Generic Approach for the Generation of Stable Humanized Single-chain Fv Fragments from Rabbit Monoclonal Antibodies

Despite their favorable pharmacokinetic properties, single-chain Fv antibody fragments (scFvs) are not commonly used as therapeutics, mainly due to generally low stabilities and poor production yields. In this work, we describe the identification and optimization of a human scFv scaffold, termed FW1.4, which is suitable for humanization and stabilization of a broad variety of rabbit antibody variable domains. A motif consisting of five structurally relevant framework residues that are highly conserved in rabbit variable domains was introduced into FW1.4 to generate a generically applicable scFv scaffold, termed FW1.4gen. Grafting of complementarity determining regions (CDRs) from 15 different rabbit monoclonal antibodies onto FW1.4 and their derivatives resulted in humanized scFvs with binding affinities in the range from 4.7 × 10−9 to 1.5 × 10−11 m. Interestingly, minimalistic grafting of CDRs onto FW1.4gen, without any substitutions in the framework regions, resulted in affinities ranging from 5.7 × 10−10 to <1.8 × 10−12 m. When compared with progenitor rabbit scFvs, affinities of most humanized scFvs were similar. Moreover, in contrast to progenitor scFvs, which were difficult to produce, biophysical properties of the humanized scFvs were significantly improved, as exemplified by generally good production yields in a generic refolding process and by apparent melting temperatures between 53 and 86 °C. Thus, minimalistic grafting of rabbit CDRs on the FW1.4gen scaffold presents a simple and reproducible approach to humanize and stabilize rabbit variable domains.

Because of their favorable pharmacokinetic properties, single-chain Fv (scFv) 3 antibody fragments represent an attractive format for therapeutic applications (1,2). scFvs are often derived from monoclonal antibodies isolated from animal or human lymphocytes. As an alternative to hybridoma screening, in vitro display technologies, e.g. phage and ribosome display, enable the selection of high affinity-binding variable domains from natural or synthetic genetic libraries. Despite the success-ful use of in vitro randomization and selection systems, generation of antibodies by immunization and subsequent screening of full-size antibodies (e.g. hybridoma supernatants) includes conceptual advantages. For example, in contrast to in vitro display systems, in vivo methods are less prone to preferential selection of well expressed clones, which in many cases results in loss of potentially interesting antibodies. Moreover, in vivo methods are preferred in particular for addressing complex antigens, such as integral membrane proteins that are notoriously difficult to purify. However, reducing a full-length monoclonal antibody to the scFv format frequently is challenging particularly due to solubility and stability problems, which often impair expression and purification. Therefore, technologies to humanize and stabilize the scFv format following isolation of a monoclonal antibody remain critical for the generation of scFv therapeutics. Numerous approaches have been described to improve biophysical properties of the scFv format (3), which can be grouped into two categories. In the first category, variable domains of pre-existing scFvs are engineered for improved stability, either by rationally altering specific positions in the framework regions (4 -8) or by random mutagenesis of framework positions and subsequent screening by genetic selection methods that favor stable scFvs (9 -13). In the second category, stabilization of the binding moiety is achieved by loop grafting, i.e. transplantation of the complementarity determining regions (CDRs) onto acceptor frameworks with suitable biophysical properties. For example, loop grafting of rodent CDRs onto a suitable consensus human variable domain framework was shown to result in superior stability of the resulting scFv fragment (14). This approach is particularly interesting for the generation of scFvs for therapeutic applications, because it combines stabilization and humanization in one step. However, because of the high structural diversity, particularly of rodent variable domains, a relatively large repertoire of human acceptor frameworks is required to match the major subtypes (15). In addition, further amino acid substitutions in the human framework regions are often required to restore the conformation of animal CDRs (16 -20). As a consequence, humanization of antibodies is frequently subject to engineering strategies specifically designed for every individual donor sequence, and it is particularly challenging for the scFv format because these fragments tend to aggregate and are difficult to produce. As a result, the outcome of such laborious efforts is unpredictable in many cases, and the overall success rate is low when compared with humanization of Fabs or IgGs.
In contrast to humans and rodents, framework variability in rabbits is very limited because one VH germ line gene segment is preferentially used and accounts for 80 -90% of VDJ genes, which are combined with multiple but homologous VJ genes coding for the light chain. This apparent limitation of antibody diversity in rabbits is compensated by a high degree of N-nucleotide addition at VD and DJ junctions. Further VDJ gene diversification occurs by somatic hypermutation and gene conversion-like mechanisms upon antigenic stimulation (reviewed in Ref. 21). As a consequence of preferential VH1 gene segment usage, high homology among V gene segments, and) usage of gene conversion during antibody diversification, rabbit variable domain frameworks are very homologous to each other. Furthermore, following immunization, rabbit antibodies mostly show significantly higher affinities when compared with rodent antibodies. 4 Thus, because of their high affinities and the relatively low structural diversity, rabbit antibodies present an ideal starting point for the development of a generally applicable protocol to generate humanized scFv therapeutics.
In the work presented here, we used a single human scFv scaffold of the V1-VH3 subtype to generate a set of scFvs with high affinity by grafting of CDRs from 15 different rabbit monoclonal antibodies directed against tumor necrosis factor-␣ (TNF-␣) or vascular endothelial growth factor (VEGF). This scaffold was previously identified from a human library using a whole genome screening approach (22). We further identified a motif consisting of five rationally altered framework positions, which, when introduced into the human acceptor scaffold, improved protein stability and supported functional presentation of rabbit CDRs. Most resulting antibody fragments exhibited excellent solubility, thermal stability, and affinity and were successfully produced with high yields in a generic refolding process from inclusion bodies in Escherichia coli.

MATERIALS AND METHODS
Generation of Rabbit Monoclonal Antibodies-Rabbit monoclonal antibodies (mAbs) were generated in collaboration with Epitomics Inc. Briefly, New Zealand White rabbits were immunized with recombinant human VEGF 165 (PeproTech EC Ltd., London, UK) and peptides thereof or with recombinant human TNF-␣. Spleen cells of the immunized rabbits were fused with rabbit immortal B cells (240E-W2) as described previously (23).
