Pre-existing Reactivity to an IgG4 Fc-Epitope: Characterization and Mitigation of Interference in a Bridging Anti-drug Antibody Assay

Twenty percent of baseline patient samples exhibited a pre-existing response in a bridging anti-drug antibody (ADA) assay for a human IgG4 monoclonal antibody (mAb) therapeutic. In some cases, assay signals were more than 100-fold higher than background, potentially confounding detection of true treatment-emergent ADA responses. The pre-existing reactivity was mapped by competitive inhibition experiments using recombinant proteins or chimeric human mAbs with IgG4 heavy chain regions swapped for IgG1 sequences. These experiments demonstrated that the majority of the samples had reactivity to an epitope containing leucine 445 in the CH3 domain of human IgG4. The pre-existing reactivity in baseline patient samples was mitigated by replacing the ADA assay capture reagent with a version of the drug containing a wild type IgG1 proline substitution at residue 445 without impacting detection of drug-specific, treatment-emergent ADA. Finally, purification on Protein G or anti-human IgG (H + L) columns indicated the pre-existing response was likely due to immunoglobulins in patient samples.


Introduction
Therapeutic proteins have the potential to generate an unwanted immunogenic response. Immune responses to a biotherapeutic can range from transient anti-drug antibodies (ADAs) with no clinical significance to the generation of high titer, persistent ADAs, which may lead to reduced drug exposure, loss of efficacy, or serious adverse events [1,2]. Therefore, an immunogenicity assessment is an important part of the safety testing for all therapeutic proteins and detailed recommendations on immunogenicity testing during various stages of drug development, including guidance from regulatory agencies, have been published [3][4][5][6].
The current industry standard for immunogenicity testing is the bridging immunoassay. In this assay format, the drug is labeled with two different tags and ADA in the sample forms a bridge between these labeled reagents. This bridging format has several important advantages, including the ability to detect different ADA isotypes, a reduced level of matrix interference, the ability to perform high throughput analysis, and the use of surrogate positive controls that are not species specific [7]. However, there are also drawbacks with this assay format. Soluble multimeric target and Fcinteractions between IgG4 mAbs can both generate a false positive signal in bridging assays [8][9][10][11][12]. In addition, preexisting antibodies that recognize the drug can also elicit a positive response [13][14][15][16]. These include anti-allotype, antihinge region, and anti-glycan antibodies, as well as autoantibodies such as rheumatoid factor (RF) and heterophilic antibodies that bind to the Fc region of an antibody [13,[17][18][19][20][21].
The presence of pre-existing antibodies may affect the determination of an appropriate ADA assay cut point or Michael A. Partridge and Jihua Chen contributed equally to this manuscript.
complicate the assessment of treatment-emergent ADA. In some cases, they can also have important consequences for clinical safety or efficacy. For example, pre-existing IgE antibodies specific for a post-translational modification (galactose-α-1,3-galactose glycan) on cetuximab have caused life-threatening infusion hypersensitivity reactions (18). In addition, decreased in vitro complement-dependent cytotoxicity of rituximab has also been observed in the presence of RF [22]. Finally, pre-existing anti-adeno-associated viral vector (AAV) antibodies have been shown to impact transgene expression in humans and non-human primates [23][24][25].
While there is not a consistent association between the presence of pre-existing antibodies and development of post treatment ADA, rheumatoid arthritis (RA) patients with pre-existing antibodies have been associated with higher ADA rates compared to other diseased populations [16]. The underlying mechanism for this is unknown, but the presence of autoantibodies including RF in these patients could be a contributing factor. Recently, various epitopes in the constant domain of IgGs have been identified that bind RF and other autoantibodies [26]. Therefore, a more complete understanding of the impact of pre-existing antibodies on treatment-emergent ADA and clinical response is needed to develop better immunogenicity management strategies and improve patient outcomes.
In the clinical program for the IgG4 monoclonal antibody (mAb) therapeutic REGN-H098P, immunogenicity was evaluated with an electrochemiluminescent bridging ADA assay. The baseline positive response rate in the initial patient population investigated was approximately 6% in the screening assay, very close to the target 5% rate recommended in health authority guidance document [3]. However, approximately 20% of patients in a second patient population had a pre-existing response at baseline. The assay signal from some pre-dose samples was more than 100-fold greater than normal background levels, indicating a population-specific cut point would be unable to mitigate this issue. In addition, the pre-existing signal would confound detection of treatment-emergent ADA, as any treatment-induced immunogenicity could be obscured by the high background signal, potentially impacting the safety and efficacy assessment of the drug.
The pre-existing reactivity was mapped using competitive inhibition experiments by spiking various human IgG Fc-containing recombinant proteins and human mAb chimeras in the REGN-H098P ADA assay. The results indicated that the baseline responses were directed to a portion of the Fc of human IgG4. Specifically, they recognized a sequence at the C-terminus of the CH3 domain of human IgG4 at an epitope containing leucine 445 (EU numbering). To our knowledge, this is the first report of pre-existing autoantibodies that recognize specific epitopes at the C-terminus of IgG4 heavy chain.
A version of REGN-H098P was engineered with a proline substitution at residue 445, corresponding to the natural IgG1 sequence. When used as the capture reagent in the bridging ADA assay, this reagent was able to mitigate the background signal in the majority of the baseline samples without affecting detection of ADA positive samples, thus allowing the assessment of any treatmentemergent ADA responses. Patients with pre-existing reactivity had clinical pharmacokinetics (PK), efficacy, or safety outcomes that were similar to the remainder of the study population.
Finally, to identify the potential serum protein(s) contributing to the pre-existing reactivity, serum samples with or without pre-existing reactivity were fractionated by both Protein G and anti-human IgG (H + L) affinity chromatography. The pre-existing reactivities were mainly found in the column elution fraction, indicating they were likely immunoglobulins.
All solutions for the REGN-H098P ADA assay, unless otherwise specified, were prepared in 1% BSA, 1X PBS. 10% BSA Diluent/Blocking Solution was from SeraCare (Milford, MA). 1X PBS was from Life Technologies (Grand Island, NY). 1.5 M Trizma base was from Sigma (St Louis, MO). Glacial acetic acid was from Thermo Fisher Scientific (Waltham, MA). Protein G agarose beads were from G Bioscience (St Louis, MO). Rabbit anti-human IgG(H + L) used for immuno-depletion was from Life Technologies. HRP-Rabbit anti-human IgG (H + L) used for western blot was from Abcam. Streptavidin-coated SECTOR® microplates, read buffer, and Sector Imager 2400 were from Meso Scale Discovery (MSD, MD, USA).

