Introduction

Despite significant advances in cancer detection and treatment, mortality rates for certain malignancies (e.g., lung cancer) remain high (1). Pfizer has developed a bispecific antibody (PF-07257876) that targets CD47 and PD-L1 to antagonize innate and adaptive immune checkpoints (2). PF-07257876 is being tested in a Phase I study to evaluate the safety, pharmacokinetic (PK), pharmacodynamic, and clinical benefit in patients with selected advanced or metastatic solid tumors (ClinicalTrials.gov Identifier: NCT04881045). A key objective during clinical development of biotherapeutics is the detection and characterization of anti-drug antibody (ADA) responses, including neutralizing antibody (NAb) responses. Specific and sensitive ADA and NAb methods are required not only to describe the incidence of such responses, but more importantly to better understand their potential relationship to clinically meaningful impact on PK, efficacy, and safety. Because drug and target concentrations in study samples have the potential to interfere with ADA/NAb detection, key assay performance characteristics must include suitable drug and target tolerances. To support clinical development of PF-07257876, ADA and NAb assays were developed and validated for clinical study sample analysis. Because PF-07257876 is a bispecific antibody with distinct functional domains, Pfizer incorporated domain specificity testing in both the ADA and NAb assays, consistent with regulatory guidance (3). For ADA, a standard approach using drug fragments in the confirmatory assay, as previously described by others, was used (4, 5). For NAb, separate cell-based assays were developed to detect NAb against CD47 and PD-L1 binding domains of the bispecific antibody.

Due to challenges encountered in the development of functional cell-based assays, including lack of reproducibility and lack of drug tolerance, cell-based binding assays were developed to detect domain-specific NAb against PF-07257876, using electrochemiluminescence (ECL). The cell-based binding format was chosen over ligand-binding approaches to represent the pharmacologically relevant targets of the drug (i.e., in a native state, on a cell surface, and free of recombinant protein tags). While other technologies can be used to measure cell binding (e.g., flow cytometry), ECL was chosen for its specificity, sensitivity, robustness, and wide use across the bioanalytical industry. The robustness of ECL made it possible to develop two domain-specific assays in a relatively short timeframe, and its wide application provided flexibility in method externalization. While both NAb assays were based on the same platform and required similar performance characteristics, the development of each assay faced unique challenges. Among the difficulties encountered, the interference caused by NAb against an adjacent domain of the bispecific antibody was an unexpected finding that required thoughtful mitigation. Achieving drug tolerance in the CD47 NAb assay, while retaining domain specificity, was also challenging. To achieve drug tolerance, a bead-extraction and acid dissociation (BEAD) procedure was initially attempted. However, the standard BEAD procedure using whole drug to capture NAb resulted in an unacceptable loss of domain-specific sensitivity (in the form of false-negative results) in samples with mixed NAb species. To address this concern, an adaptation was made to the BEAD procedure to ensure capture of potential domain-specific NAb reactivity while retaining NAb detection even in samples containing NAb to both domains. This manuscript describes the development of the CD47 NAb assay, focusing on domain specificity challenges, solutions implemented, and implications for clinical study sample analysis.

Material and Methods

Reagents

The CD47 CHO-K1 cell line (cat. no. 60602) was purchased from BPS Bioscience (San Diego, CA) and confirmed by flow cytometry to express human CD47. The PD-L1 CHO cell line (cat. no. 60543) was purchased from BPS Bioscience (San Diego, CA). Ham’s F12 medium, heat-inactivated fetal bovine serum (FBS), penicillin-streptomycin, hygromycin B, G418, and TrypLE Express were purchased from Thermo Fisher Scientific (NY, USA). DMSO was purchased from Sigma-Aldrich (MO, USA). Read Buffer T (4X) surfactant-free and MULTI-ARRAY 96 HB plates (cat. no. L15XB-3) were purchased from Meso Scale Diagnostics (MD, USA). Domain-specific anti-PF-07257876 positive controls were produced at Green Mountain Antibodies (VT, USA). Characterization of domain specificity and neutralizing activity was performed using ligand binding assays at Pfizer. Final domain-specific positive control clones (anti-CD47 domain and anti-PD-L1 domain) were formulated in phosphate buffered saline (PBS) for long-term storage. PF-07257876, biotin-conjugated PF-07257876, biotin-conjugated anti-CD47 Fab, and Ru-conjugated PF-07257876 were produced at Pfizer (MA, USA). SpeedBeads™ streptavidin magnetic beads (cat. no. GE21152104010350) were purchased from Thermo Fisher Scientific (NY, USA). Pooled normal human serum, individual normal human serum, and individual disease human serum were purchased from BioIVT (NY, USA).

