Improving the Developability of an Antigen Binding Fragment by

May 22, 2019 - 1. Global Research Technologies. , Novo Nordisk A/S, Måløv, Denmark. 2. Department of Pure and. Applied Biochemistry, Lund Univer. si...
0 downloads 0 Views 2MB Size
Article Cite This: Biochemistry 2019, 58, 2750−2759

pubs.acs.org/biochemistry

Improving the Developability of an Antigen Binding Fragment by Aspartate Substitutions Laila I. Sakhnini,*,†,‡ Per J. Greisen,† Charlotte Wiberg,† Zoltan Bozoky,†,∥ Søren Lund,† Adriana-Michelle Wolf Perez,† Hanne S. Karkov,† Kasper Huus,† Jens-Jacob Hansen,† Leif Bülow,‡ Nikolai Lorenzen,† Maria B. Dainiak,*,† and Anja K. Pedersen*,§ †

Global Research Technologies, Novo Nordisk A/S, 2760 Måløv, Denmark Department of Pure and Applied Biochemistry, Lund University, 223 62 Lund, Sweden § Chemistry, Manufacturing and Control, Novo Nordisk A/S, 2820 Gentofte, Denmark

Downloaded via UNIV OF SOUTHERN INDIANA on July 31, 2019 at 10:22:47 (UTC). See https://pubs.acs.org/sharingguidelines for options on how to legitimately share published articles.



S Supporting Information *

ABSTRACT: Aggregation can be a major challenge in the development of antibody-based pharmaceuticals as it can compromise the quality of the product during bioprocessing, formulation, and drug administration. To avoid aggregation, developability assessment is often run in parallel with functional optimization in the early screening phases to flag and deselect problematic molecules. As developability assessment can be demanding with regard to time and resources, there is a high focus on the development of molecule design strategies for engineering molecules with a high developability potential. Previously, Dudgeon et al. [(2012) Proc. Natl. Acad. Sci. U. S. A. 109, 10879−10884] demonstrated how Asp substitutions at specific positions in human variable domains and single-chain variable fragments could decrease the aggregation propensity. Here, we have investigated whether these Asp substitutions would improve the developability potential of a murine antigen binding fragment (Fab). A full combinatorial library consisting of 393 Fab variants with single, double, and triple Asp substitutions was first screened in silico with Rosetta; thereafter, 26 variants with the highest predicted thermodynamic stability were selected for production. All variants were subjected to a set of developability studies. Interestingly, most variants had thermodynamic stability on par with or improved relative to that of the wild type. Twenty-five of the variants exhibited improved nonspecificity. Half of the variants exhibited improved aggregation resistance. Strikingly, while we observed remarkable improvement in the developability potential, the Asp substitutions had no substantial effect on the antigenic binding affinity. Altogether, by combining the insertion of negative charges and the in silico screen based on computational models, we were able to improve the developability of the Fab rapidly.

M

depending on intrinsic factors, such as amino acid sequence and structure, as well as extrinsic factors such as protein concentration, pH, temperature, and excipients.16−19 It has been reported that interfaces for protein−protein interactions, such as the complementarity-determining regions (CDRs) of antibodies, can mediate aggregation.20−25 A common strategy for decreasing the aggregation propensity of antibodies is the insertion of charged residues. For instance, Miklos et al.26 generated supercharged scFvs with increased aggregation resistance by substitution of surfaceexposed residues with charged amino acids. Non-CDR positions were targeted, and the mutations were selected on the basis of computational models in Rosetta to avoid the introduction of unfavorable interactions.26 Perchiacca et al.27 reported insertion of negatively charged residues near the edges of the hydrophobic CDR3 loop of a human variable

