Rhomboid proteases are a fascinating class of enzymes that combine a serine protease active site within the core of an integral membrane protein. Despite having key roles in animal cell signalling and microbial pathogenesis, the membrane-immersed nature of these enzymes had long imposed obstacles to elucidating their biochemical mechanisms. But recent multidisciplinary approaches, including eight crystal structures, four computer simulations and nearly 100 engineered mutants interrogated in vivo and in vitro, are coalescing into an integrated model for one rhomboid orthologue in particular, bacterial GlpG. The protein creates a central hydrated microenvironment immersed below the membrane surface to support hydrolysis by its serine protease-like catalytic apparatus. Four conserved architectural elements in particular act as ‘keystones’ to stabilize this structure, and the lateral membrane-embedded L1 loop functions as a ‘flotation device’ to position the protease tilted in the membrane. Complex interplay between lateral substrate gating by rhomboid, substrate unwinding and local membrane thinning leads to intramembrane proteolysis of selected target proteins. Although far from complete, studies with GlpG currently offer the best prospect for achieving a thorough and sophisticated understanding of a simplified intramembrane protease.

INTRODUCTION

The remarkable synthetic ability of cells relies on a vast orchestra of enzymes that transform simple chemicals from the environment into complex biomolecules and energy that sustain life. As a result, enzymes have been the focus of biochemical investigation for over a century. Although most enzymes catalyse metabolic reactions, a subset evolved to function within information-processing circuits, where they are entrusted to regulate cellular processes. To this daunting task, enzymes bring key advantages: they act rapidly and with a high degree of specificity. The modern challenge is to understand how the properties of these regulatory enzymes have been tailored to fit the specific needs of the cell.

Proteases cleave proteins by catalysing hydrolysis of peptide bonds. Because of their speed, selectivity and versatility, proteolytic mechanisms have come to regulate a broad array of cellular processes. It is often stated that evidence of their importance is encrypted directly in our genomes: up to 5% of a genome is often devoted to encoding proteolytic components [1,2]. Moreover, serine proteases in particular have served as paradigms for understanding enzyme catalysis through integration of chemical, enzymological and structural investigations for over a century [36]. Yet even this well-studied area of biochemistry is not immune to surprise: in the past decade an unforeseen class of proteases was discovered that reside immersed within the cell's membrane. Without exception, each of the three types of intramembrane proteases were discovered unintentionally, through analyses focused on cell regulation or human disease [79]. Although initially thought to be rare, intramembrane proteases are now recognized to be among the most prevalent membrane proteins known, and are conserved from bacteria to man [1012].

Rhomboid proteins are the largest family of intramembrane proteases [11]. In contrast with the site-2 protease and γ-secretase/signal peptide peptidase intramembrane protease superfamilies, rhomboid proteins are serine proteases mechanistically, and differ functionally by releasing factors to the outside of the cell, rather than into the cytosol [13]. Originally discovered through genetic analysis of Drosophila embryogenesis, rhomboid proteases initiate Drosophila EGF (epidermal growth factor) signalling by releasing TM (transmembrane) EGF precursors Spitz, Keren and Gurken from the membrane [9,14,15]. Particularly exciting is the recent implication of rhomboid proteases in central processes within human pathogens (recently reviewed in [16]). A rhomboid functions in quorum sensing by a pathogenic bacterium, although the substrate that it activates is not the signal itself, but rather a component of a transporter [17]. Malaria and related parasites use their rhomboid enzymes to cleave adhesins in order to dismantle the junction formed between parasite and host during invasion [1820]. The role of rhomboid in other protozoa is less clear, but may also involve shedding of adhesion molecules, similar to a parasitic amoeba, for example, during phagocytosis and immune evasion [21].

The immense spectrum of roles that intramembrane proteases are known to play continue to expand, but already warrants a thorough investigation of the underlying enzyme mechanisms. In addition to the general interest in intramembrane proteases, it is also becoming appreciated that such insights could have useful applications, especially for inhibitor development and its therapeutic potential [16]. Although the biochemical complexity of these enzymes has made achieving a detailed knowledge of their function difficult, recent years have witnessed a dramatic shift in our understanding of intramembrane protease mechanisms. In this regard, one enzyme stands out: currently multiple structures, defined reconstitution systems, a myriad of mutants and molecular dynamics simulations have been achieved for the bacterial rhomboid GlpG. The aim of the present review is to integrate the diverse experimental approaches that have recently been applied to study GlpG, scrutinize the picture that is starting to emerge, and evaluate to what extent these properties may be common to rhomboid proteases in general.

THE STARTING LINE: BIOCHEMICAL RECONSTITUTION

The catalytic function of Drosophila rhomboid-1 in signalling was deduced from a combination of cell-based assays, protease inhibitor profiling and mutagenesis experiments [9]. All of these approaches unified into a model whereby a serine protease-like active site, assembled from TM segments, lies immersed within the membrane. But the experimental support for this model was entirely indirect: intramembrane proteolysis could not be reconstituted outside of a living cell, making biochemical interrogation impossible.

This obstacle took nearly 4 years to overcome, and ultimately just required a better understanding of rhomboid's enzymatic properties. The initial confounding issue was whether a single polypeptide alone could catalyse intramembrane proteolysis; γ-secretase, the only intramembrane protease that had been purified in active form, required three other proteins for activity [22,23]. Ultimately, analysis of dozens of rhomboid orthologues revealed most were active when examined in heterologous cells despite sharing less than 20% sequence identity (Figure 1) [2426], and displayed common substrate requirements [27]. These studies allowed a more informed approach; simplified prokaryotic rhomboid proteases that share overall enzymatic properties and cleave surrogate substrates made attractive models for recombinant expression and biochemical reconstitution.

Sequence and structural features of the rhomboid transmembrane core

Figure 1
Sequence and structural features of the rhomboid transmembrane core

Top, all residues of E. coli GlpG that have been subjected to structure analyses (residues 87–276), are shown aligned with the corresponding regions of three other well-studied rhomboid enzymes [rhomboid-1 from Drosophila (DmRho1), human RHBDL2, and Bacillus subtilis YqgP]. As such, N-terminal domains of all enzymes, and TM7 and C-terminal domains of DmRho1, RHBDL2 and YqgP are not shown. Residues involved in nucleophilic catalysis (Ser201 and His254 for GlpG) are shaded in green, presumed electrophilic catalysis residues (His150 and Asn154 for GlpG) are shaded in blue, whereas conserved structural residues are shaded in red. Helical secondary structure is highlighted by common colour shading among the sequence alignment (top), cylinders in the topology diagram (bottom left), and crystal structure images (bottom middle and right). Note the intramembrane location of the catalytic serine residue (circled with red star), and the protruding L1 loop (boxed, bottom right). By convention, the ‘back view’ centres on TM5 and TM2, whereas the ‘front view’ centres on TM1 and the L1 loop. All structure images were rendered with MacPymol using PDB co-ordinates 2NRF (most GlpG images), 2IRV (GlpG with lipid in active site), 2IC8 (closed form of GlpG), 2NR9 (H. influenzae GlpG) and 1MOX (Spitz EGF). A 360 ° rotating image of the rhomboid as well as a three-dimensional interactive structure is available at http://www.BiochemJ.org/bj/425/0501/bj4250501add.htm.

Figure 1
Sequence and structural features of the rhomboid transmembrane core

Top, all residues of E. coli GlpG that have been subjected to structure analyses (residues 87–276), are shown aligned with the corresponding regions of three other well-studied rhomboid enzymes [rhomboid-1 from Drosophila (DmRho1), human RHBDL2, and Bacillus subtilis YqgP]. As such, N-terminal domains of all enzymes, and TM7 and C-terminal domains of DmRho1, RHBDL2 and YqgP are not shown. Residues involved in nucleophilic catalysis (Ser201 and His254 for GlpG) are shaded in green, presumed electrophilic catalysis residues (His150 and Asn154 for GlpG) are shaded in blue, whereas conserved structural residues are shaded in red. Helical secondary structure is highlighted by common colour shading among the sequence alignment (top), cylinders in the topology diagram (bottom left), and crystal structure images (bottom middle and right). Note the intramembrane location of the catalytic serine residue (circled with red star), and the protruding L1 loop (boxed, bottom right). By convention, the ‘back view’ centres on TM5 and TM2, whereas the ‘front view’ centres on TM1 and the L1 loop. All structure images were rendered with MacPymol using PDB co-ordinates 2NRF (most GlpG images), 2IRV (GlpG with lipid in active site), 2IC8 (closed form of GlpG), 2NR9 (H. influenzae GlpG) and 1MOX (Spitz EGF). A 360 ° rotating image of the rhomboid as well as a three-dimensional interactive structure is available at http://www.BiochemJ.org/bj/425/0501/bj4250501add.htm.

Despite the inherent complexity of membranes, the reconstitution assays were elegantly simple: proteolytic activity was achieved with recombinantly expressed and purified rhomboid proteins when the appropriate detergent, substrate and, for some, membrane lipid were provided [2830]. In fact, rhomboid polypeptides consisting of fewer than 200 residues could catalyse complete intramembrane turnover of a substrate, without the need for protein partners, cofactors, energy or prosthetic groups [28]. This advance proved that rhomboid proteins themselves were catalytic, and paved the way for including them in the Enzyme Commission database as the only family of intramembrane serine proteases (EC 3.4.21.105). Importantly, this work also provided the long-sought tractable system with which to interrogate their biochemical mechanism, as well as offering validated candidates for structural analysis.

RHOMBOID STRUCTURE: GUIDED TOUR OF A PROTEASE

Structures come into focus

Deducing the atomic structure of a rhomboid protease heralded a renaissance in the intramembrane proteolysis field. Long-standing disbelief in the phenomenon was dispelled, and a new era of pursuing a sophisticated knowledge of rhomboid protease mechanism could begin. The structure of a site-2 family protease soon followed [31], but the current discussion will be focused on rhomboid, since a detailed comparison has been reviewed recently [32].

