Efficient production of bispecific antibody by FAST-Ig^TM and its application to NXT007 for the treatment of hemophilia A

ABSTRACT

Efficient production of bispecific antibodies (BsAbs) in single mammalian cells is essential for basic research and industrial manufacturing. However, preventing unwanted pairing of heavy chains (HCs) and light chains (LCs) is a challenging task. To address this, we created an engineering technology for preferential cognate HC/LC and HC/HC paring called FAST-Ig (Four-chain Assembly by electrostatic Steering Technology – Immunoglobulin), and applied it to NXT007, a BsAb for the treatment of hemophilia A. We introduced charged amino-acid substitutions at the HC/LC interface to facilitate the proper assembly for manufacturing a standard IgG-type BsAb. We generated CH1/CL interface-engineered antibody variants that achieved > 95% correct HC/LC pairing efficiency with favorable pharmacological properties and developability. Among these, we selected a design (C3) that allowed us to separate the mis-paired species with an unintended pharmacological profile using ion-exchange chromatography. Crystal structure analysis demonstrated that the C3 design did not affect the overall structure of both Fabs. To determine the final design for HCs-heterodimerization, we compared the stability of charge-based and knobs into hole-based Fc formats in acidic conditions and selected the more stable charge-based format. FAST-Ig was also applicable to stable CHO cell lines for industrial production and demonstrated robust chain pairing with different subclasses of parent BsAbs. Thus, it can be applied to a wide variety of BsAbs both preclinically and clinically.

KEYWORDS:

• ART-S3
• bispecific antibody
• CHO cell line
• emicizumab
• FAST-Ig
• Hemophilia A
• KiH
• NXT007
• Orthogonal Fab
• single-cell production

Introduction

Bispecific antibodies (BsAbs) bind to two different antigens or epitopes and can have unique pharmacological effects that cannot be achieved with conventional monospecific antibodies.¹ For example, two marketed antibodies with proven clinical efficacy, blinatumomab, with T-cell redirecting function,² and emicizumab, with factor VIIIa (FVIIIa) cofactor mimetic activity,^3–5 only work because of their bispecific binding ability. In addition, the clinical efficacy of the simultaneous inhibition of two antigens by a single molecule has recently been demonstrated with amivantamab⁶ and faricimab.^7,8 Buoyed by the approval of these BsAbs by the Food and Drug Administration (FDA) and other regulatory agencies, more than 110 BsAbs have recently entered clinical trials.^9,10

The molecular formats of BsAbs can be broadly classified into standard IgG type and non-IgG type. Non-IgG type BsAb includes formats such as DVD-Ig, CODV-Ig, BiTE, and DART, which are composed of binding blocks such as single-chain variable fragments (scFv) and antigen-binding fragments (Fab).¹¹ In general, the standard IgG type is better for industrial manufacturing and clinical applications due to its superior physicochemical properties.^12–14 Also, the standard IgG type is preferable from the viewpoint of immunogenicity¹⁵ and pharmacokinetics.¹⁶

Two main methods for the efficient expression of IgG-type BsAbs have been extensively studied: the separate expression of two binding arms^17–19 or single-cell expression.^20–23 Controlled Fab arm exchange (cFAE)¹⁷ is a typical separate expression system, but it requires the construction of multiple cell lines and the removal of reducing agents, making the manufacturing process more complex. In contrast, single-cell expression requires only one cell line, making it simpler to manufacture. However, when a BsAb is conventionally expressed in a single cell, two types of heavy chains (HCs) and two types of light chains (LCs) are randomly assembled, resulting in nine types of mispairs other than the correctly assembled BsAb, which makes the subsequent purification process difficult and complicated.

Since the invention of knobs into holes (KiH) technology in 1996,²⁴ a series of similar methods have been developed by utilizing domain exchange^25,26 and electrostatic complementarity (ART-Ig,^5,27–29 DDKK,³⁰ DEKK³¹, solving most of the problems of HCs-heterodimerization. On the other hand, even under the controlled heterodimerization of HCs, the assembly of the two LCs still resulted in three types of mispairs. This issue was successfully overcome by developing a common LC format, making the two LCs identical.^5,23,28,29 Emicizumab is a BsAb that recognizes coagulation factor IXa (FIXa) and factor X (FX) and restores hemostatic activity in hemophilia A patients by exerting FVIIIa cofactor mimetic activity. It was developed using the common LC format and became the first standard IgG-type BsAb approved by FDA in 2017.^4,5,32 Although a BsAb in the common LC format is much easier to manufacture, the process of converting the lead antibody to the common LC format is complicated, and for some antibodies, it may be difficult to retain initial binding activity.^5,23,29,33 Also, because of limitations on engineering space, it is difficult to obtain compatible physiochemical properties and pharmacological activity with this format. It is especially difficult to optimize therapeutic BsAbs such as emicizumab, in which a subtle change in the affinity of both binding arms has a large impact on its activity.⁵ It is easier to improve activity by using a BsAb format with four different chains, which can be more precisely optimized than the common LC format.

Methods for efficiently expressing a four-chain BsAb by facilitating the association of HCs and LCs have been actively studied in recent years. There are three main approaches: CrossMab,³⁴ DuetMab,³⁵ and Orthogonal Fab.^20–22 CrossMab enforces the correct HC and LC (HC/LC) association through immunoglobulin domain crossover within one arm of BsAb.³⁴ DuetMab utilizes an artificially placed CH1/CL (constant heavy chain domain 1/light chain constant region) inter-chain disulfide bond.³⁵ Orthogonal Fab uses electrostatic-steering^20–22 or steric hindrance for HC/LC pairing.²¹ Although these methods have been reported to facilitate cognate HC/LC pairing in some antibodies, there are still few practical examples of antibodies that demonstrate versatility, feasibility of manufacturing, and clinical immunogenicity. Therefore, it is very important to develop other options for producing BsAbs and to demonstrate their viability as clinical candidates.

Here we report a novel engineering technology, FAST-Ig (Four-chain Assembly by electrostatic Steering Technology – Immunoglobulin), which efficiently expresses BsAbs in single cells by installing charged pair mutations at the interfaces of HC/LC as an orthogonal Fab design, combined with KiH or charge-based mutations to promote HCs-heterodimerization (Figure 1a,b). In addition, we describe how this technology was applied to NXT007, an anti-FIXa/FX bispecific humanized IgG₄ antibody with kappa and lambda LCs that has better coagulation activity than emicizumab.³⁸ We thoroughly assessed the optimal FAST-Ig design for NXT007, not only for preferential cognate HC/LC and HC/HC pairing, but also for pharmacological activity, physicochemical properties, immunogenicity, pharmacokinetics, and structure. Its applicability in a stable line of Chinese hamster ovary (CHO) cells, as well as its transient expression in Expi293F cells, were examined. Our results suggest that FAST-Ig technology will be useful for the preclinical screening and commercial production of BsAbs in drug discovery.

Figure 1. Schematic diagram of FAST-Ig concept (a) for correct pairing of HC/LC, residues in the interface of HC/LC are substituted by oppositely charged paired residues. For promoting HCs-heterodimerization, KiH or pairs of charged residues such as E356K/K439E (ART-S3), which was used in the previously reported ART-Ig design,^5,28,29 are introduced (numbering of CH was defined by Eu numbering.³⁶ VH, VL and CL were defined by Kabat numbering³⁷ in this paper). (b) Ten possible pairing patterns of BsAbs with four different chains in a single-cell expression of wild type IgG. Preferential chain pairings of BsAb with FAST-Ig mutations along with the image of the result of ion exchange chromatography are shown. The expression of nine types of mispairs will be suppressed by FAST-Ig. The antibody with desired pairing is highlighted by red square.

Results

Identification of residues to promote correct HC/LC association

In order to design a novel orthogonal Fab with preferential cognate HC/LC pairing, the crystal structure of human anti-VEGF Fab (PDB: 1CZ8,³⁹ IgG₁, kappa) was referenced as a representative model. The charge-charge interactions of amino acids can work even over relatively long distances (5–10 Å). Depending on whether the charges are the same or opposite, pairs repel or attract each other, respectively.⁴⁰ In order to utilize these electrostatic interactions for orthogonal Fab design, we then identified the amino-acid pairs within a distance of about 5–10 Å; CH1_K147 and CL_S131, Q160, T180; CH1_Q175 and CL_S131, Q160, T180; CH1_K213 and CL_E123 (Fig. S1A, S1B, S1C, S1E). In addition to those CH1 and CL interface designs, to further promote orthogonal Fab pairing, VH_Q39 and VL_Q38 at the VH and VL interface, which have been previously reported to be effective,^{20–22,27,41} were selected as candidate amino acid residues for the FAST-Ig design (Fig. S1D). These residues were conserved in amino acid sequences and Fab structures among the IgG subclasses (Fig. S1A, S1C-N). In CL of lambda, residues in CL_131, 160, 180 differed from those of kappa (Fig. S1B), but their positions in Fab structure were conserved (PDB: 8FAB⁴²) (Fig. S1F-H). Therefore, we thought these selected residues could facilitate the cognate association of HC/LC in NXT007, a human IgG₄-based BsAb with kappa and lambda LCs.

Correct HC/LC pairing by single pair of charged mutations at CH1/CL interface

To compare the pairing potency, defined here as ability to promote correct HC/LC pairing in a BsAb, of a selected pair of residues, we substituted them with lysine (Lys) as the positive charge and glutamic acid (Glu) as the negative charge. To suppress the expression of HC-homodimerized species and to make the analysis and comparison as simple as possible, we introduced the disulfide-linked KiH mutations (dKiH_v14),²³ knobs (Y349C, T366W) in the HC of anti-FIXa antibody and holes (E356C, T366S, L368A, Y407V) in the HC of anti-FX antibody. By introducing dKiH_v14 mutations, the major expressed components can be limited to the correctly assembled BsAb (H1L1H2L2: anti-FIXa composed of H1[HC] and L1[LC, kappa], and anti-FX composed of H2[HC] and L2[LC, lambda]) plus three LCs-scrambled mispairs (H1L2H2L2, H1L2H2L1, H1L1H2L1) (Fig. S2A). Therefore, pairing potency can be practically compared by measuring the percentage of the target BsAb transiently expressed in Expi293F cells transfected with four plasmids, each encoding one HC or one LC.