For VEGF binders, 23,040 hybridomas were screened for the presence of rabbit mAbs to human VEGF 165 by an enzymelinked immunosorbent assay (ELISA). Neutralizing activity of the 248 positive hybridomas was assessed using a VEGF receptor 2 (VEGFR2) blocking ELISA. Out of 92 hybridomas showing inhibition of VEGF binding to VEGFR2, 23 hybridomas were selected based on consistent results in the VEGFR2-blocking ELISA and cloned twice using limiting dilution technique. Binding affinities of cloned hybridomas toward human VEGF 165 were measured by surface plasmon resonance (SPR), using a BIAcore T-100 instrument (Biacore Inc., Uppsala, Sweden). Total RNA of the seven hybridomas secreting the most potent rabbit mAbs was isolated from hybridoma cells, and the cDNAs encoding the variable light and variable heavy chain were amplified by reverse transcription-PCR. After PCR amplification, sequenced DNA fragments were ligated into a mammalian expression vector containing rabbit CL and CH domains. Correctness of amplified VL and VH domains was confirmed by transient expression of the rabbit mAbs in human 293 cells and subsequent analysis of 293 cell supernatants using the VEGFR2 blocking assay and SPR measurements.
For TNF binders, supernatants of 5640 hybridomas were screened for binding to human TNF-␣ in ELISA. Out of 142 hits, 44 neutralized TNF-␣ induced cytotoxicity in murine L929 cells. Following cloning of the 44 confirmed hits, binding affinity to human TNF-␣ was determined by SPR measurements. Eight antibodies were selected for humanization and reformatting based on potency in the L929 assay.
Sequence Alignments-Antibody variable domain sequences were aligned using ClustalW (24) and were analyzed using the BioEdit sequence alignment editor version 7.0.9.0 (Ibis Biosciences, Carlsbad, CA). The Kabat numbering scheme was used for nomenclature of residue positions as well as for the definition of CDRs, except CDR-H1 (25). The boundaries of CDR-H1 vary substantially depending on whether loop structure or sequence variability criteria are considered for the CDR definition. Therefore, in this study CDR-H1 was defined from residue VH 26 to VH 35, which is a combination of Kabat and Chotia definitions and takes into account both structural and sequence variability.
Nomenclature and overview of the human and rabbit immunoglobulin germ line sequences were according to the International ImMunoGeneTics information system (IMGT Montpellier, France). The most closely related V-gene germ line sequences were identified based on homology of VH and VL segments of each hybridoma clone to rabbit germ line sequences. Amino acid alignments were evaluated from residues 1 to 88 (for VL) and from 1 to 92 (for VH).
Framework Selection-A pool of 88-well folding and stable scFv antibodies previously isolated from a library generated by amplification and random combination of human VH and VL domains from a naive human spleen cDNA (22) was analyzed to select a suitable acceptor framework for rabbit CDR grafting. Human scFv sequences were ranked as follows: (a) according to expression level of the respective clone in yeast, and (b) according to homology to rabbit consensus at core residues. Core residues were defined as residue positions with less than 10% average relative side-chain accessibility to the solvent using information by Honegger and Pluckthun (26). Each of the two domains selected to generate the acceptor scFv scaffold FW1.4 (clone kI27 assigned to germ line IGKV1-5 and a43 assigned to germ line IGHV3-23) corresponds to a mature human scFv clone that showed high levels of soluble expression in yeast. The amino acid sequence of FW1.4 is as follows: EIVMTQSPST-LSASVGDRVIITC*CDRL1*WYQQKPGKAPKLLIY*CDRL2* GVPSRFSGSGSGAEFTLTISSLQPDDFATYYC*CDRL3*FGQ-GTKLTVLGGGGGSGGGGSGGGGSGGGGSEVQLVESGGG-LVQPGGSLRLSCAAS*CDRH1*WVRQAPGKGLEWVS*CD-RH2*RFTISRDNSKNTLYLQMNSLRAEDTAVYYCAK*CDR-H3*WGQGTLVTVSS (underlines indicate CDRs; asterisks separate the known amino acid sequence of the framework from unknown CDR sequences).
Generation of Humanized scFvs from Rabbit Monoclonal Antibodies-A first minimalistic humanized version was generated for each of the 15 rabbit monoclonal antibodies by combining sequences of their CDRs with framework region sequences of the human scFv acceptor scaffold FW1.4. A second series of "optimized" grafts was generated by CDR transfer onto optimized derivatives of FW1.4 (referred to as FW1.4opt). Human framework residues were substituted as follows: (a) at positions that are relevant for CDR conformation by the respective amino acids used in rabbit sequences and (b) by grafting of donor amino acids that potentially are in contact with the antigen. Due to the lack of crystal structures of rabbit antibody variable domains, selection of positions relevant for CDR conformation were identified based on literature data (14,16,19,(27)(28)(29)(30). Resulting clones contained such substitutions at selected subsets of position as follows: L69, H23, H24, H49, H67, H69, H71, H73, H78, and H94. Framework residues that potentially interact with the antigen were identified by alignment of the rabbit variable domains with the nearest germ line counterpart (for VL) or the rabbit variable heavy domain consensus sequence (for VH). Differences between aligned sequences were hypothesized to result from in vivo affinity maturation. Such mutations at positions predicted to be solventexposed and in proximity to the antigen-binding site as well as the rare mutations that were found at the positions mentioned above involved in CDR conformation were transferred to the acceptor framework. Information about solvent exposure of residues was extracted from a sequence analysis by Honegger and Pluckthun (26). Pro and Gly residues introduced as result of the somatic hypermutation process were substituted if such mutations were found in the proximity of CDRs.
Molecular Cloning of scFv Expression Vectors-DNA sequences encoding CDR-grafted scFvs were optimized for E. coli codon usage, GC content, mRNA secondary structure, codon and motif repeats, and restriction sites, using LETO software package (Entelechon GmbH, Regensburg, Germany). Overlapping oligonucleotides matching the optimized DNA sequence were synthesized, and genes were generated by overlap extension techniques (27). A (Gly 4 Ser) 4 linker was used to connect VL and VH domains. All scFv genes contained 5Ј-and 3Ј-flanking NcoI and HindIII restriction sites, respectively, that allowed cloning into the proprietary E. coli inclusion body expression vector. Additional AccIII and BamHI sites were introduced in the linker sequence to enable domain shuffling. Two derivatives were made for the humanized scFvs derived from each of the rabbit monoclonal antibodies 34, 43, 511, and 578 by domain shuffling. For this, domains containing back mutations to donor residues in the framework regions (optimized grafts; FW1.4opt) were paired with the domains lacking mutations in the framework (minimalistic grafts; FW1.4) using standard DNA cloning techniques.