Generation of Anti-CD47 Fab

Variable regions of heavy and light chains from anti-CD47 Fab P01A11_75 were sub-cloned into mammalian expression vectors encoding for (1) Fab CH1 domain followed by 8xHis-tag and (2) Fab constant lambda domain. Resulting vectors encoding His-tagged Fab fragment were transfected into HEK Expi293 cells and expressed for 5 days using standard expression protocols. Culture media from expression was harvested, and Fab purified on HisTrap excel 5 mL column (Cytiva), followed by size exclusion chromatography to remove any aggregates and to buffer exchange the Fab into PBS.

Cell Culture and Cell Banking

CD47 CHO-K1 cells were cultured in standard tissue culture-treated flasks using Ham’s F12 medium, supplemented with 10% FBS, 1% penicillin-streptomycin, and 500 µg/mL hygromycin B. Cells were maintained in 37ºC, 5% CO2 humidified incubators using standard aseptic techniques. A master cell bank was generated by expanding cells purchased from BPS Bioscience. Cells were cryopreserved in FBS supplemented with 10% DMSO and stored in liquid nitrogen vapor. To support method development, a working cell bank was generated in similar fashion, using one vial of cells from the master cell bank.

Preparation of Assay Controls and Samples

Negative control (NC) pool consisting of drug naïve human serum from heathy donors was purchased from BioIVT (NY, USA). High positive control (HPC, 3 µg/mL) and low positive control (LPC, 1 µg/mL) were prepared in NC using domain-specific PC stock solutions in PBS. Drug tolerance (DT) samples were prepared by spiking a CD47 domain-specific PC in NC alone or in combination with PF-07257876. For DT samples, PC and PF-07257876 were prepared at 2× final concentrations and used to make a 1:1 mixture to achieve targeted concentrations. The PC and PF-07257876 mixtures were placed on a tube rotator at room temperature (RT) for 1 h to facilitate formation of PC:drug complexes. Following formation of PC:drug complexes, DT samples were used to prepare small aliquots and frozen for ≥ 16 h prior to use. Target interference samples were prepared by spiking soluble CD47 or PD-L1 in NC in the presence or absence of PC. Mixed NAb PC samples were prepared by spiking CD47 domain-specific NAb PC and PD-L1 domain-specific NAb PC in NC. Drug control (DC) was prepared by diluting Ru-conjugated PF-07257876 (hereafter referred to as system drug or SD) to desired concentration(s) in assay buffer (PBS + 1% BSA).

Sample Pretreatment

SpeedBeads™ streptavidin magnetic beads were washed with two buffers prior to conjugation with capture reagent. PBS + 0.05% Tween 20 (PBST) was used for the first wash. PBST + 0.1% bovine serum albumin (BSA) was used for the second wash. Following washes, beads were suspended in a solution of PBST + 0.1% BSA containing biotin-conjugated NAb capture reagent (PF-07257876 or anti-CD47 Fab). The conjugation of beads to the NAb capture reagent was carried out at RT on a tube rotator/invertor for 1.5 to 2.0 h. Following conjugation, beads were first washed with PBST, then washed briefly with 100 mM glycine, pH 3.0 (≤ 1 min), followed by PBST to ensure complete removal of any unbound reagent. Following washes, beads were suspended in PBST and transferred to 96 deep-well sample blocks.

