Báo cáo khoa học: Selection of a CXCR4 antagonist from a human heavy chain CDR3-derived phage library doc

In this study, we compared the efficacy of a phage library displaying human heavy chain complementarity determining region 3 HCDR3 repertoires with a set of conventional random peptide li

chain CDR3-derived phage library

Andy Chevigne´1, Aure´lie Fischer1, Julie Mathu1, Manuel Counson1, Nadia Beaupain1,

Jean-Marc Plesse´ria1, Jean-Claude Schmit1,2and Sabrina Deroo1

1 Laboratoire de Re´trovirologie, Centre de Recherche Public-Sante´, Luxembourg, Luxembourg

2 Service National des Maladies Infectieuses, Centre Hospitalier Luxembourg, Luxembourg

Keywords

CXCR4 antagonist; HCDR3; natural

sequence randomization; peptide repertoire;

phage display

Correspondence

A Chevigne´, Laboratoire de Re´trovirologie,

Centre de Recherche Public-Sante´, 84, Val

Fleuri, L-1526 Luxembourg, Luxembourg

Fax: +352 26970 221

Tel: +352 26970 336

E-mail: andy.chevigne@crp-sante.lu

(Received 16 March 2011, revised 27 May

2011, accepted 6 June 2011)

doi:10.1111/j.1742-4658.2011.08208.x

Phage display technology is a powerful selection approach to identify strong and specific binders to a large variety of targets In this study, we compared the efficacy of a phage library displaying human heavy chain complementarity determining region 3 (HCDR3) repertoires with a set of conventional random peptide libraries for the identification of CXCR4 antagonists using a peptide corresponding to the second extracellular loop

of the receptor CXCR4 as target A total of 11 selection campaigns on this target did not result in any specific ligand from the random peptide libraries In contrast, a single selection campaign with an HCDR3 library derived from the IgM repertoire of a nonimmunized donor resulted in nine specific peptides with lengths ranging from 10 to 19 residues Four of these HCDR3 sequences interacted with native receptor and the most frequently isolated peptide displayed an affinity of 5.6 lM and acted as a CXCR4 antagonist (IC50= 23 lM) To comprehend the basis of the highly efficient HCDR3 library selection, its biochemical properties were investigated The HCDR3 length varied from 3 to 21 residues and displayed a biased amino acid content with a predominant proportion of Tyr, Gly, Ser and Asp Repetitive and conserved motifs were observed in the majority of the HCDR3 sequences The strength and efficacy of the HCDR3 libraries reside in the combination of multiple size peptides and a naturally biased sequence variation Therefore, HCDR3 libraries represent a powerful and versatile alternative to fully randomized peptide libraries, in particular for difficult targets

Introduction

Phage display technology allows the handling of large

molecular repertoires in a small and suitable format

with a direct link between the DNA information and

the peptide displayed at the surface of the phage This

technology has become a standard approach for the

identification of strong and specific peptide binders, as

well as for the study of protein–protein interactions

To date, phage libraries displaying fully randomized

peptides of different sizes, structurally constrained or

not, have been successfully applied to a wide variety of targets, such as antibodies, proteins, enzymes and receptors, to identify target-specific peptides [1–8]

In the last decade, engineering of the human antibody repertoire has greatly facilitated the development of therapeutic and diagnostic antibodies The size of the antibodies expressed as libraries on phage has evolved from Fab fragments to single-chain antibody fragments and, finally, to heavy chain variable region

Abbreviations

ECL, extracellular loop; HA, hemagglutinin; HCDR3, heavy chain complementarity determining region 3; UPA, undecapeptidyl arch; V H , heavy chain variable region fragment.

fragments (VH) However, drawbacks in terms of poor

solubility and limited access of these large antibody

fragments to some targets have prompted a further size

reduction of the displayed antibody fragments [9–11]

Recently, we have demonstrated the feasibility of

fur-ther reducing the size of the antibody to the peptide

level by engineering phage libraries displaying heavy

chain complementarity determining region 3 (HCDR3)

repertoires representing human ‘biologically

random-ized peptide collections’ These phage libraries have

been used successfully to identify HCDR3 peptides

displaying nanomolar affinity for the anti-human

influ-enza hemagglutinin (HA) IgG [12]

These HCDR3 phage libraries reflect the natural

HCDR3 diversity mainly achieved by variable,

diver-sity and joining gene segment rearrangements and

ran-dom nucleotide addition This vast potential of

HCDR3 diversity is also associated with the central

role of HCDR3 in the determination of antigen

speci-ficity [13–15] The HCDR3 repertoire complexity also

reflects the developmental changes from fetal to adult

life of the B-cell repertoire [16,17] In the mouse

model, the HCDR3 repertoire evolved during

develop-ment to eliminate outliers in terms of length and

amino acid composition to reach an average

hydro-phobicity [18] Finally, skewed HCDR3 repertoires

reflect perturbed B-cell repertoires associated with

cer-tain pathologies The HCDR3 length distribution of

patients with multiple sclerosis displayed a reduced

complexity compared with the distribution of healthy

donors [19] For other autoimmune diseases, such as

primary biliary cirrhosis and rheumatoid arthritis,

skewed HCDR3 length distributions were observed for

particular VHgene families [20,21]