The percentage of the target BsAb produced from mammalian single cells are affected by the expression ratio of each HC/LC.²⁰ To avoid misjudging the potency of the mutations due to the differences in expression level possibly caused by the mutations, transient expression was performed under three conditions, with a plasmid mass ratio of H1:L1:H2:L2 = 1:3:1:1, 1:1:1:1, and 1:1:1:3. Antibodies expressed in each condition were purified by protein A and analyzed by cation-exchange chromatography (CIEX). When the parent antibody of NXT007, without any mutations at the HC/LC interfaces, was analyzed, four separate peaks were identified. Each elution time was matched with those of the separately prepared reference samples of the mispaired IgG antibody (Fig. S2B). Under these experimental conditions, the maximum area ratio of the target BsAb peak was 61.6 ± 1.4% (±SD) (Table 1). The expression ratio of the LC exchanger was very low in the parent of NXT007 even without FAST-Ig mutations (Figure S2B), and is expected to be further reduced with the FAST-Ig mutations. Hence, we considered it feasible to sufficiently compare the pairing potency of the mutations by calculating the area ratio of the correctly assembled BsAb and two types of common LC mispairs (H1L2H2L2, H1L1H2L1), which were verified with each reference sample. The data for the selected CH1 and CL mutations and the BsAb yield, defined here as the calculated percentage of correctly assembled BsAb, are shown in Table 1. All pairs of selected residues showed an improved BsAb yield relative to the parent, although the degree of correct assembly differed depending on the orientation of charge, even in the same positions. Based on the highest BsAb yield at each set of the substitution positions, the pairing potency of each residue pair was in the order of HC147-LC131: 91.7% (S1) > HC175-LC131: 86.3% (S7) > HC147-LC180: 85.0% (S6) > HC175-LC160: 77.8% (S9) > HC147-LC160: 75.6% (S3) > HC175-LC180: 75.4% (S11) > HC213-LC123: 69.9% (S14). When the mutation pair between VH and VL was also examined in the same way, VH39-VL38 (V1) showed an improved 91.8% BsAb yield (Table S1). However, none of the single pairs were able to consistently achieve a high BsAb yield of >95% under a range of expression conditions.

Table 1. CH1 and CL variants with single pair of charged residues to promote the correct assembly of HCs and LCs of parent NXT007 in a single host cell. The four plasmids encoding the HCs and LCs of anti-FIXa (IgG₄, kappa) and anti-FX (IgG₄, lambda) antibodies were co-transfected into Expi293F cells with three plasmid weight ratios, H1:L1:H2:L2 = 1:3:1:1, 1:1:1:1, 1:1:1:3. Then, the cell culture supernatants were purified by protein A affinity chromatography and analyzed by cation exchange chromatography. The target BsAb yield was calculated by peak area ratios of chromatogram as described in Figure S2. To promote heterodimerization of HCs, knob mutations (Y349C, T366W) and hole mutations (E356C, T366S, L368A, Y407V) were introduced in anti-FIXa HC and anti-FX HC, respectively. Data shown represent mean ± SD for two or more independent experiments.

Variant	Anti-FIXa HC (H1)	Anti-FIXa LC (L1)	Anti-FX HC (H2)	Anti-FX LC (L2)	BsAb yield (%) Plasmid ratio (H1:L1:H2:L2)
	CH1	CL	CH1	CL	1:3:1:1	1:1:1:1	1:1:1:3
Parent					61.6 ± 1.4	58.5 ± 1.6	19.2 ± 1.5
S1	K147	S131E	K147E	T131K	89.7 ± 0.3	91.7 ± 0.2	77.4 ± 0.3
S2	K147E	S131K	K147	T131E	87.0 ± 2.4	76.1 ± 1.0	34.4 ± 1.4
S3	K147	Q160E	K147E	E160K	75.6 ± 1.3	73.6 ± 0.4	31.8 ± 17.3
S4	K147E	Q160K	K147	E160	66.4 ± 8.0	64.2 ± 1.4	16.6 ± 1.9
S5	K147	T180E	K147E	S180K	80.9 ± 0.6	78.5 ± 1.6	45.6 ± 1.8
S6	K147E	T180K	K147	S180E	85.0 ± 1.6	84.4 ± 1.3	35.7 ± 0.3
S7	Q175K	S131E	Q175E	T131K	65.2 ± 2.8	82.3 ± 2.4	86.3 ± 0.9
S8	Q175E	S131K	Q175K	T131E	66.1 ± 14.4	48.9 ± 1.9	3.1 ± 0.3
S9	Q175K	Q160E	Q175E	E160K	75.1 ± 1.4	77.8 ± 2.6	26.6 ± 0.4
S10	Q175E	Q160K	Q175K	E160	53.7 ± 15	62.1 ± 11.7	7.9 ± 0.4
S11	Q175K	T180E	Q175E	S180K	71.2 ± 0.5	75.4 ± 0.7	54.6 ± 2.1
S12	Q175E	T180K	Q175K	S180E	70.4 ± 1.1	64.2 ± 0.1	7.8 ± 2.0
S13	K213	E123	K213E	E123K	69.6 ± 1.1	60.9 ± 2.3	18.5 ± 2.7
S14	K213E	E123K	K213	E123	69.9 ± 0.8	69.9 ± 0.0	29.1 ± 0.8

Correct HC/LC pairing by multiple charged mutations at HC/LC interface

To obtain a higher BsAb yield, we next investigated the combination of each mutation with confirmed preferential HC/LC pairing. We examined the combinations of effective mutation pairs in the VH/VL region (V1 or V2) with the best pair in the CH1/CL region (S1). As a result, both combinations improved the BsAb yield, especially V1+S1 offering a 98.4% BsAb yield (Table S1). The influence of the mutations in the variable region on antigen binding was then evaluated by surface plasmon resonance (SPR) assay. The binding ability to FIXa was hard to compare in detail due to the poor fitting curve in the experiment (Fig. S3), but our findings suggest that the binding ability to FX was slightly weakened by the V1 or V2 mutation pair, as evidenced by an average increase of 1.3–1.4 times in K_D values (Table S2). Despite these small changes in binding ability not posing a problem for conventional antibody drugs, they may affect bispecific antibodies with enzymatic activity. Therefore, we considered it better to use mutations only in CH1/CL to achieve strong pairing potency for NXT007. Thus, we comprehensively examined combinations of mutation pairs in CH1/CL whose pairing potency was confirmed (Table 2). The result showed that the BsAb yield was higher in all combination pairs than in single pairs, indicating that the combinations worked in an additive manner. In particular, the pairs of two mutations (C1 and C3), three mutations (C19 and C20), and four mutations (C21 and C22) showed a high BsAb yield of over 97.5% under at least one of the expression conditions.

Table 2. CH1 and CL variants with multiple pairs of charged residues to promote the correct assembly of HCs and LCs of parent NXT007 in a single host cell. The four plasmids encoding the HCs and LCs of anti-FIXa (IgG₄, kappa) and anti-FX (IgG₄, lambda) antibodies were co-transfected into Expi293F cells with three plasmid weight ratios, H1:L1:H2:L2 = 1:3:1:1, 1:1:1:1, 1:1:1:3. Then, the cell culture supernatants were purified by protein A affinity chromatography and analyzed by cation exchange chromatography. The target BsAb yield was calculated by peak area ratios of cation exchange chromatography as described in Figure S2 and Table 1 legends. Data shown represent mean ± SD for two or more independent experiments.

Variant	Anti-FIXa HC (H1)	Anti-FIXa LC (L1)	Anti-FX HC (H2)	Anti-FX LC (L2)	BsAb yield (%) Plasmid ratio (H1:L1:H2:L2)
	CH1	CL	CH1	CL	1:3:1:1	1:1:1:1	1:1:1:3
Parent					61.6 ± 1.4	58.5 ± 1.6	19.2 ± 1.5
C1	K147/Q175K	S131E/Q160E	K147E/Q175E	T131K/E160K	98.0 ± 0.5	100.0 ± 0.0	93.2 ± 5.9
C2	K147E/Q175E	S131K/Q160K	K147/Q175K	T131E/E160	94.4 ± 0.7	86.6 ± 6.3	45.2 ± 3.8
C3	K147/Q175K	S131E/T180E	K147E/Q175E	T131K/S180K	97.3 ± 1.0	99.4 ± 0.4	96.2 ± 2.5
C4	K147E/Q175E	S131K/T180K	K147/Q175K	T131E/S180E	96.5 ± 2.0	95.2 ± 2.7	84.9 ± 5.5
C5	K147/Q175K	Q160E/T180E	K147E/Q175E	E160K/S180K	88.9 ± 4.6	88.4 ± 0.5	52.1 ± 2.2
C6	K147E/Q175E	Q160K/T180K	K147/Q175K	E160/S180E	92.5 ± 1.1	90.8 ± 1.5	45.1 ± 0.5
C7	K147/K213	E123/S131E	K147E/K213E	E123K/T131K	94.7 ± 2.8	96.4 ± 1.5	85.4 ± 9.0
C8	K147E/K213E	E123K/S131K	K147/K213	E123/T131E	96.8 ± 0.1	96.7 ± 1.1	90.4 ± 2.6
C9	K147/K213	E123/Q160E	K147E/K213E	E123K/E160K	82.0 ± 3.1	78.4 ± 2.2	31.9 ± 1.4
C10	K147E/K213E	E123K/Q160K	K147/K213	E123/E160	83.1 ± 1.2	80.8 ± 2.8	43.3 ± 8.0
C11	K147/K213	E123/T180E	K147E/K213E	E123K/S180K	80.6 ± 6.7	81.5 ± 3.5	54.7 ± 1.9
C12	K147E/K213E	E123K/T180K	K147/K213	E123/S180E	84.2 ± 4.9	91.4 ± 0.7	74.0 ± 3.9
C13	Q175K/K213	E123/S131E	Q175E/K213E	E123K/T131K	80.8 ± 5.3	92.7 ± 0.3	93.9 ± 3.5
C14	Q175E/K213E	E123K/S131K	Q175K/K213	E123/T131E	81.2 ± 4.8	70.5 ± 6.2	25.8 ± 3.0
C15	Q175K/K213	E123/Q160E	Q175E/K213E	E123K/E160K	79.0 ± 4.1	83.2 ± 2.3	29.9 ± 5.6
C16	Q175E/K213E	E123K/Q160K	Q175K/K213	E123/E160	70.4 ± 1.1	65.9 ± 0.7	19.1 ± 1.3
C17	Q175K/K213	E123/T180E	Q175E/K213E	E123K/S180K	71.4 ± 11.9	78.1 ± 2.9	64.2 ± 8.4
C18	Q175E/K213E	E123K/T180K	Q175K/K213	E123/S180E	78.0 ± 1.6	77.5 ± 3.6	23.5 ± 1.9
C19	K147/Q175K	S131E/Q160E/T180E	K147E/Q175E	T131K/E160K/S180K	97.8 ± 0.2	98.5 ± 0.1	100.0 ± 0.0
C20	K147E/Q175E	S131K/Q160K/T180K	K147/Q175K	T131E/E160/S180E	97.5 ± 3.6	95.5 ± 3.0	97.2 ± 1.7
C21	K147/Q175K/K213	E123/S131E/Q160E/T180E	K147E/Q175E/K213E	E123K/T131K/E160K/S180K	99.0 ± 1.4	99.6 ± 0.6	100.0 ± 0.0
C22	K147E/Q175E/K213E	E123K/S131K/Q160K/T180K	K147/Q175K/K213	E123/T131E/E160/S180E	99.7 ± 0.4	98.8 ± 0.9	100.0 ± 0.0

