Identification of a Small Peptide That Inhibits PCSK9 Protein Binding to the Low Density Lipoprotein Receptor

Background: Therapeutic inhibition of circulating PCSK9 reduces LDL-c levels. Results: A synthetic PCSK9-binding peptide, which restores cellular LDL receptors, was identified. Conclusion: Pep2-8 is the smallest PCSK9 inhibitor with a defined inhibitory mechanism described to date and structurally mimics the EGF(A) domain of the receptor. Significance: This work demonstrates the feasibility of developing a peptide-based inhibitor of PCSK9. PCSK9 (proprotein convertase subtilisin/kexin type 9) is a negative regulator of the hepatic LDL receptor, and clinical studies with PCSK9-inhibiting antibodies have demonstrated strong LDL-c-lowering effects. Here we screened phage-displayed peptide libraries and identified the 13-amino acid linear peptide Pep2-8 as the smallest PCSK9 inhibitor with a clearly defined mechanism of inhibition that has been described. Pep2-8 bound to PCSK9 with a KD of 0.7 μm but did not bind to other proprotein convertases. It fully restored LDL receptor surface levels and LDL particle uptake in PCSK9-treated HepG2 cells. The crystal structure of Pep2-8 bound to C-terminally truncated PCSK9 at 1.85 Å resolution showed that the peptide adopted a strand-turn-helix conformation, which is remarkably similar to its solution structure determined by NMR. Consistent with the functional binding site identified by an Ala scan of PCSK9, the structural Pep2-8 contact region of about 400 Å2 largely overlapped with that contacted by the EGF(A) domain of the LDL receptor, suggesting a competitive inhibition mechanism. Consistent with this, Pep2-8 inhibited LDL receptor and EGF(A) domain binding to PCSK9 with IC50 values of 0.8 and 0.4 μm, respectively. Remarkably, Pep2-8 mimicked secondary structural elements of the EGF(A) domain that interact with PCSK9, notably the β-strand and a discontinuous short α-helix, and it engaged in the same β-sheet hydrogen bonds as EGF(A) does. Although Pep2-8 itself may not be amenable to therapeutic applications, this study demonstrates the feasibility of developing peptidic inhibitors to functionally relevant sites on PCSK9.

PCSK9 (proprotein convertase subtilisin/kexin type 9), initially identified as NARC-1 (1), regulates LDL-c 2 levels by promoting degradation of the liver LDL receptor, which results in reduced LDL receptor-dependent clearance of LDL-c particles and increased plasma LDL-c levels (2)(3)(4). PCSK9 acts by binding to the EGF(A) domain of LDL receptor on the cell surface via its catalytic domain (5). The binding affinity increases in the acidic endosomal compartment through additional interactions of LDL receptor with the C-terminal Cys-rich domain (CRD) of PCSK9 (6 -8). As a result, LDL receptor recycling to the cell surface is prevented, and instead, the PCSK9⅐LDL receptor complex is directed to lysosomes for degradation. This "sorting" of the PCSK9⅐LDL receptor complex seems to be mediated by APLP2, which interacts with PCSK9 at low pH in the endosomal compartment (9).
Natural PCSK9 mutations are associated with high and low cholesterol levels (10,11). Cohen et al. (11) found that carriers of two loss-of-function mutations had lower LDL-c levels and a dramatically reduced risk for coronary heart disease. A compound heterozygous individual with no detectable PCSK9 and extremely low LDL-c levels is healthy (12), suggesting that pharmacologic PCSK9 inhibition for lipid lowering might be safe.
PCSK9 is an ideal antibody target because it circulates in blood and has only modest binding affinity to cell surface LDL receptor and its main interaction site is an exposed ϳ500-Å 2 slightly convex region that is readily available for antibody binding. Recent clinical studies with anti-PCSK9 antibodies have borne this out, demonstrating strong and sustained LDL-c lowering after a single dose injection (13). Whereas the clinical development of antibody-based inhibitors has advanced rapidly, the development of small molecule inhibitors appears still to be at an early stage, probably due to the challenging nature of PCSK9 as a small molecule target. As its name implies, PCSK9 bears recognizable sequence homology to proteolytic enzymes of the subtilisin family. The molecular structure of PCSK9 determined using x-ray crystallography confirms its close conformational relationship with subtilisin-like proteases (14 -16). Unlike typical enzymes, whose active sites are essential for function and generally offer good binding sites for small molecules, the PCSK9 active site is non-essential for LDL receptor down-regulation (17,18) and remains in an inactive state due to occupancy by the prodomain C terminus after autoprocessing in the endoplasmic reticulum. Therefore, aside from an approach that would block maturation by inhibiting the autocatalytic enzyme activity during biosynthesis, the PCSK9 active site is not attractive as a traditional protease target.
The ability of PCSK9 to mediate LDL receptor degradation involves protein-protein interactions between LDL receptor and the PCSK9 prodomain (19), the catalytic domain (20), and the CRD (6 -9, 21-23). Protein-protein interactions are inherently challenging small molecule targets because they frequently contain large, relatively flat contact surfaces devoid of the pockets required for efficient binding of small organic chemicals. However, protein-protein interactions have been successfully inhibited by peptides, which can recapitulate key protein contacts (24). To identify a peptidic PCSK9 inhibitor, we screened both linear and disulfide-constrained phage-displayed peptide libraries with final diversities of 1.8 ϫ 10 11 and 7.8 ϫ 10 11 , respectively, and identified the 13-amino acid peptide Pep2-8. Structural and biochemical characterization provided the molecular details of its inhibitory mechanism, which includes structural mimicry of the natural binding domain of LDL receptor.