The importance of HCDR3 loops in antibodies and

their multiple characteristics make these fragments

attractive for repertoire display: (a) HCDR3 is the

smallest part of an antibody retaining antigen-binding

ability [22–24]; (b) HCDR3 is the most diverse region

in length, shape and sequence; (c) HCDR3 is flanked

by conserved framework residues allowing specific

amplification and cloning in different vectors; and (d)

naive and immune HCDR3 repertoires can be easily

isolated from peripheral blood mononuclear cells from

different species In this report, we compared the

effi-ciency of a human HCDR3 library engineered from a

nonimmunized donor with a set of conventional

con-strained and nonconcon-strained libraries displaying

pep-tides of different size These libraries were screened

against a peptide derived from the second extracellular

loop (ECL2) of the CXCR4 receptor implicated in

cancer and HIV infection This loop (residues 176–

202) adopts a short antiparallel b-strand structure at

the surface of the receptor, as shown in the recently resolved three-dimensional structure, and has been reported to be a critical feature for the interaction between the receptor and both of its natural ligands, the chemokine CXCL12 and the HIV-1 envelope pro-tein [25,26] Moreover, epitopes of several CXCR4 neutralizing antibodies are located in this loop, fur-ther emphasizing its important role in ligand binding [27–29]

In parallel with biopanning, we investigated the bio-chemical properties of the natural IgM HCDR3 reper-toire in terms of length and sequence diversity to acquire better insights into the efficiency and advanta-ges of these libraries

Results

Selection with fully randomized peptide phage libraries on ECL2

A total of nine selection campaigns was performed on linear biotinylated ECL2 peptide using two linear (12- and 15-mer) and three constrained (7-, 13- and 14-mer) fully randomized peptide phage libraries Selec-tion strategies were performed on immobilized peptide (plastic support or beads), as well as on soluble ECL2 peptide Increasing stringency was applied in all selec-tion campaigns, i.e decreasing input phage titers and increasing washing steps in consecutive selection cycles Three different elution approaches were tested: acid elution, dithiothreitol elution to recover phage binding

to ECL2 in a constrained format and competitive elu-tion with an undecapeptide corresponding to the N-terminal arch of ECL2 (undecapeptidyl arch, UPA) None of the selection campaigns resulted in positive phage clones, except for the campaign performed with the multivalent f88 Cys1 library on ECL2 peptide immobilized on beads for which four positive clones were obtained Specificity ELISA with these four puri-fied clones on ECL2 and irrelevant peptides revealed that none of these phages was target specific Finally, two selection campaigns with the constrained f88 Cys1 library were performed on cyclic biotinylated ECL2 peptide without success

Selection with an IgM-derived HCDR3 phage library on ECL2

A library displaying constrained HCDR3 loops from the IgM repertoire of a nonimmunized donor, previ-ously used to isolate nanomolar binders to an anti-HA antibody, was screened on ECL2 peptide immobilized

on beads [12] Four selection rounds with increased

stringency, equivalent to the conditions applied to the

fully randomized peptide libraries, were performed

The supernatant of 260 individual clones of the fourth

round was tested by ELISA on ECL2 and on an

irrele-vant peptide (HA mimotope) Analysis revealed 66

(25%) positive clones, 22 of which displayed a signal

on ECL2 peptide higher than 0.5 and an ECL2⁄ HA

ratio > 10 The other 44 clones displayed signals

higher than 0.2 on ECL2 peptide with an ECL2⁄ HA

ratio between 3 and 10 Although, in the first set of 22

clones, a unique sequence 95-DRGGTYPGRY-102

(Kabat numbering) was identified, the second set

com-prised 21 different sequences including

95-DRGGTYP-GRY-102 (Table 1)

Specificity analysis of the 21 purified positive phage

clones revealed that nine (clones 1, 3, 6, 24, 26, 28, 29,

30 and 39) were clearly specific for linear ECL2 and

displayed at least a four-fold higher interaction with

the target peptide than with the HA mimotope

(aver-age fold of 5.8) (Table 1) None of the specific ph(aver-age

interacted with a peptide corresponding to ECL2 of

another chemokine receptor CCR5 The phage clone

isolated at the highest frequency (clone 3) (n = 29)

displayed the strongest interaction with peptide ECL2

in both linear and cyclic format (ECL2lin⁄ HA ratio,

23; ECL2cycl⁄ HA ratio, 9.8) (Fig 1)