Human Serum Samples
Blood samples for the measurement of REGN-H098P ADA were collected from patients at baseline (prior to the administration of REGN-H098P) and at various time points after post-treatment specified in a clinical protocol. Informed consent from patients participating in clinical studies of REGN-H098P has been obtained prior to blood sample collection.

ADA Assay
The REGN-H098P ADA assay was developed and validated as per regulatory guidance documents [4,27]. The presence of ADA in patient samples was determined using a titerbased, bridging immunoassay using a 3-tiered approach. The bridging assay procedure employed biotinylated drug and ruthenium-labeled drug as the bridge components. Samples were subjected to an acid dissociation step (300 mM acetic acid) prior to incubation with the labeled drugs to improve drug tolerance (1:30 final MRD). Upon sample incubation, ADA present in a sample binds to both the labeled proteins forming immunocomplexes, which are then captured in streptavidin-coated plates and detected by an electrochemiluminescent (ECL) signal (Counts) measured on an MSD plate reader.
Competitive inhibition experiments were performed to map the REGN-H098P region recognized by the preexisting reactivity. Baseline samples were tested in the REGN-H098P ADA assay with and without spiking excess (40-100X) of competing proteins. The percent inhibition (%) of each sample was calculated using assay signals from the same sample tested with and without a spiked protein. In addition, selected baseline samples with high assay signals were serially diluted in negative control (NQC) serum and subsequently measured in the original REGN-H098P ADA assay to generate titration curves.

