Advertisement
Research Article|Articles in Press

Development and validation of an automated assay for anti-drug-antibodies in rat serum

Open AccessPublished:April 27, 2023DOI:https://doi.org/10.1016/j.slast.2023.04.001

      Abstract

      The potential immunogenicity of therapeutic human and humanized monoclonal antibodies (mAb) is a significant concern, and so preclinical testing of therapeutic mAbs routinely includes assessment of anti-drug antibody (ADA) induction. Here, we report the development of automated screening and confirmatory bridging ELISAs for the detection of rat antibodies against DH1042, an engineered human mAb for the SARS-CoV-2 receptor-binding domain. The assays were evaluated for specificity, sensitivity, selectivity, absence of a prozone effect, linearity, intra- and inter- assay precision, and robustness, and found to be suitable for purpose. The assays were then used to evaluate anti-DH1042 antibodies in the sera of rats dosed with lipid-nanoparticle (LNP)-encapsulated mRNA encoding DH1042. Rats received two doses of 0.1, 0.4 or 0.6 mg/kg/dose LNP-mRNA 8 days apart. Twenty-one days after the second dose, 50-100% of rats had developed confirmed anti-DH1042 ADA depending on dose level. No animals in the control group developed anti-DH1042 ADA. These assays reflect new applications for a non-specialized laboratory automation platform, and the methodologies and approaches reported here provide a template that can be adapted for the automated detection and confirmation of ADA in preclinical testing of other biologics.

      Keywords

      Introduction

      Therapeutic monoclonal antibodies (mAb) and other biologics have revolutionized medicine; however, unlike therapeutic small molecules that go relatively unnoticed by recipients’ immune systems, biologics are often large enough to be immunogenic. This can lead to recipients generating anti-drug antibodies (ADA). Anti-mAb ADA can cause mAb neutralization and accelerated clearance of the mAb resulting in a dramatic reduction in efficacy [
      • Schmidt E.
      • Hennig K.
      • Mengede C.
      • et al.
      Immunogenicity of rituximab in patients with severe pemphigus.
      ]. ADA have also been associated with severe acute responses such as anaphylaxis and longer-term immune complex-related diseases [
      • D'Arcy C.A.
      • Mannik M.
      Serum sickness secondary to treatment with the murine-human chimeric antibody IDEC-C2B8 (rituximab).
      ,
      • Korswagen L.A.
      • Bartelds G.M.
      • Krieckaert C.L.
      • et al.
      Venous and arterial thromboembolic events in adalimumab-treated patients with antiadalimumab antibodies: a case series and cohort study.
      ]. The use of both fully human mAb and humanized mAb from which T cell epitopes have been eliminated have helped reduce the immunogenicity of therapeutic mAbs, but it is not yet possible to engineer out the risk of ADA induction a priori [
      • Vaisman-Mentesh A.
      • Gutierrez-Gonzalez M.
      • DeKosky B.J.
      • et al.
      The molecular mechanisms that underlie the immune biology of anti-drug antibody formation following treatment with monoclonal antibodies.
      ]. Preclinical and clinical studies of mAb must, therefore, include screening for induction of ADA, and evaluation of ADA production and its clinical consequences are necessary for approval of biologicals by regulatory bodies.
      The US Food and Drug Administration (FDA) and the European Medicine Agency have released guidance regarding immunogenicity assessments [
      • Bartelds G.M.
      • Krieckaert C.L.
      • Nurmohamed M.T.
      • et al.
      Development of antidrug antibodies against adalimumab and association with disease activity and treatment failure during long-term follow-up.
      ,
      European Medicines Agency
      Committee for Medicinal Products for Human Use (CHMP).
      ,
      European Medicines Agency
      Committee for Medicinal Products for Human Use (CHMP).
      ], and there is an extensive literature describing ADA assay development and risk-based testing approaches [
      • Shankar G.
      • Arkin S.
      • Cocea L.
      • et al.
      Assessment and reporting of the clinical immunogenicity of therapeutic proteins and peptides-harmonized terminology and tactical recommendations.
      ,
      • Shankar G.
      • Devanarayan V.
      • Amaravadi L.
      • et al.
      Recommendations for the validation of immunoassays used for detection of host antibodies against biotechnology products.
      ,
      • Shankar G.
      • Pendley C.
      • Stein K.E.
      A risk-based bioanalytical strategy for the assessment of antibody immune responses against biological drugs.
      ,
      • Passey C.
      • Suryawanshi S.
      • Sanghavi K.
      • et al.
      Reporting, Visualization, and Modeling of Immunogenicity Data to Assess Its Impact on Pharmacokinetics, Efficacy, and Safety of Monoclonal Antibodies.
      ,
      • Ponce R.
      • Abad L.
      • Amaravadi L.
      • et al.
      Immunogenicity of biologically-derived therapeutics: assessment and interpretation of nonclinical safety studies.
      ,
      • Mire-Sluis A.R.
      • Barrett Y.C.
      • Devanarayan V.
      • et al.
      Recommendations for the design and optimization of immunoassays used in the detection of host antibodies against biotechnology products.
      ,
      • Koren E.
      • Smith H.W.
      • Shores E.
      • et al.
      Recommendations on risk-based strategies for detection and characterization of antibodies against biotechnology products.
      ,
      • Bloem K.
      • Hernandez-Breijo B.
      • Martinez-Feito A.
      • et al.
      Immunogenicity of Therapeutic Antibodies: Monitoring Antidrug Antibodies in a Clinical Context.
      ]. Historically, the most widely-reported assay platforms used for anti-mAb ADA testing are Enzyme Linked Immunosorbent Assay (ELISA) and radioimmunoassay (RIA) [
      • Gorovits B.
      • Baltrukonis D.J.
      • Bhattacharya I.
      • et al.
      Immunoassay methods used in clinical studies for the detection of anti-drug antibodies to adalimumab and infliximab.
      ]. RIAs are extremely sensitive, but they generate radioactive waste and are generally harder to scale and automate than ELISAs, and so ELISAs are often the platform of choice. A common strategy is to use a bridging ELISA – a form of sandwich ELISA – in which ADA in a test sample creates a “bridge” between plate-bound and free, labelled therapeutic mAb [
      • Mire-Sluis A.R.
      • Barrett Y.C.
      • Devanarayan V.
      • et al.
      Recommendations for the design and optimization of immunoassays used in the detection of host antibodies against biotechnology products.
      ]. Bridging ELISAs also have the advantages that they are species- and immunoglobulin isotype- independent.
      Non-specific binding can result in false positive signals in a bridging ELISA, and so ADA testing is a tiered process. i) All samples are assayed in a screening ELISA with a false discovery rate (FDR) of approximately 5%. ii) Samples that test positive in screening (potential positives) are re-assayed in a confirmatory assay of high specificity to eliminate false positives. The confirmatory assay is often a competitive version of the screening ELISA, with bridge formation assayed in the presence and absence of free, unlabeled therapeutic mAb. If ADA responses are detected during preclinical development then a third tier of assays such as quantitative titers, neutralization, isotyping and epitope mapping may be used to characterize the ADA. Many of these assays are amenable to automation using non-specialized liquid handling units and other common laboratory equipment, which can increase throughput, improve reproducibility, and decrease hands-on time. Herein we report the development and validation of an automated workflow for the detection of ADA against the mAb DH1042 in rat serum, a common species for preclinical ADA assessment.
      DH1042 is an engineered human IgG1 mAb that binds the Receptor Binding Domain (RBD) of the SARS-CoV-2 spike protein and neutralizes the virus’ ability to infect target cells [
      • Shen X.
      • Tang H.
      • McDanal C.
      • et al.
      SARS-CoV-2 variant B.1.1.7 is susceptible to neutralizing antibodies elicited by ancestral spike vaccines.
      ,
      • Coleman C.N.
      • Cliffer K.D.
      • DiCarlo A.L.
      • et al.
      Preparedness for a 'no-notice' mass-casualty incident: a nuclear detonation scenario.
      ]. DH1042 is currently in preclinical development as a prophylactic mAb for COVID-19. Unlike conventional therapeutic mAb delivered as proteins, DH1042 is delivered intravenously as an mRNA encapsulated in a lipid nanoparticle (LNP) and transcribed into protein within the recipient. The advantage of gene-delivered mAbs is that they are faster to develop and more cost-effective to scale up for manufacturing [
      • Sempowski G.D.
      • Saunders K.O.
      • Acharya P.
      • et al.
      Pandemic Preparedness: Developing Vaccines and Therapeutic Antibodies For COVID-19.
      ,
      • Whitley J.
      • Zwolinski C.
      • Denis C.
      • et al.
      Development of mRNA manufacturing for vaccines and therapeutics: mRNA platform requirements and development of a scalable production process to support early phase clinical trials.
      ].
      The assay described in this application note was established to facilitate preclinical toxicology evaluation of DH1042 in line with FDA guidance [
      U.S. Department of Health and Human Services Food and Drug Administration
      Immunogenicity Testing of Therapeutic Protein Products — Developing and Validating Assays for Anti-Drug Antibody Detection Guidance for Industry.
      ] using only a non-dedicated automated liquid handling unit in an academic environment. The steps described and the supplementary automation scripts provide a template for the automation and validation of assays for ADA against any mAb.