Assay controls and study sample aliquots (100% serum) were thawed at RT and mixed with 600 µL 300 mM acetic acid, pH 3.0 for an initial MRD of 1:5. Acid dissociation was carried out at RT for 1 h, followed by neutralization using Tris-HCl, pH 9.0. Following dissociation, the entire volume of each well was used to suspend the beads conjugated to NAb capture reagent. The 96 deep-well sample blocks containing beads and acid-dissociated samples were placed on a temperature-controlled shaker to prevent beads from clumping. Following overnight shaking, beads were washed using a deep-well plate washer set to three cycles with PBST, followed by two cycles with PBS. Captured antibodies were eluted from washed beads using 100 mM glycine, pH 2.0, and eluted samples were immediately neutralized using a mixture of 1 M Tris, pH 9.0, and PBST + 1% BSA prior to testing in the ECL cell-based binding assay.

ECL Cell-Based Binding Assay

ECL cell-based binding assays were performed using MSD MULTI-ARRAY 96 HB plates (hereafter referred to as “high bind plates”). For titration of SD, CD47 CHO-K1 cells were collected from continuous culture and incubated with SD in V-bottom polypropylene (PP) plates to allow binding of drug to its target and to facilitate removal of unbound drug by washing cells with PBS. Cells were washed twice by resuspension with 200–250 µL PBS per well, centrifugation at 300 × g, and aspiration of PBS. Washed cells were transferred to MSD high bind plates in PBS and allowed to settle/attach. MSD Read Buffer T surfactant-free was added to each well, and the high bind plates were read on an MSD Sector S 600 instrument. For sensitivity, DT, and interference assessments, CD47 CHO-K1 cells were incubated with a 1:1 mixture of SD and sample (i.e., product of sample pretreatment).

Data Reduction

In this assay, controls and samples were tested in duplicate. For each control or sample, the mean of the raw signal (RLU) was calculated from duplicate wells. All controls and samples were converted to signal to DC ratios (S/DC) for further analysis. The DC was prepared by diluting SD to 1 µg/mL in assay buffer and mixing it with NC serum at a 1:1 ratio to achieve a final SD concentration of 500 ng/mL. The assay performance (i.e., degree of Ru-labeled drug binding to cell-surface CD47) was monitored using the ratio of DC/assay blank (cells incubated with assay buffer). The ratio of DC/assay blank was generally ≥ 30 throughout assay development and qualification at a SD concentration of 500 ng/mL.

Statistics

JMP (SAS Institute, Inc.) was used to perform outlier and normality assessments prior to calculating the cut point (CP) following industry best practices with a 1% false-positive rate using multiple drug naïve normal serum samples. In assay qualification, the CP was calculated for the normal and disease populations combined (30 individual samples of each population) using the formula CP = antilog [mean log(S/DC) of individual matrix samples − (2.326 × log(S/DC) SD)]. In assay validation, one CP was calculated for the relevant disease populations (30 individual samples). NAb levels in the cell-based binding assay are inversely correlated with S/DC ratios. That is, higher NAb levels are correlated with lower S/DC ratios. Assay sensitivity was estimated in qualification based on several titration curves and using the formula sensitivity = antilog [mean + t0.05, df × SD].

Results

Characterization of Assay Specificity Under Initial Conditions

In developing the ECL cell-based binding assay to detect NAb, our intention was to employ as straightforward a format as possible. As shown in Fig. 1, CD47 CHO-K1 cells do not produce an ECL signal in the absence of SD. However, when CD47 CHO-K1 cells are incubated with SD, binding to cell-surface CD47 occurs and produces a robust assay signal. In the presence of a NAb PC, binding of SD to CD47 is neutralized and the assay signal is diminished. While using ECL to detect NAb may provide certain advantages (e.g., sensitivity), using intact cells as a capture reagent adds a layer of complexity beyond traditional LBA approaches. To ensure robust assay performance, our development strategy was based on achieving critical assay parameters, such as obtaining DC/assay blank ratios ≥ 3 at SD concentrations that favor an acceptable level of sensitivity for NAb detection.