Within the span of less than a year, eight structures of a bacterial rhomboid orthologue GlpG were published. Five of these were different conformers of GlpG from Escherichia coli [3335], another two resulted from a DMSO-soaked crystal [36] and a point mutant [37], and another still was of an orthologue from Haemophilus influenzae [38]. All structures were of GlpG in the detergent-solubilized state, and only of the TM core: obtaining diffracting crystals required removal, or in the case of H. influenzae absence, of the cytoplasmic N-terminal domain (Figure 1). Despite the diversity of the eight structures, which were crystallized under varying conditions and in four different laboratories, the structures overlaid with a surprising degree of congruity in all regions but one (discussed below). This remarkable observation inspired some confidence in the notion that the structure is not perturbed to a large extent by its environment, and thus could represent the cellular conformation of GlpG.

The overall shape of GlpG is a compact helical bundle comprised of six helical TM segments and an unusual lateral protrusion (Figure 1 and also see the Supplementary Multimedia at http://www.BiochemJ.org/bj/425/0501/bj4250501add.htm.). Surprisingly, nearly the entire protein appears to be immersed in the membrane: even the exceptionally long L1 loop that connects TMs 1 and 2 emanates laterally to adopt a hairpin orientation that would be partially immersed in the top leaflet of the membrane. This asymmetric shape has not been encountered previously, or since, in membrane protein structure, and the protrusion probably serves as a ‘flotation device’ to position the enzyme in the membrane (discussed below). This property may be of wider significance, because this region of GlpG shares sequence homology with the derlin family of proteins implicated in endoplasmic reticulum-associated degradation [30,39,40].

It is worth emphasizing that two structural characteristics in particular ratified the intramembrane serine protease model. The catalytic serine residue lies submerged ~10 Å (1 Å=0.1 nm) from the presumed surface of the membrane, making the site of catalysis intramembrane (Figure 1). Secondly, visualizing in all structures the catalytic serine residue hydrogen-bonded to the presumed catalytic histidine base, which is distant from the serine residue in the primary sequence (Figure 1), provided the final proof for the widely accepted serine protease model.

The structures coupled with functional analyses also offer new and deeper insights into how rhomboid enzymes are constructed and function.

Architectural features: the four keystones

Sequence analysis of rhomboid proteins identified 20 residues that are conserved in most if not all members (Figure 1). Early experimental analysis suggested catalytic functions for four residues [9]. The other conserved residues were anticipated to play important structural roles, and integrating crystal structures and functional analyses now provide a framework for understanding their roles. Although a multitude of local interactions permeate the enzyme, evolutionary conservation highlights four elements that act as architectural keystones to stabilize the structure (Figure 2).

Four keystones of E. coli GlpG rhomboid protease structure

Figure 2
Four keystones of E. coli GlpG rhomboid protease structure

The central stone in an arch is called a keystone and, although arch stability depends on many factors, the keystone occupies a key central-stabilizing position. Pictured in the top inset is the Crypte at Olympia, Greece, with the keystone false-coloured red (photo: S. Urban). The back (left) and front (right) views of GlpG are shown. By analogy, ‘keystone’ residues (highlighted in yellow, active-site residues in red) of four conserved motifs are depicted in each inset. From top right and moving clockwise: the side chain of Arg137 of the ExWRxxT motif reaches upwards to make five hydrogen bonds within the L1 loop, two to Glu134, and three to backbone carbonyls of Leu121 and Ala124, while Thr140 hydrogen bonds to the carbonyl of Leu123. This array of interactions together stabilizes the lateral hairpin. The Glu166 sidechain of the GxxxExxxG motif in TM2 makes four hydrogen bonds: three to backbone and side chain atoms of Val96 and Thr97 in TM1, and one to the hydroxy group of Ser171 in TM3, bringing the three TMs into an apex on the cytosolic side (bottom). TM2 and the L2 loop bend around Gly162 and Gly170 respectively. Gly199 in the GySG motif of TM4 hydrogen bonds with the backbone carbonyls of His141 and Met144 of the L1 loop, whereas Lys200 is buried into the ‘floor’ of the enzyme. The AHxxGxxxG motif in TM6 mediates close approach of TM6 with TM3 (via Ala253), and with TM4 (via Gly257 and Gly261). Oxygen atoms are in red, nitrogen atoms are in blue, and broken lines depict hydrogen bonds throughout.

Figure 2
Four keystones of E. coli GlpG rhomboid protease structure

The central stone in an arch is called a keystone and, although arch stability depends on many factors, the keystone occupies a key central-stabilizing position. Pictured in the top inset is the Crypte at Olympia, Greece, with the keystone false-coloured red (photo: S. Urban). The back (left) and front (right) views of GlpG are shown. By analogy, ‘keystone’ residues (highlighted in yellow, active-site residues in red) of four conserved motifs are depicted in each inset. From top right and moving clockwise: the side chain of Arg137 of the ExWRxxT motif reaches upwards to make five hydrogen bonds within the L1 loop, two to Glu134, and three to backbone carbonyls of Leu121 and Ala124, while Thr140 hydrogen bonds to the carbonyl of Leu123. This array of interactions together stabilizes the lateral hairpin. The Glu166 sidechain of the GxxxExxxG motif in TM2 makes four hydrogen bonds: three to backbone and side chain atoms of Val96 and Thr97 in TM1, and one to the hydroxy group of Ser171 in TM3, bringing the three TMs into an apex on the cytosolic side (bottom). TM2 and the L2 loop bend around Gly162 and Gly170 respectively. Gly199 in the GySG motif of TM4 hydrogen bonds with the backbone carbonyls of His141 and Met144 of the L1 loop, whereas Lys200 is buried into the ‘floor’ of the enzyme. The AHxxGxxxG motif in TM6 mediates close approach of TM6 with TM3 (via Ala253), and with TM4 (via Gly257 and Gly261). Oxygen atoms are in red, nitrogen atoms are in blue, and broken lines depict hydrogen bonds throughout.

The first of these keystones is a conserved E/QxWRxxS/T (where x is any amino acid) motif situated in the L1 loop, which protrudes laterally to form the hairpin extension. This membrane-embedded conformation is stabilized by an underbelly of hydrophobic residues, which greatly decrease enzymatic activity when mutated [41]. The residues of the E/QxWRxxS/T motif lie on the bottom face of the hairpin, yet the side chains of these residues do not point into the membrane, but rather point upwards into the loop to stabilize it (Figure 2). The arginine residue donates five hydrogen bonds to other residues within the L1 loop, including the preceding glutamate residue, providing a key stabilizing influence: mutagenesis of the single arginine residue strongly reduces or abolishes protease activity [37,41]. The conserved threonine residue lies one helical turn after the arginine residue, and also hydrogen bonds with a main chain carbonyl on the upper half of the L1 loop, providing another stabilizing interaction between the lower and upper helices of the lateral L1 loop hairpin. The role of the tryptophan residue is less clear, but could be involved in shielding the arginine residue from the membrane. In support of this hypothesis, mutating the tryptophan residue to alanine strongly reduces enzyme activity in cell membranes [42], but this mutant has enzymatic activity indistinguishable from the wild-type when freed from membranes and assayed in the detergent-solubilized state [41].

A conserved GxxxExxxG sequence is situated on the lower third of TM2 and continues into the cytosolic L2 loop (Figure 2). In the structure, these residues form and stabilize an abrupt turn, because the polypeptide chain needs to cross over to the other side of the membrane-embedded L1 loop. This cross-over allows TM3 to form a V-shape with TM1, cradling the L1 loop in the upper gap between them. The turn is first accomplished by a bend of the TM2 helix around the first glycine residue in the GxxxExxxG motif. The glutamate residue lies at the bottom of TM2, with its side chain reaching out to hydrogen bond with the bottom of TM1 and TM3. The final glycine residue of the motif forms a bend in the cytosolic turn before TM3. The overall effect is the bringing together of helices TM1, 2 and 3 into almost an apex at the cytosolic face of the enzyme's front half (Figure 2). Mutation of either of the two glycine residues or the glutamate residue dramatically reduces enzyme activity (S. M. Moin and S. Urban, unpublished work), supporting an important role in structure stabilization.

Following TM3, a conserved GySG motif centred on the catalytic serine residue in TM4 served as a signature motif in rhomboid enzymes [9,11]. TM prediction programs were ambiguous on where TM4 starts, with some predicting the catalytic serine residue to lie at the very top of TM4. This was initially confounding because the active site was anticipated to be intramembrane. The structure confirmed that the serine was indeed recessed into the plane of the membrane, but the serine residue sits as the topmost residue on TM4, because residues preceding the serine are in an extended loop conformation (Figure 2). The structure also revealed that the conserved residues around the active-site serine residue provide stabilizing points, holding the extended L3 loop in place. The backbone NH group of the first glycine residue hydrogen bonds to the backbone carbonyl oxygens of His141 and Met144 in the L1 loop. Several different residues preceding the catalytic serine residue are permitted among various rhomboid orthologues, and in GlpG this residue is a leucine that faces downward and, as such, is buried into the ‘floor’ of the enzyme. The overall consequence of this structure is a stabilized unobstructed cavity above the catalytic serine residue, which presumably allows entry of water from the outside of the cell and space for the polypeptide substrate.

Lastly, a conserved alanine residue was noted to precede the catalytic histidine residue, and a canonical GxxxG TM dimerization motif exists below it on TM6 (AHxxGxxxG). Both of these elements are universally conserved among all rhomboid proteases (Figure 1). The GxxxG motif is a particularly common and strong TM dimerization motif that mediates close association between helices in various membrane proteins [43,44]. However, it was unclear with what this segment might pair [9]. The crystal structure revealed that the GxxxG motif of TM6 forms a tight association with TM4, bringing together the catalytic serine residue on TM4 and its base partner histidine residue on TM6 (Figure 2). Accordingly, even a single valine substitution in the second glycine residue of the GxxxG abolishes detectable protease activity [41]. Similarly, the alanine residue preceding the histidine residue packs into a small hole on TM3, forming a tight association with a neighbouring helix on the other side of TM6. In this way, TM6 makes a series of tight contacts with neighbouring helices both immediately above and below the catalytic histidine residue, thus cementing it in place.