      Methods

      Vendors, catalog numbers, lot numbers, version numbers and Research Resource Identifiers [

      Research Resouce Identification (RRID). Accessed 2/10/23, 2023. https://www.rrids.org/

      ] (RRIDs) for critical and other required reagents, consumables, equipment and software are listed in supplemental tables S1-S5.

      Reagents

      DH1042 recombinant protein was produced by GenScript USA Inc. using heavy and light chain plasmids made by the Duke Pandemic Prevention Program as previously described [
      • Coleman C.N.
      • Cliffer K.D.
      • DiCarlo A.L.
      • et al.
      Preparedness for a 'no-notice' mass-casualty incident: a nuclear detonation scenario.
      ]. Individual Wistar Hannover rat sera were purchased from BioChemEd and pooled rat sera from Sigma Aldrich. Rabbit polyclonal anti-DH1042 idiotype (anti-ID) was generated by GenScript USA Inc. by immunizing rabbits with the F(ab’) [
      • D'Arcy C.A.
      • Mannik M.
      Serum sickness secondary to treatment with the murine-human chimeric antibody IDEC-C2B8 (rituximab).
      ] region of DH1042, purifying the antibody using whole DH1042, and then adsorbing against total human IgG to exclude antibodies against the constant region. Rat total IgG and human IgG1 were purchased from MilliporeSigma. LNP-encapsulated DH1042-encoding mRNA for animal studies was manufactured by the Duke Human Vaccine Institute Good Manufacturing Practice facility.

      DH1042 biotinylation

      DH1042 was biotinylated using Abcam Lightning-link Biotinylation kit (Fast, Type a) according to the manufacturer's instructions. Biotinylated DH1042 was stored in single-use aliquots at 4°C, and all experiments following minimum required dilution (MRD) determination used the same batch of biotinylated antibody.

