Abstract

Introduction

Human or humanized monoclonal antibodies have been broadly successful as therapeutic agents for human disease. Over 30 antibodies have received FDA-approval for treatment of a variety of disorders and hundreds more are currently in clinical development.¹ In addition to the exquisite selectivity and high potency that can be achieved with an antibody therapeutic, the success of antibodies as drugs has greatly benefited from their typically long circulating half-life. Slow clearance from blood enables desired drug concentrations to be realized with infrequent dosing. Antibody drugs are usually administered via intravenous infusion or subcutaneous injection; less frequent administration improves patient compliance and clinical benefit.

Antibodies can be eliminated from systemic circulation by several mechanisms,^2,3 including intracellular catabolism following fluid phase endocytosis (non-specific clearance), antigen-mediated clearance,⁴ and, in some cases, formation of anti-therapeutic antibodies (ATA). Antigen-mediated clearance is generally observed when antibodies target cell-bound antigens; it is most prominent at low concentrations of antibody and can usually be saturated by increasing the antibody dose. ATA have been observed to appear as early as 4–7 d after dosing, involve formation of immune complexes that can be rapidly cleared, and can sometimes be easily detected from the atypical shape of the antibody’s plasma concentration-time profile.

Antibody elimination is slowed through an FcRn-dependent recycling mechanism.^5,6 As an antibody traverses the endocytic pathway, it can bind to FcRn at pH < 6. Binding to FcRn protects IgG from catabolism and promotes return to the apical cell surface where it can be rapidly released at the pH (> 7) of blood. These features result in half-lifes of 6–32 d for human or humanized antibodies in humans.⁷ On the basis of pharmacokinetic (PK) data with 12 IgG1 antibodies, a mean value for clearance in humans of 3.9 ± 1.2 (s.d.) mL/day/kg with range of ~2–6 mL/day/kg was determined.⁸ Similarly, studies of these antibodies in cynomolgus monkeys yielded a mean clearance of 6.5 ± 2.9 (s.d.) mL/day/kg. Engineering of the antibody Fc region to increase binding to FcRn at pH 6.0 can increase half-life of potential therapeutic antibodies in cynomolgus monkeys and in mice.^9-12

Preclinical testing of a potential therapeutic antibody in a relevant non-human species is important to gain understanding of the expected dosing regimen in humans, and to assess potential toxicities. Given the high target antigen sequence homology between human and non-human primates (NHP), and similar binding affinities for the recycling FcRn receptor,⁹ cynomolgus monkey is the preferred species for preclinical PK studies. Previously we have shown that the non-specific clearance of therapeutic IgG1 antibodies determined in humans is commonly about half that measured in cynomolgus monkeys.⁸

One potential mechanism that can contribute to faster than expected clearance is off-target binding.^13-16 Although highly specific off-target binding can sometimes be identified and eliminated,¹⁴ off-target binding is often of unknown origin and difficult to saturate with an increase in dose. In vitro systems to evaluate and predict the in vivo absorption, distribution, metabolism, and elimination mechanisms or the in vivo PK behavior of antibodies are not yet established. We sought to develop in vitro assays of non-specific binding that would be useful for identifying antibodies likely to show fast clearance in vivo. Assays of non-specific binding based on binding to glass surfaces or elution times on a gel filtration column did not show an association with fast clearance (Figs. S1 and S2). Here, we describe an assay based on ELISA detection of binding to baculovirus (BV) particles that is useful to evaluate the non-specific binding of therapeutic proteins. Using a large panel of antibodies, we demonstrate that the assay readout is associated with the non-specific clearance measured in non-human primate (n = 45) and human (n = 16), and can be employed to mitigate the risk of fast clearance for potential therapeutic antibodies.

