The Crystal Structure of PCSK9: A Regulator of Plasma LDL-Cholesterol

Structure

Article The Crystal Structure of PCSK9: A Regulator of Plasma LDL-Cholesterol Derek E. Piper,1 Simon Jackson,2 Qiang Liu,3 William G. Romanow,1 Susan Shetterly,2 Stephen T. Thibault,1 Bei Shan,2 and Nigel P.C. Walker1,* 1

Department of Molecular Structure Department of Metabolic Disorders 3 Department of Protein Science Amgen Inc., 1120 Veterans Boulevard, South San Francisco, California 94080, USA *Correspondence: [email protected] DOI 10.1016/j.str.2007.04.004 2

SUMMARY

Proprotein convertase subtilisin kexin type 9 (PCSK9) has been shown to be involved in the regulation of extracellular levels of the lowdensity lipoprotien receptor (LDLR). Although PCSK9 is a subtilase, it has not been shown to degrade the LDLR, and its LDLR-lowering mechanism remains uncertain. Here we report the crystal structure of human PCSK9 at 2.3 A˚ resolution. PCSK9 has subtilisin-like pro- and catalytic domains, and the stable interaction between these domains prevents access to PCSK9’s catalytic site. The C-terminal domain of PCSK9 has a novel protein fold and may mediate protein-protein interactions. The structure of PCSK9 provides insight into its biochemical characteristics and biological function. INTRODUCTION Proprotein convertase subtilisin kexin type 9 (PCSK9) is a serine protease involved in regulating the levels of the low-density lipoprotein receptor (LDLR) protein (Horton et al., 2007; Seidah and Prat, 2007). In vitro experiments have shown that adding PCSK9 to HepG2 cells lowers the levels of cell surface LDLR (Benjannet et al., 2004; Lagace et al., 2006; Maxwell et al., 2005; Park et al., 2004). Experiments with mice have shown that increasing PCSK9 protein levels decreases levels of LDLR protein in the liver (Benjannet et al., 2004; Lagace et al., 2006; Maxwell et al., 2005; Park et al., 2004), while PCSK9 knockout mice have increased levels of LDLR in the liver (Rashid et al., 2005). Additionally, various human PCSK9 mutations that result in either increased or decreased levels of plasma LDL have been identified (Berge et al., 2006; Kotowski et al., 2006; Zhao et al., 2006). PCSK9 has been shown to directly interact with the LDLR protein, to be endocytosed along with the LDLR, and to coimmunofluoresce with the LDLR throughout the endosomal pathway (Lagace et al., 2006). Degradation of the LDLR by PCSK9 has not been observed, and the mechanism

through which it lowers extracellular LDLR protein levels is still uncertain. PCSK9 is a prohormone-proprotein convertase in the subtilisin (S8) family of serine proteases (Seidah et al., 2003). Humans have nine prohormone-proprotein convertases that can be divided between the S8A and S8B subfamilies (Rawlings et al., 2006). Furin, PC1/PC3, PC2, PACE4, PC4, PC5/PC6, and PC7/PC8/LPC/SPC7 are classified in subfamily S8B. Crystal and NMR structures of different domains from mouse furin and PC1 reveal subtilisin-like pro- and catalytic domains, and a P domain directly C-terminal to the catalytic domain (Henrich et al., 2003; Tangrea et al., 2002). Based on the amino acid sequence similarity within this subfamily, all seven members are predicted to have similar structures (Henrich et al., 2005). SKI-1/S1P and PCSK9 are classified in subfamily S8A. Sequence comparisons with these proteins also suggest the presence of subtilisin-like pro- and catalytic domains (Sakai et al., 1998; Seidah et al., 1999, 2003). In these proteins the amino acid sequence C-terminal to the catalytic domain is more variable and does not suggest the presence of a P domain. Prohormone-proprotein convertases are expressed as zymogens, and they mature through a multistep process. The function of the prodomain in this process is two-fold. The prodomain first acts as a chaperone and is required for proper folding of the catalytic domain (Ikemura et al., 1987). Once the catalytic domain is folded, autocatalysis occurs between the prodomain and catalytic domain. Following this initial cleavage reaction, the prodomain remains bound to the catalytic domain, where it then acts as an inhibitor of catalytic activity (Fu et al., 2000). When conditions are correct, maturation proceeds with a second autocatalytic event at a site within the prodomain (Anderson et al., 1997). After this second cleavage event occurs, the prodomain and catalytic domain dissociate, giving rise to an active protease. Autocatalysis of the PCSK9 zymogen occurs between Gln152 and Ser153 (VFAQjSIP) (Naureckiene et al., 2003) and has been shown to be required for its secretion from cells (Seidah et al., 2003). A second autocatalytic event at a site within PCSK9’s prodomain has not been observed. Purified PCSK9 is made up of two species that can be separated by nonreducing SDS-PAGE: the prodomain at

