1. Introduction
Therapeutic monoclonal antibodies (mAb) are the leading class of drugs on the biopharmaceutical market, largely due to their high specificity, affinity, potency, and their long in vivo half-life. Since approval of the first mAb (OKT3) in 1983, 79 therapeutic antibodies have been approved by FDA and are currently on the market, including 30 mAbs for cancer treatment [
1]. Innovations in antibody engineering technologies—such as humanization of murine mAbs, phage display and transgenic mice for generating fully human mAbs, Fc engineering, antibody-drug conjugates, and bispecific antibodies—have contributed to the development of these important drugs [
2,
3,
4,
5,
6,
7,
8,
9]. Therapeutic antibody discovery and development starts with early candidates (hits), followed by selection of advanced candidates (leads). A critical element of antibody development is termed chemistry, manufacturing, and controls (CMC), which includes construction of an antibody-producing cell line, development of a manufacturing process, and development of suitable analytical methods to validate the antibody’s safety and efficacy [
10,
11,
12].
During the manufacturing process, including cell culture and downstream processing, therapeutic antibodies are at risk of physical and chemical degradation through multiple pathways [
13]. Degradation may affect antigen binding, decrease antibody efficacy, or even lead to immunogenic products [
14,
15,
16]. Protein aggregation is the most common and substantial type of physical degradation associated with therapeutic antibodies [
13,
17,
18]. High concentrations are associated with increased protein–protein interaction frequency, which proportionally increases the opportunity for aggregation formation. Changes in extrinsic conditions—including pH, salt, temperature, shaking, and viscosity—can also promote protein–protein associations that can lead to aggregation events [
13,
19,
20,
21,
22].
Post-translational modifications (PTMs) may also cause problems during therapeutic antibody development. For example, asparagine (Asn) deamidation, the most common pathway for the chemical degradation of therapeutic antibodies, results from hydrolysis of the amide side-chain of Asn, which cumulatively produces a heterogeneous mixture of aspartate (Asp) and isoAsp at the affected position [
23,
24]. Asn residues are more prone to deamidation when they are in a solvent-accessible region or are followed by a small or flexible residue, such as serine (Ser) or glycine (Gly). Asn deamidation can affect function if it occurs at a binding interface, such as the complementarity determining regions (CDRs) of an antibody molecule [
19,
23,
25]. Another PTM, Asp isomerization, involves the non-enzymatic interconversion of Asp and isoAsp residues [
19]. Asp isomerization may occur more commonly when Asp is followed by Ser, Gly, or Asp; it can affect protein function when it occurs in CDRs; and it can potentially result in fragmentation [
26,
27].
Each mAb has unique biophysical properties, mainly due to differences in the CDR residues and framework scaffolds. Thus, the identification of degradation-prone or unstable regions early in antibody development could allow for re-engineering of leads. This approach is aided by computational modeling tools that predict regions susceptible to physical and chemical degradation [
28,
29,
30]. To develop a therapeutic antibody with anti-tumor activity, we previously isolated a human mAb (Ab4) that specifically binds to human and rodent L1 cell adhesion molecule (L1CAM) from a human naïve Fab library using phage display [
31]. We next generated an affinity-matured version (Ab417) of this hit through site-directed mutagenesis of CDR residues, and we validated its anti-tumor efficacy and mechanism of action in rodent models [
31,
32].
In the present study, we aimed to improve the biophysical properties of the lead antibody (Ab417). We analyzed the heavy (VH) and light (VL) chain sequences and the three-dimensional (3D) model of Ab417, using computational methods to identify potential PTMs and to calculate aggregation propensities. Next, we designed 20 variants of Ab417 with reduced aggregation propensity, fewer potential PTM motifs, and the lowest predicted immunogenicity. We constructed these Ab417 variants and analyzed their expression levels and antigen-binding activities. One variant (Ab612), which was generated by substituting the four VH residues and two VL residues of Ab417, exhibited a higher expression level and higher antigen-binding activity compared to Ab417. Further studies demonstrated that compared to Ab417, Ab612 also showed higher productivity, improved biophysical properties, and a higher affinity and in vivo anti-tumor efficacy. For the preclinical development of Ab612, we generated a highly productive and stable research cell bank (RCB), and scaled up the production process to 50 L, which yielded 6.6 g/L of the antibody. Ab612 is now considered a candidate antibody to progress to preclinical development.
