doi:10.1016/j.jmb.2004.10.050 J. Mol. Biol. (2005) 345, 211–227

Proprotein Convertase Models based on the Crystal Structures of Furin and Kexin: Explanation of their Specificity

Stefan Henrich1, Iris Lindberg2, Wolfram Bode1* and Manuel E. Than1*

1Max-Planck-Institut fu¨r In eukaryotes, many secreted proteins and peptide hormones are excised Biochemie, Abteilung fu¨r from larger precursors by calcium-dependent serine proteinases, the Strukturforschung, Am proprotein/prohormone convertases (PCs). These PCs cleave their protein Klopferspitz 18, 82152 substrates very specifically following multiple basic residues. The seven Martinsried, Germany mammalian PCs and their yeast orthologue kexin are multi-domain proteinases consisting of a subtilisin-related catalytic domain, a conserved 2Dept. of Biochemistry & P-domain and a variable, often cysteine-rich domain, which in some PCs is Molecular Biology, Louisiana followed by an additional C-terminal trans-membrane domain and a short State University Health cytoplasmic domain. The recently published crystal structures of the Sciences Center, New Orleans soluble mouse furin and yeast kexin ectodomains have revealed the LA 70112, USA relative arrangement of catalytic and P domains, the exact domain fold and the detailed architecture of the substrate binding clefts. Based on these experimental structures, we now have modelled the structures of the other human/mouse PCs. According to topology and to structure-based sequence comparisons, these other PCs closely resemble furin, with PC4, PACE4 and PC5/6 being more similar, and PC1/3, PC2 and PC7 being less similar to furin. Except for PC1 and PC2, this order of similarity is valid for the catalytic as well as for the P domains, and is almost reversed using kexin as a reference molecule. A similar order results from the number and clustering of negative charges lining the non-prime subsites, explaining the gradually decreasing requirement for basic residues N-terminal to substrate cleavage sites. The preference of the different PCs for distinct substrates seems to be governed by overall charge compensation and matching of the detailed charge distribution pattern. q 2004 Elsevier Ltd. All rights reserved. Keywords: proprotein convertases (PCs); furin; kexin; substrate specificity; *Corresponding authors homology modelling

Introduction calcium-dependent serine endopeptidases, which very specifically cleave C-terminal to multiple basic Many secreted proteins and peptides essential for residues. Because of their subtilisin-like catalytic the regulation of biological activity are initially domain,7,8 these proteinases have been called synthesized as inactive precursor proteins, and are subtilisin-like proprotein convertases (SPCs) or subsequently proteolytically converted in the proprotein convertases (PCs). Besides a number of secretory pathway to the mature active forms.1–6 non-mammalian enzymes, such as the yeast process- 9–12 Most of the proteolytic enzymes involved are ing kexin/Kex2 proteinase, this PC family (being part of the MEROPS family S8 of clan SB)13 contains seven closely related mammalian enzymes, namely Present address: S. Henrich, EML Research gGmbH, furin/SPC1/PACE, PC2/SPC2, PC1/PC3/SPC3, Schloss-Wolfsbrunnenweg 33, 69118 Heidelberg, PACE4/SPC4, PC4/SPC5, PC5/PC6/SPC6, and Germany PC7/PC8/LPC. The more distantly related endo- Abbreviations used: PC, proprotein convertase; CMK, chloromethylketone; PDB, protein data bank; TGN, trans proteinases SKI-1/S1P and NARC-1, in contrast, are Golgi network; VRs, variable regions; SCRs, structurally predicted to exhibit a more pyrolisin-like and conserved regions. proteinase K-like fold, respectively, and to cleave 14 E-mail addresses of the corresponding authors: precursors C-terminal to non-basic amino acids. [email protected]; [email protected] The pro-PCs are multi-domain proteins, which all

0022-2836/$ - see front matter q 2004 Elsevier Ltd. All rights reserved. 212 Models and Specificity of Proprotein Convertases

Figure 1. Stereo plots of the ectodomains of furin and the other modelled PCs, shown in standard orientation, i.e. the active-site cleft horizontally extends across the catalytic domain surface with a substrate polypeptide chain running from left (N terminus) to right (C terminus). (a) Stereo ribbon plot of the experimental furin ectodomain.61 The b-strands are shown in red (catalytic domain) and rainbow colours (P domain), and the helices of the catalytic domain are given in yellow. The active-site residues (dark grey) and the decanoyl-Arg-Val-Lys-Arg-CMK inhibitor residues (light-coloured ball-and-stick model) are given with all non-hydrogen atoms, the three disulfide bridges as two yellow balls each, and the two bound calcium ions as pink spheres. (b) Stereo ribbon plot superposition of the six modelled PCs (PC1 gold, PC2 dark green, PC4 pink, PC5 light green, PC7 brown, PACE4 grey) and kexin (blue, PDB105 entry 1OT5),62 superimposed with furin (thick red rope, PDB105 entry 1P8J).61 The bound decanoyl-Arg-Val-Lys-Arg-CMK inhibitor (stick model) and the bound calcium ions Ca1 and Ca2 of furin (grey spheres) are shown in addition. Only a few residues have been labelled for orientation using the furin nomenclature. (c) Solid surface representation of furin, colored according to surface conservation in the PC family.61 The colours indicate sequentially and structurally conserved (red), sequentially conserved but structurally deviating (orange), structurally conserved (yellow), structurally deviating (green), and Models and Specificity of Proprotein Convertases 213 contain a pro-domain acting as an intra-molecular hand, PC2 and PC1/3 also efficiently cleave at sites chaperone in folding,15–18 an about 340 residue lacking additional basic residues beyond P1 and P2, subtilisin-like catalytic domain, an about 130 as does kexin.21,30–32 Distinct PCs often process residue P-domain important for proteolytic activity, their substrates differentially, as exemplified by the and a quite variable segment, which in furin, different cleavage products of pro-insulin and pro- PACE4 and PC5 (and less obviously in PC4 and glucagon upon hydrolysis by PC1/3 and PC2, PC7) contains a number of Cys residues and is thus respectively.1,4 often referred to in the literature as the Cys-rich Only a very few endogenous protein inhibitors domain. Full-length furin, the PC5/6B isoform, PC7 have been identified that are directed against the and kexin exhibit an additional trans-membrane PCs (see, e.g. Rockwell et al.21). Furin can be rapidly and a C-terminal cytoplasmic domain, i.e. are trans- inactivated by the cytosolic serpin inhibitor PI8, a membrane proteinases. PC1/3, PC2 and PC4 are fraction of which under physiological conditions soluble proteinases, lacking these latter domains might enter the secretory pathway to interact with but terminating in long Ser/Thr-rich tails instead. furin.33 By replacing the P1 and the P4 residue in PC1/3 and PC2 are predominantly expressed in the reactive-site loop by Arg residues, the serpin neuroendocrine cells, and PC4 is only found in a1-PI has been re-engineered into a1-Antitrypsin testicular and ovarian germ cells, while the other Portland/a1-PDX, which quite selectively forms mammalian PCs are more broadly or even ubiqui- SDS-stable complexes with furin, and reacts more tously distributed. The mammalian PCs are often weakly with PC6B, PC1/334,35 and PACE4.36 The simultaneously expressed in various tissues, where substrate-like binding inhibitors turkey ovomucoid they are mainly localized in the trans-Golgi network third domain37 and eglin c38,39 have been mutated to (TGN), but are also found in other compartments become potent inhibitors of furin and kexin. Very of the secretory pathways and on the cell surface. recently, a Drosophila Spn4 gene product has been Kexin is required for the endogenous generation identified as a potent furin-directed serpin.40,41 of mature a-factor and killer toxin,9,10,12,19 while the The isolated PC pro-domains and, less effectively, mammalian PCs are involved in the processing of smaller constructs containing the multiple basic the precursors of virtually all neuropeptides and “primary activation cleavage segment” (Lys-Arg- peptide hormones, growth and differentiation Arg-Thr-Lys-Arg107f-Y-Asp108f-Val in furin) can factors, cell-surface receptors, extracellular metallo- block the PC active sites, with some preference for proteinases, coagulation factors and other serum their endogenous PCs.18,42–46 The binding protein proteins, adhesion molecules, plasma proteins, 7B2, which is co-expressed with proPC2 and bacterial toxins such as the anthrax toxin protective facilitates maturation to an active enzyme species, antigen20 or diphtheria toxin, and viral coat proteins contains an inhibitory sequence at the C termi- such as from lethal Ebola viruses, HIV-1, or nus.47–50 In addition, a number of synthetic inhibi- cytomegalovirus.2,3,5,6 tors have been developed. Most of these exhibit The role of the PCs is to cleave these substrates polybasic peptidyl moieties mimicking substrate or accurately, even in the presence of an excess of PC pro-domain sequences,51–53 often linked to an similar other cleavage sites.21 In most substrates, the active-site directed group such as a chloromethyl- PCs cleave C-terminal to paired basic residues, with ketone (CMK), ketomethylene, phosphonate or the first so-called P2 residue often being a Lys, and boronic acid group.21,29,54–56 Oligopeptides consist- 52 the second P1 residue strongly or even strictly ing of up to ten L-Arg or D-Arg residues strongly restrained to an Arg residue (using the nomencla- block furin activity, but inhibit the other mamma- ture of Schechter and Berger,22 with P1, P2 etc. and lian PCs much more weakly, allowing some P10,P20 etc. defining the peptide substrate residues discrimination.57 N and C-terminal to the scissile peptide bond, In order to assign the experimental observations respectively,andS1,S2etc.andS10,S20 etc. available to the structural properties of the different representing the opposing proteinase subsites). PCs, Siezen and colleagues58 as well as Lipkind and Most PC cleavage sites contain one or more Steiner59 had modelled the catalytic domain of furin additional basic residues further upstream, and the active sites of PC1/3 and PC2, respectively, especially at the P4 position, but also at P3, P5, P6 based on the distantly related subtilisin structure. and/or beyond.20,23–25 Furin, the most demanding They furthermore made (partial) PC sequence of these PCs with respect to positively charged alignments and predicted structural features of the substrate residues, has been found to require, in catalytic domains. In addition, Lipkind and Steiner addition to a P1-Arg, at least two other basic proposed a model for the P domains based on residues at P2, P4 or P6.20,23,26–29 Therefore, a basic secondary structure predictions.60 Recently, we and P2 residue is not absolutely required if another others have determined the X-ray crystal structures basic residue occurs at P4 and beyond. On the other of soluble forms of mouse furin61 and yeast kexin,56,62 strongly differing surface patches (blue). The experimental decanoyl-Arg-Val-Lys-Arg-CMK inhibitor (grey stick model) binds into the conserved active-site cleft. In the intact pro-PCs the red, conserved groove extending from the active-site cleft to the right might accommodate the C-terminal part of the primary activation cleavage segment (see Figure 3). Only a few residues have been labelled. The Figures was prepared using GRASP,106 MOLSCRIPT103 and RASTER3D.104 214 Models and Specificity of Proprotein Convertases

