The origins of nonideality exhibited by monoclonal antibodies and Fab fragments in human serum

Abstract

The development of therapeutic antibodies remains challenging, time‐consuming, and expensive. A key contributing factor is a lack of understanding of how proteins are affected by complex biological environments such as serum and plasma. Nonideality due to attractive or repulsive interactions with cosolutes can alter the stability, aggregation propensity, and binding interactions of proteins in solution. Fluorescence correlation spectroscopy (FCS) can be used to measure apparent second virial coefficient (B _2,app) values for therapeutic and model monoclonal antibodies (mAbs) that capture the nature and strength of interactions with cosolutes directly in undiluted serum and similar complex biological media. Here, we use FCS‐derived B _2,app measurements to identify the components of human serum responsible for nonideal interactions with mAbs and Fab fragments. Most mAbs exhibit neutral or slightly attractive interactions with intact serum. Generally, mAbs display repulsive interactions with albumin and mildly attractive interactions with IgGs in the context of whole serum. Crucially, however, these attractive interactions are much stronger with pooled IgGs isolated from other serum components, indicating that the effects of serum nonideality can only be understood by studying the intact medium (rather than isolated components). Moreover, Fab fragments universally exhibited more attractive interactions than their parental mAbs, potentially rendering them more susceptible to nonideality‐driven perturbations. FCS‐based B _2,app measurements have the potential to advance our understanding of how physiological environments impact protein‐based therapeutics in general. Furthermore, incorporating such assays into preclinical biologics development may help de‐risk molecules and make for a faster and more efficient development process.

1. INTRODUCTION

Therapeutic antibodies show significant clinical promise but are challenging molecules to study and develop. Antibody‐derived therapeutics are rapidly becoming the predominant treatment modality for many cancers, autoimmune disorders, and other diseases (Kimiz‐Gebologlu et al., 2018; Hafeez et al., ²⁰¹⁸; Lu et al., ²⁰²⁰). As of 2021, 100 therapeutic antibodies have been Food and Drug Administration approved with 800 in the clinical pipeline and many more in earlier stages of development (Kaplon et al., 2022; The Antibody Society, ²⁰²²). Advances in protein engineering and optimization have also driven the development of nontraditional antibody platforms—such as antibody‐drug conjugates (ADCs), bispecific antibodies (bsAb), and Fc‐fusion proteins—that are also reaching the clinic, albeit in smaller numbers (Abdollahpour‐Alitappeh et al., 2019; Godar et al., ²⁰¹⁸; Mullard, ²⁰²¹). De novo–designed proteins are entering the pipeline and the clinic as therapeutic and vaccine platforms (Bonadio & Shifman, 2021; Quijano‐Rubio et al., ²⁰²⁰; Song et al., ²⁰²²). While therapeutic proteins are more targeted and potent than most traditional small molecule therapies, their development is time‐consuming and expensive. This inefficiency reflects, in large part, major limitations in our understanding of the pharmacokinetics and pharmacodynamics of antibody‐based therapeutics.

Therapeutic antibody development requires consideration of folding stability, aggregation propensity, and protein–protein interactions, all of which can be perturbed in vivo due to thermodynamic nonideality. In highly concentrated solutions of other macromolecules, proteins may experience macromolecular crowding as well as weak, nonspecific interactions with cosolutes that lead to deviations from the ideal behaviors seen in dilute conditions. These nonideal effects can perturb stability and molecular recognition in ways that are challenging to predict and model (Bhattacharya et al., 2013; Kuznetsova et al., ²⁰¹⁵; Rivas & Minton, ²⁰²²; Sarkar et al., ²⁰¹³; Speer et al., ²⁰²²; Zhou et al., ²⁰⁰⁸).

Despite extensive characterization in vitro, little is known about the behavior of biopharmaceuticals in relevant physiological environments. Given the long circulating half‐lives of therapeutic antibodies (typically several weeks), blood serum is the biological environment to which they are exposed the longest. For this reason, there is growing interest in understanding how therapeutic antibodies interact with serum proteins, and how serum impacts target or receptor binding (Chaturvedi et al., 2020; Demeule et al., ²⁰⁰⁹; Kim et al., ²⁰¹⁹; Larsen et al., ²⁰²¹; Wright, Hayes, Stafford, et al., ²⁰¹⁸). Human serum is a concentrated, complex fluid whose composition varies with physiological or disease state, from one individual to another; in general, serum contains approximately 60–80 mg/mL protein, comprised of mainly albumin (60%) and immunoglobulin G, IgG (20%) with lower levels of other globulins, lipoproteins, transport proteins, complement factors, and smaller osmolytes (Anderson et al., 2004; Gonzalez‐Quintela et al., ²⁰⁰⁸; Leeman et al., ²⁰¹⁸). There is clear evidence that such crowded environments may disrupt binding interactions of antibodies (Kim et al., 2019; Słyk et al., ²⁰²³) and other proteins (Bhattacharya et al., 2013; Jiao et al., ²⁰¹⁰).

