Research Article

Structure of a Site-2 Protease Family Intramembrane Metalloprotease

See allHide authors and affiliations

Science  07 Dec 2007:
Vol. 318, Issue 5856, pp. 1608-1612
DOI: 10.1126/science.1150755


Regulated intramembrane proteolysis by members of the site-2 protease (S2P) family is an important signaling mechanism conserved from bacteria to humans. Here we report the crystal structure of the transmembrane core domain of an S2P metalloprotease from Methanocaldococcus jannaschii. The protease consists of six transmembrane segments, with the catalytic zinc atom coordinated by two histidine residues and one aspartate residue ∼14 angstroms into the lipid membrane surface. The protease exhibits two distinct conformations in the crystals. In the closed conformation, the active site is surrounded by transmembrane helices and is impermeable to substrate peptide; water molecules gain access to zinc through a polar, central channel that opens to the cytosolic side. In the open conformation, transmembrane helices α1 and α6 separate from each other by 10 to 12 angstroms, exposing the active site to substrate entry. The structure reveals how zinc embedded in an integral membrane protein can catalyze peptide cleavage.

Regulated intramembrane proteolysis (RIP) is a conserved signaling mechanism from bacteria to humans (18). An essential step of RIP is the site-specific cleavage of a transmembrane signaling protein by a specific membrane-embedded protease within the lipid bilayer. These intramembrane proteases are classified into four families: the metalloprotease site-2 protease (S2P), serine protease rhomboid, and aspartyl proteases presenilin and signal-peptide peptidase (24).

RIP signaling is exemplified by cleavage of the membrane-bound transcriptional factor sterol regulatory element–binding protein (SREBP) by S2P in mammals (911). In response to low levels of cellular cholesterol, SREBP is translocated from the endoplasmic reticulum (ER) to the Golgi, where it is cleaved at a site that is three amino acids into the transmembrane segment on the cytosolic side (12). This cleavage follows a prior cleavage in the lumen of the Golgi by the site-1 protease (S1P) (13). Consequently, the DNA binding and transactivation domain of SREBP is released from the Golgi membrane and translocated into the nucleus, where it activates transcription of genes that control biosynthesis and uptake of cholesterol and fatty acids. The mammalian S2P is also responsible for cleavage activation of the transcription factor ATF6 (14), which plays a central role in the ER stress signaling. In response to periplasmic stress, the Escherichia coli S2P homolog YaeL (also known as RseP) cleaves a transmembrane protein RseA, following an initial cleavage event mediated by the periplasmic serine protease DegS (15). The Bacillus subtilis S2P homolog sporulation protein SpoIVFB removes the prosequence from Pro-σK, which allows the resulting transcription factor to activate genes that are required for sporulation (16, 17).

The S2P family proteases contain a consensus HExxH sequence, in which the two histidine residues are thought to coordinate a zinc atom together with a conserved aspartate residue (1, 1820). During catalysis, the conserved glutamate residue is thought to activate a zinc-bound water molecule to initiate nucleophilic attack on the scissile peptide bond. These putative catalytic residues are predicted to be located below the lipid membrane surface. In this case, because proteolysis requires water molecules, how do hydrophilic water molecules enter the active site of S2P? More important, how do transmembrane substrate proteins gain access to the active site? Are there common principles that govern different families of intramembrane proteases? These important questions remain unanswered.

Recent structural investigations on the rhomboid serine proteases revealed tantalizing clues about how an intramembrane protease might function (2124). However, because S2P shares no apparent sequence homology with rhomboid, information derived from rhomboid proteases cannot be directly applied to the understanding of S2P. In this study, we report the crystal structure of an S2P homolog from the archaebacterial species Methanocaldococcus jannaschii.

Crystallization of mjS2P. We cloned 40 S2P homologs from 31 bacterial and archaebacterial species and examined their expression in E. coli (25). On the basis of solution behavior, we focused on the S2P homolog from M. jannaschii and generated crystals of its transmembrane core domain (residues 1 to 224), hereafter referred to as mjS2P.

We reconstituted a proteolysis assay in vitro, in which mjS2P cleaved an artificial protein substrate CED-9 (26) in detergent micelles (Fig. 1A, left). Next, quantitative element analysis revealed that zinc is bound to mjS2P in an approximately 1:1 molar ratio. Finally, the metalloprotease-specific inhibitor 1,10-phenanthroline specifically inhibited substrate cleavage in a concentration-dependent manner (Fig. 1A, right). These analyses validated the use of the transmembrane core domain for crystallographic studies.

Fig. 1.