Figure 2. Aligned sequences and negative charges in the active-site cleft. (a) Structure-based amino acid sequence alignment of the soluble ectodomains of the mammalian PCs (human furin; human PC1; human PC2; mouse PC4; human PC5; human PC7; and human PACE4) and of kexin (for references, see Methods). Furin and kexin residues assigned as structurally conserved regions (SCRs) are enboxed. PC residues shown with yellow colour (SCRs) are Models and Specificity of Proprotein Convertases 215 respectively, which consist of the catalytic and the P loops selected from the Protein Data Bank (PDB). domain (Figure 1). Due to the strong sequence The preferential usage of the furin structure is conservation of both domains among the PCs, these clearly justified by the closer sequence resemblance experimental structures now offer the opportunity and by the aim to obtain most contiguous chain to generate reliable models of the other mammalian geometries. PCs, to derive common as well as distinct properties The ectodomains of all six modelled PCs share for these PCs, and to develop novel strategies for the characteristic two-domain structures of furin inhibiting these medically important enzymes. On and kexin (Figure 1). Briefly, the polypeptide chains the basis of the closely related furin and kexin are folded into two separate but abutting domains, ectodomain structures, we have built the catalytic/ the spherical catalytic domain (in furin: Tyr110f- P domain pairs of human PC1, PC2, PC5, PC7 and Ala445f; with the suffix f indicating topologically PACE4, and of mouse PC4, using homology equivalent residues using the furin numbering, also modelling techniques. In the following sections, used in Figure 2(a)) and the barrel-like P domain we describe these PC models and correlate them (Pro446f-Thr573f). Both domains are covalently with experimentally known data and their anti- linked by the “inter-domain linker” and interact cipated or known functions. with each other through an about 1100 A˚ 2 interface.

Catalytic domain structures Results For all PCs, the core of the catalytic domains Overall structure of the PC ectodomains consists of a highly twisted b-sheet composed of seven parallel and one antiparallel b-strands As already expected from previous purely (Figures 1(a) and 2(a)). This sheet is flanked by sequence-based alignments,8,56,58,60 the model/ five adjacent and two peripheral helices and by two structure superposition (Figure 1) as well as the short b-hairpin loops. As can be derived from the structure-based sequence alignment (Figure 2(a)) alignment scheme (Figure 2(a)), all of these regular reveal a close resemblance of the six modelled PC secondary structure elements are conserved in the ectodomains to furin and to kexin, with a signifi- PCs. Only the slightly longer N termini of PC2, PC5 cantly higher similarity to furin (Table 1). The and PACE4 have been assumed to adopt a kexin- sequence identities are between 68.5 and 50.1% for like helix. These regular elements connected by the comparison of the catalytic domains of these more irregularly folded surface loops are remark- models and the furin catalytic domain, and between ably similar in sequence and structure to furin. 54.9 and 34.6% for the comparison of the corre- Examples are the N-terminal segment from Asp115f sponding P domains. Among the mammalian PCs, to Trp123f, or the Cb5–Ca4 loop forming the PC4 is most similar and PC7 most dissimilar to “substrate alignment segment” Ser253f-Pro256f furin. The comparison of the mammalian PCs with (see Figures 1(a) and 4). Considerable structural kexin leads to an almost inverted order of similarity, differences occur only at a very few exposed surface yielding, however, considerably lower overall loops. These sites include: (i) the N-terminal sequence identities (between 45.7 and 41.6% for segmentsofPC2,PC5andPACE4mentioned the catalytic domains, and between 30.9 and 23.5% above; (ii) the 126f-127f surface loop, with five for the P domains, see Table 1). Therefore, most of (PC5 and PACE4) and six (PC1 and PC2) mainly the structurally conserved regions (see Methods hydrophilic residues inserted; (iii) a one-residue and Figure 2(a)) and short connecting segments insertion in the 228f-231f loop of PC2, forming a were actually modelled based on the furin scaffold more exposed “canopy” atop the S4 subsite; (iv) a (PDB entry 1P8J, molecule A).61 Structural infor- two-residue insertion at 346f-347f in an exposed mation from the kexin structures (PDB entries 1R64 PC2 hairpin loop; (v) a four-residue insertion in an and 1OT5)56,62 was mainly used for modelling of exposed 355f-356f hairpin loop in PC7; and (vi) two- those particular segments where the kexin sequence and one-residue insertions in the broad 402f-409f indicated closer similarity, such as at the elongated loops of PC1 and PC2, respectively, with confor- N termini of PC2, PC5 and PACE4 (for further mations probably more similar to kexin. details, see Figure 2(a)). For most of the other In all mammalian PCs as well as in kexin, the variable regions, the model structures were built polypeptide chain of the catalytic domain is cross- along the corresponding furin loops or along other connected by disulfide bridges Cys211f-Cys360f modelled along the furin structure, light green and dark green residues (variable regions, VRs) are adapted from furin and from kexin, respectively, while newly modelled residues are given in italics. Cys residues (probably) involved in disulfide bridges, and the active-site residues are shown by blue and pink colours, respectively. The numbering used is for intact translated human/mouse furin. Below the sequence, b-strands and helices as occurring in the furin structure are indicated by black arrows and cylinders, and are labelled according to their location in the catalytic (C) or the P domain (P). The Figure was prepared using ALSCRIPT.107 (b) Specificity-determining critical residues in subsites S8 to S20 ordered according to their negative charges in the non-prime subsites. Negatively and positively charged residues are given in red and blue colour, while polar and hydrophobic residues are shown in orange and green, respectively. The residue numbering is for furin. 216 Models and Specificity of Proprotein Convertases

Table 1. Structure-based amino acid residue identities

Furin Kexin PC1 PC2 PC4 PC5 PC7 PACE4

Identity of model and furin (%) Total 100 39.4 55.3 52.3 64.7 60.2 45.7 59.4 Catalytic domain 100 43.9 61.4 55.5 68.5 65.0 50.1 65.6 P domain 100 27.8 39.8 44.4 54.9 48.1 34.6 43.6 Identity of model and kexin (%) Total 39.4 100 40.0 37.9 36.9 38.6 40.7 39.2 Catalytic domain 43.9 100 44.0 43.4 42.2 41.6 45.7 43.7 P domain 27.8 100 30.1 24.3 23.5 30.9 27.9 27.9

and Cys303f-Cys333f. In PC5 and PACE4, two novel sugar chains would not interfere with substrate Cys residues equivalent to furin residues 127f and binding, even for large bulky substrates. 128f are properly juxtaposed to form a disulfide bridge that closes the 127f-128f insertion loop, as had been predicted by Siezen et al.58 Noteworthy Overall structure of the P domains might be the internal unpaired Cys residues equivalent to furin residues 198f and 305f, which In all PCs except PC1, the inter-domain linker are strictly or strongly conserved in all PCs except spans across the molecular surface as in furin, for PC7. In PC7, this Cys305f is replaced, similar as connecting the last helix Ca7ofthecatalytic in kexin, by a Tyr residue. By using anomalous domain with the first strand Pb1 of the P domain scattering data on either side of the K-absorption (see Figure 1(a)). In addition to the above men- edge of calcium, we recently have been able to show tioned Asp392f, the indole moiety of the strictly that furin essentially binds two but not three conserved Trp441f seems to glue the linker to the calcium ions (unpublished results). Due to their underlying Leu/Val/Ile396f residue of the catalytic close similarity to furin, all modelled PCs have domain. In PC1, a two-residue insertion results in therefore been assumed to retain the same two the formation of a bulge between the structurally calcium binding sites Ca1 and Ca2 (Figure 1(a)) as conserved helix Ca7 and this Trp441f. In all PCs, the found in furin. Except for a few polar residues, the following chain segment is organized as a separate cores of all catalytic domains essentially consist of eight-stranded b-barrel, with a strand connectivity hydrophobic residues. A main exception is the characteristic for jelly-roll b-barrels. These eight buried Glu201f, which in all PCs forms an internal structurally conserved strands are arranged in two salt bridge with the strictly conserved residue opposing four-stranded b-sheets, which sandwich a His364f. The likewise conserved and buried small hydrophobic core. Only a few of the connect- Asp392f apparently helps to clamp the inter- ing loops located on both apices of the prolate domain linker to the catalytic domain. Also note- ellipsoid deviate significantly from the correspond- worthy are a few negatively charged but buried ing segments of furin, in particular in PC5 residues located around the S1 pocket and the Ca1 (Figure 1(b) and (c)). Except for PC7 and kexin, all and the Ca2 site, which are likewise strictly or PCs share a common disulfide bridge with furin strongly conserved (see below). The charged resi- (Cys450f-Cys477f). PC7, instead, resembles kexin in dues are unevenly distributed along the catalytic lacking this disulfide bridge, and the enclosed domain surface, with negatively charged residues segment presumably also exhibits a kexin-like particularly clustering in the non-prime part of the conformation (see Figure 1(b)). PC7 seems to active-site cleft and along its rims (Figure 2(b)). The possess a novel disulfide bridge between Cys507f few positively charged residues, in contrast, are and Cys538f, whose topologically equivalent resi- located in the prime subsites or are spread across dues in furin and kexin are properly juxtaposed the remaining molecular surface. Most of the for covalent linkage. Consequently, the PC7 model potential N-glycosidic sugar sites of the distinct has been refined to include this disulfide bridge. PCs are located at different surface patches. In the Residue 538f is also a Cys residue in PC5 and inter-domain linker, furin exhibits a proven61 PACE4, where the opposing residue 507f is a Ser, sugar attachment site at Asn440. Furin shares with however, precluding disulfide bridge formation. kexin56 a further glycosylation site at Asn387f, The P domain of PC2 exhibits potential N-glycosi- which might also be present in PC1. All other dic sugar attachment sites at the end of Pb3 (494f) potential N-glycosidic attachment sites, namely and at a slightly buried site within strand Pb4 (504f), Asn133f and Asn141f (PC7), Asn159f (PC1), while PC4 and PC7 possess such sites at the Asn166f (PC7), Asn207f (PC5, PC7 and PACE4), terminal part of Pb2 (470f). The P domain termi- Asn359f (PC2) and Asn363f (PC5), are located nates with Thr/Glu/Ser573f, which is the last towards the “top” of the catalytic domain, when residue of strand Pb8 involved in inter-strand looked at in standard orientation (i.e. with a hydrogen bonds. This presence of residue 573f, substrate chain running from left to right across and thus of an intact P domain, seems to be required the front surface, see Figure 1). Therefore, attached for enzymatic activity, as shown for PC1/3.17 Models and Specificity of Proprotein Convertases 217