Our understanding of crowding, nonideality, and their consequences relies on techniques capable of probing transient intermolecular interactions, including analytical ultracentrifugation (AUC) (Wright, Hayes, Stafford, et al., 2018), dynamic light scattering (Yadav et al., 2011), static light scattering (Roberts et al., 2014), and composition‐gradient multiangle light scattering (Ma et al., 2015). While these methods have provided valuable insight, they are impossible or technically challenging to implement in complex, multicomponent biological media. Fluorescence correlation spectroscopy (FCS), a single‐molecule fluorescence technique, is an attractive alternative because it is capable of characterizing the diffusion of molecules directly in environments such as serum and plasma (Macháň & Wohland, 2014; Schmitt et al., ²⁰²²; Yu et al., ²⁰²¹). By comparison with other methods, FCS does not require a stationary phase or large quantities of protein and is relatively cheap and easy to implement.

This work builds on a previously developed and validated in‐serum FCS approach to probe nonideality effects through apparent second virial coefficient (B _2,app) measurements (Larsen et al., 2021). Traditional self‐term (B ₂₂) and cross‐term (B ₁₂) second virial coefficients describe weak, nonspecific, pairwise interactions between two of the same or different molecules, respectively. The sign of the coefficient informs on the nature of the interactions, where a positive value indicates repulsion, a negative value attraction, and a zero value no interaction between the molecules of interest. Analogously, the apparent second virial coefficient (B _2,app) probes weak, nonspecific interactions between a labeled species and the components in complex media (such as serum) as a measure of global nonideality. In this case, a positive value indicates net repulsion and a negative value net attraction, whereas a zero value likely indicates a balance between attraction and repulsion.

Briefly, the diffusion coefficient of a labeled species scaled for bulk viscosity (D _adj) at a given concentration of unlabeled species ($c_p$) can be used to determine the second osmotic virial coefficient ($B_2$) through the following relationship:

$$D_{\mathrm{adj}}=D_0\left(1+\left(2B_{2}M–\overline{V}\right)c_p\right)=D_0\left(1+k_{\text{diff}}{c}_p\right)$$ (1)

where D ₀ is the diffusion coefficient at infinite dilution, M is the effective molecular weight of unlabeled species, $\overline{V}$ is the partial specific volume, and k _diff is the diffusion interaction parameter defined by:

$$k_{\text{diff}}=2B_{2}M–\overline{V}\approx 2B_{2}M.$$ (2)

For complex mixtures such as serum, the appropriate molecular weight of the interacting species is unclear. Reporting results in terms of k _diff (2B ₂ M), rather than B ₂, avoids this problem. The complete theory behind our apparent second virial coefficient approach can be found in our previous publication (Larsen et al., 2021) (note a change in nomenclature: here we denote the diffusion interaction parameter as k _diff rather than k _D to avoid confusion with the equilibrium binding constant K _D ).

Higher order virial terms are omitted above due to their low and likely negligible contributions at concentrations relevant to serum and plasma. Accordingly, our current interpretation of B _2,app focuses on the sign of the parameter and qualitative changes rather than the precise magnitude of the values. This approach trades formal rigor for the ability to compare our B _2,app values with analogous B ₂ values determined in similar concentration regimes (albeit less physiologically relevant matrices) using well‐established techniques (Wright, Hayes, Stafford, et al., 2018; Yang et al., ²⁰¹⁸). However, this analysis could be extended in the future following Kirkwood–Buff solution theory to account for higher order interactions (Blanco et al., 2011; Blanco et al., ²⁰¹⁴).

Classic experiments investigating the effects of complex media on protein behavior have focused on either excluded volume effects by mimicking biological environments with crowding agents (e.g., polyethyleneglycol, other polymers, or proteins) (Minton, 1998; Wills et al., ¹⁹⁹⁵), or nonideality by isolating specific components and characterizing their effects independently (Moreira et al., 2007; Quigley & Williams, ²⁰¹⁵; Velev et al., ¹⁹⁹⁸). More recent studies investigating the effects of serum‐induced nonideality on therapeutic antibodies have taken the traditional approach of completing independent second virial coefficient measurements with isolated serum proteins (Kim et al., 2019; Wright, Hayes, Sherwood, et al., ²⁰¹⁸). However, such approaches may not effectively measure global nonideality in serum, because they cannot capture the effects that serum components have on each other. B _2,app measurements present an avenue to resolve this question directly. Furthermore, B _2,app measurements can be expanded to other biological fluids to characterize nonideality effects in other biological environments.