      Automation platform

      A BioMek i7 hybrid liquid handler unit (Beckman Coulter) with span-8 and 1200 μL 96-well heads, integrated 405LS plate washer (BioTek), SpectraMax 384 plus plate reader (Molecular Devices) and Cytomat 10 plate storage tower (Thermo), was used to perform all bridging ELISA steps except the plate transfer to/from cold storage, which was performed by hand.

      Software and statistical analysis

      Pipetting parameters were designed and optimized for reproducibility and speed using BioMek software (Beckman Coulter). Assay workflows were designed and executed using SAMI Ex scheduling software (Beckman Coulter). Plate reader data were collected using SoftMax Pro Molecular Devices), stored in Data Acquisition and Reporting Tool (DART) (Beckman Coulter), and analyzed in Excel (Microsoft) and Prism (Graphpad software).

      Screening ELISA

      384-well high-binding immunoassay plates (Corning) were coated with 0.5 μg/mL DH1042 in CBC buffer (15 mM Na2CO3, 35 mM NaHCO3, pH 9.5 – JT Baker / Fisher), wrapped in Parafilm, and stored off-deck overnight at 4°C. Plates were then returned to the deck and washed four times in wash buffer (PBS (ScyTek) 0.1% Tween-20 (Millipore-Sigma)). Plates were blocked for 120 minutes (SAMI Ex timing constraints used for scheduling purposes: min: 110; no max) at room temperature in blocking buffer (CBC with 5% goat serum, 0.01% Kathon, 0.05% Tween-20 (all Millipore-Sigma)). Samples and controls were then diluted to the MRD of 1:10 in sample diluent (PBS (Gibco), 5% goat serum, 0.05% Tween-20, 0.01% Kathon) (or other dilutions as indicated) and applied in quadruplicate wells to the blocked DH1042-coated plates. The negative control was pooled rat sera and the high, medium and low positive controls were 2,500 ng/mL, 500 ng/mL and 100 ng/mL rabbit anti-DH1042 idiotype antisera (anti-ID) diluted in pooled rat sera.
      The assay plates with samples and controls were incubated at room temperature for 120 minutes (min: 120; max: 130), washed four times in wash buffer and then 0.5 μg/mL biotinylated DH1042 in secondary antibody diluent (PBS (Gibco), 1% BSA, 0.05% Tween-20 (both Millipore-Sigma) pH 7.4) was added. Plates were then incubated at room temperature for a further 105 minutes (min: 105; max: 145), washed four times, and treated with 0.5 μg/mL streptavidin-horseradish peroxidase conjugate (SA-HRP (Millipore-Sigma)) in secondary antibody diluent. After one hour, the plates were washed four times in wash buffer and a one-component HRP colorimetric substrate, TMB Ultra (3,3′,5,5′-tetramethylbenzidine; Thermo Scientific) was added. The HRP/TMB reaction was stopped after 14 minutes (min: 14; max: 16) by addition of 2N sulfuric acid (VWR) and the TMB color change measured by reading absorbance at 450 nm (A450) immediately (no minimum time; max: 3 minutes). A450 measurements for quadruplicate wells were averaged to generate a single value per sample or control.
      The HRP reaction time was selected as while A450 for all controls increased as reaction time increased (Fig S1A), the signal to noise ratio for the low control did not increase after a reaction time of around 14 minutes (Fig S1B). In addition, the signals from the high and medium controls became saturating as the reaction proceeded (Fig S1A and B). While the A450 and normalized results for the low and medium controls were stable for at least 10 minutes after acid addition, the A450 and normalized result for the high control decreased within a few minutes of acid addition, hence the need to read the plates within three minutes (Fig S1C,D).
      Pipette tips were used for up to ten transfers of the same solution before being discarded. This use limit was determined by pipetting 20μL bromophenol blue solution to every well in twenty-five 384-well plates using the same tips and then assessing the intra-plate coefficients of variance (%CV) for bromophenol blue absorbance at 590nm. Some manufacturing lots tested showed unstable intra-plate %CVs and extreme outliers after thirteen uses, hence the conservative maximum of ten uses (Fig S2).
      SAMI Ex code and labware definitions for the screening assay are provided as supplemental material.

      Confirmatory ELISA

      The confirmatory assay was performed as described for the screening assay with the exception that the samples and controls were diluted to the MRD in an untreated 96-well conical bottomed plate (Greiner-Bio-One) in both sample diluent and sample diluent containing 25 μg/mL unlabeled free DH1042 (unless otherwise indicated). The diluted samples/controls were then incubated for 30 minutes at room temperature with periodic mixing by pipette. The samples/controls were then transferred to DH1042-coated and blocked plates and the assay proceeded as above. SAMI Ex code and labware definitions for the confirmatory assay are provided as supplemental material.
      Separate serum aliquots were used for the screening and confirmatory assays. Sera with results in the screening assay higher than the high positive control were manually serially diluted with commercial pooled rat sera to ensure the signal in the absence of DH1042 fell between the high and medium control.

      Rat dosing

      Wistar rat dosing studies were conducted by Charles River Laboratories, Inc. Rats were housed as specified by the USDA Animal Welfare Act and as described in the National Research Council's Guide for the Care and Use of Laboratory Animals. All procedures were reviewed and approved by the Charles River-MWN Institutional Animal Care and Use Committee and conducted within the guidelines of the National Research Council.