The domain–domain interface hydrogen bonds with the likewise conserved Arg391f and His395f side-chains. Furin and PC4 The domain–domain arrangement and inter- possess a common salt bridge between Arg418f and actions are virtually identical in the eight PC Asp540f. All mammalian PCs except PC7 share models/structures, in agreement with experimental with furin the Arg498f-Gly499f-Asp500f “RGD- findings that chimeric PC constructs exhibit enzy- motif”, located in the Pb3–Pb4 loop. In PC7, the matic activity.17 The P domains make most of their third Asp residue is replaced by a Ser, rendering it contacts with the catalytic domains essentially via more similar to kexin (with a Thr). Due to exposure their two “edge strands” Pb6 and Pb5. The first of the leading Arg and the trailing Asp side-chain strand interacts with the bulge preceding Cb8and towards different surfaces, this site does not seem to 61 the Cb11–Cb12 hairpin loop, while the second play a role as a disintegrin. contacts the loops Cb5–Ca4 and Cb6–Cb7(Figure 1). These segments are relatively conserved in all PCs. The pro-domains Except for a hydrogen bond between Ser334f and Thr536f, all inter-domain contacts are made The only structure of a PC pro-domain available through side-chains. The domain–domain interface so far is the NMR-structure of isolated mouse resembles typical solvent-exposed domain surfaces. PC1.64,65 This PC1 pro-domain exhibits an open- The few hydrophobic interface residues contributed sandwich antiparallel-a/antiparallel-b fold, made by the catalytic domain are of special note, because up by two a-helices packed against one side of a most have polar/charged counterparts in the four-stranded twisted b-sheet (see Figure 3). Its topologically related but P domain-free subtilisins. structure resembles the pro-domains of subtilisin These hydrophobic residues are mostly conserved BPN066 and subtilisin E67 in complex with their in the PCs, obviously constrained by the common respective catalytic domains, and the pro-domain function to mediate interaction with the P domain. core of the intact pro-form of the bacterial subtilase Of particular note are a few charged interface kumamolisin.68,69 According to sequence identity, residues, which are engaged in buried salt bridges. the pro-domains of the mammalian PCs might not In all PCs, the strictly conserved Arg519f forms an resemble each other as much as do their catalytic and inter-domain salt bridge with the strictly conserved P domains. Recent structure-based PC alignments65 Asp301f (part of the Ca2 binding site), and another as well as homology modelling and mutation internal salt bridge with the conserved Asp522f, experiments on furin70 show,however,thatthese whose side-chain carboxylate group, in turn, is pro-domains seem to exhibit a similar overall connected to the surface-located guanidyl group of topology. In the initial pro-PCs, the pro-domains the conserved Arg498f (Figure 4). Furthermore, in might cap the N termini of helices Ca3 and Ca4 (see all PCs the side-chains of the strictly conserved Figures 1(a) and 3), presumably similar to the residues Glu485f and Gln447f form charged interaction of the subtilisin pro-peptides and the

Figure 3. Stereo ribbon plot of a hypothetical PC1-furin pro-enzyme chimera shown in standard orientation. This PC proform has been constructed by docking the experimental pro-domain of PC1 (dark green; PDB entry 1KN6)65 to the experimental furin structure61 (catalytic and P domain: dark gold), and modelling the intact primary activation cleavage segment into the preformed active-site cleft in a manner similar to that observed in pro-kumamolisin.69 This rearrangement requires the removal of the first w12 N-terminal residues (red colour) from their position in mature furin. The polypeptide segments preceding and following the Arg107-Asp108 primary activation cleavage site are given in light green and blue colour, respectively. The primary (Arg107-Asp108) and the secondary (Arg75-Ser76) activation cleavage sites are indicated by triangles 1 and 2. These flanking residues, the active-site residues Ser368, His194 and Asp153, and the Ca1-coordinating Asp115 of furin are given as stick models. The furin positions Asp181 and Gly229, which in PC2 are occupied by residues Tyr194 and Gln242 critical for supporting the pro-PC2–7B2 interaction,76,78 are indicated by grey spheres. The residues flanking the primary and the secondary activation sites are shown with their side-chains. The Figure was prepared using MOLSCRIPT103 and RASTER3D.104 218 Models and Specificity of Proprotein Convertases