Structure of the transmembrane core domain of an S2P homolog from M. jannaschii (mjS2P). (A) The transmembrane core domain of mjS2P is catalytically active. The membrane-associated protein CED-9 (26) was used as an artificial protein substrate for mjS2P. The proteolytic activity is inhibited by 1,10-phenanthroline, an inhibitor specific for metalloproteases (right). (B) Overall structure of mjS2P in one asymmetric unit. Two molecules of mjS2P, named A (blue) and B (green), associate with each other to form a pseudo-dimer in the crystals. The catalytic zinc atom is highlighted in red. The zinc-binding and catalytic residues are colored yellow. (C) The core domain and gate domain of mjS2P. Overlay of the two molecules of mjS2P reveals a shared core domain (left) and a diverging gate domain (right). With the exception of Figs. 3A, 4A, and 4B, all structural figures were made using MOLSCRIPT (30).

The structure of mjS2P was determined by multi-wavelength anomalous dispersion (MAD), with the use of seleno-methionine–labeled protein, and refined to 3.3 Å resolution (table S1 and fig. S1). Details of crystallization and structural determination are given in the supporting online materials (25).

Overall structure of mjS2P. There are two molecules of mjS2P in an asymmetric unit, designated A and B, which associate with each other to form an antiparallel, pseudo-dimer in the crystals (Fig. 1B). The orientation of each molecule relative to the lipid membrane was assigned on the basis of the locations of charged amino acids between adjacent transmembrane segments. Each molecule contains seven α helices and five short β strands (fig. S2). At the N terminus, two antiparallel β strands (β1 and β2) combine with a third strand β3, located between helices α2 and α3, to form a membrane-embedded β sheet. Strands β4 and β5 form a β hairpin between helices α3 and α4-N.

The secondary structural elements of mjS2P are arranged into a six-transmembrane–segment (TM1–6) topology that does not resemble any known metalloprotease (fig. S2). TM1 consists of strand β2 and helix α1, whereas TM4 contains two separate α helices, α4-N and α4-C, which are connected by a nine–amino acid bulge. TM2, TM3, TM5, and TM6 correspond to helices α2, α3, α5, and α6, respectively. As previously hypothesized, the catalytic zinc atom is coordinated by three amino acids that are invariant among all S2P family proteins: His54 and His58 from helix α2 and Asp148 from the N-terminal end of helix α4-C in mjS2P. The zinc atom is located ∼14 Å below the lipid membrane surface from the cytosolic side.

The overall structure of the two mjS2P molecules in one asymmetric unit is similar, with a root-mean-square deviation (RMSD) of 1.8 Å over 150 aligned Cα atoms out of a total of 217 amino acids. In particular, TM2, TM3, and TM4 stack closely against each other and share nearly identical conformations in these two molecules (Fig. 1C, left). However, there are also significant conformational differences. Compared with molecule B, molecule A adopts a relatively open conformation, in which TM1 and TM6 move away from each other by approximately 10 to 12 Å (Fig. 1C, right). This difference results in the exposure of the active site in molecule A, but not in B. From these observations, molecules A and B are proposed to exist in open and closed conformations, respectively.

The conserved conformation of TM2 to 4 is likely essential to the formation of the active site, because His54 and His58 reside in TM2 and Asp148 is located in TM4. TM3 plays a structural role by interacting with and supporting the conformations of TM2 and TM4. In support of this analysis, TM2, TM3, and TM4 have been predicted to be common to all S2P family members (19), which contain conserved sequences in these TMs (fig. S2). Thus, we term TM2, TM3, and TM4 the “core domain” of mjS2P. In contrast, TM1, TM5, and TM6 exhibit different conformations in the two mjS2P molecules and may represent two distinct states in the regulation of substrate entry. We thus term TM1, TM5, and TM6 the “gate domain” of mjS2P.

The active site. In molecule B, the distances between zinc and the coordinating atoms of His54, His58, and Asp148 are 2.3, 2.2, and 2.1 Å, respectively (Fig. 2A). In molecule A, these distances become 2.3, 2.3, and 2.2 Å, respectively. Coordination of the zinc atom is roughly tetrahedral, with the fourth water ligand unassigned owing to the moderate resolution. In both molecules, His58 is hydrogen-bonded to the carbonyl oxygen atom of residue 94. The closest carboxylate oxygen atom of Glu55 is 3.3 and 3.4 Å away from the zinc atom in molecules A and B, respectively. These distances are consistent with activation by Glu55 of the zinc-bound water molecule during catalysis. It is noteworthy that Asn140, another invariant residue among S2P family proteases, is located above the active site in the open space. This location suggests a critical role for Asn140 in catalysis, perhaps in binding to substrate and/or helping the formation of the oxyanion hole. In support of this notion, comparison of the active-site conformation between mjS2P and thermolysin, a HExxH-containing metalloprotease, revealed that Asn140 is located in approximately the same position as Arg203 in thermolysin (27) (Fig. 2B), which binds to the carbonyl oxygen of the scissile peptide bond and neutralizes the negative charge during catalysis (28).