Evaluation of physicochemical properties and pharmacological activity of antibodies with multiple mutations introduced at the CH1/CL interface

To determine the most appropriate FAST-Ig design for the parent antibody of NXT007, we evaluated many characteristics of the antibody variants with FAST-Ig mutations (FAST-Ig variants), such as physicochemical properties, pairing potency, and pharmacological activity. First, FAST-Ig variants C1-C4, C19-C20 were analyzed by non-reducing SDS-PAGE. C2 and C4 were analyzed together because they have the same mutation positions as C1 and C3, respectively. None of FAST-Ig variants abolished the formation of disulfide bonds between HC/LC (Figure 2a). In addition, none of FAST-Ig variants showed a significant increase of aggregation or degradation in the size exclusion chromatography (SEC) analysis (Figure 2b). In silico immunogenicity risk assessment using ISPRI from EpiVax, Inc.^43–45 showed that none of the FAST-Ig mutations significantly affected the immunogenicity-prediction score of the parent antibody (Table S3). Next, the midpoint of the thermal denaturation (Tm) of the Fabs of both arms was measured for C1 and C3, with the two pairs of mutations, and for C19 and C21, which had an even higher number of mutations. A slight decrease in Tm was observed as the number of mutations in CH1/CL increased, but it did not change significantly compared to the parent of NXT007 or the variant with the mutations of the variable region (V1, V2) (Fig. S4). On the other hand, the expression level was lower in FAST-Ig variants with more than three pairs of mutations (C19, C20, C21, and C22) than in variants with two pairs of mutations (C1 and C3), albeit at an acceptable level (Table S4). Therefore, we next compared the pairing potency and expression level of the two pairs of mutations (C1, C2, C3, and C4) in detail by widely varying the plasmid ratios (H1:L1:H2:L2 = 1:16:1:1, 1:8:1:1, 1:4:1:1, 1:2:1:1, 1:1:1:1, 1:1:1:2, 1:1:1:4, and 1:1:1:8). As a result, C1 and C3 showed equivalently high pairing potency at all plasmid ratios with expression levels equal to that of the parent NXT007 (Figure 2c,d). Although C2 and C4 facilitated cognate HC/LC assembly (>90%) under conditions with an increased anti-FIXa (L1: kappa) LC ratio, the expression level was significantly reduced compared to C1 and C3 (Figure 2d). The hydrophobicity and nonspecific binding profiles of C1 and C3 were similar to those of the parent antibody (Fig. S5 and S6). Furthermore, their comparable pharmacological activities were confirmed by FIXa catalyzed FX activation in an enzymatic assay (Figure 2e). Based on these results, we selected C1 or C3 for this antibody.

Figure 2. Detailed characterization of the selected FAST-Ig variants (a) the efficient interchain disulfide bond formation of BsAbs with FAST-Ig mutations was confirmed by non-reducing SDS-PAGE analysis. 2.5 µg BsAb/lane was applied onto Mini-Protein TGX gel 4–20% (Bio-Rad) and stained by Quick-CBB (Wako). (b) Size exclusion chromatography confirmed the same high level of monomer ratio of BsAb (about 150 kDa) among all selected FAST-Ig variants. The antibodies purified by protein A were used for the analysis. (c) Detailed analysis of the BsAb yield of parent and FAST-Ig variants (C1-C4). The antibodies were expressed by changing the plasmid mass ratio of L1 and L2 while H1 and H2 were fixed at 1:1 (Total plasmid amount was fixed as 1 µg for 1 mL Expi293F transient expression). Data was obtained from three independent experiments and are expressed as mean ± SD. Data could not be obtained in C4 with L1:L2 = 1:8 ratio due to the low expression level. (d) Antibody yields after protein A purification were compared among parent and BsAbs with FAST-Ig mutations (C1-C4) in the same expression condition as (c). Data was obtained from three independent experiments and are expressed as mean ± SD (e) NXT007 with different FAST-Ig variants showed FVIII-mimetic activity comparable to that of the parent and to BsAbs prepared using the in vitro re-constituting method, as confirmed by FIXa catalyzed FX activating enzymatic assay. The Y-axis indicates the absorbance at 405 nm of the chromogenic substrate assay. All the data were collected in triplicate and are expressed as mean ± SD. Parent and C1-, C3-assembled BsAb were prepared in ART-S3 format by in vitro re-construction in reduced condition using separately expressed monospecific antibodies, followed by desalting using dialysis cassettes. C1 and C3 were prepared in dKiH_v14 format from four chain transfected single Expi293F cells, followed by protein A purification. BsAb yields of all antibodies were validated > 95% by CIEX. Data of C1 and C3 was not collected at 1000 nmol/L due to the insufficient concentration of the prepared antibody.

Pharmacokinetic analysis of representative FAST-Ig variants

Next, we performed a pharmacokinetic study by using human neonatal Fc receptor (human FcRn) transgenic mice to confirm whether the FAST-Ig mutations (C1 and C3) chosen for NXT007 affect the antibody pharmacokinetics. These mutations showed pharmacokinetic profiles and AUC_0-7day that were comparable with the parent NXT007 (Figure 3 and Table 3). Therefore, from the pharmacokinetic perspective, either C1 or C3 can be introduced as FAST-Ig mutations for the parent NXT007.

Figure 3. Pharmacokinetic profiles of the selected FAST-Ig variants Comparable pharmacokinetics of NXT007 with different FAST-Ig mutations were confirmed in human FcRn homozyhous transgenic mice strain, B6.Cg-Fcgrttm1Dcr Tg(FCGRT)32Dcr. One mg/kg dose of each BsAb was administered by single intravenous injection via tail vein. Individual data points were collected from three mice in each group and are expressed as mean ± SD.

Table 3. AUC parameters of NXT007 with various FAST-Ig mutations and its parent from Day 0 to Day 7 in human FcRn transgenic mice. AUC profiles comparable to that of the parent were observed in tested FAST-Ig variants. A 1 mg/kg dose of each BsAb was administered by single intravenous injection via tail vein. Data were collected from three mice in each group and are expressed as mean ± SD. The BsAb yields of all antibodies used in this study were confirmed >95% by CIEX analysis.

FAST-Ig variant	AUC0-7day (µg·day/mL)
Parent	46.9 ± 5.1
C1	44.2 ± 1.4
C3	48.0 ± 6.9

Selection of C3 considering removability of mispair in purification

Next, FAST-Ig variants with C1 and C3 were compared in terms of removability of mispairs during the downstream purification process. Ion-exchange chromatography (IEC) is a standard method for purifying the target BsAb and eliminating the mispaired IgG species during the manufacturing process. This method works based on the difference in isoelectric point (pI) between the target BsAb and the mispaired species.^5,29 The final NXT007 molecule is designed so that the pI of the mispaired components of the anti-FIXa and anti-FX homodimeric antibodies differ compared to the correctly assembled BsAb. In addition, dKiH_v14, or ART-S3 mutations are introduced to favor HCs-heterodimerization. Thus, anti-FIXa and anti-FX homodimeric mispaired IgG antibodies (H1L1H1L1 and H2L2H2L2) are more likely to be removed by the IEC purification process, while the mispaired antibodies with common LC (H1L1H2L1 and H1L2H2L2) must be carefully managed. We performed CIEX analysis on antibodies that incorporated C1 or C3 on the parent NXT007, as well as their respective common LC mispaired antibodies. The results showed that H1L2H2L2 with C3 eluted at a time point further from the main peak than H1L2H2L2 with C1 (Fig. S7A and S7B). This is probably because C3 has a more negatively charged surface at CH1/CL of H1L2H2L2 than C1 (Fig. S7C). Based on these results, C3 was selected as the final FAST-Ig design for the orthogonal HC/LC pairing of NXT007. The feasibility of purification is discussed below. The importance of segregating H1L2H2L2 was also confirmed by the unintended prolongation of prothrombin time (PT) (Fig. S7D).

