Immunogenicity assay development and validation for biological therapy as exemplified by ustekinumab

Summary

Introduction of biotherapeutics has been a major milestone in the treatment of different chronic diseases. Nevertheless, the immune system can recognize the administered biological as non‐self and respond with generation of anti‐drug antibodies (ADA), including neutralizing ADA (nADA). Immunogenic responses may result in altered drug dynamics and kinetics leading to changes in safety and efficacy. However, there are several challenges with standard techniques for immunogenicity testing. Ustekinumab (UST), used in different inflammatory diseases, is a therapeutic antibody directed against the shared p40 subunit of interleukin (IL)‐12 and IL‐23, interfering in the pathogenically crucial T helper type 1 (Th1)/Th17 pathway. We established and validated different approaches for detection and quantitation of UST, UST‐specific ADA and nADA. Addressing the obstacle of complex formation of UST with nADA, we developed an acidification assay to approach the total amount of nADA. Validated methods were based on surface plasmon resonance spectroscopy (SPR), enzyme‐linked immunosorbent assay (ELISA) and a cell‐based approach to characterize neutralizing capacity of nADA. Parameters assessed were determination and quantitation limits, linearity, range, precision, accuracy and selectivity. Quantitation of ADA and UST was feasible at lower concentrations using ELISA, whereas SPR showed a wider linear range for determination of ADA and UST. Accuracy, precision and linearity for quantitation were comparable using ELISA, SPR and the cell‐based approach. All validated parameters fulfill the requirements of regulatory agencies. A combination of the testing approaches could address the increasing demand of precision medicine as it can be suitable for capturing the whole spectrum of immunogenicity and is transferable to other biologicals.

Introduction

Therapeutic monoclonal antibodies are characterized by a robust and reproducible mode of action. Targeting a defined structure, they show high therapeutic efficacy with a good safety and tolerability profile, which makes them suitable for use in different indications and in a wide patient population. Nevertheless, drug‐ and patient‐related factors modify immunogenic potential and can lead to an immune response towards the biological agent, resulting in the generation of anti‐drug antibodies (ADA). Here, a subset of neutralizing ADA (nADA) directly blocking the biological function of the drug can be distinguished. ADA can influence both kinetics and dynamics of the biological and may influence patient safety. As biotherapeutics increase in number, consequences associated with immunogenicity become increasingly relevant. In order to accurately classify the clinical relevance of the drug‐specific immunogenicity, a testing algorithm taking all relevant factors into consideration must be validated.

We present here a comparative validation of different measurement approaches and evaluate their suitability for an adequate assessment of the immunogenic potential of ustekinumab (UST). UST is a monoclonal immunoglobulin (Ig)G1 antibody directed against the p40 subunit, shared by the heterodimeric cytokines interleukin (IL)‐12 and IL‐23. It prevents p40 from binding to the receptor IL‐12Rβ1 expressed on T cells. This inhibits activation and initiation of the inflammatory T helper type 1 (Th1)/Th17 cell pathway, etiologically crucial in patients with psoriasis, psoriatic arthritis and Crohn’s disease 1. For the adequate testing of immunogenicity, the assessed methods were established or adapted to the requirements of the Food and Drug Administration (FDA) 2 and the European Medicines Agency (EMA) 3. Quality parameters validated in particular were detection and quantitation limits: limit of blank (LoB), limit of detection (LoD), limit of quantitation (LoQ), inter‐ and intra‐assay linearity (within a defined linear concentration range of the analyte), intra‐ and interprecision and accuracy of repetitive measurements, as well as the selectivity of the specific assay approach. The assay systems were optimized to cover a wide linear concentration range for the determination of UST, ADA and nADA. Focus was placed on reducing the minimal required dilution (MRD), in order to generate accurate signal responses for low analyte concentrations with a minimal background signal. We validated a newly established surface plasmon resonance spectroscopy (SPR) approach, as well as an enzyme‐linked immunosorbent assay (ELISA) which we optimized for the intended use to detect and quantify UST and UST‐specific ADA and nADA (Fig. 1). A newly developed ELISA‐based acidification assay was used to detect presence of nADA. Neutralizing capacity of nADA was characterized using an adopted and validated cell‐based assay. In summary, we present here different valid, easy‐to‐use and combinable approaches for assessing the immunogenic potential of a biotherapeutic, paving the way for a clinically relevant evaluation of immunogenicity and addressing challenges accompanied with it 4.

Figure 1.

Overview of assay set‐up using surface plasmon resonance (SPR) and enzyme‐linked immunosorbent assay (ELISA) for determination of ustekinumab (UST), anti‐drug antibodies (ADA) and neutralizing anti‐drug antibodies (nADA). In the SPR assay, UST detection was performed by immobilizing interleukin (IL)‐12 to the dextran matrix of the sensor chip surface. Binding of UST to IL‐12 induced an assay signal. Determination of ADA in SPR was performed by immobilizing UST. Binding of ADA, including neutralizing ADA (nADA), to UST led to an increase in assay signal. A competition assay was used for determination of nADA in SPR. UST at a constant concentration was incubated with varying concentrations of nADA. With increasing concentrations of nADA, UST binding to immobilized IL‐12 was reduced due to a complex formation with nADA, leading to a decrease in the assay signal. With ELISA, determination of UST was performed by coating a well plate with nADA. Binding of UST to nADA was detected with horseradish peroxidase (HRP)‐conjugated anti‐UST nADA. In order to quantify ADA in ELISA, UST was coated and binding of ADA was detected with HRP‐conjugated UST. For determination of nADA, a competition ELISA was used. UST at a constant concentration was incubated with different nADA concentrations. Binding of UST to coated IL‐12 was detected with an HRP‐conjugated mouse anti‐human IgG1 antibody. Increasing nADA concentration prevented the binding of UST and led, thereby, to a reduced assay signal.

Materials and methods

Equipment and software

Biacore™ T200, Biacore T200 Control Software version 2.0.1 and Biacore T200 Evaluation Software version 3.0 (all from GE Healthcare, Milwaukee, WI, USA) were used for running experiments and for data analysis of SPR measurements. Microsoft Excel 2013 (Microsoft Corporation, Redmond, WA, USA) and IBM spss statistics version 24.0 (IBM, Chicago, IL, USA) were applied for statistical analysis. Figures were created with GraphPad Prism 7 (GraphPad Software Inc., San Diego, CA, USA). Fluorescence intensity in ELISA and cell‐based methods was detected using EnSpire™ Multilabel Plate Reader (Perkin‐Elmer, Waltham, MA, USA).

Cells and antibodies

Human anti‐UST antibody (HCA210), horseradish peroxidase (HRP)‐conjugated detection antibodies (MCA647P and HCA210P) and the conjugation kit for HRP‐conjugation of UST were obtained from Bio‐Rad Laboratories (Irvine, CA, USA). HCA210 is a monoclonal IgG antibody directed against the binding site of UST. It was used in SPR and ELISA for validation of ADA and nADA testing. Recombinant human IL‐23 and IL‐12 were supplied by PeproTech (Rocky Hill, NJ, USA). IL‐23 assay‐ready cells and IL‐12 assay‐ready cells were purchased from Eurodiagnostica (iLite™, Uppsala, Sweden). UST (Stelara^®) was kindly provided by Janssen‐Cilag GmbH (Neuss, Germany).