Figure 4. Stereo view of the active-site cleft of furin, shown in standard orientation together with the decanoyl-free Arg-Val-Lys-Arg-CMK inhibitor.61 Furin residues are given as stick models, with grey (catalytic domain) and dark-green (P domain) carbon atoms, blue nitrogen atoms, red oxygen atoms and yellow sulphur atoms. The Ca2 ion and the inhibitor residues are shown as a purple sphere and a ball-and-stick model (light-green carbon atoms, blue nitrogen atoms and red oxygen atoms), respectively. Hydrogen-mediated salt bridges around the P1-Arg, P2-Lys and P4-Arg inhibitor residues as well as along the Asp301/Arg519/Asp522/Arg498 salt bridge network extending from the S1/Ca1 site into the P domain are shown by dotted yellow lines. Certain residues of special interest (addressed in the text) have been labelled using the one-letter-code. The Figure was prepared using MOLSCRIPT103 and RASTER3D.104 pro-domain of intact pro-kumamolisin in their The productive insertion of the primary acti- respective complexes/structures. Similar to the vation cleavage segment into the active-site cleft bacterial pro-domains,71,72 the PC pro-domains and the adjacent “conserved” groove (see Figure 1(c)) are generally believed to act as intra-molecular will certainly contribute to stabilization of the chaperones to induce the final folding step of active-site region and the entire catalytic domain. their catalytic domains.18,73 The stereo model of In furin, this rearrangement requires, however, the a PC1-furin chimeric pro-enzyme is shown in partial unfolding of w12 residues (see Figure 3), Figure 3 to indicate the presumed relative domain which after activation cleavage become the newly arrangement. formed N terminus of the generated mature PC. In the PC pro-forms, the primary activation Due to the necessary removal of the Ca1-coordinat- cleavage segments (Lys-Arg-Arg-Thr-Lys-Arg107f- ing Asp115f residue, the conformation anticipated Y-Asp108f-Val in furin; Figure 3) connecting the pro for the uncleaved pro-form would require a partial with the catalytic domain will certainly insert into disruption of the Ca1 site in furin, PC1, PC4 and the preformed active-site clefts, as exemplified by PC7 (similar to what has been suggested for the the linker peptide of pro-kumamolisin.69 This pro-subtilisins; see Gallagher et al.66). Such a insertion allows this segment to span across rearrangement is not required but may also occur the catalytic residue Ser368f, suitable for intra- in PC2, PC5 and PACE4, where up to six (PC5) more molecular attack of the scissile peptide bond (see N-terminal residues than in furin allow more Anderson et al.73). PC2 is unique among the PCs in flexibility (see Figure 2(a)). During trafficking to that it contains an acidic equivalent of furin’s the more acidic TGN/endosome compartments, the Asn295f residue in the oxyanion hole, and a basic cleaved (but still associated) pro-peptides might be (Lys) residue in the P3 position of the primary cut in trans at the flexibly exposed but less activation cleavage segment. It is tempting to susceptible “secondary activation cleavage site” speculate that this Lys-P3 residue of pro-PC2 (Val-Thr-Lys-Arg75f-Y-Ser76f in furin).18,73–75 This might pull the Asp295f side-chain out of its second cut obviously leads to destabilization of the productive position, thereby contributing to auto- pro-domain, resulting in its dissociation and release inactivation of pro-PC2, as observed in pro- of the respective active PC (see Figure 3). kumamolisin.69 Pro-PC2 is an exception among the PCs in that the Models and Specificity of Proprotein Convertases 219 neuroendocrine protein 7B2 is required for its PCs. The “upper rim” of the cleft is made by the efficient conversion to the active enzyme. 7B2 is a multiple-turn loop connecting strand Cb4with 186-residue secretory protein, which in the endo- helix Ca3 (bordering subsites S6, S4, and S2), the plasmic reticulum binds to pro-PC2 and in transit to strictly conserved segment Asp153f-Asp154f (con- the trans-Golgi network becomes proteolytically fining S2), the quite exposed loop leading into helix cleaved into a 21 kDa N-terminal fragment and a C- Ca2 (carrying the active-site residue His194f, and terminal PC2 inhibitor peptide, eventually releasing bordering S2 and S10), and the segment leading into active PC276 (for details, see also Lamango et al.31 helix Ca5 (limiting S10 and S30). In PC2, the Cb4– and Mbikay et al.50). This C-terminal peptide is a Ca3 loop, with a one-residue insertion, might bulge nanomolar inhibitor of PC2, but does not inhibit out further than in the other enzymes. The “lower any other PC. Pro-PC2 seems to interact with the rim” encompasses the exposed Cb5–Ca4 loop N-terminal part of 7B2 mainly via the catalytic following the substrate alignment segment (border- domain,30 but the pro-region and the P domain ing S7, S5, S3 and S1), the multiple-turn loop might also contribute to binding.77,78 The contact following Asn/Asp295f (limiting S1), and the sites on the PC2 catalytic domain have been indole side-chain of Trp328f (bordering S20). This mapped to an Asp-Pro-Tyr181f-Pro-Tyr-Pro latter residue is replaced in PC2, PC7 and kexin by sequence containing a critical Tyr181f residue (see Leu, Phe and Pro, respectively. Figure 2(a)), to the oxyanion hole-determining Asp295f, and to a sequence encompassing the Substrate interactions at the non-prime subsites Gln229f-Pro-Phe segment.78,79 In the PC2 model, the phenolic side-chain of Tyr181f, surface-exposed In all PC models, the Arg-Val-Lys-Arg-CMK together with the flanking phenol groups of Tyr183f inhibitor has been placed in the same conformation and Tyr158f, as well as the elongated segment as experimentally observed in the furin structure. around Gln229f are placed just aside the surface Therefore, the peptidyl moiety juxtaposes the patch, where the pro-domain should bind (for extended Ser-Trp254f-Gly-Pro alignment segment orientation, see the pro-PC1-furin chimera in and forms a twisted two-stranded antiparallel Figure 3). 7B2 (or its N-terminal part) could thus b-sheet.61 The P1-Arg carbonyl group extends into easily touch the two identified contact sites and the oxyanion hole formed by the Asn/Asp295f the pro-domain simultaneously. In addition, the N-H/O–H side-chain groups and by the Ser368f C-terminal peptide of 7B2 could reach for the main chain N–H. Furthermore, it is covalently primary activation cleavage segment of the cognate bound via its tetrahedral hemiketal group (mimick- pro-PC2, assist in intra-molecular activation clea- ing a transition state intermediate) and the follow- vage by furin, and push the C-terminal pro-domain ing methylene group to Ser368f Og and His194f Nd, peptide out of its active-site position. The residues respectively. As previously mentioned, Asn295f is that are involved in the specific inhibition of PC2 by replaced in PC2 by an Asp residue, as also found in the 7B2 C-terminal peptide have been mapped the oxyanion hole of distantly related sedolisins via mutagenesis to the PC2-specific residues 79 such as kumamolisin, derived from a Bacillus Gln229f-Pro-Phe. Our model predicts that this strain adapted to an acidic habitat.68 At acidic pH, latter segment bulges (due to a one-residue inser- the protonated PC2 carboxylic acid group of tion) slightly out of the upper rim of the PC2 active- Asp295f should be a particularly good polarizer of site cleft, changing its surface properties compared thescissilepeptidecarbonylgroup,assisting to the other mammalian PCs. catalysis in the acidic environment of the secretory granule.1,21,78,80 Overall substrate binding regions In all PCs, the P1-Arg side-chain of the inhibitor identically extends through the kinked Ser253f- Of utmost importance for the distinct specificities Asp258f “entrance frame” into the S1 pocket of the PCs are the geometry and the charge (Figure 4), sandwiched between the strictly con- distribution of their substrate binding regions. All served segments Ser253f-Gly255f and Ser293f-Asn/ PC models exhibit a deep active-site cleft with a Asp295f. The terminal guanidyl group is perfectly shape remarkably similar to furin (Figure 5), with packed into a flat groove lined by the carboxylate the Ser368f/His194f/Asp153f active-site triad groups of Asp258f and Asp306f (see Figure 2(b)) arranged at the center (Figures 1 and 4). All residues and the Ala292f and Pro256f carbonyl groups. of the loop segments contributing to this cleft are Therefore, in all PCs this S1 pocket is designed to considerably more conserved than other surface- select for and accommodate an Arg side-chain, in located loop segments (Figure 1(c)), with the agreement with the strict substrate requirement of exception of a very few characteristic, negatively all PCs for P1-Arg.15,21,23,29,52 In all PCs, the charged residues within furin (see Figure 2(b)). The carboxylate group of Asp258f coordinates Ca2, “base” of this cleft (relative to the standard placed in close vicinity to the other calcium-ligating orientation as shown in Figures 1 and 4) is formed carboxylates of Glu331f and Asp301f (Figure 4). by the Ser-Trp254f-Gly-Pro “substrate alignment This latter Asp301f carboxylate, in turn, forms an segment”, the Trp254f indole side-chain, and the internal salt bridge network via Arg519f and first turn of helix Ca5 (preceding the reactive Asp522f with the Arg498f guanidyl group belong- Ser368f), all of which are fully conserved in the ing to the above-mentioned RGD sequence of the P 220 Models and Specificity of Proprotein Convertases

