Advertisement

Establishing PK Equivalence Between Adalimumab and ABP 501 in the Presence of Antidrug Antibodies Using Population PK Modeling

Open AccessPublished:February 02, 2022DOI:https://doi.org/10.1016/j.clinthera.2021.11.016

      HIGHLIGHTS

      • ABP 501 (US: [adalimumab-atto]; EU: [adalimumab]) is an adalimumab biosimilar
      • Totality of evidence supported its approval
      • Population model assessed antidrug antibodies’ effect on pharmacokinetic parameters
      • Model-based evaluation confirmed pharmacokinetic similarity
      • Population modeling confirmed bioequivalence with reference product

      ABSTRACT

      Purpose

      ABP 501 (European Union, adalimumab; United States, adalimumab-atto) is a biosimilar to the adalimumab reference product (RP). A model was developed characterizing population pharmacokinetic (PK) variables of ABP 501 and adalimumab RP to include the impact of antidrug antibodies (ADAs).

      Methods

      Data were retrospectively analyzed from a single-dose, parallel-group bioequivalence study in healthy adults who received a single 40-mg SC injection of ABP 501 or adalimumab RP. Modeling was performed by using NONMEM 7.2. The impact of ADAs on PK similarity was assessed from population model-based AUC0–∞ values using ANCOVA.

      Findings

      Linear compartment models with various clearance pathways were compared with a one-compartment distribution, first-order subcutaneous absorption model. The final model, a one-compartment model with first-order subcutaneous absorption and linear clearance from the central compartment with an additional time-dependent linear clearance for ADA-positive subjects, described ABP 501 and adalimumab RP population PK variables. Model-derived estimates confirmed PK similarity for ABP 501 and adalimumab RP despite the impact of ADAs.

      Implications

      A traditional approach for evaluating bioequivalence based on noncompartmental analysis may be inappropriate for drugs with a high incidence of ADAs because accounting for the effect of ADAs on noncompartmental analysis parameters is challenging. Use of a population PK model to discern the effect of ADAs on drug PK variables allows for assessment of PK similarity accounting for the presence or absence of ADAs.