Sample collection

The human blood samples used were collected from patients who had given their written informed consent for participation. The protocol was approved by the local Ethics Committee (approval number of the biobank: 199/15). In order to generate realistic results, samples of diseased patients were used. Samples were obtained from patients with psoriatic arthritis, naive to treatment with UST, and from patients with rheumatoid arthritis to provide rheumatoid factor (RF)‐positive sera. Serum was collected by centrifugation at 4°C and 1000g for 15 min and stored immediately at –20°C. Serum from UST‐naive diseased patients was used for determining validation parameters for LoB, LoD, LoQ and assay selectivity as well as precision and accuracy. For calibration curves and quality samples, we used purchased human serum (Sigma‐Aldrich, Schnelldorf, Germany).

Method validation

Method validation was carried out with particular regard to selectivity, linearity, precision, accuracy, LoB, LoD and LoQ, in compliance with the requirements and recommendations set by the FDA 2 and the EMA 3. For linearity, accuracy and precision, mean values along with 95% confidence intervals (95% CI) were calculated. In order to reduce matrix effects, samples were diluted to 1 : 5 in the ELISA for UST and ADA determination and 1 : 10 for detection of nADA. In the SPR and the cell‐based assays, samples were diluted 1 : 10. Therefore, all concentrations given in the text are diluted concentrations.

Linearity

Linearity refers to the ability of the assay to generate a response that correlates proportionately, within a defined range, to the analyte concentration in the sample. Calibration curves were prepared by serial dilution of a stock solution. Intra‐assay results were generated by the same investigator, using the same laboratory equipment. Interassay data were collected by two different scientists on different days and in different laboratories. The coefficient of variation (CV) of the slope was calculated as follows:

$$\text{CV}=\frac{\text{standard deviation (SD)}}{\text{mean}}$$

Assessment of linearity was performed by measuring a minimum of seven different concentrations in duplicates to sextuplicates in one assay run.

LoB, LoD and LoQ

LoB, LoD and LoQ were calculated assuming a Gaussian distribution, covering 95% of the generated values.

LoB defines the highest expected assay signal generated after repetitive measurements of serum samples that do not contain the analyte of interest (blank samples). LoB value was determined by calculating the mean value of the signal induced by blank samples and adding the 1·645‐fold standard deviation (s.d.) of the mean: $\text{LoB}={\text{mean}}_{\text{Blank}}+\text{1}.\text{645}\left({\text{SD}}_{\text{Blank}}\right)$.

LoD describes the lowest concentration that can be distinguished reliably from LoB: $\text{LoD}={\text{mean}}_{\text{Blank}}+\text{3}\left({\text{SD}}_{\text{Blank}}\right)$.

In distinction to LoD, LoQ represents not only detection but also a reliable quantitation of the analyte of interest: $\text{LoQ}={\text{mean}}_{\text{Blank}}+\text{1}0\left({\text{SD}}_{\text{Blank}}\right)$.

For an empirical determination of LoQ, low known concentrations within the lower linear concentration range of the substrate of interest were measured repetitively (five to six times). In accordance with the FDA, LoQ was accepted if repetitive measurements had precision and accuracy with a CV ≤ 25%.

Precision and accuracy

Precision defines the proximity of several measurements of a sample to the mean value. Determinations were performed by one scientist under the same conditions (intra‐assay precision) and by two scientists on different days (interassay). ELISA measurements were carried out in different laboratories with different equipment. SPR measurements were performed with the same Biacore T200 instrument but different equipment for interassay analysis.

Accuracy describes the relative difference between the true and measured values. A difference of ≤ 20% between the measured and true values was considered accurate. Determination of precision and accuracy was carried out with six to nine repetitive measurements for five different analyte concentrations, covering the linear concentration range of the analyte (low, medium and high concentration).

According to the FDA, repetitive measurements should have inter‐ and intraprecision with a CV of ≤ 20% and an accuracy between 80 and 120% of the expected result.

Selectivity

The ability of the assay to measure only the analyte of interest, despite the presence of interfering components in the sample matrix, is defined as selectivity. As methotrexate (MTX) is a common immunomodulatory co‐medication in immune‐mediated diseases, it was spiked at its highest expected serum concentration at 0·02 µg/ml (concentration before dilution) 5 into different analyte concentrations (low, medium and high concentrations) within the linear range of the substrate of interest. This resulted in differences in the analyte concentrations used, depending on the assay. For UST‐ELISA, UST (used at 0·125, 0·3675 and 0·525 µg/ml) was diluted 1:5 leading to the final concentrations: low (0·025 µg/ml), medium (0·0735 µg/ml) and high (0·105 µg/ml). For ADA‐ELISA, ADA (used at 0·25, 0·5 and 1 µg/ml) was diluted 1:5 leading to the final concentrations: low (0·05 µg/ml), medium (0·1 µg/ml) and high (0·2 µg/ml). For UST‐SPR, UST (used at 0·12, 0·25 and 0·735 µg/ml) was diluted 1:10 leading to the final concentrations: low (0·012 µg/ml), medium (0·025 µg/ml) and high (0·0735 µg/ml). For ADA‐SPR, ADA (used at 0·6, 1·25 and 10 µg/ml) was diluted, leading to the final concentrations: low (0·06 µg/ml), medium (0·125 µg/ml) and high (1 µg/ml). For nADA determination with the cell‐based assay, nADA (used at 0·4, 0·8 and 1 µg/ml and) was diluted 1:10 leading to the final concentrations: low (0·04 µg/ml), medium (0·08 µg/ml) and high (0·1 µg/ml). In all assays, MTX was used at an initial concentration of 0·02 µg/ml, as this represents the highest expected serum concentration. However, after dilution in the SPR assay (1 : 10) and the ELISA assay (1 : 5), final concentrations of 0·002 and 0·004 µg/ml were used, respectively.

Measurements were performed six to nine times for each concentration and compared with a two‐sided t‐test, assuming unequal variances and a significance level at 5% (α = 0·05). Analysis of assay interference due to the presence of rheumatoid factor (RF) in ELISA and SPR was achieved by comparing mean assay signals, induced by serum samples of six different RF‐positive patients with a mean assay signal response of 30–64 blank serum samples from RF‐negative psoriatic arthritis (PsA) patients in a two‐sided t‐test (α = 0·05).

SPR measurements

The target of the analyte of interest was immobilized to the dextran matrix of a CM5‐Chip (GE Healthcare). Contact time and flow rate for the binding of the ligand to the CM5‐Chip is described in detail under the corresponding heading. The measured signal was reported in response units (RU) and plotted against analyte concentration. Sample measurements were set with a sample contact time of 120 s and a flow rate of 10 µl/min. Regeneration was performed with glycine HCl (pH 1.5) for 45 s at a flow rate of 30 µl/min followed by a 45‐s stabilization period. Serum samples were diluted in HISPEC assay diluent (Bio‐Rad Laboratories) containing 10% non‐specific binding (NSB)‐Reducer (GE Healthcare), which is used to reduce non‐specific binding of the carboxymethyl dextran matrix of the CM5‐Chip to matrix components.