Protein G and Anti-human (H + L) Chromatography
One baseline sample with high background signal (positive sample) and one with normal background signal (negative sample) were 1:10 diluted in dilution buffer (10 mM Tris-HCl with 0.15 M NaCl, pH 7.4) and then incubated with 360 mM acetic acid for approximately 30 min at RT. The acidified samples were neutralized with 1 M Tris solution and loaded onto Protein G or rabbit anti-human IgG (H + L) columns. Columns were then washed with 10 mL 1X Dilution Buffer under vacuum. The proteins bound to the columns were eluted with 2 mL stripping buffer (0.1 M Glycine, pH 2.5) and then neutralized with 1.5 M Tris-HCl pH 8.8. The flow through, wash, and elution fractions were subsequently concentrated before being tested in the REGN-H098P ADA assay to identify the pre-existing reactivity containing fraction. To confirm the depletion of immunoglobulin by both Protein G and anti-human IgG (H + L) columns, each fraction was loaded onto 4-20% reduced SDS gel and was analyzed by western blot analysis with HRP-Rabbit-anti-human-IgG(H + L) at a 1:20,000 dilution.

High Positivity Rate in Baseline Study Samples
During early clinical development of REGN-H098P, a human IgG4k mAb therapeutic, the baseline positive rate of patient population 1 was approximately 6% in the screening assay (Fig. 1a). However, baseline samples for patient population 2 had an approximately 20% positive response, more than threefold greater than the rate for the previous population ( Fig. 1b). Furthermore, in some cases, the assay signal observed in baseline samples from this population was ~ 100-fold higher than the assay background.

Mapping Pre-existing Reactivity by Competitive Inhibition
Initial characterization experiments were performed using a small subset of samples (≤ 16) with high baseline reactivity in the REGN-H098P ADA assay. Interestingly, the preexisting signal was also detected when samples were tested in the ADA assay with unrelated biotin-or ruthenium-IgG4 mAbs swapped for the labeled REGN-H098P reagents (not shown). In addition, high baseline signal was reduced when samples were tested with excess unrelated human IgG4 mAb (not shown). This suggested that the signal was likely specific to the constant regions of the drug and was not due to target interference.
To more closely map the reactivity, a subset of 50 baseline samples from patient population 2 were tested in a competitive inhibition experiment by spiking a range of proteins with different human immunoglobulin sequences in the REGN-H098P ADA assay. The baseline samples selected spanned the range of ADA assay response observed in the population. The spiked proteins were selected from Regeneron's protein library and were specific for a variety of targets or ligands unrelated to the target of REGN-H098P. They included the drug (REGN-H098P), an IgG4 isotype control mAb1, an IgG1 mAb2, an IgG4 Fc-receptor fusion protein, an IgG4 Fab, and two IgG4 mAbs containing either a IgG1 CH3 domain or an IgG1 proline substitution at residue 445 (EU numbering).
In the competitive inhibition experiment, a reduction in signal (percent inhibition) indicated the spiked protein contained the epitope recognized by the pre-existing reactivity. Interestingly, the pattern of signal reduction in the assay was consistent for almost all of the 50 samples. Inhibition of assay signal (> 50%) was observed in 44-46 out of 50 individual samples where the spiked protein had a wild-type IgG4 Fc. Similar inhibition was not evident with spiked proteins where the Fc was absent or with an IgG1 Fc; all 50 samples spiked with IgG4 mAbs containing IgG1 CH3 domain (Fig. 2, mAb2, mAb4), and 48 out of 50 samples spiked with IgG4 mAb with a proline substitution at residue 445 (Fig. 2, mAb5), had minimal signal inhibition (< 50%). This indicated that the motif recognized by the majority of the individual samples was found in the IgG4 CH3 domain, and specifically contained leucine 445. We also observed that at least two individual samples had reactivity to a different epitope in the IgG4 CH3 domain (e.g., samples inhibitable by mAb5 but not mAb4, Fig. 2).

