Binding of CDR-derived peptides is mechanistically different from that of high-affinity parental antibodies

Authors

  • Peter Timmerman,

    Corresponding author
    1. Pepscan Therapeutics B.V., Zuidersluisweg 2, 8243 RC Lelystad, The Netherlands
    2. Van't Hoff Institute for Molecular Sciences, Faculty of Science, University of Amsterdam, Nieuwe Achtergracht 129, 1018 WS Amsterdam, The Netherlands
    • Pepscan Therapeutics B.V., Zuidersluisweg 2, 8243 RC Lelystad, The Netherlands.
    Search for more papers by this author
    • These authors contributed equally in this work.

  • Susana G. Shochat,

    1. CNRS, University of Strasbourg, Biosensor group, ESBS, Bld Sébastien Brant, BP10413, 67412 Illkirch, France
    Current affiliation:
    1. School of Biological Sciences, Nanyang Technological University, 60 Nanyang Drive, Singapore 637551, Singapore.
    Search for more papers by this author
    • These authors contributed equally in this work.

  • Johan Desmet,

    1. Algonomics N.V., Technologiepark 4, 9052 Gent, Belgium
    Search for more papers by this author
    • These authors contributed equally in this work.

  • Rodrigo Barderas,

    1. Centro de Investigaciones Biológicas (CSIC), 28040 Madrid. Spain
    Search for more papers by this author
    • These authors contributed equally in this work.

  • Jose Ignacio Casal,

    1. Centro de Investigaciones Biológicas (CSIC), 28040 Madrid. Spain
    Search for more papers by this author
    • These authors contributed equally in this work.

  • Rob H. Meloen,

    1. Pepscan Therapeutics B.V., Zuidersluisweg 2, 8243 RC Lelystad, The Netherlands
    2. Academic Biomedical Centre, University Utrecht, Yalelaan 1, 3584 CL Utrecht, The Netherlands
    Search for more papers by this author
    • These authors contributed equally in this work.

  • Danièle Altschuh

    Corresponding author
    1. CNRS, University of Strasbourg, Biosensor group, ESBS, Bld Sébastien Brant, BP10413, 67412 Illkirch, France
    • CNRS, University of Strasbourg, Biosensor group, ESBS, Bld Sébastien Brant, BP10413, 67412 Illkirch, France.
    Search for more papers by this author
    • These authors contributed equally in this work.


  • This article is published in Journal of Molecular Recognition as a special issue on Affinity 2009, edited by Gideon Fleminger, Tel-Aviv University, Tel-Aviv, Israel and George Ehrlich, Hoffmann-La Roche, Nutley, NJ, USA.

Abstract

We present data that reveal crucial differences between the binding mode of anti-gastrin17 (G17, pyroEGPWLEEEEEAYGWMDF-NH2) monoclonal antibodies (mAbs) and their CDR-derived synthetic binders (SBs) with G17. The mAbs recognize the N-terminal sequence of G17 (pyroEGPWL) with nanomolar affinity and high sequence selectivity. Molecular simulations suggest that G17 recognition is based primarily on a multitude of weak antibody–ligand interactions (H-bonding, van der Waals, etc.) inside a structurally well-defined cleft-like binding pocket. Relatively small structural changes (e.g. G-2 to A for G17) have a drastic impact on affinity, which is characteristic for antibody-like binding. In contrast, SBs recognize various sequences, including G17-unrelated targets with affinities of 1:1 complexes estimated in the 0.1–1.0 mM range. In most cases however, the G17/SB complex stoichiometries are not well-defined, giving rise to multimer aggregate formation with high apparent complex stabilities. Mutational studies on both G17 and SBs reveal the importance of positively charged (K/R) and aromatic residues (W/Y/F) for G17/SB complex formation. We propose that the synthetic binders use combinations of electrostatic, hydrophobic, and/or cation–π interactions in a variety of ways due to their intrinsic flexibility. This may also be the reason for their relatively low target specificity. We speculate that our findings are of general relevance, in showing that high-affinity mAbs do not necessarily provide the optimal basis for functional mimics design. Copyright © 2010 John Wiley & Sons, Ltd.

INTRODUCTION

A wide variety of studies indicate that bioactive peptides can be generated using sequence information of antibody variable domains (Saragovi et al., 1991; Williams et al., 1991; Monnet et al., 1999; Levi et al., 1993, 2000; Park et al., 2000; Laune et al., 1997, 2002; Berezov et al., 2001; Tsumoto et al., 2002; Feng et al., 1998, 2005; Bès et al., 2001, 2003; Casset et al., 2003; Perosa et al., 2004; Heap et al., 2005). However, little experimental data are available to demonstrate that such peptides do indeed reproduce the structure and binding mode of the corresponding antibody paratope. The frequently observed ability of peptide-based mimics to compete with the antibody for binding to the antigen does indicate partly overlapping binding sites, but not necessarily similar binding modes. To our knowledge, studies providing direct structural evidence (NMR, X-ray) of similarities in binding interfaces were not reported. Crystallization of antigen–peptide complexes is likely to be difficult due to configurational diversity, i.e. the absence of a clearly defined complex structure. Relatively poor structural mimicry of paratopes by synthetic binders is also suggested by the fact that their biological activities are typically observed at concentrations 102–103 higher than those used with the parental antibodies (Saragovi et al., 1991; Levi et al., 2000; Tsumoto et al., 2002; Casset et al., 2003; Feng et al., 2005), with only few exceptions (Park et al., 2000; Heap et al., 2005; Qiu et al., 2007). In addition to this, binding selectivities are often poorly studied, most likely as this requires that binding and neutralization experiments were performed with a number of unrelated targets.