Determination of free UST with SPR

For UST detection, IL‐12 was covalently bound as a ligand via amine coupling to the dextran matrix of a CM5‐Chip (GE Healthcare). In brief, IL‐12 was diluted in sodium acetate buffer (pH 5.0) to 50 µg/ml. Assay temperature and compartment temperature were set at 25°C. Chip surface was activated with 1‐ethyl‐3‐(3‐dimethylaminopropyl) carbodiimide (EDC) and N‐hydroxysuccinimide (NHS). Contact time for IL‐12 was set at 840 s and the flow rate at 5 µl/min.

Determination of free ADA with SPR

For ADA detection in SPR, UST (180 µg/ml in acetate buffer pH 5.5) was immobilized in an analogous way to the amine‐coupling method to a CM5‐Chip (GE Healthcare) described above.

Determination of nADA with SPR

For nADA measurements, IL‐12 was immobilized via amine coupling to the dextran matrix of a CM5‐Chip. IL‐12 was diluted to an end‐concentration of 50 µg/ml in 10 mM sodium acetate (pH 5.0). Contact time for the ligand was 840 s, flow rate 5 µl/min. Determination of nADA in SPR was performed indirectly by measuring the decreasing assay signal induced by UST binding to the immobilized ligand IL‐12 in the presence of increasing anti‐UST nADA concentrations. The nADA HCA210 (Bio‐Rad Laboratories) was used, which specifically recognizes the monoclonal antibody drug UST and inhibits its binding to its target. Assay interference due to the presence of RF was detected by comparing mean assay signals induced by serum samples of RF‐positive patients with the mean assay signal response of blank serum samples. As IL‐12 was used in the UST and in the nADA assays as a ligand, the RF interaction was only tested in the UST assay.

ELISA measurements

Assay signal measurements in ELISA were performed with the QuantaBlu™ fluorogenic peroxidase substrate kit (Thermo Fisher Scientific, Waltham, MA USA). Quantablu substrate solution and Quantablue stable peroxidase solution were mixed at a ratio of 10:1 and 20 µl were added to the sample solution and incubated for 30 min at room temperature (RT). Peroxidase activity was stopped by adding 95 µl of the stop solution.

Determination of free UST with ELISA

For UST detection in ELISA, UST‐specific nADA [2 µg/ml in phosphate‐buffered saline (PBS)] were coated overnight at 4°C to the well plate. Well plates were blocked for 1 h with 100 µl PBS Tween 20 (PBST)/5% bovine serum albumin (BSA). Before and after the blocking step, the plate was washed five times with PBST. Binding of UST was detected with HRP‐conjugated (2 µg/ml in HISPEC diluent) anti‐UST nADA.

Determination of free ADA with ELISA

Plates were coated overnight at 4°C with 20 µl UST (1 µg/ml in PBS), washed with PBST five times and subsequently blocked for 1 h with 100 µl PBST/5% BSA and again washed five times. For assessment of calibration curves, ADA was spiked into diluted serum and detected with HRP‐conjugated UST at 2 µg/ml (diluted in HISPEC assay diluent). The relation of ADA and assay signal response was shown for a concentration range between 3 × 10^–5 µg/ml and 10 µg/ml of ADA (0·00003, 0·0003, 0·0015, 0·0031, 0·0062, 0·0125, 0·025, 0·05, 0·1, 0·2, 1 and 10 µg/ml).

Determination of nADA with ELISA

For nADA determination in ELISA, IL‐12 was coated (1 µg/ml in PBS) overnight at 4°C onto the well plate. After washing plates five times with PBST, blocking with 100 µl PBST/5% BSA for 1 h and again washing five times, the UST signal was measured in the presence of varying anti‐UST nADA concentrations. A mouse anti‐human IgG1 was used as detection antibody (2 µg/ml in HISPEC assay diluent) for UST. The signal induced by binding of UST to IL‐12 was measured at constant concentrations of UST of 1, 1·2 and 1·5 µg/ml after dilution to 1 : 10, leading to final concentrations of 0·1, 0·12 and 0·15 µg/ml. Anti‐UST nADA was used at concentrations varying between 0·006 and 2 µg/ml (0·006, 0·010, 0·015, 0·023, 0·034, 0·05, 0·078, 0·117, 0·175, 0·263, 0·395, 0·888, 1·333 and 2 µg/ml).

Determination of nADA in the acidification assay

The acidification assays were performed in ELISA. Each well of a 384‐well plate was coated with 40 µl of IL‐12 (0·5 µg/ml in PBS) and incubated overnight at 4°C. Plates were blocked with 100 µl PBST/5% BSA for 1 h. Before and after blocking, plates were washed five times with PBST. UST at a final concentration of 0·08 µg/ml was incubated for 1 h at RT with anti‐UST nADA at a concentration range between 0·005 and 0·16 µg/ml. Samples were divided into two different treatment groups. In one group, acid treatment and neutralization were performed. In the other group, the same volume of H₂O instead of acid and neutralization solution was added to each sample. Acidification was performed by adding acetic acid to a final concentration of 300 mM and incubating samples for 1 h at RT. Samples were then neutralized by adjustment with 1 M TRIS buffer (pH 9.5) to a final pH of approximately 6.5. After washing the plate twice, 40 µl of each sample were transferred in triplicate onto the coated plate and incubated at RT for 45 min. Wells were washed twice and then incubated with 40 µl HRP‐conjugated anti‐UST nADA (1 µg/ml in HISPEC assay diluent) for 1 h at RT. Determination of the fluorescence intensity was carried out as described above, using 40 µl Quantablu substrate solution and 40 µl stop solution.

Characterization assay

For characterization of the neutralizing capacity of nADA, a cell‐based approach – as recommended by the regulatory agencies 2 – was used. Genetically engineered IL‐12 iLite™ assay‐ready cells were used (Eurodiagnostica, Malmö, Sweden). Expression of firefly luciferase in this reporter cell‐line is induced by IL‐12, respectively, and inhibited in the presence of UST. It can be normalized to constitutively expressed renilla luciferase (Fig 6a). Further development and adaptation of the cell‐based assay for adequate validation of quantitation of the neutralizing extent of nADA was based on the available protocols for nADA determination with IL‐12 iLite™ assay‐ready cells. iLite™ assay‐ready cells were cultured in RPMI medium containing 10% heat‐inactivated fetal bovine serum (FBS) and 1% penicillin/streptomycin (culture medium). Serum samples were diluted 1:10 with culture medium.

Figure 6.

Cell‐based assay for characterization of neutralizing anti‐drug antibodies (nADA) directed against ustekinumab (UST). (a) (1) Binding of interleukin 12 (IL‐12) to IL‐12 receptor β1 (IL‐12‐Rβ1) induces a luminescence signal in a reporter gene assay. (2) UST can hinder receptor binding of IL‐12 and lead to a decrease in assay signal. (3) The presence of anti‐UST nADA prevents the action of UST as a blocker of IL‐12 signal induction. Assay signals are given in a firefly–renilla ratio (FRU) (ratio of firefly luciferase to renilla luciferase) plotted against the IL‐12 response shows a linear dose–response over the range 0·15–10 ng/ml. Representative data from intra‐assay measurements (n = 3) are shown. (c) UST at 0·1 µg/ml showed a linear signal for analyte concentration response using anti‐UST nADA in a concentration range between 0·029 and 0·15 µg/ml. Representative intra‐assay data are shown (n = 3). (d) Methotrexate (MTX)‐interference (0·002 µg/ml) was analyzed for low (0·04 µg/ml), medium (0·08 µg/ml) and high (0·1 µg/ml) nADA concentrations (at least n = 3). (f) Mean assay signal of rheumatoid factor (RF)‐negative (n = 27) serum samples showed no significant differences compared to RF‐positive (n = 6, determined twice) serum samples. Data are presented as mean ± standard deviation (s.d.).