Results

Predictive scaling of human clearance from clearance in monkeys

Using a slightly larger data set than previously available,⁸ we confirmed a strong correlation between the clearance values measured in cynomolgus monkeys and humans [Fig. 1, Spearman’s correlation coefficient (ρ) = 0.74]. A simple scaling guideline is obtained in that the clearance measured in humans is about two-fold slower than the clearance measured in cynomolgus monkeys, with the anti-NRP1 antibody as an outlier.

Open in a separate window

Figure 1. Correlation of antibody clearance values measured in humans and cynomolgus monkeys (ρ = 0.74, n = 16). For our analyses, antibody doses were chosen that were believed to saturate any target-dependent clearance. The solid line is a linear regression fit of the logarithm of human clearance to the logarithm of cynomolgus monkey clearance [log(Human clearance) = 0.0123 + 0.744 * log(Cyno clearance)]. For most antibodies shown, human clearance is about two-fold slower than the corresponding cynomolgus monkey clearance. By contrast, for anti-NRP1 the clearance in human (9.2 mL/day/kg) is ~2-fold faster than the clearance in cynomolgus monkey (4.3 mL/day/kg).

Characterization of structural variants and clearance in cynomolgus monkeys for a panel of therapeutic antibodies

In the normal course of preclinical development, PK data in cynomolgus monkeys were collected for 52 humanized and human antibodies. The predominant isotype of these antibodies was IgG1, κ (n = 46). The antibody panel also included 4 IgG1λ and 2 IgG4κ. The IgG4 antibodies incorporate the S228P hinge-stabilizing substitution.¹⁷ Six of the 52 antibodies are aglycosylated to reduce effector function, one is afucosylated to increase ADCC activity, and three have Fc amino acid substitutions to ablate or increase FcγR binding. The variable domains of these antibodies are either humanized (n = 32), derived from synthetic human antibody phage libraries¹⁸ (n = 16), human (n = 3), or mouse (n = 1; rituximab). Twenty-six of the antibodies recognize membrane associated targets with the remainder directed against soluble antigens. These antibodies showed a wide range of target-independent clearance values (Fig. 2), with 15 (29%) antibodies having clearance faster than 10 mL/day/kg.

Open in a separate window

Figure 2. Clearance values of antibodies (n = 52) in cynomolgus monkeys. Individual animal data and geometric mean values (red bars) are shown from dose groups where contribution of any specific clearance to the reported clearance was assumed to be negligible. Humanized (circle), human (square), synthetic human phage derived (triangle), and chimeric (diamond) antibodies are shown. Filled symbols indicate antibodies with a λ light chain (mAbs 2, 20, 29, 43), all others have a κ light chain. IgG4 isotype antibodies having the hinge-stabilizing mutant S228P are colored blue (mAbs 9, 24), all others are IgG1 isotype. Red symbol indicates an afucosylated antibody (mAb 3). Aglycosylated antibodies obtained via replacement of Asn297 with Ala are colored orange (mAbs 13, 14, 20, 33, 37, 46). Antibodies with Fc amino acid substitutions to modulate FcγR binding are colored green (mAbs 1, 19, 40).

The hypothesis that fast clearance is associated with the synthetic library antibodies was investigated. Antibodies derived from synthetic human antibody phage libraries were selected and optimized through in vitro selection.¹⁸ Most of these required affinity maturation through amino acid changes in their complementarity–determining regions (CDRs) to achieve the required affinity and potency. Although many of the humanized antibodies also have CDR substitutions relative to the rodent sequence either to improve affinity or chemical stability, in general the humanized antibodies were subjected to less in vitro optimization. A concern with in vitro selection is that it may inadvertently result in off-target interactions since the negative selection that can be performed in vitro is less extensive than can occur in vivo.¹³ Nonetheless, a comparison of humanized and synthetic library-derived human antibodies indicates an overlapping range of clearance values (Fig. 3). The median clearance value is 6.5 mL/day/kg for humanized antibodies and 9.0 mL/day/kg for human antibodies from synthetic phage libraries. These data do not provide evidence for an association between the synthetic library antibodies and fast clearance (p = 0.28, Wilcoxon test).