Here, we characterize the contributions of albumin and serum IgG antibodies to serum‐induced global nonideality effects for a panel of IgG1 antibodies. Cross‐term virial coefficient measurements with IgG antibodies isolated from human serum, human serum albumin (HSA), and IgG‐depleted serum were compared to values obtained in pooled human serum. In addition, cross‐term virial coefficient measurements were carried out with antibody Fab and Fc fragments to determine how these different domains mediate nonideal effects on mAbs. This comparison also highlights systematic differences between mAb‐ and Fab‐based therapeutic platforms. Our studies provide unprecedented, detailed insight into the impacts of serum‐induced nonideality on therapeutic antibodies and other protein‐based therapeutics.

2. RESULTS

The ability of FCS to measure diffusion coefficients in complex media enables the quantification of global nonideality effects on therapeutic antibodies in biological environments such as serum. Using our previously established approach (Larsen et al., 2021), we have transitioned from measurements in fetal bovine serum to measurements in pooled human serum as a more representative physiological model. Here, we compare how the two most abundant serum proteins, HSA, and serum IgG antibodies, contribute to serum‐induced nonideality for a panel of monoclonal antibodies and their corresponding Fab fragments. Second virial coefficients were measured in pooled human serum, IgG‐depleted serum, IgG antibodies isolated from human serum, and HSA, as summarized in Table 1. This comparative approach provides insight into component effects in isolation and the context of serum.

TABLE 1.

Summary of second virial coefficient results.

Sample	k diff or 2B 2 M (mL/g)
	Pooled human serum	IgG‐depleted serum	Serum IgG antibodies	HSA
NIST mAb
mAb	−1.1 ± 0.8	4 ± 1	−12 ± 1	3.3 ± 0.4
Fab	−5.1 ± 0.7	1 ± 1	n.d.	0 ± 1
Tocilizumab
mAb	−0.4 ± 0.2	3.8 ± 0.5	−16.0 ± 0.5	3.3 ± 0.3
Fab	−3.2 ± 0.4	4.0 ± 0.8	n.d.	0.6 ± 0.9
Anti‐RSV
mAb	−6.5 ± 0.7	0.9 ± 0.3	−13 ± 2	6.0 ± 0.2
Fab	−6.1 ± 0.3	−1.1 ± 0.2	n.d.	1.0 ± 0.3
Anti‐gp120
mAb	−2.0 ± 0.1	2.4 ± 0.3	−13 ± 1	2.1 ± 0.4
Fab	−3.6 ± 0.3	2.0 ± 0.5	n.d.	1.9 ± 0.3

Note: Values presented are mean ± S.D. from at least three independent replicates.

Abbreviation: n.d.: not determined.

Figure 1 depicts the experimental setup for determining second virial coefficient values. Alexa Fluor 488 SE labeled antibodies are diluted into varying concentrations of carrier solutions (human serum, IgG‐depleted serum, IgG antibodies isolated from human serum, or HSA), and the corresponding diffusion coefficients are determined from FCS diffusion time measurements. Diffusion coefficients were adjusted for bulk viscosity, plotted against the total protein concentration of the carrier solution, and fit to Equation 1, where the slope over the y‐intercept yields interactions parameter k _diff that is directly proportional to B ₂ via the relationship in Equation 2. Complete D _adj versus c plots are presented in Figures S1–S5 and S7–S11. At the nanomolar concentrations used for FCS measurements, self‐term interactions (i.e., labeled mAbs interacting with each other) are negligible, and observed nonideal behavior is solely due to cross‐term cosolute interactions.

FIGURE 1.

Model of k _diff (or 2B ₂ M) determination: fluorescently labeled antibodies are diluted into varying concentrations of carrier solution (serum, depleted serum, isolated serum IgG antibodies, or HSA). The diffusion coefficient at each condition is determined via FCS diffusion time measurements, and k _diff or B ₂ values are determined by fitting plots of viscosity‐adjusted diffusion coefficients against carrier concentration to Equation 1. Positive values indicate net repulsion in serum or repulsive interactions with isolated serum proteins, whereas negative values indicate net attraction in serum or attractive interactions with isolated serum proteins. A zero value likely indicates a balance between attraction and repulsion in serum as opposed to no interaction when measuring interactions with isolated proteins such as HSA.

Interpretation of cross‐term interaction values in serum and HSA are slightly different. In the case of serum, k _diff values deviating away from zero in the positive and negative direction indicate either net repulsion or attraction, respectively, whereas a value of zero likely indicates a balance between attraction and repulsion. Conversely, with HSA, positive and negative values indicate repulsive and attractive interactions, respectively, whereas a value of zero indicates there are no net interactions between IgG1 and HSA.

