doi:10.1016/j.jmb.2005.04.049
`
`JMB
`
`Available online at www.sciencedirect.com
`
`SCIENCE@DIRECT 8
`
`J. Mol. Biol. (2005) 350, 126–144
`
`ELSEVIER
`
`Ultra-potent Antibodies Against Respiratory Syncytial
`Virus: Effects of Binding Kinetics and Binding Valence
`on Viral Neutralization
`
`Herren Wu1*, David S. Pfarr1, Ying Tang2, Ling-Ling An1, Nita K. Patel1
`Jeffry D. Watkins2, William D. Huse2, Peter A. Kiener1 and
`James F. Young1
`
`1MedImmune, Inc., One
`MedImmune Way
`Gaithersburg, MD 20878
`USA
`2Applied Molecular Evolution
`3520 Dunhill Street, San Diego
`CA 92121, USA
`
`*Corresponding author
`
`Introduction
`
`We describe here the selection of ultra-potent anti-respiratory syncytial
`virus (RSV) antibodies for preventing RSV infection. A large number of
`w
`antibody variants derived from Synagis
`(palivizumab), an anti-RSV
`monoclonal antibody that targets RSV F protein, were generated by a
`directed evolution approach that allowed convenient manipulation of the
`binding kinetics. Palivizumab variants with about 100-fold slower
`dissociation rates or with fivefold faster association rates were identified
`and tested for their ability to neutralize virus in a microneutralization
`assay. Our data reveal a major differential effect of the association and
`dissociation rates on the RSV neutralization, particularly for intact
`antibodies wherein the association rate plays the predominant role.
`Furthermore, we found that antibody binding valence also plays a critical
`role in mediating the viral neutralization through a mechanism that is
`likely unrelated to antibody size or binding avidity. We applied an iterative
`mutagenesis approach, and thereafter were able to identify palivizumab
`Fab variants with up to 1500-fold improvement and palivizumab IgG
`variants with up to 44-fold improvement in the ability to neutralize RSV.
`These anti-RSV antibodies likely will offer great clinical potential for RSV
`immunoprophylaxis. In addition, our findings provide insights into
`engineering potent antibody therapeutics for other disease targets.
`q 2005 Elsevier Ltd. All rights reserved.
`
`Keywords: affinity maturation; kinetics manipulation; respiratory syncytial
`w
`virus; Synagis
`; palivizumab
`
`Respiratory syncytial virus (RSV), a pneumovirus
`of the family Paramyxoviridae, is the most common
`cause of serious lower respiratory tract disease in
`infants and young children.1 In adults, RSV
`predominantly causes a relatively mild upper
`respiratory tract disease. However, RSV infection
`in elderly adults and immunocompromised
`
`Abbreviations used: RSV, respiratory syncytial virus;
`Fab, antigen-binding fragment; CDR, complementarity-
`determining region; FR, framework; kon, association rate
`constant; koff, dissociation rate constant; Kd, equilibrium
`dissociation constant; ELISA, enzyme-linked
`immunosorbent assay; BSA, bovine serum albumin; PBS,
`phosphate-buffered saline.
`E-mail address of the corresponding author:
`wuh@medimmune.com
`
`0022-2836/$ - see front matter q 2005 Elsevier Ltd. All rights reserved.
`
`individuals may induce a severe lower respiratory
`tract disease.2,3 Since the discovery of the virus in
`the 1950s,
`the on-going efforts to develop an
`effective and safe vaccine have suffered serial
`setbacks due to safety issues4,5 and poor efficacy.6,7
`In contrast, passive immunization with neutralizing
`antibodies has been proven effective and safe. A
`polyclonal RSV hyperimmune globulin (IVIg),
`w
`RespiGam
`, purified from pooled human sera was
`approved in 1996 by the FDA for the prophylaxis of
`RSV in high-risk infants, such as those with
`bronchopulmonary dysplasia or with a history of
`w
`premature birth. In 1998, Synagis
`(palivizumab),
`was approved for
`similar
`indications. This
`humanized monoclonal antibody is directed against
`an epitope in the A antigenic site of the RSV F
`protein and shows significantly improved potency
`w
`and ease of administration over RespiGam
`. In the
`IMpact-RSV study,
`in which high-risk infants
`
`