      Results and discussion

      Minimum required sample dilution

      Non-specific binding of immunoglobulins can cause background noise in immunoassays, and so pooled rat sera spiked with anti-DH1042 idiotype polyclonal antibody (anti-ID) at various concentrations was 3-fold serially diluted with sample diluent and assayed to determine the minimum required sample dilution (MRD) (Fig 1A). A 1:10 dilution of pooled rat sera spiked with 39.06 ng/mL anti-ID had an average absorbance signal more than three standard deviations above that of non-spiked pooled sera (p=0.01 by t test), well below the FDA-recommended [
      U.S. Department of Health and Human Services Food and Drug Administration
      Immunogenicity Testing of Therapeutic Protein Products — Developing and Validating Assays for Anti-Drug Antibody Detection Guidance for Industry.
      ] sensitivity of 100 ng/mL. Increasing the sample dilution to 1:270 reduced the non-specific background from the negative sample (0 ng/mL); however, this also resulted in reduced sensitivity as the 39.06 ng/mL anti-ID could not be resolved from the negative control. Use of a sample dilution greater than 1:270 did not further reduce background signal but did further reduce sensitivity, and so a MRD of 1:10 was used for all subsequent assays.
      Fig 1
      Fig. 1Screening Assay Minimum Sample Dilution and Cut Point (A) Pooled rat sera was spiked with anti-DH1042 idiotype antibody (anti-ID) at the indicated concentrations and then assayed in triplicate wells by automated screening ELISA for anti-DH1042 antibodies. Plot shows individual well absorbance at 450nm for each dilution series plus connecting lines between the means of the technical replicates. Figure is representative data from one of two experiments. (B)(C) Sera from 15 rats were assayed twice per run in three independent runs of the screening assay (90 data points, each an average of four wells) and then normalized to the negative control to determine the assay cut point. Panels show (B) Quantile-Quantile (QQ) plot of actual distribution versus predicted normal distribution and (C) scatter of normalized data (n=90), mean ±SD in red and the 5% false discovery rate (screening assay cut point).

      Screening assay cut point

      Sera from 15 rats were used to approximate the variability of true negative samples and to establish the cut point for the screening assay. Six determinations were made for each serum sample over three independent runs (two determinations per run per sample). Each determination was the average A450 of quadruplicate wells, and any determination with a percentage coefficient of variation (%CV) for the quadruplicates of over 25% was excluded from analysis. The average A450 values for the cut point analysis were normalized to the average A450 of the negative control on each plate (normalized result = average A450 of sample/average A450 of negative control). The normalized values contained no outliers by ROUT analysis [
      • Motulsky H.J.
      • Brown R.E.
      Detecting outliers when fitting data with nonlinear regression - a new method based on robust nonlinear regression and the false discovery rate.
      ], and exhibited a normal distribution when evaluated by Shapiro-Wilk test [
      • Royston P.
      Remark AS R94: A Remark on Algorithm AS 181: The W-test for Normality.
      ] (p=0.468) and QQ-plot (Fig. 1B). The screening assay cut point was therefore set to the 5% FDR (normalized result > 3.744) (Fig. 1C) [
      • Shankar G.
      • Devanarayan V.
      • Amaravadi L.
      • et al.
      Recommendations for the validation of immunoassays used for detection of host antibodies against biotechnology products.
      ,
      • Koren E.
      • Smith H.W.
      • Shores E.
      • et al.
      Recommendations on risk-based strategies for detection and characterization of antibodies against biotechnology products.
      ].
      It was anticipated that the commercial pooled rat sera would have A450 results similar to the average of those for the individual rat sera. However, in each of the three runs over 90% of the individual rat sera had results higher than the commercial pool used for early assay optimization and as a negative control (individual rats A450±SD: 0.577±0.199 versus pool: 0.251±0.04 p=0.006). In-house sera pools prepared from the study population may, therefore, be a more appropriate matrix control than commercial sera pools for preclinical immunoassay development.

      Linearity and sensitivity

      Sensitivity is the lowest analyte concentration at which an assay reliably generates a positive result, while linearity is whether the result is proportional to the amount of analyte present. To determine sensitivity of the automated screening assay, pooled rat sera was spiked with 1,000 ng/mL anti-ID and 2-fold serially diluted with pooled rat sera down to 1.9 ng/mL. The spiked sera were then assayed in three independent runs of the screening assay. Four-parameter logistic (4-PL) regression was used to identify the concentrations at which the sensitivity curves met the plate-specific cut points for each run (Fig. 2A; each plot shows the data from one independent run). The average sensitivity across the three runs was 75.3 ng/mL. The average non-linear curve fit R2 was 0.95 and the regression curves indicated the absorbance was proportional to the concentration of anti-ID present, indicating acceptable linearity.