An IL‐12 signal‐responsive curve was generated to determine the IL‐12 concentration for an optimal cell response. For this purpose, IL‐12 was spiked into diluted serum 1:10 with culture medium at a concentration range between 0·002 and 100 ng/ml (0·002, 0·005, 0·015, 0·046, 0·137, 0·412, 1·235, 3·704, 11·11, 33·33 and 100 ng/ml). Forty µl of each dilution were mixed with 40 µl of the IL‐12 assay‐ready cell suspension (1000 cells/µl) at 900 rpm for 10 s. The cell plate was incubated for 5 h at 37°C and 5% CO₂. Substrates (Dual‐Glo^® luciferase assay system, Promega, Madison, WI, USA) for firefly luciferase and renilla luciferase were added and incubated for 10 min, and the luminescence signal was quantified. Determination of sensitivity for UST detection in the cell‐based assay was obtained by establishing an UST response curve. For this purpose, diluted serum (1:10 with culture medium) was spiked with UST and incubated 1:1 with IL‐12 (18 ng/ml) for 30 min. After 1:1 addition of the cell suspension (1000 cells/µl), incubation was prolonged for another 5 h. The UST final concentration was measured over a concentration range between 0·003 and 0·112 µg/ml. Statistical calculation of detection limits for UST measurement was performed with 31 blank serum samples of UST‐naive PsA patients. For determination of the neutralizing capacity of anti‐UST nADA, UST was incubated for 1 h (37°C, 5% CO₂) at a final concentration of 0·1 µg/ml with nADA over a concentration range between 0·0097 and 5 µg/ml. Empirical lower and upper LoQ were achieved by repetitive measurement (n = 6) of low, medium and high nADA concentrations within the linear range. MTX was spiked into diluted serum samples already spiked with preincubated complexes of UST at a final concentration of 0·1 µg/ml and low, medium or high nADA concentrations. RF influence was tested by measuring 27 serum samples of UST‐naive PsA patients and six different sera of RF‐positive patients.

Statistical analysis

Signal measurements were tested with the Kolomogorov–Smirnov test at a significance level of α = 0·05 to decide whether a parametric or non‐parametric analysis was required. Significant differences in data sets with two groups were further analysed using the t‐test (parametric), as the data revealed a normal distribution. Influence of MTX on assay measurements was tested with a two‐sided t‐test and was rejected if no significant differences (α = 0·05) at low, medium and high analyte concentrations were found. We compared the signal profile of RF‐positive serum samples with the mean signal measurements in RF‐negative serum samples for significant differences in a two‐sided t‐test (α = 0·05).

Results

UST determination

Free UST can be reliably detected with SPR

In the context of immunogenicity, changes in the dynamics and kinetics of UST actions are crucial to evaluate the potential clinical consequences. An SPR assay was developed to enable detection of UST in serum samples from UST‐treated patients. Using this method, when the UST target, IL‐12, was immobilized to the chip surface a mean baseline response of 10 545.33 RU (95% CI = 8178.21, 12 912.45) (n = 11) was obtained (Fig. 1). MRD was reached at a dilution of 1:10. Higher dilutions showed no significant difference in signal‐to‐noise ratio. Assay sensitivity was determined by calculating the LoB, LoD and LoQ (Table 1). Compared to the UST standard concentration curve, calculation for LoQ was over‐estimated, due to the relatively high baseline signal in UST‐naive serum samples. Therefore, an empirical LoQ was determined providing values closer to the result expected based on the regression curve of the UST signal response (Table 1). Regulatory agencies recommend that the linear range of the analyte should cover 80–120% of the expected clinical concentration. Patients with plaque psoriasis who underwent an UST treatment regimen (45 mg; application interval = 12 weeks) have a median trough level of 0·20 mg/l (interquartile range = 0·023–0·4) when UST was measured just prior to UST injection 6. Therefore, an UST concentration range between 0·008 and 0·45 µg/ml led to a linear assay response with high precision and accuracy (Fig. 2a). The accuracy of the standard concentration curve was determined by calculating the deviation of the true value from the measured value as a percentage.

Table 1.

Comparison of validated parameters for different measurement approaches to UST determination

	Validated parameter	SPR measurements	ELISA measurements
Analytical sensitivity*	Limit of blank	0·002 µg/ml	0·0015 µg/ml
	Limit of detection	0·025 µg/ml	0·006 µg/ml
	Limit of Quantification	0·144 µg/ml	0·01 µg/ml
	Empirical limit of quantification	0·008 µg/ml	0·008 µg/ml
Linearity	Intra‐assay linearity	5·13% CV (n = 3)	3·13% (n = 6)
	Interassay linearity	4·25% CV (95% CI = 1·69, 6·82) (n = 25)	3·79% (95% CI = 2·01, 5·57) (n = 26)
	Range of linear signal response	0·008–0·45 µg/ml	0·008–0·15 µg/ml
Analytical specificity	Methotrexate	No significant assay‐interference	No significant assay‐interference
	RF‐IgM	No significant assay‐interference	Significant interference possible
Precision** and accuracy***	Repeatability	2·18% CV (95% CI = 1·00, 3·36) (n = 29)	3·11% CV (95% CI = 2·04, 4·18) (n = 45)
	Intermediate‐precision	2·57% CV (95% CI = 1·84, 3·30) (n = 122)	1·83% CV (95% CI = 0·69, 2·97) (n = 75)
	Accuracy	102·42% (95% CI = 96·10, 108·75) (n = 53)	94·97% (95% CI = 81·40, 108·53) (n = 50)

CV = coefficient of variation 95% CI: 95% confidence interval n: number of measurements; ELISA = enzyme‐linked immunosorbent assay; SPR = surface plasmon resonance; UST = ustekinumab; Ig = immunoglobulin; RF = rheumatoid factor.

^*Food and Drug Administration (FDA) recommendation: immunogenicity testing should have a sensitivity of 0·25–0·5 µg/ml (https://www.fda.gov/downloads/Drugs/Guidances/UCM192750.pdf).

^**FDA recommendation: % CV should be <20% (https://www.fda.gov/downloads/Drugs/Guidances/UCM192750.pdf).

^***FDA recommendation: accuracy should be within 20% (https://www.fda.gov/downloads/Drugs/Guidances/ucm368107.pdf).

Figure 2.