Characterization of the antagonistic properties of ECL2-specific clones

Binding of the nine target-specific phage clones (1, 3,

6, 24, 26, 28, 29, 30 and 39) to ECL2, as presented in the native CXCR4 receptor, was analyzed in a receptor activation assay Agonistic properties were first evalu-ated by measuring the effect of the phage on cAMP production None of the ECL2-specific phage clones displayed agonistic properties The antagonistic prop-erties of the selected HCDR3 phage were evaluated by monitoring cAMP production in the presence of the chemokine CXCL12 Four clones (1, 3, 28 and 29) restored the initial forskolin-induced cAMP production (Fig 2A), indicating their recognition and interaction with native receptor Clone 3 (95-DRGGTYPGRY-102), isolated at the highest frequency (n = 29), was further analyzed in free peptide format Analysis of the cyclic peptide corresponding to the HCDR3 sequence

of clone 3 extended with framework 3 and 4 residues (89-VYYCAR-DRGGTYPGRY-WCQG-106) (peptide 3) in cAMP assays demonstrated an antagonistic activ-ity towards CXCR4, characterized by an IC50 of

23 lm (Fig 2B) No inhibition was observed with the irrelevant HA mimotope The specificity for CXCR4

of this sequence in peptide and phage format was

Table 1 Sequence, length, isolation frequency and specificity (ECL2lin⁄ HA ratio) of heavy chain complementarity determining region 3 (HCDR3) clones isolated from the IgMCys library Residues presented in bold are identical in the different sequences CAR- and -WC sequences are part of the framework 3 and 4 residues, respectively ECL2, second extracellular loop; HA, hemagglutinin.

Frequency

ECL2 lin ⁄

HA ratio

confirmed in a similar analysis using CCR5-expressing

cells (CEM.NKR-CCR5) and the chemokine CCL5

The binding affinity of peptide 3 was analyzed by

sur-face plasmon resonance A clear binding was observed

with immobilized ECL2 peptide, whereas no binding

to a truncated ECL2 peptide corresponding to the first

11 residues of ECL2 (i.e UPA) was measured These

data suggest that full-length ECL2 peptide is required

for binding Kinetic analysis using the two-state

reac-tion model revealed a KDvalue of 5.6 lm for peptide 3

(Fig 3A)

To determine the residues of clone 3 critical for

binding to the target, Ala scanning was performed on

the region spanning the HCDR3 sequence including

Cys92 and Cys104 (Fig 3B) The replacement of

Cys92 and Cys104 resulted in 60% and 90% decreases

in binding, respectively In addition, mutation of the

three positively charged residues Arg94, Arg96 and

Arg101, present in the HCDR3 sequence, reduced the

binding by 40%, 39% and 53%, respectively

Interest-ingly, mutation of the residues included in the cluster

between Tyr100 and Trp103 decreased the interaction

with the target by 32–60% Together, these results

indicate the importance of the disulfide bridge, the

positive charges and the variable sequence following

the DRGG⁄ R motif in the selected HCDR3 sequences

for target binding

Length distribution analysis of human IgM HCDR3 repertoires

To unravel the biochemical features of the HCDR3 libraries resulting in very efficient biopanning when com-pared with conventional peptide libraries, we analyzed

Fig 1 Specificity of the most frequently isolated phage clone

(clone 3) for the ECL2 peptide Specificity was determined by

ELISA on immobilized linear (ECL2X4lin) and cyclic (ECL2X4cycl)

ECL2 peptide, linear ECL2 peptide derived from the chemokine

receptor CCR5 (ECL2R5) and three irrelevant peptides (Ctrl HA, Ctrl

2 and Ctrl 3) Two-fold dilutions of phage (starting at 2 · 10 12 phage

per well per 100 lL) were added Phage binding to peptides was

detected using an anti-M13 IgG conjugated to horseradish

per-oxidase The experiment was performed three times and resulted

in equivalent profiles ECL2, second extracellular loop; HA,

hemag-glutinin.

Fig 2 Antagonistic properties of the second extracellular loop (ECL2)-specific phage ⁄ peptide monitored by cAMP modulation assay (A) Antagonistic properties of the target-specific phage were determined by monitoring the inhibition of the CXCL12-induced cAMP suppression Phage (5 · 10 12

particlesÆmL)1) was incubated with forskolin (FSK) and the chemokine CXCL12 (30 n M ) Variation

in the CXCL12-induced cAMP inhibition was detected using time-resolved fluorescence resonance energy transfer technology (B) Antagonistic properties of free peptide derived from clone 3 towards CXCR4–CXCL12 and the CCR5–CCL5 activation pathways Inhibitory properties were expressed as the percentage of inhibition

of the initial CXCL12-induced cAMP signal measured in the absence of peptide Results were compared with those recorded with an irrelevant peptide (hemagglutinin, HA) The data shown here represent the mean of triplicate measurements ± standard deviation.

the complexities of the HCDR3 libraries in terms of

length, amino acid content and position variability

To better characterize the HCDR3 length diversity,

a set of forward and backward primers based on the

nucleotide alignment of 1182 human VHsequences was

used to amplify the complete diversity of the HCDR3

loops of the donor [12,30,31] The PCR products

cor-responding to the amplified IgM HCDR3 repertoire

were separated as a function of their length by

electro-phoresis (Fig 4A) The IgM length distribution profile

was Gaussian-like, with one major peak at 11 amino acids, and the length varied between 3 and 21 amino acids To analyze the quality of the IgM-derived HCDR3 fragments when displayed on phage, the same length analysis was performed on the HCDR3 PCR products after cloning in the phagemid vector (= library) (Fig 4B) The same Gaussian-like pattern, consisting of 19 peaks with the major peak at 11 amino acids, was observed for the HCDR3 repertoire ligated in the phagemid The smallest and longest frag-ments corresponded to 3 and 21 amino acids, respec-tively