`

`Ultra-potent Anti-RSV Antibodies
`
`127
`
`received intramuscular injections of palivizumab at
`15 mg/kg every 30 days for five rounds, hospital-
`izations resulting from RSV decreased by 55% in the
`palivizumab treatment group (10.6% placebo versus
`4.8% palivizumab).8 In our current study, we have
`developed an even more potent prophylactic
`antibody than palivizumab in an effort to further
`reduce the hospitalization rate. Such an antibody
`could offer better prophylaxis, and may expand
`indications.
`The biological activity of an antibody is strongly
`influenced by its binding activity. Once an antibody
`has been identified that binds to a critical epitope of
`a target, an increase in the binding ability often
`results in higher potency9–12 or better targeting.13,14
`For example, improvement in the affinity of an
`antibody fragment against anthrax toxin from
`63 nM to 0.25 nM in equilibrium dissociation con-
`stant
`(Kd)
`led to enhanced protection against
`anthrax toxin challenge in an in vitro cell culture
`assay and in a rat model.9 Furthermore, affinity
`maturation of an avb3-specific humanized antibody
`has resulted in better ligand blockade,12 and the
`affinity-improved antibody is currently in phase II
`human clinical trials for the treatment of solid
`tumors. Based on these positive outcomes resulting
`from high binding ability, we decided to explore the
`potential of improving the potency of palivizumab
`by affinity maturation.
`Many approaches have been established to
`improve antibody affinity by introducing beneficial
`
`regions,
`into entire variable (V)
`mutations
`mostly on complementarity determining regions
`(CDRs).11,12,15–20 Although antibody affinity is
`mediated by both dissociation (koff) and association
`rate (kon), the greatest affinity improvements result-
`ing from these approaches have been driven mainly
`by reducing the koff.11,12,15–20 For examples,
`reducing the koff of a monoclonal antibody
`improved its in vitro HIV-1 neutralization,11 and
`similarly in an anti-avb3 antibody and an anti-CD40
`antibody, koff reduction led to better ligand block-
`ing.12,20 However, there is very little information
`regarding the role of kon in antibody potency. This
`may be due to the difficulty in enhancing kon
`because of the lack of good approaches for selecting
`high kon clones. In the course of optimizing the
`affinity of palivizumab, we have developed a robust
`approach that allows the easy manipulation of both
`koff and kon, and we have observed differential
`effects of koff and kon on antibody potency in a
`neutralization assay.
`Increasing the affinity to
`F protein by reducing antibody koff translated very
`well into higher RSV neutralization ability for Fab
`fragments; however, it did not work well for full-
`length antibodies, our ideal therapeutic format. In
`contrast, raising the affinity by increasing kon
`resulted in a great improvement in virus neutraliz-
`ation for both Fab and IgG forms. This highlights
`the importance of kon in RSV neutralization. In
`addition, our study has shown that bivalent binding
`0
`to F protein, in either the IgG or F(ab
`)2 format,
`
`VH Domain
`
`40
`ab
`30
`20
`10
`1
`QVTLRESGPALVKPTQTLTLTCTFSGFSLS TSGMSVG WIRQPPGKALEWLA
`
`Palivizumab
`493L1FR
`AFFF
`A4b4
`
`110
`100
`90
`80 abc
`70
`60
`50
`DIWWDDKKDYNPSLKS RLTISKDTSKNQVVLKVTNMDPADTATYYCAR SMITNWYFDV WGAGTTVTVSS
`Q
`Q
`Q
`
`VL Domain
`
`50
`40
`25 29
`20
`10
`1
`DIQMTQSPSTLSASVGDRVTITC KCQLSVGYMH WYQQKPGKAPKLLIY DTSKLAS
`SASS
`SASS
`SASSR
`
`Palivizumab
`493L1FR
`AFFF
`A4b4
`
`100
`80
`70
`90
`60
`GVPSRFSGSGSGTEFTLTISSLQPDDFATYYC FQGSGYPFT FGGGTKLEIK
`V
`V
`V
`
`Figure 1. Amino acid sequence comparison of the variable regions of palivizumab, 493L1FR, AFFF, and A4b4. CDR
`regions as defined by Kabat are in italics. Mutations decreasing koff are labeled in gray, and mutations increasing kon are
`underlined.
`
`