Calibration curve for ustekinumab (UST) determination using surface plasmon resonance (SPR) (a–c) and enzyme‐linked immunosorbent assay (ELISA) (d–f). (a) Concentration range of UST with linear assay response in SPR. A representative graph for intra‐assay measurement (n = 3) is shown. (b) Assay specificity in presence of a high methotrexate (MTX) concentration (0·002 µg/ml) was tested at low (0·012 µg/ml), medium (0·025 µg/ml) and high (0·0735 µg/ml) UST‐concentrations (at least n = 4). (c) Interference of measurements with rheumatoid factor (RF) was tested using RF‐negative (n = 29) and RF‐positive (n = 6, determined twice) serum samples. (d) Linear concentration range for UST determination using ELISA. Assay signals induced by UST at different concentrations were calculated and reported in a firefly–renilla ratio (FRU) (ratio firefly luciferase to renilla luciferase). A representative figure for intra‐assay measurement is shown (n = 3). (e) With ELISA, a MTX concentration of 0·004 µg/ml showed no significant interference with repetitive measurements of UST at low (0·025 µg/ml), medium (0·0735 µg/ml) and high concentrations (0·105 µg/ml) (at least n = 3). (f) Sera from RF‐positive (n = 6, determined twice) patients showed a significant increase in assay signal on a UST‐coated well plate compared to sera from RF‐negative (n = 30) patients. Data are presented as mean ± standard error (s.d.). ***P < 0·001 shows significant difference between RF‐positive and RF‐negative samples.

Because MTX is a common co‐medication to biologicals in several immune‐mediated diseases, its potential interference with UST determination within the SPR assay was tested. Quantitation of UST was not significantly (P > 0·05) disturbed by MTX at 0·02 µg/ml (highest expected serum concentration in patients 5) (Fig. 2b). As interference in immunogenicity assessment through presence of endogenous RF is a common phenomenon, the potential effect of RF on the UST‐SPR assay was tested by comparing the baseline signal of UST in RF‐positive serum samples (n = 6) with that in RF‐negative serum samples (n = 29). Comparison of the mean signal with UST‐naive serum samples showed no significant differences (P > 0·05) (Fig. 2c). In terms of suitability of an intelligent algorithm for immunogenicity assessment, our results indicate that the calculated sensitivity and selectivity of the assay qualify the SPR approach as a suitable assay for UST screening or confirmation.

Free UST can be reliably detected by ELISA

Because regulatory guidelines recommend using with two different assay systems for screening and confirmation, we established a second assay for the determination of UST using ELISA. In this approach, human IgG1 anti‐UST antibody (HCA210) was attached to the well plate and binding of UST was detected using an HRP‐conjugated anti‐UST antibody (Fig. 1). MRD for UST determination in serum samples was assessed for a 1:5 dilution and LoB, LoD and LoQ were determined (Table 1). LoQ was calculated with both the statistical and empirical approach as the latter showed a more realistic value in relation to the range for UST determination. UST detection was linear over a UST concentration range of 0·008–0·15 µg/ml (Fig. 2d). Repetitive measurements of different UST concentrations within the linear calibration range showed adequate precision and accuracy in terms of both intra‐ and interassay differences (Table 1). To investigate the selectivity of the assay, interference with MTX and RF was assessed. MTX at a concentration of 0·02 µg/ml showed no significant interference with UST measurement (Fig. 2e). Comparison of the mean assay signal of RF‐positive serum samples (n = 6) and that of UST‐naive and RF‐negative serum samples (n = 30) showed that the presence of RF was associated with a significant strengthening of the assay signal (Fig. 2f). In summary, free UST is detected reliably with ELISA but lacks specificity, as reflected by the possible interference with RF. The SPR approach is, thus, preferable over ELISA for RF‐positive serum samples.

Determination of ADA

Free ADA can be reliably detected with SPR

Screening of ADA is a crucial step in immunogenicity testing, as false‐negative results lead to under‐estimation of its prevalence. Furthermore, precise and accurate concentration measurement of ADA is essential to assess clinical relevance. In order to measure free UST‐specific ADA with SPR, UST was immobilized to the CM5‐Chip surface by amine coupling (Fig. 1). Mean immobilization was 15 128·96 RU (95% CI = 8178.21, 12 912.45) (n = 11). MRD for serum measurements was determined using a dilution of 1 : 10. Sensitivity of the assay was assessed by the calculation of LoB, LoD and LoQ (Table 2). A linear signal to analyte relation was shown for monoclonal anti‐UST IgG1 antibody over a concentration range of 0·009–2·5 µg/ml (Fig. 3a). Repetitive measurements showed excellent intra‐ and interassay precision and accuracy (Table 2). MTX did not result in significant interference with ADA concentration measurements (Fig. 3b). Furthermore, there was also no significant difference between assay signal induced in RF‐positive and ‐negative (n = 28) serum samples (P > 0·05) (Fig. 3c). Thus, the SPR‐approach for ADA detection satisfactorily fulfills the above‐mentioned criteria for adequate ADA determination.

Table 2.

Comparison of validated parameters for different measurement approaches to anti‐UST ADA determination

	Validated parameter	SPR measurements	ELISA measurements
Analytical sensitivity*	Limit of blank	0·001 µg/ml	0·0031 µg/ml
	Limit of detection	0·027 µg/ml	0·0034 µg/ml
	Limit of quantification	0·0806 µg/ml	0·0062 µg/ml
	Empirical limit of quantification	0·0056 µg/ml	0·003 µg/ml
Linearity	Intra‐assay linearity	5·89% CV (n = 6)	3·52% CV (n = 5)
	Interassay linearity	4·49% CV (95% CI = 2·15, 6·82) (n = 31)	3·48% CV (95% CI = 0·34, 7·31) (n = 16)
	Range of linear signal response	0·009–2·5 µg/ml	0·003–0·2 µg/ml
Analytical specificity	Methotrexate	No significant assay‐interference	No significant assay‐interference
	RF‐IgM	No significant assay‐interference	No significant assay‐interference
Precision** and accuracy***	Repeatability	2·32% CV (95% CI = 0·40, 4·24) (n = 30)	4·87% CV (95% CI = 2·88, 6·86) (n = 50)
	Intermediate‐precision	3·23% CV (95% CI = 2·25, 4·23) (n = 222)	3·8% CV (95% CI = 3·81, 2·95) (n = 97)
Accuracy	100·19% (95% CI = 92·04, 108·35) (n = 50)	94·62% (95% CI = 85·38, 103·86) (n = 61)

CV = coefficient of variation 95%CI: 95% confidence interval n: number of measurements; ELISA = enzyme‐linked immunosorbent assay; SPR = surface plasmon resonance; UST = ustekinumab; ADA = neutralizing antibodies.; Ig = immunoglobulin; RF = rheumatoid factor.

^*Food and Drug Administration (FDA) recommendation: immunogenicity testing should have a sensitivity of 0·25–0·5 µg/ml (https://www.fda.gov/downloads/Drugs/Guidances/UCM192750.pdf).

^**FDA recommendation: % CV should be <20% (https://www.fda.gov/downloads/Drugs/Guidances/UCM192750.pdf).

^***FDA recommendation: accuracy should be within 20% (https://www.fda.gov/downloads/Drugs/Guidances/ucm368107.pdf).

Figure 3.