2.1. mAb interactions with serum and serum proteins

Apparent second virial coefficient measurements were carried out in pooled human serum to probe global nonideality between labeled antibodies and serum components. Values of k _diff in serum were determined for a panel of four antibodies: NIST mAb, tocilizumab, anti‐gp120 mAb, and anti‐RSV mAb. NIST mAb is a widely used reference standard mAb, tocilizumab is a clinically approved therapeutic (an interleukin‐6 receptor inhibitor used to treat autoimmune diseases), while anti‐RSV mAb (targeting fusion [F] glycoprotein from respiratory syncytial virus [RSV]) and anti‐gp120 mAb (targeting envelope glycoprotein from human immunodeficiency virus) are research antibodies expressed in‐house. Sequences and sequence properties of all four mAbs are presented in Table S1. NIST mAb (Figure 2a) and tocilizumab (Figure 2b) both displayed a balance of attractive and repulsive interactions with serum, with k _diff values (−1.1 ± 0.8 mL/g and −0.3 ± 0.4 mL/g, respectively) not significantly different from zero. These results are comparable to those obtained previously in FBS for the same antibodies (Larsen et al., 2021). In contrast, anti‐gp120 mAb (Figure 2c) exhibited a k _diff value of −2.0 ± 0.1 mL/g, whereas anti‐RSV mAb (Figure 2d) exhibited a k _diff value of −6.5 ± 0.7 mL/g, suggesting weak‐to‐moderately attractive interactions with serum components.

FIGURE 2.

Comparison of cross‐term interactions in human serum (black circles), IgG‐depleted serum (blue squares), albumin (purple triangles), and IgG antibodies isolated from human serum (green diamonds) for NIST mAb (a), Tocilizumab (b), anti‐gp120 mAb (c), and anti‐RSV mAb (d). D _adj versus c plots (Figures S1–S4) were fit to Equation 1 to obtain k _diff values, with results presented as mean ± S.D. from three replicates with independent sample preparations.

Upon depleting the serum of IgG antibodies, k _diff values became positive, indicating net repulsion in the depleted serum. In all cases, the change from either a negative or zero value in pooled human serum to a positive value in the depleted serum indicates that serum IgG antibodies are primarily responsible for attractive interactions in serum. Values measured with HSA were very similar to IgG‐depleted serum for all mAbs except anti‐RSV. The decrease of 5.1 mL/g for anti‐RSV mAb going from HSA to IgG‐depleted serum suggests that some non‐IgG components of serum also interact with this mAb but not the others in our panel. Interestingly, measurements with serum IgG antibodies isolated from pooled human serum produced negative k _diff values ranging from −12 to −16 mL/g, indicating much stronger attractive interactions with labeled antibodies. The discrepancy in the magnitude change in k _diff values with pooled human serum to those in IgG‐depleted serum and serum IgG antibodies isolated from serum indicate that the approach of characterizing molecules of interest in the presence of isolated components is not an accurate representation of nonideality in serum.

2.2. Fc fragment interactions with serum and serum proteins

Values of k _diff were measured for IgG1 Fc fragment with pooled human serum, IgG‐depleted serum, and HSA (Figure 3a). In all cases, k _diff did not significantly deviate from 0 mL/g, indicating that there are no net interactions between IgG1 Fc and serum IgG antibodies, HSA, or whole serum. This suggests that nonideal interactions of mAbs are likely driven by differences in the Fab domain, while Fc domains may be evolutionarily optimized to avoid nonspecific interactions with other constituents of serum.

FIGURE 3.

Cross‐term interaction for IgG1 Fc fragment (a) with serum (black circle), IgG‐depleted serum (blue square), and HSA (purple triangle). Comparison of cross‐term interactions in human serum for glycosylated (closed symbols) and deglycosylated (open symbols) NIST mAb, tocilizumab, anti‐gp120 mAb, and anti‐RSV mAb (b). D versus c plots (Figures S5 and S7) were fit to Equation 1 to obtain k _diff values, with results presented as mean ± S.D. from three replicates with independent sample preparations.

2.3. Deglycosylated mAb interactions with serum and serum proteins

Glycosylation of IgG1 mAbs, which occurs predominantly at specific sites in the Fc domain, can alter interactions with Fc receptors and consequent effector functions (Cobb, 2020). To test the effect of this functional modification on mAb interactions with serum components, k _diff values were measured for NIST mAb, tocilizumab, anti‐gp120 mAb, and anti‐RSV mAb in pooled human serum following deglycosylation (Figure 3b). In all cases, k _diff values were similar between deglycosylated and glycosylated mAbs, suggesting that glycans do not significantly contribute to nonideal interactions with serum. While there is a trend of slightly more negative k _diff values for deglycosylated mAbs which could indicate increased propensity for interactions with serum, these shifts are not statistically significant. Taken together, these observations point to structural or dynamic differences in the Fab domains as drivers of nonideality in mAb/serum interactions.