3. Discussion
The early phase of drug discovery is focused on antibody selection based on specificity, affinity, and functional properties. However, as an antibody is advanced into preclinical development, it is important to consider its biophysical properties to determine whether it can be successfully developed into an efficacious drug. The biophysical properties of a therapeutic antibody can critically influence its late-stage developability, which requires high-level expression, high solubility, conformational and colloidal stability, low poly-specificity, and low immunogenicity [
33].
In our previous attempt to develop a therapeutic antibody with anti-tumor activity, we isolated a human mAb that cross-reacts with rodent L1CAM from a human naïve antibody library, and we then generated an affinity-matured version (Ab417) of this hit and validated its anti-tumor efficacy and mechanism of action in rodent models [
31,
32]. In the present study, we attempted to optimize Ab417 by improving its biophysical properties.
We designed 19 variants with reduced aggregation propensity, fewer potential PTM motifs, and lower predicted immunogenicity using computational methods. Subsequently, we constructed these variants to analyze their expression levels and antigen-binding activities. The Ab417 variant H3L7 exhibited a higher expression level compared to Ab417.
Therefore, we further changed a residue in the LCDR3 of H3V7 to generate an even more improved variant (Ab612) with higher productivity and antigen-binding affinity. The Ab612 showed 2.6-fold higher productivity and improved biophysical properties, such as 1.4-fold increased purification yield, greater stability, lower aggregation propensity, 2-fold higher affinity for human L1CAM, and enhanced in vivo anti-tumor efficacy. Moreover, for the preclinical development of Ab612, we successfully generated a highly productive and stable research cell bank (RCB) and confirmed the scalability of the production process to a pilot scale.
Overall, the present results demonstrate that we successfully improved the biophysical properties and affinity of Ab417, generating an optimized antibody (Ab612). Ab612 is considered a promising candidate antibody for preclinical development.
4. Materials and Methods
4.1. Design of Ab417 Variants with Improved Biophysical Properties
4.1.1. Sequence Analysis for Potential PTMs
Multiple alignments of Ab4 and Ab417 sequences to human germline sequences were generated using MAFFT [
34], and entries in each alignment were ordered according to the sequence identity to the parental sequence. The antibody sequences were analyzed for potential PTMs such as Asn deamidation, Asp isomerization, N- and O-glycosylation, and oxidation.
4.1.2. Construction and Comparison of 3D Models of Ab417 and Its Variants
Structural models of the Fv-region for Ab417, and variants thereof, were generated using Lonza Biologics’ modeling platform. Candidate structural template fragments for the FR and CDRs, as well as the full Fv, were scored, ranked, and selected from the antibody database based on their sequence identity to the target, as well as qualitative crystallographic measures (Å) of the template structure. In order to structurally align the CDRs to the FR templates, five residues on either side of the CDR were included in the CDR template. An alignment of the fragments was generated based on overlapping segments and a structural sequence alignment using MODELLER. An ensemble of structures that satisfy the conformational restraints derived from the set of aligned structural templates was created by simulated annealing and conjugation gradient optimization procedures. One or more model structures were selected from this ensemble based on an energy score derived from the quality of the protein structure and satisfaction of the conformational restraints. The models were inspected, and the side chains of the positions which differ between the target and template were optimized using a side chain optimization algorithm and energy minimized. A suite of visualization and computational tools were used to assess the conformational variability of the CDRs, as well as the core and local packing of the domains and regions and surface analysis to select one or more preferred models.
To assess the impact of different substitutions on affinity and stability, a number of structural criteria, including the solvent accessibility, local atomic packing, and location of the substitution relative to the predicted antigen-binding interface or the Fv dimer interface, electrostatic effects, and hydrogen bonding patterns, were used.