The catalytic core

The active-site chemistry of soluble serine proteases has been the focus of investigation for over half a century [3,4]. With defined reconstitution systems and crystal structures now available, rhomboid proteases are starting to enter this exciting phase of analysis. However, our current understanding of rhomboid catalysis is strictly limited to analogy to serine protease paradigms; structures of rhomboid with bound substrate or inhibitors, or chemical probes for interrogating its active-site chemistry, have not yet been realized. Nevertheless, soluble serine proteases, although evolutionarily and structurally distinct, serve as an instructive starting point for rationalizing rhomboid catalysis.

Two catalytic elements comprise the active sites of serine proteases: residues that directly sever the C-N amide bond and neighbouring electrophiles that stabilize the negatively charged O of the transition state (for a review, see [3]). Initial investigation of rhomboid in signalling focused attention on four TM residues in particular as key for protease activity: an asparagine residue on TM2, glycine and serine residues on TM4, and a histidine residue on TM6 [9]. The role of the serine and histidine residues were straightforward to understand: the amide bond is initially severed through concerted action by a serine nucleophile and a general base, usually a histidine or lysine residue. The unsung hero of the reaction is the base, since it catalyses nearly every step in the reaction, whereas the serine residue is simply the initial attacking moiety. By analogy, if the serine residue is the scalpel, the histidine residue is the surgeon's hand. The crystal structures revealed the serine residue interacts with the presumed catalytic histidine residue (Figure 3). Yet intriguingly the serine residue is in an unusual position, being the only residue in the entire rhomboid protein that lies in a disallowed region of the Ramachandran plot. Although the significance of this, if any, is not clear, it might serve to position the serine residue for catalysis, by analogy to the ‘nucleophile elbow’ of α/β hydrolases [45].

The two catalytic elements of the GlpG rhomboid protease active site

Figure 3
The two catalytic elements of the GlpG rhomboid protease active site

Nucleophilic catalysis (left), which is responsible for severing the peptide bond, is carried out by a serine protease-like catalytic dyad: Ser201 (on TM4) serves as the nucleophile, whereas the catalytic base is His254 (on TM6). His254 hydrogen bonding to Ser201 is depicted with a dashed red line. The phenol ring of Tyr205, emanating from TM4, stacks under the His254 imidazole ring, providing some support for the catalytic base (although a tyrosine residue could not salt bridge with the imidazolium ion or alter its pKa, as might occur in some proteases). Electrophilic catalysis (right) is required to stabilize the oxyanion of the tetrahedral intermediate. The residues that form this entity are less defined, but are suggested by interaction with a lipid phosphate that protrudes into the active site of structure 2IRV. The lipid is shown in green, with oxygen atoms in red. Hydrogen bonds between the phosphate oxygen and the side chains of His150 and Asn154 (emanating from TM2) are depicted by red dashed lines.

Figure 3
The two catalytic elements of the GlpG rhomboid protease active site

Nucleophilic catalysis (left), which is responsible for severing the peptide bond, is carried out by a serine protease-like catalytic dyad: Ser201 (on TM4) serves as the nucleophile, whereas the catalytic base is His254 (on TM6). His254 hydrogen bonding to Ser201 is depicted with a dashed red line. The phenol ring of Tyr205, emanating from TM4, stacks under the His254 imidazole ring, providing some support for the catalytic base (although a tyrosine residue could not salt bridge with the imidazolium ion or alter its pKa, as might occur in some proteases). Electrophilic catalysis (right) is required to stabilize the oxyanion of the tetrahedral intermediate. The residues that form this entity are less defined, but are suggested by interaction with a lipid phosphate that protrudes into the active site of structure 2IRV. The lipid is shown in green, with oxygen atoms in red. Hydrogen bonds between the phosphate oxygen and the side chains of His150 and Asn154 (emanating from TM2) are depicted by red dashed lines.

Prior to structure determination, a major point of debate was how water is delivered to the membrane-submerged active site. All rhomboid structures revealed that the catalytic serine residue, the region to which water must be delivered, lies at the bottom of a funnel-shaped cavity that is ~10 Å in diameter at its opening and is lined with hydrophilic groups (Figure 1). This architecture immediately suggested that water accesses the active site from the large opening that faces the periplasm. This has been confirmed in E. coli cells [46]; probe accessibility studies indicate that an ~5 kDa membrane-impermeable probe can modify several TM positions, including that of the catalytic serine residue (although not the histidine residue). The limitation of cross-linking experiments is that they are a ‘covalent trap’: even if the conformation that is susceptible to cross-linking is rare or short-lived, once cross-linked it becomes permanently labelled. Nevertheless, these studies indicate that the cavity is open at least part of the time, and to surprisingly large molecules.

What has been less apparent is the identity of the second necessary element for serine protease catalysis, the oxyanion-stabilizing pocket. Early analyses of rhomboid function put forward two possibilities [9]. As in chymotrypsin, the backbone NH of a glycine residue two residues upstream of the catalytic serine residue might stabilize the oxyanion [3,4], since its mutation to even alanine abolishes protease activity in vitro and in cells [9]. Alternatively, by analogy to subtilisin [3], in which a conserved asparagine residue stabilizes the oxyanion, the asparagine residue in TM2 of rhomboid was noted as an alternative oxyanion-stabilizing entity [9].

The oxyanion transition state is difficult to study directly because of its unstable nature. Instead, stable surrogate compounds are often designed to mimic the tetrahedral intermediate [3], but currently no rhomboid structure has been solved with a transition-state inhibitor bound. Nevertheless, one fortuitous structure provides the best alternative currently available: a structure was solved with a phosphate lipid headgroup protruding laterally into the active site [35]. The phosphate shares some properties with the transition state, being both tetrahedral and containing negatively charged oxygen. In the structure, the phosphate lies within hydrogen-bonding distance to the asparagine residue, and to a conserved histidine residue that is one helical turn above the asparagine residue on TM2 (Figure 3). It may also be bonded to the backbone nitrogen of the catalytic serine residue, but not Gly199, which lies two residues upstream of the catalytic serine residue and instead forms a ‘keystone’ interaction (Figure 2). Therefore the best candidates for oxyanion-stabilizing elecrophiles are the side chains of the asparagine and histidine residues on TM2, both of which are universally conserved. Mutation of either residue reduces catalytic activity in some, but not all, rhomboid enzymes that have been investigated [9,24,29,41,47].

It is worth noting that the structure also definitively resolved the long-standing question of whether rhomboid proteases use a catalytic triad. The role of the third residue in catalytic triads, and its conspicuous absence in some serine proteases, continues to be a point of heated debate despite four decades of investigation [3,6,4850], and rhomboid was no exception to this inevitability. The only candidate for a histidine-stabilizing residue based on sequence conservation was the asparagine residue on TM2. Asparagine residues function within catalytic triads of cysteine proteases and some lipases. But the importance of the asparagine residue for rhomboid activity was contradictory from the beginning: the asparagine residue was important for activity of Drosophila rhomboid-1 [9], but not for human RHBDL2 (rhomboid, veinlet-like 2) [9]. Subsequent analyses showed it to be required in some rhomboid proteases, but not others, and this did not correlate with whether the enzyme was studied in vivo [9,24,29,51] or in vitro [30]. This mystery had to wait until the structure revealed its position; the asparagine residue was not bonded to the histidine base, but was potentially in a position to stabilize the oxyanion (Figure 3). A notable minority of serine proteases use catalytic dyads [50].

What was completely unexpected was the visualization of a tyrosine residue, one helical turn underneath the catalytic serine residue, stacking on to the catalytic histidine ring (Figure 3). The tyrosine residue would thereby serve to orientate the histidine ring, which is thought to be one of several postulated functions of the third residue in catalytic triads [3]. Accordingly, a tyrosine (or phenylalanine) residue at this position is well-conserved among rhomboid proteases (Figure 1), and its mutation reduces protease activity by 4–5-fold both in vitro and in living E. coli cells [41,42].

It must be stressed that the current rhomboid structures are of a ‘resting’ protease. In the absence of structures with bound substrate, transition-state mimics and/or inhibitors, it is not possible to delineate the mechanistic details of the catalytic events. In particular, the key question of what triggers abstraction of the proton from the serine residue by the histidine base remains a mystery, as it does for other serine proteases. In this light, rhomboid proteases present a new and radically different, albeit challenging, context for investigating this central question of enzyme catalysis.

RHOMBOID PROTEASE FUNCTION AND DYNAMICS

The molecular architecture of rhomboid proteins illustrates the enzymatic complexity of intramembrane proteolysis: the hydrated catalytic core is submerged beneath the presumed membrane surface, but is held separate from membrane lipids by protein segments. How does a TM substrate, which normally resides in membrane lipid, gain access to the internal active site of rhomboid?

This problem emphasizes the importance of understanding protein dynamics in rhomboid catalysis, and this to a large extent remains the principal challenge in the field. Recent structure–function enzyme analyses and computer simulations have taken the first steps towards outlining these key issues. The answer lies in a complex, and as yet poorly understood, interplay between rhomboid protease gating, substrate unwinding and membrane lipid dynamics.

Substrate gating: finding a way in

Substrate gating served as a starting point to begin addressing rhomboid protein dynamics: part of the rhomboid protein needs to move to open a path for substrate entry. Addressing how this might occur was initially facilitated by the availability of multiple crystal structures revealing different enzyme conformations, but two contradictory models were put forward (Figure 4). The first structure to be solved revealed a compact molecule with TM segments packing against each other along the perimeter of the enzyme, except between TM1 and TM3. Here, a gap exists between the tops of these segments owing to their V-shaped orientation, but this gap is plugged by insertion of the overlying L1 loop. Because this gap was large enough to accommodate a polypeptide chain, and given its conserved and unusual nature, the L1 loop was assumed to be the lateral substrate gate [34,52,53].