Fig. 1
figure 1

Schematic of ECL cell-based binding assay. Binding of SD to CD47 on CHO cells produces a robust ECL assay signal. Binding of SD to CD47 is inhibited by a NAb PC, resulting in a diminished ECL assay signal

Full size image

Thus, early efforts focused on evaluating SD concentrations to establish a suitable DC/assay blank ratio while ensuring specificity of SD binding to CD47 CHO-K1 cells. As shown in Fig. 2a, CD47 CHO-K1 cells incubated with a range of SD concentrations produced dose-dependent assay signals. As the cells produced minimal assay signal (< 100 RLU) in the absence of SD, the assay DC/assay blank ratios were demonstrated to be sufficiently robust at SD concentrations amenable to sensitive NAb detection. Based on the titration data, a SD concentration of 250 ng/mL was used for the remainder of method development, which consistently produced DC/assay blank ratios between 15 and 25, ensuring reproducible assay results and providing a wide dynamic range for NAb detection. To confirm that SD binding to CD47 CHO-K1 cells was specific, a comparison was made between SD and a Ru-labeled antibody that binds to unrelated cytokine targets (CHO specificity control, [CSC]). Similar to PF-07257876, the CSC is an antibody with multiple binding domains and was labeled with Ru under similar conditions. As shown in Fig. 2b, CD47 CHO-K1 cells incubated with the CSC produced minimal signal, whereas cells incubated with SD produced a robust signal. To further demonstrate the specificity of binding between SD and cell-surface CD47, a comparison was made between CD47 CHO-K1 cells and parental CHO-K1 cells. As shown in Fig. 2c, CD47 CHO-K1 cells produced a robust signal when incubated with SD, whereas parental CHO-K1 cells, lacking human CD47, produced minimal signal.

Fig. 2
figure 2

Specificity of SD binding to CD47 CHO-K1 cells. a Titration of SD in the CD47 cell-based binding assay. CD47 CHO-K1 cells were incubated with assay buffer or different concentrations of SD. Cells were washed, and assay signals were measured. b Comparison of SD to a CHO specificity control antibody (CSC). CD47 CHO-K1 cells were incubated with assay buffer, SD, or CSC (Ru-labeled antibody with multiple binding domains). Cells were washed, and assay signals were measured. c Comparison of CD47 CHO-K1 cells to parental CHO-K1 cells. CD47 CHO-K1 and CHO-K1 cells were incubated with SD. Cells were washed, and assay signals were measured

Full size image

Having demonstrated the specificity of SD binding to CD47 CHO-K1 cells, the ability of the assay to detect domain-specific NAb was next evaluated. SD was mixed with NC, CD47 NAb PC, PD-L1 NAb PC, or NAb PC for an unrelated biotherapeutic and allowed to form drug-NAb complexes. The NAb PC for an unrelated biotherapeutic was generated against a multidomain antibody using an approach similar to that used to produce the CD47 and PD-L1 NAb PCs. The mixtures were then incubated with CD47 CHO-K1 cells to assess binding. As shown in Fig. 3a, CD47 CHO-K1 cells incubated with SD produced a robust signal, whereas cells incubated with SD and CD47 NAb PC produced a greatly reduced signal, indicating domain-specific NAb detection. Note that in this experiment assay conditions (e.g., SD concentration) had not been optimized for maximal NAb detection. Nonetheless, the top-performing CD47 NAb PC (identified through screening efforts) reduced the assay signal by approximately threefold. As expected, the NAb PC for an unrelated biotherapeutic had no effect on the ability of the SD to bind to CD47 CHO-K1 cells, as evidenced by a robust assay signal. Surprisingly, cells incubated with SD and PD-L1 NAb PC produced a signal approximately tenfold higher than cells incubated with SD and NC (Fig. 3b); note the difference in scale (blue arrows) between panels a and b. Similar to the CD47 NAb PC, the PD-L1 NAb PC used in this experiment was identified as the top-performing PC through screening efforts. These data raised the possibility that under certain circumstances, such as the presence of NAb against both binding domains of PF-07257876, the cell-binding assay may lose its ability to detect NAb. Prior to addressing this issue, however, and given the complexity of cell-based ECL binding assay, we first sought to optimize other assay conditions for the method.

