The crystal structure of the rhomboid peptidasefrom Haemophilus influenzae provides insightinto intramembrane proteolysisM. Joanne Lemieux*, Sarah J. Fischer, Maia M. Cherney, Katherine S. Bateman, and Michael N. G. James*

Group in Protein Structure and Function, Department of Biochemistry, University of Alberta, Edmonton, AB, Canada T6G 2H7

Communicated by Robert M. Stroud, University of California, San Francisco, CA, November 10, 2006 (received for review October 24, 2006)

Rhomboid peptidases are members of a family of regulated in-tramembrane peptidases that cleave the transmembrane segmentsof integral membrane proteins. Rhomboid peptidases have beenshown to play a major role in developmental processes in Dro-sophila and in mitochondrial maintenance in yeast. Most recently,the function of rhomboid peptidases has been directly linked toapoptosis. We have solved the structure of the rhomboid peptidasefrom Haemophilus influenzae (hiGlpG) to 2.2-Å resolution. Thephasing for the crystals of hiGlpG was provided mainly by molec-ular replacement, by using the coordinates of the Escherichia colirhomboid (ecGlpG). The structural results on these rhomboid pep-tidases have allowed us to speculate on the catalytic mechanism ofsubstrate cleavage in a membranous environment. We have iden-tified the relative disposition of the nucleophilic serine to thegeneral base/acid function of the conserved histidine. Modeling atetrapeptide substrate in the context of the rhomboid structurereveals an oxyanion hole comprising the side chain of a secondconserved histidine and the main-chain NH of the nucleophilicserine residue. In both hiGlpG and ecGlpG structures, a watermolecule occupies this oxyanion hole.

intramembrane peptidase � membrane protein � rhomboid protease �x-ray crystallography

Rhomboid peptidases belong to family S54 of intramembraneserine peptidases (1); these enzymes carry out proteolysis ofsingle transmembrane substrates within the environment of thelipid bilayer. Rhomboid peptidases cleave their substrates in theouter leaflet of the lipid bilayer, thereby releasing an exocellularpeptide signal that can, in turn, play a role in cell signaling (forreview, see ref. 2). Although the catalytic mechanism for solubleserine peptidases has been well characterized, both biochemicallyand structurally, we are just beginning to understand the mecha-nism of serine peptidases in the membrane environment (3).

Although the rhomboids are a newly discovered family ofpeptidases, they have been identified in all kingdoms (4, 5). Theirdiverse functions are being revealed through genetic screens anddevelopmental and cell biology studies (6). Rhomboid pepti-dases have been shown to play a role in releasing EGF from themembrane environment. In Drosophila melanogaster, whererhomboids were first discovered (7), they are expressed in atemporal and a tissue specific manner to cleave three differentsubstrates: spitz, gurkin, and keren. Rhomboid cleavage of thesesubstrates occurs in the membrane-spanning helix, and thereleased peptide then interacts with the EGF receptor (8, 9).

More recently, the function of a mitochondrial rhomboidpeptidase, presenilin-associated rhomboid-like (PARL) pepti-dase, has been associated with cytochrome c release from themitochondria, thereby affecting apoptosis (10). In addition,mitochondrial PARL has been linked to insulin resistance andType II diabetes (11). The role of rhomboid peptidases inmitochondrial maintenance has been studied in detail in yeast(12, 13). In prokaryotes, rhomboid peptidases have been shownto be important for host cell invasion by Toxoplasma gondii (14).In addition, rhomboids have been shown to cleave apical mem-

brane antigen-1 (AMA1), known to play a role in the malariamerozoite stage of Plasmodium falciparum (15). In the patho-genic Providencia stuartii, rhomboids have been shown to cleaveand release a quorum-sensing factor (16, 17).

ResultsHaemophilus influenzae GlpG (hiGlpG) Structure. We have overex-pressed various prokaryotic homologs of rhomboid peptidase forstructural analysis, namely, hiGlpG, Escherichia coli GlpG(ecGlpG), and Bacillus subtilis YqgP. Crystals were obtained forboth hiGlpG and ecGlpG; those of hiGlpG diffracted to a higherresolution. The structure of hiGlpG was solved by molecularreplacement by using the atomic coordinates from ecGlpG (3).

