Exploring molecular determinants and pharmacokinetic properties of IgG1-scFv bispecific antibodies

ABSTRACT

Bispecific antibodies (BsAbs) capable of recognizing two distinct epitopes or antigens offer promising therapeutic options for various diseases by targeting multiple pathways. The favorable pharmacokinetic (PK) properties of monoclonal antibodies (mAbs) are crucial, as they directly influence patient safety and therapeutic efficacy. For numerous mAb therapeutics, optimization of neonatal Fc receptor (FcRn) interactions and elimination of unfavorable molecular properties have led to improved PK properties. However, many BsAbs exhibit unfavorable PK, which has precluded their development as drugs. In this report, we present studies on the molecular determinants underlying the distinct PK profiles of three IgG1-scFv BsAbs. Our study indicated that high levels of nonspecific interactions, elevated isoelectric point (pI), and increased number of positively charged patches contributed to the fast clearance of IgG1-scFv. FcRn chromatography results revealed specific scFv-FcRn interactions that are unique to the IgG1-scFv, which was further supported by molecular dynamics (MD) simulation. These interactions likely stabilize the BsAb FcRn interaction at physiological pH, which in turn could disrupt FcRn-mediated BsAb recycling. In addition to the empirical observations, we also evaluated the impact of in silico properties, including pI differential between the Fab and scFv and the ratio of dipole moment to hydrophobic moment (RM) and their correlation with the observed clearance. These findings highlight that the PK properties of BsAbs may be governed by novel determinants, owing to their increased structural complexity compared to immunoglobulin G (IgG) 1 antibodies.

KEYWORDS:

• Biophysical assessment
• bispecific antibody
• FcRn
• in silico developability
• pharmacokinetics (PK)

Introduction

Bispecific antibodies (BsAbs) are therapeutic modalities that can recognize two different epitopes of an antigen or two antigens with a single molecule, allowing them to interfere with multiple disease pathways.¹ Given that many disorders display multiple mechanisms, BsAbs have the potential to provide effective therapeutic options to patients compared with natural antibodies and other therapeutic entities that modulate the activity of a single target.^1–3 There are several BsAbs currently approved and numerous in advanced stages of clinical development for the treatment of disease in different therapeutic areas including oncology, inflammation, respiratory, ophthalmology and cardio metabolic diseases.⁴

Despite their advances and potential benefits, the development of BsAbs as successful medicines has been slow. While the original concept of bispecific antibodies was described more than 50 y ago, the first bispecific antibody catumaxomab was not approved until 2009 and to date only 14 bispecific antibodies have been approved for therapeutic use.^5,6 The slow clinical success of BsAbs can be attributed to several factors, including increased structural complexity leading to greater challenges in their CMC properties, and in their pharmacokinetic (PK) profiles.^7,8

PK of mAbs is characterized by a slow systemic clearance and a long terminal elimination half-life, which is primarily governed by their ability to bind to and recycle through the neonatal Fc receptor (FcRn)-mediated recycling pathway.⁹ The FcRn-mediated recycling mechanism involves the nonspecific pinocytosis of IgG and albumin into cells, their binding to FcRn in acidic endosomes, and their subsequent recycling back to the cell surface at neutral pH. The pH dependency is key for biological function because FcRn binds to IgG with high affinity at about pH 6 but with weak or no binding at the physiological pH of 7.4.¹⁰ FcRn is a major histocompatibility complex class I-like molecule consisting of a large subunit p51 (with α₁, α₂, and α₃ domains) associated with a β₂-micorglobulin (β₂m). Two FcRn molecules bind to a single IgG with 2:1 stoichiometry,¹¹ and this binding of FcRn to IgG is mainly mediated by α₂ and α₃ and β₂m domains, whereas, the CH₂-CH₃ domains of IgG binds to FcRn.^12–14

Observations of rapid clearance of several therapeutic monoclonal antibodies (mAbs) have increased interest in exploring clearance mechanisms that may affect mAb PK. Several attributes of mAbs, including surface charge, target binding, off-target binding (specific or nonspecific), in vivo stability, pH-dependent FcRn binding, and degree and type of N-glycosylation, have been associated with faster clearance.^15–18 Previous studies have shown that charged-mediated Fab-FcRn interaction can disrupt the FcRn-mediated IgG recycling mechanism, translating into faster clearance.^14,19,20 Further non-specificity mediated by charged and hydrophobic interactions can be a key factor for faster clearance.^17,21 For instance, the link between positive charge on the variable domain (Fv) of mAbs and rapid nonspecific clearance was observed, possibly due to increased pinocytosis by endothelial cells and immune cells (e.g., monocytes, macrophages). The increased pinocytosis is likely mediated by increased interaction between positively charged antibody and negatively charged cell surfaces. Multiple studies have demonstrated unfavorable PK profiles of mAbs with high positive charge, which can be significantly improved by engineering to eliminate, disrupt, or relocate positive charges or balance them with, for example, negative charges.²² The PK of BsAbs may involve some of the same specific and nonspecific clearance mechanisms that govern mAb PK. Recent studies have aimed to understand the reasons for faster clearance and non-linear PK observed with BsAbs.^7,8

Here, we present the evaluation of three BsAbs that showed differential clearance in a PK study in cynomolgus monkeys. The BsAbs are all in the “Doppelmab format”²³ on an IgG1 backbone with Fab arms targeting different antigens appended to the same scFv at the C-terminus (Figure 1a). We hereby refer to them as IgG1-scFv. There were low levels of tissue target expression and low levels of free circulating target antigens for all three molecules in the cynomolgus monkeys, which allowed for straight-forward evaluation and interpretation of PK profiles.