Open in a separate window

Figure 3. Comparison of clearance values in cynomolgus monkeys of humanized and synthetic human antibodies derived from phage display libraries. Chimeric antibodies (n = 1) and human antibodies from non-phage sources (n = 3) are not shown because of the very small size of the data sets. Plots show individual data values overlaid by boxplots. Boxplot rectangle shows the interquartile range (IQR) between the first quartile (25th percentile) and the third quartile (75th percentile); the thicker horizontal line within the boxplot rectangle is at the median (50th percentile). The lower boxplot whisker extends to the lowest data value that is still within 1.5 IQR of the lower quartile, and the upper whisker extends to the highest data value within 1.5 IQR of the upper quartile.

Binding affinity to cynomolgus monkey FcRn

To test whether variations in FcRn binding could account for the range of clearance values observed, we used surface plasmon resonance (SPR) methods to determine K_D values for binding of these antibodies to purified cynomolgus monkey FcRn at pH 5.8. Since IgG-FcRn binding is a relatively weak interaction, we used a steady-state approach described previously¹¹ to determine binding constants. Avidity effects arising from 2:1 FcRn:IgG interaction were avoided by immobilizing each of the antibodies (n = 44) that were available for analysis. K_D values determined in this fashion ranged from about 250 nM to 1,500 nM. Such a range was not unexpected because each antibody was separately immobilized using random coupling via amino groups, with a concomitant uncertainty in the orientation on the sensor chip surface. Multiple (n = 7) determinations of the K_D for pertuzumab indicated a mean of 940 ± 140 (s.d.) nM. The measurement method can detect changes in FcRn affinity because we could reproduce binding effects for Fc mutants known to increase affinity for FcRn.¹⁹

The faster clearance observed for some antibodies was not due to an altered low pH binding affinity for cynomolgus monkey FcRn (Fig. 4). Indeed, a comparison of several antibodies with approximately identical FcRn binding affinities gave a nearly 30-fold range in clearance values. Separate experiments (Fig. S3) indicated that the rate of dissociation of cynomolgus monkey FcRn from these antibodies at pH 7.4 was too fast for accurate measurement by Biacore technology. Although these experiments are insufficient to rule out a change in the recycling pathway for some antibodies as an explanation for the fast clearance observed, it appears unlikely that a significant change in binding affinity at pH 5.8 or in release kinetics at neutral pH played a major role in the fast clearance.

Open in a separate window

Figure 4. Relationship between antibody clearance in cynomolgus monkeys and binding affinity (K_D, pH 5.8) for cynomolgus monkey FcRn. Symbols as defined in Figure 2 are used. Dissociation constants (K_D) in nM were determined from steady-state analysis of SPR data as described by Yeung et al.¹⁹

Development of in vitro assay for risk mitigation of fast clearance

Fast clearance observed for a few antibodies has been associated with specific^14,15 or non-specific^13,16 off-target binding. During antibody generation and optimization, most of the antibodies in our panel were screened for lack of binding to closely related antigen homologs. In addition, many of the antibodies derived via phage display were selected in the presence of additives to prevent enrichment for non-specific binding.²⁰ As noted by Wu et al.,¹³ screening by ELISA with a limited variety of antigens may not detect the kind of specific or non-specific off-target binding that may influence in vivo behavior. Therefore, we sought to develop an in vitro assay useful for identifying candidate antibodies likely to show fast clearance. In the course of experiments using intact BV particles bearing specific membrane bound targets as antigen in ELISA assays,²¹ we noted that some antibodies bound to BV particles that did not express the target. Many of these antibodies were also shown to bind non-specifically to non-expressing human 293 cells by FACS (Fig. S4). However, expression of some cognate antigens on 293 cells made it challenging to distinguish specific from non-specific binding.