`

`128
`
`Ultra-potent Anti-RSV Antibodies
`
`confers a substantial benefit in viral neutralization
`over monovalent binding by Fab. The benefit is
`likely not simply related to the size or increase in
`avidity of the antibody.
`
`Results
`
`Further humanization of palivizumab and
`restoration of its light chain CDR1
`
`Prior to affinity maturation of palivizumab, a few
`modifications on the antibody were made. Amino
`acids KCQL, at positions 24 through 27 of the light
`chain CDR1 (LCDR1), were changed to the original
`murine monoclonal antibody 1129 sequence, SASS.
`The KCQL sequence represents four random, non-
`human, non-mouse residues that were introduced
`by a synthetic error during the previous humaniza-
`tion process.21 In addition, we replaced the murine
`residues on the framework (FR) 4 regions with
`human residues to reduce the possibility of
`immunogenicity. An amino acid substitution,
`A105Q, was made in the heavy chain FR4 to make
`
`a fully human JH6 germline sequence; an L104V
`substitution was made in the light chain FR4 to
`make a fully human JK4 germline sequence. The
`resulting clone, 493L1FR, contains fully human FR
`sequences (Figure 1) and was expressed by a
`bacteriophage expression vector. Binding analysis
`of
`the 493L1FR Fab and palivizumab Fab by
`surface plasmon resonance using a BIAcore bio-
`sensor showed that both molecules bound RSV F
`protein with similar kinetics (Table 1). This result
`suggested that contrary to the earlier prediction
`based on structural modeling21 neither murine
`residue A105 on heavy chain FR4 nor L104 on
`light chain FR4 is involved significantly in F protein
`binding. Similarly, alteration of the first four LCDR1
`residues to SASS does not substantially affect
`binding.
`
`koff-driven affinity maturation
`
`An established directed evolution approach12
`was used to improve the affinity of 493L1FR for
`the RSV F protein. The 493L1FR Fab was subjected
`to focused mutations at each residue in each of the
`
`Table 1. Kinetics and viral neutralization of koff-improved antibodies
`
`Sequence
`
`H3
`100
`
`W
`
`–
`
`–
`–
`F
`–
`–
`–
`–
`–
`
`F
`F
`F
`F
`F
`F
`F
`F
`
`L2
`52
`
`S
`
`–
`
`–
`–
`–
`F
`Y
`–
`–
`–
`
`F
`F
`F
`–
`F
`F
`F
`Y
`
`Fab
`
`kon
`(! 105),
`
`K1 sK1
`M
`
`koff
`
`(! 10K4),
`K1
`s
`
`1.26
`1.19
`1.85
`
`1.96
`n.d.
`1.65
`2.06
`1.70
`1.63
`1.62
`1.50
`
`1.34
`1.22
`1.10
`1.13
`1.33
`n.d.
`n.d.
`n.d.
`
`6.62
`7.22
`6.51
`
`0.93
`n.d.
`0.84
`1.75
`1.25
`1.74
`1.53
`1.40
`
`%0.05e
`%0.05e
`%0.05e
`%0.05e
`%0.05e
`n.d.
`n.d.
`n.d.
`
`Microneutralization (IC50),
`mg/ml (nM)a
`
`Kd, nM
`
`Fab
`
`IgG
`
`5.25
`6.07
`3.52
`
`0.47
`n.d.
`0.51
`0.85
`0.74
`1.07
`0.94
`0.93
`
`%0.037
`%0.041
`%0.045
`%0.044
`%0.038
`n.d.
`n.d.
`n.d.
`
`27.46 (549.2)
`
`0.453 (3.02)
`
`26.30 (526.0)
`
`n.d.
`
`4.85 (97.0)
`n.d.
`2.60 (52.0)
`7.27 (145.4)
`5.99 (119.8)
`8.84 (176.8)
`6.26 (125.2)
`6.57 (131.4)
`
`0.0715 (1.43)
`0.0754 (1.51)
`n.d.
`0.0908 (1.82)
`0.249 (4.98)
`n.d.
`n.d.
`n.d.
`
`0.465 (3.10)
`n.d.
`0.876 (5.84)
`n.d.
`n.d.
`n.d.
`n.d.
`n.d.
`
`0.306 (2.04)
`0.407 (2.71)
`n.d.
`0.521 (3.47)
`0.453 (3.02)
`n.d.
`n.d.
`n.d.
`
`L3
`93
`
`G
`
`–
`
`–
`–
`–
`–
`–
`F
`Y
`W
`
`F
`Y
`F
`F
`–
`Y
`W
`F
`
`Clone
`Kabat position
`
`H1
`32
`
`–
`
`S
`
`Palivizumab
`Palivizumabb
`493L1FR
`Single mutations
`A
`S32A
`S32Pc
`P
`–
`W100F
`–
`S52F
`–
`S52Y
`–
`G93F
`–
`G93Y
`–
`G93W
`Combinatorial mutations
`AFFFd
`A
`AFFYd
`A
`PFFF
`P
`AFSFd
`A
`AFFGd
`A
`PFFYc
`P
`PFFWc
`P
`PFYFc
`P
`
`n.d., not determinded; H1, HCDR1; H3, HCDR3; L2, LCDR2; L3, LCDR3.
`a For comparison purpose, the IC50 values were converted to nM and are shown in parenthesis.
`b This palivizumab Fab was prepared by papain cleavage of palivizumab IgG. Other palivizumab Fab used here were made by
`recombinant phage expression.
`c S32P was just a moderate beneficial mutation when compared with other single mutations by ELISA titration. It was therefore not
`further characterized by surface plasma resonance. Similarly for combinatorial variants, only the best five variants judged by ELISA
`titration were further characterized by surface plasma resonance, and PFFY, PFFW and PFYF were not among them.
`d The kinetics of these combinatorial variants in IgG format were also characterized by surface plasma resonance. Similarly to what
`K6 because they have reached beyond the measurement limitation
`were observed in their Fab formats, all of their koff values are %5!10
`
`K6 sK1), and could not be measured accurately. A typical example is shown in Figure 8(b) for the global
`of BIAcore 3000 biosensor (5!10
`fitting analysis of AFFF IgG. The kon values of these variants are: AFFF, 1.27!105; AFFY, 1.44!105; AFSF, 1.47!105; AFFG, 1.47!105.
`
`K6 sK1),
`e The koff value of these combinatorial clones reached beyond the measurement limitation of BIAcore 3000 biosensor (5!10
`and could not be measured accurately.
`
`