4.1.3. Calculation of Aggregation Propensity and Assessment of Potential Substitutions
Aggregation hotspots were identified based on the sequence and structure of antibodies using Lonza’s AggreSolve
TM in silico tools. The intrinsic aggregation propensity score, A
res, was calculated for overlapping 7-mer peptides, and the score was calculated over an entire amino acid sequence to generate A
tot. In addition, S
res, which reflects the aggregation propensity of a 7-mer peptide from its folded state, was calculated by applying the conformational correction to the intrinsic aggregation propensity profile [
35]. A summary score, S
tot, was calculated based on the position-specific S
res. In addition, given that non-specific protein-protein interactions can be caused by aggregation-prone hotspots on the protein’s surface, the surface aggregation propensity per position, Tres, was calculated. A summary score, T
tot, was calculated based on the position-specific Tres descriptors.
All positions outside the CDRs that were part of the hot spots were assessed based on their potential impact on binding affinity and stability. Each position was classified as either Neutral, Critical, or Contributing. A neutral position means that substituting another amino acid at this position should not affect binding affinity or stability negatively. A contributing position means that a substitution can be made, but the position may contribute to binding affinity or stability. A critical position means that the position risks a decreased binding affinity or stability, and therefore parental amino acid must be retained.
4.1.4. Analysis of Th Epitopes
The epitopes or clusters of adjoining epitopes of Ab417 and engineered variants were analyzed using Epibase
TM for substitutions that would remove or reduce binding to HLA allotypes to the greatest extent possible, with a focus on the HLA-DR allotypes, because these are known to express at a higher level than the other allotypes DQ and DP [
36]. Human germline sequences were not considered to be immunogenic as they are found in the pool of circulating antibodies. The substitutions in the engineered antibodies were analyzed for all chosen HLA allotypes (DRB1, DR3/4/5, DQ, and DP).
4.2. Construction of Expression Plasmids and Transient Expression of Antibodies
The genes coding for the VH or VL of the variants were synthesized and subcloned into the EcoRI-ApaI or HindIII-BsiWI sites of the expression plasmid (pCMV-dhfrC) containing the human Cγ1 or Cκ gene to construct heavy or light chain expression plasmid, respectively, as described previously [
31].
For transient expression of antibodies, HEK293F cells (Gibco, Thermo Fisher Scientific, Waltham, MA, USA) or ExpiCHOTM cells (Thermo Fisher Scientific, Waltham, MA, USA) were grown in FreeStyle 293 Expression medium (Gibco, Thermo Fisher Scientific, Waltham, MA, USA) at 8% CO2, 37 °C, 125 rpm or in ExpiCHO Expression medium (Thermo Fisher Scientific, Waltham, MA, USA) at 5% CO2, 37 °C, 110 rpm on a shaker. For transfection, the heavy and light chain expression plasmids (two vector system) or expression plasmid containing heavy and light chain (one vector system) were introduced into HEK293F using polyethyleneimine (Polysciences) or ExpiCHO™ cells (Thermo Fisher Scientific, Waltham, MA, USA) using the ExpiCHO Expression System. On day 6 post-transfection, the cell culture supernatant was centrifuged, filtered using a bottle top filter (0.22 µm PES, Sartorius, Germany), and subjected to quantitative ELISA and indirect ELISA to analyze the expression level and antigen binding activity.
4.3. ELISAs
A quantitative ELISA was performed to determine the concentration of antibodies. Anti-human IgG kappa antibody (Invitrogen, Waltham, MA, USA) was coated on the 96-well immunoplate (Thermo Fisher Scientific, Waltham, MA, USA) at 4 °C overnight. After blocking with 2% skim milk in 0.05% PBST (PBS with 0.05% tween 20), a standard IgG sample (400 ng/mL) or cell culture supernatant serially diluted in 0.1% PBA was added to each well and incubated at 37 °C for 1 h. The bound antibody was detected using anti-human IgG Fc-HRP (1:10,000 (
v/
v); Invitrogen, Waltham, MA), as described previously [
31].