onoclonal antibody (mAb)-based reagents are widely used as therapeutic drugs,1,2 probes in diagnostic applications,3−5 and affinity ligands in preparative downstream processes6 in the biopharmaceutical industry. In addition, fragments of mAbs, such as single-chain variable fragment (scFv) and antigen binding fragment (Fab), have attracted a great deal of interest as their smaller size can reduce steric hindrance,7−9 improve column capacity during downstream processing,6 and be used for multivalent formatting.10 In the development of pharmaceutical antibodies, there are several important attributes, apart from desired antigen binding and functionality, that need to be achieved such as low aggregation propensity, low nonspecificity, and high thermodynamic stability.11 The measurement of these biophysical properties during early stage screenings is known as developability assessment.11−13 Protein aggregation is considered a major challenge as it can compromise product quality, safety, and efficacy,14 e.g., enhanced immune responses in patients.15 Protein aggregation is a complex process as proteins often form different types of aggregates following different pathways © 2019 American Chemical Society

Received: March 22, 2019 Revised: May 16, 2019 Published: May 22, 2019 2750

DOI: 10.1021/acs.biochem.9b00251 Biochemistry 2019, 58, 2750−2759

Article

Biochemistry

modeling using MODELER (version 9.15, BIOVIA). The best antibody model was selected among the models on the basis of their energy minimization scores calculated by Rosetta version 3.7. The following mutations were fully combined as double and triple substitutions: 28/30/31/32/33/35D in heavy chain 1 (CH1) and 24/45/49/50/51/52/53/56D in light chain (LC). The mutations were introduced into the homology model, and the relative difference in Gibbs free energies (ddG) between the mutant and wild type (WT) was calculated using Rosetta version 3.7 with an energy function31 optimized toward thermodynamic properties. Single and double/triple substitutions causing increases in ddG of more than 5 and 1.5 Rosetta energy units (REU), respectively, were eliminated. Amino acid numbering was based on Kabat numbering.32 Protein Expression. DNA sequences encoding WT antiprotein X LC and WT anti-protein X CH1 were designed on the basis of the DNA sequences of a full-length anti-protein X antibody raised in mice. DNA sequences containing the following mutations were ordered in the pBOK85_S250 vector as plasmid preparations (0.5−10 mg) from Thermo Fisher Scientific GENEART (Germany): [T28D], [A30D], [S31D], and [T28D;S31D] in HC and [K45D], [K50D], [S52D], [N53D], [S56D], [K50D;S56D], [S52D;S56D], [N53D;S56D], [K50D;N53D], and [K50D;S52D] in LC. The plasmid constructs encoding anti-protein X LC were cotransfected with plasmid constructs encoding anti-protein X HC into the Expi293 Expression System (Thermo Fischer Scientific) in a 1:1 ratio. This was done for the WT and mutant sequences. The transfections were made following the instructions of the supplier of the expression system in a 0.5 L expression scale. The transfected HEK293 EXPI cells were grown for 5 days at 8% CO2 and 36.5 °C in a shaker incubator. On day 5, supernatants were harvested by centrifugation and sterile filtered through a 0.22 μm poly(ether sulfone) filter system (EMD Millipore). Protein Purification. Harvested Fab variant supernatants from HEK293 cells were adjusted to pH 4.5−5.0 using 1 M hydrochloric acid, centrifuged (1000g for 20 min), and sterile filtered for removal of the precipitate. Fab variants were captured by multimodal chromatography using MEP HyperCel (26 mm × 5.7 cm, column volume of 30 mL) with a linear flow velocity of 34 cm/h (3 mL/min). Running buffers were equilibration buffer [54 mM sodium dihydrogen citrate and 45 mM disodium phosphate (pH 7.5)], wash buffer [54 mM sodium dihydrogen citrate, 45 mM disodium phosphate, and 1 M sodium chloride (pH 7.5)], and elution buffer [54 mM citric acid and 45 mM sodium dihydrogen phosphate (pH 3.0)]. Prior to loading, sodium chloride was added to the harvest materials to a final concentration of 1 M to promote hydrophobic interactions. The Fab variants were eluted from the multimodal resin by isocratic elution (90% elution buffer), and the collected elution fractions were pooled and adjusted to pH 4.5−5.0 using 0.5 M disodium phosphate (pH 9.0). Elution fractions that contained the precipitate were centrifuged (1000g for 30 min) and sterile filtered prior to pooling. Captured Fabs were purified by cation exchange (CIEX) chromatography using SP Sepharose HP (16 mm × 2.7 cm, column volume of 5.4 mL) with a linear flow velocity of 90 cm/min (3 mL/min). Running buffers were equilibration buffer [25 mM sodium acetate and 10 mM acetic acid (pH 4.5−5.0)] and elution buffer [25 mM sodium acetate, 10 mM acetic acid, and 1 M sodium chloride (pH 5.0)]. Prior to