Structural analysis of anti-FIXa and FX-Fabs of NXT007 with C3 mutations introduced

To understand how C3 facilitates the correct association of HC/LC, crystal structure analysis was performed on the parent and C3-introduced anti-FIXa and anti-FX Fabs of NXT007 (Table S5). Fab-FIXa and Fab-FX are the parent Fab structures of NXT007 without the FAST-Ig mutations, consisting of human IgG₄/kappa and human IgG₄/lambda. The crystal structures of Fab-FIXa and Fab-FX were determined at 3.19 Å (PDB: 8GV0) and 1.27 Å (PDB: 8GV2) resolution, respectively (Figure 4a,b). Fab-FIXa-C3 and Fab-FX-C3 were the final Fab structures of NXT007, in which HC_Q175K, LC_S131E/T180E and HC_K147E/Q175E, LC_T131K/S180K were introduced, respectively. The crystal structures of Fab-FIXa-C3 and Fab-FX-C3 were determined at 1.79 Å (PDB: 8GUZ) and 1.19 Å (PDB: 8GV1) resolution, respectively (Figure 4a,b). The crystallographic asymmetric unit of Fab-FIXa contained four molecules (Mol A, B, C, and D), and that of Fab-FIXa-C3 contained two molecules (Mol A and B). The structures of Fabs contained in the asymmetric units are quite similar, so we selected Mol A for further discussion of both cases. The overall structures of the C3-introduced Fabs are essentially identical to that of parent Fabs. The backbone root mean square deviations (RMSD) are 1.077/0.347 Å between Fab-FIXa-C3/Fab-FX-C3 and the respective parent Fabs, as calculated from equivalent Cα atoms. This indicates that the introduced FAST-Ig charge-pairs have little effect on the overall Fab structure.

Figure 4. Overall structure of NXT007 Fabs with C3 mutations (a) Structural superposition of the anti-FIXa Fab structure of NXT007 containing CH1_(K147), Q175K, and CL_S131E, T180E mutations (Fab-FIXa-C3: PDB 8GUZ) and the parent anti-FIXa Fab (Fab-FIXa: PDB 8GV0, light green). In the Fab-FIXa-C3, HC and LC are orange and light orange, respectively. The key residues are shown as stick models. (b) Structural superposition of the anti-FX Fab structure of NXT007 containing CH1_K147E, Q175E, and CL_T131K, S180K mutations (Fab-FX-C3: PDB 8GV1) and the parent anti-FX Fab (Fab-FX: PDB 8GV2, light purple). In the Fab-FX-C3, HC and LC are blue and light blue, respectively. The key residues are shown as stick models. (c, d) Close-up views of the charge-pairs of FAST-Ig C3 mutations. (c) The three key interactions in Fab-FIXa-C3 (CH1_K147/CL_S131E, CH1_Q175K/CL_S131E, and CH1_Q175K/T180E) are shown with light blue dashed lines. (d) The three key interactions in Fab-FX-C3 (CH1_K147E/CL_T131K, CH1_Q175E/CL_T131K, and CH1_Q175E/S180K) are shown with light blue dashed lines. Molecular graphic images were prepared using CueMol (http://www.cuemol.org).

On the other hand, when the individual FAST-Ig mutations in the CH1/CL interface are closely inspected, salt bridges at HC_K147/LC_S131E, HC_Q175K/LC_T180E, HC_K147E/LC_T131K, and HC_Q175E/LC_S180K are formed as intended in both the Fab-FIXa-C3 and Fab-FX-C3 structures, which was confirmed using MOE (Molecular Operating Environment) software (Figure 4c,d). In the Fab-FIXa-C3 structure, the nearest neighbor distance between amino acids HC_K147/LC_S131E, HC_K147/LC_T180E, HC_Q175K/LC_S131E, and HC_Q175K/LC_T180E are 2.70 Å, 6.55 Å, 5.20 Å, and 3.05 Å, respectively. When similarly analyzed in the Fab-FX-C3 structure, the distance between amino acids HC_K147E/LC_T131K, HC_K147E/LC_S180K, HC_Q175E/LC_T131K, and HC_Q175E/LC_S180K are 2.75 Å, 7.96 Å, 4.51 Å, and 3.43 Å, respectively. Thus, the selected residues were placed in structural positions that allowed them to interact with their respective pairs.

Comparison of dKiH_v14 and ART-S3 in terms of stability after acidic treatment and feasibility of purification

In previous studies that assessed the pairing potency between HC/LC, a dKiH_v14 construct was used because it provides rigid heterodimerization of HCs, thus simplifying the analysis. However, to be suitable for a clinical trial, the final molecule should be designed based on a variety of experimental evidence, including its physicochemical characteristics and immunogenicity risk. In the commercial production, monoclonal antibodies are generally exposed to low pH during the protein A affinity purification and virus inactivation process.^46,47 It has been reported that exposure to low pH can cause IgG to denature and aggregate; IgG₄ molecules are particularly susceptible to an acidic environment.⁴⁷ ART-S3, used in emicizumab, is another set of mutation candidates that can heterodimerize HCs. Their efficiency and developability were validated in late-stage development and in manufacturing.⁵ Here we compared dKiH_v14 and ART-S3, namely their stability and efficiency in HCs-heterodimerization, to determine which mutations should be used for NXT007 (human IgG₄-based, kappa, lambda).

First, the stability of the parent NXT007 antibody with dKiH_v14 or ART-S3 was examined after acid treatment at pH 3.6 and pH 3.2, considering the viral inactivation process.⁴⁸ Each sample was analyzed by SEC with the fluorescence probe 8-Anilino-1-naphthalenesulfonic acid (ANS) and FcRn binding ability was assessed by SPR. In general, ANS can bind to the exposed hydrophobic surface of the protein,^49,50 so the degree of the denaturation of antibodies can be measured by the fluorescence intensity in SEC spiked with ANS.⁵¹ We compared the fluorescence intensity in each sample before and after acid treatment. As a result, the dKiH_v14 antibody showed more ANS florescence than the ART-S3 antibody (Figure 5a-c). Furthermore, the FcRn binding ability significantly decreased to ~80% in dKiH_v14 compared to a decrease to ~98% in ART-S3 (Figure 5d). These results indicate that the ART-S3 antibody was more stable than the dKiH_v14 antibody under acidic conditions. This observation was consistent with the CH3 domain of ART-S3 displaying superior thermal stability compared to that of dKiH_v14 (Figure S8). The hydrophobicity and nonspecific binding profiles of dKiH_v14 and ART-S3 were comparable (Figures S5 and S6).

Figure 5. Comparison of stability after low pH exposure and pairing potency between dKiH_v14 and ART-S3 (A-C) Acid-induced denature of NXT007 with different Fc mutations, dKiH_v14 (a) and ART-S3 (b), was compared by SEC analysis at pH 6.0 (indicated as initial), pH 3.6, and pH 3.2 for 3 hours, followed by neutralization with or without incubation for 24 hours at room temperature (RT). (c) Y axis indicates the extent of denature of antibody as the whole area ratio (arbitrary unit) of ANS fluorescence at 510 nm divided by UV detection at 280 nm. (d) Stability of the human FcRn (hFcrn) binding of NXT007 with dKiH_v14 and ART-S3 mutations was compared by SPR analysis. The hFcrn binding ability was calculated as relative binding response (Resonance Unit) per antibody capture amount. Samples were same as in (a-c). (e, f) CIEX chromatography analysis with an optimized CIEX method (pH and NaCl concentration gradient, described in materials and methods) for NXT007 with dKiH_v14 (e) or ART-S3 (f). The same chromatograms with reference mispairs are shown in Figure S9. Antibodies were expressed at the optimized plasmid mass ratio H1:L1:H2:L2 = 1.3:2:1:2 in Expi293F and purified by protein A.

We next compared the pairing potency of dKiH_v14 and ART-S3. When both antibodies were transiently expressed in 100 mL culture of Expi293F cells at an optimized plasmid ratio, the BsAb yields after protein A purification were 98.4% for dKiH_v14 and 84.6% for ART-S3 (Figure 5e,f). The yields were 233 mg/L with dKiH_v14 and 284 mg/L with ART-S3, which were comparable. The most commonly mispaired species expressed in ART-S3 were the anti-FIXa homodimeric antibody (H1L1H1L1) and the anti-FX homodimeric antibody (H2L2H2L2). Similar to the results of the previous study with dKiH_v14 (Fig. S7B), we confirmed that the elution time of the H1L2H2L2 mispaired antibody did not overlap with that of the BsAb (H1L1H2L2) when the set of HCs-heterodimerization mutations was changed from dKiH_v14 to ART-S3 (Fig. S9).

Next, we compared the feasibility of NXT007 purification with either dKiH_v14 or ART-S3. Both antibodies purified by protein A affinity column were further purified by IEC (Fig. S10A and S10B). CIEX analysis demonstrated ~100% BsAb yield in both antibodies (Fig. S10C and S10D). To further confirm the removal of the mispairs, LC/MS analyses were conducted. The molecular mass of the antibodies was measured under a non-reducing condition, and only the mass corresponding to the correctly assembled BsAb (H1L1H2L2) was detected in each BsAb (Fig. S11). However, the theoretical mass of the LC exchanger (H1L2H2L1) has the same mass as the correctly assembled BsAb. Therefore, by measuring the masses of the BsAbs’ Fabs using SEC/MS, we examined the contamination of the mispaired LC exchanger. To prepare the Fabs, disulfide bonds at the hinge region of F(ab’)2 were reduced to under 200 mmol/L tris(2-carboxyethyl)phosphine (TCEP). For the F(ab’)2 preparation, the BsAbs were treated with FabRICATOR. The obtained mass values of the Fabs corresponded only to the correctly assembled BsAb, and the mass values of the LC exchanger were not detected in either sample (Fig. S12). Thus, the ART-S3 variant of NXT007 can be purified as highly as the dKiH_v14 variant using protein A affinity purification and further IEC purification at the laboratory scale of expression. Based on these results, we selected the ART-S3 for the final design because of its superior stability after acid treatment and comparable feasibility of purification.

Development of stable CHO cell line for NXT007 production

In the experiments described so far, we used transient expression in Expi293F cells to determine the appropriate FAST-Ig mutations for NXT007. However, since CHO cells are generally used for industrial antibody production, we must also achieve efficient BsAb production in CHO cells.⁵² Therefore, we created and evaluated a stable CHO cell line for producing NXT007. The CHO cell lines were developed using a random integration method and the dihydrofolate reductase (DHFR)/methotrexate (MTX) gene amplification system. Sixty stable monoclonal cell lines were produced and screened based on titer, BsAb yield, monoclonality, and production and genetic stability. To simplify the measurement of BsAb yield, an IEC method that was faster than that used for transiently expressed Expi293F was used to rank the CHO cell lines (described in Materials and Methods). The resulting BsAb yields and titers for 60 clones are shown in Figure S13A and the representative IEC chromatograms of these samples are shown in Figure S13B. Based on the overall evaluation, which included monoclonality, production stability, and genetic stability, clone 46 was selected as the best strain for commercial production (data not shown).