Structural and functional analyses of GlpG substrate gating

Figure 4
Structural and functional analyses of GlpG substrate gating

Models for substrate entry posited either lifting away of the L1 loop (boxed in top left), or displacement of TM5 and the overlying Cap (shown already in the open position). In addition to the conformational differences between structures, B factors also serve indirectly to suggest regions of conformational flexibility. The three structures are coloured by B factor values (see underlying spectrum). In the first structure (2IC8, first two views), the B factors are relatively low throughout, whereas in subsequent structures the higher B factors tend to centre on TM5 and connecting loops. Bottom panels summarize effects of GlpG mutants on protease activity from a screen to identify gating residues [41]; from an analysis of ~50 mutants engineered in all TMs and outer loops, mutants that increased protease activity only localized in the TM5, TM2 and Cap interface (in yellow). Locations of targeted mutants that did not increase activity are shown in magenta (the catalytic serine residue is circled in white). In the middle of the lower panel, the interface that prohibits substrate access is shown in the ‘closed’ form, with protruding sidechains and their electron density in yellow or orange (for Trp236). The bottom right panel magnifies the interface formed by Trp236 (TM5), Phe245 (Cap) and Phe153 (TM2). The greatest stimulatory effect of any single mutant within GlpG occurs when Trp236 (shown in orange) is mutated to alanine, resulting in a ~6-fold (~6×) increase in enzyme activity, compared with an approximate doubling (~2×) when either Phe153 or Phe245 are mutated to alanine. White brackets summarize the increase in protease activity relative to wild-type of all pairwise double mutants [42]: note that only the F153A+W236A mutant combination is synergistic, resulting in an ~11-fold increase in protease activity.

Figure 4
Structural and functional analyses of GlpG substrate gating

Models for substrate entry posited either lifting away of the L1 loop (boxed in top left), or displacement of TM5 and the overlying Cap (shown already in the open position). In addition to the conformational differences between structures, B factors also serve indirectly to suggest regions of conformational flexibility. The three structures are coloured by B factor values (see underlying spectrum). In the first structure (2IC8, first two views), the B factors are relatively low throughout, whereas in subsequent structures the higher B factors tend to centre on TM5 and connecting loops. Bottom panels summarize effects of GlpG mutants on protease activity from a screen to identify gating residues [41]; from an analysis of ~50 mutants engineered in all TMs and outer loops, mutants that increased protease activity only localized in the TM5, TM2 and Cap interface (in yellow). Locations of targeted mutants that did not increase activity are shown in magenta (the catalytic serine residue is circled in white). In the middle of the lower panel, the interface that prohibits substrate access is shown in the ‘closed’ form, with protruding sidechains and their electron density in yellow or orange (for Trp236). The bottom right panel magnifies the interface formed by Trp236 (TM5), Phe245 (Cap) and Phe153 (TM2). The greatest stimulatory effect of any single mutant within GlpG occurs when Trp236 (shown in orange) is mutated to alanine, resulting in a ~6-fold (~6×) increase in enzyme activity, compared with an approximate doubling (~2×) when either Phe153 or Phe245 are mutated to alanine. White brackets summarize the increase in protease activity relative to wild-type of all pairwise double mutants [42]: note that only the F153A+W236A mutant combination is synergistic, resulting in an ~11-fold increase in protease activity.

Conversely, other structures, resulting from different crystallization conditions, revealed major conformational differences in TM5 and the L5 Cap connected to it, rather than in the L1 loop [33,35]. The TM5 tilted laterally away from the core by approx. 35 ° at its top and the Cap moved upwards, opening a potential path for substrate entry on the side opposite the L1 loop. This ‘open’ conformation was not observed initially because in the first structure TM5 was involved in crystal packing against a neighbouring rhomboid molecule in the asymmetric unit. However, the physiological significance of this ‘open’ conformer was equally unclear, because all structures were solved in the absence of membrane, which would normally apply a lateral pressure on the enzyme that might hold TM5 in place. Moreover, substrate entry from the TM5 direction seemed unlikely, because it would place the substrate into the active site in the opposite orientation compared with nearly all serine proteases (discussed below) [38].

Crystal structures only capture stable conformations as snapshots of a dynamic enzyme. Ultimately, an enzymatic approach was used to evaluate these alternatives [41]. The rationale relied on the induction that gate opening would be a slow rate-limiting step; therefore mutating the contacts that stabilize the gate, wherever it might be, in the closed conformation might lead to gate opening more readily and therefore rhomboid variants with higher protease activity. To avoid bias on the one hand, and ‘one-hit wonders’ on the other, ~50 different mutants of rhomboid were engineered that disrupted key interactions within all TM segments and outer loops in multiple ways (Figure 4) [41].

The analysis converged on the interface between TM5 and TM2 that is mediated by interdigitating large hydrophobic residues as the substrate gate (Figure 4). Mutants in all other parts of the enzyme, especially those that disrupted the stability or interactions with the L1 loop, resulted in decreased enzyme activity [41]. Recent molecular dynamics simulations also support structural roles for the L1 loop, both within the enzyme and how the enzyme sits in the membrane [54]. Conversely, not one but three classes of engineered variants in TM5 displayed dramatically enhanced protease activity: reducing the size of the three residues on the helical face of TM5 that interfaces with TM2, changing pairwise interacting residues on TM5 and TM2 to alanine, or placing a proline residue on the opposite side of TM5 to kink it outwards but without affecting the TM2:TM5 interface (Figure 4). All of these classes increased enzyme activity compared with the wild-type enzyme, from 4-fold for the triple valine mutant on TM5 to a full order of magnitude increase when the uppermost residues on TM5 and TM2 were changed to alanine in a double F153A+W236A mutant [41].

What is equally striking is how a simple perturbation can produce an enzyme with dramatically increased activity: the strongest single activating mutant identified was W236A on TM5 (Figure 4), which by itself increased activity ~6-fold [42]. Other neighbouring residues failed to have such activating effects, including Phe245, which emanates from the overlying L5 Cap and dips to a position by Trp236 (Figure 4). Moreover, synergistic effects were observed only when mutations on TM5 were combined with those on TM2, rather than with the L5 Cap phenylalanine residue (Figure 4) [42]. These observations suggest that upper residues of TM5 have the greatest effect on substrate gating. Remarkably similar stimulatory effects were observed when these gate-open mutants were analysed and quantified in living E. coli cells, arguing that the gate-open mutations both have the same effect in natural cell membranes [42].

It is important to realize that a defining feature of a gate is that it acts to limit the activity of an enzyme. In this regard, mutations in TM5 are strongly activating, whereas those in the L5 Cap are weak, arguing that TM5 movement, not Cap movement, acts as the rate-limiting gate [41,42]. This is also consistent with recent molecular dynamics simulations that indicate the L5 Cap to be highly mobile [54]. As such, the L5 Cap may not be able to occlude substrate access as would be required of a gate, and this may explain why mutations in the L5 Cap are only weakly activating. In contrast with the mobility of the loops, the TM segments are much more stable, but notably TM5 is unique as it is the only TM that is not hydrogen bonded, thus appearing to be poised to open. Beyond these observations, it should be noted that the timescale of current simulations is too limited to provide any information on the interchange between various conformers.

Although these structure–function enzymatic analyses provide a compelling model, little is known about the gating mechanism in detail. In particular, the order of events in substrate binding, gate opening and substrate access remain entirely unknown. Even the extent of the motions required for substrate access and how far they extend into the membrane are unclear [36]. Motion in the TM5/L5 Cap region is likely to be a continuum, and the extent of gate opening may depend on the requirements of the substrate being cleaved. Nevertheless, direct visualization of a phospholipid headgroup in the GlpG active site in one structure suggests that exposure of the active site to lipid deeper within the membrane may not be that uncommon or unfavourable [35].

Substrate specificity and diversity

The notion of a rhomboid TM segment tilting at its top to clear a path for substrate entry into the active site is also attractive because it fits well with previously defined substrate requirements [27]. Unlike most soluble proteases, rhomboid proteins display a notable lack of dependence on a particular sequence or single residue in substrates. Instead, rhomboid proteases rely primarily (but not exclusively) on residues that disrupt α-helical structure. This property implies that TM helix unwinding to facilitate entry into the active site is a defining requirement for rhomboid intramembrane proteolysis [27].

Substrate specificity was first assessed with Drosophila rhomboid-1 and Spitz, the first identified natural enzyme–substrate pair, but these principles were found to be widely applicable [24,27,28]. Chimaeras between cleaved and uncleaved proteins revealed that neither the extracellular nor cytosolic regions were required for substrate recognition; specificity relies on only the topmost seven TM residues of Spitz (Figure 5), which alone are sufficient to convert other type I TM proteins into rhomboid substrates. Extensive mutagenesis of the seven-residue substrate region revealed two determinants [27]: there was a requirement for the first two residues to be small and hydrophilic, whereas distal residues had to have low helical propensity. This helix-breaking glycine-alanine pair in Spitz could be moved, but its placement was constrained and correlated with remaining on the same face of the helix [27]. These general features were ultimately evaluated with a panel of over 15 other rhomboid proteases across evolution (including GlpG) both in vivo and in vitro [24,25,27], and in general were found to apply to most rhomboid enzymes, although some rhomboid proteases display specializations [18].

Substrate requirements for rhomboid processing

Figure 5
Substrate requirements for rhomboid processing

The first identified rhomboid substrate, the Drosophila EGF ligand Spitz, is depicted in the membrane. Note that structural information is available for only the EGF domain (PDB: 1MOX). The ‘substrate motif’ in Spitz that is necessary and sufficient for rhomboid cleavage is highlighted in yellow. The right panel shows all of the cleavage sites, depicted with lightning bolts, currently known for different rhomboid protease substrates (excluding the mitochondrial rhomboid substrates). Approximate TM and substrate regions are depicted, while residues experimentally determined to act as helix-breaking are in red font. The cleavage sites for the top seven substrates were determined with natural proteins, whereas the bottom two are artificial substrates. The artificial substrates are chimaeric proteins harbouring the TM segments from the E. coli LacY permease and Drosophila Gurken. Note that in natural substrates, the cleavage site always follows an alanine residue. Organism designations are: Dm, D. melanogaster; Tg, T. gondii; Pf, P. falciparum; and Ps, Providencia stuartii.

