Increased Fab thermoresistance via V_H-targeted directed evolution

Abstract

Antibody aggregation is frequently mediated by the complementarity determining regions within the variable domains and can significantly decrease purification yields, shorten shelf-life and increase the risk of anti-drug immune responses. Aggregation-resistant antibodies could offset these risks; accordingly, we have developed a directed evolution strategy to improve Fab stability. A Fab-phage display vector was constructed and the V_H domain targeted for mutagenesis by error-prone PCR. To enrich for thermoresistant clones, the resulting phage library was transiently heated, followed by selection for binding to an anti-light chain constant domain antibody. Five unique variants were identified, each possessing one to three amino acid substitutions. Each engineered Fab possessed higher, Escherichia coli expression yield, a 2–3°C increase in apparent melting temperature and improved aggregation resistance upon heating at high concentration. Select mutations were combined and shown to confer additive improvements to these biophysical characteristics. Finally, the wild-type and most stable triple variant Fab variant were converted into a human IgG1 and expressed in mammalian cells. Both expression level and aggregation resistance were similarly improved in the engineered IgG1. Analysis of the wild-type Fab crystal structure provided a structural rationale for the selected residues changes. This approach can help guide future Fab stabilization efforts.

Introduction

The therapeutic monoclonal antibody market is estimated to grow to nearly $58 billion in 2016 (Reichert, 2015), while other antibody applications include reagents, diagnostics and crystallization chaperones. Antibody aggregation can result from the chemical and physical stresses of the production process and can have significant detrimental effects on manufacturing, shelf life, efficacy and especially risk of immunogenicity (Rouet et al., 2014). Changes in buffer formulation can help to prevent degradation or aggregation (Daugherty and Mrsny, 2006), as can post-translational modifications and the addition of charged fusion tags (Schaefer and Pluckthun, 2012). The observation that the variable domain sequence often dominates aggregation propensity (Rouet et al., 2014) has led to efforts aimed at improving stability by in vitro evolution (Hoogenboom, 2005) or by grafting complementarity determining regions (CDRs) onto a stabilized scaffold (Jung and Pluckthun, 1997; Willuda et al., 1999). For instance, after the initial development of the stable human acceptor framework 4D5 (Carter et al., 1992), multiple approved antibody therapeutics have been developed using the same scaffold (Carter, 2006). Alternate stabilized acceptor frameworks and readily accessible methods to reduce aggregation propensity of existing Fabs would aid in antibody development for many applications.

One approach used to confer antibody resistance to aggregation employed heat denaturation of a single-domain antibody when fused to the surface of the M13 phage, followed by selection of correctly folded variants using a conformation-specific binding partner (Jespers et al., 2004). This approach was subsequently used to improve the thermoresistance of both human (Arbabi-Ghahroudi et al., 2009) and camelid V_H-only antibodies (Turner et al., 2014), and was extended to isolated human variable light chain (V_L) domain (Dudgeon et al., 2012), diabody (Rodriguez-Rodriguez et al., 2012), single-chain variable fragment (scFv) (Fennell et al., 2013) and single-chain T cell receptor (scTCR) (Gunnarsen et al., 2013) formats. Variants selected from these libraries exhibited improved behavior as soluble protein, including increased resistance to heat inactivation, improved thermal stability and higher expression levels. However, this approach has not yet been applied to Fabs. When produced in Escherichia coli, Fabs are typically composed of four domains in two polypeptides: V_L–C_L and V_H–C_H1. Biophysical experiments have demonstrated that a single weakly stable variable domain can impact global Fab stability (Worn and Pluckthun, 2001; Rothlisberger et al., 2005). Fabs employ intramolecular and intermolecular disulfide bonds, typically resulting in co-operative unfolding (Demarest et al., 2006). It thus appears that both the molecular architecture and the choice of binding partner are critical to ensuring selection of productive variants from a thermally challenged Fab library.

We previously developed EE peptide (sequence EYMPME)-binding scFv (Pai et al., 2011) and Fab (Johnson et al., 2015) antibody fragments. Peptide-binding monoclonal antibodies are used as therapeutics (Chu et al., 2014), are widely employed as affinity purification/detection reagents, and their fragments have been proposed for use as crystallization chaperones (Lieberman et al., 2011). The EE-binding Fab, αEE (Johnson et al., 2015), is well-behaved and crystallizes readily but may have greater value as a crystallization chaperone and antibody scaffold if expression and stability/aggregation resistance were further increased. To improve the behavior of this Fab, we pursued a directed evolution strategy. First, the expression and purification protocols were optimized to improve the yield of soluble monomers. By comparing the biophysical characteristics of related scFvs and Fabs that differed only in their heavy chain CDR sequences, we concluded that this domain was the most likely to benefit from engineering optimization. Next, a V_H-targeted library was created by error-prone PCR, displayed on phage and transiently heated followed by selection for binding to a C_L-binding antibody. After four rounds of panning, five unique variants were identified and characterized in phage display and soluble formats. All exhibited higher activity levels and reduced aggregation upon heating compared with wild-type Fab. By converting the top engineered variant to IgG format, we demonstrated that this effect is independent of antibody format. Analysis of the αEE crystal structure revealed that many selected variants likely possessed reduced flexibility in CDR H1. We demonstrate that thermoresistance engineering by phage display extends to the Fab format and have identified structural mechanisms for conferring aggregation resistance that can guide future antibody engineering efforts.

Materials and methods

Molecular biology

The pFabs vector for soluble Fab expression in E. coli was a generous gift from George Georgiou at the University of Texas at Austin (Levy et al., 2001). The pFab dicistronic vector uses a pelB leader sequence at the N-terminus of both the heavy and light chains for periplasmic expression, contains a decahistidine tag appended to the C-terminus of the C_L domain and a FLAG peptide appended to the C-terminus of the C_H domain. A native C-terminal inter-chain disulfide bond precedes the peptide tags. The FLAG peptide was first removed by standard QuikChange mutagenesis to ensure that it would not interfere with other applications, creating the vector pFabF. The V_L domain was first amplified from the previously described 3D5/EE_48 scFv (αEE scFv) construct (Pai et al., 2011) by gene-specific primers appending NcoI and NotI restriction sites. The resulting PCR product was digested by NcoI and NotI and ligated into a similarly digested pFabF plasmid. The cloning procedure was repeated with the V_H domain using NheI and HindIII restriction sites to create the EE peptide-binding Fab termed αEE. The hexa-histidine binding Fabs 683 and 3D5 (Pai et al., 2011), the FLAG-peptide (DYKDDDDK) binding Fab EEh14.3 (K. Entzminger, in preparation) and the pertussis toxin-binding Fab hu1B7 (Sutherland and Maynard, 2009) were cloned similarly into the pFabF vector. To construct derivative Fabs containing the thermoresistance mutations, Kunkel mutagenesis was performed as described previously (Pai et al., 2012) using a modified annealing step (Huang et al., 2012) and three 26 bp oligonucleotides targeting each region for mutagenesis.

To create the Fab phage display vector, a multiple cloning site was first added at the C-terminus of the C_H domain using standard QuikChange mutagenesis, introducing the restriction sites SgrAI and RsrII. This appends the peptide AGGSGGSRSG in-frame with the Fab and creates the vector pFabFS. The cassette including the c-myc peptide tag, amber stop codon and truncated pIII gene was PCR-amplified from the plasmid pMopac24 (Hayhurst et al., 2003) using oligonucleotides appending an N-terminal SgrAI restriction site and C-terminal stop codons in triplicate followed by an RsrII restriction site. The resulting PCR product was digested by SgrAI and RsrII and ligated into a similarly digested pFabFS plasmid, creating the phage display vector p3Fab. All oligonucleotides were purchased from Integrated DNA Technologies, and the fidelity of all constructs was confirmed by DNA sequencing at the University of Texas Core Facility.