Figure 5. Solid surface representation of the extended active-site cleft of the PCs shown in standard orientation. The colouring was made according to the calculated electrostatic surface potential, extending from red (K27 e/kT) to blue (27 e/kT). The Figures were prepared with GRASP,106 MOLSCRIPT103 and RASTER3D.104 First row: Stereo view of the Y active-site cleft of furin superimposed with the modelled (Arg)3-Val-Lys-Arg- -Ser-Leu octapeptide, docked to subsites S6 to S20 on the basis of the experimental decanoyl-Arg-Val-Lys-Arg-cmk inhibitor,61 the subtilisin-eglin c complex85 and pro-kumamolisin,69 and energy-minimized with MAIN.96 Rows 2 to 5: Active-site clefts in the order of decreasing negative charges, i.e. of furin, PACE4, PC4, PC5, PC1/3, PC7, PC2 and kexin, shown together with the decanoyl-free Arg- Val-Lys/Arg-Arg-CMK inhibitor, bound to subsites S4 to S1 as in furin. Models and Specificity of Proprotein Convertases 221 domain. All of these charged residues engaged in addition to making a strong hydrogen bond to the salt bridge network and in Ca2 ligation are Tyr308 Oh (Figure 4). This tight hydrogen bond strictly conserved in the PCs. This residue con- embedding is in agreement with the strong pre- servation as well as systematic mutational studies16 ference of furin for Arg and, much weaker, Lys-P4 point to the importance of this Ca2 site for the residues.21,27,28 All PCs share the Trp254f and the stabilization of the extremely negatively charged S1 Tyr308f residue as well as the “internal” Glu236f side- pocket, leading to an apparently delicately balanced chain, and all but PC1 and kexin exhibit a surface- juxtaposition of charged groups. Stabilization of located Asp264f, conferring some preference for the heavily charged active-site cleft by P domain basic P4 residues (see Figure 2(b)). The carboxylate residues might be the main function of the P group of the equivalent but more flexible Glu264f domain in the PCs/kexin. side-chain of PC1 is less suitably positioned to The P2-Lys side-chain, in contrast, runs along a clutch the guanidyl group of a bound P4-Arg surface crevice, at the end of which its 3-ammonium residue, which might in part explain the weaker group is tetrahedrally surrounded by the oxygen PC1 preference for basic P4 residues.31 atoms of the Asp154f carboxylate, the Asn192f The S4 subsite of the Ac-Arg-Glu-Lys-boroArg- carboxamide and the Asp191f carbonyl groups to kexin structure56 strongly differs from that of furin, allow efficient hydrogen bond formation (Figure 4). mainly due to a deviation of its Asp230f-Ile-Thr-Thr Accommodation of this positively charged P2 loop. While Ile231f is oriented towards the protein residue is further favoured by the nearby strictly interior in kexin, the equivalent Val231 of furin lines conserved Asp228f (Figure 2(b)). Thus, geometry the S4 specificity pocket. As a consequence, the P4- and charge of this S2-subsite are beneficial for Arg side-chain is not kinked in the kexin structure occupation by Lys, but also permit acceptance of the as it is in furin, but has an extended conformation side-chains of Arg and other residues, as found in occupying the area filled by the side-chain of Val231 several PC in vivo substrates.81 In kexin,62 Asn192f is in furin. In the kexin structure, the P4 guanidyl replaced by an Asp residue, and the surface-located group forms hydrogen bonds only with one Asp191f carboxylate is (presumably due to the (Glu236f) but not with two carboxylates (Glu236 shortened preceding loop) directed towards the and Asp264) compared with furin. This weaker P4– 3-ammonium group of the P2-Lys side-chain, S4 binding geometry of kexin seems to be in enhancing the importance of the electrostatic agreement with the dual recognition of aliphatic/ P2–S2 interactions compared with furin.21,56 Both basic P4-residues by this yeast enzyme.21 This structural differences might explain the much more altered P4–S4 geometry is a direct consequence of stringent requirement of kexin for a basic P2 residue the different placement of the 231f residues, compared to furin.56,82 In the other mammalian presumably caused by different amino acids at PCs, furin’s Asp191, whose side-chain is not position 235f. In contrast to the polar Asp235f side- directly involved in P2 binding, is replaced by a chain of kexin, which points towards the bulk Glu (PC1, PC4, PC5, PACE4), a Gly (PC7) or a Phe solvent, the Val235 side-chain of furin occupies a residue (PC2). Thus, in PC2 the 3-ammonium group hydrophobic cluster (which in kexin is filled with of a bound P2-Lys might be even more shielded the Ile231f side-chain). Because residue 235f is from the bulk water, thus possibly strengthening also hydrophobic (Val, Ile or Met) in the other the charged/hydrogen bond interactions in the S2 mammalian PCs, the modelling procedure implies site. that their 230f-233f segments and therefore their The P3-side-chain extends away from the active- overall S4 architecture should resemble that of site cleft toward the bulk solvent, consistent with furin. The S4 subsites might nevertheless differ the lack of a clear amino acid preference of most slightly from each other. In PC2, for example, the PCs (see Figure 4). Alkaline P3-side-chains, never- one-residue insertion in the Gln229f-Met231f loop theless, could make favourable contacts with the mentioned above might affect the shape of the S4 surface-located, quite exposed Glu/Asp257f car- pocket; in addition, the Val231-OMet replacement boxylate groups in furin/PC4 and PC5/PACE4, will probably narrow this pocket, which might respectively (Figures 2(b), 4 and 5). This is in make the accommodation of long P4 (Arg) side- agreement with the presence of alkaline P3-residues chains in PC2 less favourable and thus contribute in in vitro and in vivo substrates of furin21,83 and to the comparatively weaker preference for alkaline PC4.84 PC1 and PC2, with exposed Asn and Thr257f P4 residues. residues, show an even lower preference for basic The substrate–PC interactions will presumably P3 groups.31 Kexin, due to an Ala257f residue extend to beyond P4–S4 and P1–S1. Furin, in exhibiting a flat and hydrophobic S3 site (see particular, shows a strong P6 dependence, proving Figure 5), still selects against acidic P3 residues.21 the importance of specific interactions with a “S6” In the furin–inhibitor complex,61 the kinked site. The decanoyl-substituted N terminus of the P4-Arg side-chain extends into the surface-located tetrapeptidyl-CMK inhibitor in the furin complex61 S4 cleft. The proximal part of the Arg side-chain is as well as the corresponding acetyl-substituted N tightly packed against the hydrophobic side-chains terminus of the tetrapeptidyl-boroArg inhibitor of of Val231, Pro256, and Trp254, while its distal the kexin structure56 are directed towards the guanidyl moiety is favourably framed by the crevice between the catalytic and the P domain carboxylate groups of Glu236 and Asp264, in (see Figure 1). Their detailed locations in the furin 222 Models and Specificity of Proprotein Convertases and kexin structures are certainly affected by the substrates, while P10-Lys and Arg residues are decanoyl/acetyl group–enzyme interactions and by generally not accepted by PCs.19,42 The less charged the unconstrained adaptability of their P4 side- PC2, as an exception, can also process substrates chains, contributing to the observed differences.56 with P10-Phe and Tyr residues.31,86 The S20 subsite The substrate alignment segment of the PCs does of furin is a small depression extending towards the not extend beyond S4. Hence, N-terminal to P4 the exposed indole moiety of Trp328. This latter residue exact orientation of a bound substrate chain is is a Trp in most PCs, but can also be a Leu (PC2) difficult to predict. However, it is tempting to and a Phe (PC7). Thus, in all mammalian PCs the speculate that many substrates might bind their S20 subsite is well equipped for accommodation cognate PC in an overall extended conformation, as of hydrophobic medium-sized residues21,23,24 exemplified by the C-terminal pro-domain tail of (Figures 2(b) and 5). In contrast, the S20 subsite of the corresponding subtilisin BPN0-complex66 or by kexin, with a Pro at residue 328f, differs signifi- the interaction between the reactive-site loop of cantly with respect to surface contour and charges. eglin c and the active site of subtilisin.85 The The S30 grooves of all PCs look similar, with the structure of bacterial pro-kumamolisin, with an strictly conserved Arg197f forming the border at the uncleaved primary activation cleavage segment upper rim. running in a substrate-like manner through the entire active-site cleft,69 seems to represent an even better example for the more peripheral subsite Discussion interactions. Accordingly, basic P5-side-chains b a would run along the C 5–C 4 loop of furin, placing Our modelling studies have shown that the their distal guanidyl groups adjacent to the surface- mammalian PCs, in agreement with their strong located carboxylates of Glu257f and Asp259f. PC4, sequence similarity with furin, presumably exhibit PC5, PC7 and PACE4, containing similar Glu/Asp spatial atomic structures that seem to be more residues, seem to share the preference for basic P5- similar to furin than to kexin. The domain cores as side-chains with furin. In contrast, PC1 and PC2 well as the domain–domain interfaces are virtually with Asn/Thr257f and Asp/Asn259f residues, 31,86 identical in all PCs; in addition the topology of most respectively, accept other P5-residues as well. surface structures, in particular around the active- In kexin, this “S5 site” is even slightly basic, due to site cleft, is remarkably similar. According to the the unique presence of His262f replacing the structure-based sequence comparison (Table 1), equivalent Thr residue of the other PCs. PC4 is most similar to furin, PACE4 as well as The distal groups of P6 side-chains, correspond- PC5 are very similar to furin, and PC1, PC2 and PC7 ingly, should either intercalate between the surface- are increasingly different from furin. This order located Glu230f and the more distant Asp233f, or, in essentially holds for the catalytic as well as for the case of short P4 residues, could bend backwards to P domain, and is also relatively similar when thread into the S4 pocket opposite to the P4 side- 61 determined from a comparison of the less con- chain. The majority of furin substrates indeed 70 21,23 served pro-domains. Remarkable exceptions from possess alkaline residues either at P4 or P6. this general trend are PC1 and PC2, where the Furin shares the Asp233f residue with all other PCs, comparison of the catalytic and the P domains while kexin has a Thr (Figure 2(b)). Indeed, a with those of furin results in opposite degrees of recently reported Thr233fAsp kexin mutant was 87 homology, probably reflecting evolutionary shown to specifically recognize P6-Arg residues. restraints for the binding of specific substrates/ Glu230 is unique to furin, while PC5 and PACE4 (as cofactors such as 7B2 and pro-SAAS, respectively kexin) exhibit an Asp at this position. Residue 230f (see Cameron et al.32). The structural and sequence is a Pro in PC2 and PC7, and an Ile and Ala in PC1 similarities thus seem to indicate an early evolu- and in PC4, respectively, in accordance with their 42 tionary separation of a common multi-domain PC much reduced requirement for basic P6 residues. ancestor from kexin, and subsequent gene dupli- P7 side-chains might extend along the conserved cation events, with a further divergence of PC7, PC1 Asp264f side-chain, while P8 side-chains could reach and PC2 from a more homogeneous group contain- across the variable Ala267f towards the quite distant, ing furin, PC4, PC5 and PACE4.89 The occurrence of likewise variable Arg268f (Figures 2(b) and 4). a furin and a PC2-like orthologue in Drosophila also underscores this grouping. Substrate interactions at the prime subsites The substrate binding region of all PCs is characterized by the clustering of an extremely The positively charged S10-subsites are bordered large number of negatively charged residues, which by the side-chains of the strictly conserved residues enable the PCs to recognize and to cleave following His194f and His364f and of residue 193f, which is an multiple basic amino acids. However, according to Arg in furin as well as in PC4, a Lys in PACE4, PC5 the number and distribution of these negative and PC1, a His in PC7, but a Ser and a Tyr in PC2 charges the PCs differ slightly from each other. and kexin (see Figure 2(b)). P10-Asp and Glu side- Furin, with 16 acidic residues (with Asp153, chains could favourably interact with the His364f Asp228, Glu299, Glu301, Glu331 and Asp522 in and the Arg/Lys193f side-chains. Indeed, Ser and addition to the ten shown in Figure 2(b)) in the Asp residues occur frequently at P10 positions in PC immediate vicinity of the active-site cleft, possesses Models and Specificity of Proprotein Convertases 223 the highest number of negative charges. PACE4, cleft, able to bind (mainly via the S1–P1, S2–P2, PC5 and PC4, with 15 acidic residues each, exhibit a S4–P4 and P10–S10 side-chain and the S1/S2–P1/P2 similar negative charge density, while PC7 and PC1 and S3–P3 main chain interactions) the cognate (13 such acidic residues) as well as PC2 (11) peptide substrate and span it across the active site resemble kexin (13) more closely. Therefore, the in such a manner that the P1–P10 scissile peptide overall requirement for positively charged residues bond can be optimally attacked by the reactive N-terminal of the cleavage site of bound substrates Ser368f Og. The number of basic substrate/inhibitor would be expected to decrease in the same order, in groups required should thus somewhat depend on fair agreement with experimental results.21,24,29,51 the number of negatively charged residues in the In addition to the global charge compensation, active site. As shown for furin, a suitable charge the surface distribution patterns of these charges compensation at one site (such as S4 and S6) can and their matching with substrate counter-charges replace the charge compensation at another site are important for the cleavage specificity as well. (S2). The selective inhibition of furin, PC6 and All PCs share a virtually identical S1 pocket, which PC1/3 by a1-PDX35 and re-engineered eglin c39 is designed and equipped to exclusively accommo- indicates that the reactive-site architectures of the date a P1-Arg side-chain, and seems to be stabilized distinct PCs are sufficiently unique to discriminate by a salt bridge network involving the nearby Ca2 between inhibitors, presumably also due to site and also the P domain. The S2 grooves, due to additional exosite interactions. The overlapping geometric restraints, clearly prefer Lys over Arg-P2 specificities of most PCs explain the redundancy residues, but also allow (in particular in furin) in processing often observed in co-expression 3 accommodation of other side-chains, if charge studies and knock-out mouse models. Often, compensation is possible elsewhere. The architec- however, the frequently observed in vivo selectivity ture of the S4 subsites of the mammalian PCs seems of PCs also results from their distinct tissue and to resemble that of furin, explaining the general subcellular localization and expression. preference of most mammalian PCs for positively These small but significant differences between charged P4 residues. The less stringent requirement the mammalian PC family members should allow of PC1 and PC2 for basic residues might result from discrimination by elongated inhibitors stretching the lower negative charge accumulation in the across several subsites. Indeed, L-Arg and D-Arg vicinity of their S4 site, but could also be due to oligopeptides have been shown to be potent, 52 slightly modified S4 sites. Due to their positively relatively selective inhibitors of furin. Recently, charged S10 and hydrophobic S20 subsites, all the inhibitory potency of these D-poly-Arg peptides 0 has been found to be proportional to the poly-Arg mammalian PCs should prefer polar/acidic P1 57 and hydrophobic P20 residues and should disfavor length. It has already been possible to block the 0 0 lethal effects of Pseudomonas aeruginosa toxin in basic P1 and P2 residues, in agreement with many 90 experimental results.23,25,32 mice with hexa-D-Arg compounds, and to The PCs differ in particular with respect to the delay anthrax toxin-induced toxaemia in mouse cells and in living rats and mice with even other substrate-binding subsites. Due to the pre- 91 sence of the acidic residues E257f and D259f, furin, longer D-Arg oligomers. Use of the small but PC4, PACE4 and PC5 seem to favour the binding of significant differences in the vicinity of the active additional basic P3 and/or P5 residues. In contrast, site should permit generation of inhibitors more PC1, PC2 and kexin accept other residues at P3 and specific towards the distinct PCs. P5, due to less negatively charged residues in and around the S3 and S5 subsites. In PC4 and PC5, but in particular in furin, the preference for basic Methods residues at P6, P7 and beyond parallels the increased number of charges in the more peripheral All amino acid sequences of the proprotein subsites (Figure 2(b)). The slightly different require- convertases to be modelled (PC1: spjP29120jNEC1_HU- ment for basic residues is in part also reflected by NEC1_HUMAN; PC2: spjP16519jNEC2_HUMAN, PC4: j j j j the variations in the non-prime residues of their tr Q62094 PCK4_MOUSE; PC5: sp Q92824 PCK5_HU- PCK5_HUMAN, PC7: spjQ16549jPCK7_HUMAN, primary activation cleavage segments. PACE4: spjP29122jPAC4_HUMAN) were taken from the The catalytic domain can juxtapose these many ExPASy server.92 Sequence manipulations, alignments evenly charged residues only at the cost of free and energy minimization were performed with the energy of folding outside of the active-site cleft. HOMOLOGY93 and DISCOVER94 modules of INSIGHT Proper stabilization might require the presence of a II.95 The models were inspected on the display and were supporting P domain. In the free enzymes, the manually corrected with MAIN.96 The sequences of the catalytic and of the P domain of active-site clefts may be partially unfolded, so that 97 98 the clustered negatively charged residues might human furin and of yeast kexin were aligned based on move to their functional position only after docking the topological equivalence of the two-domain crystal structures of mouse furin61 and yeast kexin62 (see of a basic substrate. In line with their varying charge Figure 2(a)). Structurally conserved (SCRs) and variable density, the PCs presumably require different regions (VRs) obtained automatically were optimized degrees of charge compensation to function effi- manually. Subsequently, for each of the model PCs the ciently: only a multiple basic substrate moiety sequence was aligned to the furin sequence by pairwise might be able to induce a functional active-site alignment99 and manually adapted according to a 224 Models and Specificity of Proprotein Convertases multiple sequence alignment of all PC sequences with 2. Thomas, G. (2002). Furin at the cutting edge: from Clustal W.100 After specification of the SCRs and VRs, protein traffic to embryogenesis and disease. Nat. initial PC models were built by transferring the SCR Rev. Mol. Cell Biol. 3, 753–766. coordinates from the reference protein (mostly furin) to 3. Taylor, N. A., van de Ven, W. J. M. & Creemers, the model proteins, and taking the VR coordinates either J. W. M. (2003). Curbing activation: proprotein from (mostly) furin or kexin, or (if none of the two convertases in homeostasis and pathology. FASEB J. template sequences had the same length as the target 17, 1215–1227. sequence) from selected loop structures with similar 4. Zhou, A., Webb, G., Zhu, X. R. & Steiner, D. F. (1999). lengths and anchor residues derived from the PDB (italic Proteolytic processing in the secretory pathway. letters against white background, see Figure 2(a)). The J. Biol. Chem. 274, 20745–20748. main chains were smoothed at splice points, where 5. Khatib, A. M., Siegfried, G., Chretien, M., Metrakos, different structures had been used as templates. Upon P. & Seidah, N. G. (2002). Proprotein convertases in visual inspection of each PC model, obvious side-chain tumor progression and malignancy: novel targets in clashes and deformations were relieved under selection cancer therapy. Am. J. Pathol. 160, 1921–1935. of energetically more favourable rotamers, the disulfide 6. Rockwell, N. C. & Thorner, J. W. (2004). The kindest bonds were established, hydrogen atoms were inserted cuts of all: crystal structures of Kex2 and furin reveal (calculated for a pH of 7.0), and the two calcium ions and secrets of precursor processing. Trends Biochem. Sci. the Arg-Val-Lys-Arg-chloromethylketone inhibitor were 29, 80–87. 61 built as observed in the dec-RVKR-furin structure. 7. Siezen, R. J., de Vos, W. M., Leunissen, J. A. & These PC models were subjected to different numbers Dijkstra, B. W. (1991). Homology modelling and (usually several thousand) of energy minimization cycles protein engineering strategy of subtilases, the family in vacuo by molecular mechanics using DISCOVER with 4 101 of subtilisin-like serine proteinases. Protein Eng. , the consistent valence force field (CVFF) and visual 719–737. inspections as well as manual interventions. In all cycles 8. Siezen, R. J. & Leunissen, J. A. M. (1997). Subtilases— the steepest descent method, a dielectric constant of 1, the superfamily of subtilisin-like serine proteases. and different constraints for main and side-chains of VRs Protein Sci. 6, 501–523. K1 ˚ K1 and SCRs (500 to 3000 kcal mol A ) were used. Due to 9. Fuller, R. S., Sterne, R. E. & Thorner, J. (1988). better convergence, the cross and Morse terms were Enzymes required for yeast prohormone processing. applied to the minimization steps, when a maximum Annu. Rev. Physiol. 50, 345–362. K1 ˚ K1 derivative of 10 kcal mol A was reached. This 10. Fuller, R. S., Brake, A. J. & Thorner, J. (1989). procedure was continued until convergence, normally Intracellular targeting and structural conservation K1 ˚ K1 below a maximum derivative of 5 kcal mol A .After of a prohormone-processing endoprotease. Science, each cycle and at the end, the models were checked by 102 246, 482–486. PROCHECK and visually inspected, bad conformations 11. Mizuno, K., Nakamura, T., Ohshima, T., Tanaka, S. & and clashes were relieved manually, and the minimization Matsuo, H. (1989). Characterization of KEX2-encoded procedure was repeated if necessary. The final average K1 ˚ K1 endopeptidase from yeast Saccharomyces cerevisiae. absolute derivatives were about 0.06 kcal mol A . Biochem. Biophys. Res. Commun. 159, 305–311. In the final PC models, about 98% of the non-glycine 12. Brenner, C. & Fuller, R. S. (1992). Structural and and non-proline main chain angles are (similar to the enzymatic characterization of a purified prohormone- reference structures) within the most favored and processing enzyme: secreted, soluble Kex2 protease. additionally allowed regions of the Ramachandran plot, Proc. Natl Acad. Sci. USA, 89, 922–926. while three to nine residues are in the generously allowed 13. Barrett, A. J., Rawlings, N. D. & Woessner, J. F. (1998). regions, and zero to three just at the border to these Handbook of Proteolytic Enzymes (2nd edit.), Academic regions.102 Of particular note is Cys211f forming a Press, London. disulfide bridge with Cys360f, which also exhibits a 14. Seidah, N. G., Benjannet, S., Wickham, L., likewise unfavorable conformation in the experimental Marcinkiewicz, J., Jasmin, S. B., Stifani, S. et al. furin and kexin structures. (2003). The secretory proprotein convertase neural The annotated coordinates of the modelled PC apoptosis-regulated convertase 1 (NARC-1): liver structures are available from the authors upon request. regeneration and neuronal differentiation. Proc. Natl Acad. Sci. USA, 100, 928–933. 15. Creemers, J. W. M., Siezen, R. J., Roebroek, A. J. M., Ayoubi, T. A. Y., Huylebroeck, D. & Van de Ven, W. J. M. (1993). Modulation of furin-mediated Acknowledgements proprotein processing activity by site-directed muta- genesis. J. Biol. Chem. 268, 21826–21834. We thank R. Huber for his continuous support. 16. Lusson, J., Benjannet, S., Hamelin, J., Savaria, D., The financial support by NIH grant DA05084 (I.L.), by Chretien, M. & Seidah, N. G. (1997). The integrity of grant TH 862/1-1 from the Deutsche Forschungsge- the rrgdl sequence of the proprotein convertase pc1 meinschaft (M.E.T.), by the Sonderforschungsbereich is critical for its zymogen and c-terminal processing 596 of the LM Universita¨tMu¨ nchen (M.E.T. & W.B.), and for its cellular trafficking. Biochem. J. 326, EU grants and the Fonds der Chemischen Industrie 737–744. (W.B.) are greatly acknowledged. 17. Zhou, A., Martin, S., Lipkind, G., Lamendola, J. & Steiner, D. F. (1998). Regulatory roles of the P domain of the subtilisin-like prohormone convertases. J. Biol. Chem. 273, 11107–11114. References 18. Muller, L., Cameron, A., Fortenberry, Y., Apletalina, E. V. & Lindberg, I. (2000). Processing and sorting of 1. Steiner, D. F. (1998). The proprotein convertases. the prohormone convertase 2 propeptide. J. Biol. Curr. Opin. Chem. Biol. 2, 31–39. Chem. 275, 39213–39222. Models and Specificity of Proprotein Convertases 225

