Stabilization and humanization of a single-chain Fv antibody fragment specific for human lymphocyte antigen CD19 by designed point mutations and CDR-grafting onto a human framework.
ABSTRACT A single-chain Fv (scFv) fragment derived from the murine antibody 4G7, specific for human lymphocyte CD19, was engineered for stability and expression in Escherichia coli in view of future use as a therapeutic protein. We compared two orthogonal knowledge-based procedures. In one approach, we designed a mutant with 14 single amino-acid substitutions predicted to correct destabilizing residues in the 4G7-wt sequence to create 4G7-mut. In the second variant, the murine CDRs were grafted to the human acceptor framework huVkappa3-huV(H)3, with 11 additional point mutations introduced to obtain a better match between CDR graft and acceptor framework, to arrive at 4G7-graft. Compared to 4G7-wt, 4G7-mut showed greater thermodynamic stability in guanidinium chloride-induced equilibrium denaturation experiments and somewhat greater stability in human serum. The loop graft maintained the comparatively high stability of the murine loop donor, but did not improve it further. Our analysis indicates that this is due to subtle strain introduced between CDRs and framework, mitigating the otherwise highly favorable properties of the human acceptor framework. This slight strain in the loop graft is also reflected in the binding affinities for CD19 on leukemic cells of 8.4 nM for 4G7-wt, 16.4 nM for 4G7-mut and 30.0 nM for 4G7-graft. This comparison of knowledge-based mutation and loop-grafting-based approaches will be important, when moving molecules forward to therapeutic applications.
- [show abstract] [hide abstract]
ABSTRACT: A cell line, designated SEM, was established from the peripheral blood of a 5-year-old girl in relapse with acute lymphoblastic leukaemia (ALL). Both the lymphoblasts of the patient and the cells of the cell line SEM showed the t(4;11) chromosomal rearrangement. The analysis of the immunophenotype of the SEM cell line revealed the B-cell differentiation antigens CD19, CD22 and CDw75 in the absence of CD20, CD24 and immunoglobulin expression. Besides B-lineage antigens, SEM cells were positive for the myeloid antigens CD13, CD15, CD33 and CDw65. Immunogenotypic analysis of SEM cells showed a monoclonal rearrangement of immunoglobulin heavy-chain (IgH). T-cell receptor (TCR) gamma and delta genes. Addition of interleukin (IL)-7 promoted the growth of the patient's lymphoblasts in culture and enhanced the proliferation of SEM cells. The SEM cells also express messenger RNA (mRNA) for the IL-7 receptor (IL-7R), but no evidence for autocrine production of IL-7 by the cell line was found. Addition of IL-4, tumour necrosis factor (TNF)-alpha, interferon (IFN)-alpha, or IFN-gamma resulted in a profound inhibition of SEM growth. Thus, these cytokines may have important growth regulatory activities for biphenotypic leukaemic ALL cells.British Journal of Haematology 03/1994; 86(2):275-83. · 4.94 Impact Factor
- [show abstract] [hide abstract]
ABSTRACT: A single chain antibody (scFv) was constructed from a hybridoma expressing the anti-CD34 monoclonal antibody My10. The scFv was expressed in the mouse fibroblast cell line NIH 3T3, and purified from culture supernatant via an epitope tag fused to the C-terminus of the protein. The scFv equilibrium dissociation constant (KD) was determined to be 2.4 X 10(-7) M using a fluorescence based flow cytometry assay involving recognition of the epitope tag, and bound approximately 24-fold less avidly to CD34 expressing KG-1a cells than the native antibody My10. This novel and previously unreported method for determining antibody binding affinity offers several advantages over alternative methods. It is rapid and simple, and unlike methods that directly label the antibody, it involves no covalent modifications of antibody variable domain residues that could potentially interfere with antigen binding. The KD for the anti-CD33 antibody HuG1 (Caron et al. (1992) The biological and immunological features of humanized M195 (anti-CD33) monoclonal antibodies. Cancer Res. 52, 6761-6767) was determined as well. The close agreement of this value and the previously reported value, determined by a radioligand competition method, validates the use of this assay for antibody affinity determination. We discuss various potential applications for this anti-CD34 scFv.Journal of Immunological Methods 03/1997; 201(2):223-31. · 2.23 Impact Factor
- [show abstract] [hide abstract]
ABSTRACT: Antibody-drug conjugates are targeted anticancer agents consisting of a cytotoxic drug covalently linked to a monoclonal antibody for tumor antigen-specific activity. Once bound to the target cell-surface antigen, the conjugate must be processed to release an active form of the drug, which can reach its intracellular target. Here, we used both biological and biochemical methods to better define this process for antibody-maytansinoid conjugates. In particular, we examined the metabolic fate in cells of huC242-maytansinoid conjugates containing either a disulfide linker (huC242-SPDB-DM4) or a thioether linker (huC242-SMCC-DM1). Using cell cycle analysis combined with lysosomal inhibitors, we showed that lysosomal processing is required for the activity of antibody-maytansinoid conjugates, irrespective of the linker. We also identified and characterized the released maytansinoid molecules from these conjugates, and measured their rate of release compared with the kinetics of cell cycle arrest. Both conjugates are efficiently degraded in lysosomes to yield metabolites consisting of the intact maytansinoid drug and linker attached to lysine. The lysine adduct is the sole metabolite from the thioether-linked conjugate. However, the lysine metabolite generated from the disulfide-linked conjugate is reduced and S-methylated to yield the lipophilic and potently cytotoxic metabolite, S-methyl-DM4. These findings provide insight into the mechanism of action of antibody-maytansinoid conjugates in general, and more specifically, identify a biochemical mechanism that may account for the significantly enhanced antitumor efficacy observed with disulfide-linked conjugates.Cancer Research 05/2006; 66(8):4426-33. · 8.65 Impact Factor
Stabilization and humanization of a single-chain Fv
antibody fragment specific for human lymphocyte
antigen CD19 by designed point mutations
and CDR-grafting onto a human framework
Markus Ku ¨gler1, Christoph Stein1, Michael Schwenkert1,
Domenica Saul1, Lena Vockentanz1,3, Thomas Huber2,4,
Svava K. Wetzel2, Oliver Scholz2, Andreas Plu ¨ckthun2,
Annemarie Honegger2,5and Georg H. Fey1
1Chair of Genetics, Institute of Biology, University of Erlangen-Nuremberg,
Erwin-Rommel-Strasse 3, D-91058 Erlangen, Germany,2Biochemisches
Institut der Universita ¨t Zu ¨rich, Winterthurerstrasse 190, CH-8057 Zu ¨rich,
Switzerland,3Present address: Max-Delbru ¨ck-Centrum fu ¨r Molekulare
Medizin, Robert-Ro ¨ssle-Straße 10, D-13125 Berlin-Buch, Germany and
4Present address: Novartis Biologics/Protein Design, WSJ-506.3.13, Novartis
Pharma AG, CH-4002 Basel, Switzerland.
5To whom correspondence should be addressed.
A single-chain Fv (scFv) fragment derived from the
murine antibody 4G7, specific for human lymphocyte
CD19, was engineered for stability and expression in
Escherichia coli in view of future use as a therapeutic
protein. We compared two orthogonal knowledge-based
procedures. In one approach, we designed a mutant with
14 single amino-acid substitutions predicted to correct
destabilizing residues in the 4G7-wt sequence to create
4G7-mut. In the second variant, the murine CDRs were
grafted to the human acceptor framework huVk3-huVH3,
with 11 additional point mutations introduced to obtain a
better match between CDR graft and acceptor frame-
work, to arrive at 4G7-graft. Compared to 4G7-wt,
4G7-mut showed greater thermodynamic stability in gua-
experiments and somewhat greater stability in human
serum. The loop graft maintained the comparatively high
stability of the murine loop donor, but did not improve it
further. Our analysis indicates that this is due to subtle
strain introduced between CDRs and framework, mitigat-
ing the otherwise highly favorable properties of the
human acceptor framework. This slight strain in the
loop graft is also reflected in the binding affinities for
CD19 on leukemic cells of 8.4 nM for 4G7-wt, 16.4 nM
for 4G7-mut and 30.0 nM for 4G7-graft. This comparison
of knowledge-based mutation and loop-grafting-based
approaches will be important, when moving molecules
forward to therapeutic applications.
Keywords: antibody engineering/CDR graft/immunoglobulin
variable domains/scFv fragment/stability
The surface antigen CD19, a 95 kDa transmembrane glyco-
protein of the immunoglobulin superfamily, is selectively
expressed on mammalian B-lymphoid cells in most stages of
differentiation from the early pro-B-cell stage on, but not on
plasma cells and hematopoietic stem cells. Together with
CD21, CD81 and Leu13 (CD225), CD19 forms the B-cell
receptor co-complex and plays a role in the control of differ-
entiation, activation and proliferation of B-lymphoid cells
(Sato et al., 1997). The antigen is present on the blasts of
different types of human B-cell malignancies including pro-
common ALL (cALL) of children and young adults,
non-Hodgkin-lymphomas (NHL), chronic B-lymphocytic
leukemia (B-CLL) and hairy-cell leukemia (HCL) (Nadler
et al., 1983). It is not shed from malignant cells and is inter-
nalized after binding of antibodies and single-chain Fv
(scFv) antibody fragments (Press et al., 1989). Antigen
density ranges from 10 000 to 30 000 molecules per cell on
healthy peripheral B-cells, and from 7000 to 30 000 mol-
ecules per cell on malignant cells from a variety of lymphoid
cancers (Olejniczak et al., 2006). Due to these properties,
CD19 is an attractive target for the design of antibody-based
therapeutics for the treatment of B-cell neoplasias (Grossbard
et al., 1992). In addition, agents targeting CD19 may also
become useful for the treatment of chronic inflammatory dis-
eases and some autoimmune disorders.