Gastrin peptides are hormones that regulate the secretion of gastric acid, and may act as growth factors for pancreatic, stomach and colorectal cancers (for review Grabowska and Watson, 2007; Rehfeld et al., 2007). Gastrin17 (G17, pyroEGPWLEEEEEAYGWMDF-NH2) is one of the most bioactive forms. The efficacy of immunization with G17DT (a portion of G17 attached to diphtheria toxin) has been established in colon and gastric tumor models (Gilliam and Watson, 2007), and vaccination with G17DT was shown to double the survival time of pancreatic cancer patients (Brett et al., 2002). Unfortunately, not all individuals respond to the vaccine, and the development of antibodies may be too slow for a fast progressing disease such as pancreatic cancer. These problems might be circumvented by administration of therapeutic gastrin antagonists.

In an earlier paper, a combinatorial synthetic approach allowed the selection of CDR-derived G17 binders that neutralized G17 in cell-based proliferation assays (Timmerman et al., 2009). Analysis of high-throughput screening data from >10 000 CDR-based binders derived from five anti-G17 antibodies (antibodies with KD's ranging from 500 pM to >1 mM) indicated that G17 binding activities correlated primarily with the presence of positive charges in the synthetic binders (SBs). Indeed, negatively charged or neutral peptides derived from high-affinity anti-G17 monoclonal antibody (mAb) 189DB3 (Barderas et al., 2008a) did not bind G17. In contrast to this, positively charged peptides derived from either the low-affinity PAR10C3 and PAR10D10 anti-G17 scFvs (Barderas et al., 2006), or from the high-affinity mAb 189DB3, or from D1.3 anti-HEL mAb (Harper et al., 1987), or those containing random sequences, also displayed significant G17 binding (Timmerman et al., 2009). These observations implied a lack of correlation between G17 binding affinity of CDR-derived peptide binders and parental antibodies, and prompted us to conduct a detailed analysis of SBs binding properties. In this paper, we compare the binding properties of two anti-G17 mAbs and of a set of CDR-derived peptides. The surface plasmon resonance (SPR) analysis of binding affinity, stoichiometry, and selectivity reported here, together with the identification of G17 residues important for binding, provide clear evidence that synthetic binders and antibodies use different binding mechanisms.

MATERIALS AND METHODS

Design and synthesis of CDR-derived synthetic binders

The names and sequences of all CDR-derived SBs used in our studies are listed in Table 1; the synthesis of these was described before (Timmerman et al., 2009). SBs were derived from the CDR sequences of three anti-G17 murine antibodies (Figure 1), of which one, mAb 189DB3, binds G17 with high affinity (KD = 11.3 nM) (Barderas et al., 2008a) and the two others (scFvs PAR10C3, PAR10D10) with lower affinities (KD ≥1 µM) (Barderas et al., 2006). The synthetic peptides correspond to various fragments of CDR sequences derived from the antibodies that were separated by Cys residues, which allows their covalent coupling to a bromomethylated aromatic scaffold (T3: 1,3,5-tris(bromomethyl)benzene; T2: 1,3-bis(bromomethyl)benzene) using CLIPS™ technology (Chemically LInked Peptides on Scaffolds) (Timmerman et al., 2005). Double loops (format CT3-CDRA-CT3-CDRB-CT3; Table 1: 1d–r), a triple loop (format cyclic [CT3-CDRA-CT3-CDRB-CT3-CDRC-]; Table 1: 2a), and a single loop (format CT2-CDRA-CT2; Table 1: 3b), were used in the present study. Controls correspond to linear peptides (Table 1: 5b–5c), and double loops with random sequences (1t–1w). 1s represents a double loop derived from the anti-HEL mAb D1.3 (Harper et al., 1987).

Table 1. Sequence information for CDR-derived synthetic binders (SBs)
SBPeptide SequenceaConstraint type (CTC)TargetbParent mAb/scFvCDRs involved
  • a

    CT” represents cysteines interconnected via either a T3- or a T2-scaffold. All linear peptides were acetylated at the N-terminus and have a C-terminal amide (CONH2); “cyclic[XXXX]” represents backbone-cyclized peptides. “S” (in boldface) represents serines substituting (T3-connected) cysteines.

  • b

    The target is the immunogen to which the parent mAb/scFv was elicited or selected.