Calibration curve for anti‐drug antibodies (ADA) determination using surface plasmon resonance (SPR) (a–c) and enzyme‐linked immunosorbent assay (ELISA) (d–f). Mean assay signals induced by ustekinumab (UST)‐specific ADA at different concentrations were calculated and are reported in response units (RU) for SPR or in fluorescence intensities for ELISA. (a) Concentration range of ADA with a linear assay response in SPR. Representative data for intra‐assay measurements are shown (n = 2). (b) Interference with measurements in SPR was tested for methotrexate (MTX) at low (0·06 µg/ml), medium (0·125 µg/ml) and high (1 µg/ml) concentrations of ADA (at least n = 3). (c) Assay signals for rheumatoid factor (RF)‐negative (n = 28) serum samples were tested on a sensor chip immobilized with UST and compared to RF‐positive (n = 6, determined twice) samples. (d) Linear concentration range for ADA determination in ELISA. Representative data of intra‐assay measurements are shown (n = 3). (e) MTX interference was tested using repetitive ELISA‐measurements at low (0·05 µg/ml), medium (0·1 µg/ml) and high (0·2 µg/ml) ADA concentrations (at least n = 3). (f) RF interference was tested using RF‐negative (n = 64) and RF‐positive (n = 6, determined in triplicate) samples on a well plate coated with UST. Data are presented as mean ± standard deviation (s.d.).

Free ADA can be reliably detected by ELISA

A further approach to assess UST‐specific ADA was taken with a bridging ELISA. Here, UST was coated onto the well‐plate and binding of UST‐specific ADA was detected with an HRP‐conjugated secondary UST antibody (Fig. 1). First, MRD was determined for a 1:5 serum dilution. A linear signal response was shown over an ADA concentration range of 0·003–0·2 µg/ml (Fig. 3d). LoB, LoD and LoQ as well as precision and accuracy of repetitive ADA measurements fulfilled the criteria of the regulatory agencies for immunogenicity testing (Table 2). Furthermore, MTX had no significant influence on ADA concentration measurements in ELISA (Fig. 3e). Serum from RF‐positive patient samples (n = 6) also showed no significant assay interference when compared to 64 RF‐negative serum samples (Fig. 3f). In conclusion, ELISA meets the requirements for a valid and user‐friendly tool for ADA detection and quantitation.

Determination of nADA

Free nADA can be reliably detected with SPR

UST‐specific nADA were determined with a competitive assay developed for SPR. IL‐12 was covalently bound to the dextran matrix of a CM5‐Chip, allowing UST‐specific induction of an assay signal (Fig. 1). Addition of nADA prevented binding of UST to IL‐12 in a concentration‐dependent manner and resulted in a decrease in assay signal. The detectable amount of nADA is dependent on UST concentration. A UST concentration providing reproducible results when spiked into a sample already containing UST was used. As determination of nADA was congruent over a UST concentration range between 0·1 and 0·15 µg/ml (Fig. 4a), UST was incubated at a constant concentration of 0·1 µg/ml with anti‐UST nADA at varying concentrations (0·008–1 µg/ml) for at least 1 h at RT to simulate the in‐vivo situation. For nADA detection at 0·1 µg/ml UST (Fig. 4b), high linearity, precision and accuracy in intra‐ and interassay determinations was shown for a range of 0·008–0·1 µg/ml anti‐UST nADA. Interference for MTX was not tested, as non‐influence for UST and HCA210 detection was already shown in the above‐described SPR‐experiments (Fig. 2b, Fig. 2b). Summarizing the assessments, SPR approach qualifies as a suitable confirmatory assay.

Figure 4.

Calibration curve for anti‐drug antibodies (ADA) determination using surface plasmon resonance (SPR) (a,b) and enzyme‐linked immunosorbent assay (ELISA) (c,d). (a) SPR assay signal (response units, RU) was induced through unobstructed binding of ustekinumab (UST) to the immobilized interleukin (IL)‐12. Influence on the linear concentration range in neutralizing antibodies (nADA) determination in relation to different UST concentration was compared. UST at a constant concentration of 0·1 µg/ml, 0·12 µg/ml or 0·15 µg/ml was incubated with varying concentrations of anti‐UST nADA. The same linear range for an indirect anti‐UST nADA (0·052 µg/ml–0·39 µg/ml) detection could be determined for all three UST concentrations. (b) UST at 0·1 µg/ml was incubated with varying concentrations of neutralizing antibodies (nADA) directed against the active binding region of UST. Assay signal was induced through unobstructed binding of UST to the assay target. (b) Concentration range of anti‐UST nADA providing a linear assay response induced by UST binding in SPR. Assay signals were calculated and reported in response units (RU). Representative data from intra‐assay measurements (n = 5) are shown. (c) Linear concentration range for anti‐UST nADA determination in ELISA. Assay signals were calculated and reported in fluorescence intensities. Representative data of intra‐assay measurements (n = 9) are shown. (d) Interference by RF was tested in the ELISA assay for anti‐UST nADA coated well plates comparing rheumatoid factor (RF)‐negative (n = 64) and RF‐positive (n = 6, determined twice) serum samples. Data are presented as mean ± standard deviation (s.d.).

Free nADA can be reliably detected by ELISA

With the objective of offering an alternative approach to nADA confirmation, we developed an ELISA assay for nADA determination. Here, IL‐12 was coated to the well plate and, again, a constant concentration of UST was incubated with varying concentrations of anti‐UST nADA. In theory, binding of UST to the well plate should be inhibited in direct relation to the amount of nADA in the sample (Fig. 1). Inhibition of UST binding was detected as a reduction of the assay signal. MRD was determined at a 1 : 10 dilution. At a constant UST concentration of 0·1 µg/ml, inhibition was linear over an anti‐UST nADA concentration range between 0·052 and 0·395 µg/ml (Fig. 4c). Empirical lower and upper LoQ were determined. Linearity, precision and accuracy were calculated for intra‐ and interassay measurements (Table 3). RF‐positive sera did not interfere with the assay on the IL‐12‐coated well plate when compared to RF‐negative sera (Fig. 4d). Thus, ELISA provides an accurate and specific determination of free UST‐specific nADA.

Table 3.

Comparison of validated parameters for different measurement approaches to anti‐UST nADA determination

	Validated parameter	SPR measurements	ELISA measurements	Cell‐based‐measurements
Linearity and range	Intra‐assay linearity	5·24% CV (n = 6)	4·57% CV (n = 5)	4·29% CV (n = 3)
Interassay linearity	1·44% CV (95% CI = –1·27; 4·17) (n = 17)	8·25% CV (95% CI = 2·86, 13·61) (n = 14)	4·39% CV (95% CI = –1·36, 10·14) (n = 9)
Linear range for nADA measurement in presence of 0·1 µg/ml UST	0·008–0·1 µg/ml	0·052–0·395 µg/ml	0·029–0·155 µg/ml
Precision* and accuracy**	Repeatability	1·44% CV (95% CI= –1·28, 4·17) (n = 62)	2·21% CV (95% CI = 2·21, 1·01) (n = 141)	4·74% CV (95% CI = 2·02, 0·71) (n = 45)
Intermediate‐precision	1·67% CV (95% CI = 1·17, 2·16) (n = 110)	3·67% CV (95% CI = 2·48, 4·86) (n = 224)	5·07% CV (95% CI = 3·72, 6·42) (n = 91)
Accuracy	99·92% (95% CI = 96·01, 103·83%) (n = 96)	103·12% (95% CI = 98·10, 108·14) (n = 30)	99·86% (95% CI = 94·99, 104·74) (n = 75)

CV = coefficient of variation 95%CI: 95% confidence interval n: number of measurements; ELISA = enzyme‐linked immunosorbent assay; SPR = surface plasmon resonance; UST = ustekinumab; nADA = neutralizing anti‐drug antibodies.