Protein purification

The scTCR DO11.10 (Entzminger et al., 2012) was used to control for nonspecific binding or used as a carrier protein to present EE peptides and test for Fab binding. DO11–EE₁ and DO11–EE₂ contain one or two EE peptides respectively inserted into the inter-domain linker (Pai et al., 2011). These scTCRs were expressed from the pAK400 (Krebber et al., 1997) in E. coli BL21 in 500 ml culture volumes as previously described (Maynard et al., 2005). Following Ni²⁺-based immobilized metal affinity chromatography (IMAC) purification, the proteins were further purified by size exclusion chromatography (SEC) using an ÄKTA FPLC system (GE Healthcare) and a Superdex S75 (GE Healthcare) column with a phosphate-buffered saline (PBS; pH 7.4) running buffer. The previously described maltose-binding protein (MBP) variants containing C-terminal EE and hexa-histidine peptides (MBP-EE-His) or a hexa-histidine peptide only (MBP-His) (Pai et al., 2011) were identically expressed and purified. The scFvs αEE and 683 (Pai et al., 2011) and EEh14.3 (KCE, unpublished) were expressed similarly, with the exception of using the optimized growth and IMAC buffer formulation changes described later.

Fab expression was performed similarly using the pFabF vector and 250 ml culture volumes with the following modifications. Overnight cultures containing antibiotics and 2% glucose were used to inoculate 250 ml Terrific Broth (TB) containing antibiotics and 0.1% glucose at an OD₆₀₀ of 0.1 and grown with shaking at 25°C until OD₆₀₀ reached 0.5. Isopropyl β-d-1-thiogalactopyanoside (IPTG) was then added to a final concentration of 1 mM and growth continued for 4 h. To reduce aggregation, IMAC buffers were replaced with 25 mM Tris pH 9.0 containing 10% glycerol and 100 mM NaCl, with the wash buffer containing 20 mM imidazole and the elution buffer augmented with 0.5 M imidazole. To further improve monomer yield, the pFabF vector was co-transformed into BL21 with the compatible pBAD33 (Levy et al., 2001) vector containing the protein chaperone Skp. This vector boosts Skp expression levels in the periplasm, which can increase active purification yields of some antibody fragments (Entzminger et al., 2012). To induce Skp expression, l-arabinose was added to a final concentration of 0.2% 30 min prior to IPTG induction. A Superdex S200 (GE Healthcare) column was used with hepes-buffered saline (HBS) (pH 7.4) as running buffer to further resolve the monomer fraction. Analytical SEC was performed using a Superdex S200 column and by injecting 50–100 µl of previously purified monomer. For all proteins, concentration was measured using the Micro-BCA protein assay kit (Pierce) and purity confirmed by 12% SDS–PAGE. All graphical data were prepared using GraphPad Prism 6 and all experiments repeated at least twice.

Phage production and enzyme-linked immuno-sorbent assay

The p3Fab phagemid was transformed into E. coli XL1-Blue, and Fab-bearing phage produced by M13 super-infection, harvested and titered as described previously (Sidhu and Weiss, 2004). Briefly, starter cultures were used to inoculate 30 ml TB containing 200 µg/ml ampicillin and 0.1% glucose in baffled flasks and grown at 37°C with shaking. At an OD₆₀₀ of 0.5–1.0, IPTG was added at 1 mM final concentration for an additional 2 h incubation at 25°C with shaking, after which M13KO7 helper phage (NEB) were added at a multiplicity of infection >10 and cultures incubated similarly for 2 h. Kanamycin was then added at 70 µg/ml final concentration to select for cells possessing the helper phagemid and growth continued at 25°C overnight with shaking.

To measure activity, enzyme-linked immuno-sorbent assay (ELISA) was performed as described previously (Entzminger et al., 2012) with the following modifications: 96-well plates (Costar) were coated with 0.2 µg/ml DO11–EE₁ or control DO11 in PBS overnight at 4°C. Fab-bearing phage at a concentration of 10¹¹–10¹² cfu/ml were heated for 10 min at the indicated temperature then cooled to 25°C in a thermocycler, followed by 10-min incubation at room temperature and centrifugation to remove precipitate prior to addition to ELISA wells. After the final wash step, signal was developed using 3,3′,5,5′-tetramethylbenzidine substrate (Vector Laboratories), and absorbance at 450 nm read using a SpectraMax M5 microplate reader (Molecular Devices) after quenching with an equal volume of 1 M H₂SO₄. Data represent the average of two replicates, and GraphPad Prism 6 was used to fit data to a three-parameter logistic model to derive T₅₀ values, representing the temperature at which 50% activity loss was observed.

Library construction and phage panning

A V_H library was constructed using the megaprimer-based approach described previously (Pai et al., 2012). Briefly, error-prone PCR was performed using the Mutazyme II polymerase (Agilent Technologies) according to commercial instructions and with oligonucleotides targeting the V_H domain. Nested primers were then used to generate single-stranded mutagenic megaprimers by PCR using Vent polymerase (NEB) according to the commercial instructions. For the template strand, library contamination by wild-type was prevented by introduction of three consecutive stop codons in CDR H3 of αEE in the p3Fab vector by QuikChange mutagenesis. A modified annealing procedure (Huang et al., 2012) was used to improve the percentage of recombinant clones. Prior to panning, an aliquot of library DNA was transformed into E. coli XL1-Blue, and DNA sequencing of 10 random clones used to estimate the efficiency of incorporation of the mutagenic primers and to confirm the targeted error rates.

Library DNA was transformed into E. coli XL1-Blue and phage produced and purified as described above. Phage were heated at 58°C at a concentration of 10¹⁰–10¹² cfu/ml for 10 min then cooled to 4°C in a thermocycler, followed by a 10-min incubation on ice and centrifugation to remove precipitate. Next, 10⁹–10¹¹ cfu/ml treated phage in blocking buffer (5% nonfat milk in PBS) was added to eight replicate blocked wells in a high-binding 96-well plate (Costar) and allowed to incubate for 30 min to remove nonspecific phage. The supernatant was then transferred to eight replicate blocked wells coated overnight with a 1:1000 dilution of anti-human kappa light chain antibody (αHuCκ, Sigma K3502) and allowed to incubate for 1–2 h. Wells were washed multiple times with PBS-0.5% Tween, eluted with 0.1 M glycine, pH 2.2 and neutralized with 2 M Tris, pH 7.5. Escherichia coli XL1-Blue at an OD₆₀₀ of 0.5–1.0 was infected with eluted phage, and phage were amplified and purified for the following round as described above. Washing stringency was increased for each successive round of panning, with four total rounds performed. Individual phage clones from the last two panning rounds were grown in 250 µl cultures in 96-well plates (Costar) as described previously (Pai et al., 2011) and analyzed by ELISA for retention of binding to αHuCκ after heating at 58 or 68°C. The top clones with the highest signal were identified by DNA sequencing, grown in 30 ml cultures and tested by monoculture phage ELISA as above.