1dCT-GTHFPR-CT-VASIKS-CTT3G17189DB3L3/H2
1eCT-KSGGS-CT-VASIK-CTT3G17189DB3H2/H2
1hCT-VASIKS-CT-GTHFPR-CTT3G17189DB3H2/L3
1mCT-GSTT-CT-KPKK-CTT3G17PAR10C3H2/H3
1nCT-AKKPKK-CT-KHYRPPT-CTT3G17PAR10C3H3/L3
1oCT-STIQP-CT-KPKKF-CTT3G17PAR10C3H3/H2
1pCT-RRRK-CT-AKRG-CTT3G17PAR10D10L3/H3
1qCT-AKRG-CT-VSAI-CTT3G17PAR10D10H3/H2
1rCT-KRGR-CT-VSAI-CTT3G17PAR10D10H3/H2
1sCT-HFWSTPRT-CT-HFWSTPRT-CTT3HELD1.3L3/L3
1tCT-KPKPMKIE-CT-KPKPMKIE-CTT3random
1uCT-KPKPMKIEM-CT-KPKPMKIEM-CTT3random
1vCT-PKPMKIE-CT-KPKPMKI-CTT3random
1wCT-PMKI-CT-KPKP-CTT3random
2acyclic[GTHFPR-CT-RSDRY-CTVASIKS-CT]T3G17189DB3L3/H3/H2
3bCT-GTHFPR-CTT2G17189DB3L3
5bS-GTHFPR-S-VASIKS-SnoneG17189DB3L3/H2
5cGTHFPR-S-RSDRY-S-VASIKS-SnoneG17189DB3L3/H3/H2
Figure 1.

Schematic structures of antibody and CDR-derived SBs. (A) The paratope of an antibody is composed of six CDRs (red), of which three (H1, H2, H3) are carried by the heavy chain (yellow) and three (L1, L2, L3) by the light chain (blue). (B) The synthetic platform is a bromomethylated aromatic scaffold. (C) Synthetic binders are composed of 1, 2, or 3 CDRs, coupled or not to the scaffold.

Names and corresponding peptide sequences of all synthetic targets are given in Table 2. The theoretical pI values were calculated using the ProtParam tool of the ExPASy Proteomics Server (Gasteiger et al., 2003).

Table 2. Sequence information for synthetic targets
Target nameSequenceTheoretical pIMW (Da)
Gastrin 17pyroEGPWLEEEEEAYGWMDFC-NH23.402201.5
Gastrin1-12pyroEGPWLEEEEEAYC-NH23.511564.8
Gastrin6-17biotin-EEEEEAYGWMDF-NH23.401760.0
cholestocystokinin-8 (CCK-8)biotin-DYMGWMDF-NH23.561289.6
gonadotropin releasing hormone (GnRH)pyroEHWSYGLRPGC-NH28.231285.6
TMVP137–151mutH-GSYNRSDFYDSSGLV-OH4.211770.0

Molecular simulations

The 3-D structural model of the variable domains of mAb 243BA5 was generated using a proprietary, semi-automated tool for antibody construction (Barderas et al., 2008b). Simulations were performed in a purely deductive mode in order to rationalize the data from the experimental Ala-replacement studies.

SPR experiments

All experiments were performed on a Biacore 2000 instrument (GE-Healthcare Biacore), at 25°C, using HBS (10 mM HEPES pH 7.4, 150 mM NaCl, 3.4 mM EDTA) containing 0.005% surfactant P-20 as running and dilution buffer, except for salt sensitivity experiments in which the buffer contained 400 mM NaCl. CM5 sensor surfaces (BR-1000-14) and chemicals for immobilization (EDC, NHS, PDEA, ethanolamine-HCl) were purchased from GE-Healthcare Biacore.

Surface preparation

Proteins (recombinant human insulin I2643 from Sigma-Aldrich, HEL L-6876 from Sigma and streptavidin S47 from Sigma) and Cys-containing peptides (G17, gastrin1-12, Ala-substitution variants of G17 and GnRH from Pepscan Therapeutics B.V., The Netherlands) were covalently coupled to carboxyl groups of the dextran matrix using amine and thiol coupling chemistries (BiaApplications Handbook, GE-Healthcare Biacore), respectively. Briefly, surfaces were activated for 10 min with EDC/NHS, followed in case of thiol coupling, by 7 min injections of PDEA in 100 mM borate buffer pH 8.5 and ethanolamine-HCl. Proteins or peptides were injected at a concentration between 10 and 100 µg/ml in 10 mM sodium acetate pH 5.5 or sodium formate pH 3.5, until the desired coupling level was achieved. Subsequently the surfaces were deactivated by injection of ethanolamine (amine coupling) or 50 µM Cys in 100 mM formate pH 4.3, 1.0 M NaCl (thiol coupling), and washed with 1 min pulses of 50 mM HCl. The reference surface was treated like the other surfaces except that no peptide was injected. Biotin-containing peptides (gastrin6-12, CCK-8 from Pepscan Therapeutics B.V., The Netherlands) were captured on streptavidin that had been previously immobilized in a covalent fashion through amine coupling. The reference surface corresponded to streptavidin alone. These procedures resulted in surfaces for SB analysis with 1000–1640 RU of insulin, 1700–5000 RU of HEL, and 300–1700 RU of peptides. For analysis of mAb binding to Ala-substitution variants of G17, surfaces with less than 25 RU of the peptides were prepared.