heavy domain (VH) to reduce aggregation propensity. The key finding was that the surface net charge of the scaffold is the determinant of the charge type of the mutations in the CDRs. Thus, if the scaffold has a positive surface net charge, then positively charged mutations in the CDRs will increase aggregation resistance and vice versa.27 In another report, Perchiacca et al.28 found that aggregation resistance was not dependent on only net surface charge; highly specific charge mutations at positions 29 and 31−33 of VH conferred high aggregation resistance of human VH antibodies.28 Dudgeon et al.29 proposed a general strategy for the generation of human variable domains with increased aggregation resistance by aspartate (Asp) substitutions at specific positions clustering in CDR1 of heavy chain (CDR1-H) and CDR2 of light chain (CDR2-L). These positions were identified by phage display screening of which the library design was based on the most common germline families in the human repertoire (VH3 and Vκ1). High-affinity binding to antigen was retained by exclusion of mutations in CDR3-H, where the main antigenic contacts were located. The study showed a preference for Asp over other charged residues for the reduction of aggregation propensity, which was measured as the binding activity after heating to 80 °C.29 In this study, we investigated whether Asp substitutions at the specific positions reported by Dudgeon and co-workers could improve the developability of a murine Fab. The approach consisted of the following steps: (i) assembly of a full combinatorial in silico library of single, double, and triple Asp substitutions in the FRs and CDRs, (ii) a priori selection of mutations from the assembled library by in silico structural modeling in Rosetta for retention of thermodynamic stability, and (iii) production of selected Fab variants for developability assessment of antigenic affinity, thermodynamic stability, nonspecificity, and aggregation propensity. We report on the developability potential of Asp substitutions for decreasing nonspecific binding and aggregation propensity while retaining or improving antigenic binding affinity. We show that the number of variants for in vitro assessment can be considerably reduced by performing a structural prescreen based on computational models.



MATERIALS AND METHODS Materials. The Expi293 Expression System Kit was obtained from Life Technologies. The Corning Disposable Vacuum Filter (1 L scale, 0.22 μm), the CaptureSelect Biotin Anti-LC-κ (Murine) Conjugate, and materials for sodium dodecyl sulfate−polyacrylamide gel electrophoresis (SDS− PAGE) and IEF gel were obtained from Thermo Fisher Scientific. IEF Marker 3−10 (Liquid Mix) was obtained from Serva. Vivaspin 6 [molecular weight cutoff (MWCO) of 10 kDa] was obtained from Sartorius Stedim Biotech. MEP HyperCel and Eshmuno HCX Media were obtained from Pall and Merck, respectively. SP Sepharose HP, HiLoad 16/600 Superdex 200 pg, PD-10 desalting, and PD MiniTrap G-25 columns and the Biotin CAPture Kit (Series S) were obtained from GE Healthcare. Chemicals used for buffer preparation were purchased from Sigma-Aldrich unless otherwise stated. Structural Modeling. To generate a reliable homology model, the top 20 closest sequential homologue antibody structures were searched in the Protein Data Bank (PDB) for both the heavy and the light chain variable domain sequences. Sequence alignments were carried out using Muscle (version 3.8.31).30 These sequences served as inputs for homology 2751