2.4. Fab fragment interactions with serum and serum proteins

B _2,app values for Fab fragments generated from NIST mAb, tocilizumab, anti‐RSV mAb, and anti‐gp120 mAbs were measured in pooled human serum, IgG‐depleted serum, and HSA (Figure 4). In most cases, k _diff values were more negative in pooled human serum for Fab fragments than for the corresponding full‐length antibodies. This indicates that Fab fragments experience greater net attraction to serum, which could be attributed to the loss of interactions between the Fab and Fc domains, exposing more potential sites for nonspecific interactions. Because Fab fragments are known to be less stable than full‐length antibodies (Röthlisberger et al., 2005), experiments were carried out within 48 h of fragmentation to avoid the risk of aggregation. Furthermore, observed diffusion times of Fab fragments were consistent with expected values, providing confidence that aggregation of A488‐labeled Fab fragments was not occurring.

FIGURE 4.

Comparison of cross‐term interactions in human serum (black circles), IgG‐depleted serum (blue squares), and albumin (purple triangles), for full‐length antibodies (closed symbols) and Fab fragments (open symbols). NIST mAb (a), Tocilizumab (b), anti‐gp120 mAb (c), and anti‐RSV mAb (d) D _adj versus c plots (Figures S8–S11) were fit to Equation 1 to obtain k _diff values, with results presented as mean ± S.D. from three replicates with independent sample preparations.

As with intact mAbs (Figure 2), IgG depletion decreased or eliminated attractive interactions with serum. Fab k _diff values in IgG‐depleted serum were increased by 5–7 mL/g relative to pooled human serum, suggesting that serum IgGs are key mediators of attractive interactions for Fab fragments. However, for NIST mAb (Figure 4a) and anti‐RSV mAb (Figure 4d), k _diff in IgG‐depleted serum was significantly lower for Fabs than for the parental mAbs, indicating that non‐IgG components were also playing a role in attraction. The overall trends in the data suggest that Fab fragments display stronger attractive interactions with more diverse serum cosolutes and greater variability than intact mAbs, all of which could carry negative functional consequences.

The anti‐gp120 Fab (Figure 4c) and anti‐RSV Fab (Figure 4d) are modestly repelled by HSA (as evident by positive k _diff values), whereas NIST Fab (Figure 4a) and tocilizumab Fab fragment (Figure 4b) exhibit no interaction with albumin (k _diff values close to 0 mL/g). The latter result is striking, given that tocilizumab displays a positive k _diff value in IgG‐depleted serum, which is predominantly HSA. It would seem that the minor fraction of non‐HSA, non‐IgG components may be able to increase the repulsion between HSA and the tocilizumab Fab fragment to generate the behavior observed in IgG‐depleted serum. This further supports the idea that the net interaction with serum is not equal to the sum of the individual interactions with isolated serum components. Moreover, these nonadditive effects can vary widely from one molecule to the next.

3. DISCUSSION

Our understanding of protein–protein interactions in biological environments is limited by the inability of most analytical techniques to complete measurements in complex media. For this reason, experiments aimed at investigating serum‐induced nonideality through second virial coefficient measurements have focused on independent cross‐term (B ₁₂) measurements with isolated serum components (Correia et al., 2018; Kim et al., ²⁰¹⁹). Our results, however, suggest that this reductionist approach does not sufficiently model nonideality in serum. The second virial coefficient results for the panel of antibodies in Figure 2 with serum, HSA, and pooled IgG antibodies isolated from serum clearly support this inference. Repulsion is observed with HSA (as evidenced by positive k _diff values), while either attraction or a balance between attraction and repulsion is observed in serum (values ranging from 0 to −6.5 mL/g). The lower k _diff values in serum compared to HSA suggests that attractive interactions likely occur simultaneously in addition to the repulsion observed with albumin, a phenomenon that could not have been probed using measurements with individual components. When depleting serum of IgG antibodies, the k _diff values become positive indicating that the attraction observed in serum is likely due to serum IgG antibodies. However, IgG antibodies isolated from serum were very strongly attracted to mAbs, with k _diff values ranging from −12 to −16 mL/g, much greater than the change in k _diff when going from human serum to IgG‐depleted serum. This is likely due to interactions among serum proteins that contribute to global nonideality but are lost when specific components are studied in isolation. These results mirror AUC studies (Yang et al., 2018) where weak attractive interactions were observed between mAbs and pooled serum IgGs. While the magnitude of k _diff values measured here differs from reported k _s values (as expected when comparing parameters based on sedimentation vs. diffusion), the trends and overall conclusion are consistent.