For a more detailed analysis, this clone was further expanded and cultured for 57 days and the expressed antibody was purified by protein A. The titer at day 14 was 3.83 g/L, which is consistent and within the range of typical titers obtained with CHO cells.⁵²

When these protein A purified antibodies were analyzed using the optimized IEC method for NXT007 (Fig. S9), the actual percentage of BsAb was 78.4%, mostly comparable to that (~85%) obtained in the transient expression study using Expi293F (Figure 5f and Table 4). We confirmed that the mispaired species that lowered the BsAb yield were mainly derived from the HC mispairing (Table 4).

Table 4. Characterization of the stable CHO cell line for production of NXT007. The cell line derived from clone 46 in Figure S13 was characterized. Titer was obtained from the protein A purified NXT007 pool on Day 14, which was 57 days after thawing the master cell bank (MCB). The percentage of the target BsAb (H1L1H2L2), major homodimeric mispairs (H1L1H1L1, H2L2H2L2), and the other mispairs were measured using same CIEX analysis methods described in Figure S9.

	BsAb yield (%) [H1L1H2L2]	Anti-FIXa (%) [H1L1H1L1]	Anti-FX (%) [H2L2H2L2]	Other mispairs (%)	Titer (g/L)
Clone 46 derived MCB	78.4	17.2	2.2	2.2	3.83

These results indicate that the pairing potency of C3 with ART-S3 is stable over a long period of time, and that the pairing potency in transient expression in Expi293F cells can be generally extrapolated to the stable line of CHO cells.

Versatile application of FAST-Ig technology to diverse antibody pairs

Our previous experiments have focused on installing FAST-Ig on NXT007 (IgG₄ kappa/lambda). To demonstrate the versatility of the FAST-Ig technology, we extended our analysis to NXT007 with different formats (IgG₁ kappa/lambda, IgG₄ kappa/kappa) and four IgG₁ kappa BsAbs (anti-HER2/MS4A1, anti-IL−6 R/IL−12; IL-23A, anti-NRP1/IGF1R, anti-RSV glycoprotein F/IgE). All antibodies were transiently expressed in Expi293F cells using dKiH_v14 format. We observed a high BsAb yield (>95%) for C1 or C3 in some expression conditions, even in NXT007 with different formats (Tables S6 and S7). This result is consistent with our previous observation that the positively charged LC of the anti-FX arm (C1 and C3) achieved better pairing potency than C2 and C4 in NXT007 (Figure 2c and Table 2). Among the four pairs of BsAbs, the anti-HER2/MS4A1 BsAb exhibited less parental cognate HC/LC pairing (49.3%) than that of NXT007 (61.6%). However, we achieved a high BsAb yield (>95%) for all four parent BsAbs, including anti-HER2/MS4A1, by combining the FAST-Ig mutations of both the CH1/CL and the VH/VL interfaces (Table S8-S11). These results demonstrate the robust pairing potency of the FAST-Ig mutations and emphasize the importance of selecting appropriate mutations based on parental cognate HC/LC pairing.

Discussion

In this study, we examined novel orthogonal Fab designs based on electrostatic complementarity with the combination of reported mutations for preferential HCs-heterodimerization. Thus, we established the FAST-Ig platform to efficiently express the IgG-type BsAb with native IgG structure in single mammalian cells. In addition, by comprehensively evaluating a variety of factors that affect developability, we identified the most suitable FAST-Ig variants for the clinical candidate antibody NXT007 (IgG₄ kappa/lambda), a treatment for hemophilia A.

Generally, single-cell production of standard IgG-type BsAbs has advantages in manufacturing and these standard IgG-type BsAbs have better physicochemical properties than non-canonical IgG formats. Among the technologies for single-cell production, CrossMab³⁴ has advanced the furthest in clinical trials;^7,8,11 however, it might cause slight differences in the binding orientation of both Fabs compared to parental BsAbs. These differences can affect the pharmacological activity of BsAbs such as emicizumab, whose mechanism of action depends stringently on binding orientation.⁵³ Compared to DuetMab³⁵ and Orthogonal Fab technologies^20–22 that retain the native Fab structure, FAST-Ig has unique features such as the ability to select the most optimal mutations for parental Fab sequences from a range of options with various levels of pairing potency and the capacity to alter the surface charge of “mispairs”.

To facilitate the cognate pairing of HCs and LCs, we first focused on the residues of CH1_147, 175, 213 and CL_123, 131, 160, 180 located at the CH1/CL interface and substituted them with Glu and Lys pairs in a range of about 5–10 Å. Aside from Glu and Lys used in this study, aspartic acid (Asp) and histidine (His)/arginine (Arg) are other options for positively and a negatively charged residues, respectively. Although not exhaustively validated, we obtained data showing that these residues can also promote the correct HC/LC pairing (data not shown). However, Asp should generally be avoided because it can potentially induce degradation pathways such as succinimide intermediate and isomerization formation, which leads to the heterogeneity of the molecule.^54,55 Also, the charge state of His may fluctuate in the in vivo environment because its pKa is around 6.0–6.5, leading to a lack of uniformity in pI. Also, there are reports showing that Lys is superior than Arg in terms of solubility and aggregation.^56–58 Other reports using a peptide tag showed that Arg-tag had more cellular uptake than Lys-tag and also showed more immunogenicity.⁵⁹ Although the circumstances might differ when these residues are introduced at the interface of HC/LC in antibodies, these factors motivated us to prioritize Lys and Glu as substitutions in this study.

The amino acid substitutions in the CH1/CL interface may affect the antibody folding process in the endoplasmic reticulum (ER). It has been reported that CH1 interacts with BiP, a chaperone which plays a crucial role in the folding of antibodies.^60,61 The CH1_147, 175, 213 positions used in the FAST-Ig design differed from those of critical BiP binding residues.^62,63 In fact, the expression level and thermal stability of BsAbs bearing C1 or C3 mutations, which included both CH1_147 and 175 mutations, were comparable to the parent antibody of NXT007 (Figure 2d and S4). The IgG structures of NXT007 Fabs with FAST-Ig mutations were also similar to that of the parent (Figure 4). These results further suggest that the correct folding process occurred in ER for the BsAb with the FAST-Ig mutations.

The design strategy for determining which FAST-Ig mutations should be used for new templates is critical for versatile application. First, the purpose and criteria for the use of FAST-Ig must be clarified. When applying FAST-Ig to a clinical candidate antibody, one of the key factors is the parental cognate HC/LC pairing preference. It has been reported that the HC/LC paring preference of parent antibodies varies greatly depending on their Fab sequences.^20,64 The parent of NXT007 had a relatively strong HC/LC preference that achieved up to 61.6% BsAb yield in the dKiH_v14 format without FAST-Ig mutations. For template BsAbs with strong cognate HC/LC preference, two pairs of mutations in the CH1/CL interface (C1-C4) would be ideal for reducing artificial mutations as much as possible. When the HC/LC pairing preference of the parent antibody is weak, three or more pairs of mutations should be selected (C19-22). Furthermore, if they do not provide sufficient pairing potency, VH_39, VL_38 mutations (V1 or V2 in Table S1) should be added in addition to the CH1/CL mutations. Whether to incorporate VH/VL mutations or not also depends on the mode of action (MoA) of the BsAb. Although the introduction of VH/VL mutations in NXT007 had a slight impact on affinity (Table S2), this is acceptable when the template has a simple MoA, such as dual blockers.^7,8 The overall FAST-Ig design strategy is summarized in Figure S14. By following this strategy, the FAST-Ig platform can be flexibly adjusted to achieve the appropriate pairing potency for each parent antibody (Table S8-S11). However, if the parental cognate HC/LC pairing is too weak, even full FAST-Ig mutations might not provide sufficient pairing potency. In such cases, there may be a way to reacquire a lead BsAb with strong cognate HC/LC pairing. For example, Gong et al. obtained two pairs of BsAbs with >75% BsAb yield without any mutations in the HC/LC interface by screening 8 × 4 = 32 combinations of antibody A and B.⁶⁵ If such parent antibodies are obtained, only a small number of CH1/CL mutations would be needed to facilitate the cognate HC/LC association. In addition to the high BsAb yield, it is also important to maintain a sufficient expression level. While designing NXT007, the expression level decreased as the number of FAST-Ig mutations in the CH1/CL interface increased (Table S4). However, several templates other than NXT007 retained an acceptable expression level even when more than two mutation pairs were added in CH1/CL (data not shown). Therefore, increasing the number of FAST-Ig mutations is not to be unconditionally avoided, but should be considered after comparing several mutations (Figure S14).

Interestingly, in the process of designing FAST-Ig mutations to create NXT007, a positively charged anti-FX arm LC (lambda) and negatively charged anti-FIXa arm LC (kappa) resulted in a higher pairing potency than the oppositely charged pattern (Figure 2c). Because this tendency was reproduced in different formats (IgG₁ kappa/lambda, IgG₄ kappa/kappa), the preferred charge orientation in NXT007 depends more on Fab sequences than on IgG subclasses (Tables S6 and S7). Changing the charge orientation also changed the pairing potency in other template antibodies, even with the same pair of FAST-Ig mutations (Tables S8-S11). We also confirmed that reversing the charge orientation of FAST-Ig affected expression levels (Figure 2d). These results highlight the importance of trying both HC/LC charge patterns even with the same number of FAST-Ig mutations.

Previously, the CH1/CL interface was comprehensively screened by introducing single paired charge modifications.²⁰ The selected mutation pairs, CH1_S183K(E) and CL_V133E(K), were located in the central core of the CH1/CL interface. Compared to these mutations, FAST-Ig mutations are located slightly near the rim of the CH1/CL interface (Figure 4, S1, and S7C). Consistent with the generally accepted view that solvent exposure weakens charge interactions by canceling the ionic effect,⁴⁰ a single pair of FAST-Ig mutations in CH1/CL was not sufficient to achieve a stronger pairing potency than the reported core mutations (Table 1 and Table S12). However, we have successfully achieved a high BsAb yield (>95%) by combining two or more pairs of mutations (Table 2 and Table S8-S11), possibly due to the formation of charge clusters. Similar mutation pairs to FAST-Ig, CH1_K147E, K213E(D) and CL_E123K, Q124R(K) have been reported in combination with CrossMab.⁶⁶ While the pivotal two pairs of FAST-Ig mutations, C1-C4, have charge interactions in all four possible pairs between CH1/CL (Table 1), the reported mutation pairs are in independent structural positions. This suggests that achieving strong pairing potency without domain crossover may be challenging. Future studies should compare these mutations with FAST-Ig.