REGN-H098P mAbs Engineered with Reduced Pre-existing Reactivity
Different versions of REGN-H098P were subsequently engineered that contained wild-type IgG1 sequences in the Fc region (Fig. 3a). These REGN-H098P constructs included a chimera with an IgG1 CH3 (IgG1-CH3) domain and a molecule with a single proline substitution at residue 445 (L445P, Fig. 3b).   To confirm the pre-existing reactivity did not recognize these constructs, they were also tested in a competitive inhibition experiment with a small subset of 6 representative samples from the 50 tested in the initial mapping experiment. In addition, these six samples were tested with other human IgGs spiked in the ADA assay, including antibodies with different heavy chains (IgG1, IgG2, IgG3, containing a natural P445, Fig. 3b), different light chains (lambda), and wild-type IgG4 with the "CPSC" sequence that was not present in the hinge-stabilized mAb REGN-H098P [28]. Consistent with the previous mapping data, the three other human IgG subclasses (G1, G2, and G3) and the two REGN-H098P constructs without leucine at residue 445 in the CH3 domain failed to inhibit the pre-existing signal (Fig. 4).

Minimizing Pre-existing Reactivity-ADA Assay with REGN-H098P L445P Capture Reagent
Having identified the residue/motif responsible for the pre-existing reactivity in this population, we next wanted to mitigate these responses in the ADA assay. To do this, the REGN-H098P with L445P substitution (REGN-H098P L445P) was labeled with biotin or ruthenium and tested in the bridging ADA assay. When one or both of these engineered labeled drugs was substituted for the original labeled REGN-H098P, the pre-existing assay signal in tested samples were substantially reduced (not shown). To minimize changes to the original bridging assay, a modified assay was developed with only one alteration: incorporation of biotinylated REGN-H098P L445P as the capture reagent ( Fig. 5a and b). The same ruthenium-labeled REGN-H098P reagent and the same assay procedure were used to maintain consistency of signal between the two methods.
When the 137 baseline samples with a range of assay signal in the original ADA assay (135 samples with signal greater than the screening cut point) were tested in the modified method, with biotinylated REGN-H098P L445P and ruthenium labeled REGN-H098P, the reactivity was substantially decreased in the majority of the samples (25 samples with signal greater than the screening cut point). Importantly, the pre-existing reactivity was not unilaterally reduced overall, as the pattern of baseline signal for these samples in the modified method was not correlated with the response in the original assay ( Fig. 5c and d). This indicated that the change to the assay capture reagent specifically mitigated the pre-existing reactivity.
Cut points for the modified assay were established by analyzing a random selection of 150 baseline samples from patient population 2. The large number of samples used in the cut point experiments ensured that biological variability was captured from samples with a variety of signal in the original assay.

Performance Evaluation of Modified ADA Assay
To ensure the modified assay was still capable of detecting ADA formation after REGN-H098P treatment, a set of 22 post-dose samples from patients with confirmed treatmentemergent ADA responses in the original assay was subsequently tested in the modified method. All 22 samples were confirmed positive in the modified assay. Furthermore, the signal/noise (S/N) ratios for the samples were highly correlated between both assays (R 2 : 0.96, Fig. 6a). The ADA titer values for all samples were either the same or differed by only one dilution between the two assays (Fig. 6b).
In order to minimize the changes to the immunogenicity testing strategy for patient population 2, a two-assay analysis approach was implemented. First, all patient baseline samples were tested in the original ADA screening assay. Samples from patients that were negative at baseline continued to be tested in the original assay. However, samples from patients that were positive at baseline in the original assay were routed for testing in the modified assay (including retesting of the baseline sample; Fig. 7).
This two-step analysis approach permitted a comparison of the ADA incidence (and other clinically relevant endpoints) for patients with or without pre-existing reactivity. Importantly, there was no substantive difference in the treatment-emergent ADA incidences for the two groups of patients, with treatment-emergent incidence of 7.5% for patients tested in the original assay (pre-existing negative) and 6.6% for patients tested in the modified assay (preexisting positive).