Our results may provide insight into the evolutionary pressures on nonideal behavior exhibited by mAbs in biological environments such as serum. Conserved Fc domains have likely evolved to optimize interactions with Fc‐receptors and complement proteins that mediate effector functions, while simultaneously disfavoring nonspecific interactions with other constituents of serum. This is supported by our results for IgG1 Fc fragment in Figure 3, which showed no net interactions with HSA, IgG‐depleted serum, or whole serum, as evident by k _diff values that did not significantly deviate from 0 mL/g. Conversely, the hypervariable sequences in Fab domains are not subject to this selective pressure (particularly true of artificial or engineered mAbs) and therefore drive nonideal interactions with serum. Fab fragments for the panel of antibodies in Figure 4 display attraction to serum IgG antibodies, as evident by the changes in k _diff from negative values in pooled human serum to positive values in IgG‐depleted serum. Furthermore, the more negative k _diff values observed for Fab fragments compared with corresponding full‐length antibodies in Figure 4 suggest that the Fc domain essentially buffers nonideal interactions in the Fab domains. Supporting the idea of such long‐range cooperativity within the mAb scaffold, it has been shown that Fab domain properties can tune the binding of the Fc domain to the neonatal Fc receptor (Schlothauer et al., 2013; Suzuki et al., ²⁰¹⁰; Wang et al., ²⁰¹¹).

Our observations highlight a potentially unappreciated advantage of mAbs over Fab fragments in therapeutic development, in terms of decreased risk of cosolute interactions. More heavily engineered mAb‐based therapeutic platforms such as ADCs and bsAbs may, like Fabs, be more prone to cosolute interactions in vivo. De novo–designed therapeutic candidates, which are not based on mAbs and have not undergone either rational or evolutionary optimization to avoid cosolute interactions (Bonadio & Shifman, 2021; Quijano‐Rubio et al., ²⁰²⁰), may be at even greater risk of aberrant behavior.

Expanding B _2,app measurements to a larger, more representative panel of mAbs and mAb‐derived therapeutics could illuminate these fundamental questions. The present panel only includes four mAbs, comprising one approved therapeutic, one reference product, and two nonclinical mAbs. Intriguingly, anti‐RSV mAb, which displays atypically strong attractive interactions with serum, also displays unusual sequence properties. The anti‐RSV variable domain is the only one of the four with a negative net charge per residue (Figure S12), largely due to the high negative charge of its light chain (LC) (Table S1). Low isoelectric point values have been linked to nonspecific binding and aberrant clearance. (Gupta et al., 2022; Rabia et al., ²⁰¹⁸) However, systematic characterization of many more clinical and nonclinical mAbs would be necessary to draw any firm mechanistic conclusions. Sequence patterns (beyond simply net charge or hydrophobicity) might be influential determinants of cosolute interactions.

Our approach differs from current applications of second virial coefficient measurements in the drug development process, which are predominantly focused on characterizing protein aggregation. During manufacturing, biologics are subject to a variety of stresses (i.e., temperature, pH, ionic strength, etc.) that can promote aggregation. In addition, highly concentrated antibody formulations are susceptible to aggregation that can negatively impact the overall efficacy, safety, and half‐life of the therapeutic product (Le Basle et al., 2020). Therefore, aggregation continues to be a major obstacle and focus during the development process. As a result, self‐term virial coefficient (B ₂₂) measurements have been widely implemented in biologics development to probe aggregation propensity (Baek & Zydney, 2018; Le Brun et al., ²⁰¹⁰; Obrezanova et al., ²⁰¹⁵; Pham & Meng, ²⁰²⁰; Saluja et al., ²⁰¹⁰). Cross‐term virial coefficient measurements are not as common but have been used to probe protein‐excipient interactions during formulation development (Kamerzell et al., 2011). Such applications of the second virial coefficient, while useful, only report on the behavior of the therapeutic before administration; little is known about the behavior of therapeutic proteins in biological environments. B _2,app measurements provide a tool to better understand how therapeutic proteins interact with cosolutes in crowded, complex biological environments and close this knowledge gap.

Better methods for early identification of proteins more likely to fail would make biologics development faster and more efficient. Although a variety of in silico, in vitro, and in vivo tools are currently used to de‐risk molecules in development, the pharmacokinetics of therapeutic antibodies remain complex and difficult to predict (Dostalek et al., 2017). Many biochemical and biophysical properties have been shown to impact the pharmacokinetics of proteins, such as hydrophobicity (Sharma et al., 2014), net charge (Igawa et al., 2010), off‐target binding (Hötzel et al., 2012), glycosylation (Yu et al., 2012), antigen binding (Shi, 2014), and interactions with Fc receptors (Liu, 2018; Wu et al., ²⁰⁰⁷). Although the contribution of serum‐induced nonideality on mAb efficacy or disposition has not yet been directly investigated, we have observed a case suggesting a potential correlation between net attraction in serum and lower in vivo binding affinity. In our previous study, Carlumab, a discontinued antibody from Janssen Therapeutics, exhibited more attraction (k _diff = −11 mL/g) to fetal bovine serum than clinical (tocilizumab) or reference standard (NIST mAb) antibodies (Larsen et al., 2021). Interestingly, Carlumab's failure during clinical trials was attributed to a discrepancy in the in vitro and in vivo binding affinity for its target, CCL2, that resulted in limited therapeutic efficacy (Majety et al., 2018). This indirectly supports the idea that nonideal solutions can alter the functional and therapeutic properties of biologics. However, the relationship between attractive cosolute interactions and perturbed antigen binding needs further investigation.

