrate does not alter the binding mode of EVGR at P1–P4, but reveals or establishes additional contacts at P5 and P6. I211 at the P5 position packs against W1797 of the helical insert (Fig. 3e). ctScc1 I211A was less efficiently cleaved by ctSPD with or without Plk1 (Extended Data Fig. 8a). Thus, as reported previously28, the hydrophobic residue at P5 contributes to substrate specificity. Phospho-S210 at P6 makes favourable electrostatic interactions with R1794 of the helical insert and R2148 of APD in ctSPD (Fig. 3e and Extended Data Fig. 8b). Single mutation of R1794 or R2148 reduced the stimulation of Scc1 cleavage by Plk1, whereas the double mutation abolished the effect (Fig. 3f and Extended Data Fig. 8c, d). Unlike R2148A, R1794E does not affect the cleavage of unphosphorylated Scc1. Therefore, R1794 specifically serves as a receptor for pS210. R2148 contributes to the recognition of both pS210 and E212. The serine at P6 is conserved in fungal Scc1 and other separase substrates (Fig. 1b). The N-terminal separase cleavage site in vertebrate Scc1 contains a phospho-mimicking, acidic residue at that position. Our structures thus explain the phosphorylation dependency of cohesin cleavage, and further suggest that this phospho-regulation might apply to other separase substrates.

Finally, we probed the mechanism by which securin inhibits separase. Securin blocks substrate access to the active site of separase3,6. Expectedly, the ctsecurin–ctseparase complex was less active in ctScc1 cleavage, compared with ctSPD (Extended Data Fig. 9a, b). A conserved EVE motif in securin matches the separase cleavage consensus at positions P2–P6, but lacks the arginine at P1 and often has a proline at P0 instead of a hydrophilic residue (Fig. 4a). A securin mutant with three residues in this motif mutated was cleaved by separase in fission yeast cells13. We thus mutated P164 and P165 in ctsecurin to R and D, the matching ctScc1 residues at P1 and P0. The resulting ctsecurinRD mutant was efficiently cleaved by ctSPD, and this cleavage was inhibited by the AMK inhibitor (Fig. 4b). Mutating the phospho-serine-binding residues in ctSPD or E159 in ctsecurinRD reduced cleavage (Fig. 4c and Extended Data Fig. 9c), indicating that this artificial substrate bound at the canonical substrate-binding sites of separase. ctsecurin bound tightly to the N-terminal ARM domain of ctseparase (Extended Data Fig. 9d). A synthetic EVE-containing securin peptide did not inhibit ctSPD (Extended Data Fig. 9e). We propose that securin acts as a pseudo-substrate to competitively block substrate binding to separase (Fig. 4d). Securin binding to the ARM domain of separase provides the necessary avidity for securin to outcompete authentic substrates for access to the active site. Securin is not cleaved because of incompatible residues at the site of cleavage.

Figure 4: Securin as a pseudo-substrate of separase.
figure 4

a, Sequence alignment of the EVE motif of securin, with the separase cleavage consensus shown above. The KEN and destruction boxes (D) are indicated. Φ/ζ, hydrophobic/hydrophilic residues. b, Autoradiograph of the cleavage reactions of 35S-ctsecurin WT or P164R/P165D (RD) by ctSPD WT or C2110S with or without the AMK inhibitor. c, Autoradiograph of the cleavage reactions of 35S-ctsecurinRD by the indicated ctSPD proteins. d, Model depicting specificity determinants, phospho-regulation, and securin inhibition of separase-dependent cohesin cleavage. SA, stromal antigen. Cohesin cleavage by separase can be stimulated by DNA30. The ARM domain of separase might contact DNA.

PowerPoint slide

Full size image

As a crucial protease that triggers chromosome segregation, separase is a potential oncoprotein29. Because of the conserved principles of substrate recognition (Extended Data Fig. 7), our structure of an active fungal separase can guide the rational design of chemical inhibitors of human separase, which may have therapeutic potential.

Methods

No statistical methods were used to predetermine sample size. The experiments were not randomized. The investigators were not blinded to allocation during experiments and outcome assessment.