Fig. 3
figure 3

Ability of the CD47 CHO-K1 cell-based binding assay to detect NAb PC. Data shown in panels a and b were from the same experiment. a SD was mixed with NC (assay buffer sample), CD47 NAb PC, or NAb PC for an unrelated biotherapeutic and allowed to form complexes. Mixtures were then incubated with CD47 CHO-K1 cells at a density of 100,000 cells per well. Cells were washed, and assay signals were measured. b SD was mixed with NC (assay buffer sample) or PD-L1 NAb PC and allowed to form complexes as in panel a. Mixtures were then incubated with CD47 CHO-K1 cells at a density of 100,000 cells per well. Results were presented separately to illustrate the significant difference in signal response

Full size image

Optimization of Cell Density, System Drug Exposure, and Cell Settling Times

Cell-based NAb assays often require significant optimization to ensure the assays meet key performance characteristics. Cell density, duration of cell stimulation, and cell passage number are among the parameters that can significantly impact assay performance. Consequently, several assay parameters were optimized to ensure the assay was fit for purpose. As shown in Fig. 4a, greater cell densities (between 30,000 and 100,000 cells per well) produced higher signals with robust DC/assay blank ratios achieved at SD ≤ 0.6 µg/mL. However, to reduce SD concentration and deliver an acceptable assay sensitivity of approximately 300 ng/mL (Fig. 4b), a cell density of 200,000 cells per well was used in the final method. The ECL cell-based binding assay also involved two incubation steps: the first to expose cells to SD and the second to allow cells to settle to the bottom of the plate prior to ECL measurement. As shown in Fig. 4c, maximum signal was achieved when cells were incubated with SD for 60 min and allowed to settle for 60 min. By contrast, the greatest reduction in signal was observed when cells were incubated with SD for 30 min and allowed to settle for 30 min. However, the second incubation step, which allowed cells to settle prior to ECL measurement, had a greater effect on assay signal than the duration of cell incubation with SD. Based on these data, we completed development of the assay using 60 min for both steps. Lastly, cell passage number can affect assay performance. As shown in Fig. 4d, cells that had been cultured for approximately 3 months (passage 26) produced a reduced assay signal compared to cells that had more recently been introduced into culture. While the reduction in signal was not dramatic, we restricted the duration of cell culture to 3 months to ensure consistent assay performance.

Fig. 4
figure 4

Optimization of cell-based assay parameters. a Comparison of cell densities. Different numbers of CD47 CHO-K1 cells were incubated with assay buffer or SD at various concentrations. Cells were washed, and assay signals were measured. b Assay sensitivity for CD47 NAb PC. Pooled normal human serum was spiked with CD47 NAb PC at a range of concentrations between 3 and 0.05 µg/mL. Samples were processed using the sample pretreatment procedure outlined in Fig. 8. Eluents of NAb PC were mixed with SD prior to incubation with CD47 CHO-K1 cells. Cells were washed, and assay signals were measured. c Evaluation of SD incubation and cell settling times. CD47 CHO-K1 cells were incubated with assay buffer or SD for 60 or 30 min and allowed to settle for 60 or 30 min. Cells were washed, and assay signals were measured. Assay buffer data were omitted from the graph to facilitate illustration. d Comparison of cell passage numbers. CD47 cells that had been passaged 2 or 26 times were incubated with assay buffer or SD. Cells were washed, and assay signals were measured

Full size image

Pretreatment Optimization to Enhance NAb Detection and Achieve Drug Tolerance

To achieve drug tolerance in the CD47 cell-based binding assay, a sample pretreatment procedure to isolate and enrich NAb was necessary. Depending on the degree of drug tolerance required, bead-extraction and acid dissociation (BEAD) or solid-phase extraction with acid dissociation (SPEAD) may be employed (6, 7). To ensure accurate NAb analysis in clinical study samples, we targeted a drug tolerance level of 300 µg/mL, which was the highest projected residual drug concentration in serum at relevant sample collection times. As shown in Fig. 5, for the CD47 cell-based binding assay, we chose to incorporate a BEAD procedure to enrich NAb, initially using biotin-labeled drug as the capture reagent.

