J. Mol. Biol. (1995) 254, 392–403
CDR Walking Mutagenesis for the Affinity Maturation
of a Potent Human Anti-HIV-1 Antibody into the
Picomolar Range
Wei-Ping Yang1, Kimberly Green1, Sally Pinz-Sweeney1
Amelia T. Briones1, Dennis R. Burton1,2 and Carlos F. Barbas III1*
1
Department of Molecular
Biology, The Scripps Research
Institute, 10666 North Torrey
Pines Road, La Jolla, CA
92037, USA
2
Department of Immunology
The Scripps Research
Institute, 10666 North Torrey
Pines Road, La Jolla, CA
92037, USA
We describe the investigation of methodologies for the creation of very high
affinity human antibodies. The high affinity human antibody b4/12 was
optimized for its affinity to the human envelope glycoprotein gp120 of
human immunodeficiency virus type 1 (HIV-1). Five libraries of b4/12
were constructed by saturation mutagenesis of complementarity-determining regions (CDRs). Libraries of antibody Fab fragments were displayed on
the surface of filamentous phage and selected in vitro for binding to
immobilized gp120. Sequential and parallel optimization strategies of
CDRs were examined. The sequential CDR walking strategy consistently
yielded b4/12 variants of improved affinity in each of the four different
optimization sequences examined. This resulted in a 96-fold improvement
in affinity. Additivity effects in the antibody combining site were explored
by combining independently optimized CDRs in the parallel optimization
strategy. Six variants containing optimized CDRs were constructed.
Improvement of affinity based on additivity effects proved to be
unpredictable but did lead to a modest improvement in affinity. Indeed,
only one of the six combinations demonstrated additivity. The highest
affinity Fab prepared using this strategy was improved 420-fold in affinity.
The affinity of this Fab was 15 pM as compared to 6.3 nM for b4/12.
Examination of the kinetics of Fab binding to gp120 revealed that
improvements in affinity were dominated by a slowing of the off-rate of
the Fab. The methodology presented here provides a route for the
improvement of the affinities of antibodies typical of tertiary immune
responses into the picomolar range. Such improvements may have
profound effects on the utility of antibodies as therapeutic and prophylatic
agents.
7 1995 Academic Press Limited
*Corresponding author
Keywords: phage display; human immunodeficiency virus;
structure/function; passive immunization; additivity effects
Introduction
Antibodies represent the most molecularly diverse and structurally studied family of proteins
(Padlan, 1994). Their therapeutic use in humans has
been studied for over a hundred years (Behring,
1893) and their role as adaptable binding molecules
of defined specificities has aided the flowering of
molecular biology in the latter half of this century.
Abbreviations used: CDR, complementaritydetermining region; HIV-1, human immunodeficiency
virus type 1; gp, glycoprotein; sCD4, soluble CD4,
PCR, polymerase chain reaction; Ig, immunoglobulin.
0022–2836/95/480392–12 $12.00/0
Hence, antibodies play a central role in basic science
and in our own survival. Our interest has been to
develop a more thorough understanding of antibodies through the purposeful manipulation of
their affinity and specificity.
The current HIV-1 pandemic demands the
development of potent anti-viral molecules. In this
regard we have utilized phage display of combinatorial antibody Fab fragment libraries to select a
large number of human anti-HIV-1 antibodies
(Barbas et al., 1993a; Burton & Barbas, 1994). The
most promising of these is antibody b4/12, which
when prepared as an IgG1 has been demonstrated
to efficiently neutralize a large number of primary
7 1995 Academic Press Limited
393
Affinity Maturation of a Human Antibody
isolates of HIV-1 (Burton et al., 1994). To improve the
likelihood that this molecule could succeed in
prophylactic and therapeutic application we have
developed methods for the improvement of its
affinity. A number of methods and reports have
been published which are devoted to the improvement of antibody affinity (for a review, see Burton
& Barbas, 1994). We have sought methods for the
development of very high affinity antibodies
wherein modifications to the parental antibody are
constrained to the complementarity determining
regions. Changes in primary sequence in these
regions are less likely to generate immunogenic
antibodies than changes in the more sequence-constrained framework regions. In a previous report
we improved the affinity of b4/12 for gp120 about
eightfold and demonstrated that affinity was
correlated with neutralization potency (Barbas et al.,
1994).
In this report we extend and develop our
preliminary examination of the CDR walking
mutagenesis strategy. Sequential and parallel
optimization strategies have been explored and
the general applicability of additivity effects in the
combining site examined. We have improved the
affinity of Fab b4/12 420-fold in its affinity for
the HIV-1 envelope protein gp120. This study
suggests a route to the creation of very high affinity
antibodies.
Results
Design of antibody libraries
No specific structural information on the antibody b4/12 or its antigen gp120 was available to
guide the design of Fab libraries. Antibodies,
however, constitute one of the most structurally
studied classes of proteins (Padlan, 1994; Wilson &
Stanfield, 1993). General information regarding the
chemistry of this class of proteins, together with
previous results obtained by chain shuffling
experiments with this antibody, has served to guide
the experiments under consideration here (Barbas
et al., 1993b). Five segments within four CDRs were
targeted for saturation mutagenesis and affinity
optimization using phage display. Within the heavy
chain, the entire CDR1-H region consisting of
residues H31 to H35 was targeted in library H1
(numbering convention of Kabat et al. (1991)). The
CDR3-H region of b4/12, 18 amino acid residues
long, is significantly longer than the average
observed in human antibodies, which is approximately 12 residues (Wu et al., 1993). Due to technical
limitations in the size of libraries which may be
constructed and surveyed with confidence using the
phage display approach, six codons or less are
generally targeted for saturation mutagenesis using
NNK or NNS doping strategies (Lowman & Wells,
1993; Barbas & Barbas, 1994). For this reason, the
nine residues at the core of the VDJ fusion region
were targeted in a two-step sequential optimization
strategy; H97-100 and H100A-E. Chain shuffling
experiments suggested this region to be a natural
hotspot in the somatic evolution of this antibody
(Barbas et al., 1993b). In previous studies, a
sequential selection of library H1 followed by
randomization and selection of residues H97-100 of
CDR3-H yielded a clone designated 3B3 with a
8.2-fold improvement in affinity (Barbas et al., 1994).
As detailed below, the increase in affinity of the 3B3
clone was due almost exclusively to optimization of
the H97-100 region. In this report, the adjacent
residues in CDR3-H, H100A-E, were chosen for
optimization in the context of the clone 3B3; library
H3B3. In the light chain, CDRs 1 and 3 were
targeted for random mutagenesis. Interaction of
CDR2-L with antigen is, in general, less frequent
and was not considered for optimization here
(Wilson & Stanfield, 1993). In CDR1-L six residues
predicted to be solvent exposed, L27A to 32, were
targeted for random mutagenesis in library L1. In
CDR3-L residues L90, and L92 to 95 were targeted
for randomization. Residue L91 which is a highly
conserved tyrosine in light chains in general (Wu
et al., 1991) and in b4/12 variants obtained by chain
shuffling and was not targeted. The CDRs which
have been targeted are four of the five most heavily
utilized contact regions observed in structures of
antibody–antigen complexes (Padlan, 1994; Wilson
& Stanfield, 1993). These regions are predicted to
form a significant portion of the core of the binding
surface of the antibody. This is demonstrated in
Figure 1 by highlighting the regions under
consideration on the combining site of the human
antibody Kol, for which a crystal structure has been
determined (Marquart et al., 1980).
Sequential optimization of CDRs
To assess the sequential CDR walking strategy,
four sequential CDR optimization experiments were
performed. Selected clones were characterized by
sequencing of the entire variable domain and
determination of affinity of purified Fab for antigen
using the BIAcore biosensor from Pharmacia
(Karlsson et al., 1991). The first of these experiments
extends the previous two-step sequential walk
which targeted CDR1-H in the H1 library and
residues H97 to 100 of CDR3-H (Barbas et al., 1994).