Fig. 2.

Conformation of the active site. (A) A stereo view of the active-site conformation in molecule B. The catalytic zinc atom is coordinated by His54 and His58 on helix α2 and Asp148 at the N-terminal end of helix α4-C. Glu55 likely facilitates the activation of a water molecule during catalysis. The conserved Asn140 may contribute to the formation of the oxyanion hole. Helix α3 does not directly participate in catalysis but interacts with helices α2 and α4 to form the conserved core domain. (B) A stereo comparison of the active-site conformation between mjS2P (blue) and a HExxH-containing metalloprotease thermolysin [gold, PDB code 2TLX (27)]. Asn140 of mjS2P is located in a similar position as Arg203 in thermolysin, which is part of the oxyanion hole. (C) Mutation of the active-site residues compromised proteolytic activity. Shown here is an SDS-PAGE gel visualized by Coomassie staining. CED-9 was used as an artificial substrate in these assays.

To assess the contribution of the catalytic residues, we generated missense mutations in mjS2P. The mutant proteins H54A (in which Ala replaces the His at residue 54), E55A, H58A, D148A, and H54A/D148A (20) exhibited markedly compromised protease activity compared with the wild-type (WT) mjS2P (Fig. 2C). Quantitative element analysis by inductively coupled plasma-emission spectrometry revealed that, compared with the WT protein, the zinc content was 8.2, 44.2, 5.1, and 15.1% for the mutant proteins H54A, E55A, H58A, and D148A, respectively.

An important structural feature of mjS2P is that all active-site residues are contained within TM2 and TM4, with TM3 stabilizing the active-site conformation from the opposite side of where potential substrate proteins are cleaved (Fig. 2A). This arrangement immediately rules out the possibility that conserved residues in TM3 may directly participate in catalysis. Consistent with this analysis, Ala97 and Gly98 in the conserved AGxxxN/S/G sequence of TM3 (19) appear to play a structural role in mjS2P. For example, Gly98 stacks closely against His54 and His58, and its substitution for any other amino acid is predicted to perturb the conformation of the active site because of steric hindrance from the side chain.

Access to water molecules. How do water molecules gain access to the active site of mjS2P? Analysis of the van der Waals surface in molecule B reveals a channel that originates from the zinc atom to the cytosolic side (Fig. 3A). The narrowest point in this channel measures ∼2.6 Å in diameter (Fig. 3B), which is large enough to allow passage of a water molecule. Notably, the inner surface of the channel contains a number of polar groups and charged amino acids that may facilitate water entry. Five carbonyl oxygen atoms from residues 79, 81, 83, 84, and 146 point into the channel, poised to coordinate water molecules (Fig. 3C). In addition, the side chains of two charged amino acids Arg151 and Glu207 also point into the channel. Thus, water molecules appear to have constant access to the active site of mjS2P in the closed conformation.

Fig. 3.

Access of water molecules to the active site of mjS2P. (A) The van der Waals surface in molecule B reveals a channel that leads to the active site from the cytosolic side. The calculation was performed with the program HOLE (31) and the image was generated using VMD (32). The center line of the channel is colored magenta. (B) The channel is large enough to allow passage of water molecules. Distance from the zinc atom along the center line of the channel is plotted against the minimal radius at each point. The red line indicates the minimal radius required for passage of water molecules. (C) The channel is lined with polar groups that may help facilitate water entry. Shown here is a view of the polar groups in the channel approximately along the center line. Five carbonyl oxygen atoms, Arg151, and Glu207 are positioned along the inside of the channel.

Mechanism of substrate entry. Molecules A and B exhibit markedly different conformations in TM1. In the closed molecule B, exclusively hydrophobic amino acids from TM1 and TM2 interdigitate to form an extensive network of van der Waals interactions (fig. S3A). Compared with molecule B, TM1(helix α1 and strand β2) pivots outward around helix α2 by ∼35° to reach its position in molecule A (fig. S3B). This movement results in major alteration of the hydrophobic interface between TM1 and TM2. TM6 and TM5 also have different conformations in molecules A and B, although such differences are small compared with those in TM1. Unlike TM1, conformational changes in TM6 and TM5 do not result in the repacking of hydrophobic interfaces, because helix α4-C also slightly adjusts its position to maintain the same packing interactions with α6 and α5.