A number of antibody derivatives targeting CD19 have
been tested for the treatment of B-cell cancers, so far with
modest success. Unconjugated CD19 antibodies produced
promising results in pre-clinical tests, but no therapeutic
benefit in clinical trials (Hekman et al., 1991; Pietersz et al.,
1995; Lang et al., 2004; Yazawa et al., 2005). Immunotoxins
consisting of plant or bacterial toxins coupled either to whole
antibodies or fused to scFv fragments specific for a number
of surface antigens of B-lymphoid cells were designed
(Pastan et al., 2007). Ricin derivatives coupled to whole IgGs
specific for CD19 were extensively tested, but have not pro-
gressed beyond phase II clinical studies due to serious side
effects (Messmann et al., 2000; Schindler et al., 2001).
Recombinant immunotoxins consisting of CD19-specific
scFv-fragments coupled to saporin and diphtheria toxin are
under current investigation (Vallera et al., 2005; Flavell et al.,
2006), but have not been advanced to clinical testing. A
fusion protein between an anti-CD19 scFv and a fragment of
Pseudomonas Exotoxin A was effective in pre-clinical tests
(Schwemmlein et al., 2007), and investigators have conju-
gated potent novel small-molecule toxins such as duocarmy-
cins and maytansinoids (Erickson et al., 2006; Chari, 2008) to
anti-CD19 antibodies. Finally, CD19-specific scFv fragments
have been coupled to death-effectors such as FAS- and
TRAIL-ligands, and a CD19 scFv fused to a soluble
TRAIL-ligand (sTRAIL) had attractive pre-clinical activities
on human malignant cells (Stieglmaier et al., 2008).
A number of different bispecific formats targeting CD19 in
combination with different specificities were examined as
# The Author 2009. Published by Oxford University Press. All rights reserved.
For Permissions, please e-mail: email@example.com
Protein Engineering, Design & Selection vol. 22 no. 3 pp. 135–147, 2009
Published online February 1, 2009 doi:10.1093/protein/gzn079
well. Full-length antibodies and Fab-fragments targeting
CD19 were effective in pre-clinical studies, but not in clinical
trials (Peipp and Valerius, 2002). Recombinant bispecific
scFv-derivatives targeting CD19 have also been tested for
recruiting Natural Killer (NK) cells as effectors for tumor cell
lysis by binding to the low affinity FcgRIIIa receptor (CD16)
on NK-cells, monocytes and macrophages (Kipriyanov et al.,
2002; Bruenke et al., 2005). This has been achieved either in
a diabody format (Kipriyanov et al., 2002) or by tandemly
arranged scFv fragments, each encoding an interchain disul-
fide bond (Bruenke et al., 2005). A variant of this format are
tandem diabodies (‘tandabs’), non-covalently associated two-
chain molecules, which form two scFv binding sites each for
the tumor antigen and the trigger molecule on the effector
cell. Such a molecule specific for CD19 and CD3 was effec-
tive in pre-clinical tests (Bruenke et al., 2005). Another
CD19-specific scFvs plus a central CD16-specific scFv in a
tandem linear arrangement (Kellner et al., 2008) (termed
triplebody). In pre-clinical studies, this molecule produced
potent ADCC (antibody dependent cellular cytotoxicity) reac-
tions with human leukemic cells as targets and enriched
human NK-cells as effectors, and the format is therefore
attractive for further development towards clinical testing.
The bispecific molecule targeting CD19 that currently has
advanced furthest is Blinatumomab, a tandem scFv carrying
specificities for CD19 and CD3, a cofactor of the T-cell
receptor on human T lymphocytes. The anti-CD3 specificity
recruits and activates T-cells for tumor cell lysis, regardless of
the antigen-specificity of their T-cell receptors, and the
T-cells have high cytotoxic potential. The molecule produced
non-Hodgkin lymphoma patients and is promising for clinical
development (Bargou et al., 2008). Nevertheless, for some
applications, such as the treatment of pediatric ALL patients
in a post-transplantation stage, T-cells may not be present in
sufficient numbers to serve as an efficient effector cell popu-
lation. Therefore, a need remains to develop CD19-specific
agents in other formats, such as immunotoxins, immunocon-
jugates with other types of death-effectors and bispecific mol-
ecules recruiting effectors other than T-cells. Moreover, all of
the above formats will profit from well designed, stable scFv
derivatives with human sequences, in order to avoid pro-
duction problems for clinical trials, and loss of activity after
prolonged incubation in human serum due to denaturation and
aggregation. It should be pointed out that many of the above
molecules could not even be produced in Escherichia coli,
but the investigators had to resort to production in mammalian
cells, suggesting an aggregation propensity of the molecules.
For therapeutic applications in humans, CD19-specific
scFv antibody fragments need to have high thermodynamic
stability and should be humanized to avoid a neutralizing
antibody response (Mirick et al., 2004). The antigenicity
problem is less severe for agents used to treat B-cell neopla-
sias, because these patients have an impaired B-cell compart-
ment and produce fewer neutralizing antibodies, allowing the
use of bacterial toxins (Pastan et al., 2007). However, some
patients may require CD19-specific agents for extended
periods of time. While human antibodies can be derived
from genetically modified mice carrying immunoglobulin
loci (Lonberg, 2005) or by screening synthetic phage display
libraries carrying natural or designed human antibody
fragments (Knappik et al., 2000; Hoogenboom, 2005), in this
study, we wished to start from a well characterized murine
hybridoma, termed 4G7 (Meeker et al., 1984).
The scFv derived from the antibody 4G7 had already
shown favorable biological activities in bispecific scFv con-
structs (Bruenke et al., 2005). We analyzed its biophysical
properties in the present study and found them to be already
rather favorable. Nonetheless, to test whether the molecule
can be improved further in its biophysical properties, we
took two orthogonal approaches. We first analyzed the
murine scFv for sequence features associated with poor
folding and low stability and designed a series of 14 point
mutations addressing two major and multiple minor pro-
blems, while preserving the structural characteristics of the
original antibody variable domains. In parallel, we designed
a CDR graft to a human framework. We used the most stable
of the human germline family consensus frameworks,
huVk3-huVH3, despite the fact that the murine and human
VLsequence showed only 53% sequence identity, and the
two VH domains shared a mere 51% sequence identity.
Additionally, in this graft, we had a mismatch between the
structural subtypes (Honegger et al., 2009) of VHCDR donor
and acceptor framework, which may be responsible for the
fact that such grafts frequently fail to reach the degree of
stabilization expected from the superior properties of the
acceptor framework (Willuda et al., 1999). In the companion
paper, we have directly addressed the antibody engineering
problems caused by such a mismatch in a series of chimeric
VHconstructs (Honegger et al., 2009). In the present paper,
we compare the stabilization that can be achieved with a
limited number of point mutations introduced into the orig-
inal framework to the stability achieved by grafting to a mis-
matched framework of otherwise proven stability, and
compare both of these proteins to the 4G7-wt.
Materials and methods
Homology modeling of the 4G7 Fv fragment
A homology model of the 4G7 Fv fragment was built
based on the structures of the VLdomain of the esterolytic
murine antibody Fab fragment MS6-12 [PDB entry 1MJU
(Ruzheinikov et al., 2003)] and the VHdomain structures of
an anti-ssDNA Fab fragment [1P7K (Schuermann et al.,
(Piscitelli et al., 2006)] and the Fab fragment of bactericidal
antibody MN12H2 [1MNU (van den Elsen et al., 1999)].
Modeling of the huVk3-huVH3 germline consensus Fv has
been reported previously (Knappik et al., 2000). The two
models were combined to produce the model of the CDR graft
construct 4G7-graft in a relative domain orientation intermedi-
ate between those observed in the two antibody fragments
serving as primary template for the VLand the VHdomain.
Culture of eukaryotic cells
Leukemia-derived SEM (t(4;11)-positive pro-B-ALL; (Greil
et al., 1994)) and CEM cells (T-ALL; DSMZ, German
Collection of Microorganisms and Cell Lines, Braunschweig,
Germany) were cultured in RPMI 1640-Glutamax-I medium
(Invitrogen, Karlsruhe, Germany), supplemented with 10%
fetal calf serum and penicillin and streptomycin (Invitrogen)
at concentrations of 100 U/ml and 100 mg/ml, respectively.
M.Ku ¨gler et al.
Bacterial strains and plasmids
Escherichia coli XL1-Blue (Stratagene, Amsterdam, The
Netherlands) was used for the amplification of plasmids and
cloning. The expression vector pAK400 (Krebber et al.,
1997) was used for the expression of antibody fragments in
E.coli HB2151 (Carter et al., 1985) (from Dr G. Winter,
MRC Cambridge, UK) and the vector pASK4 (IBA,
Go ¨ttingen, Germany) was used as an intermediate cloning
vector for the construction of the mutant 4G7-mut.
Construction of expression vectors
The sequence coding for the murine anti-CD19-4G7 wild-
type scFv (4G7-wt) was excised as an SfiI cassette from a
previously cloned pAK100 vector construct (Krebber et al.,
1997; Peipp et al., 2004). It is of the format VL-(Gly4Ser)4-
VH. The cassette was inserted by directional cloning into the
prokaryotic expression vector pAK400, which provides an
N-terminal FLAG-tag and a C-terminal His6-tag. For con-
struction of the mutant 4G7-mut, the SfiI cassette was trans-
ferred from 4G7-wt into the intermediate vector pASK4. The
14 amino acid substitutions were introduced by PCR reac-
tions (site-directed mutagenesis) and oligonucleotide anneal-
ing using the QuickChangewMulti SiteDirected Mutagenesis
Kit (Stratagene, La Jolla, CA, USA) according to the manu-
facturer’s instructions. For the oligonucleotide annealing
experiments, complementary primers were designed contain-
ing the desired mutations and specific restriction site over-
laps. After annealing by heating a primer mixture to 968C
and subsequent cooling to 258C for 30 min, these overlaps
were used to ligate the complementary primers into a corre-
spondingly digested vector. Correct incorporation of the
desired mutations was checked by DNA sequencing before
the sequence coding for 4G7-mut was re-cloned into the
expression vector pAK400.