Fig. 5
figure 5

Schematic of bead-extraction and acid dissociation (BEAD) procedure, using biotin-labeled PF-07257876 as the capture reagent. Samples containing drug and NAb PC to one or both binding domains of PF-07257876 are treated with an acid to dissociate drug-NAb complexes. NAb are captured using biotin-labeled PF-07257876 coupled to streptavidin (SA) magnetic beads. After extensive washing to remove unbound molecules, SA magnetic beads are captured on a magnetic surface and NAb are eluted using a second acid treatment

Full size image

Using the approach shown in Fig. 5, the BEAD procedure would capture not only the CD47 NAb PC, but also the PD-L1 NAb PC from samples containing a mixture of both species. As suggested previously, capture of both NAb PCs had the potential to produce false-negative results in the CD47 cell-based binding assay. As shown in Fig. 6, CD47 CHO-K1 cells produced a robust assay signal when incubated with SD that had been mixed with negative control (NC) serum processed using BEAD. As expected, the assay produced a positive result (S/DC < CP) when cells were incubated with SD that had been mixed with only the CD47 NAb PC. Incubating cells with SD that had been mixed with a mixture of the two NAb PCs produced false-negative results (S/DC > CP), masking the effect of the CD47 NAb PC. The effect of the PD-L1 NAb PC was dose-dependent with a threshold at which interference was lost.

Fig. 6
figure 6

Interference caused by the PD-L1 NAb PC in the detection of the CD47 NAb PC. Negative control (NC) serum was left unspiked (0 µg/mL PC) or spiked with the CD47 NAb PC alone or in combination with the PD-L1 NAb PC at different concentrations. Samples were processed using BEAD to enrich NAb. Processed samples were mixed with SD to form complexes. Mixtures were then incubated with CD47 CHO-K1 cells. Cells were washed, and assay signals were measured. Blue arrows indicate positive results; red arrows indicate negative results. Dashed line indicates CP value (0.770) determined in qualification. The CP value was determined based on analysis of 30 normal and 30 disease samples

Full size image

Because the PD-L1 NAb PC caused interference in the CD47 cell-based binding assay, we wondered whether the CD47 NAb PC would have a similar effect in the PD-L1 cell-based binding assay. Using the sample pretreatment strategy described in Fig. 5, we evaluated the potential effect of the CD47 NAb PC alone on the cell-based binding assay (similar to Fig. 3b for the PD-L1 domain NAb PC). Interestingly, the CD47 domain NAb PC did not dramatically increase the assay signal as did the PD-L1 domain NAb PC. By contrast, the CD47-domain NAb PC caused a reduction in assay signal, resulting in a potential for false-positive results in samples lacking domain-specific NAb for the PD-L1 binding domain of PF-07257876 (Fig. 7). Similar to the PD-L1 domain NAb PC, the effect of the CD47 domain NAb PC was dose-dependent with a threshold at which interference was lost. However, unlike the PD-L1 NAb PC, significant interference in the assay required higher concentrations of the CD47 NAb PC.

Fig. 7
figure 7

Interference caused by the CD47 NAb PC in the PD-L1 cell-based binding assay. Negative control (NC) serum was left unspiked or spiked with the CD47 NAb PC at different concentrations. Samples were processed using BEAD to enrich NAb. Processed samples were mixed with SD to form complexes. Mixtures were then incubated with PD-L1 CHO cells. Cells were washed, and assay signals were measured. Dashed line indicates CP value determined in qualification (0.822). The CP value was determined based on analysis of 32 normal and 42 disease samples

Full size image

Adapting the BEAD Procedure to Selectively Enrich the CD47 NAb PC

As shown in Fig. 8, to avoid enrichment of NAb against both binding domains of PF-07257876, biotin-labeled anti-CD47 Fab was used to replace biotin-labeled PF-07257876 as the capture reagent. Using the approach shown below, the adapted BEAD procedure would enrich only the CD47 NAb PC, eliminating the potential for false-negative results in samples that contain a mixture of both NAb PCs.