      Specificity, selectivity, matrix interference and prozone effect

      Specificity is the ability of the assay to detect singularly the ADA of interest, while selectivity is the ability of the assay to detect ADA in the presence of other anticipated sample components.
      To evaluate specificity, pooled rat sera was spiked with 800 ng/mL anti-DH1042 alone or with 1,000 ng/mL rat IgG; 5,000 ng/mL rat IgG; 1,000 ng/mL human IgG1 or 5,000 ng/mL IgG1 and assayed in duplicate (four wells per replicate) in one run of the screening assay. In all cases the results were higher than the cut point for the plate, indicating that the presence of non-anti-DH1042 rat IgG or human IgG1 did not interfere with the assay (Fig. 2B).
      Fig 2
      Fig. 2Sensitivity, Specificity and Prozone Effect Testing of the Automated Screening Assay (A) Pooled rat sera was spiked with anti-DH1042 idiotype (anti-ID) at the indicated concentrations and then assayed in quadruplicate wells in three independent runs of the screening assay. The three plots show absorbance at 450nm for each well in each run of the assay plus the 4-PL curve fits of the plotted data and the run-specific cut-points. (B) To evaluate specificity, pooled rat sera was spiked with the indicated antibodies in duplicate and then each duplicate assayed in four wells in one run of the automated screening assay. Data show normalized results for each well and the mean ±SD in red. (C) To test for evidence of the prozone effect, pooled rat sera was spiked with 0.5 μg/mL anti-ID (one aliquot) or 100 μg/mL anti-ID (two aliquots) and every aliquot assayed in four wells of the screening assay. Data show normalized result for each well and mean ±SD in red.
      Selectivity was evaluated using sera from 15 rats. Individual sera were left unadulterated or spiked with 500 ng/mL anti-ID and assayed in two independent runs of the screening assay. 30/30 (100%) of the spiked sera were above the cut point, and 29/30 (97%) of the unadulterated samples were below the cut point (Fig. S3A and S3B). The single unadulterated sample (1/30; 3%) with a result above the cut point (Fig. S3B rat 15) was in line with expectations given a planned 5% FDR.
      The prozone or hook phenomenon is a false negative result that can arise in some immunoassays due to interference by extremely high ADA titers. Pooled rat sera spiked with 100 μg/mL anti-DH1042 and tested in duplicate using the screening assay gave a positive (above cut point) result, and so there was no evidence of a prozone effect (Fig. 2C).

      Intra-plate and Inter-assay Precision

      Precision reflects the ability of an assay to generate reproducibly the same result for the same sample when assayed on the same plate (intra-plate) or different batch (inter-assay). To evaluate intra-plate precision of the screening assay, six aliquots of the high, medium, low and negative controls were assayed in quadruplicate wells per aliquot in a single plate and single run (Fig. 3A). The resulting %CVs were: high control: 2.1%, medium: 9.2%, low: 17.3%, negative: 20.2%. Similarly, to evaluate inter-assay precision, single aliquots of the three positive controls were tested in quadruplicate wells in four independent runs of the screening assay and then normalized to the negative control in each run. The resulting %CVs for the normalized positive controls were: high: 18.7%, medium: 31.8%, low: 21% (Fig. 3B).
      Fig 3
      Fig. 3Intra-plate, Inter-plate and Inter-assay Variance (A) To evaluate the intra-plate variability, six aliquots of the high, medium, low and negative controls (2,500 ng/mL, 500 ng/mL, 100 ng/mL and 0 ng/mL anti-DH1042 idiotype diluted in pooled rat sera) were assayed in quadruplicate wells in a single plate of the automated screening assay. Panel shows the absorbance at 450nm per well plus mean ±SD in red for all controls. (B) To evaluate inter-assay variability, aliquots of each control were assayed on four runs of the screening assay performed on separate days. Each data point is the normalized result obtained from quadruplicate wells in a single run plus the mean ±SD across all runs. (C) Four aliquots of each control were assayed in quadruplicate wells per aliquot on four assay plates in a single run. Panel shows the A450 per well for each of the four aliquots on each plate plus mean ±SD for each technical replicate. (D) Aliquots of the controls were prepared and then either assayed fresh or after the indicated number of freeze/thaw cycles using a single run of the screening assay. Data show A450 from quadruplicate wells plus mean±SD of the technical replicates. (E) Aliquots of the controls were prepared and stored under the indicated conditions and then assayed using a single run of the screening assay. Data show A450 from quadruplicate wells plus mean±SD of the technical replicates.

      Within-run inter-plate precision and robustness of SAMI ex scheduling

      All assays described above were performed using single-plate batches; however, the screening assay was designed to automate testing of up to 368 samples in quadruplicate using four plates per batch. SAMI Ex-controlled automation dynamically schedules operations to minimize the total run time given the batch size and user-set acceptable timing ranges. This means that operation timing and order can vary depending on the batch size. There is, therefore, a potential for results to vary between plates in a single batch (inter-plate precision), and to vary depending on the use of a 1-plate versus a 4-plate schedule.
      To evaluate the inter-plate precision of the screening assay, four aliquots of the high, medium, low and negative controls were assayed in quadruplicate wells per aliquot on four plates in a single run of the screening assay. On all plates, A450 were proportional to the level of anti-ID, and the inter-plate %CVs for A450 were below 10% for all controls (high: 1.8%, medium: 3.8%, low: 9.5%, negative: 7.9%) (Fig. 3C). These %CVs were close to the intra-plate precision results and appreciably lower than the inter-assay %CVs. There were no significant differences between the normalized results for the positive controls on the plates in the single four-plate run compared to four independent single-plate runs when evaluated by two-way ANOVA (p=0.17) with Bonferroni's multiple comparison testing (Fig. S4). The screening assay therefore give reproducible results regardless of batch size.

      Robustness - sample storage

      The high, medium, low and negative controls were prepared, stored at -80 ± 15°C for 1-6 days, and then repeatedly thawed at room temperature and refrozen at -80 ± 15°C for 1-3 days to generate samples that had undergone one, two or three freeze/thaw cycles. The thawed controls were then evaluated in a single run of the screening assay. All three controls were above the cut point after one to three freeze/thaw cycles (Fig. 3D) and the A450 were always proportional to the level of anti-ID in the control. There were, however, differences in the absolute A450 values between each freeze/thaw cycle. These data indicated that samples and controls for screening can be stored at -80°C without destroying reactivity in the assay and that both controls and samples should be stored in multiple single-use aliquots to avoid differences in the number of freeze/thaw cycles between vials under test.
      To evaluate the impact of short term sample storage on the bench or fridge, controls were prepared and stored at 4±2°C or room temperature (RT) for four hours, 48 hours and 7 days and then assayed in a single run of the screening assay. Samples stored at RT for up to 48 hours or at 4°C showed a signal that was proportional to the anti-ID concentration. However, the absolute result for each control showed increased variability following prolonged storage when compared to samples assayed immediately after thawing (e.g. compared to Fig. 3A-D), and controls stored at RT for 7 days showed a loss of signal from the positive controls and a false-positive signal from the negative control (Fig. 3E). These data indicated that sample storage conditions contributed to the reproducibility of the assay and so samples and controls should be stored at -80°C immediately after collection until immediately prior to assay.