EXPERIMENTAL PROCEDURES
Construction of Phage-displayed Peptide Libraries-Two groups of phage-displayed peptide libraries, the linear peptide library called Linear-lib and cysteine-restrained cyclic library called Cyclic-lib, were constructed by fusing randomized peptides to the N terminus of M13 major coat protein p8 (25). Linear-lib consisted of random peptides with 8, 10, 12, 14, or 16 amino acids, and Cyclic-lib consisted of 14-mer random peptides with varied length between two invariant cysteines. The final diversities for Linear-lib and Cyclic-Lib were 1.8 ϫ 10 11 and 7.8 ϫ 10 11 , respectively.
The soft randomized library was constructed using degenerate oligonucleotides synthesized with 70:10:10:10 mixtures of nucleotide bases, in which the wild type base was in excess. This results in the wild type amino acids occurring at ϳ50% frequency at the targeted position.
Selection of Peptide Ligands for PCSK9 and Affinity Maturation of Pep2-Phage pools of Linear-lib and Cyclic-lib library were cycled through rounds of binding selections using 130, 65, 26, and 26 nM biotinylated PCSK9 in a buffer of PBS, 0.5% BSA, and 0.1% Tween 20 for round 1, 2, 3 and 4, respectively (26). After four rounds of binding selection, individual phage clones were analyzed in a high throughput spot phage ELISA using plate-immobilized PCSK9 as the target (27). Nonspecific binding was defined as phage particle binding to NeutrAvidin. Clones with phage binding signal to PCSK9 over 0.5 and signal/ noise ratio of Ͼ5 were considered to be positive clones and were subjected to DNA sequence analysis.
The soft randomized library was used for affinity maturation of clone 2. Phage pools were subjected to four rounds of binding selection with decreasing concentrations of biotinylated PCSK9 from round 1 to round 4 at 130, 20, 2, and 2 nM, respectively, and produced Pep2-2.
Peptide Binding Affinity for PCSK9 Measured by a Phage Competition ELISA-The sequence of Pep2-2 was fused to the N terminus of M13 major coat protein (p8) using Kunkel mutagenesis (28). The resulting construct was transformed to Escherichia coli XL1 blue, and single colonies were grown in 1 ml of 2YT supplemented with 50 g/ml carbenicillin, 10 g/ml tetracycline, and M13 KO7 helper phage at 37°C for 2 h. After the addition of kanamycin (25 g/ml) and a 6-h incubation at 37°C, the culture was transferred to 30 ml of 2YT supplemented with 50 g/ml carbenicillin and 25 g/ml kanamycin and grown at 37°C overnight. Phage were harvested and purified using the standard protocol (27). The serially diluted phage solution was added to a 384-well MaxiSorp ImmunoPlate containing biotinylated PCSK9 immobilized to NeutrAvidincoated wells. The plate was incubated at 4°C for 1 h and extensively washed with PBS, 0.05% Tween 20, and bound phage was quantified after the addition of anti-M13-HRP and HRP substrate on a microplate reader at A 450 . After subtracting background binding (i.e. phage binding to NeutrAvidin-coated plates), the data were fitted to a binding curve using Kaleida-Graph, and the subsaturating phage concentration that gave 80% of maximum binding was calculated.
For the peptide affinity determinations, biotinylated PCSK9 was immobilized to NeutrAvidin-coated MaxiSorp Immuno-Plates and incubated with a mixture of serially diluted synthetic peptide (Pep2-2 or Pep2 at 0 -100 M) and the corresponding peptide-displaying phage at the earlier determined subsaturating concentration (see above). After a 1-h incubation in PBT buffer (PBS, 0.5% BSA, 0.05% Tween 20), the plate was washed extensively, and the bound phage was detected by anti-M13-HRP. The data were fitted to a four-parameter equation (KaleidaGraph) to determine the IC 50 values. For IC 50 measurements of the Pep2-2 derivatives (Fig. 1C) and for the truncated Pep2-8 peptides (Table 1), we used the peptides in a concentration range of 0 -250 M and the Pep2-2-displaying phage at subsaturating concentrations. The results were the average Ϯ S.D. of three independent experiments.
Construction, Expression, and Purification of Wild Type and Mutant PCSK9 Proteins-Human PCSK9 cDNA containing a His 8 C-terminal tag was cloned into the mammalian expression vector pRK5. Using this expression construct as a template, the PCSK9 mutants S153A, P155A, L158A, R194A, D238A, D367A, I369A, D374A, F379A, V380A, S381A, and Q382A were constructed by ACTG Inc. (Wheeling, IL) using site-directed mutagenesis. Mutations were confirmed by DNA sequencing and, in the case of S153A, P155A, and L158A, by additional N-terminal amino acid sequencing of the purified proteins. The recombinant PCSK9 proteins (wild type and mutants) were transiently expressed in Chinese hamster ovary (CHO) cells and purified from conditioned media by affinity chromatography using a nickel-nitrilotriacetic acid-agarose column (Qiagen, Germantown, MD) followed by gel filtration on a Sephacryl S 200 column (GE Healthcare) equilibrated with 50 mM Tris, pH 8.0, 500 mM NaCl, 10% glycerol. Proteins were aliquoted and stored at Ϫ80 ºC.