Characterization of Pre-existing Reactivity
For the majority of patient samples, two lines of evidence suggested that endogenous immunoglobulins were causing the pre-existing reactivity in patient population 2. First, the causative agent was bivalent (at a minimum), as replacing one reagent in the REGN-H098P ADA assay mitigated the pre-existing response (Fig. 5b and d). Second, a subset of nine samples was analyzed in REGN-H098P titration assay and all were successfully titered. Furthermore, some samples demonstrated evidence of a hook effect in their titration curves, even for samples where the maximum signal was relatively low (~ 1500 counts, Fig. 8a and b).
To further investigate the nature of the pre-existing reactivity, samples from one patient with relatively high baseline signal in the original assay (but negative in the modified assay) were fractionated on either Protein G or rabbit anti-human IgG (H + L) columns. As a control, serum from a patient negative at baseline in the original assay was subjected to the same chromatographic procedures.
Western blot analysis of the flow-through, wash, and elution fractions confirmed that immunoglobulins were predominantly in the elution fractions from the Protein G and the anti-human IgG (H + L) columns. Furthermore, there was no obvious difference in the pattern of bands in the elution fraction from the negative and positive samples ( Fig. 8c and d). An aliquot of each of these three fractions, as well as a 1:10 dilution of the neat serum sample (mimicking the column loading solution), was then tested in the original ADA assay at minimum required dilution (MRD). For the positive sample, only the elution fraction from the Protein G or anti-human IgG(H + L) columns generated signal in the ADA assay similar to the loading solution ( Fig. 8e and  f). There was little or no reactivity in the ADA assay for all fractions from the negative sample. Protein G and antihuman IgG(H + L) columns are both specific for IgG, and bind poorly to IgA, IgD, and IgM [29,30]. However, the