Expression and Protein Purification of scFv Fragments-E. coli BL21(DE3) transformed with the respective inclusion body expression plasmids were grown at 37°C in dYT medium containing the appropriate antibiotics. Protein expression was initiated by addition of 1 mM isopropyl 1-thio-␤-D-galactopyran-oside (final concentration) at an absorbance (A 600 ) of about 2.0. Three hours after induction, E. coli cells were harvested and disrupted by sonication, and inclusion bodies were isolated by repeated washing and centrifugation steps. Inclusion bodies were solubilized at a concentration of 10 mg/ml in the presence of 6 M Gdn-HCl and reduced by addition of 20 mM dithiothreitol. Basic refolding screenings were performed to select best pH, redox system (cystine/cysteine), and salt concentrations from the range of tested conditions. Best conditions for each individual scFv were then used for a lab-scale refolding process. For this, the scFv proteins were renatured by rapid dilution into a 50-fold volume of refolding buffer. After up-concentration and dialysis against PBS buffer, pH 6.0, proteins were purified using size-exclusion chromatography. Content and purity of eluted fractions were assessed by SDS-PAGE and size-exclusion HPLC. Refolding yield was expressed as amount (milligrams) of refolded protein obtained out of 1 liter of refolding solution.
For SPR measurements, periplasmic fractions containing anti-VEGF "wild type" (WT) scFvs and anti-TNF WT scFvs were prepared. Overnight starter cultures were made by inoculating single E. coli BL21(DE3) colonies from LB plates into 2-ml cultures of dYT medium with suitable antibiotics and 1% glucose in a 37°C shaker. 1.5 ml of expression medium (dYT with 45 mM K 2 HPO 4 , antibiotics, 0.1% glucose) was inoculated with 150 l of the overnight cultures (in triplicate). The bacterial cultures were incubated in a 30°C shaker until A 595 reached 1.5-2. Expression was induced by the addition of isopropyl 1-thio-␤-D-galactopyranoside to a final concentration of 0.5 mM. Three hours after induction, cultures were harvested by centrifuging for 10 min at 4500 rpm. The pellets were resuspended in 300 l of fractionation buffer (200 mM Tris-HCl, 1 mM EDTA, 20% sucrose, 500 g/ml lysozyme), and each set of triplicates was pooled. After a static incubation at room temperature for 15 min, an equal volume of cold water was added, and the suspension was incubated for a further 15 min. The supernatant was then recovered as a periplasmic fraction by centrifuging for 15 min in a bench top centrifuge (13,000 rpm and 4°C).
Thermostability Measurements-Thermostability measurements were performed using a differential scanning calorimeter (DSC) and a Fourier-transformed infrared (FTIR) spectrophotometer. Samples were first dialyzed against phosphate-buffered saline (50 mM Na 2 HPO 4 , 150 mM NaCl, pH 6.5). DSC was performed using a MicroCal high throughput VP-Capillary-DSC (Microcal, Northhampton, MA). Measurements of the difference in heat capacity between the scFv samples in solution and reference buffer were performed using protein concentrations of ϳ1.0 mg/ml. Measurements were performed over a temperature range from 25 to 95°C at a scan rate of 3.3°C/min. Data were analyzed using the Origin plotting software (OriginLab, Northhampton, MA). Apparent melting temperatures were estimated from DSC scans after concentration normalization and subtraction of the buffer-buffer baseline.
FTIR spectra were obtained on a Bruker Tensor 27 FTIR spectrometer equipped with an attenuated total reflectance Bio-ATR cell (Bruker Optics, Faellanden, Switzerland). Changes in secondary structure of the samples were assessed by heating from 25 to 95°C using 5 and 2.5°C steps for the dynamic range of unfolding. At each temperature, a total of 200 scans was recorded for each spectrum at a resolution of 1 cm Ϫ1 . All spectra manipulations were performed using OPUS spectroscopy software (Bruker Optics, Faellanden, Switzerland). Buffer reference and transient atmospheric (CO 2 and H 2 O) background were subtracted from the spectra. Second derivative spectra were obtained for the amide I band using a third degree polynomial function with smoothing. Degree of unfolding was assessed by multifactorial analysis of second derivative amide I band shape. A linear calibration curve was generated, assuming 0% unfolding for the spectra at 25, 30, and 35°C and 100% unfolding for the spectra at the three highest temperatures (85, 90 and 95°C), which is adequate for most scFvs. Thermal unfolding curves were determined by fitting the FTIR spectra to a linear regression as a function of temperature using the calibration curve. The reported apparent melting temperatures (T m ) for the various scFvs correspond to the temperature of 50% unfolding.
Binding Kinetics and Affinity of VEGF and TNF Antagonists-For binding kinetics measurements, SPR measurements with BIAcore TM -T100 were employed. All measurements were performed at 25°C. Carboxymethylated dextran biosensor chips (CM4, GE Healthcare) were activated with N-ethyl-NЈ-(3-dimethylaminopropyl) carbodiimide hydrochloride and N-hydroxysuccinimide according to the supplier's instructions, and recombinant human VEGF 165 (PeproTech EC Ltd., London, UK) was immobilized on a CM4 sensor chip using a standard amine-coupling procedure to achieve a response of ϳ200 resonance units. 2-Fold serial dilutions of VEGF antagonists (20 -0.16 nM) in HBS-EP buffer (10 mM HEPES, 150 mM NaCl, 3 mM EDTA, and 0.05% surfactant P20, pH 7.4) were injected into the flow cells at a flow rate of 30 l/min for 5 min. Dissociation of the anti-VEGF scFv from the VEGF on the CM4 chip was allowed to proceed for 10 min. After each injection cycle, surfaces were regenerated with two injections of 100 mM NaOH.
The binding kinetics of anti-TNF scFvs were measured using a nitrilotriacetic acid (NTA) sensor chip (Series S Sensor Chip NTA, GE Healthcare) and His-tagged human TNF-␣ (produced in-house). The chip was loaded with 500 M NiCl 2 diluted in HBS-EP buffer (10 mM HEPES, 150 mM NaCl, 50 M EDTA, and 0.05% surfactant P20, pH 7.4). Human TNF-␣ (2 nM) was captured through the N-terminal His tag via Ni 2ϩ NTA chelation. 3-Fold serial dilutions of TNF antagonists (90 to 0.014 nM) diluted in HBS-EP buffer were injected into the flow cells at a flow rate of 30 l/min for 5 min. Dissociation of the TNF-␣ antagonists was allowed to proceed for 10 min. Regeneration of the chip surface was performed by injection of regeneration solution (10 mM HEPES, 150 mM NaCl, 350 mM EDTA, and 0.05% surfactant P20, pH 8.3) followed by injection of 50 mM NaOH.