Soluble Fab biophysical measurements and ELISA

Solubility of αEE, 683 and EEh14.3 scFvs and Fabs was measured by concentration using a single 10 kDa MWCO amicon filter to 5–26 mg/ml, incubation for 4 days at 4°C, centrifugation for 10 min at high speed (16 000 rcf) to pellet insoluble aggregates and measurement of the remaining soluble protein using a BCA assy. Thermoresistant Fab variants were cloned into the pFabF (variants E2, E3 and E5) or pFabFS (E1 and E4) vectors by NcoI/HindIII or NheI/SgrAI digestion, respectively, and purified as above with Skp co-expression. The use of the pFabFS vector for E1 and E4 was necessary due to the presence of C_H1 mutations in these clones. Control experiments with αEE demonstrated that the additional C-terminal peptide appended by the pFabFS vector does not significantly change any of the measured biophysical characteristics (data not shown). Thermal stability was measured using a fluorescent assay as described previously (Lavinder et al., 2009). Briefly, 0.1–0.5 mg protein or a PBS buffer blank was assayed using the Protein Thermal Shift Dye Kit (Applied Biosystems) according to the manufacturer's instructions. Twenty microliter volumes were heated in quadruplicate using a 7900HT Fast Real-Time PCR System (Applied Biosystems) from 25 to 90°C at a rate of 0.016°C/s. Melting temperature was calculated by fitting normalized curves to a four-parameter logistic model using GraphPad Prism 6, and average values reported from two separate experiments using independently prepared proteins.

Fabs were tested for peptide binding by ELISA as above, with the following exceptions: DO11–EE₁ or control DO11 was coated at a concentration of 50 µg/ml, Fabs or an scFv control was added in a dilution series from 80 to 0.63 µg/ml, a 1:1000 dilution of αHuCκ–HRP conjugate (Sigma A7164) in blocking buffer was used for detection and all steps were performed at room temperature. To test for retention of peptide binding after heat treatment, the ELISA was repeated with Fabs incubated at 25 or 61°C for 1 h then cooled to 25°C in a thermocycler at a concentration of 40 µg/ml. Samples were incubated at room temperature for 10 min, centrifuged then diluted to a concentration within the dynamic range of the ELISA prior to incubation. DO11–EE₂ was coated instead to increase sensitivity. Percentage activity was reported by normalizing between A₄₅₀ of the 25°C treatment sample for each Fab and background A₄₅₀ values. Aggregation resistance was measured as described previously (Turner et al., 2014), with Fabs heated similarly as above but at a concentration of 0.5 mg/ml. Precipitate was pelleted by centrifuge and the A₂₈₀ of the remaining soluble protein measured in triplicate using a Nanodrop 2000. Values were normalized to A₂₈₀ values measured prior to heating for each Fab.

Surface plasmon resonance

To measure binding kinetics, DO11–EE₁ was immobilized to a CM5 sensor chip (GE Healthcare) by amine coupling chemistry at 200–250 RU using a BIAcore 3000 (GE Healthcare). The signal from a flow cell coupled with control scFv was used to correct for nonspecific binding to the matrix. Fabs or an unrelated scFv control was injected in duplicate for 60 s at each concentration in a dilution series from 1000 to 62.5 nM in HBS-0.005% Tween-20 at a flow rate of 50 µl/min to minimize mass transport effects, with dissociation monitored for 5 min. Two 30 s injections of 4 M MgCl₂ were used to remove any remaining Fab from the surface between each run. The association rate constant (k_on), dissociation rate constant (k_off) and equilibrium dissociation constant (K_D; K_D = k_off/k_on) were calculated by fitting to a Langmuir 1:1 binding model using BIAevaluation 3.2 software.

IgG expression and characterization

The αEE variable chains were subcloned from the pFabF vector into the pMAZ-IgL and pMAZ-IgH vectors (Mazor et al., 2007) to create wild-type IgG/αEE. The IgG/αEE.3 variant was constructed by Kunkel mutagenesis as described above. Plasmid DNA was prepared according to the manufacturer's instructions (Qiagen Midiprep kit). Light and heavy chain plasmids were both transfected at 1 µg/ml into CHO-K1 cells using Lipofectamine 2000 (Life Technologies) and dulbecco's modified eagle medium (Sigma) media supplemented with 10% low IgG fetal bovine serum. Transfections were performed at a cell density of 1 × 10⁶ cells/ml and cells grown for 5 days, with culture supernatant harvested and replenished each day.

An Fc capture ELISA was performed to measure expression yields. Briefly, goat anti-human IgG Fc antibody (Southern Biotech 2047-05) was coated overnight at 1:500 dilution. After blocking wells with PBS/5% milk, 3.2-fold dilutions in blocking buffer were performed with a human antibody standard starting at 5 µg/ml concentration or with undiluted culture supernatants. After 1-h incubation, wells were washed and a 1:1250 dilution of αHuCκ–HRP in blocking buffer was added for an additional hour. Plates were washed and developed as above, and concentration determined by comparing to the standard. Total yield is reported for each day, normalizing to the cell density at the time of transfection.

To measure resistance to thermal inactivation, an activity ELISA was performed similarly as described above. Briefly, MBP-EE-His or the control protein MBP-His, which lacks the EE peptide, was coated at a concentration of 50 µg/ml. Culture supernatants were left at room temperature or heated for 1 h at the indicated temperature, followed by cooling for 10 min at room temperature. The samples were then diluted in blocking buffer to within the dynamic range of the assay and the ELISA repeated as described above for the Fab variants.

Results

Improving the purification yield of Fab monomer

Noting the success of crystallization chaperone strategies for solving membrane protein structures, we previously proposed using peptide-binding antibodies as generic chaperones and toward that aim engineered a unique EE peptide-binding scFv (Lieberman et al., 2011). Since the majority of antibody-based crystallization fragments is Fabs, we converted the αEE scFv to Fab format (Johnson et al., 2015). However, isolation of monomer exhibited some batch-to-batch variability primarily due to aggregate formation during purification. To improve monomer yield, we first sought to optimize the purification protocol.

The original method involves periplasmically targeted expression in E. coli and oligohistidine-tagged protein isolation by sequential IMAC and SEC. This has been robustly applied to a variety of scFvs, scTCRs and other recombinant proteins (Maynard et al., 2005), but resulted in low-level production of monomer here (Method 1, Fig. 1A). Method 1 involves overnight growth to stationary phase, which can hinder expression of toxic or aggregation-prone proteins. Inclusion of glucose in starter cultures to inhibit leaky expression and controlled growth to OD₆₀₀ = 0.5 prior to IPTG addition slightly increased monomer yield but significantly increased co-purification of aggregates and higher-order oligomers (Method 2, Fig. 1A).

Fig. 1.