An indirect ELISA was performed to determine the antigen binding activity of the antibody. Purified human recombinant L1CAM (hL1-s1) and mouse recombinant L1CAM (mL1-s1) were prepared as described previously [
31]. hL1-s1 or mL1-s1 (100 ng/well) was coated on each well at 4 °C overnight. Serially diluted cell culture supernatant or antibody was incubated with the hL1-s1 or mL1-S1 at 37 °C for 1 h. The bound antibody was detected using anti-human IgG(Fc-specific)-HRP (1:10,000, Invitrogen, Waltham, MA, USA).
A competitive ELISA was performed to determine the antigen-binding affinity of the antibody. hL1-s1 or mL1-s1 (100 ng/well) were coated on each well at 4 °C for 12 h. Ab417 or Ab612 antibody (10 ng/mL) in 0.1% PBA solution was pre-incubated with various concentrations (10−12~10−7 M) of hL1-s1 or mL1-s1 as a competing antigen at 37 °C for 3 h. The reaction mixture was added to each well coated with the hL1-s1 or mL1-s1, and indirect ELISA was carried out. Affinity (KD) was defined as the antigen concentration required to inhibit 50% of the antigen-binding activity.
A sandwich ELISA was performed to compare the antigen-binding activities of the Ab612 antibody samples produced from the transient expression and the RCB culture. Anti-human Fc antibody (Invitrogen, Waltham, MA, USA, 100 ng/well) was coated on each well at 4 °C overnight. Ab417 or Ab612 (50 ng/mL) was added to each well, incubated at 37 °C for 1 h, and further incubated with hL1-s1 serially diluted from 4 μg/mL at 37 °C for 1 h. The hL1-s1 captured by Ab417 or Ab612 was incubated with mouse anti-s1 antibody KR127 [
37] at 37 °C for 1 h. The bound KR127 antibody was detected using anti-mouse IgG(Fc-specific)-HRP (1:10,000, Invitrogen, Waltham, MA, USA).
4.4. Purification of Antibodies
Ab417 and Ab612 were purified using a Protein A affinity column. Harvested Cell Culture Fluid (HCCF) containing a monoclonal antibody was loaded on to pre-packed protein A column (HiTrap MabSelectTM Sure, Cytiva, Marlborough, MA, USA) equilibrated with binding buffer (20 mM sodium citrate, pH 6.0). The column was then re-equilibrated with binding buffer followed by a wash step at pH 6.0 and finally elution with sodium citrate buffer at pH 6.0 and 2.5 (gradient elution). The gradient elution experiments were carried out using a 30 column volume (CV) linear gradient from 0% to 100% buffer B (20 mM sodium citrate, pH 6.0). For virus inactivation, the elution fraction was incubated at 4 °C for 1 h and then neutralized with 1 M Tris-HCl (pH 9.0) to minimize the effect on the structure of the antibody under low pH conditions. The column was then regenerated and sanitized using 0.5 N NaOH.
Desalting chromatography (HiPrep™ 26/10 Desalting, Cytiva, Marlborough, MA, USA) was performed to replace the buffer in the primary purified product (affinity eluate) suitable for the loading condition (50 mM Na-Acetate, pH 5.0). Further purification of Ab417 and Ab612 was performed using cation-exchange chromatography. The sample was applied to a HiTrap Capto SP column (Cytiva, Marlborough, MA, USA) and eluted with a linear gradient of 0 to 500 mM NaCl, 30 CV.
Pooled fractions were further concentrated using Amicon Ultra-4 Centrifugal Filter Unit with Ultracel-10 membrane (Merck Millipore, Burlington, MA, USA). A concentrated protein sample was loaded onto the HiLoad Superdex 200 pg column (Cytiva, Marlborough, MA, USA) equilibrated with Phosphate-buffered saline (PBS) buffer at the rate of 1.0 mL/min. The elution profile was analyzed by the absorbance at 280 nm.