The gene fortheloopgrafted
(huVH3-4G7-graft), including codon optimization for the
(Regensburg, Germany). An SfiI cassette containing the
huVH3-4G7-graft coding sequence was then excised from
the commercial standard vector and subcloned into the
expression vector pAK400.
Soluble periplasmic expression
For soluble expression of the 4G7 scFvs, expression plasmids
were propagated in E.coli HB2151. Overnight cultures were
incubated at 378C and diluted in Tryptone Broth medium
[TB; 12 g/l bacto-tryptone, 24 g/l bacto-yeast extract, 4 ml/l
glycerol, 100 ml/l 10? TB phosphate (0.17 M KH2PO4, 0.72
M K2HPO4)] supplemented with 1% glucose and 30 mg/ml
chloramphenicol until an extinction of 0.2 at 600 nm (OD600)
was reached. Cultures were incubated at 288C and expression
was induced at an OD600of 0.8 by addition of isopropyl-b-
D-thiogalactopyranoside (IPTG) to a final concentration of
1 mM. After 18 h of incubation, cells were collected by cen-
trifugation, resuspended in 1/10 volume of homogenization
buffer containing 50 mM NaH2PO4,pH 8.0, 300 mM NaCl
and 10 mM imidazole. After disruption of cells by high-
Canada), the crude extract was centrifuged to remove debris
(28 000g, 60 min at 48C) and the supernatant was used for a
Two-step purification of scFvs
Purification of His6-tagged scFv fragments was achieved by
affinity chromatography with a nickel-nitrilotriacetic acid
(Ni-NTA, Qiagen) matrix. For this purpose, 400 ml NiNTA
slurry was added per 100 ml of supernatant, and the mixture
was incubated for 30 min at 258C. The suspension was trans-
ferred to an empty column (BioRad) and the matrix was
washed extensively with MHA high-salt wash buffer [20 mM
MHA (6.6 mM MES, 6.6 mM HEPES, 6.6 mM sodium
acetate), pH 7.0, 1 M NaCl], then with MHA low-salt wash
buffer (20 mM MHA, pH 7.0, 100 mM NaCl) and finally
with MHA imidazole wash buffer (20 mM MHA, pH 7.0,
30 mM imidazole) to remove unspecifically bound proteins.
The His-tagged antibody fragments were then eluted by
adding 250 mM imidazole. The eluate was extensively dia-
lyzed against 20 mM MHA, containing 60 mM imidazole,
and was applied to a HiTrap SP XL column (GE Healthcare,
Freiburg, Germany) for ion exchange chromatography.
Loading was performed in 20 mM MHA (pH 7.0, 60 mM
imidazole) using an A¨KTA Prime chromatography system
(Amersham Pharmacia Biotech, Freiburg, Germany). A NaCl
gradient from 0 to 500 mM was used to elute the scFvs from
the column. Collected fractions were checked for the pre-
sence of scFvs by SDS–PAGE analysis and pooled fractions
were extensively dialyzed against 50 mM Tris–HCl (pH
7.2). The purity of the antibody fragments was checked by
SDS–PAGE analysis and the concentration was determined
by measuring the absorbance at 280 nm. The second purifi-
cation step removed the 22 kDa SlyD, which carries a natural
his-rich sequence (Wu ¨lfing et al., 1994).
Measurement of expression yields
To determine expression yields of the 4G7 scFvs, aliquots of
the Ni-NTA (Qiagen) purified and dialyzed proteins were
analyzed by SDS–PAGE. The purity was assessed and the
concentration was estimated by densitometric comparison to
bovine serum albumin standards. Expression yields were cal-
culated as mg of purified scFv per liter of E.coli culture and
microgram of purified scFv per gram of E.coli pellet.
SDS–PAGE and western blot analysis
SDS–PAGE was carried out under reducing conditions
following standard procedures. Gels were stained with
Coomassie Brilliant Blue R250. For western-blot analysis,
scFv fragments were detected with a Penta-His antibody
(Qiagen) and a secondary goat anti-mouse IgG antibody
coupled to horseradish peroxidase (Dianova, Hamburg,
Germany). Blots were developed with the enhanced chemilu-
Analytical Gel-filtration chromatography
For the determination of the aggregation tendency of purified
scFvs, samples were analyzed by size exclusion chromato-
graphy on a Superdex75 column 10/300GL (GE Healthcare)
equilibrated with 50 mM sodium phosphate (pH 7.0) and
100 mM NaCl on an Agilent 1100 HPLC system (Agilent).
Samples were injected at a concentration of 0.5 mg/ml
(17 mM) in a volume of 50 ml, the flow rate was 0.5 ml/min
and elution was followed by the detection of the absorbance
at 280 nm. A mini-DAWN multi-angle light scattering
Humanization and stabilization of a CD19-specific scFv
(MALS) detector (Wyatt Technology) and a differential
refractometer (Optilab rEX, Wyatt Technology) were used
for the determination of the absolute molar mass of the scFv
variants. The concentration was calculated assuming a dn/dc
Flow cytometric analysis and competition binding studies
Binding of scFvs to CD19 positive SEM cells was analyzed
on a FACS Calibur instrument using CellQuest software
(Becton Dickinson, Heidelberg, Germany) following pub-
lished procedures (Bruenke et al., 2004). Ten thousand
events were collected for each sample and all analyses of
whole cells were performed using appropriate scatter gates to
exclude cellular debris and aggregates. The scFvs were
detected using an Anti-Penta-His antibody and a phycoery-
thrin (PE)-conjugated goat anti-mouse IgG antibody (DAKO
Diagnostica, Hamburg, Germany). For competition binding
experiments, cells were pre-incubated with a 50-fold molar
excess of either the parental monoclonal antibody 4G7 or a
mouse IgG1 isotype control antibody for 15 min before
addition of the scFvs.
Determination of equilibrium dissociation constants (KD)
by flow cytometry
Equilibrium dissociation constants (KD) were measured by
flow cytometry following published procedures (Benedict
et al., 1997). Experiments were repeated six times and arith-
metic mean values+standard error of the mean are reported.
Data were fitted to the approximate equation MFI ¼ MFImax.
(GraphPad Software Inc., San Diego, CA, USA), where MFI
is the mean fluorescent intensity and MFImax is the fitted
plateau value at high concentrations of scFv fragment.
Measurement of in vitro serum stability
The scFvs were separately incubated in human serum at con-
centrations of 1.5 mg/ml in a final volume of 30 ml at 378C.
After defined time periods of up to 4 weeks, the residual
binding activity of the incubated scFvs was determined by
flow cytometry. Binding was compared to samples with-
drawn right after the incubation period had started, which
were set to 100%. The experiment was repeated four times
and the average half-life values were calculated from fitting
to a monoexponential decay.
A FluoroMax-4 spectrofluorometer (HORIBA Jobin Yvon,
Inc., NJ, USA) was used to record fluorescence spectra at
258C with slit widths of 2 nm for excitation and emission.
Protein samples dissolved in guanidinium chloride (GdmCl)
(1.0 ml) with a final protein concentration of 0.3 mM and
denaturant concentrations from 0 to 5.0 M were prepared
from a GdmCl stock solution (6.0 M, in 0.05 M Tris–HCl
(pH 7.2), 100 mM NaCl). The final GdmCl concentration of
each sample was calculated from its refractive index.
Fluorescence emission spectra from 320 to 370 nm were
recorded with an excitation wavelength of 280 nm after over-
night incubation of the samples at 258C. Fluorescence emis-
sion maxima were determined by fitting the fluorescence
emission spectra to a Gaussian function, and plotted against
the GdmCl concentration. Relative emission maxima were
calculated by setting the highest value to 100 and the lowest
to 0. Alternatively, fluorescence emission maxima were
determined by fitting the fluorescence emission spectra with
a Taylor expansion to the fourth power term according to
(Monsellier and Bedouelle, 2005) as described in the
Supplementary data available at PEDS online. Protein stab-
ilities (DG, m-values) were evaluated by the linear extra-
polation method according to published procedures (Santoro
and Bolen, 1988; Ja ¨ger et al., 2001) but it was found that a
two-state model does not hold.
Homology modeling of 4G7-wt
The VLdomain of 4G7 is most likely derived from murine
germline family muVk2, presumably germline muVk2–137
[IMGT database (Lefranc et al., 2005), AJ231263 (Schable
et al., 1999), 98% sequence identity, 99% sequence simi-
larity to 4G7] (Fig. 1), and was modeled against template
structure 1MJU (96% identity, 98% similarity, 1.2 A˚resol-
ution) (Ruzheinikov et al., 2003). The high degree of
sequence identity made modeling of this domain unproble-
matic, with only three amino acid substitutions in the puta-
tive antigen binding site and two in the framework: the fully
exposed His L33 [All residue numbers are given in the AHo
[(Honegger and Plu ¨ckthun, 2001a) numbering scheme, in
which structurally equivalent positions in VLand VHcarry
the same residue number. The correspondence between the
AHo and the Kabat numbering schemes (Kabat et al., 1991)
is indicated in the header of the sequence alignment (Fig. 1)]
in CDR-L1 was replaced in the model by Asn, Leu L107
in CDR-L3 by Met, Phe L137 in CDR-L3 by Leu.
Framework substitutions Val L11 Ile and Leu L147 Ile, com-
pared to the template, both affected semi-exposed positions
that could be accommodated in the model without additional
To model the VHdomain of 4G7, derived from germline
(Lefranc et al., 2005) (AF304557, 91% identity, 93% simi-
larity) (Fig. 1), multiple templates were used. PDB entry 1P7K
[1.8 A˚ resolution, 86% sequence identity, 87% similarity
(Schuermann et al., 2004)] was used as the primary template.