Binding experiments

CDR-derived peptides were solubilized in H2O to a concentration of 1.0 mM, diluted in running buffer at about 100 µM and filtered through a 0.22 µm MILLEX GP filter unit (Millipore). The OD260 nm was measured before and after filtration to verify the presence of the peptide. The SBs were injected on target and reference surfaces at a flow rate of 20 µl/min during 1 min at concentrations between 10 and 100 µM. Sensor surfaces were regenerated by a 10 µl injection of 50–100 mM HCl. Responses were recorded 10 s before the end of injection and were normalized as 100 × Response/MW (MW-adjusted responses) to correct for differences in MWs between the binders. In order to compare data recorded on surfaces with different immobilization levels, responses were normalized as %Rmax, where Rmax is the response expected at surface saturation if one SB binds one immobilized target (Rmax = RUimmob × MWbinder/MWtarget). The BIA simulation software (GE-Healthcare Biacore) was used for simulating SPR curves expected for interactions with different kinetic constants.

Evaluation of binding affinities

CDR mimics were injected over G17 and corresponding reference surfaces at 5–7 different concentrations at a 30 µl/min flow rate for 60 s. The surfaces were regenerated by a 10 µl injection of 100 mM HCl. Equilibrium responses were recorded 10 s before the end of SB injection.

Ala-scan analysis of G17 by SPR

SBs were injected on sensor surfaces with high levels (800–1700 RU) of immobilized Ala-substitution variants of G17 (immobilization via N-terminal cysteine) for 1 min. Responses were expressed as % of the calculated Rmax. Solutions of 4.4 and 3.0 nM of mAbs 189DB3 and 243BA5, respectively, were injected on sensor surfaces with low levels (5–22 RU) of immobilized Ala-variants of G17. Since the amount of peptide immobilized cannot be known precisely for such low response levels, the Rmax for each surface was determined experimentally by injecting saturating concentrations of mAbs directed to the C- or N-terminus of G17 (Barderas et al., 2008a). The responses recorded with mAbs 189DB3 and 243BA5 were measured 1–2 min after injection and normalized to the experimental Rmax.

Ala-scan analysis of mAb binding to G17 (Pepscan microarrays)

Synthesis of Ala-variants of G17 in microarrays on polypropylene support was performed as described (Timmerman et al., 2004). After sidechain deprotection using TFA and scavengers, the microarrays were washed with excess of ACN/H2O 1:1 (3 × 10 min) and sonicated in disrupt-buffer [1% SDS/0.1% β-mercaptoethanol (BME) in phosphate-buffered saline (PBS, pH 7.2)] at 70°C for 30 min, followed by sonication in milliQ-H2O for another 45 min. Microarrays were then pretreated with PBS for 30 min, followed by precoating with incubation buffer (PBS containing 5% ovalbumin, 5% horse serum, and 1% Tween-80) for 1 h. Then, the microarrays were incubated with anti-G17 mAb (typically 1/1000, diluted in incubation buffer) overnight at 4 °C. After washing (3 × 10 min) with PBS/Tween-80 (0.05%), the microarrays were incubated with peroxidase labeled rabbit anti-mouse antibody for 1 h at 25°C (1/1000; Dako, Glostrup, Denmark) and, after first washing again (3 × 10 min) with PBS/Tween-80 (0.05%), with the peroxidase substrate 2,2-azino-di-3-ethylbenzthiazoline sulfonate (ABTS; 50 mg in 100 ml 0.1 M citric acid/sodium phosphate (McIlvaine) buffer, pH 4.0, containing 20 µl of a 30% H2O2 solution in water). After 1 h, the absorbance (at 405 nm) was measured using a CCD-camera (XC-77RR, Sony, Japan). Bound mAb was removed by sonication in disrupt-buffer as described above. The micro-arrays were subsequently re-used for screening with different mAbs, until the responses with control mAbs were no longer seen (approximately 10–15 times).

RESULTS

Analysis of SPR binding curves

Sequence information and names for G17 binders (Timmerman et al., 2009) used in this study are given in Table 1 (see methods for a summary of design strategy). SBs that represent only a fraction of the paratope with imperfect structural mimicry are expected to display low gastrin binding affinities. Figure 2 illustrates the shape of expected SPR binding curves for low affinity interactions (KD = koff/kon in a range 1 µM–0.1 mM; Figure 2A) and when varying the kinetic parameters koff and kon, while keeping KD = 10 µM (Figure 2B). Conditions of the simulations are close to our SPR experimental conditions: binders at concentrations 6, 12, and 25 µM (Figure 2A) and 25 µM (Figure 2B); G17 surfaces with maximal binding capacity (Rmax) of 1000 RU. The shapes of the experimental SPR curves were rarely consistent with the expected curves shown in Figures 2A and B. A typical example is shown in Figure 2C for the injection of the random bi-cycle 1t at concentrations 3–50 µM on a G17 surface with Rmax 1285 RU. The injection phases indicate slow association (compare curve shapes in Figure 2C with those simulated in Figure 2B for kon 103–105 M−1 s−1), and post-injection phases are clearly multiphasic. For most binders, long injections at high concentrations yielded responses above the Rmax calculated assuming formation of a simple 1:1 complex, and surface saturation could not be reached (not shown). Altogether, these typical behaviors reflect the binding of heterogeneous and/or aggregated material. Another sign of aggregation is the dependence of curve shapes on G17 immobilization level. The proportion of stable complexes is lower on a surface with Rmax 306 RU (Figure 2D) than that observed on a surface with Rmax 1285 RU (Figure 2C). A likely explanation for this is that the spacing between G17 molecules on a low-density surface is less favorable to multimeric binding than that on a high-density surface.