Discussion
Immunogenicity assessment plays a critical role in the development of biotherapeutics, especially at the clinical stage. Treatment-emergent ADA responses can have a significant impact on the safety and efficacy of the drug. In this case study, using a bridging immunoassay, we identified an unusually high percentage of baseline samples from a study population with pre-existing reactivity to the drug. Multiple experiments were performed to map the epitope recognized by the pre-existing reactivity and to mitigate its impact on the evaluation of REGN-H098P immunogenicity by using a modified ADA assay incorporating a version of the drug with a wild-type IgG1 proline substitution at residue 445.
An alternative solution to mitigate pre-existing Fc reactivity would be to use the Fab fragment of REGN-H098P as the labeled reagents in the ADA assay. However, by using the approach described in this paper of modifying only one amino acid in REGN-H098P, changes to the full-length drug were minimized enabling capture of treatment-emergent/ boosted responses to all other epitopes on the drug. In addition, the use of Fab fragments in the assay may complicate immunogenicity interpretation by the detection of responses to neo epitopes, for example, anti-hinge antibodies [31], not present in the drug.
An important question is whether patients with pre-existing responses to a protein therapeutic are pre-disposed to develop ADA after dosing. One approach to examine this is to assess ADA incidence in treatment-emergent (baseline ADA negative) versus treatment-boosted (baseline ADA positive) patients. However, this comparison is imperfect because the treatment-boosted category requires samples to meet different and more stringent criteria (typically a ≥ four-fold increase in titer) than the treatment-emergent classification.
In this case study, the use of a two-assay testing strategy allowed direct comparison of the incidence of treatmentemergent ADA in patients with or without pre-existing responses. Importantly, patients with the pre-existing responses had very similar rates of post-dose ADA as the pre-existing negative population, despite the observed high baseline signal in the original ADA assay (and therefore elevated levels of autoantibodies).
The modified assay does not detect treatment-boosted responses to the epitope containing the L445 residue. However, the observed pre-existing reactivity was due to a naturally occurring autoantibody, not a typical ADA. The similarity in treatment-emergent ADA incidence in patients with or without pre-existing reactivity suggested that there was no evidence of epitope spreading or that patients with pre-existing responses were pre-disposed to develop REGN-H098Pspecific antibody responses. This is consistent with published data indicating the presence of pre-existing antibodies is generally not predictive of the development of post-dose ADA against human mAb therapeutics [16,32]. In addition, most of the ADA positive patients with pre-existing antibodies had low titer ADA responses with similar PK, safety, and efficacy outcomes (not shown).
Reported examples of autoantibody Fc epitopes in RA patients describe a range of specificities for different IgG isotypes [21]. In contrast, for this study, patients with preexisting autoantibodies were (almost) uniformly specific to one particular IgG4 Fc epitope. The L445 containing sequence is unique to the IgG4 isotype, as the IgG1, IgG2, and IgG3 heavy chains all have a proline at this residue. Proline contains a conformationally rigid cyclic side chain that generates a very different secondary structure at the C-terminus of IgG1/2/3 subclasses compared to the epitope in the IgG4 heavy chain [33,34]. The combination of a unique epitope created by leucine 445 residue and the exposed nature of this motif at the C-terminus of the IgG4 heavy chain likely contributed to the uniform pattern of reactivity in these patients.
Autoantibodies such as RF that bind to other antibodies have been documented in RA patients for decades [35,36]. Development of autoantibodies has also been documented in asthma patients [37]. Recently, patterns of RF reactivity to Fc epitopes have been associated with particular diseases [26]. In addition, elevated levels of IgG4 have been associated a rare fibroinflammatory illness [38]. However, although the pre-existing responses in one study population bound to an epitope on human IgG4, the autoantibodies were not themselves IgG4. Therefore, it remains unclear whether the pre-existing antibodies are a marker of a particular disease phenotype in these patients.
Analysis of significant numbers of baseline study samples is routinely performed to establish ADA assay cut points. We have observed the pre-existing reactivity reported here, to the IgG4 L445 residue, in numerous diseased and normal populations (albeit at lower levels), as well as responses to other IgG1 and IgG4 Fc motifs (e.g., Fig. 2 mAb 4 vs mAb 5). Furthermore, reactivity of autoantibodies to subclass specific sequences in human IgG has been reported in individuals with rheumatoid arthritis, including the proline found in the IgG1 heavy chain at residue 445 [21].
Statistical evaluation of ADA assay cut point data frequently demonstrates that baseline responses are not normally distributed [39]. Furthermore, it has been reported that outlier removal of these skewed responses may result in elimination of up to 15% of the values in a population [40]. These documented statistical data are consistent with the observed biological phenomena, suggesting that skewed cut point responses are at least in part derived from human Fc-specific reactivity of immunoglobulins (autoantibodies) found in the sera of various populations. There is variability in the frequency and intensity of the responses in different groups of patients, but the evidence suggests anti-Fc autoantibodies may be relatively common in various diseased populations (in addition to RA) as well as healthy individuals [34,35,37]. Removal of these values from cut point statistics will likely result in lower cut points, increasing the false positive rate detected and reducing the ability to identify clinically relevant responses [41]. For low-risk mAb therapeutics, modification of the ADA capture (and/ or detection) reagent to minimize pre-existing signal from autoantibodies may allow for better detection of clinically meaningful drug-specific ADA responses.

Conclusions
Pre-existing immunoreactivity is frequently detected in immunogenicity assays for clinical and nonclinical studies. In the REGN-H098P clinical program, approximately 20% of patients were positive in the screening ADA assay at baseline in one population. The presence of an elevated incidence of baseline positives, with high assay signal in some cases, would make assessment of true treatment-emergent ADA especially challenging, potentially confounding the safety and efficacy assessment of REGN-H098P in this patient population.
Using competitive inhibition experiments, the pre-existing reactivity was mapped to a motif at the C-terminus of human IgG4 heavy chain containing leucine at residue 445. Employing a combination of immunoassay and immunodepletion experiments, the reactivity was further characterized, and is likely due to an interaction with an endogenous immunoglobulin. To our knowledge, this is the first description of autoantibodies recognizing specific epitopes at the C-terminus of IgG4, predominantly directed to the L445 containing sequence.
Finally, a modified ADA assay was developed that replaced the biotinylated capture reagent with a version of REGN-H098P containing a proline substitution at residue 445, corresponding to the natural IgG1 (and IgG2/IgG3) heavy chain sequence at the C-terminus. The modified assay was able to mitigate the pre-existing response while still detecting treatment-emergent ADA directed to unique epitopes of REGN-H098P, permitting effective immunogenicity assessment of REGN-H098P in this patient population.