Figure 1. Three bispecific antibodies showed different half-life and clearance in cynomolgus monkey. (a). Schematic representation of bispecific antibodies IgG1-scFv used in this work. The BsAbs contain identical scFv (red) which are linked by a short polypeptide linker to the C-terminus of the IgG1. They differ in the fv of fab binders (light blue). (b). Half-life and clearance of the IgG1-scFv a and IgG1-scFv B measured from 10 mg/kg and IgG1-scFv measured from 8 mg/kg IV administration in cynomolgus monkey, respectively.

In our studies, we aimed to investigate the molecular factors governing the differences in the PK profiles, focusing on biochemical, physiochemical, and in silico properties of the three BsAbs. The results reveal that a combination of factors, including isoelectric point (pI), FcRn interaction, and nonspecific binding by charged and hydrophobic surfaces, can contribute to faster clearance. Interestingly, in addition to previously published Fab-FcRn interactions, we observed that scFv-FcRn binding also potentially influences PK by disruption of the IgG1-FcRn recycling mechanism.

Results

Bispecific antibody design

The bivalent, bispecific antibodies used in this work are categorized according to their structures based on the International Nonproprietary Name nomenclature as IgG1-scFv (type 1).²⁴ A schematic structure of these IgG1-scFv Doppelmabs is illustrated in Figure 1a. The three BsAbs all contain identical Fc regions, including a C_H1, a C_H2, and a C_H3, the latter of which is C-terminally fused to the scFv specific for the same antigen for all three molecules. Furthermore, they differ in the Fv of the Fabs (hereby referred to as Fv_(Fab)), which bind to different antigens. The germlines of Fab of the three BsAbs are V_ĸ3–11 and V_H3–23 for IgG1-scFv A, V_ĸ3–20 and V_H5–51 for IgG1-scFv B, V_ĸ1–39 and V_H1–69 for IgG1-scFv C, respectively. The scFv contains V_ĸ3–19 and V_H3–21 domains linked by a 20 amino acid intramolecular linker (GGGGS)4. Furthermore, the IgG1 Fc bears the L234A and L235A (LALA) substitutions to prevent the interaction with Fc gamma receptors and C1q binding, precluding ADCC and CDC. The BsAbs were produced in transiently transfected Chinese hamster ovary (CHO) cells and highly purified to >97% monomer final purity.

Summary of the PK data

The PK studies in cynomolgus monkeys of three BsAbs demonstrated varying clearance. PK studies were conducted in cynomolgus monkeys with a single intravenous (IV) dose of 10 mg/kg in three animals per dosing group for IgG1-scFv A and IgG1-scFv B and with a single IV dose of 8 mg/kg in three animals per dosing group for IgG1-scFv C. The data revealed slow clearance for IgG1-scFv B of 4.3 mL/d/kg, moderate clearance for IgG1-scFv A (11.8 mL/d/kg), and fast clearance for IgG1-scFv C (63.8 mL/d/kg, Figure 1b). With regard to serum half-life, these BsAbs show a similar trend as was observed for clearance, indicating that IgG1-scFv A and IgG1-scFv C had some intrinsic liabilities that resulted in their increased clearance and shorter half-life. The reason for this discrepancy between the clearance of the three BsAbs is explored in the sections below.

Small differences in FcRn binding of IgG1-scFv molecules do not correlate with clearance

A previous study of two IgG1-scFv molecules in which the Fv and scFv were interchanged indicated inefficient FcRn recycling of one IgG1-scFv molecule, causing differential PK profiles.^8,25 Based on this observation, we investigated the influence of FcRn binding on the clearance of the three IgG1-scFv molecules. We used retention times from FcRn chromatography to characterize how effectively the three molecules undergo FcRn-mediated recycling.^15,17,26 The FcRn chromatography was performed with a linear pH gradient monitoring pH-dependent FcRn binding properties relative to a reference IgG1 mAb, with germlines V_ĸ3 and V_H3 containing L234A and L235A substituents within the Fc. The results of the FcRn chromatography showed differences in retention times between the IgG1-scFv molecules (Figure 2a). The retention time for IgG1-scFv A was comparable to the reference molecule, but higher for IgG1-scFv B and IgG1-scFv C. Relative to the reference molecule, the retention times of the IgG1-scFv B and IgG1-scFv C were similar.

Figure 2. FcRn chromatography of the bispecific antibodies IgG1-scFv revealed additional fab and scFv interactions with FcRn of the bispecific antibodies IgG1-scFv. A. Comparison of FcRn chromatography elution profiles of the three bispecific antibodies IgG1-scFv. B. Fc-scFv molecule was compared to bispecific antibodies IgG1-scFv to investigate the influence on fab and scFv on FcRn binding. C. Position of fv influences FcRn interaction. Parental IgG1 contain the same fv in fab like the fv in fc-scFv derived from IgG1-scFv.