Structure 15, 545–552, May 2007 ª2007 Elsevier Ltd All rights reserved 545

Structure The Crystal Structure of PCSK9

Table 1. Data Collection and Refinement Statistics Crystal Name

Form 1-Native

Form 1-Hg

Form 1-Xe

Form 2-Native

Wavelength (A˚)

1.0717

1.0063

2.0

1.0

Space group

P213

P213

P213

P212121

a = b = c = 139.77

a = b = c = 140.33

a = b = c = 140.15

63.11, 70.71, 150.69

Data Collection

Cell dimensions a, b, c (A˚) 

a = b = g = 90

a = b = g = 90

a = b = g = 90

a = b = g = 90

Resolution (A˚)

a, b, g ( )

40–3.1 (3.21–3.1)a

40–3.35 (3.47–3.35)

40–3.15 (3.26–3.15)

40–2.3 (2.38–2.3)

Completeness

99.8 (99.9)

99.9 (100)

99.9 (100)

97.3 (81.7)

Redundancy

5.3

12.1

5.8

4.9

Rsym

6.2 (50.3)

8.4 (46.7)

7.3 (43.7)

12.4 (37.6)

I/sI

20.1 (2.5)

30.9 (5.1)

19.3 (4.0)

10.8 (2.2)

Refinement Resolution (A˚)

40–2.3

Molecules/ASU

1

Reflections Total

29,916

Working set

28,405

Test set

1511

Rwork/Rfree

0.196/0.232

Number of atoms Protein

4,359

Ion

10

Water

215

B factors Protein

35.83

Ion

95.54

Water

38.06

Rms deviations Bond lengths (A˚)

0.005



Bond angles ( ) a

0.846

Numbers in parenthesis are for the highest resolution shell.

17 kDa and the catalytic plus C-terminal domains at 65 kDa. PCSK9 has not been isolated without its inhibitory prodomain, and measurements of PCSK9’s catalytic activity have been variable (Naureckiene et al., 2003; Seidah et al., 2003). In order to gain insight into the regulation of PCSK9’s catalytic activity and how PCSK9 lowers the level of the LDLR protein, we have determined the crystal structure of full-length wild-type PCSK9. RESULTS AND DISCUSSION Overall Structure of PCSK9 We solved the structure of PCSK9 by multiple isomorphous replacement with anomalous scattering and refined

it to 2.3 A˚ resolution (see Experimental Procedures and Table 1). The crystal structure reveals that PCSK9 has subtilisin-like pro- and catalytic domains, and a novel Cterminal domain (termed V domain) (Figure 1). Although full-length PCSK9 (31–692) was used for crystallization, electron density is present only for residues 61–683. The prodomain extends from residues 61 to 152, and autocatalysis has broken the peptide bond between Gln152 and Ser153. The catalytic domain extends from residues 153 to 447 but has two loops lacking electron density. The V domain extends from residues 452 to 683 with two disordered regions. While there is no electron density for the polypeptide chain that connects the catalytic domain to the V domain, the placement of the V domain in the overall

546 Structure 15, 545–552, May 2007 ª2007 Elsevier Ltd All rights reserved

Structure The Crystal Structure of PCSK9

the assigned position of the V domain is consistent in both of the crystal forms obtained with this protein. Although the PCSK9 protein used for crystallization is glycosylated at residue Asn533, there is very poor electron density for the oligosaccharide and it was not included in the modeled structure.

Figure 1. Overall Structure of the PCSK9 Protein Ribbons diagram of the structure with the prodomain in magenta, the catalytic domain in wheat, and the V domain in blue. Thr61 marks the first observed residue, and Gln152 marks the C terminus of the prodomain. Ser153 marks the N terminus of the catalytic domain.

model is certain for two reasons. First, with only four disordered residues in the linker between the catalytic and V domains, the linker is not long enough to reach symmetry-related V domains within the crystal lattice. Second,

The PCSK9 Prodomain Inhibits Catalytic Activity The core of PCSK9’s prodomain closely resembles the prodomain of subtilisin (Gallagher et al., 1995; Jain et al., 1998), with two a helices and a four-stranded antiparallel b sheet. The b sheet forms the interface between the prodomain and the catalytic domain through an extensive network of both hydrophobic and electrostatic interactions, and in PCSK9 this interface buries 1300 A˚2 of solvent-accessible surface from each molecule. The four Cterminal amino acids of the prodomain (PCSK9 residues 149–152) bind at the catalytic site as a b strand that lies antiparallel to a strand from the catalytic domain, in an analogous fashion to an inhibitory peptide (P4–P1, respectively) (Figure 2A). Extending from the prodomain core are 14 additional amino acids (Thr61–Leu74) ordered at PCSK9’s N terminus (Figure 2B). This 14 amino acid extension begins with a b strand that packs parallel against the core b sheet, extends into a 310 helix, and turns back Figure 2. The Prodomain Inhibitory Peptide Binds in the Catalytic Site (A) Stereo view of the catalytic site with 2Fo  Fc electron density contoured at 1 s. P1 residue Gln152 and P3 residue Phe150 are labeled. His226 and Ser386 of the catalytic triad are also shown. (B) The N-terminal extension interacts with the core b sheet and inhibitory peptide. The first ordered residue in the subtilisin prodomain corresponds with PCSK9 amino acid residue Pro75.