αEE Fab aggregates during purification and possesses only moderate thermoresistance. (A) αEE Fab was expressed using a variety of expression and purification conditions and isolated by sequential IMAC and SEC (shown). Low levels of Fab monomer were obtained by (1) expression induced after overnight growth to stationary phase. Recovery of monomer was additively improved by (2) expression induced after controlled growth to OD₆₀₀ = 0.5 and (3) change in IMAC buffer formulation (addition of 10% glycerol, increasing pH from 8.0 to 9.0 and reducing NaCl from 0.5 to 0.1 M) and (4) co-expression of the protein chaperone Skp. All experiments used 250 ml culture volumes. Purified Fab proteins were separated on a Superdex 200 column calibrated with high MW Gel Filtration Calibration Kit (indicated by arrowheads: blue dextran, ferritin 440 kDa, beta amylase 200 kDa, aldolase 158 kDa, conalbumin 65 kDa, ovalbumin 44 kDa, carbonic anhydrase 29 kDa, cytochrome c 12 kDa). Based on these, the expected elution volume for a 50 kDa Fab is 15.6 ml. (B) Activity remaining after thermal stress. Purified αEE Fab protein was heated for 10 min at the indicated temperature, ranging from 25 to 65°C, cooled and binding to the EE peptide assessed by ELISA. Incubation at 55°C results in 50% activity remaining after cooling. A carrier protein presenting a single EE peptide within an inter-domain linker (DO11–EE₁, solid circles and line) was used as ligand. No background binding was observed to a control protein lacking the EE peptide (DO11, open circles, dashed line).

Upon closer inspection, amorphous precipitate that clarified with agitation was observed during IMAC elution. IMAC wash and elution buffers were next optimized for reduced formation of soluble aggregates, with 10% glycerol added, pH increased from 8.0 to 9.0, NaCl concentration decreased from 0.5 to 0.1 M and elution performed using imidazole instead of EDTA (Method 3, Fig. 1A). This dramatically improved monomer yield and decreased the amount of soluble aggregates eluting in the void fraction. Glycerol preferentially stabilizes the native state for some proteins (Gekko and Timasheff, 1981), and higher salt conditions may screen surface charges, leading to reduced charge–charge repulsion between molecules and favoring aggregation (Kumar et al., 2011). The αEE Fab possesses a calculated pI of 8.6, and maximal monomer isolation was only observed with buffer pH values above the pI, conditions in which the protein is negatively charged and charge–charge repulsion can further minimize aggregation.

Finally, we co-expressed the protein chaperone Skp from a compatible plasmid (Method 4, Fig. 1A). We have previously observed that Skp co-expression helps improve active yields for some antibody fragments (Entzminger et al., 2012), and in this case, it greatly reduced formation of aggregates and oligomers. By screening a limited set of buffer formulations and expression conditions, we successfully and reproducibly improved overall Fab monomer yield from 18 to 65%.

Selecting thermoresistant variants

Resistance to thermally induced inactivation has been shown to correlate with improved expression level and increased solubility (Dudgeon et al., 2012). To measure thermoresistance of the αEE Fab, we next performed an activity ELISA. Purified αEE Fab was incubated at room temperature or heated to a single temperature, ranging from 48 to 65°C for 1 h, cooled and the remaining EE peptide-binding activity measured by ELISA (Fig. 1B). Only 50% activity was observed after heating to 55°C, and no activity was detected after treatment at temperatures >60°C. Antibody fragments have been routinely engineered to survive exposure to temperatures well above this threshold (Miklos et al., 2012; Turner et al., 2014), suggesting that αEE only possesses moderate thermoresistance.

We next pursued a directed evolution approach to improve αEE biophysical characteristics. Previous studies have demonstrated that even a single weakly stable variable domain can limit the stability of an entire Fab (Rothlisberger et al., 2005). To identify the least stable domain, we examined a panel of three related scFvs, each binding a different peptide: αEE (EE peptide), 683 (hexa-histidine peptide) and EEh14.3 ((Pai et al., 2011), K.C.E., unpublished, FLAG peptide). All three scFvs share an identical V_L domain and differ primarily in the three heavy chain CDRs. Each scFv was converted to a Fab, which allowed us to explore the effects of V_H sequence in multiple formats. The scFvs were expressed from the pAK400 plasmid using the Method 3 conditions described above, and Fabs using the optimized Method 4 conditions.

Conversion to the Fab format increased the total and relative monomer yields only for αEE and EEh14.3 (from 0.74 to 1.8 mg/l culture and 9–27% monomer; Table I). The apparent melting temperature (T_m) was measured by differential scanning fluorimetry (DSF), and indicated that conversion to the Fab format increased the T_m for both αEE and 683 (54–65 and 60.8–66.7°C, respectively). Solubility measurements were limited by total protein yield in some cases, as both αEE and 683 scFvs and Fabs were fully soluble even at the highest concentration tested (13 and 26 mg/ml, respectively). However, conversion to the Fab format dramatically improved and solubility of EEh14.3 (from 3.8 to 13.6 mg/ml). These results demonstrate that αEE possesses sub-optimal biophysical characteristics when compared with 683 in both formats. This led us to reason that the V_H sequence could likely be improved through directed evolution.

Table I.

Biophysical characteristics of scFvs/Fabs with identical V_L domains

	scFv			Fab
	αEE	683	EEh14.3	αEE	683	EEh14.3
Expression level (mg/l culture)	1.51	3.63	0.74	1.92	2.30	1.83
% monomera	53	52	9.4	56	56	27
Solubility (mg/ml)	12.8 (Pai et al., 2011)	>26	3.8 ± 0.2	>13	>18	13.6 ± 0.1
Tm (°C)	54.5 ± 0.1	60.8 ± 0.2	NDb	65.2 ± 0.2	66.7 ± 0.2	ND

^aMeasured upon initial purification.

^bNot detected.

To perform directed evolution using a V_H-targeted library, we first constructed a vector for phage-based selection. A C-terminal FLAG peptide was removed to prevent interference with use in other applications resulting in the vector pFabF (Fig. 2), which was used for soluble expression as described above. A multiple cloning site was then appended to the C_H domain's C-terminus, followed by cloning of a cassette containing the c-myc peptide and the bacteriophage M13 coat protein pIII to create the phage display vector p3Fab. Phage production in E. coli produces the heavy chain tethered to pIII, with the light chain paired post-translationally via a native inter-chain disulfide bond. To confirm that the p3Fab vector produces active phage-tethered Fab and to identify conditions within the dynamic range of the assay, a phage ELISA was performed with DO11–EE₁ coated at various concentrations and phage serially titrated (Supplementary Fig. S1A). Coating with 200 ng/µl DO11–EE₁ provided sufficient signal over background without signal saturation even at high phage concentrations. The phage ELISA was then repeated to optimize the coating concentration for αHuCκ (Supplementary Fig. S1B).

Fig. 2.

Fab vectors used in this study. (A) Schematic diagram of Fab expression and phage display vectors. The dicistronic parental pFab vector targets each chain to the E. coli periplasm using a pelB leader sequence. The FLAG peptide was removed from the C-terminus of the C_H1 domain to create the pFabF soluble expression vector. A multiple cloning site containing SgrAI and RsrII restriction sites was appended at the same location to create pFabFS. The cassette including a c-myc peptide tag, an amber stop codon that allows read-through in E. coli amber suppression strains and a truncated pIII phage coat protein variant was amplified from the vector pMopac24 and added by SgrAI/RsrII digestion to create the phage display vector p3Fab. (B) Detail of the p3Fab vector containing αEE and showing all restriction sites (underlined) used for cloning variable regions.

Heat challenge was chosen to select for Fabs with improved aggregation resistance and expression levels, as these properties have been identified in variants isolated from similarly screened V_H-only phage libraries (Jespers et al., 2004; Arbabi-Ghahroudi et al., 2009; Turner et al., 2014). To identify optimal screening conditions, αEE-bearing phage were transiently heated to a range of temperatures, cooled and tested for retention of activity after heat treatment by ELISA (Fig. 3A). The temperature at which 50% activity is observed, T₅₀, was measured as 58°C. Temperature-dependent loss of binding to αHuCκ was also demonstrated by phage ELISA at multiple temperatures and multiple αHuCκ concentrations tested (Fig. 3B). If heat stress is to be used during selections, it is important that it not disrupt the ability of the phage to infect E. coli, which would result in clonal loss during selection. To test if heat treatment affects phage infectivity, an equal amount of untreated or 68°C-challenged phage was used to infect E. coli, with no difference in phage titers observed (Supplementary Fig. S1C).