`

`Ultra-potent Anti-RSV Antibodies
`
`129
`
`six CDR regions. Separate libraries for each CDR
`were generated using a modified codon-based
`mutagenesis approach that consists of a codon
`doping strategy that allows the segregation of
`diversity into pools based on the degree of
`mutagenesis.22,23 Each CDR library was constructed
`to contain all possible single mutations at each CDR
`residue. These focused libraries, containing 140 to
`320 variants, allowed us to explore easily the
`potential affinity improvements in all possible
`amino acids at every CDR position.
`M13 plaques expressing 493L1FR Fab variants
`were screened for increased affinity to F protein,
`first by a filter-based capture lift method,24 and
`second by a semi-quantitative ELISA assay.25 The
`improved affinity of the identified clones was
`confirmed by an ELISA titration on immobilized F
`protein. DNA sequencing of the affinity-enhanced
`clones revealed eight distinct beneficial mutations
`at four CDR positions: S32A and S32P at heavy
`chain CDR1 (HCDR1), W100F at heavy chain CDR3
`(HCDR3), S52F and S52Y at
`light chain CDR2
`(LCDR2), and G93F, G93Y and G93W at light chain
`CDR3 (LCDR3) (Figure 2(a)). To help visualize the
`three-dimensional positions of the CDR residues
`important for koff or kon (to be discussed later) and to
`assist with comparison of their relative locations,
`these beneficial positions are shown in a molecular
`model26 based on the crystal structure of the
`palivizumab Fab (Figure 3). Analysis by BIAcore
`biosensor of seven of these mutants showed a three
`to sevenfold improvement in affinity compared
`with the 493L1FR Fab (Table 1). The affinity
`improvement was mainly driven by a lower koff.
`The best single mutation, S32A had a Kd at 0.47 nM,
`while palivizumab Fab had at Kd at 5.25 nM.
`During this analysis we did not identify any
`significant mutations in heavy chain CDR2
`(HCDR2) or LCDR1 that were beneficial. This may
`imply that both HCDR2 and LCDR1 are either not
`involved in significant binding or that their binding
`to F protein had already been optimized in vivo in
`the mouse during immunization and B cell selec-
`tion. For LCDR1 residues, it is likely that their role
`in binding to F protein is not significant, since the
`alteration of the first four LCDR1 residues out of a
`total of nine residues (from KCQL to SASS) does not
`affect the binding significantly. However, we cannot
`rule out the possibility that HCDR2 and LCDR1
`may still play minor roles in binding, since we set
`our screening threshold sufficiently high so that
`only clones with a substantial increase in affinity
`would be identified and selected for further
`characterization.
`Indeed, we did identify and
`discard two very minor beneficial mutations,
`A25L and S27V, in LCDR1 (data not shown). The
`A25L mutation was later identified again in a kon-
`biased screening approach (Table 2).
`A combinatorial library combining these eight
`beneficial mutations was constructed by site-
`directed mutagenesis using degenerate oligo-
`nucleotides. Plaque
`lifts
`that detected the
`expression of the kappa light chain and a deca-
`
`(a)
`
`HCDRl
`
`T@GM S V G
`A
`p
`
`HCDR3
`
`S M IT N@Y FD V
`F
`
`LCDR2 DT@KLAS
`F
`y
`
`LCDR3
`
`FQGS@Y P F T
`F
`y
`w
`
`(b)
`
`HCDRl T@GMSVG
`p
`
`HCDR2
`
`D I w w D@ K@@ Y N P S L K@
`G G H
`D
`s
`
`HCDR3 @M Icr}N@Y F D V
`F w
`D
`
`LCDRl @@s S@VGYMH
`R
`L L
`p
`F
`
`LCDR2
`
`D T@@@@s
`Y G H SP
`RR Q KT
`MY
`RD
`H
`F
`
`LCDR3
`
`F Q G S@Y P F T
`G
`
`Figure 2. Beneficial koff and kon mutations (highlighted
`in bold). (a) Single mutations in 493L1FR that result in
`increased affinity to F protein due to the reduction in koff.
`(b) Single mutations in AFFF, the best koff-improved
`palivizumab variant, that result in increased affinity to F
`protein due to the increase in kon. AFFF contains four
`beneficial koff mutations, which are circled in gray.
`
`peptide tag fused at the end of the heavy chain CH1
`indicated that w27% of the combinatorial library
`clones express Fab. Sequencing of the DNA of 25
`random functional clones showed that the distri-
`bution of the majority of the mutations was as
`expected, except that S52Y in LCDR2, S32A in
`HCDR1, and W100F in HCDR3 were potentially
`under-represented. A capture lift screening of
`R2400 clones followed by screening by ELISA led
`to the identification of 48 variants that had
`higher affinity than clone S32A, the best single-
`mutation variant. Further characterization by
`antigen titration and DNA sequencing revealed
`20 unique combinatorial variants. Titrations of
`antigen showed that combinatorial variants have
`
`