The HCDR3 length distribution of the subpopula-tion of phage isolated after each selecsubpopula-tion round was monitored and compared with the length distribution

of the initial library Analysis of the phage population isolated in the fourth selection round clearly showed a selection for the HCDR3 fragments with a length between nine and 14 amino acids and an enrichment

of HCDR3 loops of 18 residues HCDR3 fragments with a length between five and eight were less repre-sented, suggesting that a minimal size of nine is required for efficient interaction with the target (Fig 4C) Indeed, analysis of the individual length of the ECL2-positive HCDR3 loops revealed a length ranging from eight to 19 residues, with an average value of 12 ± 2.4

Analysis of the global amino acid frequency

To further characterize the overall amino acid content

of the IgM HCDR3 library, a set of 128 randomly iso-lated clones from the library was sequenced Analysis

of the global frequency of the 20 individual amino acids of the IgM HCDR3 sequences revealed that Tyr, Asp, Gly and Ser represented more than 45% of the total content (47.6%) The second class of residues for which each amino acid represented 5% of the overall content comprised Ala, Arg, Leu, Phe, Thr and Val The residues Cys, Gln, His and Met were less abun-dant and represented maximally 2.5% of the total con-tent The residue present at the lowest frequency (1%)

in the analyzed set of HCDR3 sequences was Lys (Fig 5A)

Analysis of the individual amino acid frequency

as a function of the HCDR3 length The frequency of each individual amino acid was calcu-lated as a function of the HCDR3 length (Table 2) The frequency of Cys and Lys increased with increasing HCDR3 length (Pearson’s r = 0.864 and 0.922 with two-tailed P values of 0.059 and 0.0259, respectively)

Fig 3 Characterization of the interaction between the heavy chain

complementarity determining region 3 (HCDR3) clone 3 and target

second extracellular loop (ECL2) (A) Kinetic analysis of the binding

of peptide 3 on immobilized biotinylated ECL2 peptide Black

sen-sorgrams represent the experimental data obtained on the ECL2

surface subtracted from the signal recorded on an irrelevant

pep-tide using different concentrations of peppep-tide 3 (31.2–2000 n M ).

Red sensorgrams represent the fitted curves on a two-state

inter-action model (v2= 0.276) using the BIAEVALUATION 4.1 program.

SPR, surface plasmon resonance (B) Ala scanning of the

interac-tion between phage clone 3 (2 · 10 12 phage per well) and the

ECL2 target peptide Data are represented as the percentage of

the initial binding signal recorded with wild-type (WT) phage (clone

3) and correspond to an average of three measurements ±

stan-dard deviation.

Although less significant, a positive correlation was

observed between frequency and HCDR3 length for

Asn, Ala, Gln and Thr (Pearson’s r = 0.534, 0.552,

0.424 and 0.496, respectively; two-tailed P < 0.5) The

frequency of the amino acids Gly, Pro, Phe and Leu

tended to decrease with increasing HCDR3 length

(Pearson’s r =)0.711, )0.532, )0.678 and )0.522,

respectively; two-tailed P < 0.5) A weak negative

cor-relation between frequency and HCDR3 length was

observed for Tyr and Asp (Pearson’s r =)0.350 and

)0.349, respectively; two-tailed P < 0.6) The

fre-quency of Ile tended to increase with increasing

HCDR3 length (Pearson’s r= 0.5884; two-tailed

P < 0.6) The residues His, Arg, Glu and Met displayed a very weak positive correlation (Pearson’s

r = 0.201, 0.217, 0.171 and 0.238, respectively; two-tailed P < 0.8) The frequency of Trp, Ser and Val displayed no correlation with the HCDR3 length (Pearson’s r < 0.05; two-tailed P > 0.9) (Table 2)

Analysis of the position-dependent amino acid variability in the HCDR3 loops

The position-dependent amino acid distribution and variability were analyzed as a function of the HCDR3 loop length (11–15 residues) The last three positions

160 nt

150 nt 180 nt 190 nt

10 aa

120 nt 140 nt 160 nt 180 nt

11 aa

A

120 nt 150 nt 160 nt 180 nt 190 nt

200 nt 120 nt 140 nt 160 nt 180 nt 200 nt

120 nt

Fig 4 Length distribution of human IgM- and IgG-derived heavy chain complementarity determining region 3 (HCDR3) repertoires (A) IgM-derived HCDR3 distribution displaying one major peak at 11 amino acids (aa) (B) Initial IgM HCDR3 distribution displayed on phage (= phage library) (C) Distribution of the HCDR3 fragments selected at the end of the fourth round of the selection campaign (dithiothreitol elution) (D) IgG-derived HCDR3 distribution displaying a skewed Gaussian profile The red profile corresponds to the ROX-labeled (A, D) and DS33-labeled (B, C) size standards expressed in nucleotides (nt) used to calculate the length of the HCDR3 fragments of the profile.