Binding kinetics measurements of anti-VEGF WT-scFvs and 578-WT-IgG were performed as described above for humanized VEGF antagonists, using a CM4 sensor chip with immobilized human VEGF 165 . Binding kinetics of anti-TNF WT-scFvs were measured using a CM5 sensor chip with immobilized human TNF-␣. Periplasmic fractions containing His-tagged WT-scFvs (or the 578-WT-IgG hybridoma supernatant) were serially diluted in 2-fold steps in HBS-EP buffer. For VEGF binders, surfaces were regenerated with two injections of 100 or 75 mM NaOH after each cycle, depending on the WT-scFv. For TNF binders, single cycle kinetic measurements were done by sequential injections of an scFv concentration series without any regeneration steps.
The apparent dissociation (k d ) and association (k a ) rate constants and the apparent dissociation equilibrium constant (K D ) were calculated with BIAcore T100 evaluation software version 2.0.1 using one-to-one Langmuir binding model (Biacore Inc., Uppsala, Sweden). Because the concentration of the scFvs in the periplasmic fractions was unknown, only the apparent dissociation rate constants were calculated by fitting the dissociation curves to a one-to-one dissociation model VEGF Receptor 2 Blocking Assay-Recombinant human VEGFR2-Fc chimera (R & D Systems Inc., Minneapolis, MN), consisting of amino acid residues 1-764 of the extracellular domain of human VEGFR2 fused to a His 6 -tagged Fc domain of human IgG 1 , was coated on a 96-well Maxisorp ELISA plate (Nunc, Langenselbold, Germany) at 0.2 g/ml in PBS and blocked using PBS with 0.01% BSA and 0.2% Tween 20 (PBST). Biotinylated human VEGF 165 (2.6 nM) was first incubated with 2-fold serially diluted anti-VEGF scFvs and ranibizumab (15-0.01 nM) in PBST. After 24 h of incubation at room temperature, the mixtures were transferred to the VEGF receptor-immobilized plate and incubated for 1.5 h at room temperature. Binding of unblocked human VEGF 165 to the immobilized VEGFR2 was detected with streptavidin-HRP conjugates (Stereospecific Detection Technologies, Baesweiler, Germany) followed by addition of substrate (BM Blue POD substrate, Roche Diagnostics). Absorbance at 450 nm was measured using a Sunrise microplate reader (Tecan, Maennedorf, Switzerland). The dose-response curves of the scFvs were fitted to a four-parameter logistic fit to calculate EC 50 values.
HUVEC Proliferation Assay-Human umbilical vein endothelial cells (Promo Cell, Heidelberg, Germany) were maintained in supplemented endothelial cell growth medium (Promo Cell, Heidelberg, Germany) with 1% penicillin/streptomycin. HUVECs were seeded in poly-D-lysine-coated 96-well plates (BD Biosciences) at a density of 2000 cells per well and incubated for 24 h at 37°C. 3-Fold serial dilutions of anti-VEGF scFvs or ranibizumab (150 -0.007 nM) and recombinant human VEGF 165 (PeproTech, London, UK) (0.16 nM final concentration) were prepared in starving medium (endothelial cell growth medium without supplement containing 0.5% heat-inactivated fetal calf serum and 1% penicillin/streptomycin) and preincubated for 60 min at room temperature. VEGF concentrations that stimulate submaximal HUVEC proliferation (EC 90 ) were used. Cells were then washed once with starving medium, and the agonist/antagonist mixtures were added to the cells and incubated for 4 days in a 37°C, 5% CO 2 -humidified incubator. Cell proliferation was assessed by measuring absorbance at 450 nm as described above after addition of WST-1 cell proliferation reagent (Roche Diagnostics). Data were analyzed using a four-parameter logistic curve fit, and the molar concentration of VEGF inhibitor required to reduce HUVEC proliferation to 50% (EC 50 ) was derived from inhibition curves.
TNF-␣-induced Apoptosis in L929 Fibroblasts-Mouse L929 fibroblasts (Deutsche Sammlung von Mikroorganismen und Zellkulturen GmbH, Braunschweig, Germany) between passages 6 and 15 were seeded in 96-well plates (Nunc, Langensebold, Germany) at a density of 60,000 cells per well in assay medium (phenol red-free RPMI with L-glutamine ϩ 5% fetal calf serum). Cells were sensitized to TNF-␣-induced apoptosis by addition of 1 g/ml actinomycin D (Sigma). 2-Fold serial dilutions of anti-TNF-␣ scFvs (14.2-0.014 nM) and recombinant human TNF-␣ (PeproTech EC Ltd., London, UK) (1000 pg/ml or 19.16 pM final concentration) were prepared in assay medium and preincubated for 30 min at room temperature. After addition of the agonist/inhibitor mixtures, the cells were incubated for 20 h in a 37°C, 5% CO 2 -humidified incubator. Cell proliferation was assessed by measuring absorbance at 450 nm using a microplate reader (Genios TECAN, Maennedorf, Switzerland) after addition of a solution containing 1 mg/ml XTT (Applichem, GmbH, Darmstadt, Germany) in phenol redfree RPMI 1640 medium and 25 M phenazine methosulfate (Sigma). Data were analyzed using a four-parameter logistic curve fit, and the concentration (in mass units) of anti-TNF-␣ scFvs required to neutralize TNF-␣-induced apoptosis by 50% (EC 50 ) was calculated.
Temperature-induced Oligomerization and Degradation-578-FW1.4, 578-FW1.4opt, and 578-FW1.4gen were concentrated up to 20, 40, and 60 mg/ml in formulation buffer (20 mM trisodium citrate, 125 mM NaCl), pH 6.5, and incubated for 2 weeks at 40°C. Samples were analyzed before and after 14 days of incubation for degradation using 12.5% SDS-PAGE under reducing and nonreducing conditions. Size-exclusion HPLC was used to determine monomer content and soluble aggregates of the samples before and after the incubation period. Monomers were resolved from nonmonomeric species on a TSKgel Super SW2000 column (TOSOH Bioscience), and the percentage of monomeric protein was calculated as the area of the monomer peak divided by the total area of all product peaks.