Structure 15, 545–552, May 2007 ª2007 Elsevier Ltd All rights reserved 547

Structure The Crystal Structure of PCSK9

to reach the N terminus observed in the subtilisin prodomain. These additional N-terminal residues form extensive contacts with the inhibitory peptide. The side chain of Trp72 from the 310 helix makes a hydrophobic interaction with the side chain of Phe150, the P3 residue. The side chain of Arg73 forms hydrogen bonds with the side chain of Glu145. Backbone hydrogen bonds are found between the carbonyl of Trp72 and the amide nitrogen of Ser148, and the amide nitrogen of Leu74 and the carbonyl of Asp146. The structure of the PCSK9 prodomain/catalytic domain complex exhibits a significant difference to the four previous subtilase complex structures. Crystals for all previous complexes were grown using proteases with active-site mutations that lessen or abolish their catalytic activity (Comellas-Bigler et al., 2004; Gallagher et al., 1995; Jain et al., 1998; Kim et al., 2004). In the case of subtilisin and kumamolisin, these mutations were required to prevent autocatalytic degradation of the prodomain and maturation of the protease into its active form. In the case of fervidolysin, inactive protease was required to obtain a homogenous sample for crystallization. While fervidolysin and kumamolisin also contain additional ordered amino acids at the N termini of their prodomains, neither of these extensions form extensive contacts with their prodomain C termini. In our structural work, wild-type PCSK9 protein with a functional catalytic triad was used. It is likely that the extra stabilization of the inhibitory peptide in the catalytic site, provided by the N-terminal extension of the prodomain, blocks further access to the catalytic triad and prevents a subsequent autocatalytic reaction from occurring. PCSK9’s prodomain retains its structure and remains tightly associated to the catalytic domain, where it inhibits further catalytic activity. In endeavors to obtain measurable catalytic activity, multiple methods to facilitate the release of the prodomain were investigated. Amino acid residues at the interface of the pro- and catalytic domains were mutated in an attempt to weaken the interaction between these two domains. Amino acid residues at the interface of the N-terminal extension and inhibitory peptide were mutated in an attempt to weaken the stabilization of the inhibitory peptide in the catalytic site and permit a second autocatalytic event. Wild-type and correctly folded mutant proteins (as determined by secretion of nonaggregated protein) were exposed to various buffer conditions (pH changes, high salt, detergents, and denaturing reagents) in hopes of destabilizing the complex. In all nondenaturing conditions tested, the prodomain remained associated with the catalytic domain. Under denaturing conditions dissociation was observed, but the resulting catalytic domain formed high molecular weight aggregates (unpublished data). The Catalytic Domain The architecture of PCSK9’s catalytic domain is similar to that observed for other members in the subtilisin family. The core of this domain is composed of a seven-stranded parallel b sheet, flanked on both sides by a helices. The N-terminal a helix of the catalytic domain has undergone

a substantial shift in conformation, moving more than 25 A˚ from Gln152 (Figure 1), and is found in a position similar to that observed with other mature subtilases (Gallagher et al., 1995). It is possible that this conformational change, brought about by autocatalysis, is the trigger required to permit secretion. The loop that links the N-terminal helix to the catalytic domain (residues 168–175) and a second loop (residues 212–218) both lack electron density and were not modeled into the structure. Three disulfide bonds are found within the catalytic domain. As previously seen in related structures, two a helices from the catalytic domain form the interface for interaction with the prodomain. The catalytic triad of PCSK9 is made up of residues Asp186, His226, and Ser386. When compared to other subtilases, minor differences can be seen in the positions of the catalytic triad residues due to differences in the surrounding amino acid residues (Betzel et al., 1988; Henrich et al., 2003; Siezen and Leunissen, 1997). In PCSK9, Pro288 causes a slight shift in the orientation of the side chain of Asp186, and Ala390 causes a slight shift in the position of Ser386. Normally, the amino acid at the position corresponding with Pro288 is a serine, whose free backbone amide nitrogen plays a role in the orientation of the catalytic aspartic acid through a water-mediated hydrogen bond. The amino acid at the position corresponding to Ala390 is normally a proline. The proline at this position breaks an a helix and pushes the catalytic serine slightly out toward solvent. In PCSK9 the a helix is able to continue, pulling the catalytic serine in toward the protein where its side chain hydroxyl is 2.8 A˚ from the carbonyl of Pro288. Although these differences are clearly not enough to prevent PCSK9’s catalytic activity (as exemplified by its autocatalysis in vitro), we find it interesting that these normally highly conserved amino acids differ when compared to other subtilases. The S1 pocket of PCSK9 is shallower when compared to other subtilases. Based on sequence alignments (Siezen and Leunissen, 1997), subtilisin-like serine proteases almost exclusively have glycine residues on both sides of their S1 pockets. In PCSK9, one of these residues is an alanine rather than a glycine. The side chain of Ala290 protrudes into the S1 pocket, thereby raising the floor of this pocket. Gln152 from the prodomain binds at the S1 site where its side chain curls up and out of the pocket and forms hydrogen bonds to the backbone carbonyl of Phe150 and the backbone amide nitrogen of Ala315. The V Domain The V domain of PCSK9 is a novel, cylindrically shaped domain with quasi 3-fold internal symmetry (Figure 3 and Figure S1). The principal fold for each of the three subdomains within this domain is a b sandwich made up of six antiparallel b strands in a truncated jelly-roll motif. Each b sandwich is held together by three internal disulfide bonds (from strands 1 to 6, 2 to 6, and 3 to 5), for a total of nine disulfide bonds within the V domain. Other than the conserved cysteine residues, there is low sequence identity between the three subdomains. While there is clear overall 3-fold symmetry within this domain, the second b sandwich