The highest affinity clone that resulted from these
studies was clone 3B3. The subsequent step was to
CDR1-L in library L1. Library L1 consisted 5 × 108
independent variants in this region built within the
context of the 3B3 heavy chain. Following four
rounds of selection, eight clones were sequenced
following verification of binding activity with the
soluble protein. Five clones were identical to the
starting clone at the nucleic acid level and resulted
from contamination of the original sequence in the
construction of the library. Contamination was at a
low level since sequencing of ten unselected
variants revealed ten unique sequences. The
affinities of the three remaining unique clones
394
Affinity Maturation of a Human Antibody
Figure 1. The tube structure of a
Fab on which the regions targeted
for optimization have been highlighted. The antibody Kol has been
utilized as a model here because of
its extended CDR3-H region which
is similar in length to antibody
b4/12 under consideration here. The
light chain variable region is shown
on the left in pink and the heavy
chain variable region on the right in
blue. The individual regions
targeted within the CDRs have been
color coded with the key given in
the Figure. This view highlights the
relative disposition of the CDR
regions targeted for optimization.
Molecular models were constructed
by Mike Pique using AVS software.
3B3/L1.1-3 were characterized. The clones as
compared to their parent 3B3 were found to be of
lower or similar affinity (Table 1). Selection was then
continued for an additional four rounds resulting in
the isolation of a single clone following the
sequencing of ten clones, 3B3/L1.4. This clone was
improved 29-fold in affinity with respect to b4/12
and 3.5-fold with respect to 3B3 (see Table 1). In the
next step, five residues within the CDR3-L of
3B3/L1.4 were mutagenized and a library of 2 × 108
variants was constructed; library L3A. Following
five rounds of selection, nine clones were characterized by sequence analysis and affinity determination (Table 2). Each clone contained a unique
CDR3-L sequence and an affinity for antigen that
was generally lower than the parent clone of this
series, 3B3/L1.4. To improve the selection of higher
affinity variants in the sub-nanomolar range using
a solid phase selection format, an off-rate biased
selection was developed. In five subsequent rounds
of selection, 8 mM 3B3 was added to the selection
well following the washing step. The selection plate
was then placed at 37°C for three hours, the well
washed, and bound phage eluted at low pH. This
selection resulted in the isolation of clones
3B3/L.14L3.10 to 12. These clones differ in sequence
only at position L95. Each clone was improved in
affinity for antigen. The highest affinity clone
3B3/L1.4L3.11 was improved 3.3-fold in affinity
above 3B3/L1.4 and 96-fold above b4/12. This clone
is the product of a four-step sequential CDR walk.
A parallel experiment without off-rate selection
with antibody revealed a multitude of clones similar
to that seen following the initial five rounds of
Table 1. Light chain CDR1 mutants isolated from the L1 library and by chain shuffling
Clone
b4/12
3B3
3B3/L1.1
3B3/L1.2
3B3/L1.3
3B3/L1.4
CS
CS
CS
CS
27A
S
S
K
T
P
Q
N
R
R
N
Residue position
28
29
30
31
32
Kd (parent)
Kd (mutant)
Kd (b4/12)
Kd (mutant)
I
I
E
V
L
L
I
I
I
I
R
R
R
D
A
S
R
R
R
R
R
R
R
R
R
R
R
R
R
R
1
8.2
0.16
0.7
1.1
3.5
NA
NA
NA
NA
—
8.2
1.3
5.7
9.0
29
ND
ND
ND
ND
R
R
F
Y
H
D
R
S
G
W
S
S
G
R
R
G
S
S
S
S
The clone designated 3B3 has been described. 3B3 contains mutations in CDR1-H and CDR3-H
only. Clones isolated from the L1 library contain the L1 designation in their name and in the
examples are paired with the heavy chain from 3B3. Dissociation constants were determined with
the BIAcore instrument from Pharmacia. The ratio of parent and mutant Kd values gives the fold
increase in affinity for this mutagenesis step which is attributed to changes in this region. The ratio
of Kd b4/12 and mutant Kd gives the overall improvement in affinity from the starting Fab b4/12.
Clones isolated from chain shuffling experiments have been described and characterized (Barbas
et al., 1993b) and are given the CS designation. The CS clones contain multiple mutations in the
light chain variable region paired with the original heavy chain of b4/12 and are of an affinity
similar to b4/12. For these CS clones ratios of Kd values were not applicable (NA) and were not
determined (ND) with the BIAcore instrument.
395
Affinity Maturation of a Human Antibody
Table 2. Light chain CDR3 mutants isolated from L3 libraries and by chain shuffling
Clone
89
90
91
Residue position
92
93
94
95
96
97
Kd (parent)
Kd (mutant)
Kd (b4/12)
Kd (mutant)
b4/12
3B3/L1.4
3B3/L1.4L3.1
3B3/L1.4L3.2
3B3/L1.4L3.3
3B3/L1.4L3.4
3B3/L1.4L3.5
3B3/L1.4L3.6
3B3/L1.4L3.7
3B3/L1.4L3.8
3B3/L1.4L3.9
3B3/L1.4L3.10
3B3/L1.4L3.11
3B3/L1.4L3.12
3B3/L3.13
3B3/L3.14
3B3/L3.15
3B3/L3.16
CS
CS
CS
Q
Q
Q
Q
Q
Q
Q
Q
Q
Q
Q
Q
Q
Q
Q
Q
Q
Q
Q
Q
Q
V
V
V
L
T
T
S
K
M
M
Q
T
T
T
V
Q
V
K
T
T
T
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
G
G
G
G
G
G
G
G
G
G
G
G
G
G
G
G
G
G
G
G
G
S
S
Q
N
V
G
D
F
D
T
L
S
H
I
S
F
A
T
S
T
S
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
T
T
T
T
T
T
T
T
T
T
T
T
T
T
T
T
T
T
T
T
S
1
3.5
0.43
0.52
0.18
0.49
0.44
1.0
0.05
0.36
0.50
3.0
3.3
1.7
1
5.6
0.23
3.2
NA
NA
NA
1
29
12
15
5.2
14
13
29
1.5
10
15
87
96
49
8.2
46
1.9
26
ND
ND
ND
A
A
W
R
R
W
G
D
G
G
D
R
R
R
A
W
G
G
G
G
G
S
S
S
G
G
S
R
S
R
F
S
G
G
G
S
P
S
G
S
A
S
Clones containing CDR1-L mutations contain the L1 designation in their name and in the examples above are paired with the heavy
chain from 3B3. Clones containing CDR3-L mutations contain the L3 designation in their name. Dissociation constants were determined
with the BIAcore instrument from Pharmacia. The ratio of parent and mutant Kd values gives the fold increase in affinity for this
mutagenesis step which is attributed to changes in this region. The ratio of Kd b4/12 and mutant Kd gives the overall improvement
in affinity from the starting Fab b4/12. Clones isolated from chain shuffling experiments have been described and characterized (Barbas
et al., 1993b) and are given the CS designation. The CS clones contain multiple mutations in the light chain variable region paired
with the original heavy chain from b4/12 and are of an affinity similar to b4/12. For these CS clones ratios of Kd values were not
applicable (NA) and were not determined (ND) with the BIAcore instrument.
selection, indicating the increased efficacy of the
off-rate selection procedure. Previously, an antigenbased off-rate selection was reported by Hawkins
et al. (1992). This technique was not considered here
due to the very high cost of the gp120 antigen.
The second sequential walk involved CDR1-H to
CDR3-H to CDR3-L steps. In this experiment,
4 × 108 variants in CDR3-L were produced, paired
with the heavy chain from 3B3; library L3B.