Selection of Rabbit Monoclonal Antibodies from Hybridomas-
Monoclonal antibodies binding either to human VEGF 165 or human TNF-␣ were selected from rabbits either immunized with recombinantly produced human VEGF 165 or TNF-␣, respectively. Hybridoma screening (see under "Materials and Methods") resulted in the selection of seven VEGF neutralizing antibodies (375, 435, 509, 511, 534, 567, and 578) that potently blocked binding of VEGF to hVEGFR2, and eight TNF-␣ neutralizing antibodies (1,6,15,19,34,35,42, and 43) that potently inhibited TNF-␣-induced apoptosis in the murine L929 cell line. Sequence analysis of the rabbit monoclonal antibodies showed that 7 germ lines (out of 65 functional VL-gene segments) were represented in the selected binders (Table 1). Based on homology assessment, no germ line VH gene segments could be assigned, probably because of homologous recombination occurring in rabbit V-genes. The fact that an additional disulfide bond linking CDRH1 and CDRH2 was present in 6 of the 15 antibodies, together with results from sequence distance analysis (data not shown), indicates that at least two different germ line VH gene segments have been used.
Identification of a Human scFv Scaffold Suitable for Humanization of Rabbit Variable Domains-A yeast-based genetic screening of the human variable domain repertoire performed earlier resulted in a number of mature human scFvs antibody fragments that were well and solubly expressed in the cytoplasm of yeast, mammalian cells, and E. coli (22). From this pool of stable human scFv sequences, a variable light and variable heavy chain combination was chosen to serve as acceptor framework for rabbit CDRs. Selection of this human acceptor framework from the pool of stable scFv clones was based on the following: (a) on the level of soluble expression in yeast, and (b) on its homology to the rabbit variable region consensus sequence at specific core positions (see "Materials and Methods"). In contrast to rodents and humans, these core and interface residues frequently involved in CDR conformation and relative disposition of VL and VH, respectively (14,19,(27)(28)(29)(30)(31), are highly conserved in rabbits. Therefore, a stable human scaffold sharing high homology to rabbit sequences at such core positions was thought to generally support functional conformation of rabbit CDRs. The chosen framework, termed FW1.4, is of the V1-VH3 type. In combination with human CDRs, FW1.4 was well characterized in terms of thermal stability, in vitro folding, and expression in microbial and mammalian systems (data not shown). At the abovementioned core positions, this framework shares high percentage of identity with the respective consensus residues of rabbit light chains (88%) and heavy chains (85%).
Generation of Humanized scFvs from Rabbit Monoclonal Antibodies by Loop Grafting onto the Human scFv Scaffold FW1.4 or Derivatives Thereof-In one series of minimalistic grafts, CDRs (as defined under "Materials and Methods") from 15 independent rabbit monoclonal antibodies against TNF-␣ and VEGF were grafted onto the human acceptor framework 1.4 (FW1.4). Resulting clones were designated by the number of their parental rabbit IgG and the human acceptor framework FW1.4 (e.g. 578-FW1.4). In a second series, optimized grafts were designed (e.g. 578-FW1.4opt) by additional substitution of specific framework residues. Such substitutions were introduced at positions conserved in rabbits, which are involved in CDR conformation at positions potentially involved in direct contact with the antigen (see "Materials and Methods"). Of all framework positions considered, mainly L69, H23, H24, H49, H67, H68, H69, H71, H73, H78, and H94 frequently differed between the various rabbit antibodies and FW1.4 (see Table 3). The limited number of such positions and, more importantly, their relatively high degree of conservation in rabbit sequences (Table 5) suggests that only a small set of highly conserved framework substitutions is generally required for the humanization of rabbit antibodies. In fact the motif consisting of Thr-H23, Gly-H49, Thr-H73, Val-H78, and Arg-H94 is highly conserved and may thus present a generic solution to prepare a human acceptor framework (e.g. FW1.4) for minimalistic CDR grafting. To test this hypothesis, a third set of humanized scFvs was generated and characterized as detailed under "Motif Consisting of a Few Conserved Framework Residues in VH Accounts for Stability and CDR Structure of Humanized scFvs." Improved Production Yields following Loop Grafting onto FW1.4 and Its Derivatives-Purification of scFv antibody fragments is often limited by their relatively low expression yields and their tendency to form aggregates. We earlier experienced that stable scFvs can be efficiently produced by refolding from purified inclusion bodies. Humanized rabbit TNF-␣ and VEGF inhibitory scFvs were expressed as inclusion bodies in E. coli. Inclusion body expression levels were very similar between all molecules tested (data not shown). Following harvesting, inclusion bodies were subjected to a generic refolding process, and monomeric proteins were subsequently purified by preparative size-exclusion chromatography (for details see "Materials and Methods"). Most humanized scFvs based on FW1.4 or its derivatives were well producible by refolding in a generic lab scale process (see "Materials and Methods"), resulting in yields of up to 60 mg of purified protein/liter of refolding solution (see Table 1). Only two molecules (534-FW1.4 and 34-FW1.4) could not be purified in significant amounts. Optimization of framework regions by substitution of residues L69, H23, H49, H71, H73, H78 and H94 in 534-FW1.4opt and L15, L40, L72, H23,

TABLE 1 Pharmacodynamic and biophysical characterization of minimalistic and optimized grafts
Affinities of anti-VEGF and anti-TNF scFvs were measured with BIAcore using sensor chips with immobilized human VEGF 165 or human TNF-␣, respectively. Potencies of VEGF antagonists were measured in the VEGFR2 blocking assay, and the ability of TNF antagonists to neutralize TNF-␣-induced apoptosis was assessed in mouse L929 fibroblasts. Potencies of VEGF antagonists are compared with ranibizumab (relative potency ϭ EC 50, ranibizumab /EC 50, acFv ), and potencies of TNF antagonists are compared with infliximab (relative potency ϭ EC 50, infliximab /EC 50, scFv ). Thermostability measurements were performed by DSC. Refolding yield of the respective scFvs is expressed as amount (in milligrams) of refolded protein obtained out of 1 liter of refolding solution. NB denotes no binding detected; ND denotes analysis not done.
H49, H68, H69, H71, H73, H78, and H94 in 34-FW1.4opt only slightly improved production yield (see Table 1). Although production of some optimized grafts resulted in higher yields when compared with their minimalistic counterparts, the observed differences were minor in most cases. Indeed, the impact of the highly diverse CDRs on refolding yield seems to be more significant than the few substitutions in the framework regions. In contrast, attempts to purify the scFv fragments consisting of the parent rabbit variable domains were not successful. Minor quantities of secreted His-tagged scFv fragments were thus produced with poor purity from E. coli culture supernatants by Ni-NTA affinity chromatography (data not shown). These results indicate that refolding and purification of scFvs based on FW1.4 and its derivatives are particularly efficient.