In the region between Ser H7 and Cys H106, template and
target sequences were identical. However, an additional tem-
plate [1XGY, 2.7 A˚resolution (Piscitelli et al., 2006)] had to
be used to model the conformation of the N-terminal residues,
since the N-terminal sequence of the VHdomain in structure
1P7K does not fit a muVH1-derived sequence. With a gluta-
mate in position H6 and a Pro in H10, 1P7K shows the
sequence signature and conformation of a type I framework 1
kink conformation (Honegger and Plu ¨ckthun, 2001b), while
4G7 shows the H6 Gln, H7 Ser, H10 Pro signature sequence of
a type III kink. Such N-terminal sequence mismatches are fre-
quently caused by the degenerate primers used for the cloning
of antibody variable domains (Honegger and Plu ¨ckthun,
2001b; Jung et al., 2001). Although these sequence differences
were concentrated in the region from H1 to H6, they predomi-
nantly affect the framework backbone conformation in the
region from H7 to H11. Therefore, the conformation of frame-
work 1 residues H1–H22 was taken from template 1XGY.
CDR-H3 and framework 4 of 4G7 were modeled using 1MNU
[2.5 A˚resolution (van den Elsen et al., 1999)] as template.
M.Ku ¨gler et al.
The two template Fvs were aligned by a least squares fit
of the structurally conserved positions of both the VLand the
VHdomains, and the VLdomain model of 4G7 aligned to
the VLdomain of 1MJU, the VHdomain model to the VH
domain of 1P7K, resulting in a relative domain orientation of
the model intermediate between those of the two templates.
Some adjustments in side-chain rotamers of residues in the
VL/VH interface and in the CDR-H3 conformation were
necessary to avoid steric clashes.
Identification of potentially destabilizing features in 4G7-wt
Sequence and structure model of 4G7 were scrutinized to
identify potentially destabilizing features (Honegger, 2008)
Fig. 1. Sequence alignment of the 4G7-wt VLand VHsequences (denoted 4G7-wt) to their presumed parental germline precursors (muVk2-137(AJ231263),
muVH1S136 (AF304557)) and germline family consensus sequence (muVk2 GL consensus, muVH1 GL consensus), to the point mutant construct 4G7-mut
(4G7-mut), the CDR-grafted variant 4G7-graft (huVk3*4G7-graft, huVH3*4G7-graft) and the graft acceptor framework (huVk3 consensus, huVH3 consensus)
representing the human germline family consensus sequences for the huVk3 and the huVH3 families, with CDR-L3 and -H3 derived from humanized antibody
hu4D5–8. In the header, the residue numbering according to Kabat et al. (1991) and AHo (Honegger and Plu ¨ckthun, 2001a) are indicated, the extent of the
CDRs according to the definition of Kabat et al. (1991) (white background) and the structurally least variant positions used for the least-squares superposition
of variable domain structures (dark gray background). A color code (line Protein Ag Contact) indicates the probability of the different sequence positions to be
involved in contacts to a protein antigen based on the average reduction of solvent accessible surface area (Honegger and Plu ¨ckthun, 2001a): white; no contact,
yellow; low probability to red; very high probability. Sequence positions identical in 4G7-wt and in huVk3-huVH3 are indicated in white letters on a dark blue
background, sequence positions identical to 4G7-wt but different from huVk3-huVH3 by black letters on magenta, positions identical to huVk3-huVH3 and
different from 4G7-wt by cyan, positions differing from either sequence are highlighted in yellow. Closed triangles point at sequence positions altered in
4G7-mut, open triangles at framework positions deviating from the sequence of the acceptor framework in 4G7-graft.
Humanization and stabilization of a CD19-specific scFv
(Figs 1 and 2A). Two principal problem spots became appar-
ent in the 4G7-wt scFv.
Light chain positions l7, l8 and l9 forming the framework 1
kink The muVk2 light-chain germline family shows an
unusual conformation of the framework 1 kink between
strand a0and a00: Residue L8 is a cis-Pro in .99% of the
human and 84% of the murine Vk sequences. In muVk2-
derived VL domains, L8 is an Ala, however, and L9 a
trans-Pro. In the 4G7-wt sequence, L7 is an Ala, L8 an Ala,
L9 a Pro. The presence of a cis-Pro in L8 has been shown to
improve the folding behavior and stability of scFvs: from a
library of an scFv fully randomized in positions L7, L8 and
L9, the S-P-S amino acid combinations most frequently seen
in natural antibodies also prevailed (Spada et al., 1998).
Sequences containing Pro in position L8 were preferred over
other amino acids in this position and were produced with a
higher yield of native protein upon periplasmic expression.
Reanalysis of the raw data (Spada et al., 1998) showed that
all of the selected clones showing a Pro in position L9 also
had a Pro in L8, contrary to the situation observed in 4G7.
Similarly, in an in-vitro evolution experiment aimed at
increasing the stability of the disulfide-free scFv AB48,
clones containing the mutation from T-T-S to T-P-S were
generated and enriched (Proba et al., 1997; Proba et al.,
1998). Surface residue Ser L7, present in 96% of human and
70% of murine Vk sequences, forms a hydrogen bond to its
own main chain CO, facilitating the formation of the chain
kink between strands a0and a00. Of the sequences lacking this
Ser, the majority have a Thr in this position that can also
form this hydrogen bond. In the point mutation construct
4G7-mut, we therefore replaced the L7-L9 AAP motif of
4G7 by SPS.
Heavy chain position H45, H77 and H97 in the charge
cluster In antibody variable domains, a highly conserved
cluster of charged residues is buried or semi-buried in the
lower core of the domain. Arg H77 and Asp H100 form the
central core of this cluster, interacting in a doubly hydrogen-
bonded salt bridge. Charged residues Arg H45, Glu H53,
Arg H97 and Glu H99 surround this central salt bridge to
form the charge cluster, while highly conserved Tyr H104
helps to orient the Asp H100 side chain by forming a hydro-
gen bond to its main-chain carbonyl group. While 86% of all
human VHdomains have an arginine in position H77, 56%
of the murine VHsequences have a lysine residue in this pos-
ition. Such domains are significantly stabilized by having
Arg H77 instead of Lys (Lindner et al., 1997; Proba et al.,
1997; Proba et al., 1998; Wo ¨rn et al., 2000). The Arg
H97-Glu H99 ion pair is more solvent-exposed than the Arg
H77-Asp H100 salt bridge. However, the interaction of Glu
H99 with Arg H87 and with Arg H45 probably keeps the
glutamate from interacting with Arg H77 and thus from
opening up the charge cluster to the solvent. In 4G7-wt, H45
and H77 are both lysines, H97 a threonine. In 4G7-mut, we
therefore replaced the three residues with arginines to opti-
mize the charge cluster.
Surface mutations with a track record Besides these two
major problem spots described above, a few minor changes
of surface residues were introduced that had proven ben-
eficial in other antibody fragments: residue L15 is part of a
semi-buried cluster of hydrophobic amino acids. The substi-
tution of Ala L15 by Leu stabilized the MCPC603 VL
domain by 7.0 kJ/mol, Ala L15 Pro by 4.5 kJ/mol (Ohage
et al., 1997); the inverse substitution of Leu L15 by Pro
destabilized the LEN VL domain by 7.1 kJ/mol (Raffen
et al., 1999). In 4G7-mut, we therefore replaced Pro L15
H5 is a fully exposed surface residue. However, the well-
expressed huVH3, huVH1 and huVH5 human germline con-
sensus frameworks have a Val in this position (Ewert et al.,
2003b), while the poorly expressed huVH2, huVH4 and
huVH6 have a Lys (VH2) or Gln (VH4, VH6) in this position.
The mutation of Gln to Val in two different antibodies
derived from the human VH6 germline family resulted in
1.7- and 2.6-fold increase in production yield (soluble peri-
plasmic protein) and a stabilization shifting the midpoint of
the GdmCl denaturation curve ([GdmCl]50) by 0.1 and
0.15 M, respectively (Ewert et al., 2003a). In addition, H5
Val was selected over Gln, Leu and Glu in the phage display
selection of 4D5-Flu for improved stability (Jung et al.,
Fig. 2. Structural models of 4G7-mut and 4G7-graft. (A) Amino acid
substitutions in 4G7-mut. The model structure of the scFv fragment shows
the secondary structure of the VLdomain (magenta) and the VHdomain
(blue). The 14 amino acid substitutions are highlighted in yellow, 8 in the
VLand 6 in the VHdomain. (B) Three-dimensional model of 4G7-graft.
Residues derived from the CDR donor (murine 4G7-wt) are highlighted in
magenta, residues derived from the framework donor (human VH3 and Vk3
domains) in cyan, and identical residues between both donors are
highlighted in blue, respectively. Left: view from the ‘top’ (putative antigen
binding site), right, view from the ‘bottom’ (in a Fab fragment, the V/C
interface), bottom center: view from the ‘front’.
M.Ku ¨gler et al.
1999). A possible reason for the beneficial effect of Val in
this solvent-exposed position lies in the particularly high
b-sheet propensity of b-branched amino acids, which might
help to stabilize the short N-terminal a0b-strand of the
immunoglobulin variable domain. We therefore replaced Gln
H5 of 4G7-wt by Val in 4G7-mut.
Hydrophobic surface residues in the VH/CHinterface of
the Fab fragment are solvent exposed in scFvs, where they
can have a negative influence on the folding efficiency of an
scFv. In the anti-fluorescein antibody 4-4-20, the substitution
of Leu H12 by Ser had no significant effect, while the substi-
tution of Leu H12 by Asp significantly improved the pro-
duction yield (Nieba et al., 1997). We therefore replaced Leu
H12 by Asp both in 4G7-mut and in the 4G7-graft.