Figure 5
Substrate requirements for rhomboid processing

The first identified rhomboid substrate, the Drosophila EGF ligand Spitz, is depicted in the membrane. Note that structural information is available for only the EGF domain (PDB: 1MOX). The ‘substrate motif’ in Spitz that is necessary and sufficient for rhomboid cleavage is highlighted in yellow. The right panel shows all of the cleavage sites, depicted with lightning bolts, currently known for different rhomboid protease substrates (excluding the mitochondrial rhomboid substrates). Approximate TM and substrate regions are depicted, while residues experimentally determined to act as helix-breaking are in red font. The cleavage sites for the top seven substrates were determined with natural proteins, whereas the bottom two are artificial substrates. The artificial substrates are chimaeric proteins harbouring the TM segments from the E. coli LacY permease and Drosophila Gurken. Note that in natural substrates, the cleavage site always follows an alanine residue. Organism designations are: Dm, D. melanogaster; Tg, T. gondii; Pf, P. falciparum; and Ps, Providencia stuartii.

GlpG substrate specificity has been interrogated in greater detail using a chimaeric substrate harbouring the second TM segment from the E. coli LacY permease [55]. This is not a physiological substrate for any rhomboid, but serves as a useful surrogate because it is cleaved with high efficiency. A notable strength of these studies is that they were conducted with endogenous GlpG, by comparing effects in wild-type cells with the GlpG-deleted strain, and confirmed in vitro with purified enzyme. What is particularly remarkable is that this independent approach with a completely different substrate found all of the same features applied: the upper cleavage site required small residues, whereas the only distal glycine residue (Figure 5), presumably analagous to the Spitz glycine-alanine, was exquisitely sensitive to mutation: replacing only this glycine residue with a cysteine abolished cleavage [46].

Interestingly, two additional properties were clarified through these analyses. First, a second helix-breaking motif was discovered further down the helix (Figure 5), because the structure of the LacY permease showed that TM2 bends around a helix-breaking QP (glutamine-proline) pair [56]. Mutation of this QP pair in LacY reduced cleavage, whereas placing it into other substrates enhanced cleavage, explaining why the LacY TM segment is such an efficient substrate for GlpG compared with Spitz or other natural substrates [55]. Similar to the glycine-alanine in Spitz, the QP could be moved, but its position was constrained to the same face of the TM helix. This requirement could be partially overcome if the cleavage site was also moved, arguing that it is not a particular face of the helix by itself, but rather the placement of the helix break relative to the cleavage site that is important.

Secondly, properties of the cleavage site in LacY were also examined in greater detail [55]. A genetic screen identified eight pairs of residues flanking the cleavage site that supported efficient proteolysis. The clear preference was for small residues preceding the scissile bond (P1 position), and negatively charged residues after the cleavage site (P1′ position). The cleavage site in LacY TM2 lies at the extracellular–TM junction, rather than being embedded within the membrane. Indeed, cysteine residues introduced into this region could be labelled with a membrane-impermeable agent that reacts with thiols [46]. This is consistent with residues essential for rhomboid cleavage being at the top of TM segments (Figure 5), where water is known to permeate the lipid headgroup layer of membranes [57]. Nevertheless, the position of the cleavage site in the substrate probably does not reflect where cleavage occurs relative to the membrane: the complex geometry of the internal active-site cavity of GlpG, which lies a lateral distance from the initial point of contact with the substrate, blurs the distinction of where precisely the cleavage is received within the substrate molecule. Indeed, the cleavage site probably represents how far each polypeptide has to travel as it bends into the membrane-recessed active site.

Although these studies define general principles underlying rhomboid substrate requirements, the specific details should be treated with caution; all of the analyses were conducted with artificial substrates, since the natural substrates of GlpG remain unknown [29,51]. This artificial context could have unintended differences. For example, all known cleavage sites for rhomboid enzymes analysed in their natural environment invariantly follow an alanine residue in the P1 position, and never has a charged residue been found in the P1′ position (Figure 5) [17,19,5860]. Thus the residues that flank the cleavage site in natural substrates under physiological conditions may have properties that are different from those experienced in current assays. Although not yet complete, all current evidence agrees that substrate dynamics via helix unwinding, in addition to gate opening in rhomboid proteases, is an integral part of intramembrane proteolysis, and underlies its specificity.

Protein dynamics and conformational transitions

Perhaps the most enigmatic aspects of rhomboid function currently centre on the general dynamics of the protein. The only study in this area comes from recent molecular dynamics simulations [54]. Although in the simulations GlpG is globally stable overall, it is intrinsically dynamic, even wiggly, compared with other membrane proteins. For example, the catalytic serine residue moves up and down by several angstroms during the simulations, and multiple hydrogen bonds within the molecule are redistributed. The role of the observed flexibility in catalysis remains unknown, but warrants further scrutiny. Moreover, there appear to be two main hydrogen bond clusters within the molecule, with TM3 being shared by the two. One possible implication of this arrangement is that TM3 could relay structural conformation changes from one part of the molecule to the other. Since the structure with a lipid trapped within the active site served as the subject for the simulations, it remains to be seen whether the observed properties are common to all conformers, and whether the dynamics change when lipid is removed or substrate is bound.

Beyond the general motions within the enzyme, little is known about the conformational changes that accompany protease activity. Current molecular dynamics simulations are not instructive because the limited timeframe is insufficient to observe interchange between different conformations.

MEMBRANE IMMERSION OF A PROTEASE

In addition to being serine proteases, rhomboid proteins are integral membrane proteins with an unusual shape, and this reality raises the challenge of membrane biophysics to understanding enzyme function. Yet rhomboid proteins have been studied as proteases first, and only recently as membrane proteins.

Experimental reconstitution of rhomboid proteins into different phospholipid environments in vitro was found to have profound consequences on activity [28]. What is particularly intriguing is the diverse response of rhomboid enzymes to different reconstitution lipids: the activity of some rhomboid enzymes is dramatically increased in almost any type of phospholipid, whereas others display a clear preference for particular lipids, but these preferences are different among rhomboid proteins. In fact, the type of lipid used in reconstitution was found to be the condition that has the greatest influence on protease activity. These experiments indicate that the way in which rhomboid proteases are positioned into the membrane is likely to be an important factor in their enzymatic function, and could potentially constitute a regulatory function [28].

The asymmetric, and unprecedented, shape of detergent-solubilized GlpG made it impossible to predict from structural information alone how it might be positioned in a cellular membrane [34,37]. Recent molecular dynamics simulations have started to address this complex issue [54]. The simulations were conducted with one structure as a starting point, and in POPE (1-palmytoyl-2-oleoyl-sn-glycero-3-phosphatidylethanolamine), a natural component of bacterial membranes, and POPC (1-palmytoyl-2-oleoyl-sn-glycero-3-phosphatidylcholine), which E. coli cannot synthesize. In the simulations, GlpG sits tilted in the membrane (Figure 6). This orientation is the result of many interactions between the enzyme and surrounding lipids, as mediated primarily by hydrogen bonding of residue side chains to lipid headgroups. The L1 loop is responsible for most of the lipid interactions on the upper surface of the membrane, whereas interactions to the cytosolic surface of the membrane are mediated primarily by positively charged residues, which are concentrated on the lower surface of the enzyme. The outcome is GlpG tilting ~12 ° from the membrane normal in POPE, with the L1 loop acting as a ‘flotation device’ (Figure 6). The tilt is slightly greater in a POPC membrane, owing mainly to altered interactions with the L1 loop residues, but it should be noted that GlpG displays low enzymatic activity when reconstituted into phosphatidylcholine lipids [28].

GlpG in a membrane: protease tilt and membrane thinning

Figure 6
GlpG in a membrane: protease tilt and membrane thinning

A stylized diagram of GlpG tilt and membrane thinning as suggested by molecular dynamics simulations. Interactions with membrane lipids, including through the L1 loop acting as a ‘flotation device’, serve to stabilize GlpG in a position tilted ~12 ° from the membrane normal. This results in slightly deeper submersion of the catalytic serine residue (in yellow) and TM5. The hydrophobic mismatch between GlpG and membrane also causes deformation of the membrane in the immediate vicinity of GlpG, resulting in a ~4 Å thinner lipid annulus. The most extreme thinning occurs on the cytosolic side under the L1 loop. Enzyme tilt and membrane thinning may play roles in how substrates approach and are handled by rhomboid proteases.

Figure 6
GlpG in a membrane: protease tilt and membrane thinning

A stylized diagram of GlpG tilt and membrane thinning as suggested by molecular dynamics simulations. Interactions with membrane lipids, including through the L1 loop acting as a ‘flotation device’, serve to stabilize GlpG in a position tilted ~12 ° from the membrane normal. This results in slightly deeper submersion of the catalytic serine residue (in yellow) and TM5. The hydrophobic mismatch between GlpG and membrane also causes deformation of the membrane in the immediate vicinity of GlpG, resulting in a ~4 Å thinner lipid annulus. The most extreme thinning occurs on the cytosolic side under the L1 loop. Enzyme tilt and membrane thinning may play roles in how substrates approach and are handled by rhomboid proteases.

Many membrane proteins are known to alter the membrane environment in their vicinity [61]. The asymmetric shape and relative sturdiness of the TM core, coupled with the relatively thinner belt of hydrophobic residues that encircle GlpG, also forces lipids in the simulations to reorient their positions to the shape of GlpG [54]. This results in membrane thinning around the first two or three shells of lipid surrounding the enzyme (Figure 6). Although an extreme thinning to 20 Å was extrapolated from the crystal structure [37], the actual membrane thickness around GlpG observed in the simulations is nearly twice that estimate, approx. 35 Å for POPE [54]. Not surprisingly, the thinning is most dramatic under the partially submerged L1 loop, but overall the result is a 4 Å thinning around GlpG. This represents a ~10% thinning in comparison with the measured 39 Å glycerol–glycerol distance of membrane POPE lipids.

A necessary caveat of these simulations is that they were conducted in membranes composed entirely of a single phospholipid, POPE or POPC, to simulate the complexity of natural membranes. Cellular membranes are known to be composed of hundreds of different lipid species that vary in headgroup, acyl chain length, and degree and position of saturation [62]. It remains to be seen whether mixtures of different lipids might allow lipid species not present in the simulations to adopt preferential positions around GlpG in which the longer POPE lipids were forced to distort. Some early experimental support for this possibility comes from the observation of lipids bound to different parts of the enzyme in the crystal structures, which appear to be of sufficient affinity to be maintained throughout the detergent extraction, purification and crystallization procedure [35]. To what degree the tilt in the simulations is affected by compression of longer lipid remains to be tested. Simulations using more physiological populations of membrane lipids, and further experimental analysis of GlpG directly in lipid membranes using biophysical techniques, could further refine the current view of how lipids interact with the enzyme.