19. Julius, D., Brake, A., Blair, L., Kunisawa, R. & Kisiel, W. (1998). Inhibition of soluble recombinant Thorner, J. (1984). Isolation of the putative structural furin by human proteinase inhibitor 8. J. Biol. Chem. gene for the lysine-arginine-cleaving endopeptidase 273, 1851–1854. required for processing of yeast prepro-alpha-factor. 34. Anderson, E. D., Thomas, L., Hayflick, J. S. & Cell, 37, 1075–1089. Thomas, G. (1993). Inhibition of HIV-1 gp160- 20. Molloy, S. S., Bresnahan, P. A., Leppla, S. H., Klimpel, dependent membrane fusion by a furin-directed K. R. & Thomas, G. (1992). Human furin is a calcium- alpha 1-antitrypsin variant. J. Biol. Chem. 268, dependent serine endoprotease that recognizes 24887–24891. the sequence Arg-X-X-Arg and efficiently cleaves 35. Jean, F., Stella, K., Thomas, L., Liu, G. P., Xiang, Y., anthrax toxin protective antigen. J. Biol. Chem. 267, Reason, A. J. & Thomas, G. (1998). Alpha(1)- 16396–16402. antitrypsin portland, a bioengineered serpin highly 21. Rockwell, N. C., Krysan, D. J., Komiyama, T. & selective for furin—application as an antipathogenic Fuller, R. S. (2002). Precursor processing by Kex2/- agent. Proc. Natl Acad. Sci. USA, 95, 7293–7298. furin proteases. Chem. Rev. 102, 4525–4548. 36. Tsuji, A., Ikoma, T., Hashimoto, E. & Matsuda, Y. 22. Schechter, I. & Berger, A. (1968). On the active site of (2002). Development of selectivity of alpha 1- proteases. 3. Mapping the active site of papain; antitrypsin variant by mutagenesis in its reactive specific peptide inhibitors of papain. Biochem. site loop against proprotein convertase. A crucial Biophys. Res. Commun. 32, 898–902. role of the P4 arginine in PACE4 inhibition. Protein 23. Nakayama, K. (1997). Furin—a mammalian subtili- Eng. 15, 123–130. sin/kex2p-like endoprotease involved in processing 37. Lu, W., Zhang, W., Molloy, S. S., Thomas, G., Ryan, of a wide variety of precursor proteins. Biochem. J. K., Chiang, Y. et al. (1993). Arg15-Lys17-Arg18 turkey 327, 625–635. ovomucoid third domain inhibits human furin. 24. Hosaka, M., Nagahama, M., Kim, W. S., Watanabe, J. Biol. Chem. 268, 14583–14585. T., Hatsuzawa, K., Ikemizu, J. et al. (1991). Arg-X- 38. Komiyama, T. & Fuller, R. S. (2000). Engineered eglin Lys/Arg-Arg motif as a signal for precursor cleavage c variants inhibit yeast and human proprotein catalyzed by furin within the constitutive secretory processing proteases, Kex2 and furin. Biochemistry, pathway. J. Biol. Chem. 266, 12127–12130. 39, 15156–15165. 25. Duckert, P., Brunak, S. & Blom, N. (2004). Prediction 39. Komiyama, T., VanderLugt, B., Fugere, M., Day, R., of proprotein convertase cleavage sites. Protein Eng. Kaufman, R. J. & Fuller, R. S. (2003). Optimization of Des. Selection, 17, 107–112. protease-inhibitor interactions by randomizing 26. Bravo, D. A., Gleason, J. B., Sanchez, R. I., Roth, R. A. adventitious contacts. Proc. Natl Acad. Sci. USA, & Fuller, R. S. (1994). Accurate and efficient 100, 8205–8210. cleavage of the human insulin proreceptor by the 40. Richer, M. J., Keays, C. A., Waterhouse, J., Minhas, J., human proprotein-processing protease furin. Hashimoto, C. & Jean, F. (2004). The Spn4 gene of Characterization and kinetic parameters using the Drosophila encodes a potent furin-directed secretory purified, secreted soluble protease expressed by pathway serpin. Proc. Natl Acad. Sci. USA, 101, a recombinant baculovirus. J. Biol. Chem. 269, 10560–10565. 25830–25837. 41. Osterwalder, T., Kuhnen, A., Leiserson, W. M., Kim, 27. Watanabe, T., Nakagawa, T., Ikemizu, J., Nagahama, Y.-S. & Keshishian, H. (2004). Drosophila serpin 4 M., Murakami, K. & Nakayama, K. (1992). Sequence functions as a neuroserpin-like inhibitor of subtili- requirements for precursor cleavage within the sin-like proprotein convertases. J. Neurosci. 24, constitutive secretory pathway. J. Biol. Chem. 267, 5482–5491. 8270–8274. 42. Fugere, M., Limperis, P. C., Beaulieu-Audy, V., 28. Krysan, D. J., Rockwell, N. C. & Fuller, R. S. (1999). Gagnon, F., Lavigne, P., Klarskov, K. et al. (2002). Quantitative characterization of furin specificity. Inhibitory potency and specificity of subtilase-like Energetics of substrate discrimination using an pro-protein convertase (SPC) prodomains. J. Biol. internally consistent set of hexapeptidyl methyl- Chem. 277, 7648–7656. coumarinamides. J. Biol. Chem. 274, 23229–23234. 43. Boudreault, A., Gauthier, D. & Lazure, C. (1998). 29. Jean, F., Boudreault, A., Basak, A., Seidah, N. G. & Proprotein convertase PC1/3-related peptides are Lazure, C. (1995). Fluorescent peptidyl substrates as potent slow tight-binding inhibitors of murine an aid in studying the substrate specificity of human PC1/3 and Hfurin. J. Biol. Chem. 273, 31574–31580. prohormone convertase Pc1 and human furin and 44. Bhattacharjya, S., Xu, P., Zhong, M., Chretien, M., designing a potent irreversible inhibitor. J. Biol. Seidah, N. G. & Ni, F. (2000). Inhibitory activity and Chem. 270, 19225–19231. structural characterization of a C-terminal peptide 30. Johanning, K., Juliano, M. A., Juliano, L., Lazure, C., fragment derived from the prosegment of the Lamango, N. S., Steiner, D. F. & Lindberg, I. (1998). proprotein convertase PC7. Biochemistry, 39, Specificity of prohormone convertase 2 on proenke- 2868–2877. phalin and proenkephalin-related substrates. J. Biol. 45. Lesage, G., Tremblay, M., Guimond, J. & Boileau, G. Chem. 273, 22672–22680. (2001). Mechanism of Kex2p inhibition by its 31. Lamango, N. S., Zhu, X. R. & Lindberg, I. (1996). proregion. FEBS Letters, 508, 332–336. Purification and enzymatic characterization of 46. Anderson, E. D., Vanslyke, J. K., Thulin, C. D., Jean, recombinant prohormone convertase 2—stabiliza- F. & Thomas, G. (1997). Activation of the furin tion of activity By 21 Kda 7b2. Arch. Biochem. Biophys. endoprotease is a multiple-step process—require- 330, 238–250. ments for acidification and internal propeptide 32. Cameron, A., Apletalina, E. & Lindberg, I. (2001). The cleavage. EMBO J. 16, 1508–1518. enzymology of PC1 and PC2. Enzyms, 22, 291–332. 47. Laurent, V., Jaubert-Miazza, L., Desjardins, R., Day, 33. Dahlen, J. R., Jean, F., Thomas, G., Foster, D. C. & R. & Lindberg, I. (2004). Biosynthesis of 226 Models and Specificity of Proprotein Convertases