548 Structure 15, 545–552, May 2007 ª2007 Elsevier Ltd All rights reserved

Structure The Crystal Structure of PCSK9

Figure 3. The V Domain Has Quasi 3-Fold Internal Symmetry Subdomain 1 is in blue, subdomain 2 is in green, and subdomain 3 is in yellow. The b strands in subdomain 1 are numbered. Subdomain 2 is missing b strand 4 (disordered region between Leu571 and Pro585) that would fill the space between subdomains 2 and 3. The N(Gly452) and C termini (His683) of the V domain are shown.

lacks electron density for strand 4 and has five rather than six b strands. This strand is missing in the structures from both crystal forms. The region of missing amino acid residues (572–584) contains three proline residues that possibly prevent this segment from forming a b strand. In subdomains 1 and 3, strand 4 packs against strand 5 from the same subdomain and strand 3 from the neighboring subdomain. The lack of this strand leaves a cavity in the surface of the V domain between subdomains 2 and 3, with the space partially filled by a loop from subdomain 3. While a structure similarity search utilizing the complete V domain does not produce any good matches, SSM (Krissinel and Henrick, 2004) returns a structurally related protein when a single b sandwich is used as the search model. Each single b sandwich of PCSK9 closely resembles the N-acetylgalactosamine-binding lectin, Helix pomatia agglutinin (HPA) (Sanchez et al., 2006). The b sandwich of the HPA molecule is a six-stranded, truncated jelly roll with one internal disulfide bond. In the HPA molecule three separate b sandwiches associate to form a homotrimer, which then binds end-to-end with another homotrimer to form the biologically active molecule. The V domain of PCSK9 resembles the homotrimer of the HPA protein. The interaction between the V domain and the catalytic domain is not extensive, burying 700 A˚2 of solventaccessible surface from each molecule. There are three direct hydrogen bonds at the interface of these domains. The side chain of Asp480 from the V domain forms a hydrogen bond with the side chain of Thr407 from the catalytic domain. The side chain of Gln413 from the catalytic domain forms hydrogen bonds with the side chain of

Figure 4. The V Domain Histidine Patch The cluster of histidine amino acid residues on the V domain may meditate a pH-dependant protein-protein interaction. Twelve of the 14 histidines in the V domain are visible in the figure. The subdomains are colored as in Figure 3.

Glu481 and the backbone carbonyl of Cys526. The remaining direct interactions between these domains are hydrophobic and van der Waals in nature. The interaction surfaces of these two domains are not complementary, with more than 15 ordered water molecules in the cavity between them. Although the interaction between the V domain and the catalytic domain is comparable in both of the crystal forms obtained, it is conceivable that this domain/ domain conformation is a low-energy conformation utilized during crystal packing and is not representative of the protein in solution, or that adopted when performing its physiological function. Examination of the V domain reveals an overall large number of histidine residues, the majority of which cluster on a surface proximal to the groove between subdomains 2 and 3 (Figure 4). With a pKa of 6, histidine residues have been shown to act as switches in pH-dependent protein-protein interactions (Doi et al., 1994; Rudenko et al., 2002). Using Biacore analysis, we have found that wildtype PCSK9 interacts more tightly with the LDLR at pH 5.2 (KD = 35 nM) when compared to pH 7.4 (KD = 840 nM) (Figure S2). Unlike the LDL/LDLR interaction, which dissociates at endosomal pH (5–6) (Brown and Goldstein, 1986), the PCSK9/LDLR interaction is stronger at endosomal pH. Conclusions Many PCSK9 gain-of-function and loss-of-function mutations have been previously described in the literature