Despite these important caveats, tilting of the molecule in the membrane, coupled with a ~4 Å thinner lipid annulus, has a number of possible mechanistic implications (Figure 6). First, the active serine residue may actually be immersed even deeper into the plane of the membrane than previously anticipated: in the POPE membrane simulation, the nucleophilic oxygen lies only 5 Å above the middle of the bilayer [54]. Nevertheless, the active site is well hydrated throughout the simulations, owing to predominance of hydrophilic residue side chains within the cavity and prevalence of lipid headgroups.

Secondly, membrane thinning is known to change the tilt angle of TM helices. As such, thinning of the membrane as the substrate approaches the rhomboid enzyme may be important for positioning the substrate for cleavage. Another scenerio suggests that the upper few residues of the substrate may exit the membrane altogether [36], which was also proposed by early studies on site-2 protease [63]. The implications of these models on the mechanism of intramembrane proteolysis have yet to be evaluated functionally.

NOT YOUR TYPICAL SERINE PROTEASE: OUTSTANDING QUESTIONS

At least two further glaring discrepancies remain when comparing rhomboid to other serine proteases. The active-site chemistry of serine proteases has been exploited for the development of many classes of inhibitor warheads over nearly half a century of research, some with therapeutic applications [64]. Secondly, studying the biological roles of serine proteases has emphasized that protease activity is exquisitely regulated at the protein level; serine proteases are seldom made in active form [2]. On both levels, rhomboid proteases could not be more different from their soluble serine protease cousins: no selective inhibitors of rhomboid proteases exist, and no direct mechanism of regulation is known for any rhomboid protease.

Prospects for rhomboid inhibition

Protease inhibitors have served as exquisite probes for studying protease mechanisms, as well as being invaluable tools for pharmacological approaches for uncovering the roles proteases play in different processes and organisms [65]. These approaches are particularly important when studying such an unusual membrane-immersed active site and a family of proteases with hundreds of members conserved in all forms of life, yet whose functions are understood in less than a dozen organisms [13]. Unfortunately, rhomboid proteases have proven to be unexpectedly tricky in the area of inhibitor design.

Initial inhibitor profiling in living cells revealed that rhomboid protease activity is sensitive to only a small subset of serine protease inhibitors [9]. Many serine protease inhibitors are either highly cytotoxic, unstable or membrane impermeant, therefore a reasonable fear was that this initial inhibitor profile was incomplete despite being carefully controlled. Yet development of a defined pure enzyme reconstitution system conclusively demonstrated that only isocoumarins are consistent and potent inhibitors of all rhomboid proteases tested [28]. Although not all classes of inhibitors have yet been exhausted, classical serine protease warheads including sulfonyl fluorides {e.g. PMSF AEBSF [4-(2-aminoethyl)benzenesulfonyl fluoride]} and phosphonates [e.g. DFP (di-isopropyl fluorophonate)], frustratingly proved unable to exert any apparent effect on rhomboid activity even when preincubated in millimolar concentrations with pure enzyme. Unfortunately, isocoumarins are masked electrophiles that are unstable and quite cytotoxic, which limits their application in cellular investigations.

The limited reactivity of rhomboid with serine protease inhibitors was suspected to result from a failure of the compounds to gain access to the protease active site. However, the ensuing rhomboid structures showed that the catalytic machinery is open to the outside environment through a large aqueous cavity [3335,38]. In fact, chemical accessibility probes subsequently demonstrated that a cysteine residue in place of the catalytic serine residue can be directly labelled with both membrane-permeable and membrane-impermeable thiol-reactive compounds of a similar size as serine protease inhibitors [46].

The failure of serine protease inhibitor prototypes to influence rhomboid activity may result from its unusual active-site stereochemistry. One unexpected implication of the gating mechanism is that the substrate polypeptide approaches the catalytic apparatus from the ‘si’ face which is stereochemically opposite to most serine proteases. Indeed, α/β-hydrolases, which include β-lactamases, lipases and bacterial leader peptidases, also attack their substrates from the ‘si’ face and are refractory to inhibition by classical serine protease inhibitors [66]. The only active-site-directed warheads that were found to inhibit leader peptidase are β-lactams, and these compounds deserve consideration for testing as rhomboid inhibitor prototypes.

Compounds that do not depend on catalytic co-operation from rhomboid may provide an alternate and more likely source of inhibitors. Substrate-mimicking helical peptides are capable of high potency as competitive inhibitors of both γ-secretase and signal peptide peptidase [67,68]. Preliminary investigation indicates that helical peptides can also be developed as rhomboid protease inhibitors (S. Urban, unpublished work). These encouraging early-generation compounds warrant further experimental development and scrutiny.

Future advances in understanding the mechanistic peculiarities of rhomboid proteases is likely to provide new and more informed avenues for inhibitor development. Such compounds will undoubtedly be of high utility in future mechanistic and biological investigations, and could even find eventual applications in the therapeutic realm [16].

Regulation

A notable contradiction in our understanding of rhomboid function is the apparent lack of mechanisms for their regulation. Unlike all other intramembrane proteases, rhomboid proteins cleave full-length proteins and therefore do not depend on a prior cleavage by a site-1 protease [29,69]. It is this site-1 cleavage that usually serves as a point of regulation for intramembrane proteolysis [10,70].

Currently, the only sources of regulation for rhomboid catalysis rely on transcription of the rhomboid gene or physical separation of substrate from protease in the cell [71]. The best example of transcriptional regulation is the rhomboid-1 protease during Drosophila embryogenesis, where its strong and dynamic mRNA expression pattern has long served as an atlas for EGF signalling [72]. Separation of substrate and protease is exemplified by Toxoplasma and Plasmodium parasites, in which adhesin substrates are segregated in internal organelles, whereas rhomboid is restricted to the surface of the parasite [19,20,73]. Only at the end of the invasion programme do adhesins encounter a rhomboid protease at the parasite surface as a way of dismantling the junction formed between the host cell and invading parasite.

It is unusual for a catalytic activity to be regulated so indirectly, and it remains possible that current examples reflect circumstances in which regulation has become specialized to meet a particular need, rather than being representative of rhomboid regulation in general. This is underscored by attempting to apply current paradigms to GlpG. E. coli GlpG derives its name from being in the glycerol-3-phosphate (Glp) operon [74], but this gene structure is not conserved among bacteria, and GlpG is not thought to function in glycerol metabolism. Therefore its transcription profile could be inappropriate for regulating its activity. Separation of substrate from protease in different compartments is also unlikely to be a source of rhomboid regulation in prokaryotic organisms.

The challenge is to understand how a membrane-embedded active site can be regulated. One possibility is regulation by the membrane environment itself, and this has been demonstrated experimentally for many rhomboid proteases, but only in vitro [28]. It remains to be seen whether this potential mechanism is indeed used in a cellular context and, if so, whether it is a more common mechanism for these proteases as a family. In reality, one significant but under-appreciated obstacle to understanding rhomboid regulation is the lack of clues due to the limited biological examples where rhomboid function is defined [13].

CONCLUDING PERSPECTIVES

Despite sharing some catalytic properties with their soluble counterparts as a result of convergent evolution, intramembrane proteases are a different class of enzymes that are only beginning to be studied mechanistically. This area is particularly exciting because the enzymatic properties of intramembrane proteases are likely to be fundamentally different as a result of their immersion in the membrane. Perhaps the most conspicuous issue has been how water is delivered to the active site, but past this apparent structural novelty, more interesting and less understood are the unique properties that the unusual milieu of the membrane imparts on these enzymes. Realizing a sophisticated understanding of these properties requires detailed investigation from multiple disciplines, and for the first time the tools for this analysis are becoming available. As these exciting new avenues are undertaken, two fundamental caveats must be taken into account.

There is an inherent danger in studying the detailed mechanism of an enzyme such as GlpG whose cellular function, and even natural substrate, remain entirely unknown [29,51]. For example, all substrates that have been used to interrogate GlpG specificity have been heterologous [28,55], whereas our current view of GlpG position in a membrane is based on simulations with POPE, a lipid in which GlpG displays low activity [28] (S. M. Moin and S. Urban, unpublished work). Indeed, phosphatidylethanolamine is known as a non-bilayer lipid because it cannot form a membrane bilayer by itself due to its shape. In all of these investigations, the greater challenge is to be certain that the exact parameters that are being measured are directly relevant to the natural environment and physiological function of GlpG.

It is also important to avoid the temptation of extrapolating lessons learned with GlpG to other rhomboid proteins without deliberate experimental evaluation. Eukaryotic rhomboid proteases could have important structural differences, because they contain a seventh TM segment that can either precede or follow the six TM core, as well as longer extramembraneous domains [11]. Moreover, GlpG reacts unlike any other rhomboid protease to reconstitution into different lipid environments [28]. Mechanistic diversity among rhomboid proteases has also already been encountered: P. falciparum rhomboid-4 that has been implicated in parasite invasion cannot cleave canonical substrates such as Spitz [18], whereas radically different substrate cleavage site and requirements have been proposed for the mitochondrial rhomboid [75]. Indeed, given the fact that rhomboid proteins can even be hard to recognize based on basic sequence analysis alone [76], variation in their biochemical properties should be expected. On an optimistic note, these differences could hold opportunities to target pathogenic compared with homoeostatic rhomboid enzymes differentially for therapeutic intervention. Detailed, but cautious, biochemical interrogation may well be the best strategy for tackling the therapeutic promise that these core enzymes might hold for human health.

Abbreviations

     
  • EGF

    epidermal growth factor

  •  
  • POPC

    1-palmytoyl-2-oleoyl-sn-glycero-3-phosphatidylcholine

  •  
  • POPE

    1-palmytoyl-2-oleoyl-sn-glycero-3-phosphatidylethanolamine

  •  
  • QP pair

    glutamine-proline pair

  •  
  • RHBDL2 rhomboid

    veinlet-like 2

  •  
  • TM

    transmembrane

I thank the members of my laboratory, collaborators and the many colleagues worldwide for making this field such a vibrant and exciting area of research.