proopiomelanocortin-derived peptides in prohor- Eisenstein, E., Toedt, J. & Orban, J. (2001). Stability mone convertase 2 and 7B2 null mice. Endocrinology, and global fold of the mouse prohormone convertase 145, 519–528. 1 pro-domain. Biochemistry, 40, 5488–5495. 48. Lindberg, I., van den Hurk, W. H., Bui, C. & Batie, 65. Tangrea, M. A., Bryan, P. N., Sari, N. & Orban, J. C. J. (1995). Enzymatic characterization of immuno- (2002). Solution structure of the pro-hormone con- purified prohormone convertase 2: potent inhibition vertase 1 pro-domain from Mus musculus. J. Mol. by a 7B2 peptide fragment. Biochemistry. 34, 5486–5493. Biol. 320, 801–812. 49. Cameron, A., Fortenberry, Y. & Lindberg, I. (2000). 66. Gallagher, T., Gilliland, G., Wang, L. & Bryan, P. The SAAS granin exhibits structural and functional (1995). The prosegment-subtilisin bpn complex— homology to 7B2 and contains a highly potent crystal structure of a specific foldase. Structure, 3, hexapeptide inhibitor of PC1. FEBS Letters, 473, 907–914. 135–138. 67. Jain, S. C., Shinde, U., Li, Y., Inouye, M. & Berman, 50. Mbikay, M., Seidah, N. G. & Chretien, M. (2001). H. M. (1998). The crystal structure of an auto- Neuroendocrine secretory protein 7B2: structure, processed Ser221Cys-subtilisin E-propeptide com- expression and functions. Biochem. J. 357, 329–342. plex at 2.0 a resolution. J. Mol. Biol. 284, 137–144. 68. Comellas-Bigler, M., Fuentes-Prior, P., Maskos, K., 51. Apletalina, E., Appel, J., Lamango, N. S., Houghten, Huber, R., Oyama, H., Uchida, K. et al. (2002). The R. A. & Lindberg, I. (1998). Identification of inhibitors 1.4 A˚ crystal structure of kumamolysin: a thermo- of prohormone convertases 1 and 2 using a peptide stable serine-carboxyl-type proteinase. Structure, 10, combinatorial library. J. Biol. Chem. 273, 26589–26595. 865–876. 52. Cameron, A., Appel, J., Houghten, R. A. & Lindberg, 69. Comellas-Bigler, M., Maskos, K., Huber, R., Oyama, I. (2000). Polyarginines are potent furin inhibitors. H., Oda, K. & Bode, W. (2004). 1.2 A˚ crystal structure J. Biol. Chem. 275, 36741–36749. of the serine carboxyl peptidase pro-kumamolisin: 53. Fugere, M. & Day, R. (2002). Inhibitors of the first structure of a full-length pro-subtilase. Structure, subtilase-like pro-protein convertases (SPCs). Curr. 12, 1313–1323. Pharm. Des. 8, 549–562. 70. Bissonnette, L., Charest, G., Longpre, J. M., Lavigne, 54. Basak, A., Jean, F., Seidah, N. G. & Lazure, C. (1994). P. & Leduc, R. (2004). Identification of furin pro- Design and synthesis of novel inhibitors of pro- region determinants involved in folding and acti- hormone convertases. Intern. J. Pept. Prot. Res. 44, vation. Biochem. J. 379, 757–763. 253–261. 71. Li, Y., Hu, Z., Jordan, F. & Inouye, M. (1995). 55. Angliker, H., Wikstrom, P., Shaw, E., Brenner, C. & Functional analysis of the propeptide of subtilisin Fuller, R. S. (1993). The synthesis of inhibitors for E as an intramolecular chaperone for protein folding. processing proteinases and their action on the Kex2 Refolding and inhibitory abilities of propeptide proteinase of yeast. Biochem. J. 293, 75–81. mutants. J. Biol. Chem. 270, 25127–25132. 56. Holyoak, T., Kettner, C. A., Petsko, G. A., Fuller, R. S. 72. Eder, J., Rheinnecker, M. & Fersht, A. R. (1993). & Ringe, D. (2004). Structural basis for differences in Folding of subtilisin BPN0: role of the pro-sequence. substrate selectivity in Kex2 and furin protein J. Mol. Biol. 233, 293–304. convertases. Biochemistry, 43, 2412–2421. 73. Anderson, E. D., Molloy, S. S., Jean, F., Fei, H., 57. Kacprzak, M. M., Peinado, J. R., Than, M. E., Appel, Shimamura, S. & Thomas, G. (2002). The ordered J., Henrich, S., Lipkind, G. et al. (2004). Inhibition of and compartment-specific autoproteolytic removal furin by polyarginine-containing peptides: nanomo- of the furin intramolecular chaperone is required for lar inhibition by nona-D-arginine. J. Biol. Chem. 279, enzyme activation. J. Biol. Chem. 277, 12879–12890. 36788–36794. 74. Molloy, S. S., Thomas, L., VanSlyke, J. K., Stenberg, 58. Siezen, R. J., Creemers, J. W. & Van de Ven, W. J. P. E. & Thomas, G. (1994). Intracellular trafficking (1994). Homology modelling of the catalytic domain of and activation of the furin proprotein convertase: human furin. A model for the eukaryotic subtilisin- localization to the TGN and recycling from the cell like proprotein convertases. Eur. J. Biochem. 222, surface. EMBO J. 13, 18–33. 255–266. 75. Vey, M., Schafer, W., Berghofer, S., Klenk, H. D. & 59. Lipkind, G., Gong, Q. & Steiner, D. F. (1995). Garten, W. (1994). Maturation of the trans-Golgi network protease furin: compartmentalization of Molecular modeling of the substrate specificity of propeptide removal, substrate cleavage, and COOH- prohormone convertases SPC2 and SPC3. J. Biol. terminal truncation. J. Cell. Biol. 127, 1829–1842. Chem. 270, 13277–13284. 76. Hwang, J. R. & Lindberg, I. (2001). Inactivation of the 60. Lipkind, G. M., Zhou, A. & Steiner, D. F. (1998). A 7B2 inhibitory CT peptide depends on a functional model for the structure of the P domains in the furin cleavage site. J. Neurochem. 79, 437–444. subtilisin-like prohormone convertases. Proc. Natl 77. Martens, G. J. M., Braks, J. A. M., Vanhorssen, A. M., Acad. Sci. USA. 95, 7310–7315. Broers, C. A. M. & Kuiper, R. P. (1996). The 61. Henrich, S., Cameron, A., Bourenkov, G. P., neuroendocrine chaperone 7b2 assists transport of Kiefersauer, R., Huber, R., Lindberg, I. et al. (2003). prohormone convertase Pc2 through the secretory The crystal structure of the proprotein processing pathway. J. Physiol. 1996;, S. proteinase furin explains its stringent specificity. 78. Zhu, X., Muller, L., Mains, R. E. & Lindberg, I. (1998). Nature Struct. Biol. 10, 520–526. Structural elements of PC2 required for interaction 62. Holyoak, T., Wilson, M. A., Fenn, T. D., Kettner, with its helper protein 7B2. J. Biol. Chem. 273, C. A., Petsko, G. A., Fuller, R. S. & Ringe, D. (2003). 1158–1164. 2.4 angstrom resolution crystal structure of the 79. Apletalina, E. V., Muller, L. & Lindberg, I. (2000). prototypical hormone-processing protease Kex2 in Mutations in the catalytic domain of prohormone complex with an Ala-Lys-Arg boronic acid inhibitor. convertase 2 result in decreased binding to 7B2 and Biochemistry, 42, 6709–6718. loss of inhibition with 7B2 C-terminal peptide. J. Biol. 64. Tangrea, M. A., Alexander, P., Bryan, P. N., Chem. 275, 14667–14677. Models and Specificity of Proprotein Convertases 227