The overall structure of hiGlpG consists of a six-�-helicalbundle (H1 to H6) (Fig. 1) with residues for the active sitelocated in a central cavity between H1 and H3 near the periplas-mic surface of the protein. A long conserved loop (L1) is locatedbetween the first and second transmembrane helices, H1 and H2,at the periplasmic side of the protein membrane, thus blockingaccess to the active site. The L1 loop has a prominent amphi-pathic nature (3); it extends beyond the 6-helical bundle into thehydrophobic environment of the lipid bilayer. A short two-residue loop connects H2 and H3 at the cytoplasmic end of themolecule. H3 and H4 are connected by a loop, L3, that extendsas a coil into the internal cavity in the center of the peptidase.At the catalytic serine residue, this random coil becomes �-helical like in bacterial subtilisin. H5 is unusual in that it is notfully �-helical but rather deformed. H5 does interact throughhydrophobic interactions with H2 and H4. H6 has the catalytichistidine near its N terminus, and it extends deep into thecytoplasmic space at its C terminus.

Bound Lipid Molecules. The detergent mixture present through-out the crystallization procedure was C12E8 (AnaPOE–C12E8)with lower amounts of dodecylmaltoside (DDM) remainingafter detergent exchange. The electron density for two mole-cules of C12E8 could be identified in the electron density maps.Both molecules pack between two hiGlpG molecules (Fig. 1 Aand B). The first C12E8 interacts with helices H3 and H6 on oneof the hiGlpG molecules and with H6 on the other. The secondC12E8 interacts with helices H1 and H2 on one hiGlpG

Author contributions: M.J.L., S.J.F., and M.M.C. designed research; M.J.L., S.J.F., M.M.C., andK.S.B. performed research; M.J.L., K.S.B., and M.N.G.J. analyzed data; and M.J.L., K.S.B., andM.N.G.J. wrote the paper.

The authors declare no conflict of interest.

Abbreviations: hiGlpG, H. influenzae GlpG; ecGlpG, E. coli GlpG; PA, phosphatidic acid.

Data deposition: The coordinates for the H. influenzae GlpG structure have been depositedin the Protein Data Bank, www.pdb.org (PDB ID code 2NR9).

*To whom correspondence may be addressed. E-mail: michael.james@ualberta.ca ormlemieux@ualberta.ca.

This article contains supporting information online at www.pnas.org/cgi/content/full/0609981104/DC1.

© 2007 by The National Academy of Sciences of the USA

750–754 � PNAS � January 16, 2007 � vol. 104 � no. 3 www.pnas.org�cgi�doi�10.1073�pnas.0609981104

Dow

nloa

ded

by g

uest

on

July

9, 2

021

http://www.pnas.org/cgi/content/full/0609981104/DC1http://www.pnas.org/cgi/content/full/0609981104/DC1

molecule and with H2 on the other. There are three lipidmolecules visible in the electron density map that we havepresently interpreted as phosphatidic acid (PA); the phospho-ryl group has significantly higher electron density than theother atoms in the glycerol backbone and acyl chains. (Figs. 1A and B and 2). PA has been shown to provide a source of thesignaling lipid diacylglycerol (18). It is possible that these lipidsmay be cardiolipin (CL), phosphatidylserine (PS) or phos-phatidylethanolamine (PE), the main lipids found in the E. colilipid bilayer. Two PA molecules can be seen f lanking the L1loop, a loop that is proposed to be f lexible in order forsubstrate to bind (see below and ref. 3). Another PA is locatednear H6. It is unclear at present whether these lipids merelyplay a structural role, acting as chaperones, or whether theyplay a role in rhomboid function.

Comparison of hiGlpG and ecGlpG Structures. Superimposition ofthe C� atoms of the hiGlpG and ecGlpG structures reveals thathiGlpG and ecGlpG share a common fold as predicted, with anrmsd of 1.09 Å for 137 equivalent C� atom pairs. An overlay ofthe structures demonstrates this agreement (Fig. 3). The only

areas that appear to have variability are the L1 loop, locatedbetween H1 and H2, and the H5 helix. In fact, we see relativelyweak electron density for H5, suggesting some conformationalvariability for this helix. On the other hand, we see density forthe C terminus of the molecule that has allowed us to build H6further into the cytoplasm than that for ecGlpG. The C terminusof hiGlpG actually extends beyond the lipid bilayer boundary,similar to that seen with the GlpT structure (19).