      Key words

      Introduction

      ABP 501 (European Union, adalimumab*; United States, adalimumab-atto) was approved as the first biosimilar to the adalimumab reference product (RP). ABP 501 has the same amino acid sequence as adalimumab RP; data from analytical studies along with functional, pharmacokinetic (PK) assessment of similarity in healthy subjects, and clinical comparisons of efficacy and safety in moderate-to-severe plaque psoriasis and moderate-to-severe rheumatoid arthritis, indicate that ABP 501 is highly similar to adalimumab RP.
      • Liu J
      • Eris T
      • Li C
      • Cao S
      • Kuhns S.
      Assessing analytical similarity of proposed Amgen biosimilar ABP 501 to adalimumab.
      • Velayudhan J
      • Chen Y-F
      • Rohrbach A
      • Pastula C
      • Maher G
      • Thomas H
      • Brown R
      • Born TL.
      Demonstration of functional similarity of proposed biosimilar ABP 501 to adalimumab.
      • Kaur P
      • Chow V
      • Zhang N
      • Moxness M
      • Kaliyaperumal A
      • Markus R.
      A randomized, single-blind, single-dose, three-arm, parallel-group study in healthy subjects to demonstrate pharmacokinetic equivalence of ABP 501 and adalimumab.
      • Papp K
      • Bachelez H
      • Costazo A
      • et al.
      Clinical similarity of the biosimilar ABP 501 compared with adalimumab after single transition: long-term results from a randomized controlled, double-blind, 52-week, phase III trial in patients with moderate-to-severe plaque psoriasis.
      • Papp K
      • Bachelez H
      • Costazo A
      • et al.
      Clinical similarity of biosimilar ABP 501 to adalimumab in the treatment of patients with moderate to severe plaque psoriasis: a randomized, double-blind, multicenter, phase III study.
      • Cohen S
      • Genovese MC
      • Choy E
      • et al.
      Efficacy and safety of the biosimilar ABP 501 compared with adalimumab in patients with moderate to severe rheumatoid arthritis: a randomized, double-blind, phase III equivalence study.
      • Cohen S
      • Pablos JL
      • Pavelka K
      • et al.
      An open-label extension study to demonstrate long-term safety and efficacy of ABP 501 in patients with rheumatoid arthritis.
      • Markus R
      • McBride HJ
      • Ramchandani M
      • et al.
      A review of the totality of evidence supporting the development of the first adalimumab biosimilar.
      • *Trademark: AMGEVITA™ (Amgen Europe, Breda, the Netherlands).
      • Trademark: AMJEVITA™ (Amgen Inc, Thousand Oaks, CA, USA).
      • Trademark: Humira® (AbbVie, North Chicago, IL, USA).
      Adalimumab RP treatment has been associated with a relatively high rate of antidrug antibodies (ADAs; up to 87%) in patients with inflammatory diseases.
      • Gorovits B
      • Baltrukonis DJ
      • Bhattacharya I
      • et al.
      Immunoassay methods used in clinical studies for the detection of anti-drug antibodies to adalimumab and infliximab.
      In the human PK and comparative clinical studies of ABP 501, a relatively high rate of ADA development was observed for ABP 501 and adalimumab RP (healthy subject PK study, 58.6%; comparative study in patients with rheumatoid arthritis, 38.2%; comparative study in patients with moderate-to-severe psoriasis, 72.3%).
      • Kaur P
      • Chow V
      • Zhang N
      • Moxness M
      • Kaliyaperumal A
      • Markus R.
      A randomized, single-blind, single-dose, three-arm, parallel-group study in healthy subjects to demonstrate pharmacokinetic equivalence of ABP 501 and adalimumab.
      • Papp K
      • Bachelez H
      • Costazo A
      • et al.
      Clinical similarity of the biosimilar ABP 501 compared with adalimumab after single transition: long-term results from a randomized controlled, double-blind, 52-week, phase III trial in patients with moderate-to-severe plaque psoriasis.
      ,
      • Cohen S
      • Genovese MC
      • Choy E
      • et al.
      Efficacy and safety of the biosimilar ABP 501 compared with adalimumab in patients with moderate to severe rheumatoid arthritis: a randomized, double-blind, phase III equivalence study.
      The single-dose PK study in healthy subjects found differences in PK parameters between subjects who developed ADAs and those who did not. The AUC was reduced by 20% to 30% in subjects who developed ADAs. In addition, terminal elimination t1/2 in ADA-positive subjects decreased by approximately one half.
      • Kaur P
      • Chow V
      • Zhang N
      • Moxness M
      • Kaliyaperumal A
      • Markus R.
      A randomized, single-blind, single-dose, three-arm, parallel-group study in healthy subjects to demonstrate pharmacokinetic equivalence of ABP 501 and adalimumab.
      These findings support ADA development as a cause of interindividual variation in PK variables.
      Population PK analysis enables the identification and quantification of interindividual variation in parameters of interest and the influence of individual factors on the interindividual distribution in PK data.
      • Ternant D
      • Bejan-Angoulvant T
      • Passot C
      • Mulleman D
      • Paintaud G.
      Clinical pharmacokinetics and pharmacodynamics of monoclonal antibodies approved to treat rheumatoid arthritis.
      ,
      • Atiqi S
      • Hooijberg F
      • Loeff FC
      • Rispens T
      • Wolbink GJ.
      Immunogenicity of TNF-inhibitors.
      In this case, population PK analysis would allow quantification of the effect of ADAs on PK variables and an assessment of the differences in PK variables between ABP 501 and adalimumab RP in the presence of ADAs.
      The objectives of the present analysis were to use a population modeling approach to characterize the PK variables of ABP 501 and adalimumab RP after subcutaneous administration to healthy subjects. This included assessing the effect of ADAs on PK variables, estimating a model-based AUC0–∞ of ABP 501 and adalimumab RP, and confirming PK similarity between ABP 501 and adalimumab RP based on a population model-based AUC0–∞ in the presence of ADAs. The data used in these analyses were those from the PK equivalence study between ABP 501 and adalimumab RP.