CDR Grafting onto the Human scFv Scaffold Family Derived from FW1.4 Reproducibly Results in Functional Variable Domains-
Binding kinetics and potency of the minimalistic and optimized grafts of anti-VEGF and anti-TNF-␣ scFvs were compared. Data are summarized in Table 1. When CDRs were exclusively grafted from the rabbit sequence to the human scaffold, affinities of the resulting scFv to their target proteins were in the low nanomolar to high picomolar range (2.89 ϫ 10 Ϫ8 to 1.8 ϫ 10 Ϫ10 M) for VEGF inhibitors. One minimalistic graft (375-FW1.4) no longer showed binding to VEGF. Affinities of the TNF-␣ binding molecules were lower with dissociation constants (K D ) ranging from 2.62 ϫ 10 Ϫ7 to 5.16 ϫ 10 Ϫ10 M. Two minimalistic grafted molecules (1-FW1.4 and 6-FW1.4) did not show significant binding to TNF. In both groups, mutation of further residues in FW1.4 significantly enhanced binding, resulting in an improvement of K D by 1-3 orders of magnitude for all molecules tested. Only one molecule (1-FW1.4opt) still did not bind to its target. In terms of potency, optimized grafts were also clearly better compared with the variants generated by minimalistic CDR grafting. For example, 34-FW1.4opt, the most potent TNF-␣ antagonist, blocked TNF-␣induced apoptosis of mouse L929 fibroblasts 5.6-fold more efficiently than infliximab (compared in mass units), whereas none of the TNF-␣ binders generated by minimalistic grafting was sufficiently potent to rescue cell growth at the tested concentrations (Table 1). Among the VEGF inhibitors, 578-FW1.4opt exhibited the highest affinity in SPR experiments and the best potency in ELISA. This scFv also showed 1.3-fold stronger inhibition of VEGFinduced HUVEC proliferation compared with ranibizumab (see Fig. 1). In line with SPR analysis, the minimalistic graft 578-FW1.4 displayed a 14.8-fold lower potency (EC 50 of 2.614 nM versus EC 50 of 0.177 nM) compared with 578-FW1.4opt (see Fig. 1 and data not shown). Similarly, the potencies of the optimized grafts 511-FW1.4opt and 567-FW1.4opt were 68-fold (EC 50 of 3.832 nM versus EC 50 of 260.576 nM) and 5.7-fold higher (EC 50 of 6.694 nM versus EC 50 of 38.156 nM), when compared with the respective minimalistic counterparts (see Fig. 1  and data not shown). Surprisingly, when comparing off-rates FIGURE 1. Characterization of VEGF and TNF antagonists. A and B, binding of unblocked biotinylated human VEGF 165 (hVEGF) to ELISA wells coated with VEGFR2-Fc chimeric protein was measured in the presence of a constant amount of biotinylated hVEGF (2. 6 nM) and varying concentrations of scFvs and ranibizumab. C and D, dose-dependent inhibition of hVEGF-induced HUVEC proliferation by ranibizumab and scFvs. Cells were incubated with hVEGF (0.16 nM) and with increasing concentrations of scFv variants or ranibizumab. Cell proliferation was measured and plotted as percentages of cells treated with VEGF alone. E and F, neutralization of TNF-induced cytotoxicity by TNF antagonists. hTNF (19.16 pM) and serial dilutions of scFvs or infliximab were premixed, and L929 mouse fibroblast cell suspension was added. Proliferation was expressed as percentages of untreated cells. Curves show four-parameter fits to the data. All potency data were normalized to the standard. Triplicate data were used, and error bars represent standard deviations.
between optimized grafts of VEGF and TNF inhibitors and their progenitor rabbit scFv in SPR, the apparent loss in binding strength was only very moderate for most scFvs tested. Indeed, off-rates of humanized scFvs were equal or at most 7-fold lower than their rabbit precursors for VEGF inhibitors indicating almost complete retention of activity upon optimized grafting on FW1.4 (Table 2). For TNF inhibitors, two molecules, 1-FW1.4opt and 15-FW1.4opt, showed significant loss in binding strength, although off-rates of the remaining scFvs were equal or at most 14.6-fold lower compared with their rabbit precursors ( Table 2). Comparison of off-rates between the rabbit scFv of 578 and the full-length rabbit IgG of the same binder showed only a 4-fold decrease in the off-rate (3.66 ϫ 10 Ϫ5 versus 9.18 ϫ 10 Ϫ6 s Ϫ1 ). This difference in binding strength can most probably be attributed to avidity effects.
Human Variable Domain Scaffold Family Derived from FW1.4 Provides Drug-like Properties to Humanized scFvs-Aggregation during scFv fragment purification and storage often correlates with low thermal stability of the protein. To evaluate thermal stabilities of the humanized scFv antibody fragments, apparent melting temperatures were assessed by differential scanning calorimetry. Briefly, heat capacity changes of the scFv formulations were measured over a temperature gradient ranging from 25 to 95°C. Apparent melting temperatures (T m ) are summarized in Table 1. For the least stable molecule (6-FW1.4opt), T m was at 53.7°C. The T m of most other humanized fragments was above 60°C. Best results were obtained with the VEGF inhibitory scFvs 578-FW1.4opt, 509-FW1.4opt, and 511-FW1.4opt with T m values of 77.7, 80.1, and 86.9°C, and with the TNF inhibitory scFvs 15-FW1.4opt and 34-FW1.4opt with apparent melting temperatures at 71.6 and 78.1°C, respectively. For all scFvs, thermal stability of the optimized molecule was higher than that of the minimalistic graft, indicating that the grafted framework positions are relevant not only for CDR positioning but also for domain stability.