Further studies will reveal whether and how serum‐induced nonideality can impact the overall efficacy of therapeutic antibodies. Results from this study suggest that serum IgG antibodies are responsible for attraction in serum, as evident by the change from negative or zero k _diff values in pooled human serum to positive k _diff values in IgG‐depleted serum for four IgG1 antibodies. We cannot yet identify which particular IgGs in serum interact with a given mAb, or with what affinity. Both the properties of a given therapeutic mAb and the cocktail of serum IgGs borne by a given person would determine the nature of cosolute interactions, meaning that nonideality is a potential source of interindividual variability in mAb disposition. Interestingly, the attractive interactions were determined to occur in Fab domains, which could have a negative impact on antigen binding. Speculatively, attractive interactions may exert a destabilizing effect by promoting local unfolding in the antibody that can perturb binding. In effect, molecules that exhibit attractive interactions with serum could be at greater risk of perturbation in vivo. Therefore, B _2,app measurements have the potential to advance our understanding of the serum environment in vivo and help de‐risk therapeutic candidates during biologics development.

4. CONCLUSION

In this study, we used FCS‐based second virial coefficient measurements to determine the contributions of albumin and IgG antibodies to global nonideality in serum for a panel of IgG1 antibodies. Apparent second virial coefficient measurements with pooled human serum, IgG‐depleted serum, and in solutions of isolated serum proteins for a panel of full‐length antibodies and their respective Fab fragments provided an interesting insight into the origins of serum‐induced nonideality. Differing results for the panel of IgG1 antibodies in pooled human serum supported our previous finding that nonideality effects are antibody‐dependent. Comparing interaction parameters across the different carrier solutions for each antibody revealed trends of attraction to serum IgG antibodies and repulsive interactions with albumin. Furthermore, attractive interactions were determined to occur in the Fab domain across the entire panel of antibodies, highlighting the potential for negative impacts to antigen binding. However, the significance of these measurements needs further investigation. Finally, comparing apparent second virial coefficient results in pooled human serum and IgG‐depleted serum to those obtained with HSA and pooled serum IgG antibodies, suggests that the classical approach of characterizing the contributions of isolated serum proteins, via independent cross‐term virial coefficient measurements, does not adequately model nonideality in serum. B _2,app measurements could provide an essential tool for understanding the behavior of therapeutic antibodies in physiological environments. Furthermore, such measurements have the potential to enhance in vitro characterization used to de‐risk therapeutic candidates during biologics development.

5. MATERIALS AND METHODS

5.1. Samples

The NIST mAb humanized IgG antibody (10 mg/mL) was purchased from the National Institute of Reference Standards and Technology (RM 8617). Tocilizumab (35 mg/mL) and anti‐RSV mAb (10.91 mg/mL) IgG antibodies were provided by the Genentech Outgoing Materials Transfer Agreement program and Jannsen Pharmaceuticals, respectively. Albumin from human serum (lyophilized powder) and serum IgG antibodies were purchased from Sigma‐Aldrich. Pooled human serum (34019) and IgG‐depleted serum (34021) were purchased from Pel‐Freeze Biologics. Recombinant Human IgG1 Fc Fragment was purchased from Thermo Scientific (A42561). Transient expression of Human anti‐gp120 IgG1 mAb was carried out in Expi293F cells (Thermo Fisher A14635) using a 1:3 ratio of heavy chain to LC DNA according to the manufacturer's protocol. Purification was performed via Protein A column in 1× phosphate‐buffered saline (PBS: 150 mM NaCl, 2.7 mM KCl, 10 mM Na₂HPO₄) pH 7.2. Briefly, the supernatant was loaded onto the protein A column, washed with PBS, and eluted with 100 mM sodium citrate, pH 3.5 into 1.0‐mL fractions. Ab‐containing fractions (based on A280 UV‐signal) were pooled and immediately buffer exchanged into 1× PBS using Zeba Spin Desalting columns (Thermo Scientific), following the manufacturer's protocol. The antibodies and HSA were stored at 4°C in PBS pH 7.4. Antibody sequences were analyzed using AbRSA (Li et al., 2019), IPC (Kozlowski, 2016), and CIDER (Holehouse et al., 2017).

5.2. Protein labeling

Following the Thermo Scientific labeling protocol, NIST mAb, tocilizumab, anti‐RSV mAb, anti‐gp120 mAb, and the Fc fragment were labeled with Alexa‐Fluor 488 carboxylic acid, succinimidyl ester (Thermo Scientific A20000; A488). Desalting was carried out using Zeba Spin Desalting Columns from Thermo Scientific (89882), following the Thermo Scientific protocol. Labeling efficiency and mAb concentration were determined through UV–Vis spectroscopy (absorbance measurements at 280 nm and 494 nm) on a Nanodrop One Microvolume UV–Vis spectrometer (Thermo Scientific, ND‐ONE‐W). Additionally, diffusion time was used to verify the absence of free dye in the solution by comparing observed changes to expected changes in diffusion from free dye to protein‐bound dye. A488‐mAbs were stored in 1x PBS pH 7.4 at 4°C before dilution to ~20 nM for FCS measurements. Typically, 1–2 A488 molecules covalently attached to each mAb or Fab fragment, while 0.5 molecules of A488 covalently attached to the Fc fragment.