K

Q C H

Y 15%

M E N

P 3% %

I 3%

W 4%

D 12%

T 4%

V 5%

G 12%

R 5% %

A 5%

S 9%

L 5%

F 6%

E

I M Q C W N L

G 25%

A 4%

T 5%

V 6%

S 7%

Y 14%

D 8%

R 11%

P 10%

Fig 5 Frequency of individual amino acids in a set of IgM-derived heavy chain complementarity determining region 3 (HCDR3) sequences (A) Relative amino acid frequency in a set of IgM-derived HCDR3 sequences (n = 128) randomly selected from the initial phage library Tyr, Gly, Ser and Asp represented more than 45% of the total amino acid content (B) Amino acid distribution of the 21 target-positive HCDR3 sequences The predominant amino acids were Gly, Tyr, Arg, Pro and Asp.

at the C-terminal base of the HCDR3 loop (residues

100x to 102) showed limited variability compared with

the other positions The motif FDY was most

com-monly identified at these positions for the IgM-derived

HCDR3 loops varying from 11 to 15 residues When

analyzing the evolution of the amino acid composition

of the last three C-terminal positions as a function of

HCDR3 length, their variability tended to decrease

with increasing HCDR3 length (Pearson’s r =)0.575,

)0.606 and )0.875 with two-tailed P values of 0.310,

0.278 and 0.05, respectively)

Discussion

In this study, we have demonstrated that HCDR3

rep-ertoires displayed on phage can be efficiently used to

identify specific bioactive sequences by targeting ECL2

of the G-protein-coupled receptor CXCR4 Peptide

ECL2 was initially selected as target for its critical role

in the viral envelope protein (gp120) and chemokine

CXCL12 interactions [25] As demonstrated recently,

the binding pocket of CXCR4 is smaller and involves

the extracellular surface to a larger extent when

com-pared with other G-protein-coupled receptors [32]

Eleven selection campaigns with five conventional

fully randomized phage-displayed peptide libraries,

each displaying single size peptides on ECL2, were not

successful These results illustrate the difficulty in

screening random phage-displayed peptide libraries on

a peptide target Indeed, only rare examples have been published on the successful isolation of peptides on a peptide target [33] In contrast, only one selection cam-paign with a constrained human IgM-derived HCDR3 library from a nonimmunized donor was required to isolate four HCDR3 sequences of different lengths (10,

11 and 13 residues) recognizing the native CXCR4 receptor and interfering with the binding of the chemo-kine Among these sequences, at least one peptide cor-responding to the clone isolated at the highest frequency acted as an antagonist (IC50= 23 lm) and displayed an affinity of 5.6 lm A discrepancy in the potency of clone 3 in phage format (nanomolar range activity) and in peptide format (micromolar range activity) was observed These data suggest an impor-tant contribution of the phage scaffold to the binding Avidity effects in the phage format are less probable,

as a phagemid system resulting mainly in a monova-lent display was used

To explain the difference in efficiency between bio-logical (HCDR3) and fully randomized synthetic pep-tide libraries, the properties of the HCDR3 phage library were analyzed in terms of length, amino acid composition and sequence variability

Length analysis of amplified IgM HCDR3 loops revealed a Gaussian-like profile with lengths from three to 21 amino acids The same length profile was obtained for the repertoire of phage-displayed HCDR3 fragments, indicating that no bias occurred on cloning The observed length profiles were in agreement with those reported in the literature on very large sequence panels [34,35] In comparison with HCDR3 sequences

of adult mature B cells (n = 42), varying between six and 23 amino acids, our sequence set also comprised smaller HCDR3 loops [36] HCDR3 length profiles were markedly different between the IgM and IgG rep-ertoires; ideal versus skewed Gaussian-like profiles were observed for IgM and IgG repertoires, respec-tively (Fig 4A, D) In addition, the IgG profile dis-played a reduced range of HCDR3 lengths (6–18 residues) compared with the IgM distribution (3–21 residues) (Fig 4A, D) These findings correlate with the affinity maturation process of the IgM and IgG isotypes IgM repertoires reflect a polyclonal response, whereas IgG repertoires represent oligoclonal expan-sions corresponding to the development of high-affinity antibodies [37] These observations, together with pre-viously reported successful isolation of nanomolar binders from the IgM library, prompted us to use the IgM-derived HCDR3 library [12]

The overall amino acid distribution in our set of HCDR3 sequences revealed a biased content in the

Table 2 Analysis of the frequency of individual amino acids as a

function of the heavy chain complementarity determining region 3

(HCDR3) length.