`

`130
`
`Ultra-potent Anti-RSV Antibodies
`
`single mutation, S32A, was under-represented in
`the combinatorial library, we decided to incorporate
`this mutation with the combinatorial information
`derived from the four best clones, PFFF, AFSF,
`AFFG, and PFFY (Table 1). This led to the construc-
`tion of clones AFFF (Figure 1) and AFFY by site-
`directed mutagenesis. Both clones had a very high
`affinity, comparable to the four best clones; this
`suggests that the combinatorial
`library was not
`screened thoroughly enough to pick up the under-
`represented clones, even though many redundant
`clones were identified during the screening. The
`eight best variants are listed in Table 1. BIAcore
`analysis of the five best variants showed that their
`affinity was more than 117-fold higher than that of
`the palivizumab Fab, and the affinity increase arises
`from the koff improvement (Table 1). Clone AFFF
`Fab has a Kd at %0.037 nM, while palivizumab Fab
`has a Kd at 5.25 nM. It should be recognized that
`these combinatorial palivizumab Fab variants bind
`so tightly to the immobilized F protein on the sensor
`chip that an accurate dissociation rate could not be
`determined (the koff detection limit for BIAcore 3000
`
`K6 sK1).
`is 5!10
`To verify the binding specificity of these variants
`with improved koff, clones S32A, AFFF, AFFY, PFFF,
`AFSF, AFFG, and PFFY in periplasmic extracts were
`tested in ELISA for binding to the F protein in
`competition with palivizumab IgG. All variants
`tested competed with palivizumab and their ability
`to compete correlated with their affinity. Typical
`inhibition curves are shown in Figure 4(b).
`
`Functional characterization of koff-improved
`palivizumab variants
`
`We used microneutralization of RSV as the
`primary assay to screen the palivizumab variants
`for improvement of biological function.21,27 This
`assay has been used successfully to screen donors
`for RSV IVIg and yielded very few false positives.28
`Analysis by microneutralization of the purified
`palivizumab combinatorial Fab variants with
`improved koff showed a 110 to 384-fold greater
`potency than recombinant palivizumab Fab (Table 1
`and Figure 5(a)). Among both the single-mutation
`and combinatorial Fab variants, we observed an
`excellent correlation between their affinities and
`their ability to neutralize RSV in vitro (Table 1 and
`Figure 6(a)).
`Based on the affinity to F protein and the ability to
`neutralize virus, the two best single-mutation Fab
`variants, S32A and W100F, and the four best
`combinatorial Fab variants, AFFF, AFFY, AFSF,
`and AFFG, were converted to intact IgG1 antibodies
`and expressed in NS0 cells. These purified, full-
`length antibodies were tested in the micro-
`neutralization assay and to our surprise there was
`little to no increase in the in vitro potency when
`compared to intact palivizumab (Table 1 and
`Figure 5(b)). It should be noted that the micro-
`neutralization data of the combinatorial variants in
`Table 1 are averages from at least two independent
`
`Figure 3. Beneficial koff and kon positions in a three-
`dimensional structure of palivizumab V region model
`based on a crystal structure of palivizumab Fab. koff
`positions are coded in light blue, and kon positions are
`coded in red. Light chain V region is in green, and heavy
`chain V region is in blue. It should be noted that all four
`positions that yielded improvements in koff could also be
`mutated to improve kon. (a) Top view (the structure is
`oriented such that the antigen binding region is facing the
`reader). (b) Side view. The Figures were drawn using
`Swiss-Pdb Viewer.39 The coordinates of the crystal
`structure of the palivizumab Fab were provided by
`Bradford C. Braden at
`the Department of Natural
`Sciences, Bowie State University, Bowie, MD 20715, USA
`(data not published).
`
`S32A
`over
`enhanced affinity
`significantly
`(Figure 4(a)). The variants each contain two to
`four beneficial mutations, and there is a loose
`correlation between the affinity and the number of
`beneficial mutations (data not shown). The best
`clones contain a W100F mutation in HCDR3; no
`other obvious pattern was observed. Since the best
`
`