To investigate which molecular features of the IgG1-scFv molecules influence the FcRn interaction, the three IgG1-scFv were digested above the hinge region with the cysteine protease GingisKHAN® under native conditions to obtain the Fab and Fc-scFv. The reference IgG1 mAb was also digested to obtain an identical Fc. The successful digests were confirmed by intact liquid chromatography-mass spectrometry (LC-MS) (Table S1). The FcRn chromatography results of the digested IgG1-scFv molecules showed that the Fab did not interact with FcRn chromatography, and the Fc-scFv of the three IgG1-scFv molecules bound to the FcRn column and had similar elution profiles (Figure 2b). The retention times of the Fc-scFv were increased compared to Fc alone of the reference molecule.

Also, while IgG1-scFv B and IgG1-scFv C had higher retention times relative to their corresponding Fc-scFv, IgG1-scFv A had a lower retention time than its corresponding Fc-scFv. This observation of scFv-FcRn interaction for IgG1-scFv molecules is consistent with the observation of retention times shift between IgG1-scFv molecules and their respective parental IgG1 (Figure S1). These results suggest that both the Fab and the scFv can interact with FcRn and alter the elution profiles. Since all three BsAbs share an identical scFv sequence, these observations suggest that the Fab can have repulsive or increased FcRn interaction, and the scFv can also increase the FcRn interaction in addition to the typical Fc-FcRn interaction.

To investigate whether the Fc C-terminus position of the scFv plays a role in scFv-FcRn interaction, a IgG1-scFv derived from GingisKHAN® digestion was compared with a parental IgG1 containing the same Fv sequence of the scFv in the three BsAbs. Using FcRn chromatography, the digested Fc-scFv eluted slightly later than parental IgG1 (Figure 2c). The discrepancy in retention times observed for Fab-FcRn and scFv-FcRn interaction indicate that interaction of scFv-FcRn is specific.

Molecular stability and structural integrity of IgG1-scFv do not contribute to differential clearance

In addition to the FcRn-mediated recycling, instability of antibodies can be linked to denaturation and aggregation that can impair the PK properties of antibodies.^27–29 To investigate whether these BsAbs have poor stability profiles, their thermal stability and serum stability were evaluated. Thermal stability was measured using differential scanning calorimetry (DSC). Figure S2 shows a representative DSC thermogram with data from curve fitting of IgG1-scFv B. The melting temperature (T_m) values of the scFv and Fabs were in the ranges of 54.3–55.5°C and 75.7–79.1°C, respectively. The Fab and scFv Tms were therefore highly comparable across the three molecules (Table S2). The T_m analyses indicate that all molecules have similar thermal stabilities.

The serum stability of the bispecific antibodies was assessed by incubation over 4 d at 37°C in 90% mouse serum and compared with incubation of the BsAbs in phosphate-buffered saline (PBS) (Table S3). The stability of the BsAbs was assessed based on interaction with their antigens of the Fab and the scFv using antigen captured/immobilized biolayer interferometry. The three BsAbs revealed similarly good stability in mouse serum (Table S3). This data together suggest that the thermal stability or serum stability may not be a key contributor to the observed differences in clearance and half-life of the three BsAbs.

Significant differences in surface charge of the IgG1-scFv molecules

Previous literature has shown nonspecific interactions mediated by charged or hydrophobic interactions of an antibody can contribute to faster nonspecific clearance.^7,8,30–32 To identify potential biophysical properties that affect nonspecific interactions, a set of analytical methods was used, comprising analytical hydrophobic interaction chromatography (aHIC), heparin chromatography, and surface plasmon resonance (SPR)-based binding to positively and negatively charged surfaces to compare the three IgG1-scFv molecules (Figure 3).

Figure 3. Comparative analysis of nonspecific interactions of IgG1-scFv relative to clearance in cynomolgus monkey. Correlation of experimental HIC, positively and negatively charged surface SPR and heparin chromatography, with clearance in cynomolgus monkey, respectively. For negatively and positively charged SPR, nonspecific binding below 50 RU were considered as not relevant because there were low levels of nonspecific binding.

aHIC analyses characterize hydrophobic interaction of the IgG1-scFv molecules with the hydrophobic materials on the column under high to low salt concentrations. The analyses revealed that IgG1-scFv A and IgG1-scFv B had higher hydrophobicity than IgG1-scFv B with increased retention times.

Heparin chromatography was previously used to investigate if antibodies interact with the negatively charged heparin using a salt gradient with increasing ionic strength.²¹ Alternatively, negatively charged protein SPR that immobilizes trypsin inhibitor also indicates nonspecific interaction. A stronger interaction to negatively charged surface by SPR and negatively charged heparin for IgG1-scFv C, but not for IgG1-scFv A and IgG1-scFv B was observed.

Next, the interaction of the BsAbs with positively charged surface by SPR was evaluated. No association with positively charged surface of IgG1-scFv A and IgG1-scFv B was detected, whereas IgG1-scFv C bound marginally.