Structure 15, 545–552, May 2007 ª2007 Elsevier Ltd All rights reserved 549

Structure The Crystal Structure of PCSK9

(Berge et al., 2006; Kotowski et al., 2006; Zhao et al., 2006), and the structure of PCSK9 helps to elucidate how some of these mutations impart their effect. Loss-of-function mutations DArg97 and Gly106Arg are found in the prodomain of PCSK9 and result in a protein that is defective in autocatalysis and is not secreted from the cell. Arg97 is found within an a helix in the prodomain, and removal of the amino acid from this secondary structure element would disrupt its fold and likely prevent the correct folding of the entire prodomain. The Gly106Arg mutation occurs in a loop between an a helix and b strand, and it is likely that the flexibility allowed by a glycine amino acid residue at this position is required for correct folding of the prodomain. The Leu253Phe loss-of-function mutant also effects the autocatalysis of PCSK9. Residue Leu253 is found between the S2 and S4 pockets of the catalytic site, and it is likely that a larger phenylalanine at this position disrupts the binding of the peptide required for autocatalysis. The Cys679X loss-of-function mutation causes a 14 amino acid truncation at the C terminus of the protein and results in a protein that is retained in the endoplasmic reticulum. Cys679 from b strand 6 forms a disulfide bond with Cys608 from b strand 1 in subdomain 3 of the V domain. Loss of the disulfide bond between strands 1 and 6 likely prevents the V domain from folding correctly, which in turn prevents its secretion from the cell. The specific interactions between PCSK9 and the LDLR remain to be elucidated, and it is likely that knowledge of this interaction will be beneficial in completing the picture of how PCSK9 regulates extracellular levels of the LDLR. However, the structure of PCSK9 does shed much light on its biochemical characteristics and biological function. While the autocatalytic activity of PCSK9 has been shown to be required for the PCSK9 protein to be secreted from cells, the stabilization of the inhibitory peptide in the catalytic site and the tight association of the prodomain to the catalytic domain explain why robust measurements of PCSK9 catalytic activity have not been obtained. Although the structure does not rule out the activation of PCSK9 catalytic activity by additional proteases or under certain conditions, such as the binding of an additional regulatory protein or through displacement of the prodomain by a unique substrate protein, the results seen here agree with the current hypothesis on how PCSK9 lowers LDLR levels (Horton et al., 2007). In this model, PCSK9 autocatalysis triggers its progression through the secretory pathway, and it interacts directly with the LDLR either within this pathway or on the cell surface. The PCSK9/LDLR complex subsequently enters the endosomal pathway. The ability of PCSK9 to interact with the LDLR at endosomal pH prevents the recycling of the LDLR, and the PCSK9/LDLR complex shuttles to the lysozome, where degradation of the LDLR occurs. While we postulate that the histidine-rich patch proximal to the groove on the V domain may be important for mediating the PCSK9/ LDLR interaction, confirmation must await a detailed mapping of surface residue mutations that result in attenuated receptor-binding affinity. A therapeutic that inhibits the activity of PCSK9 anywhere along this pathway could poten-