IgM sequences Tyr, Gly, Ser and Asp were

over-rep-resented (total of 50%), whereas Glu, Lys and Cys

were under-represented (total of 3%) (Fig 5) These

results were in agreement with the analysis on a large

set of HCDR3 sequences retrieved from antibody

data-bases by Zemlin et al [35]

Analysis of individual amino acid contents as a

function of HCDR3 length demonstrated that the Gly

content decreased with increasing HCDR3 length,

whereas the Cys content increased with longer HCDR3

loops, in agreement with data obtained on large sets of

sequences [35] The absence of Cys and the higher

con-tent of Gly in the shorter HCDR3 loops render these

loops more flexible to adapt to a variety of antigens,

whereas an increase in Cys content in the longer

HCDR3 loops could favor the formation of structural

constraints In our dataset, Lys displayed a significant

positive correlation as a function of the HCDR3

length, as observed by Zemlin et al [35] For Phe,

a negative correlation was identified

In addition, HCDR3 repertoires displayed

position-dependent biased sequence diversity Clear limited

sequence diversity was observed at the C-terminal end

of the HCDR3 loops The majority of the HCDR3

sequences displayed the motif FDY at positions 100x–

102 The variability of the C-terminal end of the

HCDR3 loops was negatively correlated with the loop

length Longer HCDR3 loops displayed less variability

at these positions, and these data suggest that the loss

of sequence variability is compensated for by the

longer loop lengths, providing a higher degree of

struc-tural freedom

Remarkably, the majority of HCDR3 sequences

dis-played repetitive motifs of particular amino acids

Doublets of Ser, Gly and Tyr were identified in the

majority of the sequences Higher order repetitions

were only observed with Tyr (quadruplets and

quintu-plets) Studies on protein–protein interactions have

demonstrated the importance of Tyr to obtain high

affinity and specific protein–protein interactions [38]

Gly- and Ser-rich clusters were also frequently

observed in HCDR3 sequences, and probably act as

flexible linkers to separate ‘hot spots’ for binding

Indeed, these Gly⁄ Ser linkers are commonly used in

protein engineering to separate functional domains in

proteins [39,40]

Our HCDR3 sequence analysis clearly demonstrates

that paratopes of antibodies have evolved to yield

biased amino acid contents enriched for Tyr, Gly and

Ser Enrichment of these residues and their differential

repartition in IgM and IgG repertoires indicate that

HCDR3 fragments display remarkable biochemical

properties, particularly suited and efficient for

high-affinity antigen binding As an illustration, minimal Tyr, Ser, Ala and Asp phage libraries were engineered

in vitroand were successfully screened on globular pro-teins [41]

In addition to the advantages inherent to the biochemical properties of the HCDR3 loops, peptides corresponding to HCDR3 sequences selected from phage libraries can be easily produced in large amounts

by solid phase synthesis The results observed with the synthetic peptide VYYCARDRGGTYPGRYWCQG (peptide 3) indicate that this HCDR3 sequence retains binding and antagonistic properties outside the phage format

Together, the efficacy of HCDR3 libraries compared with conventional peptide libraries can be explained by the natural bias selected during evolution, resulting in four levels of complexity: (a) a large variety of peptide lengths; (b) a biased sequence diversity towards amino acids crucial for high affinity and specific binding (Tyr, Gly and Ser); (c) length- and position-dependent sequence bias; and (d) repetitive motifs of Ser, Gly and Tyr In conventional peptide libraries, diversity

is mainly achieved by introducing NNN or NNK codons, resulting in an equal representation of the 20 amino acids at each position However, the natural HCDR3 sequence bias leading to high-efficacy peptide libraries cannot be achieved with this approach

In addition to this natural HCDR3 variability, a supple-mentary level of complexity was added in vitro by engi-neering a disulfide bridge between Cys92 in the framework 3 residues and a Gly to Cys mutation at position 104 in the framework 4 residues, mimicking more closely the parental antibody context

The isolation of four HCDR3 sequences acting as antagonists in phage format, with at least one sequence retaining its biological properties in peptide format, indicates that soluble ECL2 peptide represents a valu-able alternative to native receptor screening on living cells when targeting the extracellular surface implicated

in ligand binding Nevertheless, further optimization and affinity maturation of the four HCDR3 sequences

on complete receptor are required to develop nanomo-lar affinity CXCR4 antagonists

Interestingly, analysis of the amino acid content of the isolated HCDR3 sequences binding to the ECL2 peptide revealed changes, in particular for Arg, Gly, Pro, Asp and Phe, in comparison with the initial reper-toire Arg and Gly contents underwent a two-fold increase, whereas the Pro content was four-fold higher

In contrast, Asp decreased from 12.3% to 7.6%, and Phe was almost absent in positive clones (Fig 5B) The enrichment for Arg (11%) is correlated with the biochemical properties of the target peptide, displaying

a net negative charge of four in its N-terminal part

(four Asp, one Glu and one Arg) In addition,

muta-tion of the three Arg residues in the most frequently

isolated HCDR3 sequence resulted in a loss of binding,

confirming their importance for target interaction This

Arg-rich content is in strong agreement with previous

studies reporting the importance of positive charges

for interaction with CXCR4 extracellular surface and

receptor antagonists, as observed in T22 and NeoR6

[42,43]