The results shown in Figure 3 suggest that differences in hydrophobicity and binding to negatively charged surfaces exist between the molecules that could potentially affect their circulating half-life and clearance.

In silico descriptors of Fv_(Fab) and scFv may predict systemic clearance

Following the observation of differences in hydrophobicity and nonspecific interaction between the three BsAbs, the molecular surface properties of Fv_(Fab) and scFv were investigated by calculating nonredundant descriptors using molecular dynamics (MD) and homology-based molecular model based on Ahmed et al. and Licari et al.^33,34 (Figure 4a). The simulations were initially validated by comparison of simulated HIC values and experimental HIC values (Figure 4b). Similar results were obtained by both methods.

Figure 4. In silico descriptors of the Fv_(Fab) and scFv of IgG1-scFv molecules. (a). Illustration of surface properties of the three molecules. The green highlights hydrophobic patches, blue highlights positively charged patches and red shows negatively charged patches. (b). Hydrophobic and charged descriptors that describe the surfaces. Note the pI of the full-length IgG1-scFv molecules were calculated based on their primary sequences, whereby the pI of Fv_(Fab) and scFv were obtained based on their homology model structure. (c). Correlation of pI of fv of fab and scFv of the three IgG1-scFv BsAbs colored according to clearance in cynomolgus monkeys (slow <8 ml/d/kg in green, moderate > 9 and <15 ml/d/kg in orange, fast <15 ml/d/kg in red). RP = ratio of charged to hydrophobic surface patches, RM = ratio of dipole moment to hydrophobic moment, HI = hydrophobic imbalance, pI = isoelectric point, pro_patch_pos = positively charged patche(s), pro_patch_neg = negatively charged patche(s).

Non-redundant descriptors that describe charged and hydrophobic properties, as described in Ahmed et al. were applied. These descriptors were the ratio of charged to hydrophobic surface patches (RP), the ratio of dipole moment to hydrophobic moment (RM), hydrophobic imbalance (HI), and structure-based pI. No clear relationships between RP and HI with the clearance of the three BsAbs were observed. The analysis of charged and hydrophobic residues described as RM was less well distributed for Fv_(Fab)-A and Fv_(Fab)-C than for Fv_(Fab)-B and scFv (Figure 4b).

In contrast to the pI values of full-length IgG1-scFv, the pI values of Fv_(Fab) and scFv varied significantly. Whereas the pI value of the Fv_(Fab) (pI = 8.3) in IgG1-scFv B was similar to the pI of the scFv (pI = 7.7), the pI value of the Fv_(Fab) (pI = 5.3) was lower than the pI of the scFv (pI = 7.7) in IgG1-scFv A and the pI of the Fv_(Fab) (pI = 9.6) was higher than the pI of the scFv (pI = 7.7) in IgG1-scFv C (Figure 4b).

To investigate the basis for the high pI of Fv_(Fab) in IgG1-scFv C and the low pI of Fv_(Fab) in IgG1-scFv A, the charge patches of the three Fv_(Fab) and the same scFv determined from MD simulations were compared. This revealed differences in the distribution of positively and negatively charged patches in the Fv_(Fab) and scFv. For the Fv_(Fab) of IgG1-scFv A, the negatively charged patches were increased relative to positively charged patches (477 Ų vs 354 Å).² In contrast, for the Fv_(Fab) of IgG1-scFv C, positively charged patches were significantly increased compared to negatively charged patches (772 Ų vs 202 Å).² In comparison to the scFv and Fv_(Fab) of IgG1-scFv B, the positively charged patches were slightly higher (536 Ų vs 395 Å,² and 465 Ų vs 306 Ų) (Figure 4b).

Overall, the difference between the pIs between the two binding domains seems to play a critical role in determining the clearance of antibodies, particularly when the charge of one of the binders is high (pI > 9.0) as observed for IgG1-scFv C (Figure 4c).

Molecular dynamics simulation reveals specific interactions of scFv of IgG1-scFv with FcRn

To investigate the hypothesis that the scFv of IgG-scFv molecules interacts with FcRn, we conducted an all-atom MD simulation to elucidate the molecular basis of this interaction using IgG1-scFv B as a representative model (Figure 5). At the beginning of the simulation, the relaxed, energy minimized, and elongated conformation of IgG1-scFv B shows no contact between the Fab and scFv with FcRn (Figure 5a). However, after 300 ns of simulation, we observed a more compact conformation of the complex. This conformation presents the Fab interacting with FcRn, confirming previous findings.^19,20 In addition, we identified a new specific interaction between both scFv and FcRn (Figure 5b). A more detailed structural analysis during the 300 ns MD simulation revealed that the Fv light chain of the scFv and the (GGGGS)4 linker between Fc and scFv interact with the large p51 subunit of FcRn, specifically with parts of the α3 and α4 helix secondary structural elements, and the loop connecting them. Similarly, to the Fab:FcRn interactions, only the Fv light chain participates in the scFv:FcRn interaction with key contributions from L-CDR1 (L1) and L-CDR3 (L3) stabilizing the scFv and FcRn interface (Figure 5c).