Additional mutations The changes suggested above had been
tested in other scFvs and shown to improve stability and/or
Additional mutants were suggested based on sequence stat-
istics and structural criteria.
Position L45 in VLcorresponds to H45 in the heavy chain
and is likewise in contact with the charge cluster in the lower
core of the domain. In 4G7-wt, L75 is a Leu, while 88% of
human and 78% of murine antibodies have a Gln in this pos-
ition. The huVk1 and huVk3 consensus domains have a Gln
in this position, the less well-behaved huVk2 a Leu (Ewert
et al., 2003b). In 4G7-mut, we placed a Gln in this position.
Sequence statistics indicate a strong preference for Asp in
position L88. Mutations from the germline consensus Asp to
His or Asn in human Vk1-derived light chains have been
reported to increase propensity for amyloid formation
(Stevens et al., 2000). However, an effect of this mutation on
antigen affinity cannot be excluded, since in protein-binding
antibodies the outer loop of VL may make contact to the
antigen. Ala L88 was nevertheless replaced by Asp, and did
not have a deleterious effect on binding (see below).
Residue L95 frequently has a positive Phi torsion angle.
Arg L95 was therefore replaced by Gly in 4G7. For the same
reason, Ala H17 was replaced by Gly. H17 is a Gly in the very
stable huVH3 consensus domain (Ewert et al., 2003b), and in
IcaH-01, the substitution of Ala H17 by Gly stabilized VH
with a shift in [GdmCl]50by 0.1 M (Wirtz and Steipe, 1999).
In the majority of murine Vk domains, an Ala is found in
position L98, as it is in 4G7-wt. In human Vk domains, Pro
predominates. In the in vitro evolution of the 4D5Flu scFv
towards improved stability, Pro was selected over Ser and Ala
(Jung et al., 1999) in this position. In addition, replacement
of Pro by other residues in Vk1 was reported to be linked to
increased propensity for the formation of amyloid (Stevens
et al., 2000). Therefore, we replaced Ala L98 by Pro.
As a result of this analysis, 14 point mutations were intro-
duced into the 4G7-wt scFv to yield the scFv 4G7-mut
(Figs 1 and 2A).
Design of the CDR graft 4G7-graft
In designing a CDR graft from a murine to a human anti-
body, one often has the choice of either grafting to the
human acceptor framework that is closest to the sequence of
the CDR donor, or to an acceptor framework with proven
good biophysical characteristics. The sequence of the VL
domain of 4G7 most closely resembles that of huVk2, with
79% sequence identity and 84% similarity. However, the
huVk2 consensus fragment shows relatively poor biophysical
characteristics (Ewert et al., 2003b). The VHdomain of 4G7
most closely resembles huVH1, with 73% identity and 79%
similarity. Based on the analysis of Ewert et al. (2003b),
who performed a detailed biophysical characterization of a
series of consensus frameworks representing the major
human germline families, huVk3 is the most stable human
Vk fragment, huVH3 the most stable human VHfragment.
We therefore chose this framework combination as acceptor
framework, despite the fact that huVk3 showed a mere 53%
sequence identity and 65% similarity to the VLdomain of
4G7 and huVH3 51% identity and 65% similarity to the VH
domain of 4G7 (Fig. 1).
To model the CDR graft, the VLand VH4G7-wt domain
models were aligned to the corresponding domains of a pre-
viously built model of the huVk3-huVH3 Fv fragment
(Knappik et al., 2000). Starting from the superposed models
of the acceptor framework and the CDR donor, a model of
the graft was built, combining the coordinates of the loops
containing the CDRs and the outer loops from the model of
the CDR donor with the rest of the coordinates from the
model of the acceptor framework. The cut-points between
the two models were placed within the adjacent b-strands
forming the structurally least variable regions in antibody
variable domains, marked by dark gray boxes in Fig. 1. This
allowed the fragments of the model to be joined without
introducing undue strain. Sequence positions that were identi-
cal in both sequences (marked in dark blue in the sequence
alignment) did not have to be changed. Since no experimen-
tal structure of the complex of 4G7 with its antigen was
available, potential antigen contact residues were identified
by an extensive analysis of known complexes of antibodies
with protein antigens (Honegger and Plu ¨ckthun, 2001a) (see
Supplementary data available at PEDS online for additional
information). These sequence position, marked as ‘Protein
Ag Contact’ in the header of the sequence alignment (Fig. 1)
were changed to the sequence of the CDR donor. In addition,
the buried residues that pack the space between the CDRs
and a highly conserved layer of residues across the domain
core (Cys 23 and Cys 106, Trp 43, Thr 143) were changed to
the sequence of the CDR donor to avoid destabilization of
the desired CDR conformation. Residue L4 was changed
from the Leu of the acceptor framework to Met found in
the CDR donor. Buried Ala L24 was replaced by the Ser
of the CDR donor, since this substitution can influence the
conformation of CDR-L1. H1 was kept a Glu, and H24 Ala,
H29 Phe and H32 Ser adjacent to CDR-H1 were likewise
adjusted to the 4G7 CDR-donor sequence. L44 was kept
a Phe like in 4G7 instead of Tyr in the acceptor framework.
Located in the VL/VH interface, L44 has been observed
to contact the antigen in hapten and peptide binding anti-
bodies, and could potentially affect the relative domain
Optimization of the packing interactions between CDR-H2
and framework residues is critical to the success of a
CDR-graft to a framework of divergent subtype. Position H56
adjacent to CDR-H2 is a buried Ser or Ala in huVH3, Gly in
most other VHdomains, including 4G7. Since this sequence
difference affects the local b-bulge backbone conformation, it
was also adjusted to the 4G7 sequence. In contrast, residues
H73 and H74 were changed to the sequence of the framework
donor, since in particular H74 interacts with the hydrophobic
Humanization and stabilization of a CD19-specific scFv
core of the framework. Interacting residues H74 and H78 are
Phe and Ala in 4G7, yet Val and Phe in huVH3. The combi-
nation of two Phe residues in the graft would have overpacked
the core of the domain, which would either result in
significant destabilization or a change in the CDR-H2 takeoff
The outer loop of the VHdomain (H83-H88) is relatively
variable in structure. In contrast to the corresponding loop in
VL, it has not been observed to directly contact the antigen,
but it can affect CDRs -H1 and -H2. Outer loop residues
H80–H89 were thus retained from the 4G7 sequence, since
buried residues H80 and H89 pack against buried CDR-H2
and CDR-H1 residues. H107, adjacent to CDR-H3, was like-
wise taken from 4G7. All other residues were taken from the
acceptor frameworks, with the exception of exposed Leu
H12 in the former VH/CHinterface, which was replaced by
an Asp (Fig. 2B) in 4G7-graft as well as in 4G7-mut, since it
can have a marked effect on folding efficiency.
4G7-mut has higher soluble expression yields than the other
The 4G7-scFv variants were expressed in E.coli HB2151 and
purified from periplasmic extracts by NiNTA affinity chrom-
atography and ion exchange chromatography (Fig. 3B and
C). scFv purification yields were determined by densitometry
of Coomassie stained SDS–PAGE gels (Table I). 4G7-mut
was expressed consistently with about 2-fold higher yields
than 4G7-wt and 4G7-graft (2.3+0.8 mg/l versus 1.2+0.3
and 1.1+0.2 mg/l, respectively), corresponding to 135+
15 mg scFv per gram wet weight E.coli, 92+16 and 80+
28 mg/g, respectively. These data suggest that the mutations
of 4G7-mut indeed lead to an improved folding efficiency
due to the 14 mutations.
All three scFv variants show similar behavior in analytical
To analyze the tendency of the purified scFvs for forming
aggregates, analytical gel-filtration experiments were per-
formed (Fig. 4). For 4G7-wt, ?92% of the protein eluted at
the expected size for a monomer, ?8% at the size expected
for a dimer and no higher aggregates were seen. The elution
profiles for 4G7-mut and 4G7-graft contained ?11.5 and
?12.5%, respectively, of protein with the mobility of dimers.
Dilution of the sample to 0.25 mg/ml and 5 hours incubation
before measurement did not alter the fraction of dimers.
Multi-angle light scattering (MALS) measurements allowed
us to determine the molecular masses of the faster eluting
material, which were consistent with the interpretation of
dimers. The values measured by MALS for the monomer
and presumed dimer peaks for 4G7-wt were 29.7+0.6
Fig. 3. scFv construct and purification of the three 4G7 variants. (A) FLAG,
N-terminal FLAG-tag; 6xHis, hexahistidine tag; VL and VH, variable
domains of light and heavy chains of the 4G7-scFv connected by a
(Gly4Ser)4 linker; molecular masses of 4G7-wt (29.1 kDa), 4G7-mut
(29.2 kDa) and 4G7-graft (28.8 kDa) were calculated from their amino acid
sequences. (B and C) Periplasmic expression and purification of the 4G7
scFvs by IMAC and subsequent ion exchange chromatography. Purified
proteins were analyzed by SDS–PAGE and (B) Coomassie staining or (C)
western blot using a primary Penta-His antibody. Lanes 1–3: 4G7-wt, 4G7-
Table I. Biophysical properties of 4G7 scFv variants
scFvSoluble yieldBinding affinitySerum half-life
(mg/l E.coli)(mg/g E.coli)nMh
Fig. 4. Analytical gel-filtration of 4G7 variants. Samples (applied at a
concentration of 0.5 mg/ml (17 mM) were analyzed on a Superdex-75
column in 50 mM sodium phosphate (pH 7.0), 100 mM NaCl with 4G7-wt
(A), 4G7-mut (B), and 4G7-graft (C).
M.Ku ¨gler et al.
62.2+3.1 kDa; and for 4G7-graft: 29.4+0.6 and 60.0+
2.4 kDa. The two different design principles used to produce
the 4G7-mut and -graft variants, therefore, had no significant
effects on the propensity of these variants to form dimers.