tially benefit patients suffering from hypercholesterolemia by decreasing their levels of plasma LDL-cholesterol. EXPERIMENTAL PROCEDURES Expression, Purification, and Crystallization of PCSK9 Human PCSK9 (residues 31–692) was cloned into a pFastBac vector (Invitrogen) with an N-terminal His6 affinity tag and TEV protease cleavage site. The honeybee melittin signal peptide was added by PCR. Recombinant baculovirus was prepared with the Bac-to-Bac system (Invitrogen). Trichoplusia ni cells (High Five, Invitrogen) at 1 3 106 cells/ml were infected at a multiplicity of infection (MOI) of 0.01 with recombinant baculovirus. Each liter of culture media was supplemented with 5 ml of Antibiotic-Antimycotic Solution (MediaTech) and 5 mM E-64 protease inhibitor (Sigma). Conditioned media was harvested after 72 hr and filtered through a 0.45 mm filter. The cleared supernatant was concentrated from 10 to 0.6 liters with a Prep/Scale TFF 6.0 ft2 10 kD cartridge according to the manufacturer’s instructions (Millipore). The concentrator retentate was bound to His-Select Nickel Affinity Gel beads (Sigma) and eluted with an imidazole gradient. The PCSK9-enriched sample was further purified by ion-exchange chromatography using a Mono Q 10/100 column (GE Healthcare) with elution by a NaCl gradient. The melittin-His6 tag was then removed by overnight cleavage with rTEV at 4 C. Following cleavage, PCSK9 was purified by size-exclusion chromatography with a Superdex 75 column (GE Healthcare). The final step of the purification again utilized a Mono Q 10/100 column, this time eluting with a pH gradient. Purified PCSK9 was buffer-exchanged into 10 mM Tris (pH 7.5) and 50 mM NaCl and concentrated to 6.5 mg/ml for crystallization. Crystals were grown using the vapor diffusion method by mixing equal volumes of protein and well solution. PCSK9 crystallized in two crystal forms, and in both cases microseeding was used to optimize crystal growth. Crystal form 1 grew from a well solution of 0.1 M NaOAc (pH 4.6), 10% glycerol, and 4.1 M NH4OAc. Crystal form 2 grew from a well solution of 0.1 M Tris (pH 7.8), 0.2 M (NH4)2SO4, and 13% PEG 3350. For both crystal forms, crystals were cryoprotected by supplementing the mother liquor with glycerol and were flash cooled in liquid nitrogen prior to data collection. Data Collection and Structure Determination Diffraction data were collected at 90K at the Berkeley Advanced Light Source beamline 5.0.2, and processed using Denzo and Scalepack (Otwinowski et al., 2003). Crystal form 1 grew in the space group P213 and was used to generate heavy-atom derivatives. Native crystals were derivatized with mercury by soaking overnight in 10 mM ethylmercuryphosphate. For the xenon derivative, native crystals were exposed to xenon gas at 200 psi for 10 min. SOLVE (Terwilliger and Berendzen, 1999) was used to determine the position of one mercury and two xenon atoms by MIRAS. CNX (Brunger et al., 1998) was used to refine the sites and calculate phases. Density modification was used to improve and extend phases to 3.1 A˚, and maps were calculated. Protein secondary structure was easily interpretable in the experimental electron density maps. Multiple rounds of manual model building were performed with Quanta (Accelrys), followed by positional and grouped B factor refinement with CNX. Crystal form 2 grew in the space group P212121 and diffracted to significantly higher resolution. The partial PCSK9 model from crystal form 1 was used as a search model in MOLREP from the CCP4 program suite (CCP4, 1994) to find a molecular replacement solution for crystal form 2. Further model building and refinement were continued against the 2.3 A˚ data set from crystal form 2. Structure improvement proceeded with multiple rounds of manual model building in Quanta, followed by positional and individual B factor refinement using CNX. Simulated annealing omit maps were calculated to confirm the PCSK9 model. Accessible surface area was calculated with AREAINMOL in CCP4. Analysis with the program PROCHECK in CCP4 shows that the model has 90.4% of the residues in the most favored region of the Ramachandran plot, with an

550 Structure 15, 545–552, May 2007 ª2007 Elsevier Ltd All rights reserved

Structure The Crystal Structure of PCSK9

additional 9.4% in allowed regions. Asp651 is in a disallowed region, but the electron density confirms its orientation in the model. The final model has an R of 19.6% and an Rfree of 23.2%, and it includes 87.2% of the PCSK9 amino acid residues, 215 water molecules, and 2 sulfate ions. Figures were made using PyMOL (DeLano, 2002). LDLR/PCSK9 Binding Assay Surface plasmon resonance experiments were performed using a Biacore 2000 (GE Healthcare). LDLR (R&D Systems) was immobilized to a research grade CM5 sensor chip by amine-coupling at levels that gave a maximum analyte binding response (Rmax) of no more than 150 RU. Binding affinities of wild-type human PCSK9 to LDLR were measured at pH 7.4 (HPS-P buffer supplemented with 0.01% BSA and 2 mM CaCl2) and acidic pH 5.2 (Tris-succinate buffer supplemented with 0.01% BSA and 2 mM CaCl2). For the pH 5.2 binding affinity measurement, the LDLR was equilibrated with pH 5.2 buffer before the PCSK9 was flown over the immobilized LDLR. Supplemental Data Supplemental Data include two figures and are available at http:// www.structure.org/cgi/content/full/15/5/545/DC1/. ACKNOWLEDGMENTS We thank M. Ayres for assistance with protein expression, J. Tang and J. Higbee for reagents, and Z. Wang and X. Min for assistance with data collection. The Advanced Light Source at the Lawrence Berkeley National Laboratory is supported by the Director, Office of Science, Office of Basic Sciences, Material Sciences Division, of the U.S. Department of Energy. The authors declare competing financial interests. All authors are employees of Amgen Inc., a company that develops human therapeutics. Received: April 2, 2007 Revised: April 23, 2007 Accepted: April 24, 2007 Published: May 15, 2007