      Confirmatory assay development

      After developing the screening assay, the concentration of non-labelled DH1042 required to compete with the binding of biotinylated DH1042 in the assay was determined. A screening assay was run using 500ng/mL anti-ID in pooled rat sera alone and spiked with unlabeled DH1042 at concentrations of 0.01 to 10 μg/mL. The signal from the anti-ID was substantially below the screening assay cut point when preincubated with DH1042 at 5 µg/mL or above (Fig. 4A). The free DH1042 concentration for the confirmatory assay was set to five times this value: 25 μg/mL.
      Fig 4
      Fig. 4Confirmatory Assay Development (A) 500ng/mL anti-DH1042 idiotype (anti-ID) in pooled rat sera was assayed in quadruplicate wells of the screening assay alone and in the presence of unlabeled DH1042 at the indicated concentrations. Data points show the result for each well from a single plate, the solid line indicates 5-PL curve fit and dashed line indicates the screening assay cut point. (B-D) To determine the confirmatory assay cut point, sera from 15 rats were assayed twice per run in three independent runs of the confirmatory assay (90 data points, each an average of four wells). Panels show: (B) scatter plot of percentage inhibition (n=90) with median and interquartile range in red, (C) quantile-quantile (QQ) plot of actual distribution of the of log-transformed inhibitory ratios versus predicted normal distribution, and (D) violin plot of log-transformed inhibitory ratios; each datapoint corresponds to transformed datapoint from panel B. Dashed line indicates the confirmatory assay 97th percentile cut point. Blue line indicates median. (E) Pooled rat sera was spiked with anti-DH1042 idiotype (anti-ID) at the indicated concentrations and then assayed in quadruplicate wells in three independent runs of the automated confirmatory assay. The three plots show the log inhibitory ratio at each anti-ID concentration for each of the three independent runs and dashed lines indicate confirmatory assay cut point. (F) To evaluate the intra-assay variability of the confirmatory assay, six aliquots of the high, medium, low, and negative controls were assayed in quadruplicate wells in the same assay run. Data show the log-transformed inhibitory ratios for each of the six aliquots plus the mean±SD in red. (G) To evaluate inter-assay variability of the confirmatory assay, the four controls were assayed in quadruplicate wells in four independent runs. Data show the log-transformed inhibitory ratios for each control in each the four independent runs plus the mean±SD in red.
      To establish the confirmatory assay cut point, sera from 15 rats were assayed in duplicate in three independent runs of the confirmatory assay to approximate the variability of true negative samples (90 determinations total). The percentage signal inhibition by DH1042 was calculated for each determination (1 – [average of quadruplicate A450 for individual serum spiked with unlabeled DH1042] / [average of quadruplicate A450 for individual serum] * 100). No outliers in percentage signal inhibition were identified by ROUT, but the data exhibited a non-normal distribution (Fig. 4B) (p<0.0001 by Shapiro-Wilk). The log-transformed inhibitory ratio (log([Average of quadruplicate A450 for individual serum spiked with unlabeled DH1042] / [average of quadruplicate A450 for individual serum])) also exhibited a non-normal distribution (Fig. 4C) (p<0.0001 by Shapiro-Wilk). Given the non-normal distributions of both these metrics, the cut point for the confirmatory assay was set to the 97th percentile of the log-transformed data (Fig. 4D), as recommended by Devanarayan et al. [
      • Devanarayan V.
      • Smith W.C.
      • Brunelle R.L.
      • et al.
      Recommendations for systematic statistical computation of immunogenicity cut points.
      ] This equaled a log-transformed inhibitory ratio of -0.8721 and a percentage inhibition of 86.58%.

      Confirmatory assay sensitivity, linearity and precision

      To test the sensitivity of the confirmatory assay, pooled rat sera was spiked with 500 ng/mL anti-ID, two-fold serially diluted with pooled rat sera to 1.95 ng/mL anti-ID and tested in three independent runs of the assay. The average sensitivity across the three runs was 150.76 ng/mL (Fig. 4E; each plot shows the data from one independent run). The average non-linear curve fit R2 was 0.95 indicating acceptable linearity.
      Intra-assay precision of the confirmatory assay was estimated by assaying six replicates of the high, medium, low, and negative controls in the same batch (Fig. 4F). The %CVs of the resulting inhibitory ratios were: high control: 1.6%, medium: 4%, low: 14.6%, negative: 79.7%. Similarly, to evaluate inter-assay precision the controls were tested in four independent runs of the confirmatory assay. The resulting %CVs were: high control: 0.8%, medium: 1.8%, low: 3.7%, negative: 73.7% (Fig. 4G). The robustness of the confirmatory assay was not assessed as sample stability was assessed during robustness testing of the screening assay, and too few samples were anticipated to require confirmatory screening to necessitate robustness screening for multi-plate batches.