Figure 5. All-atom molecular dynamics simulation of IgG1-scFv with FcRn.

IgG1-scFv B serves as a representative for BsAbs used in this study. (a). and (b). depict the models of IgG1-scFv bound to FcRn at the beginning and after 300 ns of all-atom MD simulation, respectively. (b). The box indicates the part of the molecule shown in C. C. Close-up view of the interaction between the scFv and the FcRn. The complementarity-determining regions (CDRs) of the light chain of the scFv are indicated as L1, L2, and L3. The (GGGGS)4 linker between Fc and scFv (gray) as well as L1 and L3 contribute to the interaction. Interestingly, the Fv β-strands of the Fab packs against the β-sheets of FcRn p51 sub-unit. (d). The change of solvent-accessible-surface area (SASA) of all FcRn and scFv combined shows a reduction of ~25 nm² SASA over time indicating coverage of SASA by specific interactions. (e). The increase in contact surfaces of the individual FcRn´-scFv´ and FcRn-scFv complexes was depicted. The contact surface of FcRn´-scFv´ stabilizes in the last 100 ns around 4 nm,² a smaller value compared to the FcRn:scFv contact surface that fluctuates around 6 nm.² (f). The distances between two example residues in the scFv L1 and FcRn were analyzed. Over the course of 300 ns, the simulation shows the change of the centers of mass-distances between Phe154´ (FcRn´) and Ser24´ (scFv´), and Lys143 (FcRn) and Ser24 (scFv).

A more quantitative assessment supported the findings by monitoring changes in the solvent-accessible surface area (SASA) throughout the simulation (Figure 5d,e). Figure 5d illustrates the change in SASA for both FcRn and scFv combined. Throughout the 300 ns simulation, the total SASA steadily decreases from approximately 570 nm² to around 535 nm,² resulting in a reduction of 35 nm.² The loss of SASA leads us to infer that the surface is no longer accessible to the solvent, and thus, is involved in binding. Additionally, this data encompasses the changes within each domain, emphasizing the molecular alterations that contribute to a more compact structure.

Next, we were interested to see if both scFvs exhibit similar behavior in terms of changes in SASA over the course of simulation. Figure 5e displays the formation of the scFv’:FcRn´ and scFv:FcRn contact surfaces throughout the simulation. In the initial phase of the simulation (up to ~150 ns), both scFvs exhibit either no contact or only short contact times (~10–15 ns) with their respective FcRn. In contrast, the contact surfaces of both scFv and FcRn stabilize at approximately 4 and 6 nm² during the final 100 ns. Interestingly, the scFv’:FcRn’ contact surface of 4 nm² is smaller compared to the scFv:FcRn, which fluctuates around 6 nm.² This difference may be attributed to the additional Fab:FcRn interaction. The stabilization of the scFv contact surfaces during the simulation allows us to conclude that scFv interacts with FcRn. The observed contact surface areas are consistent with those typically found in protein–protein interactions. We continued to explore the distance between scFv and their corresponding FcRn. Our residue contact analysis throughout the simulation identified recurring interactions on the contact surface between scFv and FcRn. Figure 5f illustrates the changes of the center of mass distances for two representative residue interaction pairs, Ser24´-Phe154´ and Ser24-Lys143, which are located in the scFv´:FcRn´ and scFv:FcRn complex, respectively. After approximately 150 ns of simulation, the residue contact distances remained nearly constant at around 0.6 and 1.5 nm, indicating stabilized interaction sites. This observation aligns with our previously mentioned findings.

Discussion

Many BsAbs have shown unfavorable PK that necessitates the assessment of undesired molecular properties with the aim of lowering the risks of further development.^7,8 In this study, the correlation between PK behavior and molecular properties of three bispecific IgG1-scFv antibodies was investigated. During a cynomolgus monkey PK study, the three IgG1-scFv molecules showed differential clearance, ranging from slow to fast.

It has been reported that thermal stability can be an indicator of in vivo aggregation that contributes to poor PK.¹⁸ Following the observation that the scFv of IgG1-scFv molecules had lower T_m values than the Fab, it is reasonable to propose that thermal stability is a factor impacting clearance of the three BsAbs. However, since all molecules used a common scFv, the T_m of the scFv was similar between the three BsAbs (Table S2) and a previous report suggests that it is common for scFv to have lower T_m than Fab.³⁵

Another important factor is the stability of the BsAbs containing scFv in serum and PBS.³⁶ To determine the serum stability of the three BsAbs, they were incubated in 90% mouse serum or in PBS, and their degradation over time was monitored by their binding to the target antigen(s). The data indicated that neither incubation in mouse serum nor in PBS in a time course over 4 d resulted in a loss of binding, suggesting that all molecules were stable (Table S3).