Fig. 3.

Heat treatment reduces Fab-phage binding activity. (A) Phage displaying αEE Fab were heated for 10 min at the indicated temperature, cooled and tested for ligand-binding activity by ELISA. Binding to the DO11–EE₁ ligand (solid line) shows 50% activity after incubation at 58°C. No background phage binding to the DO11 control lacking the EE peptide (open circles, dashed line) was observed. (B) Phage displaying the αEE Fab via a heavy chain-gpIII fusion exhibit temperature-dependent loss of binding to antibodies binding the light chain constant domain (αHuCκ). No binding was observed for an scFv phage control which lacks a HuCκ domain.

The V_H region was amplified by error-prone PCR with a targeted mutation rate of 1 or 2.5%, and the resulting fragments were used separately to synthesize library DNA by Kunkel mutagenesis. DNA sequencing of random clones prior to panning identified an amino acid error rate of 1.0 ± 1.2 and 1.5 ± 1.7%, respectively, and the proportion of recombinant clones ranged from 80 to 88%. Equal volumes of the two synthesis reactions were combined and transformed to E. coli for a final library size of 1.22 × 10⁸. A first attempt at panning using DO11–EE₁ capture and 60°C heat treatment failed to yield any productive clones (data not shown). Because coupled unfolding of all four Fab domains is commonly observed (Demarest et al., 2006) we reasoned that the use of a C_L-binding monoclonal antibody would be sufficient to identify V_H mutations conferring thermoresistance. Phage were re-propagated and the library subjected to four rounds of panning using heat challenge at 58°C followed by αHuCκ capture. To monitor library progress, the number of phage input and output from each round of panning was measured. Rounds 3 and 4 showed increases in the total number of output phage, suggesting enrichment of positive clones (Supplementary Fig. S2A). DO11–EE₁ binding before and after heat treatment was tested by ELISA using the unselected input phage population for each round. Rounds 3 and 4 showed almost no loss in activity upon heat treatment, again suggesting positive enrichment (Fig. 4A).

Fig. 4.

Enrichment of thermoresistant Fabs during phage library panning. (A) After heating the unselected total input phage population at room temperature or 58°C for 10 min and cooling on ice, ELISA was performed to assess the loss of DO11–EE₁ binding. Rounds 3 and 4 show increased thermoresistance. (B) Thermoresistant Fab variants (E1–E5) were purified from a large-scale 30 ml culture in phage display format and tested for temperature-dependent loss of binding to DO11–EE₁. Phage concentrations are normalized among all variants at all temperatures. E1–E5 all show increased T₅₀ and ELISA A₄₅₀ values compared with wild-type for ligand-specific binding (solid and dashed lines) with negligible nonspecific binding to uncoated, blocked well (dotted lines).

To identify thermoresistant clones, monoclonal phage were produced from 364 randomly selected colonies from the output population for rounds 3 and 4. Clones were screened by monoclonal phage ELISA under three conditions: replicate phage were incubated for 10 min at 25, 58 or 68°C, cooled and assessed for binding to an αHuCκ-coated or uncoated control plate to assess nonspecific binding (Supplementary Fig. S2B). Variants E1–E5 were selected for further screening on the basis of high specific ELISA signal after heat treatment when compared with wild-type (Table II). Each amino acid change was created by mutation of a single nucleotide, and the same codon was used for each amino acid substitution that appeared in multiple clones. Substitutions at positions 75, 76 and 82b occurred in framework region 3, while positions 32 and 98 occur in CDRs H1 and H3, respectively. Unexpectedly, since the variable heavy chain was targeted for mutagenesis, two substitutions were also selected in C_H1, at positions 116 and 187.

Table II.

Biophysical characterization of selected Fab variants

	Phage ΔT50 (°C)a	Yield (mg/l culture)	Monomer ΔTm (°C)b	% activity after heatingc	% soluble after heatingd	Selected mutations
αEE	–	1.54	–	19 ± 2	23 ± 2	–
E1	+3.4	2.52	+3.0	68 ± 6	62 ± 6	VH:S76N/CH:L187F
E2	+3.0	2.57	+2.0	70 ± 6	61 ± 6	VH:S32P
E3	+2.6	2.35	+2.2	53 ± 3	87 ± 8	VH:S75N/S76N/S98R
E4	+2.4	2.45	+2.6	68 ± 6	50 ± 4	VH:S82bN/CH:F116Y
E5	+1.8	2.75	+2.1	57 ± 5	53 ± 1	VH:S76N

^aRelative temperature at which 50% activity loss was observed as measured by phage ELISA (Fig. 4D).

^bApparent melting temperature for soluble Fab measured using DSF (Fig. 5B).

^cDO11–EE₁ binding of soluble Fab measured by ELISA after heating at 61°C for 1 h (Fig. 5C).

^dPercentage of soluble protein remaining after heating at 61°C for 1 h (Fig. 5D).

To confirm increased thermoresistance, αEE and E1–E5 phage were then propagated using 30 ml culture volumes and subjected to a monoclonal ELISA screen for DO11–EE₁ binding after heating as above. To further discriminate T₅₀ values, a smaller range of temperatures was used, in eight increments between 53 and 63°C. All variants show T₅₀ increases of 2–3°C compared with wild-type (Fig. 4B, Table II). The maximum ELISA signal for each variant was than roughly double that of wild-type. Because the number of phage was normalized among all Fabs at every temperature, this suggests increased expression level, which would lead to a higher proportion of phage incorporating Fab. Alternatively, variations in affinity could account for differences in ELISA signals.

Biophysical and antigen binding properties of selected variants

We next wanted to determine whether increased stability was also present in soluble Fabs containing these amino acid changes. Phage Fab clones E1–E5 were transferred to the soluble Fab expression vector and purified using the optimized procedure detailed above. All engineered variants expressed better than wild-type, an additional increase of up to 1.2 mg/l culture (Table II). Higher stability often confers increased E. coli expression levels for recombinant proteins (Demarest et al., 2006). Analytical SEC of the purified monomer fraction shows that all Fabs elute at the expected volume (∼16 ml, Fig. 5A). To measure relative thermal stabilities, Fabs were subjected to DSF (Lavinder et al., 2009). Fabs E1–E5 all exhibit 2–3°C increases in apparent T_m compared with αEE (Fig. 5B, Table II), in close agreement with the phage activity ELISA. This demonstrates that Fab biophysical characteristics measured in the phage display format are predictive of solution behavior, as has been observed previously with heavy chain single-domain antibodies (Dudgeon et al., 2012).

Fig. 5.