Comellas-Bigler, M., Maskos, K., Huber, R., Oyama, H., Oda, K., and Bode, W. (2004). 1.2 A crystal structure of the serine carboxyl proteinase pro-kumamolisin; structure of an intact pro-subtilase. Structure 12, 1313–1323. Cunningham, D., Danley, D.E., Geoghegan, K.F., Griffor, M.C., Hawkins, J.L., Subashi, T.A., Varghese, A.H., Ammirati, M.J., Culp, J.S., Hoth, L.R., et al. (2007). Structural and biophysical studies of PCSK9 and its mutants linked to familial hypercholesterolemia. Nat. Struct. Mol. Biol. 14, 413–419. DeLano, W.L. (2002). The PyMOL Molecular Graphics System (Palo Alto, CA: Delano Scientific). Doi, T., Kurasawa, M., Higashino, K., Imanishi, T., Mori, T., Naito, M., Takahashi, K., Kawabe, Y., Wada, Y., Matsumoto, A., et al. (1994). The histidine interruption of an a-helical coiled coil allosterically mediates a pH-dependent ligand dissociation from macrophage scavenger receptors. J. Biol. Chem. 269, 25598–25604. Fu, X., Inouye, M., and Shinde, U. (2000). Folding pathway mediated by an intramolecular chaperone. The inhibitory and chaperone functions of the subtilisin propeptide are not obligatorily linked. J. Biol. Chem. 275, 16871–16878. Gallagher, T., Gilliland, G., Wang, L., and Bryan, P. (1995). The prosegment-subtilisin BPN’ complex: crystal structure of a specific ‘foldase’. Structure 3, 907–914. Henrich, S., Cameron, A., Bourenkov, G.P., Kiefersauer, R., Huber, R., Lindberg, I., Bode, W., and Than, M.E. (2003). The crystal structure of the proprotein processing proteinase furin explains its stringent specificity. Nat. Struct. Biol. 10, 520–526. Henrich, S., Lindberg, I., Bode, W., and Than, M.E. (2005). Proprotein convertase models based on the crystal structures of furin and kexin: explanation of their specificity. J. Mol. Biol. 345, 211–227. Horton, J.D., Cohen, J.C., and Hobbs, H.H. (2007). Molecular biology of PCSK9: its role in LDL metabolism. Trends Biochem. Sci. 32, 71–77. Ikemura, H., Takagi, H., and Inouye, M. (1987). Requirement of pro-sequence for the production of active subtilisin E in Escherichia coli. J. Biol. Chem. 262, 7859–7864.

REFERENCES Anderson, E.D., VanSlyke, J.K., Thulin, C.D., Jean, F., and Thomas, G. (1997). Activation of the furin endoprotease is a multiple-step process: requirements for acidification and internal propeptide cleavage. EMBO J. 16, 1508–1518. Benjannet, S., Rhainds, D., Essalmani, R., Mayne, J., Wickham, L., Jin, W., Asselin, M.C., Hamelin, J., Varret, M., Allard, D., et al. (2004). NARC-1/PCSK9 and its natural mutants: zymogen cleavage and effects on the low density lipoprotein (LDL) receptor and LDL cholesterol. J. Biol. Chem. 279, 48865–48875. Berge, K.E., Ose, L., and Leren, T.P. (2006). Missense mutations in the PCSK9 gene are associated with hypocholesterolemia and possibly increased response to statin therapy. Arterioscler. Thromb. Vasc. Biol. 26, 1094–1100. Betzel, C., Pal, G.P., and Saenger, W. (1988). Synchrotron X-ray data collection and restrained least-squares refinement of the crystal structure of proteinase K at 1.5 A resolution. Acta Crystallogr. B 44, 163– 172. Brown, M.S., and Goldstein, J.L. (1986). A receptor-mediated pathway for cholesterol homeostasis. Science 232, 34–47. Brunger, A.T., Adams, P.D., Clore, G.M., DeLano, W.L., Gros, P., Grosse-Kunstleve, R.W., Jiang, J.S., Kuszewski, J., Nilges, M., Pannu, N.S., et al. (1998). Crystallography & NMR system: a new software suite for macromolecular structure determination. Acta Crystallogr. 54, 905–921. CCP4 (Collaborative Computational Project, Number 4) (1994). The CCP4 suite: programs for protein crystallography. Acta Crystallogr. D Biol. Crystallogr. 50, 760–763.

Jain, S.C., Shinde, U., Li, Y., Inouye, M., and Berman, H.M. (1998). The crystal structure of an autoprocessed Ser221Cys-subtilisin E-propeptide complex at 2.0 A resolution. J. Mol. Biol. 284, 137–144. Kim, J.S., Kluskens, L.D., de Vos, W.M., Huber, R., and van der Oost, J. (2004). Crystal structure of fervidolysin from Fervidobacterium pennivorans, a keratinolytic enzyme related to subtilisin. J. Mol. Biol. 335, 787–797. Kotowski, I.K., Pertsemlidis, A., Luke, A., Cooper, R.S., Vega, G.L., Cohen, J.C., and Hobbs, H.H. (2006). A spectrum of PCSK9 alleles contributes to plasma levels of low-density lipoprotein cholesterol. Am. J. Hum. Genet. 78, 410–422. Krissinel, E., and Henrick, K. (2004). Secondary-structure matching (SSM), a new tool for fast protein structure alignment in three dimensions. Acta Crystallogr. 60, 2256–2268. Lagace, T.A., Curtis, D.E., Garuti, R., McNutt, M.C., Park, S.W., Prather, H.B., Anderson, N.N., Ho, Y.K., Hammer, R.E., and Horton, J.D. (2006). Secreted PCSK9 decreases the number of LDL receptors in hepatocytes and in livers of parabiotic mice. J. Clin. Invest. 116, 2995–3005. Maxwell, K.N., Fisher, E.A., and Breslow, J.L. (2005). Overexpression of PCSK9 accelerates the degradation of the LDLR in a post-endoplasmic reticulum compartment. Proc. Natl. Acad. Sci. USA 102, 2069– 2074. Naureckiene, S., Ma, L., Sreekumar, K., Purandare, U., Lo, C.F., Huang, Y., Chiang, L.W., Grenier, J.M., Ozenberger, B.A., Jacobsen, J.S., et al. (2003). Functional characterization of Narc 1, a novel proteinase related to proteinase K. Arch. Biochem. Biophys. 420, 55–67.