Production of PCSK9 Lacking the CRD (PCSK9-⌬CRD)-PCSK9-⌬CRD (Gln 31 -Gly 452 ) was cloned by PCR amplification from the full-length PCSK9 cDNA using GC Advantage Polymerase (Clontech). The PCR product contained the sequence coding for PCSK9-⌬CRD and the restriction endonuclease sites BamHI and EcoRI at the 5Ј-and 3Ј-end, respec-tively. The PCR product was ligated into a modified pAcGP67 baculovirus transfer vector containing the DNA sequence coding for a hexahistidine affinity tag. The inserted PCR product was verified by DNA sequencing. Protein was produced in Trichoplusia ni insect cells (Expression Systems LLC, Woodland, CA) grown for 2 days after infection with baculovirus. Cells were removed from the production medium by centrifugation, and the supernatant was loaded onto a nickel-nitrilotriacetic acid resin (Qiagen) equilibrated with 50 mM Tris, pH 8.0, 500 mM NaCl and eluted with 50 mM Tris, pH 8.0, 500 mM NaCl, 250 mM imidazole. Pooled fractions were further purified on an S75 size exclusion column (GE Healthcare) equilibrated with 40 mM Tris, pH 8.0, 200 mM NaCl, 5% glycerol, 0.1 mM CaCl 2 . By use of a spin concentrator, the pooled fractions were concentrated to 8 mg in 0.9 ml. The fusion proteins were transiently expressed in CHO and purified on a Protein A resin followed by size exclusion chromatography. The identities of the proteins were confirmed by mass spectrometry, N-terminal sequencing, and SDS-PAGE. Unexpectedly, the purified LDL receptor-Fc lacked the first four LDL receptor repeats due to cleavage by an unknown protease at Gly 192 -Asp 193 in the linker region between LDL receptor repeats 4 and 5. The binding affinity of PCSK9 to this LDL receptor-Fc construct determined by biolayer interferometry at neutral pH was 106.7 Ϯ 5.8 nM (average Ϯ S.D. of three experiments), suggesting that the N-terminal truncation of LDL receptor did not impair PCSK9 binding.
Binding Affinity and Specificity Determinations by Biolayer Interferometry-The binding affinities of Pep2-8_Fc and Pep-ctrl_Fc to PCSK9 were measured by biolayer interferometry on an OctetRed 384 (ForteBio, Menlo Park, CA). Anti-hIgG-Fc capture biosensors (ForteBio) were loaded with Pep2-8_Fc or Pep-ctrl_Fc in 50 mM Tris, pH 7.5, 300 mM NaCl, 1 mM CaCl 2 , 1 mg/ml BSA, 0.05% Tween 20, washed in the same buffer, and transferred to wells containing PCSK9 at concentrations ranging from 8 to 6000 nM in the same buffer. The k on and k off values were obtained by fitting the association and dissociation data to a 1:1 model algorithm using Octet software. The K D derived from kinetic fitting was calculated as k off /k on.
Peptide Synthesis-Synthetic peptides were prepared on an automated Protein Technologies Inc. synthesizer. Peptides were assembled using standard Fmoc chemistry protocols on an Fmoc Rink amide linker attached to aminomethyl resin, and peptides were acetylated on the N terminus. Upon completion of the linear chains, peptides were cleaved off the solid support with trifluoroacetic acid/triisopropylsilane/water (95:2.5:2.5) for 1 h at room temperature. Trifluoroacetic acid was evaporated, and peptides were precipitated with ethyl ether; extracted with acetic acid, acetonitrile, and water; and lyophilized. Crude peptides were solubilized in dimethyl sulfoxide and purified by reverse phase chromatography on a C18 column using acetonitrile/water buffers. Purified fractions were analyzed by liquid chromatography mass spectrometry (PE-Sciex), pooled, and lyophilized.
HepG2 Assay for Measuring Cell Surface LDL Receptor and LDL Uptake-The effects of Pep2-8 and Pep-ctrl on PCSK9mediated LDL receptor degradation were measured in a HepG2 cell assay essentially as described (26). Peptides at various concentrations were preincubated with 15 g/ml PCSK9 in the presence of 0.5% DMSO for 30 min prior to the addition to cells. After a 4-h incubation, the cells were processed for measurements of LDL receptor surface levels by flow cytometry as described (26).
In the LDL uptake assay, HepG2 cells were seeded at a density of 3 ϫ 10 4 cells/well in a 96-well plate. After 24 h, the medium was changed to 10% lipoprotein-deficient serum (Intracel) in DMEM (Invitrogen). After 24 h, Pep2-8 or Pep-ctrl at various concentrations was preincubated with 15 g/ml PCSK9 in 0.5% DMSO for 30 min prior to the addition to the cells. After 1.5 h, 10 g/ml BODIPY-LDL (Invitrogen) was added and incubated for 3.5 h. Cells were washed three times with PBS, and fluorescence was measured on a Wallac VICTOR3 plate reader (PerkinElmer Life Sciences). For both assays, the results were the average Ϯ S.D. of three independent experiments.
Formation and Crystallization of the Complex-A 10-fold molar excess of Pep2-8 (0.5 mg) was added to 200 l of PCSK9-⌬CRD (8.9 mg/ml) and incubated for 14 h at 4°C. Crystallization trials were performed using commercially available screening reagents in 0.4-l sitting drops made from protein and reservoir in a 1:1 ratio. A drop made using reservoir containing 0.2 M LiCl and 20% PEG 6000 in 0.1 M Tris, pH 8.0, produced crystals that were optimized by growth in larger drops. A crystal was harvested and passed briefly through a solution of reservoir augmented with 30% glycerol and preserved for data collection by immersion in liquid nitrogen.
Structure Determination of PCSK9-⌬CRD⅐Pep2-8 by X-ray Crystallography-Diffraction data extending to 1.85 Å resolution were collected at synchrotron beamline SSRL 9-2 and integrated and scaled (29,30) in space group C2. Approximate phases were obtained by the method of molecular replacement (31), using a truncated version of a previously reported structure of PCSK9 (15) (Protein Data Bank code 2QTW). Electron density maps and intermediate molecular models were manipulated using Coot (32), and refinement was performed using Phenix.refine (33) and Refmac5 (32).