      Materials and Methods

      Study Design, Subjects, Dosing, and Sampling

      Data were analyzed from the randomized, single-blind, single-dose, 3-arm, parallel-group bioequivalence study of healthy volunteers comparing ABP 501 with adalimumab RP. Details of the study have been published previously.
      • Kaur P
      • Chow V
      • Zhang N
      • Moxness M
      • Kaliyaperumal A
      • Markus R.
      A randomized, single-blind, single-dose, three-arm, parallel-group study in healthy subjects to demonstrate pharmacokinetic equivalence of ABP 501 and adalimumab.
      To be in alignment with regulatory requirements for comparative PK evaluation of proposed biosimilars, the study used adalimumab RP sourced from the United States and the European Union.
      US Department of Health and Human Services, Food and Drug Administration, Center for Drug Evaluation and Research, Center for Biologics Evaluation and Research
      Quality considerations in demonstrating biosimilarity of a therapeutic protein product to a reference product.
      ,

      European Medicines Agency. Guideline on similar biological medicinal products (revision 1). 2014. Available at: https://www.ema.europa.eu/en/documents/scientific-guideline/guideline-similar-biological-medicinal-products-rev1_en.pdf. Accessed September 24, 2020.

      Subjects received a single 40-mg SC injection of ABP 501 or adalimumab RP sourced from the United States (adalimumab US) or the European Union (adalimumab EU). The study included intensive PK sampling collected as follows: before dosing; 1, 4, 8, and 12 hours’ postdose; and 2, 3, 4, 5, 6, 7, 8, 9, 11, 14, 16, 22, 29, 36, 43, 50, 57, and 63 days’ postdose.
      • Kaur P
      • Chow V
      • Zhang N
      • Moxness M
      • Kaliyaperumal A
      • Markus R.
      A randomized, single-blind, single-dose, three-arm, parallel-group study in healthy subjects to demonstrate pharmacokinetic equivalence of ABP 501 and adalimumab.
      Subjects were healthy adults aged 18 to 45 years. Mean weight at screening for the ABP 501 arm was 72.9 kg (range, 48.3–101.1), 73.1 kg (range, 53.1–96.7 kg) for the adalimumab US arm, and 75.7 kg (range, 55.2–105.4 kg) for the adalimumab EU arm.

      Assays

      Serum concentrations of ABP 501, adalimumab US, and adalimumab EU were measured by using a validated electrochemiluminescence assay based on the Meso Scale Discovery platform (Pacific BioLabs, Hercules, CA, USA). The assay had a lower limit of quantitation of 50 ng/mL and an upper limit of quantitation of 12,800 ng/mL. The assay used an anti-idiotype monoclonal antibody to capture ABP 501 and adalimumab RP from test samples and a second ruthenium-labelled anti-idiotype to detect the bound study drugs.
      • Kaur P
      • Chow V
      • Zhang N
      • Moxness M
      • Kaliyaperumal A
      • Markus R.
      A randomized, single-blind, single-dose, three-arm, parallel-group study in healthy subjects to demonstrate pharmacokinetic equivalence of ABP 501 and adalimumab.
      Binding ADAs were determined by using a validated electrochemiluminescence assay, and neutralizing ADAs were detected by using a tumor necrosis factor-alpha (TNFα)-responding cell line–based assay described previously.
      • Kaur P
      • Chow V
      • Zhang N
      • Moxness M
      • Kaliyaperumal A
      • Markus R.
      A randomized, single-blind, single-dose, three-arm, parallel-group study in healthy subjects to demonstrate pharmacokinetic equivalence of ABP 501 and adalimumab.