DOI: 10.1021/acs.biochem.9b00251 Biochemistry 2019, 58, 2750−2759

Article

Biochemistry

the different Fab variants in duplicate at a flow rate of 30 μL/ min to both flow cells for 240 s. The dissociation was monitored for 600 s. The WT Fab was included for target protein X binding in all channels before and after each run as well as in one channel in each cycle throughout the run as a control. Running buffer was 50 mM Tris-buffered saline (TBS) at pH 7.4 containing 5 mM calcium chloride, 1 mg/mL bovine serum albumin (BSA), and 0.005% Tween 20. Conditioning and regeneration of the chip were performed according to the manufacturer’s instructions. Data obtained from association and dissociation of target protein X were subjected to double referencing and fitted to a 1:1 Langmuir binding model using the Biacore 8K Evaluation Software (GE Healthcare). Isoelectric Focusing. The isoelectric point (pI) of the Fab variants was determined by isoelectric focusing (IEF) using precast Novex pH 3−10 IEF Protein Gel (1.0 mm, 10 wells), Novex IEF Sample Buffer pH 3−10 (2×), Novex IEF Anode Buffer (50×), Novex IEF Cathode Buffer pH 3−10 (10×), the Colloidal Blue Staining Kit, IEF Marker 3−10, a fixation solution (12% trichloroacetic acid and 3.5% sulfosalicylic acid), and 96% ethanol according to the manufacturer’s instructions. Invitrogen PowerEase 500 from Thermo Fisher Scientific was used as a power source to run the gel. Two micrograms of protein was loaded into each well. Differential Scanning Fluorimetry. The thermal stability was determined by differential scanning fluorimetry using Prometheus NT.48 (NanoTemper Technologies). Ten microliters of each protein sample (0.5 mg/mL) was loaded into nanoDSF grade standard capillaries and then subjected to a temperature ramp from 20 to 95 °C with a heating rate of 1.5 °C/min. Unfolding was measured with tryptophan fluorescence using the ratio of fluorescence at 350 and 330 nm. The midpoint of the thermal unfolding reaction (Tm) was determined from the first-derivative spectrum by fitting the maximum using the software PR control (NanoTemper Technologies). Nonspecificity SE-UPLC* Analyses. The nonspecific interaction was investigated by SE-UPLC* analyses, defined here as analyses in the absence of organic solvents and high concentrations of salts. The column was an ACQUITY UPLC Protein BEH SEC (200 Å, 1.7 μm, 4.6 mm × 150 mm) column from Phenomenex, and the system was a Waters Acquity H Class UPLC instrument (Waters Corp.). The running buffer was 20 mM sodium phosphate (pH 6.8). The flow rate was 0.4 mL/min, and the temperature was 25 °C. The sample volume was 10 μL. SE-UPLC* retention was monitored with ultraviolet (UV) detection at 280 nm. Cross-Interaction Chromatography. Investigation of nonspecific interaction with the IgG pool purified from human serum (Sigma-Aldrich) was performed by crossinteraction chromatography (CIC). The CIC column was prepared as described by Wolf Pérez et al.33 and set up on a Waters Alliance HPLC system (Waters Corp.). Running buffer consisted of 20 mM sodium phosphate and 140 mM sodium chloride (pH 7.4). Five micrograms of protein was injected into the column. The flow rate was 0.1 mL/min, and the run time was 30 min with UV detection at 280 nm for monitoring of column retention. Storage Stability Study. Sixty-five microliters of the Fab sample (1 mg/mL) in triplicate in 96-well polymerase chain reaction (PCR) semiskirted plates (Thermo Fischer Scientific) was incubated at 45 °C using the Applied Biosystems Veriti Thermal Cycler PCR instrument (Thermo Fischer Scientific).

loading, the capture pools were dialyzed into equilibration buffer overnight at 4 °C using Slide-A-Lyzer G2 Dialysis Cassettes (10K MWCO, 70 mL, Thermo Fisher Scientific) to achieve a conductivity of