Regarding orthogonal Fab design to remove the mispaired byproducts in IEC, it would be difficult to modify the surface charge of the mispaired antibodies when mutations are located in the core. However, if the charged modifications are located at a marginal position, it is possible to modify the surface charge of the mispaired antibodies without interfering with the correct pair because the formation of salt bridges cancels out the electrostatically charged residues (Figure 4c,d, and S7C). In fact, in the case of NXT007, the separation of one of the mispaired byproducts (H1L2H2L2) from the main peak in the IEC purification process was achieved by selecting C3 (Fig. S7A and S7B). Such controlled separation of mispaired species is especially useful for removing byproducts that may potentially lead to toxicity or a reduction in efficacy. In NXT007, because the anti-FIXa Fab has weaker neutralizing potential than the anti-FX Fab (Fig. S3 and in-house data, data not shown), contamination by mispaired species with an anti-FX binding Fab should be more carefully managed than those with FIXa binding ability. In fact, when comparing the prolonged PT time in normal plasma, reflecting FX-neutralization, the mispaired antibodies with FX binding arms prolonged PT, although the prolongation was not evident at less than 30 nmol/L of mispaired antibodies (Fig. S7D). Therefore, selecting C3 to segregate mispair with potential neutralizing activity, such as H1L2H2L2 contributes to safer drug production (Fig. S7A, S7B, and S9). In actual commercial manufacturing, small amounts of mispaired molecules can even be expressed with established BsAb production technologies. Thus, to consider not only BsAb yield but also safety, we should assess the risk of each mispair during the molecular design process.

To find a suitable design that can enforce the HCs-heterodimerization in NXT007, the stability of the dKiH_v14 and ART-S3 formats were compared in the acidic condition. The commercial production of therapeutic antibodies generally includes acidic treatment for virus inactivation. Although the KiH approach has been widely used in preclinical- and clinical-stage molecules,^7,8,11 it is important to note that there are few reports on the detailed properties of KiH applied to IgG₄-based molecules. We demonstrated the fluorescent of ANS after acid treatment was higher in the dKiH_v14 format than ART-S3 format (Figure 5a-c). We also observed that KiH variants without disulfide bonds showed higher fluorescence than dKiH_v14 (data not shown). This suggests that KiH is structurally denatured under acid conditions, resulting in the exposure of partial hydrophobic surfaces. Although the effects of these structural changes on clinical efficacy are not clear, it is possible that the IgG₄-dKiH_v14 format after acid treatment could affect antibody pharmacokinetics due to its reduced FcRn binding (Figure 5d). The partial recovery of FcRn binding along with reduced ANS binding after 24-hour incubation at RT suggests that further incubation may enforce the correct refolding (Figure 5a-d); however, further evidence is needed before choosing to apply KiH to the IgG₄-based clinical candidate.

Since NXT007 is a next-generation version of emicizumab and is intended for lifelong use with long intervals between doses in patients with hemophilia A, we wanted to minimize not only the risk of pharmacokinetic effects but also of immunogenicity. Emicizumab, which uses ART-S3 to promote HCs-heterodimerization, has already been proven to have a low risk of immunogenicity in the clinic (incidence rate of ADAs with a loss of efficacy is ≥1/1000 to <1/100).⁶⁷ Hence, we prioritized using the same mutations as emicizumab because of this robust clinical evidence. In addition, the immunogenicity of C3 was predicted in silico to be low, which was comparable to the parent (Table S3). This was consistent with the in vitro assays measuring the amount of IL−2 produced by T-cell activation (data not shown). Thus, NXT007 was designed to reduce the risk of immunogenicity. Actual immunogenicity in humans is currently being determined in the Phase 1/2 clinical trials of NXT007 (jRCT2031220050, jRCT2080224835).

When considering the commercial production of antibodies, it is essential to extrapolate their pairing potency and expression level from their transient expression in Expi293F cells to a stable CHO cell line.⁵² During the construction of the CHO cell line for the commercial production of NXT007, we generated 60 cell lines and ranked them with a simplified screening system (Fig. S13). Although the final cell line selected for production ranked 46th out of 60 in terms of BsAb yield, the BsAb yield was about 78% in the CHO cell line, similar to the BsAb yield (about 85%) transiently expressed in Expi293F with the optimized plasmid ratio. This means that, in most of CHO cell lines (~45/60), NXT007 with FAST-Ig C3 and ART-S3 format maintained about >80% BsAb yield. The majority of mispaired species generated during the screening process and ultimately in the selected CHO cell line were anti-FIXa and anti-FX homodimeric antibodies (H1L1H1L1 and H2L2H2L2). This may be due to slightly weaker heterodimerization of HCs by ART-S3, which was assumed based on the comparative study of dKiH_v14 and ART-S3 at the small scale (Fig. 5e,f). When introducing ART-S3, the cell lines should be constructed and selected so that the expression levels of both HCs are as equal as possible, reducing the potential for homologous HCs dimerization. Otherwise, if the parent antibody is a non-IgG₄ subclass with better stability in acidic conditions than IgG₄, stronger HCs-heterodimerization technologies such as KiH can be prioritized as molecular design. Regarding HC/LC paring, C3 on NXT007 showed robust pairing potency even in CHO cells since the proportion of mispaired antibodies, including the non-cognate HC/LC, was as low as 2.2%. During the selection process, the average titer was 1.38 ± 0.65 (±SD) g/L for CHO cell lines (few mL culture), and the finally selected clone had a titer 3.83 g/L in the expanded culture condition (>1000 L). Thus, this expression level was not considered to be a major problem in production with FAST-Ig, as it is expected in a CHO cell stable line for general antibodies.⁵² Although the purification processes on a commercial scale should be comprehensively validated separately, the most common mispaired byproducts generated in NXT007 production have sufficiently different pI from that of the main BsAb (Fig. S9B). Therefore, NXT007 can be produced with a BsAb yield of 99–100% on an industrial scale using an in-house purification process (data not shown).

In summary, we have established FAST-Ig, a novel antibody engineering technology that efficiently expresses IgG-type BsAb in single mammalian cells. In addition, we describe in detail how FAST-Ig was applied to NXT007 (IgG₄, kappa/lambda), a next-generation version of emicizumab for hemophilia A. Considering pairing potency and manufacturability, C3 was selected for cognate pairing in HC/LC. ART-S3, because of its stability in acidic conditions and clinically proven low immunogenicity, was selected to enforce heterodimerization of HCs. NXT007 with C3 and ART-S3 showed equivalent binding activity and pharmacological activity to that of the parent. It also had similar physiochemical properties in pharmacokinetics, structure, and immunogenicity. The pairing potency in transient expression using Expi293F cells was also largely maintained in stable CHO cell lines. In addition, FAST-Ig variants in the dKiH_v14 format showed >95% BsAb yield for several antibodies in the general IgG₁ kappa format other than NXT007. These results demonstrate the potential versatility of FAST-Ig technology; it can be flexibly used to screen a wide variety of BsAbs and can also be used for manufacturing. The design strategy reported in this study is not limited to FAST-Ig antibodies, but may also serve as a guide for the optimal creation of BsAb therapeutics engineered with other BsAb-producing technologies.

Materials and methods

Antibody expression and purification

Antibodies for FAST-Ig mutation screening were expressed transiently in Expi293F (Thermo Fisher Scientific) cells transfected with plasmids encoding immunoglobulin heavy chains and light chains, according to the manufacturer’s instructions (Thermo Fisher Scientific). Site-directed mutagenesis and sub-cloning into mammalian expression vectors were performed using an In-Fusion HD Cloning Kit (Clontech) or NEBuilder HiFi DNA Assembly Master Mix (New England BioLabs). Antibodies were purified by protein A affinity chromatography using a MabSelect PrismA (Cytiva). To assess the risk of mispaired species (Fig. S7D), plasmids encoding antibodies without FAST-Ig mutations were used for expression. Each HC contained ART-S3 mutations for enforcing the heterodimerization of HCs. Homodimeric antibodies (H1L1H1L1, H1L2H1L2, H2L1H2L1, H2L2H2L2) were prepared by co-transfecting the plasmids of each HC and LC of antibody, followed by protein A purification. To prepare the other six antibodies, pairs of parental homodimeric antibodies were separately prepared. These were mixed in 1:1 molar ratio and reconstituted in mildly reduced conditions (25 mmol/L 2-MEA), followed by dialysis to remove 2-MEA. Their BsAb yields were confirmed as >95% by CIEX analysis.

Cation exchange chromatography analysis

CIEX analysis for FAST-Ig variant screening was carried out using HPLC Column PEEK YMC-BioPro, 5 µm, non-porous, YMC-BioPro SP-F, 100 × 4.6 mm (YMC) with CX−1 pH Gradient Buffer A, pH 5.6 (Thermo Fisher Scientific) and CX−1 pH Gradient Buffer B, pH 10.2 (Thermo Fisher Scientific) at a flow rate of 0.5 mL/min on an Alliance HPLC System (Waters) at 40°C. A binary pump was used to deliver buffer A and buffer B as a linear gradient of 0%−100% buffer B over 100 min or 15%−80% buffer B over 65 min. The column was then washed for 10 min at 100% of B and further equilibrated for 25 min at 0% or 15% of B. Elution was monitored by UV absorbance at 280 nm. 100 µL of each sample diluted to be 0.1 mg/mL by buffer A was injected for analysis. If the expression levels of samples were too low, samples diluted with <0.1 mg/mL were injected.