Structure Determination of Pep2-8 in Solution Using NMR-Nuclear magnetic resonance experiments were carried out on a Bruker DRX-600 spectrometer equipped with a triple-resonance cryoprobe. Samples for NMR studies were prepared by dissolving Pep2-8 in PBS (5% D 2 O) to a final concentration of 1.2 mM. The NMR spectra were collected at 298 K and internally referenced to 4,4-dimethyl-4-silapentane-1-sulfonic acid. Complete assignment of the proton resonances was achieved using a combination of two-dimensional total correlation spectroscopy, two-dimensional double-quantum-filtered COSY, and two-dimensional NOESY experiments. The mixing times for the total correlation spectroscopy and NOESY spectra were 98 and 150 ms, respectively. For the two-dimensional experiments, 2,048 and 512 data points were used in F2 and F1, respectively. NMR data were processed using the NMRPipe/NMRDraw package (34) and analyzed with NMRView (35). The interproton distances were estimated from the intensities of the cross-peaks observed in the NOESY experiment and classified as strong (1.8 -2.8 Å), medium (1.8 -3.4 Å), and weak (1.8 -5.0 Å).
Structure calculations were performed with the program CNS, starting from an extended structure and using a combination of torsion angle dynamics and cartesian dynamics. A total of 127 NOE-derived interproton distances were used as experimental restraints to calculate a set of 30 structures, all of them satisfying the experimental constraints, with no NOE violations greater than 0.2 Å. From this set, the 20 structures with the lowest energies were selected for further analysis. The quality of the structures was assessed using Procheck-NMR (36). The figures representing the structures were generated using the programs MOLMOL and PyMOL.

Selection of PCSK9-binding Peptide by Peptide Phage Display-
Two types of phage-displayed naive peptide libraries, Linear-lib and Cyclic-lib (25), were used to select for PCSK9 binding peptides by solution sorting against biotinylated PCSK9 in calciumfree buffer (26). After four rounds of panning, several clones gave modest binding signals. The peptides derived from these clones were synthesized and tested in the HepG2 LDL receptor degradation assay. We found that Pep2 (VVFTSWQDYLFWV) derived from clone 2 was able to inhibit PCSK9-mediated LDL receptor degradation with low potency (data not shown). A competition phage ELISA showed that Pep2 was a relatively weak PCSK9 binder (IC 50 ϭ 37.7 Ϯ 1.0 M) (Fig. 1A). To improve binding affinity, a soft-randomized library based on the Pep2 parent sequence was constructed. In this experiment, each position retained the parent amino acids at a 50% frequency. After four rounds of solution sorting, 80 phage clones showed improved binding to PCSK9 compared with the parent clone as measured by phage ELISA (data not shown). The sequence alignment, summarized in the sequence logo (Fig. 1B) of the 80 peptides, showed that Val 2 , Phe 3 , Ser 5 , Trp 6 , Tyr 9 , and Leu 10 were highly conserved, suggesting that these residues are important for PCSK9 binding. Of note, many of the clones with high ELISA signals also replaced the parent residues Gln 7 and Asp 8 with a glutamate residue, indicating that the glutamate preference may account for the affinity improvements. Therefore, for further experiments, we chose Pep2-2, which gave the strongest ELISA signal and had glutamate residues in positions 7 and 8 ( Table 1). The synthesized Pep2-2 bound to PCSK9 with 50-fold greater affinity than the parent peptide Pep2 in a phage competition ELISA (IC 50 ϭ 0.76 Ϯ 0.01 M) (Fig. 1A).
Structure-Activity Relationship of Pep2-2 and Design of Its Derivative Pep2-8-A panel of Pep2-2-based peptides in which each residue was individually replaced by alanine was synthesized and tested for PCSK9 binding in a competition phage ELISA (Fig. 1C). Alanine replacements of the three C-terminal residues Leu 11 , Trp 12 , and Val 13 did not affect binding. In contrast, a dramatic binding loss (Ͼ1,000-fold) was observed for alanine substitutions of Phe 3 and Trp 6 , suggesting that these side chains are necessary either for maintaining peptide structure or providing critical PCSK9 contacts or both. Strong effects (Ͼ20-fold loss) were also found for substitutions at Ser 5 , Glu 8 , and Tyr 9 .
The peptide alanine scanning results suggested that the three C-terminal residues may not be critical for the binding interac-tion. Therefore, we measured binding of C-terminally truncated versions of Pep2-2. Surprisingly, the sequential deletion of the four C-terminal residues resulted in decreasing binding levels, and binding became undetectable when all four residues were deleted ( Table 1). Deletion of the N-terminal Thr 1 resulted in a 6-fold binding loss, consistent with the reduced binding in the alanine scanning experiment. Therefore, Pep2-2 cannot be shortened without compromising binding affinity, indicating that all residues are necessary for optimal interaction with PCSK9.
The low solubility of Pep2-2 limited its further characterization in cell-based systems. Therefore, we changed Leu 11 to Asp 11 , because the sequence logo indicated that aspartate is relatively preferred at this position (Fig. 1B). This new peptide, designated Pep2-8 (TVFTSWEEYLDWV), was synthesized and showed better solubility than Pep2-2 while maintaining a binding affinity that was comparable with that of Pep2-2 (Table  1). Pep2-8 was used for all subsequent studies.