Following the selective protocol utilized for the L3A
library, four unique clones were characterized
following sequencing. Clone 3B3/L3.14 was improved 5.6-fold in affinity as compared to 3B3 and
46-fold as compared to b4.12. This clone was the
predominant clone obtained after selection as
revealed by sequencing.
The third sequential CDR walk involved CDR1-H
to CDR3-H97-100 to CDR3-H100A-E. This involved
the construction of library H3B3 wherein 1 × 108
variants in the H100A-E region were generated in
the context of clone 3B3. Following a selective
regime involving antibody assisted off-rate selection, seven clones of highly related sequence were
isolated and characterized (Table 3). All but one
clone was improved in affinity for antigen, with the
highest affinity clone h1.3B/h3.33 being improved
7.7-fold as compared to 3B3 and 63-fold as
compared to b4/12.
The last CDR walk involved a re-investigation of
the single step walk to CDR1-H. Previously, library
H1 had been generated and selected to yield a
sub-library of functional CDR1-H variants of b4/12.
The individual clones were not characterized and
were utilized as a collection in subsequent
mutagenesis of CDR3-H generating among others,
clone 3B3 (Barbas et al., 1994). The sequence at
position H31 to H33 of 3B3 was Asn-Phe-Thr. This
sequence generates a potential glycosylation site in
CDR1-H. To circumvent this sequence, as well as to
identify a more optimal sequence in CDR1-H,
library H1 was selected for an additional six rounds
of panning beyond that previously reported. The
results of this selection are given in Table 4. Clones
lacking the potential glycosylation site were
characterized. Clone h1.1 exhibited a 3.9-fold
improvement in affinity for antigen.
Analysis of selected CDRs
Analysis of CDR1-H variants shown in Table 4
demonstrated a strong selection towards a consensus sequence. H31 was exclusively His or Asn,
suggesting a functional role for the conservation of
the d-amino group, which is a shared element in the
side chains of these residues. H32 was exclusively
an aromatic residue, most frequently Phe. H33 was
exclusively Thr. CDR1-H of b4/12 is predicted to
have a class 1 canonical structure. It contains the key
residues which define this class: Ala at 24, Gly at 26,
Tyr at 27, Phe at 29, Ile at 34, and Arg at 94 (Chothia
et al., 1989, 1992). Interestingly, selection at H34
revealed Ile, Leu, or Val, all of which fall within
those described for a class 1 structure. Position H35
revealed a selection towards the parental residue
396
Affinity Maturation of a Human Antibody
Table 3. Heavy chain CDR3 mutants isolated from the H3B3 library and by chain shuffling
b4/12
h1.3B/h3.3(3B3)
h3.3
h1.3B/h3.32
h1.3B/h3.33
h1.3B/h3.34
h1.3B/h3.35
h1.3B/h3.36
h1.3B/h3.38
h1.3B/h3.39
CS
CS
CS
CS
95
96
97
98
99
100
Residue position
A B C D E
F
G
I
J
K
101
102
Kd (parent)
Kd (mutant)
V
V
V
V
V
V
V
V
V
V
V
V
V
V
G
G
G
G
G
G
G
G
G
G
G
G
G
G
P
E
E
E
E
E
E
E
E
E
E
E
E
P
Y
W
W
W
W
W
W
W
W
W
W
W
W
Y
S
G
G
G
G
G
G
G
G
G
T
T
T
T
W
W
W
W
W
W
W
W
W
W
W
W
W
W
D
D
D
E
E
E
H
D
T
D
D
D
D
D
D
D
D
D
D
D
D
D
D
D
D
D
A
D
N
N
N
N
N
N
N
N
N
N
N
N
N
N
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
M
M
M
M
M
M
M
M
M
M
M
M
M
M
D
D
D
D
D
D
D
D
D
D
D
D
D
D
V
V
V
V
V
V
V
V
V
V
V
V
V
V
1
ND
7.9
1.8
7.7
5.9
5.6
0.43
4.4
5.5
NA
NA
NA
NA
D
D
D
Q
M
M
Q
Q
Q
Q
D
D
M
D
S
S
S
F
F
R
R
R
R
V
S
F
D
S
P
P
P
R
R
R
R
R
R
R
P
P
P
P
Q
Q
Q
F
Y
F
Y
Y
F
Y
Q
Q
Q
Q
Kd (b4/12)
Kd (mutant)
1
8.2
7.9
15
63
48
46
3.5
36
45
ND
ND
ND
ND
Heavy chain CDR1 mutants contain the h1 designation in their name and in the examples above all Fabs contain the original light
chain from b4/12. Clones containing CDR3-H mutations contain the h3 designation in their name. Clone h3.3 was created by swapping
the mutant h1.3B CDR1-H region of 3B3 with the original CDR1-H region of b4/12. Dissociation constants were determined with the
BIAcore instrument from Pharmacia. The ratio of parent and mutant Kd values gives the fold increase in affinity for this mutagenesis
step which is attributed to changes in this region. The ratio of Kd b4/12 and mutant Kd gives the overall improvement in affinity from
the starting Fab b4/12. Clones isolated from chain shuffling experiments have been described and characterized (Barbas et al., 1993b)
and are given the CS designation. The CS clones contain multiple mutations in the heavy chain variable region paired with the original
b4/12 light chain and are of an affinity similar to b4/12. For these CS clones ratios of Kd were not applicable (NA) and were not
determined (ND) with the BIAcore instrument.
His at this position. Correlation of affinity with
sequence suggests that His at H31 is a key element
in the increased affinities of clones h1.1-4. CDR1-H
of h1.1 and h1.3B differ in sequence only at position
H31 but differ by approximately fourfold in affinity.
Variation in the aliphatic side-chains of the residue
at H34 had little effect on the affinity of these clones.
Previous analysis of CDR3-H variants derived by
selection of the H97 to 100 region suggested the Pro
Table 4. Heavy Chain CDR1 mutants isolated from the H1
library and by chain shuffling
Clone
31
b4/I2
h1.1
h1.3
h1.4
h1.2
h1.5
h1.6
h1.3B/h3.3(3B3)
CS
CS
N
H
H
H
N
N
N
N
N
N
Residue position
32 33 34 35
F
F
F
F
Y
F
W
F
F
F
V
T
T
T
T
T
T
T
T
T
I
V
L
I
L
L
I
V
V
V
H
H
H
M
Q
I
M
H
H
H
Kd (b4/12)
Kd (mutant)
1
3.9
3.2
2.6
ND
ND
ND
8.2
ND
ND
Heavy chain CDR1 mutants contain the h1 designation in their
name and in the examples above all Fabs contain the original
light chain from b4/12. Clones containing CDR3-H mutations
contain the h3 designation in their name. Dissociation constants
were determined with the BIAcore instrument from Pharmacia.
The ratio of Kd b4/12 and mutant Kd gives the overall
improvement in affinity from the starting Fab b4/12 and in all
cases except clone 3B3 gives the fold increase in affinity for this
mutagenesis step which is attributed to changes in this region.
Clones isolated from chain shuffling experiments have been
described and characterized (Barbas et al., 1993b) and are given
the CS designation. The CS clones contain multiple mutations in
the heavy chain variable region paired with the original b4/12
light chain and are of an affinity similar to b4/12. For these CS
clones ratios of Kd were not determined (ND) with the BIAcore
instrument. Clones containing the potential glycosylation site
Asn-X-Thr were not characterized with the exception of 3B3.
to Glu change to be associated with the higher
affinity of clone 3B3. The CDR1-H segment of 3B3
contributes little to the increase in affinity observed
in 3B3. This is demonstrated by swapping the
CDR1-H region of b4/12 for that of 3B3. The
resulting clone h3.3 bound with approximately
the same affinity as 3B3 (Table 3). Sequence analysis
of the H100A to E region selected in library H3B3
reveals a set of highly related clones. Position
H100A was predominately Glu or Asp. H100B was
Gln or Met. Met at this position was observed in a
clone derived by chain shuffling. H100C was
predominately Arg, but Phe and Val were also
observed. Phe had been observed at H100C in a
clone derived by chain shuffling. Position H100D
was exclusively Arg and H100E was either Phe or
Tyr. Within this set of clones are several related by
single point changes. Clone h1.3B/h3.35 and
h1.3B/h3.36 differ only at H100A, which is His and
Asp in the respective clones. They differ in affinity,
however, by 13-fold. The affinity can be completely
restored as shown in clone h1.3B/h3.39 with the
mutation of Arg at H100C to Val. This result
highlights the interdependence of residues in this
region.