Efficient FcRn-mediated IgG1 recycling contributes to slower clearance of mAbs.⁹ To have longer serum persistence, it is essential to have efficient release from FcRn on the cell surface after recycling. Previous reports on IgG1 molecules revealed that charge-mediated Fab-FcRn interaction can disrupt the FcRn-mediated recycling at physiological pH.^19,20 Using FcRn chromatography, some differences in FcRn interaction between the IgG1-scFv molecules were observed. However, the similar retention times of the full length IgG1-scFv B, which had the slow clearance, and IgG1-scFv C, which had faster clearance (Figure 2a,b), suggest that the influence on unfavorable PK is marginal. In contrast to traditional mAbs, the variation in FcRn interaction of the full-length BsAbs containing identical Fc cannot be solely described by charge-mediated Fab-FcRn binding. Increased FcRn column retention times of the cleaved Fc-scFv and different retention times between IgG1-scFv and their parental IgG1 molecules indicate a scFv-FcRn interaction (Figure 2c and Figure S1).

Our MD simulation data provide some additional insights into the scFv-FcRn and Fab-FcRn interactions (Figure 5). Previous studies have reported Fab-FcRn interactions whereby the light chain of the Fab contributes to additional binding to FcRn.^19,20 Consistent with these observations, our MD simulations show that the light chain of Fab is capable of binding to FcRn. Additionally, the simulations show that 1) there is a scFv-FcRn interaction and 2) the possibility of both the scFv and Fab binding to the same FcRn molecule, which can, taken together, disrupt the FcRn-mediated recycling and lead to faster clearance. However, for the purpose of this study, only one trajectory per molecule is provided, which merely suggests the plausibility for such an scFv-FcRn interaction.

Higher hydrophobicity of IgG molecules has been associated with faster clearance.^30,32,37 Compared with published HIC results from 137 IgG antibodies, the observed HIC results for the BsAbs were only marginally different (Figure 3) and in a narrow hydrophobicity range. In addition, another in silico measure of hydrophobicity is the hydrophobicity imbalance (described as HI).³³ However, no correlation with HI and PK was observed for the three BsAbs (Figure 4).

Some cases of faster clearance of IgG1 antibodies can be attributed to extremely low or high pI of the Fv_(Fab). For example, Liu et al. showed that positively and negatively charged Fv_(Fab) variants of trastuzumab can have faster clearance.³⁸ Furthermore, the work of Grinshpun et al. reported that the pI of Fv_(Fab) ranging from 6.7 to 9.05 allowed the prediction of 64% of fast clearance of IgG1 molecules.³⁰ Investigating the pI range, we were able to flag the two IgG1-scFv molecules with faster clearance. The in silico MD simulation results showed a high pI of the Fv_(Fab) of IgG1-scFv C with a high level of positively charged patches (Figure 4b), which is consistent with the observed experimental result where there were higher levels of nonspecific interactions with negatively charged protein surface by SPR and by heparin chromatography (Figures 3 and 4).

Using this combined data, we postulate that IgG1-scFv C has positively charged patches that can induce nonspecific interactions with negatively charged extracellular matrix or cell membrane surfaces, which in turn result in internalization into cells via pinocytosis and ultimately into faster nonspecific clearance. In contrast, IgG1-scFv A, which also had an increased rate of clearance, had a lower pI for the Fv of Fab and no nonspecific interaction, was observed in our assays, indicating a different clearance mechanism for IgG1-scFv A than IgG1-scFv C, possibly target-mediated drug disposition (TMDD).

Further, in silico MD simulation results reveal additional parameters that could be used to predict the faster clearance of BsAbs. For example, trends in the ratio of dipole moment to hydrophobic moment (RM) and charge imbalance may correlate with faster clearance for IgG1-scFv A and IgG1-scFv C. To determine if we could use these values to predict faster clearance of BsAbs, we looked at RM and charge imbalance of 11 additional IgG1-scFv (Table S4). Analysis of these data revealed that molecules with 27% similar or higher RM values and 50% charge imbalance with 1 unit or larger differences in pI were a predictor of poor PK and faster clearance.

In addition to the factors that have been discussed here, there could potentially be other factors in play, including TMDD, particularly for IgG1-scFv A, where nonlinear PK was observed at high dose. Analysis of this aspect was beyond the scope of this study, which focuses mainly on the molecular attributes and biophysical properties of the formats themselves.

Overall, identification of the molecular characteristics of IgG1-scFv bispecific antibodies that are responsible for unfavorable PK profiles is challenging. Our results, along with previous published data, suggest that there is no single feature that can be pinpointed as a key driver for unfavorable PK. Rather, it is more a combination of multiple features. Improvements to clearance of IgG1-scFv can be achieved by 1) an efficient FcRn-mediated recycling of the IgG1-scFv without any impacts from Fab-FcRn and likely also scFv-FcRn interactions; 2) balanced hydrophobicity and pI, which in turn can reduce aggregation and nonspecific interactions; and 3) good thermal and serum stability that prevents aggregation under physiological conditions. In addition to these empirical parameters, we also identify additional in silico parameters, including the distribution of charged and hydrophobic residues (RM) and the charge imbalance due to the different pIs of Fv of Fab and scFv in IgG1-scFv that can be used to predict and subsequently engineer the properties of such BsAbs. It should be noted that these parameters are not fully validated and if these additional factors are to be used as predictors of disfavored PK of IgG1-scFv bispecific antibodies, further studies to investigate the impact of these factors and to understand their correlations with the potential underlying mechanism of clearance will be critical.