In summary, we have demonstrated the value of

nat-urally size and sequence randomized human HCDR3

libraries, in comparison with fully synthetic peptide

libraries, in particular for difficult targets The efficacy

of HCDR3 phage libraries was also observed in

bio-panning with more complex targets, such as enzymes

and viral gp120 protein In addition, HCDR3 libraries

derived from immunized donors were successfully

explored [44] This proof of concept opens up a very

interesting field for drug discovery

Materials and methods

Chemicals and cell lines

Biotinylated peptides were purchased from JPT (Berlin,

Germany) or Bachem (Bubendorf, Switzerland) The

bio-tinylated peptide corresponding to the predicted sequence of

ECL2 of human CXCR4 (176–201) was synthesized in linear

(NVSEADDRYICDRFYPNDLWVVFQFQ) and cyclic

(C-NVSEADDRYICDRFYPNDLWVVFQFQ-C) formats

[25] Biotinylated peptides corresponding to ECL2 of

human CCR5 (167–198)

(GGGTRSQKEGLHYTCSSHF-PYSQYQFWKNFQTLKI) and a non-G-protein-coupled

receptor-related peptide corresponding to a mimotope of

the HA epitope (GGGSPAPERRGYSGYDVPDY) were

used as negative controls [12,45] Cyclic peptide

corre-sponding to the most frequently isolated HCDR3 (clone 3)

(VYYCARDRGGTYPGRYWCQG), elongated with six

(VYYCAR) and four (WCQG) amino acids from the

framework 3 and 4 residues, respectively, was purchased

from Bachem

The cell lines MT-4 expressing CXCR4 and CEM.NKR

-CCR5 were obtained through the AIDS Research and

Ref-erence Reagent Program, Division of AIDS, NIAID,

National Institutes of Health, from D Richman and A

Trkola, respectively

Fully randomized peptide and HCDR3 phage

libraries

Five different fully randomized phage-displayed peptide

libraries were screened For minor coat protein (pIII)

dis-played libraries, linear dodecapeptide (12-mer) (PhD-12)

and disulfide bridge-constrained heptapeptide (7-mer) (CX7C) libraries were purchased from New England Biol-abs (Ipswich, MA, USA) In the CX7C library, the random-ized peptides are flanked by a pair of Cys residues which,

by oxidation during phage assembly, result in the formation

of a disulfide bridge The major coat protein (pVIII) dis-played libraries, linear pentadecapeptide 15-mer (f88) and constrained tridecapeptide Cys1 (X5CXCX5) and tetradeca-peptide Cys4 (X4CX4CX4) libraries, were kindly provided

by G P Smith (University of Missouri, CO, USA) Engineering of the phage library displaying constrained human IgM-derived HCDR3 fragments derived from a nonimmunized donor has been described previously by Deroo et al [12] Briefly, HCDR3 fragments were amplified from VH from a nonimmunized donor In this library, Gly

at position 104 was substituted by a Cys, allowing the for-mation of a disulfide bridge with Cys92 The complexity of the IgMCys library corresponded to 3· 108clones [12]

Affinity selection of fully randomized peptide and HCDR3 phage libraries

Biopanning with the fully randomized peptide phage libraries was performed according to the manufacturer (New England Biolabs) and as described previously [46] Linear or cyclic biotinylated ECL2 peptide (5 nmol) was immobilized on either magnetic Dynabeads or plastic wells (0.5 nmol) coated with streptavidin The elution of specific phage was performed with 75 mm dithiothreitol for the phage libraries displaying cyclic peptides, glycine-HCl solu-tion at pH 2.2 (acid) or by competisolu-tion with free ECL2 peptide or a peptide comprising a 12-mer sequence of ECL2 (UPA)

Biopanning with the IgMCys HCDR3 library was per-formed on linear biotinylated ECL2 peptide (5 nmol) immobilized on magnetic Dynabeads (Invitrogen, Mere-lbeke, Belgium) coated with streptavidin For the first round, beads were incubated with 2· 1012

phage particles for 2 h in 2% milk After washing the beads five times with NaCl⁄ Pi⁄ 0.5% Tween 20, the target-bound phage was first eluted with 500 lL of 75 mm dithiothreitol, and the remain-ing fraction of bound phage was eluted with glycine⁄ HCl buffer, pH 2.2, for 10 min and neutralized with Tris⁄ HCl buffer, pH 8.0 Eluates were used to infect log phase Escherichia coli TG1 cells Phage was rescued with M13K07 helper phage and employed for three additional selection rounds using 1· 1012, 1· 1011and 1· 1010phage particles and 10, 20 and 40 washing steps in rounds 2, 3 and 4, respectively