Figure 2.

SPR analysis of G17/1t interactions. SPR responses are expressed as RU on the left y-axis, and as % of the theoretical maximal binding capacity (% Rmax) of the surfaces on the right y-axis. (A and B) Simulated SPR curves for SBs interacting with G17, on a G17 surface with Rmax = 1000 RU. Kinetic rate constants (format koff/kon) are indicated on the curves. (A) Influence of KD on curve shape for [SB] = 6, 12, and 25 µM, with kon = 104 M−1 s−1 and with koff = 10−2, 10−1, and 1 s−1, resulting in KD = 10−6 (black), 10−5 (dark gray), and 10−4 M (light gray), respectively. (B) Influence of varying rate constants for KD = 10−5 M and [SB] = 25 µM, and with koff (s−1)/kon (M−1 s−1) varied between 10−2/103 (light gray), 10−1/104 (dark gray), and 1/105 (black). (C and D) Experimental set of binding curves recorded when injecting 1t at concentrations 3–50 µM on a G17 surface with Rmax = 1285 RU (C) and Rmax = 306 RU (D) (Rmax = Rligand × MWanalyte/MWligand. %Rmax = 100 if each immobilized molecule has bound one injected molecule).

SPR-based ranking of binders

Only few SBs, among which the linear peptides 5b and 5c, and the constrained bi-cycle 1d (Figures 3A, D, and B, respectively) produced binding curves that reflect the formation of low affinity 1:1 complexes (simulated by light gray curves in Figure 2A). Figure 3C shows the MW-adjusted equilibrium responses for the linear peptide 5b, and the corresponding cyclic double-loop analog 1d, plotted as a function of SB concentration. Figure 3F shows a similar plot for the linear peptide 5c and the cyclic triple-loop analogue 2a. Although the binding of 2a may be slightly affected by heterogeneity (Figure 3E), it appears that peptides that are constrained in the form of loops show higher G17 responses compared to their linear analogues. It can also be observed that the larger peptides 5c and 2a show higher G17 response levels than the corresponding smaller ones 5b and 1d, respectively, from which they are derived by the addition of the sequence “RSDRY.”

Figure 3.

SB concentration dependence of SPR responses. (A, B, D, and E) SPR curves for linear binders 5b (A) and 5c (D), and the corresponding bi-cyclic 1d (B) and tri-cyclic 2a (E) binders, respectively. The binders were injected on a G17 surface with maximal MW-adjusted response of 19 (100 × RU/Da) in the concentration ranges indicated. (C and F) Experimental MW-adjusted responses recorded 10 s before the end of injection for 5b (▴) and 1d (♦) (C), and for 5c (▪) and 2a (♦) (F). Values are compared with those expected for interactions with KD 10−4 M (○) and 10−3 M (□).

In these experiments, equilibrium responses cannot be used for KD calculation because they represent less than 25% of surface saturation at the highest SB concentration, making data fitting unreliable. However, simulations of Req values expected in these experimental conditions indicate that KD of interaction between G17 and SBs 5b, 1d, 5c, lies in the 10−3–10−4 M range (open squares and circles, respectively, in Figure 3C and F). For the other binders, strong responses clearly reflect aggregation and avidity effects, and must be interpreted with caution. Their G17 binding capacity is nevertheless unambiguous, which allowed to identify G17 residues important for SB and antibody binding.

mAb epitope identification and computer simulation of the 243BA5/G17 complex

The SPR and microarray approaches were used to evaluate binding of mAbs 243BA5 and 189DB3 (Barderas et al., 2008a) to a set of 17 surface-immobilized Ala-variants of G17 (Figure 4A). This confirmed high sensitivity (>95% decrease of G17 binding) to substitutions in the N-terminal sequence pyroEGPWL, as reported previously for mAb 243BA5 (Barderas et al., 2008a). A full replacement analysis for the pyroEGPWL-region (substitutions by 19 other natural amino acids) showed that not a single replacement was allowed at pyroE-1, G-2, and L-5 without complete loss of G17 binding (Figure 4B). For P-3 only one (F) and for W-4 only two (F/Y) replacements were allowed.

Figure 4.

Epitope identification for mAbs 243BA5 and 189DB3. (A) Ala-replacement studies (SPR and micro-array) with surface-immobilized G17 (A11 for G). Responses are given relative to that of native G17 (100%). “+++” refers to >95% decrease, “++” to >75% decrease, “+” to >25% decrease, and “−” to <25% decrease of binding signal. (B) Microarray binding data for mAbs 243BA5 and 189DB3 to a partial replacement set (substitution of native amino acid into 19 other naturally occurring amino acids) for the first five amino acids (pyroE1 to L5), A11, and W14 of G17. (C) Top- and sideview of the space-filled model of the simulated 243BA5/G17 complex, showing a deep cleft-like binding pocket for G17 and specific interactions of G17 with residues of the antibody.