Materials and methods

Antibodies

The bispecific antibodies IgG1-scFv investigated in this study were produced in-house by mammalian cell culture technology using CHO cells and purified with a standard procedure as described previously.³⁶ All bispecific antibodies are similar in size with a molecular weight between 195 and 205 kDa. Sample quality was ensured by quantification using analytical size exclusion chromatography passing the acceptance criteria (>97%).

PK studies

The three bispecific antibodies, IgG1-scFv A, IgG1-scFv B, and IgG1-scFv C, were evaluated in treatment-naïve male cynomolgus monkeys of Chinese origin. A single intravenous (iv) dose of 10 mg/kg was administered for IgG1-scFv A and IgG1-scFv B, while a single iv dose of 8 mg/kg was given for IgG1-scFv C. Serum concentrations of IgG1-scFv A and IgG1-scFv B were monitored over a 6-week period, and IgG1-scFv C concentrations were observed over a 3-week period in the cynomolgus monkey PK study. Antigen-capture ELISA was used to determine the serum concentrations using human antigen-capture reagent along with anti-human IgG-horseradish peroxidase (HRP, Cat # A0170, Sigma Aldrich, Germany) detection to quantify IgG1-scFv levels. Both PK and anti-drug antibody (ADA) analyses were conducted, and ADA-positive samples were excluded from the PK analysis.

FcRn affinity chromatography

A commercially available FcRn Affinity Column Gen2 (1 mL, Cat# 09430857001, Roche Diagnostics Deutschland GmbH, Mannheim, Germany) was pre-equilibrated with 20 mM Bis-Tris 140 mM NaCl, pH 5.8 over 20 min at 0.5 mL/min. Fifty-microgram diluted sample was injected to the FcRn column at a column temperature of 30°C. A linear gradient from 25% to 100% 20 mM Tris 140 mM NaCl pH 8.8 in 30 min was applied eluting the sample.

Heparin chromatography

A commercially available 5 mL heparin column (Cytiva, USA) was equilibrated with 50 mM Tris pH 7.4 and a flow rate of 0.8 mL/min. Forty-microgram diluted sample was looped onto the heparin column. Following the injection, the protein was eluted with a salt gradient of 50 mM Tris pH 7.4 in 50 mM Tris, pH 7.4, 1 M NaCl over 16.5 min: 0–1 min (0%), 1–1.5 min (sample was looped), 1.5–18.0 min (0–55%), 18.0–26 min (55%) and then 26–31 min (2 min/mL, 0%). The heparin chromatography data was reported as relative retention time, in which the retention time of the sample was divided by retention time of the reference control mAb.

GingisKHAN® digestion

The digestion of IgG1-scFv molecules and IgG1 with the cysteine protease GingisKHAN® (Cat# B0-GKH-020, Genovis, MA, USA) was performed and confirmed by intact LC-MS analysis using BioAccord LC-MS system. A BioAccord LC-MS system (Waters Corp., Milford, MA, USA) was operated in positive ion mode consisting of time of flight (Tof)-based ACQUITY RDa Mass detector. The reconstitution and digestion were applied according to the manufacturer’s instructions. GingisKHAN® was reconstituted in 200 µL water, and GingisKHAN® reducing agent was freshly reconstituted in 50 µL water. For each reaction, 1 mg/mL DM1 or IgG1 with 1 unit GingisKHAN®/1 µg antibody, 1/10 (v/v) GingisKHAN® reducing agent were combined in PBS with a final volume of 200 µL and then incubated for 1 h at 37°C. The reaction was injected onto a reverse phase liquid chromatography (RPC) using BioResolve Reverse Phase (RP) mAb Polyphenyl Column (450 Å, 2.7 μM, 2.1 × 50 mm, Waters Corp., Milford, MA, USA). The RPC method contains a gradient from 15% to 30% acetonitrile 0.1% (v/v) formic acid and water 0.1% (v/v) formic acid over 8 min. The data were analyzed by Unify software (Waters Corp., Milford, MA, USA).

Hydrophobic interaction chromatography

For HIC, the Waters Acquity UPLC H-class system was equipped with Sepax Proteomic HIC butyl-NP 1.7 column (Sepax Technologies, Inc., DE, USA). Twenty-four ug sample in 1 M ammonium sulfate was injected onto the HIC column, which was pre-equilibrated in 1 M ammonium sulfate, 0.1 M sodium phosphate, pH 6.5. Elution was achieved in 16 min from 1 M ammonium sulfate, 0.1 M sodium phosphate, pH 6.5 to 0.1 M sodium phosphate, pH 6.5. The elution profile was analyzed by Empower 3 software (Waters Corp., Milford, MA, USA).

Positively and negatively charged protein SPR

Standard amine-coupling according to the manufacturer’s instructions was applied to immobilize chicken egg white lysozyme (Cat # L3790, Millipore Sigma) and trypsin inhibitor type 1-S from soybean (Cat# T2327, Millipore Sigma, Germany) to Biacore T200 series S CM5 chips reaching a surface density of 3000–5000 RU, respectively. Samples were diluted to a 1 µM sample solution in running buffer 1× HBS-EP buffer (pH 7.4). The 1 µM sample solution was injected for 10 min followed by 10 min of running buffer. The data were acquired by using Biacore T200 Control Software version 2.0.1 and analyzed using Biacore T200 Evaluation Software version 3.0. The threshold for binding was at 50 RU.