An initial screen against a test set of 15 antibodies selected from the panel used to generate data shown in Figure 2 suggested that fast clearing antibodies bound non-specifically in the BV ELISA. Subsequently, a larger group of 45 antibodies was tested in the assay. While there is considerable variability, a higher score in the in vitro assay is associated with faster clearance in cynomolgus monkey (Fig. 5). Thresholds of > 10 mL/day/kg for “fast” clearance and > 5 for normalized BV score were useful for quantifying these data. A clearance value of 10 mL/day/kg is more than 1 standard deviation from the mean value (6.5 ± 2.9) determined for 12 human IgG1 antibodies in monkeys.⁸ With these thresholds, 9 antibodies were correctly identified (“true positives”) as having fast clearance in monkeys, with only 3 “false positives”—antibodies with clearance values < 10 mL/day/kg that showed a normalized BV score > 5. Four antibodies were “false negatives,” having clearance greater than 10 mL/day/kg and a BV score < 5. The remaining 29 antibodies were correctly identified as “true negatives.”

Open in a separate window

Figure 5. Association of antibody clearance value in (A) cynomolgus monkey and (B) human with normalized BV ELISA score. A BV ELISA score > 5 is associated with increased risk of fast clearance in cynomolgus monkey (ρ = 0.53, n = 45). The BV score was calculated from the mean of 6 determinations; each determination was normalized by dividing by the signal observed for non-coated wells on the same assay plate. Of antibodies with a BV score < 5, 12% have clearance > 10 mL/day/kg in cynomolgus monkey, while clearance exceeds 10 mL/day/kg for 75% of antibodies with BV score > 5. A maximum likelihood estimate for the odds ratio is 19.5 [Fisher’s Exact Test, 95% Confidence Interval (3.3, 165.7)]. The confidence interval suggests a 3.3 to 166-fold increase in the odds of faster clearance for BV > 5; that the interval does not contain 1 implies statistical significance. (B) BV ELISA score > 5 is associated with increased risk of fast clearance in human (ρ = 0.83, n = 16).

Although the data set (n = 16) was smaller, an increased risk of fast clearance in humans was also associated with high BV score (Fig. 5B; Spearman’s ρ = 0.83). Five antibodies with clearance faster than 5 mL/day/kg in humans were correctly identified in the assay. Notably, one antibody, the anti-NRP1 antibody that had clearance < 10 mL/day/kg in cynomolgus monkeys was detected as a true positive for human clearance in the BV ELISA. One antibody having clearance > 5 mL/day/kg in humans, and > 10 mL/day/kg in cynomolgus monkeys, was not identified in the assay. The remaining 10 antibodies were correctly classified as having slow clearance in humans.

We tested whether the BV ELISA could be used prospectively to help aid candidate selection through further engineering of antibody 47. This is a humanized antibody with amino acid changes in the CDRs both for affinity improvement and to remove potential sequence liabilities. Cynomolgus monkey is a non-target-binding species for antibody 47. No reproducible tissue staining was observed in monkeys with antibody 47, but a clearance of 20.2 mL/day/kg in cynomolgus monkeys was determined for this antibody. A variant of antibody 47 that retained 2 of the affinity maturation changes, and an amino acid change to remove a potential N-linked glycosylation site in one of the heavy chain CDRs (antibody 47b), showed reduced binding in the BV ELISA (Table 1). Antibody 47c retains only the amino acid change to remove the potential N-linked glycosylation site and also had reduced BV binding. Separate experiments indicated that a large contribution to the BV binding of antibody 47 was made by a single amino acid change to remove a potential Asp-Gly isomerization site. Antibodies 47b and 47c had slower clearance relative to antibody 47 in cynomolgus monkeys (Table 1).