4G7-mut has the greatest thermodynamic stability,
determined by equilibrium denaturation
Thermodynamic stabilities of the three variants were
assessed by denaturant-induced equilibrium denaturation/
renaturation experiments. To this end, the proteins were incu-
bated in the presence of various concentrations of guanidi-
nium chloride (GdmCl) overnight at 258C, and fluorescence
emission spectra were measured in the wavelength range
from 320 to 370 nm, using an excitation wavelength of
280 nm. Each of the three constructs contained four trypto-
phan residues contributing to the fluorescence spectrum: Trp
L43 and Trp H43 located in the core of the VLand the VH
domain usually are quenched by the neighboring disulfide
bond in the native state, while Trp H54 and Trp H149,
buried in the VL/VHdimer interface, make a strong contri-
bution to the fluorescence signal of the native scFv. 4G7
(and all the variants) contained only these four highly con-
served tryptophans, with no additional Trp in the CDRs.
While the wavelength of the fluorescence maximum of the
native 4G7-graft (338.6 nm) was close to that of the full con-
sensus huVk3-huVH3 (338.7 nm), the fluorescence maxima
of 4G7-wt (342.0 nm) and 4G7-mut (340.8 nm) were consist-
ently red-shifted by 2–3 nm relative to 4G7-graft (Fig. 5).
To ensure that this shift was not due to the presence of a
dimer species with different spectral properties, the spectra
of the monomer and dimer fractions were measured immedi-
ately after their elution from the gel filtration column.
Indeed, the same 2–3 nm shift was seen between the
monomer of 4G7-graft and the monomers of the other two
constructs, while no significant difference was detected
between the spectra of the monomer and the corresponding
dimer. Therefore, this shift seems indeed to be due to the
difference in the intrinsic spectral properties of the native
molecules, not to the presence of aggregated or misfolded
The equilibrium unfolding/refolding curves were evaluated
by a two-state fit (Santoro and Bolen, 1988) (Fig. 5).
Additional details are shown in the Supplementary data
available at PEDS online. Based on the homology models of
the Fv fragments, an m-value of 5.6–5.9 kcal.mol/l2would
be predicted (Myers et al., 1995). The m-values of 2.1–
2.6 kcal.mol/l2derived from the two-state fit for 4G7-wt,
4G7-graft and for the acceptor framework huVk3-huVH3,
representing the germline family consensus for the two
domains (Knappik et al., 2000), with CDR-L3 and -H3 taken
from hu4D5–8 (Eigenbrot et al., 1993), are far too low for a
protein the size of an scFv containing two disulfide bonds.
This indicates that the unfolding transition does not really
represent a two-state transition, and the extrapolation to zero
denaturant concentration inherent in the calculation of
DG(H2O) is not reliable. It is particularly problematic to
compare fragments with strongly different apparent m-value,
even qualitatively. Such a deviation from two-state unfolding
behavior is not unusual for antibody scFvs, which, depending
on the intrinsic stabilities of the domains and the stability
difference between VLand VHdomain, may show anything
from a fully cooperative unfolding of the two domains to the
formation of stable equilibrium unfolding intermediates, in
which one domain is still native while the other is fully
denatured (Wo ¨rn and Plu ¨ckthun, 1998; Ja ¨ger et al., 2001).
Similar deviations from two-state behavior have also been
observed in the scFv analyzed in the accompanying paper
(Honegger et al., 2009) and are discussed in detail in the
Supplementary data available at PEDS online for that paper.
We also discuss there why incomplete unfolding at high
denaturant concentrations—which in principle can lead to
low m-values—would not be consistent with our data.
Fig. 5. GdmCl equilibrium denaturation of the 4G7 variants. (A)
Fluorescence spectra of native and unfolded (in 4 M GdmCl) 4G7 variants.
(B) Equilibrium denaturation curves of constructs 4G7-wt, 4G7-mut and
4G7-graft. The unfolding transitions were observed by following the change
of the wavelength of the emission maximum of the Trp fluorescence
(excitation wavelength 280 nm) as a function of denaturant concentration.
The actual shift can be seen in (A), and it is expressed here as a percentage
from 0 (maximum of the native state) to 100 (maximum of the denatured
state). The data were fitted using the two-state model (Santoro and Bolen,
1988) for illustrative purposes only, simultaneously fitting the pre-transition-
baseline (lmaxN, mN), the post-transition baseline (lmaxU, mU) and the
thermodynamic parameters DG(H2O)and m, and plotted as relative spectral
shift (relative change of lmaxobs versus denaturant concentration) after
baseline correction (see Supplemental data available at PEDS online,
Table ST3). However, as explained in the text, and in more detail in the
Supplementary data available at PEDS online, the low m-values obtained for
4G7-wt and 4G7-graft indicate that the two-state assumption is not fulfilled,
and the fitted DG(H2O)values would be meaningless.
Humanization and stabilization of a CD19-specific scFv
The higher apparent cooperativity of 4G7-mut (m-value of
5.3 kcal.mol/l2) indicates that it more closely approaches the
behavior of a two-state system. This effect is found when the
difference between the intrinsic stabilities of the two
domains has been reduced and there is significant mutual
stabilization between the domains (Wo ¨rn and Plu ¨ckthun,
1998; Ja ¨ger et al., 2001; Honegger et al., 2009). In 4G7-mut,
the shift to a higher [GdmCl]50 shows that the weaker
domain has been stabilized to a larger degree than the stron-
ger one. This is consistent with the weaker domain having
become intrinsically more stable. While increased mutual
stabilization of the two domains would also lead to increased
cooperativity with concomitant shift to a higher [GdmCl]50,
none of the mutations is located in the VL/VHinterface, and
thus, this is less likely to be responsible for the observed
increase of the m-value.
While it was not possible to derive the absolute stabilities
of the constructs from such measurements, qualitatively, the
stabilities of the constructs can be ranked based on the mid-
points of the transition curves (Fig. 5B). This merely
assumes that the ranking in the absence of denaturant follows
the ranking at intermediate denaturant concentration. With a
midpoint of 2.4 M, 4G7-mut is the most stable of the three
4G7 variants. Amongst the three constructs, it is the one
whose denaturation starts at the highest denaturant concen-
tration. Despite the fact that the acceptor framework
huVk3-huVH3 with a midpoint of 2.8 M is more stable than
any of the 4G7 variants, the graft of the 4G7 CDRs to the
huVk3-huVH3 framework has a denaturation midpoint of
only 2.1 M, and is therefore slightly less stable than the
murine wild-type, 4G7-wt with 2.2 M.
All three variants have similar functional stability in human
To measure their stability in human serum, all three variants
were incubated in serum at a concentration of 1.5 mg/ml at
378C for different lengths of time before residual binding
was measured by flow cytometry. Staggered starting times
and a common endpoint were used, and a sample of scFv
freshly added to serum was then used to define 100%
binding at t0. All values were normalized to this t0-value
(Fig. 6 and Table I). The serum half-lives do not differ
greatly, with that of 4G7-mut being slightly higher with
106 h, compared to wt with 95 h and 4G7-graft with 82 h,
respectively. The reported half-lives represent the averages of
four independent measurements.
All variants retain antigen-specific binding, but 4G7-graft
has a slightly lower affinity
All three variants bound to CD19-positive SEM leukemic
cells in an antigen-specific manner (Fig. 7). They bound to
antigen-positive cells (Fig. 7A, C and E), but not to antigen-
negative cells (not shown), and binding was blocked by pre-
incubation of the cells with a 50-fold molar excess of the
parental antibody 4G7 (Fig. 7B, D and F), but not by a corre-
sponding molar excess of an isotype control antibody, which
itself did not react with CD19 (not shown). Therefore, intro-
duction of the 14 point mutations and the grafting procedure
apparently have not introduced any unspecific binding prop-
erty into the variants.
Binding affinities were determined by an immunofluores-
cence assay, in which increasing concentrations of the scFvs
were incubated with constant numbers of CD19-positive
SEM cells. Equilibrium constants (KD) were determined
from a fit of the mean fluorescence intensity to a 1:1 satur-
ation model. The 4G7-wt had an affinity of KD¼ 8.4+
0.9 nM, and 4G7-mut gave a value of 16.4+1.6 nM, while
Fig. 6. Serum stability of 4G7 variants at 378C. ScFvs were incubated in
human serum at 378C for the time periods indicated. The residual binding
activities of purified 4G7-wt (squares), 4G7-mut (triangles) and 4G7-graft
(diamonds) variants were determined by flow cytometry on CD19-positive
SEM cells. The data are expressed as a percentage of the sample at t0(taken
immediately after addition of scFvs to human serum) and present the mean
percentage of residual binding+standard error of the mean of four
independent experiments. Serum half-lives (t1/2values) deduced from these
curves are reported in Table I.
Fig. 7. Specific binding of 4G7 variants to CD19-positive cells. (A and B),
4G7-wt; (C and D), 4G7-mut; (E and F), 4G7-graft. Flow cytometric
analysis was performed by staining cells with equal concentrations of
purified CD19-specific scFvs (black line) or a non-related scFv (light gray)
(A, C, E). For competition experiments (B, D, F), cells were pre-incubated
with a 50-fold molar excess (black line) of parental 4G7 mAb.
Pre-incubation with a non-related mAb did not block binding of the 4G7
scFvs (data not shown).
M.Ku ¨gler et al.
4G7-graft gave 30.0+2.7 nM, indicating a slight effect of
the framework on the CDRs (Fig. 8 and Table I).