Engineered Fabs exhibit increased thermoresistance in soluble format. (A) Analytical SEC elution profiles of the purified monomer fraction for all Fabs show a single peak with a similar elution volume as expected for a 50 kDa protein based on the calibration curve (inset). Elution volumes of molecular weight standards are indicted by arrowheads, observed Fab elution volumes by circles. (B) Apparent T_m was measured by DSF. Variants E1–E5 exhibit 2–3°C increases in T_m compared with αEE. Averaged traces are shown for each Fab. (C) Fabs were heated at a concentration of 40 µg/ml at 61°C for 1 h, cooled and tested for DO11–EE₁ binding by ELISA. A₄₅₀ values were normalized to the A₄₅₀ of an untreated sample for each Fab separately to account for any binding affinity differences. E1–E5 all retain higher activity levels compared with αEE. (D) Fabs were heated at a concentration of 0.5 mg/ml at 61°C for 1 h and centrifuged to remove precipitate. The A₂₈₀ of the remaining soluble fraction was measured and all values normalized to A₂₈₀ values prior to heat treatment for each Fab. E1–E5 all exhibit increased soluble protein compared with αEE.

To next assess thermoresistance, Fabs were left untreated at 25°C or heated to 61°C for 1 h at a concentration of 40 µg/ml. After centrifugation to remove insoluble material, Fabs were diluted to a concentration previously determined to be within the assay's dynamic range, and binding to DO11–EE₂ was tested by ELISA. To account for potential affinity differences among the variants, the reported activity for each Fab variant after heat treatment was normalized to the signal resulting from the same variant incubated at 25°C. The wild-type αEE Fab retained only 19% activity after heating, while all engineered variants retained 53–68% activity (Fig. 5C, Table II). No binding to the control ligand was observed for any sample at any temperature (data not shown).

To assess aggregation resistance, heat treatment (61°C for 1 h) was repeated at a higher concentration of 0.5 mg/ml. The amount of soluble protein remaining was measured and normalized to the initial protein concentration. All selected variants exhibited dramatically reduced levels of insoluble aggregates upon heating, with 52–95% remaining soluble while only 25% of the wild-type remained soluble (Fig. 5D, Table I). While this fraction may include soluble aggregates, collectively the ability of selected variants to bind ligand after heating and reduced formation of insoluble aggregates support their increased thermostability.

Because selection was performed in the absence of peptide antigen, it was unclear whether binding affinity would be affected for the selected variants. All Fabs were titered for DO11–EE₁ binding by ELISA to determine room temperature EC₅₀ values. Variants E1–E5 retained antigen binding, with no background binding to the DO11 control observed (Fig. 6). Variants E1, E2 and E4 all possess calculated EC₅₀ values within error compared with αEE (data not shown). Variant E3 showed reduced binding (EC₅₀ of 165 vs. 105 nM), likely due to the S98R mutation in CDR H3.

Fig. 6.

Stabilized Fab variants retain affinity for EE peptide antigen. All engineered variants retain affinity for the EE peptide as presented by the DO11–EE₁ ligand (solid lines) in the absence of thermal stress as measured by room temperature ELISA. Variants E1, E2 and E4 appear to have similar affinity as the wild-type Fab, while E3 and E5 exhibit modest reduction in affinity. No binding to the DO11 control protein was observed for any Fab (dashed lines).

Binding affinity and kinetics were next quantified by surface plasmon resonance (SPR) for variants E1, E2 and E4. Variant E3 was excluded as this variant includes a charged residue in the CDR, which is less likely to translate to other antibodies; E5 since it only includes the residues change V_H:S76N, which is also present in E1. DO11–EE₁ was coupled to the surface as bait and Fabs flowed over as prey with all steps performed at 25°C. As expected from the ELISA data, the affinities are all very similar and within the nM range (Table III). The αEE Fab was created using the variable domains from an scFv previously reported to possess a K_D of 212 nM for DO11–EE₁ by SPR (Pai et al., 2011), which agrees well with our measured value of 229 nM. The affinities of the engineered variants are slightly lower than αEE, though the small difference may be due to increased stability. No binding was observed to an unrelated protein control by SPR (data not shown). These data demonstrate that the selected mutations did not significantly impact antigen binding, even though antigen was not used during selection.

Table III.

Characterization of Fab ligand-binding kinetics by SPR

	kon (M–1 s–1)	koff (s–1)	KD (nM)
αEE	1.6 ± 0.04 × 105	3.8 ± 0.2 × 10–2	229 ± 12
E1	1.9 ± 0.05 × 105	2.3 ± 0.1 × 10–2	126 ± 7
E2	1.2 ± 0.02 × 105	1.2 ± 0.1 × 10–2	107 ± 10
E4	3.5 ± 0.1 × 105	3.4 ± 0.2 × 10–2	96 ± 6

Structural analysis of selected mutations

We recently reported the crystal structure of wild-type αEE Fab (PDB 4X0K (Johnson et al., 2015), which provides mechanistic insights for the stabilizing amino acid substitutions identified. Almost all substitutions involved solvent-exposed residues (Fig. 7A), which can accommodate a higher mutational load than buried residues. Moreover, conformational restriction of surface loops is commonly observed during directed evolution experiments aimed at improving thermostability (Arnold et al., 2001).

Fig. 7.

Structural implications of selected mutations. (A) Molecular modeling was performed in PyMOL using the αEE Fab crystal structure (Johnson, et al., 2015). Thermoresistance mutations were modeled using the mutagenesis function, and the most common rotamer was used unless significant steric clashes were apparent. All mutations cluster at the surface with the exception of C_H:L187F. The light chain is rendered in white and the heavy chain in light gray. V_H CDRs are shown dark gray and selected mutations in black stick format. (B) V_H:P32 occurs directly within CDR H1 and likely restricts the conformation of this loop. The dihedral angle of wild-type Ser (−60°) agrees well with the dihedral angles generally adopted by proline (φ = −63 ± 15°). Wild-type side chains are rendered in white stick format. (C) V_H:S76N, the most frequently observed mutation in the sequenced population, likely allows hydrogen bonding to the hydroxyl group of S28 and backbone carbonyl of Y27 upon heating, which occupy the loop immediately preceding CDR H1. This same stabilizing hydrogen bonding network has been observed previously (Wong, et al., 2011). (D) V_H:N82b protrudes into solvent, showing no clear stabilizing mechanism. (E) C_H:F116Y decreases the hydrophobicity of a relatively solvent-exposed residue, and the hydroxyl group points toward the V_H domain. (F) C_H:F187 may better fill a hydrophobic pocket compared with wild-type Leu, though local rearrangements would be necessary to accommodate the bulkier group.

For the serine-to-proline substitution at position V_H32 (variant E2; Fig. 7B), the dihedral angles of the wild-type residue at this position agree well with those typically adopted by proline, so this mutation may not introduce a significant conformational change. During our analysis of the αEE crystal structure, a molecular dynamics simulation was performed with αEE complexed to an EE peptide-containing client protein (Johnson et al., 2015). The CDR H1 did not exhibit significant flexibility during the simulation (Supplementary Fig. S3), which was performed at the moderate temperature of 37°C. This suggests that residues conferring thermoresistance may not target inherently flexible loops but instead likely decrease the conformational flexibility of the denatured state. Many previous studies have designed or serendipitously discovered proline substitutions within or near surface loops that confer thermostability (Tian et al., 2010; Yu and Huang, 2014). The same S32P mutation has been selected from scFv and Fab affinity maturation libraries (Saviranta et al., 1998, Van Blarcom et al., 2010), as well as proline substitutions in other CDRs (Maynard et al., 2002).