Motif Consisting of a Few Conserved Framework Residues in VH Accounts for Stability and CDR Structure of Humanized scFvs-Framework positions at which human residues were substituted by rabbit amino acids to generate optimized derivatives of FW1.4 are listed in Table 3. To assess the relative contribution of framework substitutions in the light and the heavy chain, domain shuffling experiments were performed with the humanized variable domains of clones 511, 578, 34, and 43. Therefore, heavy chains and light chains of minimalistic and optimized grafts were recombined. scFv constructs consisting of a minimalistic light chain and an optimized heavy chain exhibited near equal affinities and comparable or even better thermal stabilities when compared with the optimized grafts (compare Tables 1 and 4). In the opposite situation, when optimized light chains were combined with minimalistic heavy chains, the resulting scFvs showed lower thermal stabilities and affinities, which were decreased 6-, 5-, 293-, and 4-fold for such derivatives of 578, 511, 43, and 34, respectively (see also Table  4). These results suggest that affinity and stability of the humanized scFvs mainly depends on the few conserved rabbit amino acids introduced into the human heavy chain framework. For this reason, as hypothesized earlier, the set of amino acid motifs required to support rabbit CDR structure could be further generalized and possibly reduced to the five most conserved positions (Table 5). To test this hypothesis, a new generically applicable acceptor framework termed FW1.4gen was designed based on FW1.4. In the heavy chain, five residues were substituted with conserved rabbit amino acids at positions H23, H49, H73, H78, and H94. At the other "vernier zone" positions H69 and H71, the most frequently used amino acids in rabbit heavy chains were already present in the human FW1.4. At positions H24 and H67, valine and serine, respectively, would correspond to the consensus residues according to the collection of rabbit antibody sequences in the Kabat data base. However, the positions are less conserved in rabbits (Table 5), and moreover, in the majority of the rabbit variable domain sequences identified in the course of this study, alanine and phenylalanine were more abundant and were thus not changed (compare Table 3 and Table 5).
Generic Minimalistic CDR Grafting onto a Modified Human scFv Scaffold-Variable domains of the VEGF binding monoclonal antibodies 511 and 578 as well as of the TNF-binding antibodies 34 and 43 were humanized and reformatted into an scFv fragment by minimalistic grafting of CDRs onto FW1.4gen. This framework contains the rabbit amino acid motif Thr-H23, Gly-H49, Thr-H73, Val-H78, and Arg-H94. Affinities of the VEGF inhibitory scFv fragments 511-FW1.4gen and 578-FW1.4gen, as determined by SPR, were found to be 5.7 ϫ 10 Ϫ10 and 3.9 ϫ 10 Ϫ11 M, respectively (Table  6). When compared with their optimized counterparts 511- FW1.4opt and 578-FW1.4opt, affinity dropped slightly for 511-FW1.4gen by a factor of 2.7. A minor, if at all significant, loss in affinity was observed also for 578-FW1.4gen, which was attributed to the HG94R mutation in FW1.4gen, flanking CDRH3 and possibly affecting loop conformation. Following minimalistic grafting of the TNF inhibitory rabbit antibody 34 and 43, no loss in affinity was observed. On the contrary, the affinity of 34-FW1.4gen was more than 8-fold higher than that of 34-FW1.4opt (compare Tables 1 and 6). This is in sharp contrast to minimalistic grafting of antibody 43 onto the original FW1.4, where loss in affinity was about 217-fold (Table 1). In line with results from affinity measurements, relative potencies of the scFv fragments based on the FW1.4gen framework were also in the same range as their optimized variants on FW1.4 (compare Tables 1 and 6). In the VEGFR2 blocking assay, rela-tive potency of 511-FW1.4gen slightly dropped when compared with 511-FW1.4opt but was identical when assessed in the HUVEC proliferation assay. In contrast, no significant loss in potency was observed for 578-FW1.4gen in the VEGFR2 blocking assay as well as in the cell-based assay (see Fig. 1 and Table 6). Differences in relative potencies between the cellbased and ELISA-based assays can probably be attributed to day to day variations in assay performance. Thus, potencies of 511-FW1.4gen and 578-FW1.4gen did not significantly differ when compared with their optimized counterparts. In comparison with ranibizumab, a market-approved VEGF-inhibitory Fab fragment, the EC 50 value to block VEGF-induced proliferation of HUVECs for 511-FW1.4gen was 12.7-fold higher than for the benchmark molecule, whereas the potency of 578-FW1.4gen was similar to ranibizumab in the same assay. For the TNF-   (Table 6). Therefore, thermal stabilities were higher when compared with 511-FW1.4opt, 578-FW1.4opt 34-FW1.4opt, and 43-FW1.4opt (see also Fig. 2). Temperatureinduced unfolding experiments in FTIR confirmed the exceptionally high apparent melting temperatures of the same clones on the FW1.4gen framework with 71.8, 75.8, 76.5 and 65.6°C, respectively. In line with DSC data, again, the T m was lower for the respective optimized clones on FW1.4 ( Fig. 2 and Table 6).
578-FW1.4, 578-FW1.4opt, and 578-FW1.4gen were compared in a stability study under accelerated conditions. No degradation of the molecules was observed after incubation at 40°C for 2 weeks at a concentration of 60 mg/ml as assessed by

Analysis of amino acid frequencies at positions considered relevant for CDR conformation in rabbit and mouse V H
Percentages reflect the amino acid frequency of occurrence of 505 rabbit and 1478 mouse nonredundant antibody sequences taken from the "Kabat Database of Sequences of Proteins of Immunological Interest" as of April, 2007. SI is Simpson's index as measure of calculated diversity for each species. % agree is the percentage of agreement which indicates the percentages of sequences in the Kabat Database containing the consensus residue for the position indicated above. Cons indicates the consensus residue.

TABLE 6 Pharmacodynamic and biophysical characterization of VEGF and TNF antagonists grafted onto the FW1.4gen framework
Affinities of anti-VEGF and anti-TNF scFvs were measured with BIAcore using sensor chips with immobilized human VEGF 165 or human TNF-␣, respectively. Potencies of VEGF antagonists were measured in the VEGFR2 blocking assay, and the ability of TNF antagonists to neutralize TNF-␣-induced apoptosis was assessed in mouse L929 fibroblasts. Potencies of VEGF antagonists are compared with ranibizumab (relative potency ϭ EC 50, ranibizumab /EC 50, scFv ), and potencies of TNF antagonists are compared with infliximab (relative potency ϭ EC 50, infliximab /EC 50, scFv ). Thermostability measurements were performed by DSC and FTIR. Refolding yield of the respective scFvs is expressed as amount (in milligrams) of refolded protein obtained out of 1 liter of refolding solution.  (Fig. 3). The results presented above demonstrate that highly potent and highly stable humanized scFv antibody fragments can be generated in reproducible manner by minimalistic grafting of CDRs from rabbit monoclonal antibodies onto a single optimized human acceptor framework.