CDR1-L sequences shown in Table 1 indicate that
a number of the targeted positions accept a wide
range of residues with little effect on affinity.
Compare 3B3/L1.2 and 3B3/L1.3. Conservation of
Arg at L32 is noted in b4/12 and all variants. Chain
shuffling had produced variants with Asn or Arg at
L27A and Ser, Gly, or Trp at L29. None of these
mutations is noted in the variants obtained by
saturation mutagenesis. The highest affinity variant
differs from the parental sequence in five of the six
selected regions with a significant change in the net
charge of this region from +3 to 0. CDR1-L of b4/12
and selected variants do not possess sequences
397
Affinity Maturation of a Human Antibody
Table 5. Binding kinetics of b4/12, b4/12 mutants and sCD4
Clone
kon /104
(M−1 s−1 )
koff /10−4
(s−1 )
sCD4
b4/12
h1.1
h1.3Bh3.33
h1.1h3.33
h1.1h3.33/L1.4L3.14
h1.1h3.33/L1.4L3.11
h1.1h3.33/L3.14
L1.4L3.14
L1.4L3.11
1.7 2 0.03
7.6 2 0.02
14 2 1.0
7.0 2 0.8
15 2 2.6
7.8 2 0.7
9.2 2 0.9
7.0 2 0.1
12 2 1.1
14 2 1.2
1.0 2 0.007
4.8 2 0.02
2.3 2 0.03
0.071 2 0.006
0.034 2 0.001
0.012 2 0.002
0.090 2 0.007
0.083 2 0.005
4.1 2 0.02
4.2 2 0.02
Kd (nm)
5.9
6.3
1.6
0.10
0.022
0.015
0.097
0.12
3.4
3.9
Kd (b4/12)
Kd (mutant)
NA
1
3.9
63
286
420
65
53
1.9
1.6
DDGI (kcal/mol)
0.094
−1.54
−2.33
−2.02
−1.39
−1.17
Binding kinetics were determined using the BIAcore instrument from Pharmacia. The envelope
glycoprotein gp120 was immobilized on the biosensor chip. The Kd value was calculated from koff /kon . The
ratio of Kd values indicates the fold increase in binding affinity as compared to the Fab b4/12. DDGI was
calculated using equation (1) and DDG as + RT ln(Kd b4/12/Kd mutant).
which fall within those described for classification
of this region into a canonical structure class
(Chothia et al., 1989).
CDR3-L sequences demonstrated a striking
selection for Gly at L92 in all 16 characterized
mutants (Table 2). This position was conserved in all
mutants derived by chain shuffling. L93 was
predominately Gly or Arg. Gly at this position was
the predominant mutant obtained by chain shuffling. Other positions were more permissive. Four
clones are related by amino acid changes at a single
position, L95. Clone 3B3/L1.4L3.3, which has Val at
this position binds gp120 with an affinity 18-fold
lower than 3B3/L1.4L3.11, which has His at this
position. Conversion of His to either Ser or Ile in
3B3/L1.4L3.10 and 3B3.L1.4L3.12 has little effect on
affinity. This highly related set has Thr at position
L90, which was the only mutant at this position
obtained by chain shuffling. With respect to
canonical structures predicted for this region, the
parental antibody and all variants except 3B3/L3.14
do not contain conformation defining residues of the
appropriate identity required for classification.
Selection for Gln at L90 and Pro at L94 in this clone
classifies this as a class 2 canonical structure
(Chothia et al., 1989). This was the highest affinity
clone selected in the CDR1-H to CDR3-H to
CDR3-L walk and differs radically in sequence from
3B3/L1.4L3.11, which was the optimal clone
selected in the context of the optimized CDR1-L
region L1.4.
Investigation of additivity effects in the
antibody combining site
An alternative CDR walking strategy involves a
parallel optimization strategy wherein CDRs are
independently optimized and recombined in a
single clone. This strategy is based on the
assumption that in most cases additivity will be
observed when non-interacting mutations are
combined (Wells, 1990). To assess the applicability
of additivity principles to the antibody combining
site, six new antibodies were created by shuffling
optimized CDR regions. The resulting clones were
expressed, purified, and the affinity for antigen
determined. The results are shown in Table 5.
Combination of the most optimal CDR3-H region
with the most optimal CDR1-H region produced
clone h1.1h3.33, which was improved 286-fold in its
affinity for antigen as compared to b4/12. In this
case the measured affinity is in good agreement
with that predicted assuming additivity; 286-fold as
compared to 246. Recruitment of CDR1-L and
CDR3-L segments L1.4 and L3.14 into this clone
resulted in only a 1.5-fold increase in affinity, see
clone h1.1h3.33/L1.4L3.14. These segments yielded
3.5 and 5.6-fold increases in affinity, respectively, in
the context of the antibody in which they were
originally selected. If simple additivity had been
operative, a 19.6-fold improvement in affinity
relative to h1.1h3.3 would have been observed and
an antibody with a 5600-fold improvement in
affinity would have resulted. These two light chain
CDRs were the products of independent selections.
To investigate the source of the discrepancy
between the observed and predicted results, clone
h1.1h3.3/L3.14 was constructed. This clone introduces only a single optimized CDR3-L region. A
5.5-fold decrease in affinity was observed as
compared to clone h1.1h3.33. This result demonstrates an incompatibility of the L3.14 CDR3-L
region with the h1.1h3.33 heavy chain. In clone
h1.1h3.33/L1.4L3.11, the light chain containing the
sequentially optimized CDR1 and CDR3 regions
L1.4 and L3.11 was combined with the h1.1h3.33
heavy chain. The resulting clone demonstrated an
affinity 4.4-fold lower than that observed for
h1.1h3.3 paired with the original light chain derived
from b4/12. Simple additivity would have produced a 12-fold increase in affinity. The results
obtained by combining optimized light chain
regions with heavy chain variants are unpredictable. Combination of L1.4L3.14 and L1.4L3.11
light chains with the original heavy chain of b4/12
resulted in approximately a twofold increase in
affinity in each case.
398
Changes in the binding kinetics of improved
antibodies
Insight into the mechanism of affinity improvement is provided by determination of the binding
kinetics using the BIAcore instrument. As shown in
Table 5, improvements in affinity are dominated by
decreases in off-rates. Only a twofold variation in
on-rate was observed compared to a 400-fold
variation in off-rate. This trend was maintained in
the 34 additional clones reported in Tables 1 to 4
(data not shown). Off-rate is expected to be most
affected if phage selections are performed under
conditions approaching equilibrium (Lowman &
Wells, 1993). The kinetics of soluble CD4 (sCD4)
binding to gp120 were determined for comparative
purposes. CD4 is the natural ligand for gp120. The
affinity determined by BIAcore is in close agreement with the reported value of 3 nM (Ryu et al.,
1990) This measurement serves to correlate the
affinities reported here with those determined by
other methods. The apparent Kd value of Fab b4/12
as determined by competition ELISA is 9 nM
(Barbas et al., 1992b). Affinity improvements
determined with the BIAcore instrument are
generally well correlated with those determined
with other methods (Lowman & Wells, 1993; Kelley
& O’Connell, 1993). Antibody binding kinetics
compare favorably with sCD4. Fab on-rate was
faster in all cases than that observed for sCD4. The
off-rate of sCD4 is 4.8-fold slower than b4/12 and
83-fold faster than the highest affinity Fab
h1.1h3.33/L1.1L3.14. Since small quantities of
aggregated protein can lead to artifactually slow
off-rates (van der Merwe et al., 1993), the
dissociation of Fabs b4/12 and h1.1h3.33 were
studied following gel filtration chromatography. In
both cases the off-rates determined for these
proteins were within the values determined for the
proteins that had not been subjected to this
additional purification step. In an additional study
the binding kinetics of Fab h1.1h3.33 binding to
gp120 were determined in the reverse orientation,
i.e. with Fab immobilized and gp120 as the soluble
ligand. The on-rate of this interaction was 2.8-fold
slower and the off-rate was 1.3-fold faster than that
determined with gp120 immobilized. This variation
is within the range expected since it is difficult to
match the densities and orientations of the proteins
which are immobilized. Thus, the differences in
off-rates reported herein are not artifacts resulting
from protein aggregation.