Rhomboid Active Site. Most serine peptidases are characterized bya catalytic triad consisting of the nucleophilic O� atom of theserine, a general acid/base (the imidazole ring of a histidine) thatassists in the deprotonation of the nucleophile and an asparticacid carboxylate that helps to maintain an optimal position forthe imidazole ring of the histidine during catalysis. Importantly,and in addition to these groups, there is an oxyanion hole, afeature that provides electrophilic assistance to the nucleophilicattack by the serine O� on the carbonyl carbon atom of thescissile bond. Rhomboid has these features as well, but themolecular details are different.

In the native unbound form of rhomboid, the residues com-prising the catalytic triad and the oxyanion hole are buried andinaccessible to solvent (Fig. 4). Three loops, L1, L3 and L5, haveto move substantially in order that the substrate can gain accessto the active site (3). The B-factors for these loops are high,ranging from 60 Å2 to 70 Å2 for L1, 50 Å2 for L3, and 60 Å2 forL5, supporting this hypothesis. The overall B-factor for hiGlpGis 35.5 Å2. A possible sequence of events could include: thedestabilized area of the transmembrane helix of the substrate(20) docks to rhomboid displacing the L1 gate (Gly-29 to Ser-55in hiGlpG). The L3 loop (Gly-109 to Gly-114) and the L5 loop(Gly-161 to Gly-165) are then displaced. The beginning andending residues of these loops are defined principally by glycineresidues that have large associated conformational f lexibility,which facilitates moving these three loops to provide access forthe substrate to the active site (Fig. 4).

DiscussionRhomboid Mechanism. We have explored the binding of a shortsegment of a substrate polypeptide, spitz, by superimposing the sidechain of His-169 and O� of Ser-116 from hiGlpG onto the structureof chymotrypsin complexed to the turkey ovomucoid third domain,OMTKY3 (1CHO.pdb) (21). This superposition, although not veryaccurate, gives a reasonable position for the segment of substratepolypeptide from P2 to P2� [nomenclature of Schechter and Berger

Fig. 1. Structure of hiGlpG. (A) Cartoon representation of hiGlpG colored in a rainbow fashion, with the N terminus being dark blue and the C terminus beingred. Three lipids, PA (cyan) and two detergent molecules, C12E8 (yellow), can be seen associated with hiGlpG. (B) View A rotated 90°.

Fig. 2. Electron density for lipid. 2Fo�Fc electron density for a lipid molecule,PA, bound between loop L1 and helix H3.

Lemieux et al. PNAS � January 16, 2007 � vol. 104 � no. 3 � 751

BIO

CHEM

ISTR

Y

Dow

nloa

ded

by g

uest

on

July

9, 2

021

(22)], fitting into the rhomboid active site (Fig. 4). The substrate hasbeen modeled as four residues surrounding the cleavage site ofspitz, Ala-Ser-Gly-Ala (20). The cleavage specificities of severalrhomboids indicate a preference for residues, having small sidechains, such as glycine and alanine (3).

The water molecule in our structure, W40 (Fig. 4A), whichbridges Ser116NH and His65N�2H and forms H bonds toPhe113O and Val162O, is located in the oxyanion hole; thiswater molecule would be displaced by the carbonyl-oxygen atomof the P1 residue of the substrate, thus placing the P1 carbonyl-carbon atom in an ideal position for nucleophilic attack bySer116O�. The nucleophilic attack is facilitated by the transferof the proton on O� to the N�2 atom of the imidazole ring ofHis-169. A good hydrogen bond from Ser116O�. . . N�2 His-169is already established in the native enzyme, thereby defining thepathway for this proton transfer. The oxyanion hole in rhomboidhiGlpG comprises the main-chain NH of Ser-116 and theprotonated N�2 of the imidazole of His-65. This favorabletautomeric form for the imidazole ring of His-65 is ensured bythe main-chain hydrogen bond from Leu61NH to N�1 of His-65(Fig. 4). Not only does this oxyanion hole stabilize the developingnegative charge on the carbonyl-oxygen atom in the tetrahedralintermediate, but it also assists the nucleophilic attack byelectrophilically enhancing the polarization of the carbonyl(C—O) bond of the substrate. Mutagenesis experiments havedemonstrated that, in ecGlpG and YqgP, the residues equivalentto His-169 and Ser-116 from hiGlpG are essential for activity(23). In addition, mutagenesis in YqgP of the residue equivalentto His 65 in hiGlpG, which is also highly conserved, resulted inmarkedly reduced activity.