FUNDING

Research in my laboratory is funded through grants from the National Institutes of Health [grant number R01AI066025], and career awards from the Burroughs-Wellcome Fund, the Packard Foundation and the Howard Hughes Medical Institute.

References

References
1
Lopez-Otin
C.
Overall
C. M.
Protease degradomics: a new challenge for proteomics
Nat. Rev. Mol. Cell Biol.
2002
, vol. 
3
 (pg. 
509
-
519
)
2
Lopez-Otin
C.
Bond
J. S.
Proteases: multifunctional enzymes in life and disease
J. Biol. Chem.
2008
, vol. 
283
 (pg. 
30433
-
30437
)
3
Polgar
L.
The catalytic triad of serine peptidases
Cell. Mol. Life Sci.
2005
, vol. 
62
 (pg. 
2161
-
2172
)
4
Blow
D. M.
Birktoft
J. J.
Hartley
B. S.
Role of a buried acid group in the mechanism of action of chymotrypsin
Nature
1969
, vol. 
221
 (pg. 
337
-
340
)
5
Blow
D.
So do we understand how enzymes work?
Structure
2000
, vol. 
8
 (pg. 
R77
-
R81
)
6
Blow
D. M.
The tortuous story of Asp…His… Ser: structural analysis of α-chymotrypsin
Trends Biochem. Sci.
1997
, vol. 
22
 (pg. 
405
-
408
)
7
Wolfe
M. S.
Xia
W.
Ostaszewski
B. L.
Diehl
T. S.
Kimberly
W. T.
Selkoe
D. J.
Two transmembrane aspartates in presenilin-1 required for presenilin endoproteolysis and γ-secretase activity
Nature
1999
, vol. 
398
 (pg. 
513
-
517
)
8
Rawson
R. B.
Zelenski
N. G.
Nijhawan
D.
Ye
J.
Sakai
J.
Hasan
M. T.
Chang
T. Y.
Brown
M. S.
Goldstein
J. L.
Complementation cloning of S2P, a gene encoding a putative metalloprotease required for intramembrane cleavage of SREBPs
Mol. Cell
1997
, vol. 
1
 (pg. 
47
-
57
)
9
Urban
S.
Lee
J. R.
Freeman
M.
Drosophila rhomboid-1 defines a family of putative intramembrane serine proteases
Cell
2001
, vol. 
107
 (pg. 
173
-
182
)
10
Brown
M. S.
Ye
J.
Rawson
R. B.
Goldstein
J. L.
Regulated intramembrane proteolysis: a control mechanism conserved from bacteria to humans
Cell
2000
, vol. 
100
 (pg. 
391
-
398.
)
11
Koonin
E. V.
Makarova
K. S.
Rogozin
I. B.
Davidovic
L.
Letellier
M. C.
Pellegrini
L.
The rhomboids: a nearly ubiquitous family of intramembrane serine proteases that probably evolved by multiple ancient horizontal gene transfers
Genome Biol.
2003
, vol. 
4
 pg. 
R19
 
12
Kinch
L. N.
Ginalski
K.
Grishin
N. V.
Site-2 protease regulated intramembrane proteolysis: sequence homologs suggest an ancient signaling cascade
Protein Sci.
2006
, vol. 
15
 (pg. 
84
-
93
)
13
Urban
S.
Rhomboid proteins: conserved membrane proteases with divergent biological functions
Genes Dev.
2006
, vol. 
20
 (pg. 
3054
-
3068
)
14
Urban
S.
Lee
J. R.
Freeman
M.
A family of rhomboid intramembrane proteases activates all membrane-tether EGF ligands in Drosophila
EMBO J.
2002
, vol. 
21
 (pg. 
4277
-
4286
)
15
Urban
S.
Brown
G.
Freeman
M.
EGF receptor signaling protects smooth-cuticle cells from apoptosis during Drosophila ventral epidermis development
Development
2004
, vol. 
131
 (pg. 
1835
-
1845
)
16
Urban
S.
Making the cut: central roles of intramembrane proteolysis in pathogenic microorganisms
Nat. Rev. Microbiol.
2009
, vol. 
7
 (pg. 
411
-
423
)
17
Stevenson
L. G.
Strisovsky
K.
Clemmer
K. M.
Bhatt
S.
Freeman
M.
Rather
P. N.
Rhomboid protease AarA mediates quorum-sensing in Providencia stuartii by activating TatA of the twin-arginine translocase
Proc. Natl. Acad. Sci. U.S.A.
2007
, vol. 
104
 (pg. 
1003
-
1008
)
18
Baker
R. P.
Wijetilaka
R.
Urban
S.
Two Plasmodium rhomboid proteases preferentially cleave different adhesins implicated in all invasive stages of malaria
PLoS Pathog.
2006
, vol. 
2
 pg. 
e113
 