`

`aCloneAFFFisthebestcombinatorialvariantfromkoff-drivenaffinitymaturationofpalivizumabintermsofaffinityandtheabilitytoneutralizevirus.Itwasusedasastartingtemplateforkon
`
`bSubstantiallymorecombinatorialmutantswereidentified.ThisTablelistsonlythetop17variantsbasedonkonimprovement.
`
`mutagenesis.
`
`H
`
`H
`
`H
`
`H
`
`D
`
`H
`
`H
`
`H
`
`D
`
`H
`
`H
`
`H
`
`H
`
`D
`
`D
`
`D
`
`D
`
`H
`
`58
`
`H2
`
`K
`
`K
`
`G
`
`K
`
`K
`
`K
`
`K
`
`K
`
`K
`
`K
`
`K
`
`K
`
`K
`
`K
`
`K
`
`K
`
`K
`
`K
`
`K
`
`K
`
`K
`
`G
`
`57
`
`G
`
`G
`
`G
`
`D
`
`D
`
`G
`
`D
`
`G
`
`D
`
`D
`
`D
`
`G
`
`D
`
`G
`
`G
`
`D
`
`G
`
`D
`
`D
`
`D
`
`D
`
`G
`
`55
`
`A
`
`A
`
`A
`
`A
`
`A
`
`A
`
`A
`
`A
`
`A
`
`A
`
`A
`
`S
`
`S
`
`A
`
`A
`
`P
`
`32
`
`H1
`
`Singlemutations
`D95/G93
`AFFFa
`493L1FR
`Palivizumab
`
`Kabatno.
`
`CDRs
`
`A
`
`A
`
`A
`
`P
`
`P
`
`P
`
`P12f4
`P12f2
`P11d4
`A17h4
`A17f5
`A17d4
`A17b5
`A16b4
`A14a4
`A13c4
`A13a11
`A12a6
`A8c7
`A4b4
`A3e2
`A1h5
`A1e9
`Combinatorialmutationsb
`
`Table2.Summaryofbeneficialkonmutations
`
`W
`
`W
`
`F
`
`F
`
`W
`
`W
`
`W
`
`W
`
`F
`
`W
`
`F
`
`W
`
`F
`
`W
`
`W
`
`W
`
`F
`
`W
`
`W
`
`W
`
`F
`
`F
`
`T
`
`T
`
`T
`
`T
`
`F
`
`F
`
`F
`
`F
`
`F
`
`F
`
`F
`
`F
`
`F
`
`T
`
`F
`
`F
`
`F
`
`F
`
`F
`
`F
`
`F
`
`F
`
`24
`
`100
`
`98
`
`H3
`
`D
`
`D
`
`D
`
`D
`
`D
`
`D
`
`D
`
`D
`
`D
`
`D
`
`D
`
`D
`
`D
`
`D
`
`D
`
`D
`
`D
`
`S
`
`S
`
`S
`
`D
`
`D
`
`95
`
`D
`
`D
`
`D
`
`D
`
`D
`
`D
`
`D
`
`D
`
`D
`
`D
`
`D
`
`D
`
`D
`
`D
`
`D
`
`D
`
`D
`
`S
`
`S
`
`S
`
`S
`
`D
`
`65
`
`S
`
`S
`
`S
`
`S
`
`S
`
`G
`
`G
`
`G
`
`G
`
`G
`
`G
`
`G
`
`G
`
`G
`
`G
`
`G
`
`G
`
`G
`
`G
`
`G
`
`G
`
`G
`
`G
`
`G
`
`F
`
`G
`
`G
`
`93
`
`L3
`
`A
`
`A
`
`A
`
`A
`
`S
`
`K
`
`R
`
`H P T D
`
`S
`
`R
`
`H
`
`D
`
`S
`
`S
`
`S
`
`S
`
`T
`
`A
`
`P
`
`S
`
`T
`
`A
`
`A
`
`S
`
`P
`
`55
`
`L
`
`H
`
`L
`
`L
`
`Q
`
`L
`
`H
`
`Q
`
`Q
`
`Q
`
`L
`
`Q
`
`H
`
`L
`
`L
`
`L
`
`L
`
`L
`
`L
`
`L
`
`L
`
`H
`
`Q
`
`54
`
`L2
`
`K
`
`K
`
`K
`
`K
`
`R
`
`Y
`
`F
`
`K
`
`F
`
`Y
`
`F
`
`Y
`
`K
`
`R
`
`Y
`
`Y
`
`Y
`
`Y
`
`Y
`
`R
`
`Y
`
`R
`
`Y
`
`G
`
`G
`
`53
`
`F
`
`F
`
`F
`
`F
`
`R
`
`F
`
`Y
`
`M
`
`Y
`
`M
`
`Y
`
`M
`
`F
`
`Y
`
`M
`
`F
`
`R
`
`S
`
`S
`
`F
`
`F
`
`Y
`
`R
`
`M
`
`52
`
`R
`
`S
`
`S
`
`R
`
`R
`
`R
`
`R
`
`R
`
`R
`
`R
`
`R
`
`R
`
`R
`
`S
`
`R
`
`R
`
`R
`
`S
`
`S
`
`S
`
`S
`
`R
`
`F
`
`29
`
`L
`
`L
`
`A
`
`A
`
`P
`
`A
`
`P
`
`L
`
`L
`
`L
`
`L
`
`P
`
`L
`
`P
`
`P
`
`L
`
`L
`
`C
`
`A
`
`A
`
`A
`
`L
`
`P
`
`25
`
`L1
`
`K
`
`S
`
`S
`
`S
`
`L
`
`S
`
`S
`
`S
`
`S
`
`S
`
`S
`
`S
`
`S
`
`L
`
`L
`
`S
`
`L
`
`S
`
`S
`
`S
`
`S
`
`S
`
`