Binding Affinity and Specificity of Pep2-8-For binding affinity measurements, we constructed a Pep2-8 Fc fusion protein (Pep2-8_Fc) that could be immobilized on Octet anti-Fccoated sensor tips for biolayer interferometry experiments. As a control, we made a peptide Fc fusion protein (Pep-ctrl_Fc) based on a Pep2-2 double mutant having the two important binding residues Phe 3 and Trp 6 replaced by alanine (Table 1). PCSK9 bound to Pep2-8_Fc with a K D value of 0.66 Ϯ 0.11 M (Fig. 2A), whereas the control Pep-ctrl_Fc did not give any binding signal up to 6 M (data not shown). Furthermore, the same system was used to assess the specificity of Pep2-8 by measuring the binding to the related proprotein convertases furin, PC1, PC2, PC4, and PC7. Whereas PCSK9 gave a strong and specific binding response, the tested proprotein convertases showed only small and nonspecific binding signals (Fig. 2B), indicating that Pep2-8 specifically recognizes PCSK9.

Pep2-8 Restores LDL Receptor Function in HepG2
Cells-The addition of PCSK9 to HepG2 cells reduced surface LDL receptor levels by about 75%, as determined by FACS. Co-incubation with Pep2-8 restored LDL receptor surface levels in a concentrationdependent manner (Fig. 3A). The control peptide showed no effect. Moreover, Pep2-8 restored LDL uptake of PCSK9treated HepG2 cells to about 90% of control activity at a concentration of 50 M, whereas the control peptide showed no effect (Fig. 3B). The half-maximal effect (EC 50 ) for LDL uptake was 6 -12.5 M, which was similar to the EC 50 value of ϳ12.5 M for surface LDL receptor restoration. The results indicated that Pep2-8 was able to neutralize PCSK9 activity, resulting in the functional recovery of cellular LDL receptor.
Pep2-8 Interferes with LDL Receptor Binding to PCSK9-PCSK9-mediated LDL receptor degradation involves all PCSK9 domains (5,19,20,23). To understand how Pep2-8 interfered with LDL receptor degradation, we sought to identify the peptide-binding site on PCSK9. We found that Pep2-8 bound to PCSK9 lacking the CRD (PCSK9-⌬CRD), suggesting that the CRD does not harbor the peptide-binding site (data not shown). Furthermore, in a competition binding ELISA, Pep2-8 inhibited binding of PCSK9 to LDL receptor-Fc with an IC 50 value of 0.81 Ϯ 0.08 M, whereas Pep-ctrl had no effect (Fig. 4A). A similar assay was used to examine whether Pep2-8 interfered with PCSK9 binding to the EGF(A) domain of LDL receptor. Pep2-8 completely inhibited this interaction with an IC 50 value of 0.44 Ϯ 0.04 M, whereas Pep-ctrl showed no inhibition (Fig.  4B). The Pep2-8 binding site was further elucidated by measuring binding of Pep2-8 to a panel of PCSK9 mutants with alanine changes within or close to the known EGF(A) binding region on the PCSK9 catalytic domain, including Phe 379 , which is at the center of the EGF(A) binding site (20,37). Analysis of each purified mutant by SDS-PAGE and by N-terminal sequencing indicated that the introduced mutations did not  affect pro-PCSK9 processing at Gln 152 -Ser 153 during biosynthesis, yielding mature PCSK9 proteins (data not shown). The binding ELISA showed that many of the mutant proteins, including F379A, had Ͼ50% reduced binding to PCSK9 (Fig.  4C), suggesting that the Pep2-8 binding site overlaps with that of EGF(A).
Crystal Structure of the PCSK9-⌬CRD⅐Pep2-8 Complex and Solution Structure of Pep2-8 by NMR-In order to obtain a more detailed insight into the binding site and the inhibitory mechanism of Pep2-8, we determined the crystal structure of Pep2-8 bound to PCSK9-⌬CRD at 1.85 Å resolution (Table 2). Despite numerous close intermolecular contacts characterizing our crystals (Fig. 5), the overall similarity to previously reported PCSK9 structures (14,37) is very high, with r.m.s. deviations from superposed C␣ atoms, ignoring a few outliers, of less than 0.5 Å. The entire Pep2-8 is resolved from residue acetyl-Thr 1 to amidyl-Val 13 and interacts with PCSK9 residues that are in the catalytic domain and that are also directly contacted by the EGF(A) domain of LDL receptor (20) (Fig. 6, A and  B). The surface of PCSK9 made solvent-inaccessible by Pep2-8 is smaller than that by EGF(A) (420 versus 520 Å 2 ), but there is a strong overlap in the Pep2-8 and EGF(A) binding surfaces on PCSK9 (Fig. 6, C and D). Among the PCSK9 residues central to both interfaces are Ile 369 and Phe 379 , which also showed effects on Pep2-8 binding in the alanine changes (Fig. 4C). Pep2-8 residues Thr 1 -Ser 5 adopt an extended main chain conformation, with Phe 3 main chain torsion angles appropriate for a ␤-strand. Starting with the carbonyl of Ser 5 and until the end at Val 13 , the peptide is an ␣-helix (Fig. 6C). Residues Glu 7 , Glu 8 , and Asp 11 are the only peptide residues with a formal charge at neutral pH, and thus an overall negative electrostatic potential is concentrated on one side of the helix. The opposite side of the helix contains the side chains from residues Trp 6 , Tyr 9 , Leu 10 , and Trp 12 . In addition, the side chain of Phe 3 has a small van der Waals contact with the side chains of Tyr 9 and Trp 12 , and the Val 2 side chain is adjacent, so these residues thereby contribute to the hydrophobic surface on this side of the peptide and form the major interaction site with PCSK9 as described in more detail below.