80. Lamango, N. S., Apletalina, E., Liu, J. & Lindberg, I. 93. Homology. (2000). Molecular simulations Inc. March (1999). The proteolytic maturation of prohormone 2000 edit., San Diego, CA. convertase 2 (PC2) is a pH-driven process. Arch. 94. Discover. (1994). Molecular simulations Inc. May Biochem. Biophys. 362, 275–282. 1994 edit., San Diego, CA. 81. Basak, S., Chretien, M., Mbikay, M. & Basak, A. 95. Insight_II. (2000). Molecular simulations Inc. June (2004). In vitro elucidation of substrate specificity 2000 edit., San Diego, CA. and bioassay of proprotein convertase 4 using 96. Turk, D. (1992). Weiterentwicklung eines Programms intramolecularly quenched fluorogenic peptides. fu¨ r Moleku¨ lgraphik und Elektronendichte-Manipu- Biochem. J. 380, 505–514. lation und seine Anwendung auf verschiedene 82. Bevan, A., Brenner, C. & Fuller, R. S. (1998). Protein-Strukturaufkla¨rungen. Ph.D. Thesis, Tech- Quantitative assessment of enzyme specificity nische Universita¨tMu¨ nchen. in vivo—P-2 recognition by kex2 protease defined 97. van den Ouweland, A. M., van Duijnhoven, H. L., in a genetic system. Proc. Natl Acad. Sci. USA, 95, Keizer, G. D., Dorssers, L. C. & Van de Ven, W. J. 10384–10389. (1990). Structural homology between the human fur 83. Munzer, J. S., Basak, A., Zhong, M., Mamarbachi, A., gene product and the subtilisin-like protease Hamelin, J., Savaria, D. et al. (1997). In vitro encoded by yeast KEX2. Nucl. Acids Res, 18, 664. characterization of the novel proprotein convertase 98. Mizuno, K., Nakamura, T., Ohshima, T., Tanaka, S. & PC7. J. Biol. Chem. 272, 19672–19681. Matsuo, H. (1988). Yeast KEX2 genes encodes an 84. Basak, A., Toure, B. B., Lazure, C., Mbikay, M., endopeptidase homologous to subtilisin-like serine Chretien, M. & Seidah, N. G. (1999). Enzymic proteases. Biochem. Biophys. Res. Commun. 156, characterization in vitro of recombinant proprotein 246–254. convertase PC4. Biochem. J. 343, 29–37. 99. Needleman, S. B. & Wunsch, C. D. (1970). A general 85. Bode, W., Papamokos, E. & Musil, D. (1987). The method applicable to the search for similarities in the high-resolution X-ray crystal structure of the amino acid sequence of two proteins. J. Mol. Biol. 48, 443–453. complex formed between subtilisin Carlsberg and 100. Thompson, J. D., Higgins, D. G. & Gibson, T. J. eglin c, an elastase inhibitor from the leech Hirudo (1994). CLUSTAL W: improving the sensitivity of medicinalis. Structural analysis, subtilisin structure progressive multiple sequence alignment through and interface geometry. Eur. J. Biochem. 166, 673–692. sequence weighting, position-specific gap penalties 86. Lazure, C., Gauthier, D., Jean, F., Boudreault, A., and weight matrix choice. Nucl. Acids Res. 22, Seidah, N. G., Bennett, H. P. J. & Hendy, G. N. (1998). 4673–4680. In vitro cleavage of internally quenched fluorogenic 101. Dauber-Osguthorpe, P., Roberts, V. A., Osguthorpe, human proparathyroid hormone and proparathyroid- D. J., Wolff, J., Genest, M. & Hagler, A. T. (1988). related peptide substrates by furin. Generation of a Structure and energetics of ligand binding to Potent Inhibitor. J. Biol. Chem. 273, 8572–8580. proteins Escherichia coli dihydrofolate reductase- 87. Rozan, L., Krysan, D. J., Rockwell, N. C. & Fuller, trimethoprim a drug-receptor system. Proteins: R. S. (2004). Plasticity of extended subsites facilitates Struct. Funct. Genet. 4, 31–47. divergent substrate-recognition by Kex2 and furin. 102. Laskowski, R. A., MacArthur, M. W., Moss, D. S. & J. Biol. Chem. 279, 35656–35663. Thornton, J. M. (1993). PROCHECK: a program to 89. Oliva, A. A., Jr, Chan, S. J. & Steiner, D. F. (2000). check the stereochemical quality of protein struc- Evolution of the prohormone convertases: identifi- tures. J. Appl. Crystallog. 26, 283–291. cation of a homologue of PC6 in the protochordate 103. Kraulis, P. J. (1991). MOLSCRIPT: a program to amphioxus. Biochim. Biophys. Acta, 1477, 338–348. produce both detailed and schematic plots of protein 90. Sarac, M. S., Cameron, A. & Lindberg, I. (2002). The structures. J. Appl. Crystallog. 24, 946–950. furin inhibitor hexa-D-arginine blocks the activation 104. Merritt, E. A. & Bacon, D. J. (1997). Raster3D of Pseudomonas aeruginosa exotoxin A in vivo. Infect. photorealistic molecular graphics. Methods Enzymol. Immun. 70, 7136–7139. 277, 505–527. 91. Sarac, M. S., Peinado, J. R., Leppla, S. H. & Lindberg, 105. Berman, H. M., Westbrook, J., Feng, Z., Gilliland, G., I. (2004). Protection against anthrax toxemia by Bhat, T. N., Weissig, H. et al. (2000). The protein data hexa-D-arginine in vitro and in vivo. Infect. Immun. 72, bank. Nucl. Acids Res. 28, 235–242. 602–605. 106. Nicholls, A., Bharadwaj, R. & Honig, B. (1993). 92. Gasteiger, E., Gattiker, A., Hoogland, C., Ivanyi, I., Grasp: graphical representation and analysis of Appel, R. D. & Bairoch, A. (2003). ExPASy: the surface properties. Biophys. J. 64, 166–170. proteomics server for in-depth protein knowledge 107. Barton, G. J. (1993). ALSCRIPT: a tool to format and analysis. Nucl. Acids Res. 31, 3784–3788. multiple sequence alignments. Protein Eng. 6, 37–40.

Edited by I. Wilson

(Received 10 August 2004; received in revised form 14 October 2004; accepted 15 October 2004)