`

`(a)
`
`132
`
`0 co
`in
`<(
`
`1.0
`
`0.8
`
`0.6
`
`0.4
`
`0.2
`
`0.0
`0.0001
`
`0.001
`
`0.3 (b)
`
`0.25
`
`0.2
`
`0.15
`
`0.1
`
`0.05
`
`0 co
`in
`<(
`
`0
`1E-04 0.001
`
`100
`10
`0.1
`0.01
`Fab Molar Ratio
`(palivizumab/variant)
`
`1000 10000
`
`Ultra-potent Anti-RSV Antibodies
`
`1.8 (c)
`
`0
`in
`"If'
`<(
`
`1.6
`1.4
`1.2
`
`0.8
`0.6
`0.4
`0.2
`0
`0.0001
`
`0.01
`0.1
`Fab (1,19/ml)
`
`10
`
`100
`
`0.001
`
`0.01
`
`0.1
`
`10
`
`100
`
`Fab (1,19/ml)
`
`0.4 (d)
`
`0
`in
`"If'
`<(
`
`0.3
`
`0.2
`
`0.1
`
`0
`0.01
`
`• • • • • • • •
`
`0.1
`
`100
`10
`Fab Molar Ratio
`(palivizumab/variant)
`
`1000
`
`10000
`
`Figure 4. (a) Titration of 493L1FR and koff-improved palivizumab Fab variants on immobilized RSV F protein.
`(b) Inhibition of the binding of koff-improved Fab variants to F protein by palivizumab IgG. In both (a) and (b), bacterial
`periplasmic extracts containing Fab variants AFFF (,), AFSF (:), S32A ($), 493L1FR (&), and an irrelevant Fab (B)
`were tested as described in Materials and Methods. For the inhibition study, Fab molar ratio of the palivizumab IgG (two
`Fabs per molecule) to Fab variants was plotted on the x-axis. (c) Titration of palivizumab Fab and its kon-improved Fab
`variants on immobilized RSV F protein. (d) Inhibition of the binding of kon-improved Fab variants to F protein by
`palivizumab IgG. In both (c) and (d), purified Fab variants A4b4 (6), A12a6 (C), palivizumab Fab (%), and an irrelevant
`Fab (B) were tested.
`
`experiments; the data shown in Figure 5(a) and (b)
`are from one typical neutralization curve.
`
`kon optimization with novel ELISA screen
`
`Many of the koff combinatorial mutants had high
`potency for neutralization of RSV in the Fab format
`but did not show any further increase in potency
`upon conversion of Fab to IgG. We thus next
`explored the potential of optimizing kon. We
`reasoned that theoretically an antibody with a faster
`kon should have a better chance to bind to and
`neutralize the virus before the virus has the
`opportunity to infect the cells.
`An iterative mutagenesis approach that involved
`screening of about ten CDR mutation libraries was
`used to gradually improve the kon. A schematic flow
`chart for the entire procedure is shown in Figure 7.
`Clone AFFF (Figure 1) was used as a template in the
`first round of kon mutagenesis. This combinatorial
`Fab showed one of the best improvements in koff
`and was the most potent viral neutralizing clone
`
`derived from the koff-driven affinity maturation.
`Libraries consisting of single mutations in HCDR3
`or LCDR3, double mutations in HCDR3 or LCDR3,
`and double mutations with one in HCDR3 and one
`in LCDR3 were prepared and screened. To identify
`variants with increased kon, we developed a novel
`ELISA screening method. In principle, we wanted
`to reduce the interaction time between antibody
`and antigen as much as possible to favor the
`selection of variants with higher kon. In addition,
`after the antibody–antigen complex was formed,
`both the number of washes and the washing time
`were minimized to reduce the impact of antibody–
`antigen complex dissociation. Using a BIAcore
`kinetic simulation program, we modeled several
`kinetic and interaction parameters, such as kon, koff,
`and association and dissociation times.23 Appro-
`priate association and dissociation ELISA con-
`ditions were then determined, thus allowing the
`easy selection of high kon variants over low koff
`variants in output signals. We screened using a
`ten minute incubation time for antibody–antigen
`
`

`

`Ultra-potent Anti-RSV Antibodies
`
`133
`
`{a)
`
`0.8
`
`0.6
`
`0.4
`
`0.2
`
`0
`It)
`v
`c(
`
`{b)
`
`0.8
`
`0.6
`
`0
`It)
`v 0.4
`c(
`
`0.2
`
`0
`0.001
`
`0.01
`
`0.1
`Fab (1,19/ml)
`
`10
`
`100
`
`0
`0.01
`
`0.1
`lgG (1,19/ml)
`
`10
`
`{c)
`
`0.8
`
`0.6
`
`0.4
`
`0.2
`
`0
`It)
`v
`c(
`
`{d)
`
`0.8
`
`0.6
`
`0
`It)
`v 0.4
`c(
`
`0.2
`
`0
`0.001
`
`0.01
`
`1
`0.1
`Fab (1,19/ml)
`
`10
`
`100
`
`0
`0.01
`
`0.1
`lgG (1,19/ml)
`
`10
`
`Figure 5. RSV neutralization curves of palivizumab and its variants derived from a microneutralization assay. Several
`koff-improved variants in the Fab (a) or IgG (b) format were measured for their abilities to inhibit RSV replication in HEp-
`2 cells. Variants AFFF (,), AFSF (:), AFFG (6), palivizumab (&), and BSA (B) were titrated. Several kon-improved
`variants as Fab (c) or IgG (d) were also measured. Variant A1e9 (6), A13c4 (%), A12a6 (,), A4b4 ($), and palivizumab
`(&) were titrated.
`
`interaction followed by three quick washes in less
`than 30 seconds. We also eliminated the conven-
`tional second step of applying a secondary detec-
`tion antibody.
`Instead, we precomplexed the
`biotinylated F protein with horseradish peroxidase
`(HRP)-conjugated streptavidin and used it in the
`first step. To boost the ELISA signal, we also added
`biotinylated HRP to the precomplex.
`Four heavy chain variants, S95D, S95F, S95L and
`S95N/M96S, were identified from the HCDR3
`libraries, and three light chain variants, F93A,
`F93G, and F93W, were identified from a LCDR3
`library. As estimated by BIAcore analysis, most of
`these mutations improved the association rate only
`marginally, by less than 80%. Interestingly, F93G
`mutation represents a reversion to a wild-type
`residue. It was mutated to an F in the clone AFFF,
`which was selected for its improved koff. The
`mutation at light chain position 93 to W was also
`identified earlier, in the context of 493L1FR, for its
`improved koff, with no kon benefit. A combinatorial
`library consisting of these beneficial kon mutations
`was subsequently constructed and screened. Two of
`the best combinatorial clones were the combi-
`nations of S95D with F93G or F93W.
`The variant
`that contained S95D and F93G
`mutations, denoted as D95/G93, was used as a
`template in the second round of kon mutagenesis.
`Six single-mutation CDR libraries based on
`D95/G93 were constructed and screened for F
`protein binding. Single mutations that resulted in
`
`enhanced affinity for the F protein arising from kon
`improvements of the Fabs were identified. These
`mutations and the earlier identified mutations,
`S95D and F93G, are listed in Figure 2(b) and
`Table 2. Due to the relative small increase in kon,
`we did not characterize in detail
`the kinetic
`constants of these single mutations. The mutation
`to proline at position 32 on HCDR1, which was
`identified earlier in the first affinity maturation
`attempt in the context of 493L1FR, was identified
`here for its ability to improve kon (Figure 2).
`Similarly, the mutation to tyrosine at position 52
`on LCDR2 was also identified previously.
`In
`summary, all four positions that yielded improve-
`ments in koff, including positions 32 (HCDR1), 100
`(HCDR3), 52 (LCDR2) and 93 (LCDR3), could also
`be mutated to improve kon. The three-dimensional
`locations of the kon and koff mutations are shown in
`Figure 3. The structural modeling shows that kon
`mutations are located in a broad area covering the
`entire CDR regions. In contrast, the koff mutations
`are restricted to four positions.
`Combinatorial libraries of these kon mutations
`were constructed and screened; this then led to the
`identification of Fab variants (Table 2) with mostly
`four to fivefold improvements in kon compared to
`the palivizumab Fab (Table 3). To verify the binding
`specificity of these kon variants, several purified
`combinatorial Fab variants were tested in ELISA for
`binding to the F protein with the presence of
`palivizumab IgG;
`in addition,
`titrations of the
`
`