Discussion
Antibody b4/12 represents the most potent and
broadly neutralizing human anti-HIV-1 antibody
yet described (Burton et al., 1994). This antibody has
been demonstrated to neutralize 75% of 36 primary
isolates of HIV-1 tested at concentrations which
could be achieved by passive immunization. This
extends earlier work that demonstrated potent
neutralization of laboratory adapted isolates of
Affinity Maturation of a Human Antibody
HIV-1 with monovalent Fab of this antibody (Barbas
et al., 1992b). The Fab fragment of b4/12 was
originally isolated from a phage display library
prepared from an individual who had been HIV-1
positive for six years but had no symptoms of
disease. The antibody binds a highly conformational epitope in the CD4-binding region of the
HIV-1 envelope glycoprotein gp120. The epitope
within the CD4 binding region of gp120 recognized
by b4/12 is unique as compared to five other
antibodies directed against this region and isolated
from the same individual (Roben et al., 1994).
To potentially improve the therapeutic or prophylactic efficacy of this antibody, we investigated
strategies for the evolution of its affinity for antigen.
Phage display provides a convenient format for the
refinement of the contact between receptor and
ligand or antibody and antigen which may result
from unpredictable sequence changes in the region
of interest. The limitations of the phage display
approach have been considered elsewhere (Lowman & Wells, 1993; Barbas & Barbas, 1994). We have
proposed that repeated introduction of diversity
into CDRs followed by stringent selections should
allow for the refinement of human antibodies to
levels of affinity far beyond those generated by the
immune response (Barbas et al., 1994). CDR
targeted mutagenesis is advantageous since optimization of these regions is most likely to improve
affinity and least likely to create problems of
immunogenicity. CDR walking involves targeted
mutagenesis of CDR regions and selection for
fitness with monovalent phage display (Barbas et al.,
1994).
The free energy changes that result from the
combination of sets of mutations may be expressed
by the equation:
DDGAB = DDGA + DDGB + DDGI
(1)
where DDGA and DDGB are changes in the free
energy of mutants A and B, DDGAB is the change in
free energy for the protein that combines the
mutations of A and B, and DDGI is the change in the
interaction energy between the two sets of residues
(after Lowman & Wells, 1993). In accord with
equation (1), two strategies are evident for the
application of CDR walking, either sequential or
parallel optimization of CDRs. The most appropriate route to the optimization of antibody affinity is
dependent on the term DDGI . In this study we have
investigated these two CDR walking strategies.
Sequential CDR walking takes into account that
DDGI may not always be negligible and that optimal
binding may result from the interdependence of
CDR loops. Such interdependence could result from
coordinated structural changes on binding antigen
and is supported by recent evidence which suggests
induced-fit mechanisms may best describe the
binding observed in some antibody–antigen complexes (Wilson & Stanfield, 1993). Together with a
previous study we have performed five sequential
CDR walking experiments. In the previous study a
library of CDR1-H variants, H1 library, with five
Affinity Maturation of a Human Antibody
residues randomized was selected by panning
against gp120. The collection of clones which
resulted was then used in the construction of a
CDR3-H library where four additional residues
were randomized. The resulting library was
selected by an additional six rounds of panning. A
variety of clones with improved binding were
produced. Correlations within this set of selected
progeny were examined to establish functional
trends which may be predictive of the activity of
subsequent generations of b4/12 (Barbas et al.,
1994). With the gp120 envelope protein derived
from the MN isolate of HIV-1, affinity was well
correlated with neutralizing ability. Correlation of
affinity with neutralizing ability in the HIV-1
system is very dependent on the epitope recognized
(Roben et al., 1994) and has been observed
elsewhere (Nakamura et al., 1993). A 54-fold
improvement in neutralizing potency was determined for the highest affinity clone 3B3. The ability
of these evolved monovalent Fabs to neutralize with
potencies equivalent or better than sCD4 is unique.
In a recent multi-center study of human and mouse
anti-HIV-1 antibodies, no bivalent antibody has
demonstrated such potency (D’Souza et al., 1994).
Furthermore, four primary clinical isolates of HIV-1
not neutralized by the parent b4/12 were neutralized by the highest affinity clone selected, 3B3.
Clone 3B3 was improved 8.2-fold in binding gp120
for a final affinity of 0.77 nM. These experiments
suggest that improvement in the affinity of b4/12
should improve its therapeutic potency. To dissect
the relative contribution of the mutant CDR1-H and
CDR3-H of 3B3, CDR1-H was converted to
wild-type sequence in clone h3.3. This clone
retained the binding affinity of 3B3. This experiment suggested that the use of the pool of selected
CDR1-H variants in this experiment was not the
best starting point for further optimization. With
stringent selection higher affinity antibodies from
the original H1 library were selected as shown in
Table 4. Based on this result all subsequent CDR
walks reported in this study use a single optimized
clone on which diversity is incorporated, as
opposed to a collection of clones of selected but
ill-defined binding characteristics. In every case
examined, saturation mutagenesis within the CDRs
followed by selection using phage display resulted
in the improvement of affinity. Selection of clones
with sub-nanomolar binding affinities was facilitated by incorporating an off-rate biased selection in
the panning step (vida supra). Affinity gains at each
step of the walks were generally in small
increments, about fourfold, with the exception of the
two steps in CDR3-H which produced about an
eightfold improvement in each case. Three different
walks yielded clones of similarly high affinity
137-65 pM; 3B3/L1.4L3.11, 3B3/L3.14, and h1.3/
h3.33. The net improvement in affinity of these
clones was 46 to 96-fold.
If the free energy change of a multiple mutant (in
this case several optimized CDRs combined) is
nearly equal to the sum of the free energy changes
399
observed in the point mutants (or singly optimized
CDRs) the free energy changes are said to be
additive; DDGI is negligible. The reader should note
that additivity in free energy changes results in
multiplicative changes in Ka ; a fourfold improved
affinity in CDR1 combined with a fourfold
improved affinity in CDR2 results in a clone with a
16-fold improvement in affinity. In protein engineering studies this is generally the case unless
residues interact with each other by direct contact or
indirectly through electrostatic or structural perturbations (Wells, 1990). This (interaction) will certainly be the case for optimization within the long
CDR loops of CDR1-L, CDR2-H and CDR3-H
which require optimization of more than five
residues and at least two sequential optimizations
within the given CDR. Parallel CDR walking makes
the assumption that the optimized loops will exhibit
additivity in free energy changes when the
individually optimized loops are combined. The
advantage of a parallel approach is the speed with
which the final desired clone might be obtained. In
a study guided by a high resolution X-ray structure
and extensive alanine scanning mutagenesis an
additivity-based approach has been convincingly
demonstrated in the human growth hormone
system with the generation of a hormone with a
380-fold improvement in affinity for its receptor
(Lowman et al., 1991; Lowman & Wells, 1993).