Expression and purification of ctSPD

The ctseparase cDNA (GenBank identity 18261092) was synthesized at GenScript USA. For the expression of the ctSPD, the cDNA fragment of ctSPD1632–2223 was subcloned into a modified pET bacterial expression vector. The pET-ctSPD vector encoded ctSPD1632–2223 with an N-terminal His6 tag of the following sequence: MGSSHHHHHHSQLEVLFQGPLGSGRP. The pET-ctSPD vector was transformed into Escherichia coli strain BL21(DE3). Protein expression was induced with isopropylthiogalactoside (IPTG) at 18 °C overnight. The bacteria were harvested and resuspended in the lysis buffer (50 mM Tris-HCl, pH 8.0, 200 mM NaCl, 5% glycerol, 1 mM DTT, and 0.05% Triton X-100). After sonication and centrifugation, the supernatant was applied to Ni2+-NTA resin (Qiagen). After extensive washing, His6-ctSPD was eluted from the Ni2+-NTA column. His6-ctSPD was further purified with a mono Q 5/50 GL anion-exchange column (GE Healthcare) and a Superdex 200 10/300 GL column. The point mutants of ctSPD were generated with a QuikChange Lightning Site-Directed Mutagenesis kit (Agilent Technologies). The truncated variants and point mutants of ctSPD were expressed and purified similarly. Because ctSPD1632–2223 underwent autocleavage at the 1643ELAR1646 site, we generated a non-cleavable 1643RLAE1646 mutant to prevent autocleavage and increase yield. All ctSPDs in this study, except that in Extended Data Fig. 2f, contained the non-cleavable mutation.

The selenomethionine (SeMet)-labelled ctSPD1663–2223 was produced with the methionine biosynthesis inhibition method31. Briefly, bacteria transformed with pET-ctSPD cultured overnight were pelleted, washed, and resuspended with M9 minimal media. The bacteria were further incubated at 37 °C until the absorbance at 600 nm reached about 1.0. Methionine biosynthesis was inhibited by the addition of the amino-acid solution containing 50 mg l−1 of Leu/Ile/Val and 100 mg l−1 of Phe/Lys/Thr/SeMet. Protein expression was induced with 0.4 mM ITPG at 18 °C overnight. The SeMet-labelled protein was subsequently purified through the same procedure as described above.

Expression and purification of ctseparase–securin complex and ctSPD1501-2223

The ctsecurin cDNA (GenBank identity 18256826) was cloned from a cDNA library of C. thermophilum. The cDNAs of ctseparase and ctsecurin were separately subcloned into a modified pFastBac HT vector (Invitrogen). The final constructs encoded an N-terminal His6-Strep-tagged ctseparase and an N-terminal His6-tagged ctsecurin. Baculoviruses of ctseparase (full-length or residues 1-1500) and ctsecurin were constructed with the Bac-to-Bac system (Invitrogen) according to the manufacturer’s protocols. Sf9 cells were co-infected with ctseparase and ctsecurin baculoviruses and harvested at 48 h after infection. Cells were resuspended in the lysis buffer containing 50 mM Tris-HCl, pH 8.0, 200 mM NaCl, 5% glycerol, 1 mM DTT, and 0.05% Triton X-100, followed by sonication and centrifugation. The supernatant was applied onto a Strep-Tactin Superflow column (Qiagen). After extensive washes with the lysis buffer, the ctseparase–securin complex was eluted with the elution buffer containing 5 mM d-Desthiobiotin (Sigma-Aldrich), 50 mM Tris-HCl, pH 8.0, 200 mM NaCl, 5% glycerol, and 1 mM DTT. The His6-ctSPD1501–2223 protein was expressed in Sf9 cells with a similar strategy and purified through a Ni2+-NTA column.

Separase activity assay

The ctScc1 cDNA (GenBank identity 18259702) was synthesized by GenScript USA and was cloned into a modified pCS2 vector with a SP6 promoter. To produce 35S-ctScc1 or its mutants, the pCS2-ctScc1 plasmids were added to a TNT Quick Coupled Transcription Translation System (Promega) and incubated in the presence of 35S-methionine at 30 °C for 90 min. (The 35S-ctScc1 proteins migrated as a doublet on SDS–polyacrylamide gel electrophoresis (SDS–PAGE), possibly owing to proteolysis or internal methionine initiation during the in vitro translation reaction.) Then, 2 μl of 35S-ctScc1 was added to 18 μl of ctSPD (~1.5 μM) or ctseparase–securin protein solution containing 25 mM HEPES (pH 7.5), 75 mM KCl, 5 mM MgCl2, 1 mM DTT, 15 mM NaF, 1 mM EGTA, 10% glycerol, and 0.05% Triton X-100, and incubated at 30 °C for 60 min. For assays in Fig. 1c and Extended Data Fig. 2b, a higher concentration of ctSPD (3.0 μM) was used, resulting in more complete ctScc1 cleavage. For AMK inhibition assay, ctSPD was pre-incubated with the ctScc1-AMK inhibitor (synthesized by KareBay Biochem) at room temperature (25 °C) for 30 min and further incubated with 35S-ctScc1 at 30 °C for 60 min. In the Plk1 stimulation assay, 35S-ctScc1 was pre-treated with recombinant GST-hsPlk1T210D in the kinase buffer for 30 min at 30 °C, in the absence or presence of 10 μM BI2536, and further incubated with ctSPD (0.5 μM) for 60 min. The reaction mixtures were separated on SDS–PAGE gels, which were stained, destained, dried, and analysed with Fuji or GE phosphoimagers.