In hiGlpG there is no direct equivalent to the third memberof the catalytic triad, i.e., an AspCOO2�, or an AsnCONH2. InecGlpG, the side-chain amide of Asn-251 forms a hydrogenbond to a water molecule, and the water, in turn, forms ahydrogen bond to N�1 of His-254, the general base in thatenzyme. hiGlpG has no equivalent water molecule, and theclose contact of the imidazole ring of His-169 to Phe-155 likely

precludes the binding of water or the side chain of Asn-166. Amore likely candidate for a third member of the catalytic triadis the packing of the side chain of Tyr-120 against the imidazolering of His-169, thereby ensuring some stability in the positionof the general base (3).

It has not escaped our attention that the nucleophilic Ser-116is at the beginning of helix H4 and that the NH of Ser-116 wouldform a good hydrogen bond to the oxyanion of the tetrahedralintermediate. Therefore, additional stabilization of the oxyanionwill come from the partial positive charge on the helix dipole ofhelix H4. This situation is analogous to the peptide dipolestabilization provided by the helix bearing the Ser-221 in bac-terial subtilisin (24). Subtilisin has the side-chain amide fromAsn-155 that contributes to the formation of the oxyanion hole.Rhomboid is similar to subtilisin, but it uses a histidine side chainrather than an asparagine carboxamide.

We have solved the structure of a full-length rhomboidpeptidase, namely hiGlpG. The superimposed structures ofhiGlpG and ecGlpG are similar; however, their active sites havesubtle differences and our structure has allowed us to identify theoxyanion hole, an essential feature for serine peptidases that isrequired to propose confidently a sound catalytic mechanism.Perhaps the most unexpected result of our structural analysis isto find the catalytic machinery of a soluble globular serine andthe general peptidase like �-chymotrypsin or subtilisin �9 Åbelow the surface of the bilayer. It is also surprising to find acatalytic site in hiGlpG resembling that of bacterial subtilisinwith the nucleophilic serine and general base histidine residuesnear the N termini of two separate �-helices. Similarly, theoxyanion hole is formed by the side chain of an asparagine insubtilisin. There is clearly an evolutionary relationship betweenthese two families of serine peptidases.

Materials and MethodsCloning and Expression. H. influenzae DNA was purchased fromAmerican Type Culture Collection (Manassas, VA). With theuse of restriction digestion, PCR products were then ligated

Fig. 3. Superimposition of hiGlpG and ecGlpG rhomboid structures. Wall-eyed stereoview of structural superimposition of rhomboid structures. hiGlpG is shownin magenta, and ecGlpG is shown in green. Structural alignment was carried out with ALIGN (27). See supporting information (SI) Fig. 5 for sequence alignment.

752 � www.pnas.org�cgi�doi�10.1073�pnas.0609981104 Lemieux et al.

Dow

nloa

ded

by g

uest

on

July

9, 2

021

http://www.pnas.org/cgi/content/full/0609981104/DC1

into pBAD-MycHisA vector (Invitrogen, Burlington, Canada).Expression was carried out in Top10 cells (Invitrogen) inLuria–Bertani medium supplemented with ampicillin. Induc-tion of the various expression constructs were carried out asdescribed below with arabinose.