Table 1. BV ELISA signal and clearance of antibody 47 variants in cynomolgus monkeys

Variant	Number and purpose of CDR amino acid substitutions	Normalized BV ELISA Score	Clearance1 (mL/day/kg)
antibody 47	3-affinity maturation 2-chemical stability 1-removal of potential N-linked glycosylation	8.6	20.4 ± 3.6 (n = 4)
antibody 47b2	2-affinity maturation 1-removal of potential N-linked glycosylation site	3.1	9.3, 8.2 (n = 2)
antibody 47c	1-removal of potential N-linked glycosylation site	2.0	4.2 ± 1.4 (n = 4)

Open in a separate window

¹Pharmacokinetic data were collected for a single IV dose of 5 mg/kg, mean (± s.d.) values for clearance are shown. Serum concentrations of antibody were determined using ELISA with antigen coat. The non-specific clearance was calculated from the serum concentration data to day 42 by using a two-compartment model. ²Antibody 47b: 2 of 4 animals had evidence of ATA and data from these animals were not used in the calculation of clearance.

Discussion

The broad distribution of clearance values in cynomolgus monkeys measured for our panel of 52 antibodies, as well as the faster clearance (> 10 mL/day/kg) observed for 15 of the antibodies, was unexpected. Fast clearance likely involves internalization followed by catabolism, but we have not performed experiments to address this question. Since 40 out of the 52 antibodies have identical human IgG1 Fc sequences, it was expected that faster clearance would not be associated with changes in FcRn interaction. Indeed, although the K_D measured for FcRn binding at pH 5.8 varied over a seven-fold range, this was not associated with a trend in clearance values. Given that relatively large increases in pH 5.8 FcRn affinity lead to modest improvement in clearance,¹⁹ it is not surprising that the small differences in FcRn affinity determined here did not have an effect on clearance. All of the antibodies tested showed equivalent and rapid release from FcRn at neutral pH. In contrast to the results reported by Wang et al.,²² we did not detect an association between clearance and an affect of the Fab portion of the antibody on FcRn release kinetics at neutral pH. Our study used a larger panel of antibodies and a different orientation of the SPR experiments to minimize avidity effects on binding.

Given that the greatest difference between antibodies in our panel is in the variable domains, and in particular in the CDRs, it is likely that differences in the composition of these domains contributed to the variability in clearance behavior. Other groups have observed effects on clearance associated with changes in CDR sequence that were not related to differences in binding affinity to the target antigen.^13,16 We did not find that clearance was associated with isoelectric point (pI) or hydrophobicity of the intact antibody (Fig. 6A and B). Igawa et al.²³ have previously shown that clearance of a human IgG1 antibody in cynomolgus monkeys can be reduced through changes in the variable domain that result in a lower pI. This effect was observed at a dose where significant antigen-dependent clearance was operating. The mechanism was presumed to be non-FcRn dependent. For our large panel of antibodies, PK data was collected at doses where half-life is FcRn-dependent, and we found that clearance rates were not dependent on pI. A caveat to this conclusion is that the pI values of a majority of antibodies in this analysis are clustered in a relatively narrow range (7.5–9.5); it is possible that a trend would be observed if a wider range of pI was tested. In addition, we find that clearance is not correlated with total charge calculated for the antibody Fv (Fig. 6C). Although the absolute value of calculated and experimentally determined charge may differ,²⁴ the relative charge ranking of antibodies by experiment and calculation is similar,²⁵ suggesting that calculated charge is sufficient for this comparison. Thus, the physiochemical properties of the antibody that result in fast clearance are not easily recognized from sequence comparisons alone and may require more detailed structure-function analysis.

Open in a separate window

Figure 6. Relationship between antibody clearance in cynomolgus monkeys and antibody (A) pI, (B) hydrophobicity as measured by HIC, and (C) charge calculated for Fv domain.