Screening of positive clones Culture supernatants of phage rescued from the fourth selection round were tested by ELISA on biotinylated ECL2 and a control peptide (10 lgÆmL)1) immobilized via

streptavidin After blocking plates with NaCl⁄ Pi⁄ 2% milk

for 1 h, plates were washed with NaCl⁄ Pi⁄ 0.05% Tween 80

and incubated with culture supernatants in NaCl⁄ Pi⁄ 2%

milk for 2 h After washing with NaCl⁄ Pi⁄ 0.05% Tween 80,

phage binding was detected with an anti-M13 IgG

conju-gated to horseradish peroxidase (GE Healthcare, Diegem,

Belgium) The plates were developed with

ortho-phenyl-enediamine (Sigma-Aldrich, St Louis, MO, USA) and read

at 492 nm

Purified phage was prepared by PEG⁄ NaCl precipitation

from the individual positive supernatant cultures and tested

by ELISA on a panel of biotinylated immobilized peptides

(linear and cyclic ECL2 of CXCR4, linear ECL2 of CCR5,

HA epitope) as described above to confirm target specificity

Length distribution analysis of the amplified

HCDR3 repertoires

HCDR3 fragments corresponding to the IgM and IgG

rep-ertoires of a nonimmunized donor were PCR amplified with

the 6-carboxyfluorescein (FAM)-labeled pool of backward

primers and a pool of forward primers, as described

previ-ously [12] A total of 1 lL of labeled PCR product was

mixed with 12 lL of HiDi formamide (Applied Biosystems,

Nieuwerkerk a⁄ d Ijssel, the Netherlands) and 0.5 lL of

ROX and DS33 size standards (Applied Biosystems) The

fluorescent HCDR3 fragments were separated by

electro-phoresis on an ABI3100 capillary sequencer, and their

lengths (Kabat positions 95–102) were calculated using the

ROX and DS33 size standards

Analysis of amino acid sequences

HCDR3 loops of randomly picked colonies of the amplified

IgM HCDR3 repertoire of the nonimmunized donor were

sequenced on an ABI3100 capillary sequencer using the

BigDye Terminator Cycle Sequencing Ready Reaction Kit

v3.1 (Applied Biosystems)

cAMP assay

The inhibition of primary intracellular cAMP production

induced by binding of the chemokine CXCL12 to CXCR4

was evaluated in the presence of the ECL2-specific HCDR3

phage clones using time-resolved fluorescence resonance

energy transfer LANCE cAMP assay (Perkin-Elmer,

Waltham, MA, USA) adapted for a 96-well plate format

MT-4 cells expressing CXCR4 and CEM.NKR.CCR5 cells

(National Institutes of Health AIDS Program) were

har-vested, washed with simulation buffer [Hank’s buffered salt

solution (1· ) containing 5 mm Hepes, 0.1% BSA, 0.5 mm

3-Isobutyl-1-methylxanthine (IBMX), pH 7.4] and

resus-pended in simulation buffer containing Alexa Fluor

647-labeled antibodies (1 : 100 dilution) Cells (20 000 per

well) were incubated for 30 min at room temperature with

2· forskolin CXCL12 per phage or RANTES per phage (30 nm and 5· 1012 phageÆmL)1) Substrate containing Eu-W8044-labeled streptavidin and biotin–cAMP was added and incubated for 1 h at room temperature The LANCE signal was recorded at 665 nm and compared with cAMP standard curves (10)6–10)11m) Experiments were per-formed in triplicate

Surface plasmon resonance Biotinylated ECL2 and UPA were immobilized on a strep-tavidin chip (GE Healthcare) by injecting peptide (1 lm) at

a flow rate of 5 lLÆmin)1 (10 min) in 0.01 m Hepes,

pH 7.4, containing 0.15 m NaCl, 3 mm ethylenediaminetet-raacetic acid (EDTA) and 0.005% (v⁄ v) surfactant P20 (HBS-EP) on a BIAcore 3000 (GE Healthcare, Diegem, Belgium) Typically, a signal ranging from 1000 to 1500

RU was obtained Kinetic analysis was performed by inject-ing peptide 3 at various concentrations (30–2000 nm) at a flow rate of 70 lLÆmin)1 Regeneration of the surface was performed by a single injection of 15 lL of 10 mm glycine,

pH 2 For all sensorgrams, the signal obtained on a control surface (irrelevant peptide) was subtracted from the signal obtained on the relevant surface The presence of mass transfer phenomena was excluded by performing the con-trol assays, as recommended by BIAcore Kinetic data anal-ysis was performed using biaevaluation 4.1 software employing a two-state reaction model in agreement with the presence of linked reactions

Ala scanning Ala mutations were introduced at each position of the HCDR3 sequence of clone 3 (positions 92–104) by overlap-ping PCR using a set of specific primers Mutated frag-ments were digested by BglI and NotI and cloned into the phagemid vector

Sequence analysis was performed to ensure the presence

of the mutation, and all individual Ala mutated clones were amplified and tested at a single concentration (2· 1012

phage per well) in phage-ELISA on ECL2 peptide The sig-nal recorded with each Ala mutant was compared with the signal recorded with wild-type phage clone 3 and irrelevant phage (HA) Experiments were performed in triplicate

Acknowledgements

This study was financially supported by the ‘Fonds National de la Recherche’, Luxembourg [BIOSAN⁄ 07 ⁄

19 (PEPSAR project) and C09⁄ BM ⁄ 20 (MIMOKINE project)] and the ‘Centre de Recherche Public-Sante´’, Luxembourg (grant SAN⁄ 03 ⁄ 00, 20070115) The authors thank Charle`ne Verschueren for technical assistance