The V_H:S76N mutation was represented singly in variant E5 and in combination with other mutations in E1 and E3, and was the most highly represented mutation identified after screening. A previous study compared the crystal structures of a germline antibody containing S76 and an affinity-matured variant generated possessing N76 (Wong et al., 2011). Molecular dynamics simulations showed that the germline antibody possessed a highly flexible CDR H1, especially at increased temperatures, and that the asparagine mutation significantly reduced CDR flexibility, likely through hydrogen bonding to the backbone carbonyl of Y27 and the hydroxyl group of T28. The αEE possesses a serine at position 28, retaining the hydroxyl group and suggesting S76N may also reduce CDR H1 flexibility (Fig. 7C). A more rigorous analysis at this position using the mutagenesis function in Coot (Emsley et al., 2010) did not predict hydrogen bonding with Y27 or S28 (Supplementary Fig. S4A and Table S1), again supporting the hypothesis that thermostabilizing interactions predominantly impact the partially unfolded state. Finally, the S76N mutation was also selected from an IgG affinity maturation library (Bowers et al., 2011).

Variant E4 possessed a mutation in both the V_H and C_H1 domains. V_H:S82b protrudes directly into the solvent, and the asparagine mutation does not appear to confer any additional hydrogen bonding or optimized packing interactions with nearby residues (Fig. 7D), making it unlikely that this mutation improved thermoresistance. Similar results were observed with the V_H:S75N and S98R mutations (data not shown). Analysis was repeated in Coot and identified a few potential hydrogen bonding interactions conferred by the mutations (Supplementary Fig. S4A–C and Table S1) but did not readily identify stabilizing mechanisms. In contrast, C_H:F116Y decreases hydrophobicity at this relatively solvent-exposed location (Fig. 7E), and the hydroxyl group of tyrosine points toward the V_H domain. Analysis in Coot identified a putative interaction with V_H:D11 within acceptable hydrogen bonding distance (3.0 Å, Supplementary Fig. S4D and Table S1). This interaction could potentially serve to stabilize this interface, even prior to heating.

The C_H:L187F mutation appears to additively improve thermoresistance in combination with V_H:S76N as measured by most biophysical characteristics (E1 vs. E5). Mutation at this site appears to improve hydrophobic packing compared with wild-type leucine (Fig. 7F), though local structural rearrangements would be necessary to accommodate this bulkier residue. Interestingly, the C_H:L187F mutation was chosen in a previous study to improve Fab stability based on structural comparison to an ultra-stable V_H domain (Teerinen et al., 2006). However, it was only tested in combination with another C_H1 mutation and did not improve Fab expression level. By design, neither C_H mutation was present during the initial error-prone PCR step but instead must have arisen during amplification from the template strand. The ability of these low-frequency clones to emerge from selection hints at their functional importance.

Transferability of stabilizing residue changes

Because structural analysis suggested that multiple unique mechanisms were responsible for conferring thermoresistance to the engineered Fabs, we next wanted to test if select mutations could be combined to further improve biophysical characteristics. While V_H:S32P and V_H:S98R were some of the most robust solutions, these positions occur within CDRs. Because we wanted to identify a set of optimized residues that could be transferred to other Fabs without impacting antigen binding, we focused on framework positions. The Asn mutations at positions V_H:S75 and S82b were not selected for reasons detailed above. Variant E1, which contains both V_H:S76N and C_H:L187F mutations, possesses increased T_m and increased thermoresistance compared with E5, which only contains V_H:S76N; thus both residues were chosen for use in the final engineered variant. According to the structural analysis, the substitution V_H:F116Y appears to confer thermoresistance via a unique mechanism. Thus, this set of three substitutions was combined to create the triple variant αEE.3. The T_m and solubility analyses were repeated, and the triple mutant conferring further biophysical improvements: a 3.6°C increased T_m and 9.4-fold increased aggregation resistance compared with wild-type without loss of antigen binding (Fig. 8, Table IV).

Fig. 8.

Triple variant αEE.3 Fab exhibits further increased thermoresistance. (A) The apparent T_m of purified, soluble αEE and variant containing all three stabilizing substitutions, αEE.3, Fab protein was measured by DSF. The αEE.3, exhibits a further enhanced thermostability for a total 3.6°C increase compared with αEE, suggesting additive effects of each change. Averaged traces are shown for each Fab. (B) Purified αEE and αEE.3 Fab protein was heated at a concentration of 0.5 mg/ml at 61°C for 1 h and centrifuged to remove precipitate. The A₂₈₀ of the remaining soluble fraction was measured and all values normalized to A₂₈₀ values prior to heat treatment for each Fab. (C) The binding affinity of ELISA αEE.3 appears unchanged vs. the original αEE, as measured by ELISA. Data for αEE.3 are shown with circles, αEE with squares; binding to DO11–EE₁ ligand represented by solid lines and, while dotted lines and hollow icons (αEE.3 (circle) and αEE (square)) show binding to a control surface coated with DO11.

Table IV.

Biophysical characterization of thermoresistant Fabs employing alternative CDRs and scaffolds

	Tm (°C)	ΔTm (°C)	% soluble after heat treatmenta	Fold change
αEE	63.7 ± 0.1	–	8 ± 2	–
αEE.3b	67.3 ± 0.1	+3.6	76 ± 4	+9.4
683	65.5 ± 0.1	–	13 ± 5	–
683.3	65.8 ± 0.1	+0.3	31 ± 7	+2.5
3D5	58.1 ± 0.2	–	18 ± 1	–
3D5.3	58.8 ± 0.2	+0.7	24 ± 1	+1.3
hu1B7c	60.5 ± 0.2	–	24 ± 2	–
hu1B7.2	60.7 ± 0.3	+0.2	27 ± 1	+1.1

^aPercentage of soluble protein remaining after heating 0.5 mg/ml solution to 61°C for 1 h.

^b‘0.3’ variants contain V_H:S76N/C_H:F116Y/C_H:L187F mutations compared with wild-type.

^cWild-type hu1B7 already possesses an Asn at position V_H76.

We next wanted to test if these substitutions would similarly enhance thermostability for other antibodies. The Fabs 683 is a related protein that differs essentially only in the heavy chain CDR sequences. These were cloned to our soluble expression vector, and purified along with their triple mutants (denoted ‘0.3’ in Table IV). The T_m improvements were negligible compared with wild-type for all stabilized variants. Aggregation resistance was improved, but this effect decreased the further the Fab sequence diverged from αEE (Fabs are listed in order of similarity to αEE in Table IV). We also tested the effect for the Fab hu1B7, which possesses human, as opposed to murine, variable framework regions and thus could be used to test the effect of these mutations on highly divergent Fab sequences. Antibody hu1B7 already contains a V_H:N76 residue, so only the C_H1 mutations were transferred for the hu1B7.2 variant. The effect of the thermoresistance mutations on both T_m and aggregation resistance was negligible for this Fab. These results demonstrate that these mutations are dependent on Fv sequence and that the engineering procedure would likely need to be repeated for each new Fab. All of these Fabs retain a hydroxyl group at position V_H28, which we expect would be necessary for hydrogen bonding with the V_H:N76 mutation.

Thermoresistance-conferring mutations are independent of antibody format

We next wanted to determine whether these identified substitutions would also enhance biochemical features in the full-length IgG1 antibody format. The αEE variable regions were cloned into heavy and light chain mammalian expression vectors to create IgG/αEE, and the triple mutations added to the heavy chain to create the optimized IgG/αEE.3 variant. The vectors were transfected to CHO-K1 cells for transient expression, and the culture supernatant harvested daily to collect secreted IgG for up to 5 days. ELISA was used to quantify total expression yield, and IgG/αEE.3 expressed at higher levels at each time point compared with wild-type, as was observed for bacterial Fab expression (Fig. 9A).