We show here that a simple assay of non-specific binding can be used to identify antibodies with increased risk of having fast clearance in both humans and cynomolgus monkeys. Preliminary results indicate a similar correlation with clearance measured in rodents; however, the data set of clearance values in rodents is smaller, precluding a robust statistical analysis of the data. This assay, based on antibody binding to virus particles prepared from BV-infected insect cells, employs a simple ELISA format and can be applied in high throughput fashion. Because PK data are determined in mammalian species, an assay based on non-specific binding to mammalian cells might have been expected to show a stronger correlation with fast clearance. Preliminary results indicated a correlation between fast clearance and non-specific binding to human 293 cells detected by FACS, but broad application of this assay was precluded because many of the cognate antigens for this collection of antibodies are expressed on 293 cells. The BV ELISA assay we developed does not suffer from this target-expression limitation. For antibodies that can be evaluated with both assays, a similar trend in non-specific binding was observed (Fig. S4). Budded BV virions are stable nanoparticles that mimic infected cell surfaces, presenting a complex mixture of phospholipid, carbohydrate, glycoproteins, extracellular matrix and nucleic acids, as well as the viral capsid.^26,27 The antigen presumably also contains some insect cell debris, including protein, membranes and nucleic acids. Such a mixture that lacks the specific targets of therapeutic antibodies while retaining the same overall biochemical complexity as human cells and tissues has advantages in a non-specific binding assay because it allows detection of different kinds of interactions, e.g., electrostatic, hydrophobic, with different antibodies.

An important caveat to the association between antibody clearance in cynomolgus monkeys and binding to BV particles is that, other than for antibody 47, the observation is based on data collected through normal preclinical development, rather than a designed experiment in which BV-particle binding was systematically varied. Clearance values were obtained from separate studies with widely-varying dose levels and using varying PK assay types; however, where sufficient data was available to make an assessment, the association between fast clearance in cynomolgus monkeys and BV binding was found to persist across dose levels, assay types and antibody sources and target types.

The risk of fast clearance can be reduced by using our assay to minimize introduction of non-specific binding during antibody engineering. In antibody 47, both charge reduction and increased hydrophobicity were associated with increased non-specific binding to BV particles. Variants with decreased non-specific binding had slower clearance. The BV ELISA, however, is not expected to be a reliable assay of specific off-target effects. For example, the humanized antibody previously shown to have fast clearance in mice due to unintended specific binding to a complement protein¹⁴ had no detectable binding to the BV particles. In our experience, specific off-target effects are less prevalent than antibodies with broader off-target binding. Moreover, specific off-target effects can usually be identified through mechanistic studies of fast clearance.

The BV particle ELISA detects most, but not all, antibodies with rapid clearance in cynomolgus monkeys, with few false positives. Therefore, the assay should be considered as a screening tool to reduce the number of therapeutic antibodies that need to be tested in NHP PK experiments and not as an assay that consistently predicts clearance in cynomolgus monkeys. When multiple candidates with similar potencies against a target are available, such that a small but non-zero rate of false positives is acceptable, this assay is a cost effective tool to minimize risk of fast clearance. The BV ELISA assay could be enhanced if the basis for the false-negative results could be identified. Relative to membranes from mammalian cells, the phospholipid composition of insect cell membranes is enriched in phosphatidyl inositol, phosphatidyl choline, and phosphatidyl ethanolamine, with lower levels of the acidic phospholipid phosphatidyl serine, cholesterol, glyco- and sphingolipids.²⁶ The higher amounts of neutral phospholipids may account for the false negatives in the assay if the faster clearance of these antibodies is related to binding to acidic phospholipids on mammalian membranes. Two of the four false negative antibodies have a high net positive charge calculated for the Fv domain. It will be interesting to explore whether the phospholipid composition of the BV particles can be manipulated through growth conditions or cell line differences with a concomitant improvement in the veracity of the non-specific binding assay. These studies could further illuminate features to be avoided during antibody engineering to maintain acceptable PK behavior.

Materials and Methods

Supplementary Material

Acknowledgments

Disclosure of Potential Conflicts of Interest

All authors are current or former employees of Genentech, a member of the Roche group, and hold a financial interest in Roche.

Footnotes

References