^*Food and Drug Administration (FDA) recommendation: % CV should be <20% (https://www.fda.gov/downloads/Drugs/Guidances/UCM192750.pdf)

^**FDA recommendation: accuracy should be within 20% (https://www.fda.gov/downloads/Drugs/Guidances/ucm368107.pdf).

Complexed nADA can be reliably determined in an acidification assay

In vivo, complex formation between UST and UST‐specific ADA may occur. In this case, the binding sites of both antibodies would be covered and thus unable to bind to the assay target and induce a signal. In this scenario, the amount of UST and nADA would be undetected or under‐estimated with SPR or ELISA.

If nADA/UST complexes are present in a sample, acidification of the samples induces a dissociation of the complexed UST and this would lead to an increase in assay signal compared to the signal produced in the untreated sample.

For acidification, we coated the well plate with IL‐12. UST at a constant concentration of 0·08 µg/ml, a concentration expected in patients treated with UST 6, was incubated with varying concentrations of nADA. In order to screen for the presence of nADA, we included an acidification step that resulted in dissociation of the complex between UST and anti‐UST nADA and enabled UST binding to IL‐12 (Fig. 5a). Acid treatment of the samples was followed by a neutralization step. We compared the signal profile of acidified with untreated samples. After acid treatment of serum samples, increased tolerance of UST measurement towards the presence of high nADA concentrations was observed (Fig. 5b). UST detection in untreated samples was not distinguishable from treated samples if the nADA : UST ratio was below 1:10. UST detection steadily decreased in non‐acidified samples above a 1:10 ratio and was not detectable above a 1:1 ratio. In treated samples, UST detection was possible at a nADA:UST ratio from 1:10 to 10:1. Thus, our results indicate that acidification is a suitable method to disengage complex formation and determine otherwise under‐reported UST or nADA. However, exact quantification is not possible with this method.

Figure 5.

Determination of bound neutralizing anti‐drug antibodies (nADA). (a) Bound nADA were measured indirectly by integrating an acidification step in a sandwich enzyme‐linked immunosorbent assay (ELISA). Ustekinumab (UST) was incubated at a constant concentration with varying concentrations of nADA. UST bound to an interleukin (IL)‐12‐coated well plate was detected by horseradish peroxidase (HRP)‐conjugated UST‐specific nADA. Induced signal was measured before and after acidification; (b) 0·08 µg/ml UST were incubated with anti‐UST nADA (0·0015–1·6 µg/ml). Samples were divided into two standard curves, one generated on acidification treatment with acidic acid (●) and one in the absence of treatment (○). Acidification led to a dissociation of the UST/anti‐UST nADA complex and released the active binding site of UST. UST bound to the coated IL‐12 and induced an assay signal. Representative data for intra‐assay measurements are shown (n = 3). Data are presented as mean ± standard deviation (s.d.).

Characterization of nADA (cell‐based assay)

In general, a cell‐based assay adequately mimics in‐vivo drug neutralization. Thus, the FDA recommends the use of a cell‐based assay for the characterization of nADA. We used a commercially available, cell‐based assay (iLite™ IL‐12 assay‐ready cells; Eurodiagnostica) containing IL‐12‐responsive cells that induce a detectable firefly luciferase response when IL‐12 binds to their over‐expressed IL‐12Rβ1 receptor (Fig. 6a/1). Addition of UST captures IL‐12, and therefore prevents expression of the luminescence signal in a concentration‐dependent manner (Fig. 6a/2). Conversely, the presence of nADA prevents capturing of IL‐12 by UST and thus enables induction of the assay signal (Fig. 6a/3). MRD for serum measurements with the cell‐based assay was determined for a 1:10 dilution in culture medium. As an IL‐12 concentration of 4·5 ng/ml results in a signal response within the saturation area, this concentration was used for further experiments (Fig. 6b). At a constant IL‐12 concentration of 4·5 ng/ml and a constant UST concentration of 0·1 µg/ml, assay signal increase was linear over an anti‐UST nADA concentration range between 0·029 and 0·155 µg/ml (Fig. 6c). Parameters for intra‐ and interassay precision, as well as accuracy, were all in line with the recommendations for immunogenicity testing (Table 3). Determination of nADA at three different concentration levels (low, medium and high) in the presence of MTX showed no significant interference with samples without MTX (Fig. 6d). Comparison of mean signal values in RF‐positive serum samples with RF‐negative sera from diseased patients showed no significant difference (Fig. 6e) and, thus, no indication of the generation of false‐positive signals in the presence of RF. In conclusion, the cell‐based assay is an appropriate approach for the characterization of nADA, not only because it imitates the crucial mechanism of biotherapeutic neutralization, but also because it allows a reliable quantitation of UST‐specific nADA.

Discussion

Covering a wide spectrum of diseases, the introduction of biological therapy has led to significant improvements in treatment regimens of cancer, autoimmune diseases or replacement therapies. In parallel, the challenge of specific immune reactions with generation of ADA against biologicals has become increasingly relevant. As immunogenicity is associated with alterations in drug efficacy and patient safety, validated detection of drug and anti‐drug antibodies, including neutralizing anti‐drug antibodies, should be provided in order to characterize the implications and possible causes of a drug‐induced immune reaction. This would not only facilitate prevention strategies for immunogenicity, but would also allow a personalized treatment algorithm for patients. Here we have validated different approaches to the determination, quantitation and characterization of ADA which represent user‐friendly and cost‐effective methods, in compliance with regulatory requirements for assessing UST‐associated immunogenicity. The approaches presented can also be modified and used for immunogenicity testing of other biotherapeutics.

The FDA recommends a multi‐tiered approach to immunogenicity assessment. Screening‐positivity of ADA should be confirmed and further characterized. Characterization of antibodies includes definition of the isotype of the ADA and quantitation of the neutralizing capacity of nADA. The aim of our study was to establish and validate different testing approaches to the assessment of the immunogenic potential of UST. In contrast to the majority of other immunogenicity testing systems, assays were validated for detection of UST, ADA and/or nADA, considering all potentially relevant contributors to immunogenic reactions. Specific, validated quality parameters included selectivity, detection and quantitation limits, linearity, precision and accuracy. The focus was laid on a coherent quantitation of drug product and determination of ADA, with special emphasis on the detection and characterization of nADA. We used serum samples from treatment‐naive, diseased patients to provide information relevant to clinical studies. Data obtained from all assay formats fulfilled the requirements for immunogenicity assessment set by the FDA and EMA 2, 3.

The LoQ (0·008 µg/ml) for UST quantification determined by the empirical approach appears to contradict the LoD (0·25 µg/ml) (Table 1). However, we repeatedly showed that adequate results were generated at 0·008 µg/ml. We assume that the difference is due to high selectivity of the SPR. The LoD is determined without the presence of the analyte. While a high background signal is produced by the different serum components, for example as a result of electrostatic binding, the presence of the analyte of interest leads to a significantly distinguishable assay signal.