The ctsecurin WT and P164R/P165D (ctsecurinRD) cDNAs were subcloned into pCS2-Myc vector with an SP6 promoter. The 35S-ctsecurinWT and ctsecurinRD proteins were produced and assayed as described above. The ctsecurin153–177 (DPLQVEEVEYAPPKPKEMPYESDVF) and ctsecurin153–177 3A (DPLQVEAAAYAPPKPKEMPYESDVF) peptides were chemically synthesized and tested for their ability to inhibit the cleavage of ctScc1 by ctSPD as described above.

Crystallization and data collection

All crystallization experiments were performed at 20 °C. Initial screens were performed with a Phoenix crystallization robot (Art Robbins Instruments), using the commercially available screening kits from Hampton Research, Qiagen, and Molecular Dimensions. Conditions obtained from the initial screens were optimized using hanging-drop vapour diffusion method. Diffraction-quality crystals were obtained by repeated microseeding. All crystals were cryoprotected with a reservoir solution supplemented with 15% glycerol.

Both native and SeMet-labelled ctSPD1663–2223 crystals were grown by mixing equal volumes of the protein solution (11 mg ml−1) with the precipitant solution containing 0.2 M ammonium citrate tribasic (pH 7.0), 20% PEG3350, and 10 mM DTT. Diffraction data were collected at beamline BL8.2.1 at the Advanced Light Source (Lawrence Berkeley National Laboratory) at the wavelength of 0.9786 Å at 100 K and processed with HKL3000 (ref. 32).

For crystallization of ctSPD1632–2223–AMK and ctSPD1693–2223–pAMK complexes, the purified ctSPD proteins were mixed with the ctScc1-AMK or phospho-ctScc1-AMK peptide inhibitors (KareBay Biochem) at a molar ratio of 1:2.5, and incubated overnight at room temperature to form covalent complexes as monitored by SDS–PAGE. The complexes were further purified with a Superdex 200 10/300 GL size-exclusion column in the buffer containing 20 mM Tris-HCl (pH 8.0), 200 mM NaCl, and 5 mM DTT. Crystals of ctSPD1632–2223–AMK were grown by mixing equal volumes of the protein solution (13 mg ml–1) with the precipitant solution containing 0.1 M ammonium citrate tribasic (pH 7.0) and 12% PEG3350. For the crystallization of ctSPD1693–2223–pAMK complex, the 11 mg ml–1 protein solution was mixed with an equal volume of the precipitant solution containing 0.2 M KCl, 50 mM HEPES (pH 7.5), 32% pentaerythritol propoxylate (5/4 PO/OH), and 10 mM DTT. Diffraction data for ctSPD1632–2223–AMK and ctSPD1693–2223–pAMK were collected at beamline 19-ID (SBC-CAT) at the Advanced Photon Source (Argonne National Laboratory) at 100 K at wavelengths of 0.9793 Å and 0.9795 Å, respectively, and processed with HKL3000.

Structure determination and refinement

The crystal of SeMet-labelled ctSPD1663–2223 diffracted to a minimum Bragg spacing of 2.20 Å and exhibited the symmetry of space group P212121 with cell dimensions of a = 55.67 Å, b = 98.79 Å, c = 107.76 Å. Phases were obtained from the selenium single-wavelength anomalous diffraction method. With data truncated to 2.5 Å, nine of ten possible selenium sites were located and refined with PHENIX AutoSol33, resulting in an overall figure of merit of 0.323. The experimental electron density map was used to construct an initial model with automated building with PHENIX AutoBuild. As a result, 414 of total 587 residues were built in the initial model, with Rwork and Rfree of 27.74% and 32.79%, respectively. Iterative model building and refinement were performed with JLigand34, COOT35, and PHENIX. Phases of native ctSPD1663–2223, ctSPD1632–2223–AMK, and ctSPD1693–2223–pAMK were obtained by molecular replacement with Phaser using the SeMet crystal structure as the search model. Data collection and structure refinement statistics are summarized in Extended Data Table 1. Ramachandran statistics (favoured/allowed/outlier (%)) calculated by MolProbity36 for ctSPD1663–2223, ctSPD1632–2223–AMK, and ctSPD1693–2223–pAMK were 98.0/1.6/0.4, 98.1/1.9/0.0, and 98.1/1.7/0.2, respectively. All structural figures were generated with the program PyMOL (http://www.pymol.org/) using the same colour and labelling schemes.