Within the present system we are guided by the
data base of available antibody structures and hints
at somatic hotspots observed in chain shuffling
experiments. Within antibodies of known structure,
CDR-CDR contacts are well documented. However,
from crystallographic studies it is not clear how
significant these interactions are or to what extent
change in sequence in interacting regions will affect
affinity for antigen. Examples of the utilization of
additivity in the affinity optimization with antibodies have been reported (Foote & Winter, 1992;
Hawkins et al., 1992, 1993; Riechmann & Weill,
1993). The extent of mutation in the combined
segments in these reports is generally much lower
than that reported here.
Encouraged by these results, a parallel CDR
walking mutagenesis strategy was examined where
independently optimized CDRs were combined.
Table 5 summarizes the binding data for six
combinations of CDRs. Only in the combination of
CDR1-H with CDR3-H was additivity observed.
This result could not be rationalized by knowledge
of structures of antibodies, since these CDRs are
known to form many van der Waals’ contacts.
Changes within the CDR1-H in h1.1 are, however,
modest in comparison to selected changes in other
CDRs. Recently interchain CDR-CDR contacts in
antibodies of known structure have been tabulated
(Padlan, 1994). This study revealed that CDR3-L
generally forms van der Waals’ contacts with all the
CDRs of the heavy chain whereas CDR3-H
generally interacts with all the CDRs of the light
chain. CDR1-L, however, contacts predominately
CDR3-H. Overall the results of combinations were
400
unpredictable but did lead to some modest
improvements over the clones derived from
sequential selection. The large negative interaction
energies observed for most combinations given in
Table 5 are sufficient to undermine the small
0.82 kcal/mol increases which would accompany
the about fourfold increases in affinity of the
individual optimized CDRs. The failure of the
additivity-based strategy is most likely due to direct
interactions though long-range non-additive effects
are also possible (LiCata & Ackers, 1995). The
−2.02 kcal/mol interaction energy observed in the
construction of h1.1h3.33/L3.14 suggests that to a
large degree non-additivity is observed in these
cases due to the incompatibility, perhaps due to
steric interaction, directly or through antigen, of the
optimized CDR3-L regions with independently
optimized CDR3-H region. The highest affinity
clone produced by parallel optimization, h1.1h3.3/
L1.4L3.1 was improved 420-fold in affinity for
antigen. This is only a 4.4-fold improvement over
3B3/L1.4L3.11, the highest affinity clone obtained
in the sequential approach, and would likely have
been achieved in the next sequential optimization
step. The highest affinity Fab generated in these
studies, h1.1h3.3/L1.4L3.14, bound gp120 with a
15 pM dissociation constant. Since most of the CDR
sequence of this antibody remains to be optimized,
the limit to which one may improve affinity remains
to be reached.
Examination of the kinetics of binding using the
BIAcore instrument revealed that improvements in
affinity were dominated by slowing of the off-rate.
The highest affinity clone demonstrated a 400-fold
decrease in off-rate as compared to b4/12. The
BIAcore sensorgrams for Fabs b4/12 and h1.1h3.3/
L1.4L3.14 are given in Figure 2. The slowing in
off-rate observed here is in agreement with that
observed in the affinity maturation of human
growth hormone (Lowman & Wells, 1993) and is
expected since experimentally selective pressure is
provided to enrich for the most stable Fab/gp120
Affinity Maturation of a Human Antibody
complex. In comparison to the natural ligand for
CD4, the Fab fragments bound with a four to
eightfold faster on-rate than sCD4 and in the best
case, with an off-rate 83-fold slower than sCD4.
Further improvements in off-rates cannot be
measured with the BIAcore instrument which is
limited to off-rates faster than 10−7 s−1
This study demonstrates that the sequential CDR
walking strategy is an effective route to the
generation of very high affinity antibodies. The
large affinity increases observed in the optimization
of the CDR3-H region suggest that the most
expeditious route to improve affinity may be
optimization of this region. This is likely to be
particularly true for an antibody such as b4/12,
which has an extended CDR3-H of 18 residues.
Since human and murine CDR3-H regions have
mean lengths of approximately 12 and 9 residues,
respectively (Wu et al., 1993), which are generated
by mechanisms most analogous to saturation
mutagenesis (Sanz, 1991), it should not be surprising that the humoral response has not selected the
combinatorially most optimized segment in this
region.
The ability to create very high affinity antibodies
has implications for their use as catalysts, since
optimization of affinity for the transition-state of a
reaction is directly correlated with enhanced
catalytic efficiency (Lerner et al., 1991). For
therapeutic and prophylactic application, very high
affinity antibodies have important theoretical
advantages if potency, for example in viral
neutralization assays, is strictly correlated with
affinity. In therapeutic applications the quantity of
protein required could be drastically reduced. For
example, if a serum level of 25 mg/ml were required
for a given indication, a 420-fold improvement
would reduce the quantity of protein required to
0.06 mg/ml. In prophylactic applications, increased
potency could dramatically extend the time in
which the individual is protected from disease.
Given the 21 day half-life of an antibody in the body
Figure 2. Comparison of BIAcore
sensorgrams obtained for the binding of Fabs b4/12 and h1.1h3.33/
L1.4L3.14 to immobilized gp120.
RU, resonance units.
401
Affinity Maturation of a Human Antibody
(Capon et al., 1989), an antibody which provided
protection through a single half-life would, if
improved 420-fold and given in the same quantity,
provide protection for six months.
Materials and Methods
Reagents, strains and vectors
Oligonucleotides were from Operon Technologies
(Alameda, CA). Escherichia coli, phage, and the phagemid
vector pComb3H are as described (Barbas et al., 1991;
Barbas & Wagner, 1995)). Restriction enzymes were
obtained from Boehringer Mannheim. Taq polymerase
and DNA ligase were obtained from Promega and
Gibco-BRL, respectively. The recombinant protein gp120
IIIB was purchased from Intracel Corp. (Cambridge, MA).
Reagents for surface plasmon resonance experiments
were obtained from Pharmacia.
Construction and selection of libraries
The construction of library H1 has been described
(Barbas et al., 1994). This library was selected for gp120
binding by an additional six panning cycles beyond the
four cycles performed in the previous report. For
selections, gp120 IIIB was immobilized on Costar 3690
microtiter wells. One well, coated with 1 ag of gp120
overnight, was used for each round of selection. To
construct library H3B3 clone 3B3 was utilized as a
template for overlap PCR mutagenesis as described
(Barbas et al., 1994). The two fragments required for this
procedure were obtained with oligonucleotide primer
pairs FTX3 (Barbas et al., 1992a) and 3B3C350B
(5' - CCAGACGTCCATATAATAATTGTCMNNMNNMN NMNNMNNCCAACCCCACTCCCCCACTCT - 3') and
with R3B (Barbas et al., 1993b) and H4H3OF (5'-GACAATTATTATATGGACGTCTGG-3'), respectively. N = A, C, G,
or T,: M = A or C which gives K = T or G in the
complementary strand (NNK doping strategy). Following
fusion and gel purification, the fragment was digested
with restriction enzymes XhoI and SpeI, purified and
ligated into the phagemid vector pComb3H containing the
b4/12 light chain from which a 300 base-pair stuffer
fragment in the heavy chain cloning sites XhoI and SpeI
had been removed. The ligation products were introduced into XL1-Blue cells by electrotransformation to
yield 1 × 108 independent transformants. The library was
selected by panning against gp120 IIIB for four cycles. In
the fifth through the tenth cycles of selection, 8 mM Fab
3B3 was added to the well following removal of unbound
phage by washing. The plate was then incubated with the
Fab for three hours at 37°C (off-rate selection). The well
was then washed five times and bound phage were eluted
and propagated as usual. Library L1 was prepared in a
similar procedure using 3B3 as a template for overlap PCR
mutagenesis with oligonucleotide primer pairs KEF
(Barbas et al., 1993b) and HIVCDR1-OV-B (5'-AGGTTTGTGCTGGTACCAGGCTACMNNMNNMNNMNNMNNMNNGTGACTGGACCTACAGGAGAAGGT-3') and T7B
(Barbas et al., 1993b) and HIVCDR1-FO (5'-GTAGCCTGGTACCAGCACAAACCT-3'). Following fusion and gel
purification, the fragment was digested with restriction
enzymes SacI and XbaI, purified and ligated into the
phagemid vector pComb3H containing the 3B3 heavy
chain wherein a 1.2 kb stuffer fragment between SacI and
XbaI had been removed. The ligation products were
introduced into XL1-Blue cells by electrotransformation to
yield 5 × 108 independent transformants. Libraries L3A
and L3B were prepared with primer pairs KEF and
H4KCDR3-B0 (5'-CAGTTTGGTCCCCTGGCCAAAAGTGTAMNNMNNMNNMNNATAMNNCTGACAGTAGTACAGTGCAAAGTC-3') and T7B and HVKFR4-FO (5'TACACTTTTGGCCAGGGGACCAAACTG-3').