This study aimed at comparing two orthogonal strategies for
improving the stability and expression yield of a scFv, using
a CD19-specific scFv derived from murine monoclonal 4G7,
an antibody with very favorable biological properties as an
example. To this effect, two variants of 4G7-wt were con-
structed: 4G7-mut with 14 single amino acid substitutions at
selected critical positions, and 4G7-graft with the murine
CDR sequences grafted to a human framework. These var-
iants were produced and analyzed in parallel with the wt
protein, and two main results were obtained: first, 4G7-mut
had approximately 2-fold greater expression yields and
greater thermodynamic stability than the wt (which turned
out to be on the upper end of the usually observed values for
murine antibodies), and an almost identical affinity (2-fold).
Second, the 4G7-graft had approximately equal yield and
thermodynamic stability as the wt, but also a 3–4-fold
The variants did thus not differ dramatically in their prop-
erties from the wt, less than expected from the design, and
therefore, these results require some comments. The murine
wt scFv already had comparatively favorable properties, and
thus this study concentrated on testing whether they can be
improved even further.
CDR-grafts, the transfer of the antigen binding loops from
an antibody with a desired antigen specificity to a different
antibody framework for the dual purpose of increasing the
stability and reducing the immunogenicity of an antibody
construct, represent by now a routine technique in antibody
engineering. The huVk3 and huVk1 germline consensus for
the VLdomain and the huVH3 germline consensus for the
VHdomain are the most frequently used human acceptor fra-
meworks, both because they represent the most frequently
utilized germline families in the human antibody repertoire
and because these consensus domains show the most favor-
able biophysical properties (Ewert et al., 2003b). However,
while CDR grafts from problematic antibody scFvs to these
stable frameworks usually yield chimeric scFvs of greater
stability than the CDR donor, they frequently fail to reach
the level of stability observed for the consensus frameworks
themselves (Willuda et al., 1999). This reduced stability
could be due to destabilizing interactions within the CDRs or
between CDRs and framework. In the consensus scFv used
for comparison (Knappik et al., 2000; Ewert et al., 2003b),
CDR-1 and -2 sequences were derived from the same
V-gene consensus as the framework sequences themselves,
and thus presumably represent an optimal fit to these frame-
works, while in a CDR graft, and especially in a CDR graft
from a highly divergent framework, a mismatch between
CDRs and framework may lead to a destabilization of the
individual CDR-grafted domains (Honegger et al., 2009)
(accompanying paper). In addition, the CDRs contribute to
the interface between the VLand the VHdomain and, there-
fore, to the degree of mutual stabilization between the
two domains (Wo ¨rn and Plu ¨ckthun, 1998; Ro ¨thlisberger
et al., 2005).
In this study, we compared the biophysical properties of
an scFv derived from the murine monoclonal antibody 4G7
(4G7-wt) both to those of a variant in which a number of
destabilizing features in the 4G7 framework have been
repaired by point mutations (4G7-mut) and to those of a
variant in which the CDRs of 4G7 have been grafted to the
frameworks of the most stable human consensus domains,
huVk3 and huVH3 (4G7-graft). With a [GdmCl]50of 2.2 M,
the murine 4G7-wt scFv is already quite stable in equili-
brium denaturation, e.g. compared to the hu4D5-8 scFv
(Herceptin) (Eigenbrot et al., 1993), which is itself a
CDR-graft. hu4D5–8 scFv shows a first step of equilibrium
denaturation, the unfolding of the VH domain, with a
[GdmCl]50of only 1.4 M (Ja ¨ger et al., 2001), even though it
expresses well and is not aggregation prone. Indeed, the
hu4D5-8 scFv shows significantly lower stability than its
own human consensus acceptor framework, huVk1-huVH3.
This acceptor, even with CDR-L3 and -H3 derived from
hu4D5-8, unfolds with a [GdmCl]50of 2.8 M (Ewert et al.,
2003b), pointing to an important role of CDR-1 and CDR-2
The acceptor framework used in the construction presented
here, huVk3-huVH3, has a [GdmCl]50of 2.6 M (Honegger
et al., 2009) to 2.8 M (Ewert et al., 2003b). The CDR-graft
4G7-graft with a [GdmCl]50 of 2.1 M turned out to be
slightly less stable than the parental murine scFv 4G7-wt.
The m-values (cooperativity) of the equilibrium unfolding
transitions of 4G7-wt, 4G7-graft and of the acceptor frame-
work huVk3-huVH3 were all significantly lower than
expected for a protein the size of an scFv containing two dis-
ulfide bridges, indicating deviation from a two-state tran-
sition, and this is why we resort to reporting [GdmCl]50and
not DG values. These [GdmCl]50-values are to be understood
as qualitative midpoints of the unfolding trace, since the pre-
sence of a hidden intermediate does not allow to quantify the
fraction of molecules that are unfolded at a given GdmCl
concentration. This low apparent m-value can be explained
by the scFv showing a difference between the intrinsic stab-
ilities of the VLand the VHdomain large enough that the
Fig. 8. Determination of equilibrium binding constants of 4G7 variants on
CD19 bearing cells. CD19 positive SEM cells were incubated with
increasing concentrations of purified 4G7-wt (squares), 4G7-mut (triangles)
and 4G7-graft (diamonds). Cell-bound scFvs were detected by staining with
a secondary PE-conjugated antibody and analyzed by flow cytometry. The
mean fluorescence intensity (MFI) is plotted as a percentage of its maximal
plateau value, and the KD is estimated by MFI ¼ MFImax.[scFv]/(KDþ
[scFv]). Averaged over 6 independent experiments, the KD values were
8.4+0.9 nM, 16.4+1.6 nM, and 30.0+2.7 nM for 4G7-wt, 4G7-mut and
Humanization and stabilization of a CD19-specific scFv
stabilizing influence of the VL/VHinterface is insufficient to
enforce fully cooperative unfolding of the scFv. This point is
illustrated in the Supplementary data available at PEDS
online, of this paper and the accompanying manuscript
(Honegger et al., 2009).
While various additional mutations outside the classical
CDR boundaries have been introduced in order to preserve
the upper core packing of the CDR donor in the graft and to
minimize steric conflicts between framework and CDR resi-
dues that potentially could alter the CDR conformation or
destabilize the scFv, the high stability of the human acceptor
framework (with its own CDRs) could not be maintained.
This is due to residual strains between the grafted CDRs and
the acceptor framework, similarly found in the Herceptin
construct itself and in a case described in the accompanying
paper (Honegger et al., 2009).
In the second approach presented here, we addressed indi-
vidual destabilizing features in the 4G7-wt sequence with a
series of point mutations located within the frameworks of
the two domains. In a knowledge-based approach, we
scanned the sequence of 4G7-wt against our database of
mutations known to affect the stability and/or folding effi-
ciency of antibody variable domains, and we identified 14
positions requiring mutation. Ten of these mutations are
mutations towards the huVk3 or the huVH3 sequence consen-
sus and thus are also present in 4G7-graft. One of the
mutations, the replacement of hydrophobic surface residue
Leu H12 by Asp, was also introduced in 4G7-graft, since
this mutation had previously been shown to have a beneficial
effect on the folding efficiency of scFv without affecting
stability (Nieba et al., 1997), increasing the ratio of soluble
protein to aggregates in the periplasm. The largest effect on
stability was expected from the optimization of the charge
cluster in the lower core of the VHdomain (position H77,
H45 and H97). Other mutations were targeted at optimizing
the framework 1 kink of the VLdomain and at non-Gly pos-
itions that carry positive phi angles in a majority of the
known antibody structures.
With a [GdmCl]50of 2.4 M, 4G7-mut turned out to be the
most stable of the three 4G7 constructs. In addition, the sig-
nificantly increased cooperativity indicates that by predomi-
nantly stabilizing the weaker of the two domains, we
succeeded in reducing the difference between the intrinsic
stabilities of the VLand VHdomains. Measures of serum
half-life, production yield and ratio of the production of
soluble protein to aggregates in the periplasm of E.coli all
favored 4G7-mut over 4G7-wt and 4G7-graft.
The observation that a limited number of point mutations
in the framework was capable of stabilizing 4G7-wt, while a
graft to the most stable human consensus framework combi-
nation failed to do so, suggests that the relatively low stab-
ility of the graft cannot be blamed on destabilizing
influences located within the CDRs themselves. More likely,
despite the care taken in the design of the CDR-graft, the
structural difference between the muVH1 framework of 4G7
and the huVH3 framework serving as acceptor framework led
to some strain between CDRs and framework that limits the
stability of the graft. Nonetheless, the stability of the loop
donor was maintained.
The knowledge-based redesign with the individual amino
acid substitutions was, therefore, capable of producing a
variant of greater stability, and based on the biophysical
properties, 4G7-mut is thus a candidate for the development
of further derivatives intended for clinical use. However, the
final choice will need to take into account the issue of immu-
nogenicity, although for the therapy of human B-cell malig-
nancies immunogenicity is not as severe a problem as for the
treatment of solid tumors such as carcinomas, and the scien-
tific basis of a lower immunogenicity of humanized mol-
ecules has been questioned (Clark, 2000). Additionally, the
use as a fusion protein with e.g. protein toxins of plant or
bacterial origin, may lower the importance of the immuno-
genicity of the scFv portion.
The final choice of the variant to be promoted further will
thus depend on the exact therapeutic format chosen. The
results of this study underscore the powers of the knowledge-
based approach, but also illustrate some remaining difficul-
ties in correctly predicting strain when bringing pieces from
different proteins together.
The authors thank Prof. R. Levy from Stanford University for making the
4G7 hybridoma available, and PD Dr R. Slany for size exclusion chromato-
graphy. C.S. was the recipient of a stipend from the Bavarian Government
(Bayerische Universita ¨ten e.V.), and M.S. of a stipend from the Kind Phillip
Foundation for leukemia research. We thank Hugues Bedouelle for providing
the Kaleidagraph macros used to evaluate the equilibrium unfolding data.
This study was supported by grants from the W. Sander
Foundation (No. 03.015.2), Deutsche Krebshilfe e.V. (No.