The two contrasting conformations of mjS2P molecules provide a plausible explanation for the question of how substrate proteins gain access to the active site. In the closed molecule B, the active site is inaccessible to transmembrane protein substrate (Fig. 4A). In the open molecule A, there is a deep groove that is roughly parallel to the transmembrane helices (Fig. 4B). This groove traverses through the entire molecule and exposes the active site to potential substrate peptide. These observations suggest a lateral gating mechanism, in which TM1 and TM6-TM5, as two sides of the gate, move away from each other to allow substrate entry and catalysis (Fig. 4C).

Fig. 4.

Mechanism of substrate gating in mjS2P. (A) Surface representation of the closed state of mjS2P in two perpendicular views. Note the closure of the active site. (B) Surface representation of the open state of mjS2P in two perpendicular views. In this conformation, an extended polypeptide can be readily fitted into the cleft between the two gating helices (α1 and α6). (A) and (B) were prepared using GRASP (33). (C) A proposed general model for the S2P family of intramembrane proteases. In this model, substrate entry to the active site is gated by two transmembrane segments, TM1 and TM6-TM5.

Discussion. How is a substrate protein recognized by S2P? Although the current study does not provide a direct answer to this question, it reveals some tantalizing clues. For example, a number of buried amino acids in the closed conformation (molecule B) become exposed in the open conformation (molecule A), thus creating novel surface features that might be involved in binding to substrate proteins. Compared with the bottom half of the putative substrate-binding groove, the top half is much wider (Fig. 4B) and could accommodate an intact α-helix. We speculate that the funnel-shaped groove in the open conformation may play an active role in unwinding the transmembrane helix of the substrate protein.

Previous studies show that substrate cleavage of the putative transmembrane helix occurs in various positions. These observations appear to contrast with the prediction that the catalytic zinc atom and the active-site residues are likely located in approximately the same depth into the membrane for S2P proteases. This discrepancy can be reconciled by the hypothesis that specific cleavage of a substrate protein is determined by specific recognition.

Structure determination of mjS2P allows comparison with the rhomboid proteases, which represent the only other structurally characterized intramembrane protease (2124). Despite the superficial similarity of 6TMs, the structures of mjS2P and the rhomboid serine protease GlpG exhibit different topologies and share no apparent features. Nonetheless, water molecules appear to gain access to the active sites of rhomboid and mjS2P via a similar mechanism. GlpG from E. coli contains a water-filled cavity that converges on the active-site residue Ser201 and opens to the extracellular side (2123). In contrast, mjS2P has a polar channel that allows water entry to the catalytic zinc atom in the closed conformation. Both classes of intramembrane proteases also appear to share a common feature in mechanisms of substrate entry—gating by transmembrane helices, although significant differences exist. In GlpG, bending of the C-terminal half of TM5 was proposed to open the gate for substrate entry (22). This hypothesis is consistent with other structural observations (21, 23, 24) and biochemical characterization (29). In mjS2P, the rotation and translocation of TM1 and the translocation of TM6-TM5 may allow substrate entry.

The full-length mjS2P protein contains 339 amino acids; but only the transmembrane core domain (residues 1 to 224) was used for crystallization. Could the structure be altered or disrupted by not including the C-terminal sequences? Two lines of evidence argue against such a possibility. First, the C-terminal sequences of mjS2P were frequently lost during bacterial expression, which suggests that they are unlikely to be part of the structural core domain. Indeed, the transmembrane core domain of mjS2P was identified by using limited proteolysis. Second, the transmembrane core domain retained full proteolytic activity compared with the full-length mjS2P (fig. S4).

Because mjS2P was crystallized in the presence of detergents, it is possible that the observed conformations are, in part, induced by interactions with the detergent molecules or crystal-packing interactions. Although we could not rule out this possibility, the observed active-site geometry and the conserved core domain structure in the two molecules do not suggest anything had gone awry. Nonetheless, further experiments will determine whether the proposed closed and open conformations of mjS2P in the crystals are physiologically relevant. The large conformational differences of the gate domain in the two molecules are consistent with the fact that TM1 and TM6-TM5 make up highly divergent sequences among members of the S2P family. This feature, also observed in TM5 of GlpG, would better allow specificity for recognition and cleavage of substrate proteins.

References and Notes

View Abstract

Stay Connected to Science

Navigate This Article