Library
L3A was prepared with 3B3/L1.4 as the PCR template
and library L3B was prepared with 3B3 as the template.
Library sizes were 2 × 108 and 4 × 108, respectively.
Library L3A was selected for binding to gp120 with five
cycles of panning followed by five cycles of panning with
3B3 used in an off-rate selection as described above.
Library L3B was selected with three panning cycles
followed by four cycles with Fab 3B3 off-rate selection.
Expression of soluble Fab
Following selection, phagemid DNA was isolated and
digested with SfiI and the approximately 1500 bp cassette
was ligated with SfiI-digested pPhoA. pPhoA was
constructed by insertion of the phoA promoter and
antibody encoding cassette between the AatII and HindIII
sites of pBR322. The phoA promoter was isolated from
genomic E. coli DNA by PCR. The antibody encoding
expression cassette was derived from pComb3H. The
pPhoA vector was designed to be compatible with
pComb3H and is based on the previously reported
pAK19 expression vector by Carter et al. (1991). This
vector is designed for use in fermentation, though for the
experiments reported here fermentation was not necessary. Expression of Fab in shaker flasks with this vector was
achieved by phosphate starvation as described (Carter
et al., 1992). In some cases transfer to pPhoA was not
performed. In these cases NheI/SpeI restriction enzyme
digestion and religation of the pComb3H vector leads to
efficient expression of soluble Fab as described (Barbas
et al., 1991). Following screening for binding activity in an
enzyme linked immunosorbant assay (ELISA), the
complete amino acid sequences of the variable regions of
ELISA positive clones were deduced by dideoxy
sequencing as described. Fab was purified by affinity
chromatography as described (Barbas et al., 1992b). To
examine if protein aggregates are isolated using this
protocol, an additional gel filtration purification step was
performed on Fabs b4/12 and h1.1h3.33. Gel filtration
chromatography was performed using a standardized
HiPrep 16/60 Sephacryl S-100 column equilibrated and
run (0.5 ml/min) in PBS buffer. Fabs eluted at 50 kDa and
were collected and used for BIAcore studies.
Recruitment of optimized complementarity
determining regions
Shuffling of optimized CDR regions was facilitated by
an encoded SacII site in framework three of the heavy
chain and a KpnI site in framework two of the light chain.
The heavy chain h1.1h3.33 was constructed from the
210 bp XhoI/SacII fragment of h1.1 and the 470 bp
SacII/SpeI fragment of h1.3B/h3.33. Light chain L14L3.11
was constructed from the 100 bp SacI/KpnI fragment of
L1.4 and the 550 bp KpnI/XbaI fragment of L3.14.
Combinations of mutant light and heavy chains were
constructed in the pPhoA vector.
Surface plasmon resonance
The kinetic constants for the binding of Fab to gp120
IIIB were determined by surface plasmon resonance-
402
based measurements using the BIAcore instrument from
Pharmacia. The sensor chip was activated for immobilization with N-hydroxysuccinimide and N-ethyl-N '-(3-diethyl aminopropyl) carbodiimide. The protein, gp120, was
coupled to the surface by injection of 50 ml of a 50 mg/ml
sample. Excess activated esters were quenched with 15 ml
of 1 M ethanolamine (pH 8.5). Typically 4000 resonance
units were immobilized. Binding of Fab fragments to
immobilized gp120 was studied by injection of Fab in a
range of concentrations (0.5 to 10 mg/ml) at a flow rate of
5 ml/min. The binding surface was regenerated with HCl,
1 M NaCl (pH 3) and remained active for 030
measurements. Each immobilization surface was standardized by determination of the binding kinetics of clone
3B3 so that relative values could be determined. The
association rate (kon ) was determined by studying the rate
of binding of the protein to the surface at five different
protein concentrations ranging from 10 mg/ml to 0.5 mg/
ml. The dissociation rate (koff ) was determined by
increasing flow rate to 50 ml/min after the association
phase. The koff value is the average of three measurements.
The kon and koff values were calculated using BIAcoreTM
kinetics evaluation software. The derived equilibrium
dissociation constants (Kd ) were calculated from the rate
constants. To exclude protein aggregation as a cause for
the slowing of the off-rates of mutant Fabs, the off-rates
of Fabs b4/12 and h1.1h3.33 were determined following
gel filtration purification to remove any aggregated
protein. The binding of gp120 to immobilized Fab b4/12
was also studied. Fab was immobilized and kinetics
determined as described above using five concentrations
of gp120 ranging from 20 mg/ml to 1.25 mg/ml to
determine association rate and 50 mg/ml injections to
determine off-rate.
Acknowledgements
We thank Mike Picque for molecular graphics, Richard
Lerner and Robyn L. Stanfield for helpful discussions,
and Ray Sweete for sCD4. We thank a reviewer for calling
the van der Merwe et al. reference to our attention. C.F.B.
is a Scholar of the American Foundation for AIDS
Research and recipient of an Investigator Award from The
Cancer Research Institute. This work was supported by
the National Institutes of Health grant RO1 AI 37470 to
C.F.B.
References
Barbas, C. F., III & Wagner, J. (1995). Synthetic human
antibodies: selecting and evolving functional proteins. Methods, 8, 94–103.
Barbas, C. F., III, Kang, A. S., Lerner, R. A. & Benkovic,
S. J. (1991). Assembly of combinatorial antibody
libraries on phage surfaces: The gene III site. Proc.
Natl Acad. Sci. USA, 88, 7978–7982.
Barbas, C. F., III, Bain, J. D., Hoekstra, D. M. & Lerner,
R. A. (1992a). Semisynthetic combinatorial antibody
libraries: a chemical solution to the diversity
problem. Proc. Natl Acad. Sci. USA, 89, 4457–4461.
Barbas, C. F., III, Björling, E., Chiodi, F., Dunlop, N.,
Cababa, D., Jones, T. M., Zebedee, S. L., Persson,
M. A. A., Nara, P. L., Norrby, E. & Burton, D. R.
(1992b). Recombinant human Fab fragments neutralize human type 1 immunodeficiency virus in vitro.
Proc. Natl Acad. Sci. USA, 89, 9339–9343.
Affinity Maturation of a Human Antibody
Barbas, C. F., III, Amberg, W., Simoncsits, A., Jones, T. M.
& Lerner, R. A. (1993a). Selection of human
anti-hapten antibodies from semisynthetic libraries.
Gene, 137, 57–62.
Barbas, C. F., III, Collet, T. A., Amberg, W., Roben, P.,
Binley, J. M., Hoekstra, D., Cababa, D., Jones, T. M.,
Williamson, R. A., Pilkington, G. R., Haigwood,
N. L., Cabezas, E., Satterthwait, A. C., Sanz, I. &
Burton, D. R. (1993b). Molecular profile of an
antibody response to HIV-1 as probed by combinatorial libraries. J. Mol. Biol. 230, 812–823.