`

`134
`
`Ultra-potent Anti-RSV Antibodies
`
`(a)
`
`5501
`200 =
`
`~ 150
`.s
`g_"' 100
`
`0
`
`6
`
`6
`
`6
`~
`6
`
`50
`
`6
`
`0
`
`Fab
`
`k011 (x 10-4)
`Palivizumab -6.62
`
`Single k.itt
`mutations
`
`-1 .74
`
`-1 .75
`} 1.25-1 .53
`
`-0.93
`
`-0.84
`
`Combinatorial -S0.05
`k0tt mutations
`
`(c)
`10
`
`8
`
`~ 6
`.s
`"'
`g_
`
`4
`
`0
`
`2
`
`0
`
`(b)
`5501
`200 =
`
`Palivizumab, single
`and combinatorial
`k0tt mutations
`IC50 = -3 nM
`
`~ 150
`.s
`"'
`g_ 100
`
`0
`
`50
`
`0
`
`lgG
`
`(d)
`10
`
`8
`
`~ 6
`.s
`0
`"'
`0
`
`4
`
`2
`
`0
`
`Combinatorial k.,,
`mutations
`IC50 = -0.1-0.2 nM
`
`lgG
`
`(cid:144)
`
`(cid:144)
`
`Combinatorial
`k0, mutations
`
`k011 {x 10-4)
`
`}2.40-3.89
`
`}o.84-3.oo
`
`} 0.52-1 .12*
`
`the
`Figure 6. Summary of
`beneficial effects of koff, kon and
`bivalence of the antibody on RSV
`neutralization as indicated by
`the reduction in IC50 (determined
`in a microneutralization assay).
`(a) Comparison of the IC50 of
`palivizumab Fab with its koff-
`improved Fab variants. In Fab
`format, a strong correlation was
`observed between the IC50 and
`koff. Combinatorial koff variants
`with two log reduction in koff
`have w300-fold improvements
`in the ability to neutralize virus
`compared with palivizumab.
`(b) Conversion to IgG of palivizu-
`koff-improved
`mab
`and
`its
`variants. The bivalent binding
`effect has increased significantly
`the ability to neutralize virus for
`the palivizumab and its single koff
`mutation variants, but not
`the
`combinatorial koff variants. The
`IC50 values of palivizumab IgG
`and all of its koff-variants converge
`at w3 nM. (c) The IC50 of the
`§
`combinatorial kon Fab variants.
`s
`These variants have about four to
`fivefold improvements in kon,
`8
`which resulted in substantial
`enhancements in viral neutraliz-
`ation compared with palivizu-
`Fab
`mab. The differences in IC50
`
`K4 sK1 is not
`among these kon variants are in part due to their differences in koff. * One outlier with a koff of 2.19!10
`included. (d) Conversion to IgG of the combinatorial kon-improved variants. Upon conversion to IgG, the IC50 values of
`all the combinatorial kon variants converge at w0.1–0.2 nM, despite their differences observed in Fab formats. This
`bivalent effect was similarly observed in koff variants. Overall, the kon improvement resulted in a 15 to 30-fold
`enhancement in viral neutralization compared with palivizumab IgG.
`
`purified combinatorial Fab variants for binding to
`immobilized F protein were carried out. All the
`variants tested competed with palivizumab. Typical
`ELISA titration curves are shown in Figure 4(c), and
`inhibition curves shown in Figure 4(d).
`The building of the improvement in kon in AFFF
`significantly diminished the improvement in koff.
`AFFF Fab has a koff two log better than that of the
`palivizumab Fab, while these kon combinatorial
`Fabs have a koff only two to 13-fold better. This
`result was not surprising because some of the
`beneficial koff mutations in