ELISA and SPR are both assays that are easy to use with a high (ELISA) or moderate throughput (SPR). We found the least non‐specific binding using ELISA for UST and ADA determination, allowing less dilution (1:5 in ELISA versus 1:10 in SPR) of serum samples and more precise and accurate quantitation at lower analyte concentrations compared to the SPR approach. Higher non‐specific binding of serum using SPR in comparison to ELISA was also observed by other groups 7. This is due probably to the multiple washing steps included in the ELISA. This may also contribute to the finding that ELISA showed high sensitivity for high‐affinity analytes and lower sensitivity for low‐affinity antibodies, while SPR is suitable for detection of high‐ and low‐affinity antibodies 7, 8. Nevertheless, it should be borne in mind that ADA are polyclonal and show variability, among others, in affinity and valence. Thus, validation parameters for ADA or nADA detection are only valid for the antibody analyzed and cannot be generalized. In both assays, the linear range for UST detection covered the concentration range expected in serum samples from treated patients 6. Concentrations calculated with repeated measurements were comparable for SPR, ELISA and the cell‐based approach. Differences were observed within the range for linear analyte determination. SPR showed a wider linear range for analyte measurements (Tables 1, 2 and 3). In contrast, with nADA determination in the presence of 0·1 µg/ml UST, indirect measurement of nADA was possible over a wider linear range using ELISA (0·05–0·4 µg/ml compared to 0·008–0·1 µg/ml), while with SPR, accurate quantitation of nADA was possible at lower concentrations (Table 3). Empirically determined LoQ should be used cautiously, as false‐positive results might be encouraged. Particularly in a confirmatory assay, use of the calculated should thus be considered instead of the empirical LoQ. Determination and especially characterization of anti‐UST nADA was performed with a cell‐based assay reflecting the mechanism of action of UST and its inhibition by the presence of nADA. Validated parameters for linearity, precision and accuracy yielded comparable results for both SPR and ELISA measurements (Table 3).

Different factors such as co‐medication, pre‐existing antibodies or autoantibodies may interfere with the assay format 9, 10, 11, 12. Thus, regulatory agencies recommend testing for measurement disturbances before clinical use of an immunogenicity assay. We showed that detection of UST, ADA and nADA was not significantly influenced by MTX at low, medium or high analyte concentrations (P < 0·05), while RF may lead to a significant increase in assay signal when determining UST by ELISA (Fig. 2f).

All Ig subtypes should be taken into account when drawing conclusions concerning the origin and role of the ADA. Here, detection of IgG4 is of special interest as it is associated with chronic inflammatory conditions and is reported to have a higher neutralizing capacity than other IgG subclasses 13. The bridging ELISA format is described in the literature as being unsuitable for detecting monovalent IgG4 antibodies 14, 15, 16. Therefore, when using ELISA as a screening assay, true ADA prevalence could be under‐estimated. Nevertheless, a bridging ELISA should not be excluded as a screening tool as ADA are expected to be polyclonal and screening positivity, even in the absence of certain Ig subtypes, remains realistic.

If a complex formation occurs, UST and ADA could mutually prevent each other from binding to the assay target by blocking their respective binding regions or through steric hindrance. In this case, the assay analyte will not be detectable and the complex should first be dissociated to make the analyte accessible for detection. A method already reported by other groups is acidification of samples 17, 18. We integrated an acidification step into the ELISA assay, and in this way increased tolerability towards high concentrations of drug and/or ADA. A clear UST signal was detectable in the acidification assay we established. An increase in the assay signal from pre‐ to post‐acidification indicated release of UST from binding sites and thus the presence of nADA. Accordingly, this method could be used as a screening test for nADA. Furthermore, determination of the total amount of a biological medication is essential in order to investigate the correlation between drug concentration and clinical impact. Considering the very narrow therapeutic range of biotherapeutics, therapeutic drug monitoring is recommended in order to minimize side effects and simultaneously increase therapeutic effectiveness 19. L’Ami et al. showed that quantitation of the therapeutic anti‐TNF‐α antibody, adalimumab, can be used to determine dosing intervals and to individualize disease control in patients with rheumatoid arthritis 20. This approach requires reliable drug quantitation, even in the presence of nADA.

The data presented here indicate that the immunogenic potential of UST can be assessed. Moreover, the assay concept is also transferable to other biological agents. We provide the first report on the precise quantitation of biotherapeutic‐specific nADA, and thus enable the correlation between nADA concentration and its impact on clinical evaluation to be quantified. As the methods described here show distinct advantages over those previously employed, combination of these approaches in a multi‐tiered manner permits assessment of the whole spectrum of ADA generation and may lead to a better understanding of the impact on patient immune responses. Correlation between low drug concentrations and high ADA titer has previously been described without claiming a cause–effect relationship 21, 22, as the study designs failed to address this issue. ADA and nADA can alter pharmacokinetic and pharmacodynamic behavior which may lead to reduced or enhanced drug efficacy 23. Quantitation of ADA in a clinical context could, therefore, be used to describe potential correlation between concentration and patient outcome with regard to therapeutic efficacy, adherence or adverse events. For further clarification, clinical consequences of immunogenicity should be evaluated within prospective cohort studies.

Validated immunogenicity testing could be considered as part of various therapy algorithms (Fig. 7), as in primary or secondary TNF‐α‐inhibitor failure 24. However, although the value of additional information provided by immunogenicity monitoring has been well established mainly for TNF‐blocking therapies, especially in inflammatory bowel disease (IBD), rheumatoid arthritis and psoriasis, a general consensus on its application in routine care has not yet been reached. However, a recent consensus of an international expert panel in IBD based on a critical meta‐analysis of the literature has rated the assessment of anti‐TNF drug and antibody concentrations as appropriate in the following clinical scenarios: at the end of induction therapy in primary non‐responders, in secondary non‐responders, at least once during the first year of maintenance therapy and following a drug holiday 25. Compared to the wealth of data available on the clinical relevance of immunogenicity to TNF blockers, the available information for UST is still somewhat limited, although already approved in 2009 for psoriasis and since 2016 for Crohn’s disease. One reason for the somewhat limited insight into the scope of its immunogenicity is due to the lack of general accessibility of reliable and coherent testing for UST‐specific ADA as a prerequisite for systematic studies of their impact on drug serum trough levels and clinical parameters.

Figure 7.

Immunogenicity may lead to failure of ustekinumab therapy. An algorithm consisting of a combination of different measurement approaches could help to detect neutralizing anti‐drug antibodies (nADA). This enables adjusting therapies.

Conclusion

The present investigation fills a critical gap, as our results provide a sensitive, reliable, valid and robust assay system for the detection of UST‐specific ADA in clinical routine. Beyond the assay specificity for UST as a prototypical application, the newly established general methodology of our investigation is also transferable to the monitoring of other biopharmaceuticals. Accordingly, it is a valuable tool to support intelligent strategies for dealing with adverse immunogenicity or to reduce uncertainties in this respect when switching between a biotherapeutic originator and biosimilar drug 26, and thus helps to improve the quality of patient care.

Disclosures

The authors declare that there are no competing interests associated with the manuscript.

Author contributions

S. M., L. G., M. H. and T. U. performed the practical research. S. M. and S. S. analyzed the data and wrote the manuscript. M. P., H. B., G. G., M. K. and F. B. critically reviewed the contents of the manuscript. All authors have read and approved the manuscript.