Structure 15, 545–552, May 2007 ª2007 Elsevier Ltd All rights reserved 551

Structure The Crystal Structure of PCSK9

Otwinowski, Z., Borek, D., Majewski, W., and Minor, W. (2003). Multiparametric scaling of diffraction intensities. Acta Crystallogr. A 59, 228–234.

et al. (1999). Mammalian subtilisin/kexin isozyme SKI-1: a widely expressed proprotein convertase with a unique cleavage specificity and cellular localization. Proc. Natl. Acad. Sci. USA 96, 1321–1326.

Park, S.W., Moon, Y.A., and Horton, J.D. (2004). Post-transcriptional regulation of low density lipoprotein receptor protein by proprotein convertase subtilisin/kexin type 9a in mouse liver. J. Biol. Chem. 279, 50630–50638.

Seidah, N.G., Benjannet, S., Wickham, L., Marcinkiewicz, J., Jasmin, S.B., Stifani, S., Basak, A., Prat, A., and Chretien, M. (2003). The secretory proprotein convertase neural apoptosis-regulated convertase 1 (NARC-1): liver regeneration and neuronal differentiation. Proc. Natl. Acad. Sci. USA 100, 928–933.

Rashid, S., Curtis, D.E., Garuti, R., Anderson, N.N., Bashmakov, Y., Ho, Y.K., Hammer, R.E., Moon, Y.A., and Horton, J.D. (2005). Decreased plasma cholesterol and hypersensitivity to statins in mice lacking Pcsk9. Proc. Natl. Acad. Sci. USA 102, 5374–5379.

Siezen, R.J., and Leunissen, J.A. (1997). Subtilases: the superfamily of subtilisin-like serine proteases. Protein Sci. 6, 501–523.

Rawlings, N.D., Morton, F.R., and Barrett, A.J. (2006). MEROPS: the peptidase database. Nucleic Acids Res. 34, D270–D272.

Tangrea, M.A., Bryan, P.N., Sari, N., and Orban, J. (2002). Solution structure of the pro-hormone convertase 1 pro-domain from Mus musculus. J. Mol. Biol. 320, 801–812.

Rudenko, G., Henry, L., Henderson, K., Ichtchenko, K., Brown, M.S., Goldstein, J.L., and Deisenhofer, J. (2002). Structure of the LDL receptor extracellular domain at endosomal pH. Science 298, 2353–2358.

Terwilliger, T.C., and Berendzen, J. (1999). Automated MAD and MIR structure solution. Acta Crystallogr. 55, 849–861.

Sakai, J., Rawson, R.B., Espenshade, P.J., Cheng, D., Seegmiller, A.C., Goldstein, J.L., and Brown, M.S. (1998). Molecular identification of the sterol-regulated luminal protease that cleaves SREBPs and controls lipid composition of animal cells. Mol. Cell 2, 505–514.

Zhao, Z., Tuakli-Wosornu, Y., Lagace, T.A., Kinch, L., Grishin, N.V., Horton, J.D., Cohen, J.C., and Hobbs, H.H. (2006). Molecular characterization of loss-of-function mutations in PCSK9 and identification of a compound heterozygote. Am. J. Hum. Genet. 79, 514–523.

Sanchez, J.F., Lescar, J., Chazalet, V., Audfray, A., Gagnon, J., Alvarez, R., Breton, C., Imberty, A., and Mitchell, E.P. (2006). Biochemical and structural analysis of Helix pomatia agglutinin. A hexameric lectin with a novel fold. J. Biol. Chem. 281, 20171–20180. Seidah, N.G., and Prat, A. (2007). The proprotein convertases are potential targets in the treatment of dyslipidemia. J. Mol. Med., in press. Published online March 10, 2007. 10.1007/s00109-007-0172-7. Seidah, N.G., Mowla, S.J., Hamelin, J., Mamarbachi, A.M., Benjannet, S., Toure, B.B., Basak, A., Munzer, J.S., Marcinkiewicz, J., Zhong, M.,

Accession Numbers Atomic coordinates have been deposited in the Protein Data Bank under accession code 2PMW.

Note Added in Proof After this manuscript was submitted, Cunningham et al. (2007) published a paper describing the same structure.

552 Structure 15, 545–552, May 2007 ª2007 Elsevier Ltd All rights reserved