Membrane Fraction Isolation. Cells were grown to an OD600 of 0.4and induced with 0.0002% arabinose at 24°C for 16 h. Cells wereharvested at 12,227 � g for 10 min by using an AvantiJ1.8000rotor (Beckman, Fullerton, CA). Cells were resuspended in fourvolumes of TBS supplemented with an EDTA-free peptidase-

inhibitor mixture (NEB, Beverly, MA), 1 mM PMSF, and 0.1mg/ml DNase and lysed by using an TEmulsiFlex-C3T, (Avestin,Ottawa, Canada). Unbroken cells were pelleted in a JA17 rotorat 10,000 � g for 20 min. Membrane fractions were collected byultracentrifugation in a L8–80 ultracentrifuge at 100,000 � g ina 45Ti rotor (Beckman).

Protein Purification. Membrane fractions were homogenized in 50mM Tris, 300 mM NaCl, 30 mM imidazole, 20% glycerol, and 1%DDM (pH 8.0). The solution was stirred for 30 min, followed byultracentrifugation for 30 min at 110,000 � g in a 45Ti rotor

Fig. 4. Mechanism of H. influenzae GlpG. (A) Top view from the periplasmic space of hiGlpG. Residues in the active site are labeled in gray. Water (WAT40)located in the active site between Ser-116 and His-65 is shown in red. (B) Similar view to A but with loops L1, L3, and L5 removed. A model of the D. melanogasterrhomboid substrate spitz (magenta) has been docked manually into the active site. (C) Wall-eyed stereoview of spitz docked into the active site of hiGlpG. Thehydroxyl oxygen of S116 is hydrogen boding with H169. H169 is stabilized by Y120. The carbonyl oxygen from the P1 residue is stabilized by H65 and the backboneamide of S116, forming the oxyanion hole. (D) Proposed catalytic mechanism of hiGlpG (ChemDraw).

Lemieux et al. PNAS � January 16, 2007 � vol. 104 � no. 3 � 753

BIO

CHEM

ISTR

Y

Dow

nloa

ded

by g

uest

on

July

9, 2

021

(Beckman, USA). The supernatant was incubated with Ni-NTAresin (Qiagen, Ontario, Canada) for 2 h. The resin was thencollected and washed with 20 column volumes (CV) of 50 mMTris, 300 mM NaCl, 30 mM imidazole, 20% glycerol, and 0.1%DDM (pH 8.0), followed by 20 CV of the above stated bufferwith 35 mM imidazole. Protein fractions were eluted in astep-wise manner with 3 � 2 CV of the above-described buffercontaining 250, 500, and 1,000 mM imidazole. Protein was thenconcentrated by using a 30K centrifugal filter (Millipore, Bed-ford, MA) and subjected to detergent exchange on a Spe-hadex200 (16/60) column (Amersham, Piscataway, NJ) in 50 mMTris (pH 8.0), 0.05% C12E8, 20 mM NaCl, and 10% glycerol (seeSI Fig. 6).

Crystallographic Analysis. Protein for crystallization was concen-trated by using Millipore Ultrafree centrifugal concentrators, 30kDa molecular mass cutoff to a concentration of 5 mg/ml.Crystals of hiGlpg were obtained in 25% PEG 4000, 0.1 M citrate(pH 6.0), 1 M NaCl, 3% ethanol, and 15% glycerol and grew todimensions of 100 �m � 50 �m � 20 �m in 1–2 weeks. Crystalswere directly f lash-cooled in liquid nitrogen. Data were collectedat the Advanced Light Source beam line 8.3.1. Two differentspace groups were obtained: monoclinic C2 and orthorhombicP212121. The monoclinic data diffracted to higher resolution and

had one molecule in the AU. Molecular replacement was carriedout by using MolRep (25) with the ecGlpG coordinates. Re-finement was carried out with Refmac5 (26). See SI Table 1 forstatistical results.