CIEX analysis for CHO cell line screening was carried out using the ProPac WCX−10 column (4.0 mm ID × 250 mm, particle size, 5 µm) with mobile phase A (9.6 mmol/L Tris, 6.0 mmol/L piperazine, 11.0 mmol/L imidazole buffer, pH 6.0) and mobile phase B (9.6 mmol/L Tris, 6.0 mmol/L piperazine, 11.0 mmol/L imidazole, 150 mmol/L NaCl buffer, pH 9.9) at a flow rate of 1.0 mL/min on an Alliance HPLC System (Waters). A binary pump was used to deliver buffer A and buffer B as a linear gradient of 0%−100% buffer B over 20 min. The column was then washed for 15 min at 100% of B and further equilibrated for 15 min at 0% of B. Elution was monitored by UV absorbance at 280 nm. The sample was diluted with mobile phase A to prepare a solution containing 1 mg of protein per mL. CIEX analysis optimized to separate to mispairs of NXT007 was carried out using BioResolve SCX 3 µm, 4.6 mm ID × 50 mm with 3.9 mm ID VanGuard Fit (Waters) with mobile phase A (CHES, HEPES, MES, NaCl-based buffer, pH 6.0) and mobile phase B (CHES, HEPES, MES, NaCl-based buffer, pH 9.9) on an Alliance HPLC System (Waters) at 40°C. A binary pump was used to deliver buffer A and buffer B as a linear gradient of 0%−100% buffer B over 45 min. Elution was monitored by UV absorbance at 280 nm.

Purification of bispecific antibodies by cation exchange chromatography

Using an ÄKTA pure™ 25 (Cytiva), HiTrap™ Capto™ SP ImpRes 1 mL columns (Cytiva) were connected and equilibrated by 20 mmol/L sodium phosphate, pH 6.0. One mol/L of NaCl was added to the equilibration buffer, which was used as the elution buffer. The load sample was first prepared from culture supernatant using HiTrap MabSelect SuRe pcc 5 mL (Cytiva) and was then dialyzed with equilibration buffer. The sample antibody solution was applied to the HiTrap™ Capto™ SP ImpRes column at 3–6 mg/mL resin. After washing with equilibration buffer, antibodies were eluted by a linear gradient up to 50% of the elution buffer for 20 column volumes (CV), equivalent to NaCl concentration of 500 mmol/L. After that, the percentage of elution buffer was raised in a stepwise manner to 100% for 3 CV.

Binding kinetics analysis by surface plasmon resonance

Kinetic analysis of anti-FIXa and anti-FX antibodies was performed by SPR using a Biacore T200 system (Cytiva) at pH 7.4 at 25°C. The Anti-Human IgG (Fc) antibody (Cytiva, BR100839) was immobilized onto a CM4 sensor chip (Cytiva). Then, anti-FIXa or anti-FX monospecific antibodies with various FAST-Ig mutations were captured in amounts of approximately 1600 RU and 450 RU, respectively. The human FIXa (Enzyme Research Laboratories, HFIXa 1080) (0, 80, 160, 320, 640, 1280 nmol/L) and human FX (Enzyme Research Laboratories, HFX 1010) (0, 10, 20, 40, 80, 160 nmol/L) dissolved in running buffer (HBS-P+,1.2 mmol/L CaCl₂, 1 mg/mL bovine serum albumin (BSA), 1 mg/mL Carboxymethyl-dextran (CMD), 0.05% v/v Tween 20, pH 7.4) was injected at a flow rate of 30 µL/min to monitor the association phase for 180 s and the dissociation phase for 180 s. The sensor chip surface was regenerated twice using 3 mol/L MgCl₂. Kinetic parameters were determined by fitting the sensorgrams with 1:1 binding model using Biacore T200 evaluation software, version 2.0 (Cytiva). For anti-FIXa monospecific antibodies, accurate kinetics parameters could not be calculated due to poor fitting caused by relatively large bulk response and nonspecific binding of antigens. Therefore, only the kinetic parameters of anti-FX antibodies are listed and overall sensorgrams of anti-FIXa antibodies are shown as reference data.

Mass spectrometry

Mass spectra were obtained using a SYNAPT G2-Si HDMS mass spectrometer equipped with an electrospray ion source and MassLynx data processor (Waters Corp., Milford, MA). Samples were injected into the mass spectrometer through an ACQUITY UPLC I-class system (Waters Corp., Milford, MA). To measure the molecular mass of NXT007 with C3-dKiH_v14 and C3-ART-S3, the antibody solution (0.5 mg/mL) was treated with peptide-N-glycosidase F (MERCK KGaA, Darmstadt, Germany), and 3 mL of the sample was injected into the above LC/MS equipped with a MassPrep desalting column (Waters Corp., Milford, MA). The samples were eluted by a linear gradient of 10% to 85% v/v acetonitrile containing 0.1% v/v formic acid and 0.02% v/v trifluoroacetic acid for 10 min at 100 mL/min, and the column temperature was set at 70°C. For the mass spectrometer, an electrocapillary voltage of 3.0 kV and sample cone voltage of 40 V were employed. To examine contamination by the mis-paired antibody species of the purified NXT007 solution, Fab of the antibody was prepared, and molecular mass was measured as follows. The 0.2 mg/mL antibody was incubated with FabRICATOR (Genovis AB, Lund, Sweden) at 37°C for 4.5 h to generate F(ab’)2 and then the solution was incubated with 200 µmol/L TCEP at room temperature for 1 h to prepare Fab of the sample antibody. The resulting sample solution was injected into LC/MS equipped with the MabPac SEC−1 column (5 mm, 300 Å, 2.1 × 300 mm, Thermo Fisher Scientific, Sunnyvale, CA). The sample was eluted by 250 mmol/L ammonium acetate at 50 mL/min and the column oven was set at 30°C. For the mass spectrometer, an electrocapillary voltage of 2.5 kV and sample cone voltage of 140 V were used.

Pharmacokinetic study in human FcRn transgenic mice

One mg/kg doses of each bispecific antibody were administered to human FcRn homozygous transgenic mice, line #32 (B6.Cg-Fcgrt < tm1Dcr > Tg(FCGRT)32Dcr/DcrJ, Jackson Laboratories) by single intravenous injection via the tail vein (n = 3 for each group). The drug solution consisted of antibody (Parent, C1, and C3) in phosphate-buffered saline (PBS) containing Tween 20 and was administered at a volume of 10 mL/kg. Five minutes, 7 h, 1 day, 2 days, 3 days, and 7 days after injection, blood was collected from the cervical vein without anesthesia and mixed with heparin sodium. Plasma was obtained by centrifuging the blood (12,000 rpm, 4°C, 5 min). Plasma concentration of bispecific antibodies was determined by sandwich enzyme-linked immunosorbent assay (ELISA) using anti-idiotype antibodies (in-house). The AUC_0-7day was calculated from the plasma concentration – time data using non-compartmental analysis with WinNonlin Professional software (Pharsight).

In silico evaluation of immunogenicity

The potential immunogenicity of each FAST-Ig variants was evaluated in silico using an EpiMatrix score and Tregitope-adjusted EpiMatrix score (EpiVax).⁴⁴

Prothrombin time assay

We used RecombiPlasTin 2 G (Instrumentation Laboratory) as the PT reagent and CS−2000i (Sysmex) for measurement as the recommended protocol. We used commercially available control human plasma (Sysmex) for the PT assay, which was treated with each concentration of BsAb or mis-paired Ab species. Data were collected in triplicate.

Enzymatic assay for FIXa-catalyzed FX activation

We measured the conversion rate of FX to FXa in an enzymatic assay using purified coagulation factors as described previously with a slight modification.³ The assay system consisted of 1 nmol/L human FIXa (Enzyme Research Laboratories, HFIXa 1080), 140 nmol/L human FX (Enzyme Research Laboratories, HFX 1010), 4 µmol/L phospholipid (10% phosphatidylserine, 60% phosphatidyl-choline, and 30% phosphatidylethanolamine), and bispecific antibodies. FXa generation was measured at room temperature for 1 min in TBS containing 1 mmol/L CaCl₂, 1 mmol/L MgCl₂, and 0.1% w/v BSA (pH 7.6). We stopped the reaction by adding EDTA. After adding S−2222 chromogenic substrate (Cromogenix), we measured absorbance at 405 nm to determine the rate of FXa generation. Data were collected in triplicate.

Differential scanning calorimetry

Differential scanning calorimetry (DSC) experiments were performed on a MicroCal VP-Capillary DSC (MicroCal). All antibodies were dialyzed against PBS (pH 7.4) and then diluted to 1 mg/mL (Fig. S4) or 0.75 mg/mL (Fig. S8). Thermal scanning was performed at a rate of 1°C/min from 30°C to 110°C. The midpoint of the thermal denaturation (Tm) of Fab fragments were analyzed using Origin 7 software (OriginLab Corporation). The background obtained from the PBS buffer scan was subtracted from the sample measurement.

Size exclusion chromatography

SEC was performed by HPLC using a ACQUITY UPLC H-Class PLUS system (Waters) equipped with a ACQUITY UPLC Protein BEH SEC Column 200Å, 1.7 µm, 4.6 mm × 150 mm (Waters). A running buffer of 50 mmol/L Na phosphate, 300 mmol/L NaCl, 200 mmol/L Arg, 0.5% w/v NaN3, pH 6.7 was used at a flow rate of 0.35 mL/min. Ten µL of samples diluted to 0.36 mg/mL were injected. Samples were excited at 280 nm and emission at 330 nm was recorded for fluorescence (FLR) detection.

Hydrophobic interaction chromatography

Each 100 µL of sample diluted to 0.1 mg/mL in the buffer, 80% mobile phase A (25 mmol/L Na-phosphate, 1.5 mol/L (NH4)₂SO₄, pH 7.0) and 20% mobile phase B (25 mmol/L Na-phosphate, pH 7.0), was injected to TSKgel Ether−5PW 7.5 × 75 mm (TOSOH), connected to the Alliance HPLC System (Waters). The column was equilibrated with 100% mobile phase A. The samples were eluted using an inverted gradient from mobile phase A to mobile phase B over 25 min at a flow rate of 1.0 mL/min with fluorescence detection (Excitation: 280 nm; Emission: 330 nm). The temperature for column compartment was set at 25°C.