108242), the association ‘Kaminkehrer helfen krebskranken
Kindern e.V.’, the Beitlich Foundation, Tu ¨bingen and the
association of supporters of the University of Erlangen
Children’s Hospital (to G.H.F.). Part of this work was funded
by an intramural grant from the ELAN fond and the Training
Grant GK592 from the German Research Community (DFG).
Bargou,R., et al. (2008) Science, 321, 974–977.
Benedict,C.A., MacKrell,A.J. and Anderson,W.F. (1997) J. Immunol.
Methods, 201, 223–231.
Bruenke,J., et al. (2004) Br. J. Haematol., 125, 167–179.
Bruenke,J., et al. (2005) Br. J. Haematol., 130, 218–228.
Carter,P., Bedouelle,H. and Winter,G. (1985) Nucleic Acids Res., 13,
Chari,R.V. (2008) Acc. Chem. Res., 41, 98–107.
Clark,M. (2000) Immunol. Today, 21, 397–402.
Eigenbrot,C., Randal,M., Presta,L., Carter,P. and Kossiakoff,A.A. (1993)
J. Mol. Biol., 229, 969–995.
Hoffman,K., Lutz,R.J., Goldmacher,V.S. and Blattler,W.A. (2006) Cancer
Res., 66, 4426–4433.
Ewert,S.,Honegger,A. andPlu ¨ckthun,A.
Ewert,S., Huber,T., Honegger,A. and Plu ¨ckthun,A. (2003b) J. Mol. Biol.,
Flavell,D.J., Warnes,S.L., Bryson,C.J., Field,S.A., Noss,A.L., Packham,G.
and Flavell,S.U. (2006) Br. J. Haematol., 134, 157–170.
Greil,J., Gramatzki,M., Burger,R., Marschalek,R., Peltner,M., Trautmann,U.,
Br. J. Haematol., 86, 275–283.
Nadler,L.M. (1992) Blood, 80, 863–878.
Hekman,A., Honselaar,A., Vuist,W.M., Sein,J.J., Rodenhuis,S., ten Bokkel
Huinink,W.W., Somers,R., Rumke,P. and Melief,C.J. (1991) Cancer
Immunol. Immunother., 32, 364–372.
M.Ku ¨gler et al.
Honegger,A. (2008) In Chernajovsky,Y.(eds.), Therapeutic Antibodies.
Vol. 181. Springer-Verlag, Berlin-Heidelberg, pp. 47–68.
Honegger,A. and Plu ¨ckthun,A. (2001a) J. Mol. Biol., 309, 657–670.
Honegger,A. and Plu ¨ckthun,A. (2001b) J. Mol. Biol., 309, 687–699.
Honegger,A., Malebranche,A.D., Ro ¨thlisberger,D. and Plu ¨ckthun,A. (2009)
Protein Eng. Des. Sel., (in press).
Hoogenboom,H.R. (2005) Nat. Biotechnol., 23, 1105–1116.
Ja ¨ger,M., Gehrig,P. and Plu ¨ckthun,A. (2001) J. Mol. Biol., 305, 1111–1129.
Jung,S., Honegger,A. and Plu ¨ckthun,A. (1999) J. Mol. Biol., 294, 163–180.
Jung,S., Spinelli,S., Schimmele,B., Honegger,A., Pugliese,L., Cambillau,C.
and Plu ¨ckthun,A. (2001) J. Mol. Biol., 309, 701–716.
Kabat,E.A., Wu,T.T., Perry,H., Gottesmann,K. and Foeller,C. (1991)
Sequences of Proteins of Immunological Interest. Fifth edition, ed. NIH
Publication No. 91-3242.
Kellner,C., et al. (2008) J. Immunother., 31, 871–884.
Kipriyanov,S.M., Cochlovius,B., Schafer,H.J., Moldenhauer,G., Bahre,A., Le
Gall,F., Knackmuss,S. and Little,M. (2002) J. Immunol., 169, 137–144.
Hoess,A., Wo ¨lle,J., Plu ¨ckthun,A. and Virneka ¨s,B. (2000) J. Mol. Biol.,
Bosshard,H.R. and Plu ¨ckthun,A. (1997) J. Immunol. Methods, 201,
Pfeiffer,M., Handgretinger,R., Niethammer,D. and Fey,G.H. (2004) Blood,
Scaviner,D., Ginestoux,C., Clement,O., Chaume,D. and Lefranc,G. (2005)
Nucleic Acids Res., 33, D593–D597.
Honegger,A., Klinger,B., Mocikat,R.
BioTechniques, 22, 140–149.
Lonberg,N. (2005) Nat. Biotechnol., 23, 1117–1125.
Meeker,T.C., Miller,R.A., Link,M.P., Bindl,J., Warnke,R. and Levy,R.
(1984) Hybridoma, 3, 305–320.
Messmann,R.A., et al. (2000) Clin. Cancer Res., 6, 1302–1313.
Mirick,G.R., Bradt,B.M., Denardo,S.J. and Denardo,G.L. (2004) Q. J. Nucl.
Med. Mol. Imaging, 48, 251–257.
Monsellier,E. and Bedouelle,H. (2005) Protein Eng. Des. Sel., 18, 445–456.
Myers,J.K., Pace,C.N. and Scholtz,J.M. (1995) Protein Sci., 4, 2138–2148.
Nadler,L.M., Anderson,K.C., Marti,G., Bates,M., Park,E., Daley,J.F. and
Schlossman,S.F. (1983) J. Immunol., 131, 244–250.
Nieba,L., Honegger,A., Krebber,C. and Plu ¨ckthun,A. (1997) Protein Eng.,
Ohage,E.C., Graml,W., Walter,M.M., Steinbacher,S. and Steipe,B. (1997)
Protein Sci., 6, 233–241.
Olejniczak,S.H., Stewart,C.C., Donohue,K. and Czuczman,M.S. (2006)
Immunol. Invest., 35, 93–114.
Pastan,I., Hassan,R., FitzGerald,D.J. and Kreitman,R.J. (2007) Annu. Rev.
Med., 58, 221–237.
Peipp,M. and Valerius,T. (2002) Biochem. Soc. Trans., 30, 507–511.
Peipp,M., Saul,D., Barbin,K., Bruenke,J., Zunino,S.J., Niederweis,M. and
Fey,G.H. (2004) J. Immunol. Methods, 285, 265–280.
Pietersz,G.A., Wenjun,L., Sutton,V.R., Burgess,J., McKenzie,I.F., Zola,H.
and Trapani,J.A. (1995) Cancer Immunol. Immunother., 41, 53–60.
Lawrence,C.M. (2006) J. Biol. Chem., 281, 6813–6825.
Press,O.W., Farr,A.G., Borroz,K.I., Anderson,S.K. and Martin,P.J. (1989)
Cancer Res., 49, 4906–4912.
Proba,K., Honegger,A. and Plu ¨ckthun,A. (1997) J. Mol. Biol., 265,
Proba,K., Wo ¨rn,A., Honegger,A. and Plu ¨ckthun,A. (1998) J. Mol. Biol., 275,
Raffen,R., Dieckman,L.J., Szpunar,M., Wunschl,C., Pokkuluri,P.R., Dave,P.,
Wilkins Stevens,P., Cai,X., Schiffer,M. and Stevens,F.J. (1999) Protein
Sci., 8, 509–517.
Ro ¨thlisberger,D., Honegger,A. and Plu ¨ckthun,A. (2005) J. Mol. Biol., 347,
Ruzheinikov,S.N., et al. (2003) J. Mol. Biol., 332, 423–435.
Santoro,M.M. and Bolen,D.W. (1988) Biochemistry, 27, 8063–8068.
Sato,S., Miller,A.S., Howard,M.C. and Tedder,T.F. (1997) J. Immunol., 159,
Schable,K.F., et al. (1999) Eur. J. Immunol., 29, 2082–2086.
Schindler,J., Sausville,E., Messmann,R., Uhr,J.W. and Vitetta,E.S. (2001)
Clin. Cancer Res., 7, 255–258.
Proteins, 57, 269–278.
and Plu ¨ckthun,A.
Schwemmlein,M., et al. (2007) Leukemia, 21, 1405–1412.
Spada,S., Honegger,A. and Plu ¨ckthun,A. (1998) J. Mol. Biol., 283,
Stevens,F.J., Pokkuluri,P.R. and Schiffer,M. (2000) Biochemistry, 39,
Stieglmaier,J., et al. (2008) Cancer Immunol. Immunother., 57, 233–246.
Vallera,D.A., Todhunter,D.A., Kuroki,D.W., Shu,Y., Sicheneder,A. and
Chen,H. (2005) Clin. Cancer Res., 11, 3879–3888.
van den Elsen,J., Vandeputte-Rutten,L., Kroon,J. and Gros,P. (1999) J. Biol.
Chem., 274, 1495–1501.
Zangemeister-Wittke,U. and Plu ¨ckthun,A. (1999) Cancer Res., 59,
Wirtz,P. and Steipe,B. (1999) Protein Sci., 8, 2245–2250.
Wo ¨rn,A. and Plu ¨ckthun,A. (1998) Biochemistry, 37, 13120–13127.
Wo ¨rn,A., Auf der Maur,A., Escher,D., Honegger,A., Barberis,A. and
Plu ¨ckthun,A. (2000) J. Biol. Chem., 275, 2795–2803.
Wu ¨lfing,C., Lombardero,J. and Plu ¨ckthun,A. (1994) J. Biol. Chem., 269,
Yazawa,N., Hamaguchi,Y., Poe,J.C. and Tedder,T.F. (2005) Proc. Natl Acad.
Sci. USA, 102, 15178–15183.
Received October 3, 2008; revised November 28, 2008;
accepted November 30, 2008
Edited by Hugues Bedouelle
Humanization and stabilization of a CD19-specific scFv