A 3-D structural model of the variable domains of mAb 243BA5 revealed a groove in between VL and VH and a deep, hydrophobic pocket in between L1, L3, and H3 (Figure 4C), caused mainly by the short H3 (five residues) in combination with a small side chain at VL S-91. The W-4 side chain of G17 was expected to be contained in this pocket, in view of its experimentally determined importance (Figure 4A) and explicit hydrophobic character. Considering also the importance of L-5 for G17 binding (Figure 4A), the 4-W-L-5 motif was oriented in a turn-like conformation with the side chains facing the pocket. Then, the G17 chain was extended by a combination of alternating conjugate gradient minimization (typically 200 steps) and short molecular dynamics simulations (typically 10–20 ps). This resulted in a docked conformation with the following features (Figure 4C): (1) full burial and near-optimal packing of W-4, including a H-bond with VH D-98, suggesting that it forms the dominant anchor for G17 binding, (2) the L-5 side chain of G17 is buried in the same hydrophobic pocket, but packing is qualitatively less optimal, (3) the side-chain ring of P-3 of G17 points towards the solvent, while inducing a kink in the chain direction, (4) the HA2 pseudo-side-chain atom of G-2 of G17 is fully buried and sterically constrained, (5) the pyroE-1 is sandwiched between VH Y-99 and VH F-100 of mAb 243BA5 and forms a H-bond with VH R-97, (6) E-6 forms an ionic bridge with VL K-50 of mAb 243BA5, which is H-bonded to the carbonyl oxygen of P-3, (7) E-7 of G17 is largely solvent exposed, but it can possibly H-bond to VH Y-99 of mAb 243BA5. None of the remaining residues of G17 can appreciably interact with the Fv domains. These observations are fully in accordance with the Ala-replacement data for mAb 243BA5 as described (Figure 4A). Based on a qualitative assessment of the modeled complex, it was deduced that G17 binding is mainly driven by the short H3 creating a pocket for W-4 and by strong, local interactions of pyroE-1 of G17.

Identification of the epitope recognized by SBs

Binding to the 17 surface-immobilized Ala-variants of G17 was studied for five bi-cycles, of which two are derived from anti-G17 mAb 189DB3 (1d, 1h), one from anti-HEL mAb D1.3 (1s), and two have random sequences (1t, 1u) (none of the 243BA5-based peptides showed any binding to G17; Timmerman et al., 2009). For all five, it appeared that binding to G17 was almost exclusively affected by replacement of aromatic residues (W-4, Y-12, W-14, F-17) (Figure 5A). In contrast with the drastic effect of mutations on mAb binding, the replacements decreased the response level only partly (mostly 25–75% of WT response). The effect of substituting negatively charged residues (E-6 to 10 or D-16) was negligible, despite the importance of positive charges in the SBs. This may be explained by the small changes in global charge (from −6 to −5) upon E or D to A substitution, which would not impede electrostatic attraction. Alternatively, positively charged residues in the SBs may also interact with aromatic residues through cation–π interactions, instead of interacting with negative charges on G17 (Ma and Dougherty, 1997). The role of cation–π interactions for stabilizing peptide–peptide complexes has been demonstrated (Pletneva et al., 2001; Woods, 2004).

Figure 5.

Epitope identification for SBs. (A) Ala-replacement studies (SPR) with surface-immobilized G17 (A11 into G) performed with two bi-cyclic SBs derived from mAb 189DB3, (1d and 1h), one from anti-HEL mAb D1.3 (1s), and two random mimics (1t and 1u). Responses are given relative to that of native G17 (100%). “++” refers to >75% decrease, “+” to >25% decrease, and “−” to <25% decrease of binding signal. (B–D) SPR responses of 3b, 1d, and 1h to immobilized G17, three G17 related peptides (light gray), and two unrelated peptides (black). Peptide sequences are given in Table 2. Responses are expressed as %Rmax.

Our data show that the SBs do not recognize the N-terminal pyroEGPWL sequence, which is the epitope of mAb 189DB3. They further suggest that SBs do not recognize a well-defined epitope on G17, but instead a fuzzy combination of various hydrophobic and/or negatively charged patches, indiscriminately from their N- or C-terminal location on G17. To verify this hypothesis, we analyzed the interaction of SBs with five peptides (Table 2), of which three present sequence identities with G17: the 12 N-terminal residues of G17 (gastrin1-12), the 12 C-terminal residues of G17 (gastrin6-17), and cholestocystokinin-8 (CCK-8) that shares the 5 C-terminal residues with G17. The two unrelated peptides are cysteine-extended gonadotropin releasing hormone (GnRH) and a triple mutant derived from the coat protein of tobacco mosaic virus (TMVP137–151mut,). Figure 5B shows SPR responses recorded when injecting the mono-cycle 3b, and bi-cycles 1h, and 1d over the six peptides. The SBs showed no binding to the two G17-unrelated peptides, and 3b recognized peptides sharing the five C-terminal residues of G17 (sequence GWMDF), suggesting some binding selectivity. However, 1d and 1h bound gastrin1-12 and CCK-8 that have no sequence identities, which indicates a lack of sequence-specific recognition.