Extracellular matrix binding assay

The binding to extracellular matrix (ECM) was measured using electrochemiluminescence assay. Matrigel Basement Membrane Matrix, LDEV-free (Corning, 354234) was dispensed onto a MULTI-ARRAY 96 HB (High Bind) Plate Pack, SECTOR Plate (Meso Scale Discovery) and incubated overnight at 4°C. Three µg/mL concentration of test antibodies diluted with ACES buffer pH 7.4 (20 mmol/L ACES, 150 mmol/L NaCl, 0.05% Tween 20, and 0.1% BSA) were dispensed onto ECM-coated plate and incubated for 1 hour at 30°C with gentle shaking at 600 rpm. SULFO-Tag labeled goat anti-human antibody (Meso Scale Discovery, R32AJ–1) was used as detection antibody. The signal was detected by MESO Sector S 600 MM (Meso Scale Discovery). The data were further processed to calculate the ratio of detected signal for each sample to that of tocilizumab (assay control). Tocilizumab was used as the standard of a low nonspecific binding antibody.

Stability study after low pH exposure

Each antibody was dialyzed to histidine buffer (20 mmol/L Histidine-HCl pH 6.0, 30 mmol/L NaCl), and buffer pH was adjusted to pH 3.2 or 3.6 by mixing with citric acid solution. For control, ultrapure water was added instead of citrate buffer to prepare pH 6.0 sample. Each antibody’s concentration was around 2.1–2.4 mg/mL. These samples were placed at room temperature for 3 hours. Then, samples were neutralized with 1 mol/L Tris buffer. Each antibody’s concentration was around 1.8–2 mg/mL. The final measured pH of the solution was around 7.4. Half of the neutralized samples were preserved at −80°C until experimental use and the other half were incubated at room temperature for 24 hours then preserved at −80°C until experimental use. The increased hydrophobicity of low-pH buffer-exposed antibodies was measured by SEC analysis using a Prominence HPLC (Shimazu) equipped with a TSKgel G3000SWXL Column, 5 µm, 7.8 mm × 300 mm (Tosoh Bioscience). A running buffer of 50 mmol/L Na phosphate, 300 mmol/L NaCl, 1 mg/L 8-Anilino-1-naphthalenesulfonic acid (ANS), pH 7.0 was used at a flow rate of 0.5 mL/min. For endpoint measurement, samples were excited at 380 nm and emission at 510 nm was recorded for fluorescence (FLR) detection in addition to the UV detection at 280 nm. The FLR/UV ratio of total peak area in each sample was calculated as an indicator of increased hydrophobicity. Stability of IgG for human FcRn binding after low-pH buffer exposure was evaluated with SPR using a Biacore T200 system (Cytiva). The CaptureSelect™ Human Fab-lambda Kinetics Biotin Conjugate (Thermo Fisher Scientific) was immobilized onto a Series S Sensor Chip SA (Cytiva). Then, NXT007 in the ART-S3 or dKiH_v14 format was captured in an amount of approximately 400 RU. Then, 0 and 300 nmol/L recombinant human FcRn (In-house prepared) dissolved in running buffer (50 mmol/L sodium phosphate, 150 mmol/L NaCl, 0.05% v/v surfactant P20, pH 6.0) was injected at a flow rate of 10 µL/min to monitor the association phase for 120 s and the dissociation phase for 120 s at 25°C. The sensor chip surface was regenerated using 10 mmol/L Glycine-HCl, pH 1.5. The relative response binding/capture was calculated by utilizing the result plot function in the Biacore T200 evaluation software, version 3.2.1 (Cytiva). The Fc sequence of ART-S3 format was exactly same as the clinical compound, NXT007 (development code). The dKiH_v14 format only differed from ART-S3 in the mutations for heterodimerization of HCs.

Crystallization and structure determination

Anti-FIXa antibodies (for Fab-FIXa, Fab-FIXa-C3 preparation, part of hinge region was replaced with human IgG₁) and anti-FX antibodies (for Fab-FX, Fab FX-C3 preparation) were digested by papain or Lys-C at 35°C for several hours and the four Fabs were purified as described below. To stop the reaction, buffer A (20 mmol/L sodium acetate pH 4.7) and the Tablet Protease Inhibitor Cocktail Complex (Roche) were added and the samples were incubated on ice for several hours. Each Fab fragment was purified by cation exchange chromatography with HiTrapSPHP (Cytiva) using buffer A and buffer B (20 mmol/L sodium acetate pH 4.7 and 1 mol/L NaCl), and purified by affinity chromatography with HiTrap MabSeect SuRe (Cytiva). The Fabs were further purified by SEC using HiLoad Superdex75 (Cytiva) (SEC buffer: 20 mmol/L HEPES pH 7.1 and 100 mmol/L NaCl). The three purified Fab proteins, Fab-FIXa-C3, Fab-FX, and Fab-FX-C3, were concentrated to 10–12 mg/mL and Fab-FIXa was concentrated to 19 mg/mL for crystallization. Fab-FIXa-C3 was crystallized in 20% w/v PEG 3350, 0.2 mol/L calcium chloride dihydrate. Fab-FIXa was crystallized in 20% w/v PEG 3350, 0.1 mol/L CBTP buffer (pH 6.4). Fab-FX was crystallized in 20% w/v PEG 3350, 0.2 mol/L potassium formate. Fab-FX-C3 was crystallized in 20% w/v PEG 3350, 0.2 mol/L sodium acetate trihydrate. All crystals were obtained using the vapor diffusion method. For data collection, crystals were transferred to cryo-protectant solutions containing around 30% v/v ethylene glycol and the crystallization solutions, and then frozen in liquid nitrogen. For Fab-FIXa and Fab-FIXa-C3, the diffraction data were collected at 95 K with a beam wavelength of 1.045 Å at BL1A in Photon Factory. In Fab-FX and Fab-FX-C3, the diffraction data were collected at 100 K with a beam wavelength of 1.000 Å at BL45XU in SPring-8. All diffraction data were processed by autoPROC (Global Phasing), and all structures were determined by the molecular replacement method using PDB 5DK3 as the search model by Phaser. Model building was performed with PHENIX AutoBuild and Coot. Refinement was done by PHENIX and BUSTER (Global Phasing). The structure factors and final models were deposited into the PDB with the accession codes 8GUZ (Fab-FIXa-C3), 8GV0 (Fab-FIXa), 8GV1 (Fab-FX-C3), and 8GV2 (Fab-FX). MOE (Molecular Operating Environment) 2022.02 software was used to calculate RMSD and to identify salt bridge-forming residues.

Stable CHO cell line development

The following two plasmids were transfected into DHFR-deficient CHO cells to produce NXT007: Plasmid 1 containing the DHFR gene, the puromycin resistance gene, and one copy of the L1 gene followed by one copy of the L2 gene, and plasmid 2 containing the DHFR gene, the neomycin resistance gene, and one copy of the H1 gene followed by one copy of the H2 gene. Cells were cultured in proprietary medium and those that integrated both plasmids were selected by hypoxanthine and thymidine deprivation and using puromycin and neomycin. In addition, the DHFR gene was introduced into the two plasmids to allow the amplification of the target NXT007 genes by using MTX. Clones were single-cell sorted using fluorescence-activated cell sorting and 60 were selected after expansion and pre-screening. Monoclonal cells were used for fed-batch production culture on either a small scale (few mL) (data in Figure S13A) or large commercial production scale (>1000 L) (data in Table 4). The titer of NXT007 was measured from cell culture supernatants using Octet Red (ForteBio) (data in Figure S13A) or from protein A purified pool using HPLC (data in Table 4).

Abbreviations

ANS	=	8-Anilino-1-naphthalenesulfonic acid
ART-S3	=	asymmetric re-engineering technology – E356K/K439E
BiP	=	binding immunoglobulin protein
BsAb	=	bispecific antibody
CH1	=	constant heavy chain domain 1
CIEX	=	cation exchange chromatography
CL	=	light chain constant region
CHO	=	chinese hamster ovary
DHFR	=	dihydrofolate reductase gene
dKiHv14	=	disulfide-linked KiH variant 14
ECM	=	extracellular matrix
ELISA	=	enzyme-linked immunosorbent assay
Fab	=	antigen-binding fragment
FAST-Ig	=	four-chain assembly by electrostatic steering technology – immunoglobulin
Fc	=	fragment crystallizable
FcRn	=	neonatal Fc receptor
FIXa	=	factor IXa
FX	=	factor X
HC	=	heavy chain
HIC	=	hydrophobic interaction chromatography
KiH	=	knobs into holes
LC	=	light chain
MS	=	mass spectrometry
MTX	=	methotrexate
PBS	=	phosphate-buffered saline
PDB	=	protein data bank
pI	=	isoelectric point
PT	=	prothrombin time
RMSD	=	root mean square deviation
SEC	=	size exclusion chromatography
VH	=	heavy chain variable region
VL	=	light chain variable region

Acknowledgments

We thank our colleagues at Chugai Pharmaceutical Co., Ltd. and Chugai Research Institute for Medical Science, Inc. We thank Ryota Kitada for assisting with the CIEX analysis of a large number of samples and Haruna Shinba for constructing plasmids and confirming their quality. We thank Shiho Ohtsu for analyzing physicochemical properties and Noyuri Imai for the SPR analysis. We thank Miho Mamiyoda and Shotaro Watanabe for providing mixtures of the plasmids used in transfections. We thank Yui Kaneta for assisting with the development of the stable CHO cell line. We thank Jacob Davis for English language editing. This work was fully supported by Chugai Pharmaceutical Co. Ltd.

Disclosure statement

All the authors are current employees of Chugai Pharmaceutical Co., Ltd. TI and TK are named as inventors of the patent, “Antigen-binding molecule having regulated conjugation between heavy-chain and light-chain” (WO/2013/065708). HK, KY, YT, and TI are inventors of the patent, “Multispecific antigen-binding molecule having blood coagulation factor VIII (FVIII) cofactor function-substituting activity, and pharmaceutical formulation containing said molecule as active ingredient” (WO/2019/065795). FAST-Ig is a trademark of Chugai Pharmaceutical Co., Ltd. These facts do not alter the authors’ adherence to all the mAbs policies on sharing data and materials.

Supplementary material

Supplemental data for this article can be accessed online at https://doi.org/10.1080/19420862.2023.2222441

Additional information

Funding

This work was fully supported by Chugai Pharmaceutical Co., Ltd. Funding to pay the Open Access publication charges for this article was provided by Chugai Pharmaceutical Co., Ltd.