The chemical shift dispersion of the NMR resonances and the presence of a substantial number of interresidue NOE contacts indicated that Pep2-8 does adopt a preferred fold when free in solution ( Table 3). The solution conformation of Pep2-8 determined from the NMR data shows a remarkable correspondence to the Pep2-8 structure bound to PCSK9 (Fig. 7, A  and B). The lowest energy solution structure determined using NMR, after superposition with the peptide from the x-ray structure using 13 C␣ atoms, has an r.m.s. deviation of only 0.78 Å. The largest deviations between pairs of C␣ atoms occur for Thr 1 (1.3 Å) and Glu 8 (1.2 Å). Superpositioning of only the ␣-helical residues yields an r.m.s. deviation of 0.31 Å, and superpositioning of residues 1-5 yields an r.m.s. deviation of only 0.27 Å. Most of the main chain torsion angles (,) in the  x-ray peptide coordinates are very close to those in the NMR coordinates, the two outliers being Thr 4 (⌬ ϭ 36°) and Trp 6 (⌬ ϭ 32°). Thus, the conformation of Pep2-8 when free or when bound to PCSK9 is a short extended N-terminal segment followed by an ␣-helix. Similar strand-loop-helix conformations have been described previously for small phage-derived peptides (38 -40), although these examples were stabilized by a disulfide bond to tether the N terminus to the center of the helix. Pep2-8 lacks such a covalent stabilizing element, and in lieu of a disulfide bond, van der Waals contacts from Phe 3 to Tyr 9 and Trp 12 serve to tether the N-terminal region and probably stabilize the helix.
Pep2-8 Interaction with PCSK9 and Its Structural Mimicry of the EGF(A) Domain-The hydrophobic face of Pep2-8 contacts PCSK9, being within 4 Å of residues in PCSK9 ␤-strands ␤13 (Ile 368 -Ala 371 ) and ␤14 (Phe 379 -Gln 382 ) and the loop between them, the loop between ␣4 (Ser 225 -Ser 235 ) and ␤7 (Ser 246 -Arg 251 ), and Pro 155 near the start of ␣5 (Fig. 8). The Pep2-8 Trp 6 side chain is within 4 Å of carbon atoms of Asp 238 , Ala 239 , and    (Table 3)). The lowest energy NMR structure is in red, and the next 19 lowest energy models from NMR are shown in black. B, the Pep2-8 structure from the Pep2-8⅐PCSK9-⌬CRD x-ray structure (green) is essentially the same as the lowest energy solution structure from NMR (red).
Phe 379 . The Pep2-8 Tyr 9 side chain is within 4 Å of Ile 369 . Of the remaining peptide side chains comprising the hydrophobic side of the peptide, Thr 1 is 3.7 Å from the C␤ atom of Cys 378 , and Phe 3 is 4.1 Å from Ile 369 , but Trp 12 and Val 13 do not contact PCSK9. There are only three hydrogen bonds directly between Pep2-8 and PCSK9; two reciprocal ␤-sheet hydrogen bonds between Pep2-8 Phe 3 and PCSK9 Phe 379 , and one between the main chain oxygen atom of Pep2-8 Thr 4 and the side chain of PCSK9 Thr 377 . There is a single instance of one water molecule bridging Pep2-8 and PCSK9 (Thr 1 and Ser 381 , respectively) and also several multiwater bridges (e.g. that involving the Pep2-8 Tyr 9 side chain hydroxyl). The binding site identified by a PCSK9 alanine scan is imperfectly consistent with the structural contact region seen in the x-ray structure. For instance, the two largest hits, Asp 367 and Gln 382 , are not contact residues. Rather they are at the start and the end, respectively, of ␤-stands ␤13 and ␤14, which together with the loop between them provide several contact residues to Pep2-8. However, Pro 155 is a contact residue for both Trp 6 and Tyr 9 , and it seems likely that the effects of alanine substitutions for Ser 153 or Leu 158 are best understood in terms of an indirect effect on Pro 155 , inasmuch as the native side chains stabilize and/or orient the catalytic domain N terminus, which is also important for LDL receptor binding (20,37).
Of the 10 PCSK9 residues within 4 Å of Pep2-8, nine are also among the 13 residues within 4 Å of EGF(A) (i.e. Pro 155 , Asp 238 , Ile 369 , Ser 372 , Asp 374 , Thr 377 , Cys 378 , Phe 379 , and Val 380 (Fig. 8). Pep2-8 and EGF(A) use the same combination of secondary structure elements, a strand and ␣-helix, to establish complementary surfaces capable of binding strongly overlapping surfaces on PCSK9 (Fig. 9A). Structural mimicry between Pep2-8 and EGF(A) is strongest where they each form ␤-sheet hydrogen bonds with PCSK9 ␤-strand ␤14 (Fig. 9, B and C). The fidelity of this correspondence is quite high, as it must be if both involve interactions characteristic of a ␤-sheet, and accordingly the main chains, rather than side chains, dominate contacts. Additional interactions for both Pep2-8 and EGF(A) come from helical segments via their respective side chains, which project toward PCSK9. Here the correspondence is far from perfect; the Pep2-8 helix is longer, and the helical axes diverge by about 45°from parallel (Fig. 9A). The best side chain correspondence is seen between Pep2-8 Trp 6 and EGF(A) Leu 298 , which project into a closely similar region of PCSK9 (Fig. 9A).