Fig. 8
figure 8

Schematic of adapted bead-extraction and acid dissociation (BEAD) procedure, using biotin-labeled anti-CD47 Fab as the capture reagent. Samples containing NAb PC to one or both binding domains of PF-07257876 are treated with an acid to dissociate drug-NAb complexes. NAb are captured using biotin-labeled anti-CD47 Fab coupled to streptavidin (SA) magnetic beads. After extensive washing to remove unbound molecules, SA magnetic beads are captured on a magnetic surface and NAb are eluted using a second acid

Full size image

As shown in Fig. 9, CD47 CHO-K1 cells produced a robust assay signal when incubated with SD that had been mixed with an NC serum sample processed using the adapted version of BEAD. Using biotin-labeled anti-CD47 Fab as the capture reagent did not negatively impact the ability of the adapted procedure to enrich the CD47 NAb PC, as the sample containing only the CD47 NAb PC tested positive. Importantly, incubating CD47 CHO-K1 cells with SD that had been incubated with a mixture of the two NAb PCs did not produce CD47 NAb PC false-negative results (S/DC < CP), in contrast to the use of biotin-labeled PF-07257876 as the capture reagent (Fig. 6, S/DC > CP) at identical PD-L1 NAb PC concentrations.

Fig. 9
figure 9

Lack of interference caused by the PD-L1 NAb PC in the detection of the CD47 NAb PC. Negative control (NC) serum was left unspiked or spiked with the CD47 NAb PC alone or in combination with the PD-L1 NAb PC at different concentrations. Samples were processed using an adapted version of BEAD employing biotin-labeled anti-CD47 Fab to selectively enrich the CD47 NAb PC. Processed samples were mixed with SD to form complexes. Mixtures were then incubated with CD47 CHO-K1 cells. Cells were washed, and assay signals were measured. Dashed line indicates CP value (0.770) determined in qualification

Full size image

Having resolved the interference caused by the PD-L1 NAb PC, we wanted to confirm that the adapted sample pretreatment procedure resulted in adequate drug and soluble target tolerance. Drug interference in this assay would have the potential to produce false-negative results. As shown in Fig. 10a, the CD47 NAb PC tested positive in the assay when spiked into NC serum alone or in combination with a range of PF-07257876 concentrations up to 400 µg/mL. Soluble target interference in this assay would have the potential to produce false-positive results in samples lacking NAb. As shown in Fig. 10b, soluble CD47 or PD-L1 spiked into NC serum at concentrations up to 10 ng/mL did not produce false-positive results. In addition to drug and target interference, other sources of potential interference, including hemolysis and lipemia, were evaluated in validation.

Fig. 10
figure 10

Drug and soluble target tolerance of the CD47 cell-based binding assay. a Drug tolerance of the assay. Negative control (NC) serum was left unspiked or spiked with the CD47 NAb PC alone or in combination with PF-07257876 at various concentrations. Samples were processed using the adapted version of BEAD to selectively enrich the CD47 NAb PC. Processed samples were mixed with SD to form complexes. Mixtures were then incubated with CD47 CHO-K1 cells. Cells were washed, and assay signals were measured. b Soluble target tolerance of the assay. NC serum was left unspiked or spiked with soluble CD47 or soluble PD-L1 at various concentrations. Samples were processed using the adapted version of BEAD and mixed with SD to form complexes. Mixtures were then incubated with CD47 CHO-K1 cells. Cells were washed, and assay signals were measured. Dashed line indicates CP value (0.770) determined in qualification

Full size image

Discussion

A key objective during clinical development of biotherapeutics is the detection and characterization of potential ADA responses, including NAb responses. Specific and sensitive ADA and NAb methods are required not only to describe the incidence of such responses, but more importantly to better understand their potential relationship to clinically meaningful impact on PK, efficacy, and safety. ADA assays are primarily plate-based, while NAb assays can be plate- or cell-based, depending on several factors, including the makeup of the biotherapeutic, location of the drug target, and intended pharmacology (8). In general, for cell-surface drug targets, cell-based functional NAb assays are preferred as they may more closely represent the biological function of both the target and therapeutic agent. However, not all cell-based functional NAb assay formats are successful. This can be due to the complexity of the biotherapeutic (e.g., multidomain) as well as its intended pharmacology (e.g., multicellular interactions). These factors, among others, can result in a failure to achieve adequate specificity, sensitivity, tolerance to drug and soluble forms of the target, and overall reproducibility of the method. Significant efforts were made to develop functional cell-based NAb assays for PF-07257876 using reporter systems. However, the assays were ultimately unsuccessful due to lack of reproducibility in one system and lack of drug tolerance in the other.