Fig. 9.

Selected substitutions confer thermoresistence independent of antibody format and expression host. (A) V_L and V_H–C_H1 domains for wild-type αEE and the thermoresistant variant αEE.3 were cloned to mammalian expression vectors in IgG format for transient transfection and expression in CHO cells. The culture supernatant was harvested each day, and IgG concentration was measured by anti-Fc ELISA after comparison to standards. αEE.3 shows increased expression yield compared with wild-type. (B) The CHO culture supernatant containing αEE and αEE.3 IgG (∼10 μg/ml) were heated to the indicated temperature for 1 h followed by cooling at room temperature and testing for DO11–EE₁ binding by ELISA. A₄₅₀ values were normalized to the A₄₅₀ of the untreated samples for each IgG. αEE.3 shows increased thermoresistance compared with wild-type.

We next wanted to determine whether IgG/αEE.3 exhibits aggregation resistance properties when compared with IgG/αEE. CHO culture supernatant was left at room temperature or heated to a single temperature between 25 and 67°C for 1 h, followed by cooling. Insoluble material was pelleted and the samples diluted to within a concentration determined to be within the dynamic range of the ELISA assay, then tested for binding to EE peptide (Fig. 9B). Each IgG was normalized to the ELISA signal measured for the untreated sample, and IgG/αEE.3 retained higher activity levels compared with IgG/αEE at elevated temperatures (42 and 15%, respectively, at 62°C). The optimized residues are not constrained to the Fab format, but confer improvements in a full-length IgG format as well. These results are likely to be useful for applications requiring monoclonal antibodies, almost any of which could benefit from stability optimization.

Discussion

Directed evolution to enhance stability

Based on comparison to a panel of related antibodies, we reasoned that αEE Fab possessed a sub-optimal V_H domain. Most stability engineering studies target the V_H domain (Rouet et al., 2014) as it tends to be less stable compared with V_L (Caravella et al., 2010), and scFv folding experiments support this approach (Hoyer et al., 2002). One Fab engineering study used a germline consensus-based approach targeting all four domains, but stabilizing mutations were almost exclusively found in the V_H chain (Demarest et al., 2006). Except in special circumstances employing weakly stable V_L or ultra-stable V_H domains, future Fab stabilization efforts will likely benefit from a similar V_H-targeted approach.

It has been previously shown that aggregation resistance on phage is a good predictor of solution behavior for single-domain antibodies (Dudgeon et al., 2012), and we observed this for our Fabs as well. Multiple previous studies have used phage display to improve thermoresistance, melting temperature and/or expression level for a variety of antibody formats (Jespers et al., 2004; Arbabi-Ghahroudi et al., 2009; Rodriguez-Rodriguez et al., 2012; Fennell et al., 2013; Turner et al., 2014). To the best of our knowledge, this technique has not yet been extended to Fabs. For capture during selection, these previous studies used either antigen or the conformation-specific binders protein A (human V_H and camelids), protein L (human V_L) or a variable domain-binding antibody (scTCR) (Gunnarsen et al., 2013).

Extending this technique to the four-domain Fab format required careful consideration to ensure that selected V_H mutations provided overall conformational stability instead of local stabilization that could disrupt the V_L–V_H interface or otherwise negatively impact global structure. To ensure that V_H mutations conferred global Fab stabilization, a C_L-binding antibody was used for selections. Fabs that are disulfide bonded both intramolecularly and intermolecularly, as is the case in our expression vectors, commonly exhibit coupled unfolding of all four domains (Demarest et al., 2006). Pilot experiments with phage-displayed αEE confirmed temperature-dependent loss of binding to αHuCκ, supporting the choice of this binding partner. Using a similar approach, scTCR stability was enhanced using a conformational Vβ antibody to selected improved clones (Richman et al., 2009).

Mechanisms of stabilization

Conformational restriction of CDR H1 appeared to be the most likely mechanism by which many of the clones exhibited thermoresistance; this played a role in all of the engineered variants except E4. Restricting loop flexibility is a well-characterized method used to engineer protein thermostability (Yu and Huang, 2014) and has been successfully applied to many different enzymes (Tian et al., 2010; Wijma et al., 2013) and even scTCRs (Gunnarsen et al., 2013). Optimization of CDR H1 specifically is also common in many antibody stabilization studies. A previous study using camelids found that the most stabilizing mutation improved hydrophobic packing of CDR H1 (Turner et al., 2014). Others have noted that charged residues within or adjacent to this CDR can reduce aggregation propensity of single-domain antibodies (Perchiacca et al., 2011; Dudgeon et al., 2012). While multiple asparagine mutations were identified in our thermoresistant Fabs, no charged residues were selected with the exception of S98R. This is in contrast to previously developed scFv thermoresistance engineering strategies (Miklos et al., 2012). Careful biophysical experiments by Perchiacca et al. (2014) have demonstrated that the charged residue effect is likely dependent on the net charge of the antibody scaffold.

Why was antigen binding retained in all of the thermoresistant Fabs? For this specific antibody, heavy chain CDR2 and 3 residues are expected to make direct contact with peptide residues, but we do not expect significant side-chain interactions involving CDR H1 or any of the other residues selected in this work (Pai et al., 2011; Johnson et al., 2015). In particular, V_H residue His 50 appears to stabilize peptide residue Tyr 2 through hydrophobic interactions, while V_H residue Arg 95 forms key polar interactions with multiple peptide side chains (Tyr 2, Met 3, Glu 6). This latter interaction may explain the reduced binding of variant E3, which includes a CDR H3 S98R substitution. Notably, neither CDR H1 residue S32 nor adjacent residues are expected to form peptide contacts, which is relevant as the S32P change was selected as a stabilizing substitution.

Although changes in CDR conformation can negatively impact binding, restriction of CDR flexibility is a general mechanism attributed to the affinity maturation process by allowing preconfiguration of the antigen binding site (Schmidt et al., 2013). By comparing molecular dynamics simulations using germline and mature antibody structures, Wong et al. showed that this mechanism extends to CDR H1 (Wong et al., 2011). Affinity maturation studies support these observations, as the same V_H:S32P and V_H:S76N mutations identified here were also selected in other antibodies (Saviranta et al., 1998; Bowers et al., 2011). Because antigen was absent during selection, preconfiguration of the binding site was not the basis for selection of these residues from our library. Instead, CDR conformational restriction was likely selected for its stabilizing effects on the unfolded state. It has been previously shown that mutations can improve colloidal stability without affecting thermodynamic stability (Rouet et al., 2014) and, with the exception of mutant side chains, without imparting significant structural changes to the native state (Dudgeon et al., 2012). Even the proline mutation, which occurs directly within CDR H1, likely does not change native loop conformation due to the matching dihedral angles of the wild-type residue at this position.

This is to the best of our knowledge the first demonstration of Fab thermoresistance engineering by phage-based selection. We used this method to engineer Fabs with improved expression level and increased aggregation resistance, properties desirable for crystallization chaperones or therapeutics in general. Our methods and the general stabilizing mechanisms that we identified can help the pursuit of more robust antibodies.

Supplementary data

Supplementary data are available at PEDS online.

Funding

This work was supported by the National Institutes of Health (1R01 GM095638 to J.A.M. and DK091357 to R.L.L.), National Science Foundation (0845445 to R.L.L.) and the Welch Foundation (#F-1767 to J.A.M.).