Antigen selectivities of CDR-derived binders

The epitope identification studies clearly indicate that the mAbs and SBs recognize different features on G17, and highlight loose sequence requirements for G17 recognition by SBs, in sharp contrast with the stringent structural requirements of mAbs. To further investigate this point, two proteins were chosen as targets: hen eggwhite lysozyme (HEL; MW∼14.7 kDa, pI = 11.0) and human insulin (MW = 5808 Da, pI = 5.4). Figure 6A shows that all 12 SBs showed some binding to HEL, while binding to surface-immobilized insulin was only observed for the +4 and +6-charged SBs 1n, 1p, and 1v. The response levels (%Rmax) were only 1/2 to 1/3 of that observed for G17 when responses are normalized with respect to the binding capacities of the surfaces (calculated assuming 1:1 complex formation).

Figure 6.

Binding selectivity studies for bi-cycles at NaCl concentrations of 150 nM (A) and 400 nM (B). Comparison of SPR binding levels (%Rmax) for a set of bi-cyclic binders derived from mAb 189DB3 (1d/1h), scFv PAR10C3 (1m1o), scFv PAR10D10 (1p1r), mAb D1.3 (1s), and random bi-cyclic binders (1t1w) to surface-immobilized G17 (black), HEL (light gray) and human insulin (dark gray). Interactions were analyzed only once at [NaCl] = 400 mM and at [NaCl] = 150 mM for 1v/1w on the HEL and insulin surfaces (indicated by *). Other values represent mean values over 2–5 different experiments.

The experiments were repeated at a higher salt concentration (400 mM instead of 150 mM NaCl) in order to evaluate the importance of electrostatics in recognition (Figure 6B). Binding to HEL was similar at both salt concentrations, while binding to insulin was abolished, suggesting that electrostatics play a major role in SB/insulin interactions, but not in SB/HEL interactions. Binding to G17 was decreased but not abolished, suggesting a contribution of various contact types. Therefore, the synthetic binders not only bind G17 in different ways, but also showed different binding activities and modes for different targets.

Residues of the SBs important for binding

Mutational analyses of the SBs was not performed by SPR, because binding levels recorded when injecting SB variants on immobilized antigens are influenced both by affinity and aggregation propensity, making it difficult to ambiguously identify residues important for binding. An Ala-scan of SB 1e with sequence CT-KSGGS-CT-VASIK-CT was performed on microarrays with immobilized 1e variants, which indicated the importance of the two Lys residues in 1e (Timmerman et al., 2009). An Ala-scan of 1d on microarrays also indicated that charged residues R and K influence binding (17 and 24% of WT binding), followed by F and A (replaced by G) (33 and 31% of WT binding, respectively, data not shown).

DISCUSSION

Synthetic G17 binders (Timmerman et al., 2009; Figure 1) were developed based on the CDR sequences of anti-G17 antibodies (Barderas et al., 2006, 2008a), and the binding properties of a sub-set of the SBs have been analyzed in detail by SPR (this study). Our experimental observations are consistent with those reported in other studies of paratope mimicry: (1) synthetic antigen binders can be obtained based on antibody sequences; (2) they recognize the antigen with, for some of them, high apparent complex stabilities; (3) binding strength can be improved by peptide cyclization or by increasing their sequence length; and (4) they neutralize G17 in a cell-based assay. Based on these observations, one could easily assume that the SBs are decent structural mimics of the antibody paratopes.

However, paratope mimicry in our system was initially questioned by the observation that G17 binders can be constructed from the CDR sequences of anti-G17 antibodies, but also from other antibodies or random sequences (Table 1), and that binding activity is strongly correlated with the presence of positive charges in the SBs (Timmerman et al., 2009). We show here that the anti-G17 mAbs and SBs recognize different features of G17: while the antibodies recognize the N-terminal sequence pyroEGPWL with high (low nanomolar) affinity and little tolerance to amino-acid replacements (Figure 4), the SBs recognize various sequences with 1:1 affinity in the 0.1–1 mM range (Figure 3), and sometimes form stable complexes due to aggregation and avidity effects (Figure 2). Mutational studies (Figure 5) suggest that an appropriate combination of charges and/or hydrophobic residues is sufficient to cause detectable binding, although with relatively low binding selectivity. The ability of the synthetic peptides to bind other targets (Figure 6A) and with different salt sensitivities (Figure 6B), indicates that they use different binding modes depending on the target. Clearly, SBs do not mimic the highly specific binding mode of mAbs.