We thank Dr. Ya Ha for generously making available the E. coli GlpGcoordinates (2IC8.pdb); Drs. James Holden (ALS, BL 8.3.1), ErnstBergmann, and Jonathan Parrish for assistance with data collection; allmembers of the M.N.G.J. laboratory for their support; and WendyKasinec for administrative assistance. X-ray diffraction data were col-lected at beam line 8.3.1 of the Advanced Light Source (ALS) at theLawrence Berkeley National Laboratory, Berkeley, CA, under an agree-ment with the Alberta Synchrotron Institute (ASI). The ALS is operatedby the Department of Energy and supported by the National Institutesof Health (Bethesda, MD). Beam line 8.3.1 was funded by the NationalScience Foundation, the University of California, and Henry Wheeler.The ASI synchrotron access program is supported by grants from theAlberta Science and Research Authority (ASRA) and the AlbertaHeritage Foundation for Medical Research (AHFMR). This work hasbeen supported by the Canadian Institute for Health Research (CIHR).M.J.L is supported by a fellowship scholarships from CHIR, CIHRstrategic training program for membrane proteins and cardiovasculardisease, and Alberta Heritage Foundation for Medical Research.M.N.G.J. acknowledges support from the Canada Research Chair’sProgram.

1. Rawlings ND, Morton FR., Barrett AJ (2006) Nucleic Acids Res 34:D270–D272.

2. Urban S, Freeman M (2002) Curr Opin Genet Dev 12:512–518.3. Wang Y, Zhang Y, Ha Y (2006) Nature 444:153–155.4. Wasserman JD, Urban S, Freeman M (2000) Genes Dev 14:1651–1663.5. Pascall JC, Brown KD (1998) FEBS Lett 429:337–340.6. Freeman M (2004) Nat Rev Mol Cell Biol 5:188–197.7. Urban S, Lee JR, Freeman M (2001) Cell 107:173–182.8. Lee JR, Urban S, Garvey CF, Freeman M (2001) Cell 107:161–171.9. Urban S, Lee JR, Freeman M (2002) EMBO J 21:4277–4286.

10. Cipolat S, Rudka T, Hartmann D, Costa V, Serneels L, Craessaerts K, MetzgerK, Frezza C, Annaert W, D’Adamio L, et al. (2006) Cell 126:163–175.

11. Walder K, Kerr-Bayles L, Civitarese A, Jowett J, Curran J, Elliott K, TrevaskisJ, Bishara N, Zimmet P, Mandarino L, et al. (2005) Diabetologia 48:459–468.

12. Herlan M, Vogel F, Bornhovd C, Neupert W, Reichert AS (2003) J Biol Chem278:27781–27788.

13. McQuibban GA, Saurya S, Freeman M (2003) Nature 423:537–541.14. Brossier F, Jewett TJ, Sibley LD, Urban S (2005) Proc Natl Acad Sci USA

102:4146–4151.

15. Howell SA, Hackett F, Jongco AM, Withers-Martinez C, Kim K, CarruthersVB, Blackman MJ (2005) Mol Microbiol 57:1342–1356.

16. Urban S, Schlieper D, Freeman M (2002) Curr Biol 12:1507–1512.17. Rather PN, Ding X, Baca-DeLancey RR, Siddiqui S (1999) J Bacteriol

181:7185–7191.18. Dillon DA, Wu WI, Riedel B, Wissing JB, Dowhan W, Carman GM (1996)

J Biol Chem 271:30548–30553.19. Huang Y, Lemieux MJ, Song J, Auer M, Wang DN (2003) Science 301:616–620.20. Urban S, Freeman M (2003) Mol Cell 11:1425–1434.21. Fujinaga M, Sielecki AR, Read RJ, Ardelt W, Laskowski M, Jr, James MN

(1987) J Mol Biol 195:397–418.22. Schechter I, Berger A (1967) Biochem Biophys Res Commun 27:157–162.23. Lemberg MK, Menendez J, Misik A, Garcia M, Koth CM, Freeman M (2005)

EMBO J 24:464–472.24. Hol WG, van Duijnen PT, Berendsen HJ (1978) Nature 273:443–446.25. Vagin A, Teplyakov A (1997) J Appl Crystallogr 30:1022–1025.26. Steiner RA, Lebedev AA, Murshudov GN (2003) Acta Crystallogr D 59(Pt

12):2114–2124.27. Cohen GH (1997) J Appl Crystallogr 30:1160–1161.

754 � www.pnas.org�cgi�doi�10.1073�pnas.0609981104 Lemieux et al.

Dow

nloa

ded

by g

uest

on

July

9, 2

021

http://www.pnas.org/cgi/content/full/0609981104/DC1http://www.pnas.org/cgi/content/full/0609981104/DC1