      Rats dosed with mRNA-DH1042 generate anti-DH1042 ADA

      Wistar rats were dosed intravenously with either saline (control – 31 animals) or LNP-encapsulated mRNA encoding DH1042 (0.1, 0.4 or 0.6 mg/kg/dose – 30 animals per group). The volume of saline used for the control was the same volume as the volume for the highest dose administered. Animals were dosed on study day 0 and study day 8, and sera collected for ADA testing at study days 7 (all animals) and 29 (subset of 8 controls plus 10 of each treatment group). Sera were stored at -80°C immediately after collection.
      The automated screening and confirmatory assays for DH1042 ADA reported above were used to evaluate the collected sera. Any screening assay plate on which all the positive controls were not above the cut point was excluded and repeated using different sample aliquots; any sample result with a %CV across quadruplicate wells greater than 25% was re-assayed using a different sample aliquot.
      In the lowest dose group only about 13% percent of the rats had developed anti-DH1042 antibodies seven days after dosing (Fig. 5). However, 21 days after the second dose (study day 29) over 50% of animals tested had developed confirmed anti-DH1042 ADA regardless of mRNA dose. The assay never indicated the presence of ADA in sera from the control (saline only) group of rats, supporting the finding that the assay reported here is specific for anti-DH1042. Of note, two of the 39 control samples tested as potential positives in the screening assay but were subsequently eliminated as false positives on testing of a second aliquot in the confirmatory assay. This reflects a 5.1% FDR, matching the 5% FDR used for development.
      Fig 5
      Fig. 5Detection of anti-DH1042 ADA in Rats Following Administration of LNP Encapsulated mRNA-DH1042

      Broader applicability

      Expansion of this assay to other laboratories, adaption for other species or biologics, or the bridging to a new lot of a critical reagent (table S1), especially the animal-derived polyclonal anti-ID and biotinylated detection antibody, requires re-validation of the key performance metrics. Of particular importance is assessment of the cut point, linearity and sensitivity determinations as these are critical for data interpretation and performance consistency. Assessment of these metrics can be accomplished by performing four independent runs of each assay.
      This bridging ELISA only detects ADA that bind both non-labelled and chemically biotinylated DH1042, and biotinylation has the potential to disrupt binding epitopes [
      • Von Grunigen R.
      • Schneider C.H.
      Epitope analysis: biotinylated short peptides as inhibitors of anti-peptide antibody.
      ]. Indeed, surface plasmon resonance comparisons of mAb binding to unlabeled and biotinylated peptides indicate that in some contexts chemical biotinylation can alter binding kinetics [
      • Peter J.C.
      • Briand J.P.
      • Hoebeke J.
      How biotinylation can interfere with recognition: a surface plasmon resonance study of peptide-antibody interactions.
      ]. A subset of ADA within a polyclonal response may, therefore, be undetectable in the current assay. This issue may be avoided through the use of a detection antibody enzymatically biotinylated outside the idiotype region. Similarly, bispecific ADA, e.g. human IgG4 that has undergone Fab-arm exchange with a non-ADA antibody, cannot be detected by bridging ELISA.
      DH1042 is under consideration for short-term single-dose use, and this assay was designed to detect ADA in rat serum after completion of pharmacokinetic studies. The performance of the assay was, therefore, not tested in the presence of a circulating DH1042 within the test sera. This potential cause of false negatives must be evaluated if samples are likely to contain significant concentrations of drug.
      In conclusion, here we report the development, validation, and application of an automated assay for the detection of anti-therapeutic mAb anti-drug antibodies using a general purpose BioMek liquid handling unit driven by SAMI Ex dynamic scheduling software. The SAMI Ex protocols provided as supplemental material are readily adaptable to the development of ADA for other species and therapeutics.

      Declaration of Competing Interests

      The authors declare the following financial interests/personal relationships which may be considered as potential competing interests:
      Gregory D. Sempowski reports financial support was provided by Defense Advanced Research Projects Agency. Gregory D. Sempowski reports financial support was provided by National Institutes of Health. Gregory D. Sempowski has patent #PCT/US2021/050552 issued to Duke University.

      Acknowledgements

      The authors thank Amelia Karlsson for program management.

      Funding

      Work for this study was performed in the Duke Regional Biocontainment Laboratory, which received partial support for construction from the National Institutes of Health, National Institute of Allergy and Infectious Diseases (UC6-AI058607; G.D.S). The work was supported by a cooperative agreement with DOD/DARPA (HR0011-17-2-0069; G.D.S).
      Wistar rats received either saline alone (n=31) or the indicated doses of LNP-encapsulated mRNA-DH1042 (n=30 per group) at day 0 and day 8. Sera were collected at day 7 (all animals) and day 29 (n=8 controls; n=10 per mRNA-DH1042 group) and then tested for anti-DH1042 ADA in the automated assay. Plots show the percentage of animals tested in each group that gave positive results in both the screening and confirmatory assays.