Other CDR mimicry studies

Mutational studies have been used by others to assess whether antibodies and CDR-derived peptides recognize the same epitope, or to check whether corresponding residues in the parental antibody and synthetic peptide are involved in antigen binding. The data mostly suggest at least partial differences in binding mode between antibody and CDR-derived binders (Bès et al., 2002; Heap et al., 2005; Williams et al., 1991). For example, Bès et al. (2002) identified 19 amino acid positions as apparently critical for binding via Ala-replacement studies with hexapeptides derived from the CDR regions of the anti-CD4 antibody fragment Fab 13B8.2, and then analyzed the effect of replacing 16 of the corresponding positions in the Fab by Ala. Eleven replacements indeed impaired Fab/CD4 complex formation, but five did not, indicating differences in CD4 binding mechanisms between Fab and CDR-derived peptides. In another study, residues important for binding of peptides derived from the CDR domains of mAb HyHEL-5, to HEL were identified by Ala-replacement studies (Laune et al., 1997). From all residues identified as important for HEL-binding, only 14 involved antibody residues located at the HEL/HyHEL5 complex interface in the X-ray structure, while 16 corresponded to non-interface residues (Laune et al., 1997), and were thus identified for reasons other than their contribution to the paratope.

Other common observations include an improvement of binding strength by cyclization of peptides (Figure 3), which has been attributed to an increase in the number of structural constraints (Berezov et al., 2001; Laune et al., 2002; Levi et al., 1993; Saragovi et al., 1991; Tsumoto et al., 2002). However, binding strength may be improved by decreasing the entropic cost related to the mobility of the free peptide, without necessarily achieving structural mimicry. Also in our case, the higher responses of the larger peptides are likely to result from an increase in net charge and/or contact surface area or increased hydrophobicity, and not from a better structural mimicry of the paratope.

Can synthetic peptides mimic protein-binding sites?

An extensive literature deals with peptidomimetics and peptides displaying biologic activity (for review, Sillerud and Larson, 2005; Watt 2006; Robinson et al., 2008). Synthetic peptides may obviously mimic binding sites on proteins, in particular if they consist of amino-acids that are contiguous in the sequence as linear epitopes. For example, the antibody fragment Fab 57P, elicited to the tobacco mosaic virus protein (TMVP), recognizes TMVP, and a TMVP-derived 15-mer with similar low nM affinities (Choulier et al., 1999, 2001).

In contrast with linear binding sites, an efficient mimicry of highly structured sites, such as paratopes, may require the synthesis of relatively large peptides. A possible example is provided by the design of mimetics of antibody Z12 to tumor necrosis factor-alpha (TNFα), using the structure of the antigen and molecular docking approaches. The mimetics were assayed for their capacity to inhibit the TNFα/Z12 interaction and to reduce the TNFα mediated cytotoxicity. A 19-mer corresponding to one CDR was used in mM amounts in the biological assays (Feng et al., 2005), while a peptide based on two CDRs was used in µM amounts (Qin et al., 2006). Finally, the peptides were displayed on a scaffold, in the form of a designed recombinant scFv able to inhibit TNF-induced cytotoxicity at concentrations 50-fold higher only when compared to the parent antibody (Chang et al., 2007). Clearly, short peptides are not structured like a mAb paratope. They expose at the surface amino-acids that are part of the hydrophobic interior in the mAb, which is likely to promote binding and/or aggregation.

CONCLUSION

Our data indicate that the CDR-derived peptides bind G17 and other targets by combinations of electrostatic, hydrophobic, and/or cation–π interactions, in sharp contrast with the highly selective binding mode of the parental mAb. Aggregate formation is observed in most cases and may be relevant for G17-neutralizing activity. In addition to increasing the binding strength, other challenges for the development of improved SBs will consist in achieving higher solubility and selectivity. We believe that our findings are of general relevance and propose that, in view of the drastic differences in binding mode, high-affinity mAbs do not necessarily provide the optimal basis for the design of potent synthetic binders.

Acknowledgements

This work was financially supported by a European Commission FP6 cooperative research project (COOP-CT-2004-512691). RB was recipient of a contract from the Fondo de Investigaciones Sanitarias of the Spanish Ministry of Health. DA and SGS acknowledge support from the CNRS (Centre National de la Recherche Scientifique) and UdS (University of Strasbourg). PT and RHM acknowledge also financial support from the Dutch Ministry of Economic Affairs (grant TSGE2017). RB and IC acknowledge financial support from the Spanish Ministry of Science (grant BIO2006-07689).

Abbreviations used:
CCK-8

cholestocystokinin-8

CDR

complementarity determining region

EDC

N-ethyl-N′-[3′-(diethylamino)propyl]carbodiimide

Fab

antibody binding fragment

G17

gastrin17

GnRH

gonadotropin releasing hormone

HEL

hen egg white lysozyme

KD

equilibrium dissociation constant

kon

association rate constant

koff

dissociation rate constant

L1,

L2, L3, H1, H2, H3: respectively CDR1, CDR2, CDR3 of antibody light and heavy chains

mAb

monoclonal antibody

NHS

N-hydroxysuccinimide

PDEA

2-(2-pyridinyldithio)ethaneamine hydrochloride

Rmax

maximal response

RU

resonance unit

scFv

single-chain variable fragment

SB

synthetic binder

SPR

surface plasmon resonance

TMVP

tobacco mosaic virus protein

VL

variable region of the light chain

VH

variable region of the heavy chain

WT

wild type

Ancillary