DISCUSSION
Antibody fragments offer particular advantages over full-size antibodies. Most fragments can be produced in microbial expression systems. Due to their low molecular weight, smaller fragments such as scFv antibody fragments freely pass kidney filtration and are cleared from the circulation with terminal half-lives (t1 ⁄ 2 ) of a few hours, whereas t1 ⁄ 2 of full-size antibodies ranges up to several weeks. In contrast to IgGs, scFvs have excellent tissue penetration properties and were shown to efficiently penetrate into cartilage and even across certain epithelial barriers (1,2,32). Thus, from a pharmacokinetic perspective, the scFv format meets requirements for local and superficial therapies to achieve high local concentrations. Because of the short half-life in the circulation, local application of scFvs leads to low systemic exposures reducing the risk for systemic side effects. Thus, local therapies with scFv antibody fragments represent a promising approach to cope with side effects related to systemic therapies with IgGs. As a consequence, a superior efficacy/safety profile is expected.
However, in many cases scFvs do not possess the required drug-like properties. Low solubility and high aggregation rates are considered major drawbacks of the scFv format. Besides this, high affinity binding of the progenitor variable domains is required to compensate for the lack of avidity of monovalent scFvs, a prerequisite that is only rarely met with rodent antibod-  Table 6. ies. Our results demonstrate that humanization of rabbit variable domains by simple grafting of antigen-binding loops onto the FW1.4gen scaffold reproducibly results in humanized scFvs with drug-like biophysical properties that bind with high affinity to their targets.
The rabbit antibody repertoire represents an attractive source for antibodies for several reasons. First, rabbit antibodies mostly show significantly higher affinities when compared with rodent antibodies. Second, generation of antibodies that are cross-reactive toward mouse antigens is possible in many cases. This is of particular interest for the preclinical evaluation of therapeutic antibodies in mouse models of human diseases. Third, framework variability in rabbits is very limited leading to the assumption that a generic framework suitable for generation of humanized scFvs by simple grafting of CDRs could be designed. Rabbit antibodies have been humanized before. For example, Rader and co-workers (33, 34) have applied phage display to humanize rabbit antibodies. In their work, selection of successfully humanized molecules was based on binding activity of the Fab fragment presented on a phage. Although biophysical properties of these Fab fragments were not characterized in detail, it is likely that the corresponding scFv fragments would exhibit considerable variability in terms of stability and solubility, requiring further characterization and possibly even engineering to identify scFvs with drug-like properties. Moreover, the use of in vitro display systems for the humanization of larger numbers of variable domains is timeconsuming as follows: on one hand because of the screening procedure, and on the other hand because of laborious characterization of the numerous positives resulting from each individual progenitor antibody. In contrast, the FW1.4gen framework offers a technically simple humanization solution that reproducibly results in humanized scFv fragments with favorable biophysical properties.
As an alternative to humanization of animal antibodies, genetic libraries have been generated in the past that contained diverse sets of synthetic CDRs on drug-like human scFv scaffolds (15). Although such libraries have been applied to generate binders against a variety of targets (35)(36)(37), this approach is limited by the typical shortcomings of in vitro display technologies as follows: (a) potential loss of weakly expressed high affinity binders, due to preferential selection of well expressed molecules, and (b) the need for purified target molecules. Particularly for the generation of antibodies against complex and difficult to purify antigens, such as GPCRs or ion channels, animal immunization may still be advantageous. For example, immunization with transfected host cells or vaccination with cDNA allows specific presentation of integral membrane proteins in their native conformation and natural environment to the immune system. In contrast to in vitro display systems, such an approach does not co-select for unspecific binding to other proteins on the cellular surface. For these reasons, animal immunization remains an important starting point for the generation of antibodies.
In this study we demonstrated that a broad spectrum of monoclonal antibodies derived from rabbit immunization can be successfully humanized and reformatted to an scFv by simple transfer of antigen-binding loops to a human framework.
This framework was specifically selected and optimized to provide a universal acceptor scaffold for rabbit CDRs and to confer drug-like properties to the resulting scFv. In a first stage, humanized scFvs were generated from 15 independent rabbit monoclonal antibodies directed against two different protein targets, by exclusive transplantation of CDRs onto the human acceptor framework FW1.4. In a second stage, optimized set and specific framework residues were additionally substituted. These residues were assumed to interact with the antigen or to be involved in defining CDR structures. Although the first set of fragments already showed relatively strong interaction with the antigen, all but one optimized scFv bound to their target with very high affinities ranging from 4.7 ϫ 10 Ϫ9 to 1.5 ϫ 10 Ϫ11 M. More than 50% of them displayed equilibrium dissociation constants in the subnanomolar range. These scFvs were well producible in a generic production process and exhibited excellent thermal stabilities.
A generically applicable acceptor framework for rabbit CDRs, termed FW1.4gen, was created by substituting a set of five amino acids in FW1.4 with residues that are conserved in rabbit heavy chains (HT23, HG49, HT73, HV78, and HR94). Exclusive grafting of CDRs onto the FW1.4gen-scaffold resulted in humanized scFvs with binding strength similar to optimized grafts and superior biophysical properties with apparent melting temperatures between 69.2 and 85.6°C in DSC. Moreover, comparison of potencies of these VEGF and TNF-␣ inhibitory antibody fragments to ranibizumab and infliximab showed that the VEGF inhibitory 578-FW1.4gen and the TNF inhibitors 34-FW1.4gen and 43-FW1.4gen were as potent as the benchmark molecules. Therefore, biochemical and biophysical properties of these scFvs demonstrate that the method described here reproducibly results in single-chain antibody fragments that have the potential to be developed for therapeutic applications.
More recent methods to isolate cDNA sequences coding for target-specific monoclonal antibodies from animal and human sources (e.g. B-cell isolation techniques (38)) will possibly generate significantly more hits as compared with the hybridoma technique. Higher numbers of clones would in turn increase the chance to identify rare events, such as binders against difficult targets, which would broaden the application spectrum for therapeutic antibodies. A challenge for hybridoma-independent methods, however, is the reliable generation of sufficient protein amounts to perform functional and biophysical screenings, which is essential for the identification of drug-like antibody fragments. For this reason, methods that allow fast cloning, reproducible production, and purification are required. The use of generally applicable frameworks enabling high throughput humanization of wild type scFvs, fast cloning, and generic production of fragments with favorable biophysical properties in microbial systems presents a promising approach for the development of future scFv-based therapeutics.