DISCUSSION
PCSK9 has emerged as an important regulator of LDL-c metabolism, and current clinical studies with injectable PCSK9 inhibitors, such as antibodies, RNAi, and locked nucleic acids, hold great promise for the treatment of coronary heart disease (reviewed in Refs. 13 and 41). Undoubtedly, an orally available small molecule inhibitor of PCSK9, due to lower cost and ease of administration, would be a highly desirable alternative therapeutic agent. Pharmacologic approaches to inhibit PCSK9 with a small molecule appear challenging, because PCSK9 function is governed by protein-protein interactions. Phage display has been used successfully to identify peptides that disrupt protein-protein interactions (25, 38 -40, 42, 43), and for peptides targeting the antiapoptotic Bcl-2 family members, it established a structural basis to synthesize organic small molecules for therapeutic applications (44,45). By screening linear and constrained phage-displayed peptide libraries against PCSK9, we were able to identify Pep2-8, which neutralized PCSK9 activity and restored LDL receptor function in cellular assays.
Structural and biochemical studies revealed the molecular details of how Pep2-8 inhibits PCSK9. The peptide binding site largely overlapped with that of the EGF(A) domain; therefore, Pep2-8 acts a competitive inhibitor of LDL receptor binding. This is akin to the inhibition mechanism of anti-PCSK9 monoclonal antibodies, which also disrupt the interaction of the LDL receptor-EGF(A) domain with PCSK9 (46 -49). However, unlike the extended epitope region of the antibodies, the Pep2-8 contact area is merely 420 Å 2 , which is even smaller than that of the EGF(A) domain. Therefore, the 13-amino acid peptide is, to our knowledge, the smallest PCSK9 inhibitor with a defined mechanism of inhibition. Other described polypeptide inhibitors include the EGF(A) domain and variants thereof (26,37,50,51), the PCSK9 C-terminal domain (52), and annexin A2 (23). Peptides derived from the annexin A2-R1 domain directly inhibited PCSK9 binding to LDL receptor, but they required a length of 73 amino acids for binding interactions in the submicromolar concentration range (23). Another report showed activity of PCSK9 prodomain-derived linear peptides but without any insight into the molecular mechanisms (53).
Despite the small size of the Pep2-8 peptide, it has a well defined structure free in solution, comprising an N-terminal extended strand and turn followed by a two-turn helix. Moreover, the peptide fold is very similar in the absence or presence of PCSK9. Adopting the bound conformation in the absence of target protein allows this particular peptide to minimize the entropic losses ordinarily associated with transferring a ligand from a conformationally diverse free state to a restricted conformation when bound. The phage selection method seems particularly adept at selecting such conformationally constrained peptides, with the N-terminal strand/turn followed by a C-terminal helix being a particularly common motif (38 -40). In three reported cases (38 -40), the peptides also have well defined conformations in the absence of their binding partner. When comparing these strand-turn-helix peptides, a number of common structural features are found that probably contribute to their structural stability: (i) tethering the N-terminal strand to the helix to position the helix capping motif; (ii) helix capping motifs to help initiate the helical conformation; and (iii) predominantly hydrophobic residues along one face of the helix (Fig. 10). The recurrence of this motif suggests that it is one of the few ways in which a stable fold can be obtained in a peptide of ϳ15 residues.
In the three cases described above (38 -40), the peptides bind at a protein site ordinarily used for protein-protein interactions. Given the often flat and hydrophobic nature of such surfaces, the strand-turn-helix peptide motif seems to be particularly adept at complementing these regions, with the surface formed by the strand and the hydrophobic face of the helix providing a relatively flat surface with which to interact with the target protein. The N-terminal strand size and composition and the particular residues on the helix provide sufficient diversity for such peptides to bind selectively to a variety of protein targets. In the case of Pep2-8, the strand-turn-helix conformation positions the side chains of three key interacting residues, Phe 3 , Trp 6 , and Tyr 9 , in close proximity to form a central hydrophobic binding area. The important contributions of these residues to PCSK9 binding are also reflected in the Ͼ1,000-fold binding loss as a result of Alanine substitutions of Phe 3 and Trp 6 . Thus, the hydrophobic patch on one side of the Pep2-8 serves a dual role to stabilize the structure (with entropic benefit for binding) and to also form van der Waals contacts with the partner protein. The length of Pep2-8 could not be reduced without inflicting a substantial affinity loss, but alanine substitutions of the three C-terminal residues Asp 11 , Trp 12 , and Val 13 had no binding effects. The crystal structure explains this paradoxical finding and shows that side chains from these peptide positions do not contact PCSK9, consistent with the Ala scan results. None-theless, these residues extend the ␣-helix and thereby stabilize it and in this way indirectly influence the preceding residues in the helix that do contact PCSK9.
The structural details revealed an unexpected mimicry between Pep2-8 and important PCSK9-interacting elements of the EGF(A) domain. Pep2-8 uses its N-terminal extended segment to recreate the anti-parallel ␤-sheet main chain interactions between EGF(A) residues 305-310 and PCSK9 residues 379 -382 (␤14). In addition, the Pep2-8 ␣-helix provides additional contacts to PCSK9, as does the EGF(A) helical segment running from 295 to 301, although we note that the backbone atoms of these helical segments are not well aligned structurally. The peptide geometry allows Phe 3 and Tyr 9 of Pep2-8 to make favorable hydrophobic contacts with Ile 369 and Phe 379 of PCSK9. Asn 301 in the helical segment of EGF(A) is involved in two intramolecular hydrogen bonds, but the polar nature of its side chain suggests that it is not making productive contacts with PCSK9. Interestingly, replacing Asn 301 by a leucine has been shown to improve the binding to PCSK9 (26). Thus, the phage selection process led to a peptide with a binding site on PCSK9 that strongly overlaps that of a biological binding partner and in doing so yielded two structural elements that mimic the key structural elements of the biological binding partner while improving some of the specific contacts that are unproductive in the native complex.