The purity and aggregation of purified protein were determined using high-performance size exclusion chromatography (SE-HPLC). A Waters HPLC (Alliance 2695) system was used with a Bio SEC-3 column (3 μm, 300 Å, 4.6 × 300 mm, Agilent, Santa Clara, CA, USA) at 0.3 mL/min flow rate (isocratic) using a mobile phase buffer of 20 mM Na-phosphate w/150 mM NaCl, pH 6.8.
4.5. Dynamic Light Scattering (DLS)
DLS was performed using Zetasizer (Malvern, Herrenberg, Germany). After the Z-average of antibody sample was measured at 25 °C, 50 μL of the sample was added to disposable cuvettes (ZEN0040, Malvern, Herrenberg, Germany) and gradually the temperature was increased from 30 °C to 80 °C with 3 °C of temperature interval and a fixed angle of θ = 173°. The Z-average and intensity were calculated using Zetasizer Software version 7.02 (Malvern, Herrenberg, Germany).
4.6. Affinity Determination of Antibodies Using Octet red384 System
The affinity of the antibody was determined using the Octet Red384 system (Sartorius, Goettingen, Germany). Anti-human IgG sensor AHC (ForteBio, Fremont, CA, USA) was firstly soaked in 0.1% PBA for 20 min. The antibody (0.2 mL of 0.5 µg/mL) was captured for 10 min followed by washing with 0.1% PBA for 2 min. hL1-s1 or mL-s1 (25, 12.5, 6.25, 3.125, or 1.5625 nM in 0.1% PBA) was then incubated with the antibody captured on the sensor. Association and dissociation rates were measured for 10 min and 30 min, respectively. For correction of baseline drift, a control sensor was designated as an antibody-captured AHC sensor exposed to running buffer only. All analytes were recalculated by subtraction of the rate of a control sensor. The operating temperature was maintained at 30 °C and agitated at 1000 rpm. Data were analyzed using a 1:1 interaction model (fitting global, Rmax unlinked by a sensor) with analysis software (ForteBio, ver. 8. 2).
4.7. Capillary Isoelectric Focusing (cIEF) Analysis
cIEF was conducted according to the SCIEX application protocol. All cIEF experiments were performed on SCIEX PA800 plus instrument with a 50 µm i.d. neutral coated capillary (SCIEX P/N 477441) at the length of 30.2 cm. The UV detector was used to detect absorbance at 280 nm wavelength. A cIEF master mix solution was composed of 3 M urea-cIEF gel solution, pharmalyte 3–10 carrier ampholytes, cathodic stabilizer (500 mM arginine), anodic stabilizer (200 mM iminodiacetic acid), and five pI markers (10.0, 9.5, 7.0, 5.5, 4.1). Analytes (5 mg/mL) were also mixed with 10 µL of the master mix solution. All experiments were performed in triplicate. The pI values of the sample were calculated using qualitative analysis of a 32 Karat software.
4.8. In Vivo Antitumor Activities of Ab417 or Ab612
All the animals were housed under a 12/12 h light/dark cycle (light phase, 8:00 A.M. to 8:00 P.M.) with a standard laboratory diet and water ad libitum. All animal handling and experiments were conducted with the approval of the Institutional Animal Care and Use Committee (IACUC) of preclinical CRO Biotoxtech (180678). Nude mice (BALB/cSlc-nu, 5 weeks old) were obtained from Japan SLC, Inc (Shizuoka, Japan). Choi-CK cells (1 × 106) were inoculated into the right flank of each mouse. Constructed Choi-CK tumor tissue (3 × 3 × 3 mm3) was subcutaneously inoculated into the back of mice. After tumor volume reached a mean of 100 mm3 (n = 8 per group), the antibody at a dose of 10 mg/kg was i.v. injected 3 times per week for 3 weeks. Tumor growth was monitored by measuring the length and width of the tumor with a caliper and calculating tumor volume based on the following formula; TV (mm3) = L (mm) × W2 (mm2) × 1/2, where L is length and W is width. Body weight was measured twice a week, and tumor tissues were taken out and weighed at the end of the experiment. Tumor growth inhibition rate (IR) was calculated as the following formula; IR (%) = (1 − T/C) × 100. T is the mean tumor weight of the antibody treated group, and C is the mean tumor weight of the mock control group.