A common challenge in the development of ADA and NAb assays is matrix interference. Two common sources of matrix interference are residual drug and soluble drug target (9, 10). However, other sources have been reported (11). While addressing matrix interference can be challenging in ADA assays, it is a particularly daunting task in NAb assays. The NAb assay platform (e.g., luciferase reporter, ECL) and format (e.g., functional cell-based, binding) can greatly influence the likelihood of successful mitigation of matrix interference. Due to challenges encountered with functional systems, a cell-based binding format based on the ECL platform was used for the NAb assays for both binding domains of PF-07257876. Separate cell-based binding NAb assays were developed for each of the bispecific antibody’s two binding domains to support clinical development.

While the development workflow was similar for the CD47 and PD-L1 NAb assays, we chose to focus on the development of the former to highlight a unique example of matrix interference. During the initial stages of development of the CD47 NAb assay, it became apparent that serum spiked with NAb PCs against both domains of PF-07257876 had the potential to produce false-negative results. Serum was spiked with both NAb PCs to simulate a patient sample containing a mixture of ADA to both binding domains of PF-07257876, as may be the case in a polyclonal antibody response. As assay development progressed, we tested the use of biotin-labeled whole drug as the capture reagent in a BEAD procedure to achieve drug tolerance (12). However, the use of biotin-labeled PF-07257876 captured CD47 and PD-L1 NAb PCs in samples containing a mixture of both, producing false-negative results in the cell-based binding assay. To specifically enrich the CD47 NAb PC, we substituted biotin-labeled whole drug with biotin-labeled anti-CD47 Fab as the capture reagent. We reasoned that the use of anti-CD47 Fab in sample pretreatment was justifiable, as the Fab contained the functional CD47 binding domain and was identical in amino acid sequence to that domain of the whole drug. Consequently, the use of the anti-CD47 Fab was relevant to potential neutralization and ensured fidelity of the sample pretreatment procedure to capture appropriate NAb. Furthermore, since the detection reagent was Ru-labeled PF-07257876 whole drug, the assay would retain its ability to detect NAb, while minimizing the potential for false-negative results in the event of an in vivo NAb response to both binding domains. Our adaptation to the BEAD procedure eliminated the potential for false-negative results, as serum samples spiked with the CD47 NAb PC and a range of PD-L1 NAb PC concentrations all produced positive results in the cell-based binding assay.

The choice of ECL as a platform for the CD47 cell-based binding NAb assay proved beneficial in several ways. Due to the low background and wide dynamic range characteristic of ECL, we were able to achieve robust DC/assay blank ratios at SD concentrations that favored assay sensitivity. Additionally, the robustness of the ECL platform enabled high well-to-well precision and assay-to-assay reproducibility, which facilitated efficient assay development. Finally, the use of ECL as an alternative to flow cytometry provided a measure of flexibility in externalization of the CD47 NAb assay. However, the ECL platform may have played a role in the interference caused by the PD-L1 NAb PC. We hypothesized that the interference caused by the PD-L1 NAb PC was due to a combination of the ECL platform and the binding characteristics of CD47 and PD-L1 binding domains. Our hypothesis was supported by the observation that the CD47 NAb PC also caused interference in the PD-L1 NAb assay (Fig. 7). The different nature and degree of interference in each assay may relate to the different affinity of each binding domain of PF-07257876 for its target. Irrespective of the mechanism of interference, our results underscore the importance of evaluating samples containing individual and mixtures of PCs in the development of NAb assays for multidomain biotherapeutics. Despite the challenges posed by the interference, we successfully developed a robust CD47 ECL cell-based binding assay capable of detecting 1 µg/mL of a NAb PC in the presence of clinically relevant drug concentrations up to 400 µg/mL.