The correspondence between PCSK9 residues contacted by Pep2-8 and EGF(A) is high (Fig. 8), and the amount of PCSK9  (40), and FVIIa (D) (38). In each case, the N terminus is on the left, and the helix is on the right. The top panels show the side view of the helix, and the bottom panels show the "end on." Side chains are depicted for those residues involved in tethering the N-terminal strand to the helix, participating in helix-cap motifs, or that form the hydrophobic surface on one face of the helix; other side chains are omitted for clarity. Helix-cap hydrogen bonds are shown in magenta, whereas helical-hydrogen bonds are in yellow. Side chains in the N-terminal strand involved in hydrophobic contacts with the helix are in green. These structures reveal a number of common features that probably contribute to the stability of the peptide fold: (i) tethering the N-terminal strand to the helix to position the helix capping motif (in the literature examples, this is achieved via a disulfide bond, whereas in Pep2-8, the hydrophobic interaction of Phe 3 and hydrogen bonds from Thr 4 to the side chain of Glu 8 achieve the same result); (ii) helix capping motifs to help initiate the helical conformation (e.g. in Pep2-8, hydrogen bond from Ser 5 side chain to the backbone amide of Glu 8 and the presence of the Glu 7 side chain to hydrogen-bond to its own backbone amide NH and to provide a favorable interaction with the helix dipole); (iii) predominantly hydrophobic residues along one face of the helix; the Van der Waals contacts between residues on adjacent turns help to propagate the helix and also strengthen the i, i Ϫ 4 hydrogen bonds by sequestering them from solvent. The arrangement of strand-loop-helix provides a scaffold from which to present a variety of relatively flat hydrophobic surfaces (bottom panels) for interacting with partner proteins.
surface buried by Pep2-8 is about 80% of that buried by EGF(A), yet Pep2-8 is only 13 amino acids long versus about 40 residues for EGF(A). Thus, Pep2-8 is a minimized version of EGF(A) with roughly equivalent binding affinity (26,37). The breadth of solutions to in vitro discovery of EGF(A) mimics is suggested by the reported structure of a phage-derived anti-PCSK9 antibody fragment 1D05 in complex with PCSK9 (48). This antibody also contacts PCSK9 residues contacted by EGF(A) and Pep2-8 and places key structure elements in rough analogy to the Pep2-8 extended N-terminal segment (for 1D05, the ␤-strand following the complementarity-determining region (CDR-H3) and C-terminal ␣-helix (for 1D05, CDR-H2) (Fig. 11). In this case, the structural analogy between the Pep2-8 helix and CDR-H2 from 1D05 (a ␤-hairpin loop) is obviously quite poor, and 1D05 has many additional aspects of its recognition of PCSK9 because it buries about twice the PCSK9 surface area of Pep2-8 and has an about 150-fold better binding than Pep2-8. Nonetheless, even as large and high affinity a PCSK9-binding protein as 1D05 has some elements in common with Pep2-8 and EGF(A). In PCSK9, the presence of a hydrophobic patch (centered on Ala 239 , Ile 369 , and Phe 379 ) adjacent to backbone NH and CO groups with unsatisfied hydrogen bonds (strand ␤14) clearly provides an attractive binding site for peptides, proteins and antibodies. Due to the non-directional nature of van der Waals contacts, there is considerable variety in the side chain vectors that ligands use to interact with the hydrophobic patch, whereas the specific geometric constraints required for hydrogen bond interactions dictate that the optimal interactions require a complementary ␤-strand on the ligand.
Peptide phage display provides a robust tool to identify ligands for targets of biological interest. The libraries are economical to establish (utilizing DNA synthesis and molecular biology manipulations), selection experiments are straightforward to perform, and the use of DNA sequencing methods provides a facile route to identify binding elements within complex mixtures of up to 10 11 unique members. However, phage display is limited by accessing only the diversity represented by the 20 naturally occurring amino acids. In contrast, synthetic combinatorial libraries offer a far greater scope for molecular diversity (non-natural amino acids, including D-stereochemistry; non-peptidic backbones; inclusion of moieties not present in naturally occurring biopolymers) with the caveat that they are more expensive to prepare and require more complex schemes to identify those molecules that bind to the target.
The work herein describes the identification of novel peptide ligands from phage libraries that are able to bind selectively to PCSK9 with an affinity comparable with the endogenous ligand EGF(A). The small size of the peptide (13 residues) serves to identify a focused epitope on PCSK9 and highlights key interactions that ligands must make to obtain a high binding energy. Although peptides are not usually considered as orally bioavailable drug candidates, several examples exist in the literature where peptides have served as the starting point for the successful development of small molecule drugs that exhibit oral bioavailability (44,45,54). Conceivably, Pep2-8 could provide such a steppingstone to go from protein to small molecule drug antagonists of the LDL receptor-PCSK9 interaction. However, the specific geometry of the EGF(A) binding site suggests that this will be challenging because the PCSK9 surface is relatively flat without concavities that are a prominent feature of potent, small molecule ligand binding sites. Indeed, significant optimization of Pep2 family members failed to identify peptides with IC 50 much below 1 M, suggesting that there may be an intrinsic limit on the energy available through interaction with this site. On the other hand, it is possible that the biostability of Pep2-8 could be improved by chemical modifications or the introduction of non-peptidic hydrocarbon backbones. Nonetheless, we remain optimistic that approaches such as the one described here may yield reagents and insights into ways of affecting PCSK9 function, possibly through interaction at allosteric sites or by binding to other domains of this large protein.