5.3. Antibody deglycosylation

NIST mAb, tocilizumab, anti‐RSV mAb, and anti‐gp120 mAb deglycosylation was carried out by adding 2 μL of PNGaseF (New England Biolabs P0711S) to 60 μg of antibody (in 1× PBS, pH 7.4) in a reaction volume of 20 μL following manufacturer's suggestions. After 2 h at room temperature, PNGaseF was removed from samples using Zeba Spin Desalting Columns from Thermo Scientific (87766). Deglycosylation products were confirmed by SDS‐PAGE (Figure S6) using 4%–20% polyacrylamide gels from Bio‐Rad (4561094). Deglycosylated antibodies were labeled with Alexa Fluor 488 SE using the labeling protocol above. Samples were stored in 1× PBS, pH 7.4 at 4°C.

5.4. Antibody fragmentation

NIST mAb, tocilizumab, anti‐RSV mAb, and anti‐gp120 mAb were first labeled with Alexa Fluor 488 SE using the labeling protocol above. A488‐labeled antibodies were fragmented following the Thermo Scientific protocol using a Pierce Fab Micro Preparation kit (Thermo Scientific 44685). Fab fragments were stored in 1x PBS pH 7.4 at 4°C and used within 48 h of fragmentation. Fc fragments were not salvaged using this protocol due to low yield. Recombinant IgG1 Fc was ordered from Thermo Scientific for experiments.

5.5. FCS

All experiments were carried out at room temperature on a home‐built instrument based on a Zeiss Axio Observer D1 microscope equipped with HydraHarp 400 detection electronics, Tau‐SPAD photon counting detector, and pulsed 485 nm laser line driven by a PicoQuant PDL 828 Sepia II driver (PicoQuant GmbH, Berlin, Germany). Sample aliquots of 50 μL were placed on a 22 × 22 cover glass (VWR 48366‐067). Five 30‐s measurements of 10 nM A‐488 were used to calibrate the instrument at the start of each experiment. The average diffusion time was used to determine the ω² parameter needed for second virial coefficient calculations. Antibody and Fab fragment measurements (n = 5) were carried out for 60s each.

A488‐labeled antibodies or antibody fragments were diluted to ~20 nM in varying concentrations of carrier protein (HSA, pooled IgG antibodies from human serum, pooled human serum, or IgG‐depleted human serum) ranging from 0% to 100%, where the 100% condition ranged from 10 to 62 mg/mL depending on the experiment.

FCS intensity time traces were imported into Prism 9.0 (Graphpad Software, Boston, MA) and fit to a single component FCS equation to yield diffusion time, ${\tau }_D$, using the following equation (Haustein & Schwille, 2007):

$$G_{\mathrm{FCS}}\left(\tau \right)=\frac{1}{N}\left(\frac{1}{1+\raisebox{1ex}{$\tau $}\!\left/ \!\raisebox{-1ex}{${\tau }_D$}\right.}\right)\sqrt{\frac{1}{1+\raisebox{1ex}{$s^2\tau $}\!\left/ \!\raisebox{-1ex}{${\tau }_D$}\right.}}$$ (3)

where N is the mean number of molecules in observation volume, ${\tau }_D$ is the correlation decay time due to translational diffusion, and s is the axial ratio of the detection volume (0.2 for our instrument). Translational diffusion coefficients (D) were determined from diffusion times and used to calculate interaction parameters (k _diff), that are directly proportional to the second osmotic virial coefficient (B ₂). The methodology and validation behind these measurements are outlined in our previous publication (Larsen et al., 2021).

AUTHOR CONTRIBUTIONS

Hayli A. Larsen, Abhinav Nath, and William M. Atkins designed experiments. Hayli A. Larsen performed experiments and analyzed data. Abhinav Nath and William M. Atkins provided analytical tools. Hayli A. Larsen, Abhinav Nath, and William M. Atkins wrote the manuscript.

CONFLICT OF INTEREST STATEMENT

The authors declare no conflicts of interest.

Supporting information

Supporting Information S1: Larsen_ComponentAnalysis SI.pdf contains D _adj versus C plots for all molecules described in the manuscript, mAb sequence and analysis, and characterization of deglycosylated mAbs.

Larsen HA, Atkins WM, Nath A. The origins of nonideality exhibited by monoclonal antibodies and Fab fragments in human serum. Protein Science. 2023;32(12):e4812. 10.1002/pro.4812