19
O'Donnell
R. A.
Hackett
F.
Howell
S. A.
Treeck
M.
Struck
N.
Krnajski
Z.
Withers-Martinez
C.
Gilberger
T. W.
Blackman
M. J.
Intramembrane proteolysis mediates shedding of a key adhesin during erythrocyte invasion by the malaria parasite
J. Cell Biol.
2006
, vol. 
174
 (pg. 
1023
-
1033
)
20
Brossier
F.
Jewett
T. J.
Sibley
L. D.
Urban
S.
A spatially localized rhomboid protease cleaves cell surface adhesins essential for invasion by Toxoplasma
Proc. Natl. Acad. Sci. U.S.A.
2005
, vol. 
102
 (pg. 
4146
-
4151
)
21
Baxt
L. A.
Baker
R. P.
Singh
U.
Urban
S.
An Entamoeba histolytica rhomboid protease with atypical specificity cleaves a surface lectin involved in phagocytosis and immune evasion
Genes Dev.
2008
, vol. 
22
 (pg. 
1636
-
1646
)
22
Edbauer
D.
Winkler
E.
Regula
J. T.
Pesold
B.
Steiner
H.
Haass
C.
Reconstitution of γ-secretase activity
Nat. Cell Biol.
2003
, vol. 
5
 (pg. 
486
-
488
)
23
Kimberly
W. T.
LaVoie
M. J.
Ostaszewski
B. L.
Ye
W.
Wolfe
M. S.
Selkoe
D. J.
γ-Secretase is a membrane protein complex comprised of presenilin, nicastrin, Aph-1, and Pen-2
Proc. Natl. Acad. Sci. U.S.A.
2003
, vol. 
100
 (pg. 
6382
-
6387
)
24
Urban
S.
Schlieper
D.
Freeman
M.
Conservation of intramembrane proteolytic activity and substrate specificity in eukaryotic and prokaryotic Rhomboids
Curr. Biol.
2002
, vol. 
12
 (pg. 
1507
-
1512
)
25
Kanaoka
M. M.
Urban
S.
Freeman
M.
Okada
K.
An Arabidopsis Rhomboid homolog is an intramembrane protease in plants
FEBS Lett.
2005
, vol. 
579
 (pg. 
5723
-
5728
)
26
Gallio
M.
Sturgill
G.
Rather
P.
Kylsten
P.
A conserved mechanism for extracellular signaling in eukaryotes and prokaryotes
Proc. Natl. Acad. Sci. U.S.A.
2002
, vol. 
99
 (pg. 
12208
-
12213
)
27
Urban
S.
Freeman
M.
Substrate specificity of rhomboid intramembrane proteases is governed by helix-breaking residues in the substrate transmembrane domain
Mol. Cell
2003
, vol. 
11
 (pg. 
1425
-
1434
)
28
Urban
S.
Wolfe
M. S.
Reconstitution of intramembrane proteolysis in vitro reveals that pure rhomboid is sufficient for catalysis and specificity
Proc. Natl. Acad. Sci. U.S.A.
2005
, vol. 
102
 (pg. 
1883
-
1888
)
29
Maegawa
S.
Ito
K.
Akiyama
Y.
Proteolytic action of GlpG, a rhomboid protease in the Escherichia coli cytoplasmic membrane
Biochemistry
2005
, vol. 
44
 (pg. 
13543
-
13552
)
30
Lemberg
M. K.
Menendez
J.
Misik
A.
Garcia
M.
Koth
C. M.
Freeman
M.
Mechanism of intramembrane proteolysis investigated with purified rhomboid proteases
EMBO J.
2005
, vol. 
24
 (pg. 
464
-
472
)
31
Feng
L.
Yan
H.
Wu
Z.
Yan
N.
Wang
Z.
Jeffrey
P. D.
Shi
Y.
Structure of a site-2 protease family intramembrane metalloprotease
Science
2007
, vol. 
318
 (pg. 
1608
-
1612
)
32
Urban
S.
Shi
Y.
Core principles of intramembrane proteolysis: comparison of rhomboid and site-2 family proteases
Curr. Opin. Struct. Biol.
2008
, vol. 
18
 (pg. 
432
-
441
)
33
Wu
Z.
Yan
N.
Feng
L.
Oberstein
A.
Yan
H.
Baker
R. P.
Gu
L.
Jeffrey
P. D.
Urban
S.
Shi
Y.
Structural analysis of a rhomboid family intramembrane protease reveals a gating mechanism for substrate entry
Nat. Struct. Mol. Biol.
2006
, vol. 
13
 (pg. 
1084
-
1091
)
34
Wang
Y.
Zhang
Y.
Ha
Y.
Crystal structure of a rhomboid family intramembrane protease
Nature
2006
, vol. 
444
 (pg. 
179
-
180
)
35
Ben-Shem
A.
Fass
D.
Bibi
E.
Structural basis for intramembrane proteolysis by rhomboid serine proteases
Proc. Natl. Acad. Sci. U.S.A.
2007
, vol. 
104
 (pg. 
462
-
466
)
36
Wang
Y.
Ha
Y.
Open-cap conformation of intramembrane protease GlpG
Proc. Natl. Acad. Sci. U.S.A.
2007
, vol. 
104
 (pg. 
2098
-
2102
)
37
Wang
Y.
Maegawa
S.
Akiyama
Y.
Ha
Y.
The role of L1 loop in the mechanism of rhomboid intramembrane protease GlpG
J. Mol. Biol.
2007
, vol. 
374
 (pg. 
1104
-
1113
)
38
Lemieux
M. J.
Fischer
S. J.
Cherney
M. M.
Bateman
K. S.
James
M. N.
The crystal structure of the rhomboid peptidase from Haemophilus influenzae provides insight into intramembrane proteolysis
Proc. Natl. Acad. Sci. U.S.A.
2007
, vol. 
104
 (pg. 
750
-
754
)
39
Lilley
B. N.
Ploegh
H. L.
A membrane protein required for dislocation of misfolded proteins from the ER
Nature
2004
, vol. 
429
 (pg. 
834
-
840
)
40
Ye
Y.
Shibata
Y.
Yun
C.
Ron
D.
Rapoport
T. A.
A membrane protein complex mediates retro-translocation from the ER lumen into the cytosol
Nature
2004
, vol. 
429
 (pg. 
841
-
847
)
41
Baker
R. P.
Young
K.
Feng
L.
Shi
Y.
Urban
S.
Enzymatic analysis of a rhomboid intramembrane protease implicates transmembrane helix 5 as the lateral substrate gate
Proc. Natl. Acad. Sci. U.S.A.
2007
, vol. 
104
 (pg. 
8257
-
8262
)
42
Urban
S.
Baker
R. P.
In vivo analysis reveals substrate-gating mutants of a rhomboid intramembrane protease display increased activity in living cells
Biol. Chem.
2008
, vol. 
389
 (pg. 
1107
-
1115
)
43
Senes
A.
Ubarretxena-Belandia
I.
Engelman
D. M.
The Cα—H···O hydrogen bond: a determinant of stability and specificity in transmembrane helix interactions
Proc. Natl. Acad. Sci. U.S.A.
2001
, vol. 
98
 (pg. 
9056
-
9061
)
44
Senes
A.
Gerstein
M.
Engelman
D. M.
Statistical analysis of amino acid patterns in transmembrane helices: the GxxxG motif occurs frequently and in association with β-branched residues at neighboring positions
J. Mol. Biol.
2000
, vol. 
296
 (pg. 
921
-
936
)
45
Li
J.
Ahvazi
B.
Szittner
R.
Meighen
E.
Mutation of the nucleophilic elbow of the Lux-specific thioesterase from Vibrio harveyi
Biochem. Biophys. Res. Commun.
2000
, vol. 
275
 (pg. 
704
-
708
)
46
Maegawa
S.
Koide
K.
Ito
K.
Akiyama
Y.
The intramembrane active site of GlpG, an. E. coli rhomboid protease, is accessible to water and hydrolyses an extramembrane peptide bond of substrates
Mol. Microbiol.
2007
, vol. 
64
 (pg. 
435
-
447
)
47
Lemberg
M. K.
Bland
F. A.
Weihofen
A.
Braud
V. M.
Martoglio
B.
Intramembrane proteolysis of signal peptides: an essential step in the generation of HLA-E epitopes
J. Immunol.
2001
, vol. 
167
 (pg. 
6441
-
6446
)
48
Carter
P.
Wells
J. A.
Dissecting the catalytic triad of a serine protease
Nature
1988
, vol. 
332
 (pg. 
564
-
568
)
49
Dodson
G.
Wlodawer
A.
Catalytic triads and their relatives
Trends Biochem. Sci.
1998
, vol. 
23
 (pg. 
347
-
352
)
50
Ekici
O. D.
Paetzel
M.
Dalbey
R. E.
Unconventional serine proteases: variations on the catalytic Ser/His/Asp triad configuration
Protein Sci.
2008
, vol. 
17
 (pg. 
2023
-
2037
)
51
Clemmer
K. M.
Sturgill
G. M.
Veenstra
A.
Rather
P. N.
Functional characterization of Escherichia coli GlpG and additional rhomboid proteins using an aarA mutant of Providencia stuartii
J. Bacteriol.
2006
, vol. 
188
 (pg. 
3415
-
3419
)
52
White
S. H.
Rhomboid intramembrane protease structures galore!
Nat. Struct. Mol. Biol.
2006
, vol. 
13
 (pg. 
1049
-
1051
)
53
Freeman
M.
Structural biology: enzyme theory holds water
Nature
2006
, vol. 
444
 (pg. 
153
-
155
)
54
Bondar
A. N.
del Val
C.
White
S. H.
Rhomboid protease dynamics and lipid interactions
Structure
2009
, vol. 
17
 (pg. 
395
-
405
)
55
Akiyama
Y.
Maegawa
S.
Sequence features of substrates required for cleavage by GlpG, an Escherichia coli rhomboid protease
Mol. Microbiol.
2007
, vol. 
64
 (pg. 
1028
-
1037
)
56
Abramson
J.
Smirnova
I.
Kasho
V.
Verner
G.
Kaback
H. R.
Iwata
S.
Structure and mechanism of the lactose permease of Escherichia coli
Science
2003
, vol. 
301
 (pg. 
610
-
615
)
57
White
S. H.
Wimley
W. C.
Hydrophobic interactions of peptides with membrane interfaces
Biochim. Biophys. Acta
1998
, vol. 
1376
 (pg. 
339
-
352
)
58
Zhou
X. W.
Blackman
M. J.
Howell
S. A.
Carruthers
V. B.
Proteomic analysis of cleavage events reveals a dynamic two-step mechanism for proteolysis of a key parasite adhesive complex
Mol. Cell. Proteomics
2004
, vol. 
3
 (pg. 
565
-
576
)
59
Opitz
C.
Di Cristina
M.
Reiss
M.
Ruppert
T.
Crisanti
A.
Soldati
D.
Intramembrane cleavage of microneme proteins at the surface of the apicomplexan parasite Toxoplasma gondii
EMBO J.
2002
, vol. 
21
 (pg. 
1577
-
1585
)
60
Howell
S. A.
Hackett
F.
Jongco
A. M.
Withers-Martinez
C.
Kim
K.
Carruthers
V. B.
Blackman
M. J.
Distinct mechanisms govern proteolytic shedding of a key invasion protein in apicomplexan pathogens
Mol. Microbiol.
2005
, vol. 
57
 (pg. 
1342
-
1356
)
61
Phillips
R.
Ursell
T.
Wiggins
P.
Sens
P.
Emerging roles for lipids in shaping membrane-protein function
Nature
2009
, vol. 
459
 (pg. 
379
-
385
)
62
Munro
S.
Lipid rafts: elusive or illusive?
Cell
2003
, vol. 
115
 (pg. 
377
-
388
)
63
Ye
J.
Dave
U. P.
Grishin
N. V.
Goldstein
J. L.
Brown
M. S.
Asparagine-proline sequence within membrane-spanning segment of SREBP triggers intramembrane cleavage by site-2 protease
Proc. Natl. Acad. Sci. U.S.A.
2000
, vol. 
97
 (pg. 
5123
-
5128
)
64
Walker
B.
Lynas
J. F.
Strategies for the inhibition of serine proteases
Cell. Mol. Life Sci.
2001
, vol. 
58
 (pg. 
596
-
624
)
65
Paulick
M. G.
Bogyo
M.
Application of activity-based probes to the study of enzymes involved in cancer progression
Curr. Opin. Genet. Dev.
2008
, vol. 
18
 (pg. 
97
-
106
)
66
Paetzel
M.
Dalbey
R. E.
Strynadka
N. C.
Crystal structure of a bacterial signal peptidase in complex with a β-lactam inhibitor
Nature
1998
, vol. 
396
 (pg. 
186
-
190
)
67
Bihel
F.
Das
C.
Bowman
M. J.
Wolfe
M. S.
Discovery of a subnanomolar helical D-tridecapeptide inhibitor of γ-secretase
J. Med. Chem.
2004
, vol. 
47
 (pg. 
3931
-
3933
)
68
Sato
T.
Nyborg
A. C.
Iwata
N.
Diehl
T. S.
Saido
T. C.
Golde
T. E.
Wolfe
M. S.
Signal peptide peptidase: biochemical properties and modulation by nonsteroidal antiinflammatory drugs
Biochemistry
2006
, vol. 
45
 (pg. 
8649
-
8656
)
69
Lee
J. R.
Urban
S.
Garvey
C. F.
Freeman
M.
Regulated intracellular ligand transport and proteolysis control EGF signal activation in Drosophila
Cell
2001
, vol. 
107
 (pg. 
161
-
171
)
70
Urban
S.
Freeman
M.
Intramembrane proteolysis controls diverse signalling pathways throughout evolution
Curr. Opin. Genet. Dev.
2002
, vol. 
12
 (pg. 
512
-
518
)
71
Freeman
M.
Rhomboid proteases and their biological functions
Annu. Rev. Genet.
2008
, vol. 
42
 (pg. 
191
-
210
)
72
Bier
E.
Jan
L. Y.
Jan
Y. N.
rhomboid, a gene required for dorsoventral axis establishment and peripheral nervous system development in Drosophila melanogaster
Genes Dev.
1990
, vol. 
4
 (pg. 
190
-
203
)
73
Dowse
T. J.
Pascall
J. C.
Brown
K. D.
Soldati
D.
Apicomplexan rhomboids have a potential role in microneme protein cleavage during host cell invasion
Int. J. Parasitol.
2005
, vol. 
35
 (pg. 
747
-
756
)
74
Yang
B.
Larson
T. J.
Multiple promoters are responsible for transcription of the glpEGR operon of Escherichia coli K-12
Biochim. Biophys. Acta
1998
, vol. 
1396
 (pg. 
114
-
126
)
75
Herlan
M.
Bornhovd
C.
Hell
K.
Neupert
W.
Reichert
A. S.
Alternative topogenesis of Mgm1 and mitochondrial morphology depend on ATP and a functional import motor
J. Cell Biol.
2004
, vol. 
165
 (pg. 
167
-
173
)
76
Lemberg
M. K.
Freeman
M.
Functional and evolutionary implications of enhanced genomic analysis of rhomboid intramembrane proteases
Genome Res.
2007
, vol. 
17
 (pg. 
1634
-
1646
)