In silico descriptor analyses

The homology-based molecular models of Fv_(Fab) and scFv were prepared using the Molecular Operating Environment ((MOE), 2022.02 Chemical Computing Group ULC, MontrealCanada) version 2022 with an Amber10: EHT force field, and classical MD simulations were performed using NAMD2 and NAMD3 software developed by the Theoretical and Computational Biophysics Group in the Beckman Institute for Advanced Science and Technology at the University of Illinois at Urbana-Champaign.³⁹ All models were prepared at pH 7.4 and 137 mM NaCl to mimic physiological conditions. Finally, protein properties were averaged along the last 180 ns of the trajectories. Detailed descriptions of each step can be found in Licari et al. (2023).³⁴

MD simulations of FcRn with IgG1-scFv B

A homology model for IgG1-scFv B representing the IgG1-scFv BsAbs tested in our study was generated using standard computational approaches. The complex structure of the human neonatal Fc receptor and human IgG1 Fc was obtained from the RCSB PDB database with PDB accession code: 4N0U.⁴⁰ The structure (omitting albumin) was grafted (including the glycans) onto the homology model of the IgG1-scFv BsAb using MODELLER.^41,42 The AMBER a99SB-disp force field⁴³ was used for the ternary FcRn:IgG1-scFv B complex and water, and the glycan moieties were described with the GLYCAM06j force field.⁴⁴ The titratable amino acid side chains were protonated according to their standard charge states at pH 7.4 to reflect physiological condition. The rhombic dodecahedron simulation boxes were solvated with water, counter ions added, and the total charge of the simulation box was neutralized by adding sodium and chloride ions at a concentration of 137 mM. In total, one simulation box contained about 800,000 atoms.

All-atom MD simulations were carried out with the GROMACS code (version 2021.1).⁴⁵ The interactions were implemented with CHARMM-GUI⁴⁵ using the Charmm36m force field.⁴⁶ First, the systems were energy-minimized for 200 steps, using a steepest descend algorithm. Afterwards, the systems were equilibrated for 20 ns in the NpT (isothermal-isobaric) ensemble with harmonic position restraints on all Cα-atoms (with force constants of 1000 kJ mol-1 nm-2). The stochastic velocity rescaling thermostat⁴⁷ with a coupling constant of 0.1 ps was used to keep the temperature constant at 298 K. Constant pressure at 1 bar was realized using the stochastic cell rescaling barostat⁴⁸ with a coupling time constant of 2 ps and a compressibility of 4.5 × 10–5 bar-1. The integration of the equations of motion was solved with a 2 fs time step, using the LINCS and SETTLE⁴⁹ algorithms to constrain the H-bond involving protein bonds and the internal degrees of freedom of the water molecules, respectively. Lennard-Jones 12–6 and short-range Coulomb interactions were computed up to a cutoff of 1.2 nm using a Verlet pair list with buffering, with forces smoothly switched to zero between 1.0 nm and 1.2 nm.⁵⁰ The smooth particle-mesh Ewald (PME) approach with cardinal B-spline interpolation of fourth order and a grid spacing of 0.12 nm was used to compute the long-range Coulomb interactions.⁵¹ The simulations used for analysis were performed with a simulation time of 300 ns each.

Serum stability assay

For serum stability assay, 1 mg/mL of three bispecific antibodies were incubated in the presence of 90% mouse serum (Cat # 10410, Invitrogen, USA) or in PBS, at 37°C for 4,2,1, and 0 d, respectively. Prior to the serum stability assay, the antibodies were prepared as 10 mg/mL antibody solution in PBS and the mouse serum was sterile filtrated. The samples were diluted by 1:100 in 1× PBS. The interaction of the bispecific antibodies with their antigens in 90% mouse serum or 1× PBS was measured using a Biolayer interferometry (Octet™, Sartorius, Göttingen, Germany). The Octet™ was equipped with the recombinant antigens that were immobilized onto amine reactive second Generation (AR2G) biosensors (Sartorius, Göttingen, Germany) or captured onto anti-human IgG Fc (AHC) capture biosensors (Sartorius, Göttingen, Germany). The association was performed for 600 sec, at room temperature with agitation at 1000 rpm. The time points of the responses in the sensorgrams (590 sec ±10 sec) were referenced with the serum control or PBS. The referred responses were then normalized to the respective-binding data of day 0.

Acknowledgments

The authors would like to thank Gale Hansen, Nikolai Prill, Cynthia Kenny and Joschka Bauer for their technical support, and Lars V. Schäfer for useful discussions. This work was supported by Deutsche Forschungsgemeinschaft (DFG) under Germany’s Excellence Strategy - EXC 2033 - 390677874 - RESOLV.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Supplementary material

Supplemental data for this article can be accessed online at https://doi.org/10.1080/19420862.2024.2318817

Additional information

Funding

The work was supported by the Deutsche Forschungsgemeinschaft (DFG) under Germany’s Excellence Strategy [EXC 2033 - 390677874 - RESOLV].