Data were validated using SAS (Version 9.3, SAS Institute Inc., Cary, NC, USA). Each point represents the mean ± SD. Statistical comparison between groups were performed by one-way analysis of variance (ANOVA) followed by Dunnett’s t-test.
4.9. Production of Ab612 from Research Cell Bank (RCB)
4.9.1. Construction of RCB
RCB was generated by Shanghai OPM Biosciences Co., Ltd. (Shanghai, China). CHO-K1 cells were grown in CD CHO Medium (GIBCO, Thermo Fisher Scientific, Waltham, MA, USA) containing 6 mM L-glutamine (Sigma, St. Louis, MO, USA). Cells were incubated under the condition of 120 rpm, 37 °C, and 8% CO2. Freshly prepared linearized plasmid was transfected into CHO-K1 cells by electroporation using a Bio-Rad system. For each sample, 1 × 107 cells were transfected with a total of 40 μg of the linear plasmid. At 24 h later, the transfected cells were plated into a 96-well plate with 4000 cells/well. The culture medium was CD CHO Medium containing 50 μM L-Methionine sulfoximine (MSX) (Sigma, St. Louis, MO, USA) and 1 × GS-Supplement (Sigma, St. Louis, MO, USA). Cells were statically cultured in an incubator with 8% CO2 and 37 °C. The plasmid was transfected into cells twice by electroporation, and then the cells were plated into 20 pieces of a 96-well plate. Mini pool fed-batch assay medium was OPM-CHO CD07 Medium (OPM, China) with 1:200 anti-clumping Agent (ACA, OPM, China). The CDF18 (OPM, China) at 3%, 5%, 6%, 6% and 5% concentration and CDF26 (OPM, China) at 0.3%, 0.5%, 0.6%, 0.6% and 0.5% were fed on day 3, 5, 7, 9, and 11, respectively. The glucose was fed as needed to maintain at 2–6 g/L. The fed-batch was stopped when the viability is about 60% or on day 12.
4.9.2. Stability Study of Research Cell Bank (RCB)
In suspension culture from RCB stock, cells were passaged once every 3 days for 69 days with 0.4 × 106 cells/mL seeding density in OPM-CHO CD07 Medium with 50 μM MSX and 1:200 Anti-Clumping Agent. In a fed-batch assay, cells were incubated at the density of 0.8 ± 0.1 × 106 cells/mL with 30 mL culture volume. The medium was OPM-CHO CD07 Medium with 1:200 ACA. The CDF18 was fed on day 3, 5, 7, 9 and 11 at 3%, 5%, 6%, 6% and 5%, respectively, and CDF26 was fed at 0.3%, 0.5%, 0.6%, 0.6% and 0.5%. The glucose was fed as needed to maintain it at 2–8 g/L. The fed-batch was stopped when the viability is about 60% or on 12 days. The culture supernatants were harvested by centrifuging at 3500 rpm for 30 min under 4 °C.
4.9.3. Production and Purification of Antibodies
The HCCF obtained from RCB production was purified using protein A (Repligen, Waltham, MA, USA) packed in XK26/20 column (Cytiva, Marlborough, MA, USA) affinity chromatography. The purified antibody was stored in 10 mM sodium phosphate buffer containing 5% sorbitol and 0.01% tween 20 and dialyzed using HiPrepTM 26/10 Desalting Column (Cytiva, Marlborough, MA, USA). The concentration of purified antibody was determined with a Nanodrop (Thermo fisher scientific, Nanodrop 2000) based on the molar extinction coefficient. The produced Ab612 from RCB was used for analyzing purification by SEC-HPLC and antigen-binding activity by ELISA.