      Appendix. Supplementary materials

      References

        • Schmidt E.
        • Hennig K.
        • Mengede C.
        • et al.
        Immunogenicity of rituximab in patients with severe pemphigus.
        Clin Immunol. 2009; 132: 334-341
        • D'Arcy C.A.
        • Mannik M.
        Serum sickness secondary to treatment with the murine-human chimeric antibody IDEC-C2B8 (rituximab).
        Arthritis Rheum. 2001; 44: 1717-1718
        • Korswagen L.A.
        • Bartelds G.M.
        • Krieckaert C.L.
        • et al.
        Venous and arterial thromboembolic events in adalimumab-treated patients with antiadalimumab antibodies: a case series and cohort study.
        Arthritis Rheum. 2011; 63: 877-883
        • Vaisman-Mentesh A.
        • Gutierrez-Gonzalez M.
        • DeKosky B.J.
        • et al.
        The molecular mechanisms that underlie the immune biology of anti-drug antibody formation following treatment with monoclonal antibodies.
        Front Immunol. 2020; 11: 1951
        • Bartelds G.M.
        • Krieckaert C.L.
        • Nurmohamed M.T.
        • et al.
        Development of antidrug antibodies against adalimumab and association with disease activity and treatment failure during long-term follow-up.
        JAMA. 2011; 305: 1460-1468
        • European Medicines Agency
        Committee for Medicinal Products for Human Use (CHMP).
        Guideline on Immunogenicity assessment of therapeutic proteins. 2017 (EMEA/CHMP/BMWP/14327/2006 Rev 1)
        • European Medicines Agency
        Committee for Medicinal Products for Human Use (CHMP).
        Guideline on immunogenicity assessment of monoclonal antibodies intended for in vivo clinical use. 2012 (EMA/CHMP/BMWP/86289/2010)
        • Shankar G.
        • Arkin S.
        • Cocea L.
        • et al.
        Assessment and reporting of the clinical immunogenicity of therapeutic proteins and peptides-harmonized terminology and tactical recommendations.
        AAPS J. 2014; 16: 658-673
        • Shankar G.
        • Devanarayan V.
        • Amaravadi L.
        • et al.
        Recommendations for the validation of immunoassays used for detection of host antibodies against biotechnology products.
        J Pharm Biomed Anal. 2008; 48: 1267-1281
        • Shankar G.
        • Pendley C.
        • Stein K.E.
        A risk-based bioanalytical strategy for the assessment of antibody immune responses against biological drugs.
        Nat Biotechnol. 2007; 25: 555-561
        • Passey C.
        • Suryawanshi S.
        • Sanghavi K.
        • et al.
        Reporting, Visualization, and Modeling of Immunogenicity Data to Assess Its Impact on Pharmacokinetics, Efficacy, and Safety of Monoclonal Antibodies.
        AAPS J. 2018; 20: 35
        • Ponce R.
        • Abad L.
        • Amaravadi L.
        • et al.
        Immunogenicity of biologically-derived therapeutics: assessment and interpretation of nonclinical safety studies.
        Regul Toxicol Pharmacol. 2009; 54: 164-182
        • Mire-Sluis A.R.
        • Barrett Y.C.
        • Devanarayan V.
        • et al.
        Recommendations for the design and optimization of immunoassays used in the detection of host antibodies against biotechnology products.
        J Immunol Methods. 2004; 289: 1-16
        • Koren E.
        • Smith H.W.
        • Shores E.
        • et al.
        Recommendations on risk-based strategies for detection and characterization of antibodies against biotechnology products.
        J Immunol Methods. 2008; 333: 1-9
        • Bloem K.
        • Hernandez-Breijo B.
        • Martinez-Feito A.
        • et al.
        Immunogenicity of Therapeutic Antibodies: Monitoring Antidrug Antibodies in a Clinical Context.
        Ther Drug Monit. 2017; 39: 327-332
        • Gorovits B.
        • Baltrukonis D.J.
        • Bhattacharya I.
        • et al.
        Immunoassay methods used in clinical studies for the detection of anti-drug antibodies to adalimumab and infliximab.
        Clin Exp Immunol. 2018; 192: 348-365
        • Shen X.
        • Tang H.
        • McDanal C.
        • et al.
        SARS-CoV-2 variant B.1.1.7 is susceptible to neutralizing antibodies elicited by ancestral spike vaccines.
        Cell Host Microbe. 2021; 29 (e3): 529-539
        • Coleman C.N.
        • Cliffer K.D.
        • DiCarlo A.L.
        • et al.
        Preparedness for a 'no-notice' mass-casualty incident: a nuclear detonation scenario.
        Int J Radiat Biol. 2021; : 1-5
        • Sempowski G.D.
        • Saunders K.O.
        • Acharya P.
        • et al.
        Pandemic Preparedness: Developing Vaccines and Therapeutic Antibodies For COVID-19.
        Cell. 2020; 181: 1458-1463
        • Whitley J.
        • Zwolinski C.
        • Denis C.
        • et al.
        Development of mRNA manufacturing for vaccines and therapeutics: mRNA platform requirements and development of a scalable production process to support early phase clinical trials.
        Transl Res. 2022; 242: 38-55
        • U.S. Department of Health and Human Services Food and Drug Administration
        Immunogenicity Testing of Therapeutic Protein Products — Developing and Validating Assays for Anti-Drug Antibody Detection Guidance for Industry.
        Center for Drug Evaluation and Research (CDER); Center for Biologics Evaluation and Research (CBER). 2019; (Eds)
      1. Research Resouce Identification (RRID). Accessed 2/10/23, 2023. https://www.rrids.org/

        • Motulsky H.J.
        • Brown R.E.
        Detecting outliers when fitting data with nonlinear regression - a new method based on robust nonlinear regression and the false discovery rate.
        BMC Bioinformatics. 2006; 7: 123
        • Royston P.
        Remark AS R94: A Remark on Algorithm AS 181: The W-test for Normality.
        J R Statis Soc. Series C (Applied Statistics). 1995; 44: 547-551
        • Devanarayan V.
        • Smith W.C.
        • Brunelle R.L.
        • et al.
        Recommendations for systematic statistical computation of immunogenicity cut points.
        AAPS J. 2017; 19: 1487-1498
        • Von Grunigen R.
        • Schneider C.H.
        Epitope analysis: biotinylated short peptides as inhibitors of anti-peptide antibody.
        J Immunol Methods. 1989; 125: 143-146
        • Peter J.C.
        • Briand J.P.
        • Hoebeke J.
        How biotinylation can interfere with recognition: a surface plasmon resonance study of peptide-antibody interactions.
        J Immunol Methods. 2003; 274: 149-158