Barbas, C. F., III, Hu, D., Dunlop, N., Sawyer, L., Cababa,
D., Hendry, R. M., Nara, P. L. & Burton, D. R. (1994).
In vitro evolution of a neutralizing human antibody
to human immunodeficiency virus type 1 to enhance
affinity and broaden strain cross-reactivity. Proc. Natl
Acad. Sci. USA, 91, 3809–3813.
Barbas, S. M. & Barbas, C. F., III (1994). Filamentous phage
display. Fibrinolysis, 8, 245–252.
Behring, E. A. (1893). Die Geschichte der Diphtherie.
pp. 186–187. Thieme, Leipzig.
Burton, D. R. & Barbas, C. F., III (1994). Human antibodies
from combinatorial libraries. Advan. Immunol. 57,
191–280.
Burton, D. R., Pyati, J., Koduri, R., Sharp, S. J., Thornton,
G. B., Parren, P. W. H. I., Sawyer, L. S. W., Hendry,
R. M., Dunlop, N., Nara, P. L., Lamacchia, M.,
Garratty, E., Stiehm, E. R., Bryston, Y. J., Cao, Y.,
Moore, J. P., Ho, D. D. & Barbas, C. F., III (1994).
Efficient neutralization of primary isolates of HIV-1
by a recombinant human monoclonal antibody.
Science, 266, 1024–1027.
Capon, D. J., Chamow, S. M., Mordenti, J., Marsters, S. A.,
Gregory, T., Mitsuya, H., Byrn, R. A., Lucas, C.,
Wurm, F. M., Groopman, J. E., Broder, S. & Smith,
D. H. (1989). Designing CD4 immunoadhesins for
AIDS therapy. Nature, 337, 525–531.
Carter, P., Kelley, R. F., Rodrigues, M. L., Snedecor, B.,
Covarrubias, M., Velligan, M. D., Wong, W. L. T.,
Rowland, A. M., Kotts, C. E., Carver, M. E., Yang, M.,
Bourell, J. H., Shepard, H. M. & Henner, D. (1991).
High level Escherichia coli expression and production
of a bivalent humanized antibody fragment. Bio/Technology, 10, 163–167.
Chothia, C., Lesk, A. M., Tramontano, A., Levitt, M.,
Smith-Gill, S. J., Air, G., Sheriff, S., Padlan, E. A.,
Davies, D., Tulip, W. R., Colman, P. M., Spinelli, S.,
Alzari, P. M. & Poljak, R. J. (1989). Conformations of
immunoglobulin hypervariable regions. Nature, 342,
877–883.
Chothia, C., Lesk, A. M., Gherardi, E., Tomlinson, I. M.,
Walter, G., Marks, J. M., Llewelyn, M. B. & Winter,
G. (1992). The structural repertiore of human VH
segments. J. Mol. Biol. 227, 799–817.
D’Souza, M. P., Geyer, S. J., Hanson, C. V., Hendry, M.,
Milman, G. & Collaborating Investigators (1994).
Evaluation of monoclonal antibodies to HIV-1 by
neutralization and binding assays: an international
collaboration. AIDS, 8, 169–181.
Foote, J. & Winter, G. (1992). Antibody framework
residues affecting the conformation of the hypervariable loops. J. Mol. Biol. 224, 487–499.
Hawkins, R. E., Russell, S. J. & Winter, G. (1992). Selection
of phage antibodies by binding affinity mimicking
affinity maturation. J. Mol. Biol. 226, 889–896.
Hawkins, R. E., Russell, S. J., Baier, M. & Winter, G.
(1993). The contribution of contact and non-contact
residues of antibody in the affinity of binding to
antigen. J. Mol. Biol. 234, 958–964.
403
Affinity Maturation of a Human Antibody
Kabat, E. A., Wu, T. T., Perry, H. M., Gottesman, K. S. &
Foeller, C. (1991). Sequences of Proteins of Immunological Interest. 5th edit., US Department of Health and
Human Services, Public Health Service, National
Institutes of Health (NIH Publication no. 91-3242).
Karlsson, R., Michaelsson, A. & Mattson, A. (1991).
Kinetic analysis of monoclonal antibody-antigen
interactions with a new biosensor based analytical
system. J. Immunol. Methods, 145, 229–240.
Kelley, R. F. & O’Connell, M. P. (1993). Thermodynamic
analysis of an antibody functional epitope. Biochemistry, 32, 6828–6835.
Lerner, R. A., Benkovic, S. J. & Schultz, P. G. (1991). At
the crossroads of chemistry and immunology:
catalytic antibodies. Science, 252, 659–667.
LiCata, V. J. & Ackers, G. K. (1995). Long-range, small
magnitude nonadditivity of mutational effects in
proteins. Biochemistry, 34, 3133–3139.
Lowman, H. B. & Wells, J. A. (1993). Affinity maturation
of human growth hormone by monovalent phage
display. J. Mol. Biol. 234, 564–578.
Lowman, H. B., Bass, S. H., Simpson, N. & Wells, J. A.
(1991). Selecting high-affinity binding proteins by
monovalent phage display. Biochemistry, 30, 10832–
10838.
Marquart, M., Deisenhofer, J., Huber, R. & Palm, W.
(1980). Crystallographic refinement and atomic
models of the intact immunoglobulin molecule Kol
and its antigen-binding fragment at 3.0 Å and 1.9 Å
resolution. J. Mol. Biol. 141, 369–391.
Nakamura, G. R., Byrn, R., Wiles, D. M., Fox, J. A., Hobbs,
M. R., Hastings, R., Wessling, H. C., Norcross, M. A.,
Fendly, B. M. & Berman, P. W. (1993). Strain
specificity and binding affinity requirements of
neutralizing monoclonal antibodies to the C4 domain
of gp120 from human immunodeficiency virus type
1. J. Virol. 67, 6179–6191.
Padlan, E. A. (1994). Anatomy of the antibody molecule.
Mol. Immunol. 31, 169–217.
Riechmann, L. & Weill, M. (1993). Phage display and
selection of a site-directed randomized single-chain
antibody Fv fragment for its affinity improvement.
Biochemistry, 32, 8848–8855.
Roben, P., Moore, J. P., Thali, M., Sodroski, J., Barbas, C. F.,
III & Burton, D. R. (1994). Recognition properties of
a panel of human recombinant Fab fragments to the
CD4 binding site of gp120 that show differing
abilities to neutralize human immunodeficiency
virus type 1. J. Virol. 68, 4821–4828.
Ryu, S.-E., Kwong, P. D., Truneh, A., Porter, T. G., Arthos,
J., Rosenberg, M., Dai, X., Xuong, N.-H., Axel, R.,
Sweet, R. W. & Hendrickson, W. A. (1990). Crystal
structure of an HIV-binding recombinant fragment of
human CD4. Nature, 348, 419–426.
Sanz, I. (1991). Multiple mechanisms participate in the
generation of diversity of human H chain CDR3
regions. J. Immunol. 147, 1720–1729.
van der Merwe, P. A., Brown, M. H., Davis, S. J. & Barclay,
A. N. (1993). Affinity and kinetic analysis of the
interaction of the cell adhesion molecules rat CD2
and CD48. EMBO J. 12, 4945–4954.
Wells, J. A. (1990). Perspectives in biochemistry:
additivity of mutational effects in proteins. Biochemistry, 29, 8509–8517.
Wilson, I. A. & Stanfield, R. L. (1993). Antibodyantigen interactions. Curr. Opinion Struct. Biol. 3,
113–118.
Wu, T. T., Johnson, G. & Kabat, E. A. (1993). Length
distribution of CDRH3 in antibodies. Protein: Struct.
Funct. Genet. 16, 1–7.
Edited by F. E. Cohen
(Received 10 April 1995; accepted in revised form 5 September 1995)