      Dataset

      The analysis dataset was generated as comma delimited files according to the format required by NONMEM software (ICON Development Solutions, Ellicott City, MD, USA) and according to the data specifications in NONMEM. Concentration values for ABP 501, adalimumab US, or adalimumab EU that were below the lower limit of quantitation were retained in the dataset.
      • Beal SL
      • Sheiner LB
      • Boeckmann R
      • Bauer RJ.
      NONMEM user's guides (1989-2009).

      Software

      ABP 501, adalimumab US, or adalimumab EU serum concentration–time profiles were used for nonlinear mixed effect modeling (NONMEM) by extended least squares regression using NONMEM 7.2 on the NONMEM High Performance Computing System. First-order conditional estimate, first-order conditional estimate interaction, or stochastic approximation expectation maximization methods were evaluated.
      • Beal SL
      • Sheiner LB
      • Boeckmann R
      • Bauer RJ.
      NONMEM user's guides (1989-2009).
      ,
      • Beal SL.
      Population pharmacokinetic data and parameter estimation based on their first two statistical moments.
      Geometric mean ratios for model-estimated AUC0–∞ were calculated for comparisons of ABP 501 versus adalimumab US, ABP 501 versus adalimumab EU, and adalimumab US versus adalimumab EU. Point estimates and 90% CIs were estimated by using an ANCOVA model adjusted for region and weight with SAS version 9.4 (SAS Institute, Inc, Cary, NC, USA), with and without protein concentration adjustment.

      Analysis Plan

      The population PK analysis was performed by using a nonlinear mixed effects modeling approach according to a 4-phase plan: exploratory analysis, model development, model assessment/validation, and model-based simulations and statistical analysis of bioequivalence.

      Exploratory Analysis

      An initial exploratory analysis was performed to graphically assess individual concentration–time profiles to inspect for dose and/or time effects and assist in identifying potential outliers. Summary plots for ABP 501, adalimumab US, and adalimumab EU serum concentrations were plotted over time in a semi-log scale to aid in identification of potential effects of dose and/or time and to facilitate the identification of subjects deemed to be outliers. Data for individual subjects’ study drug serum concentrations versus time were generated together with the individual and population predictions based on the final population PK model.

      Model Development

      The initial model used for describing the PK profile of ABP 501 and adalimumab RP was a linear one-compartment absorption and disposition model based on previous analyses.
      • Doshi S
      • Krishnan E
      • Wang H
      • Zhang N
      • Chow V.
      Establishing pharmacokinetic equivalence between adalimumab and ABP 501 in the presence of anti-drug antibodies using population PK modeling.
      This simple model sufficiently described the PK profile of these agents. The model was parameterized in terms of first-order absorption, apparent linear clearance, and apparent volume of distribution for the central compartment.
      This simple one-compartment model did not account for the impact of ADAs on the PK profiles of ABP 501 and adalimumab RP. To do so, a structural model was developed using a data-driven process. More complex clearance pathways, including Michaelis-Menten clearance and time-dependent and concentration-dependent clearances, were tested during the analysis. Michaelis-Menten clearance was parameterized in terms of maximum rate and concentration associated with 50% of maximum rate (Km). Time-dependent clearance was parameterized as an additional linear clearance with a time of onset.
      Study drug values below the lower limit of quantitation (BQL) were retained in the analysis and analyzed with the M3 method.
      • Beal SL.
      Ways to fit a PK model with some data below the quantification limit.
      ,
      • Ahn JE
      • Karlsson MO
      • Dunne A
      • Ludden TM.
      Likelihood based approaches to handling data below the quantification limit using NONMEM VI.
      To account for differences in protein content between treatments, nominal doses (40 mg) were updated in the analysis based on the protein content obtained from study products of the respective treatment: 38.32 mg for ABP 501, 39.84 mg for adalimumab US, and 42.84 mg for adalimumab EU.
      Visual inspection of individual observed serum concentrations versus time profiles suggested significant variability in drug concentration between subjects. Therefore, between subjects (interindividual), variability in a model parameter, P, was included in the model and was assumed to be log normally distributed, per Equation 1:
      Pj=P·exp(ηj)


      where Pj is the PK parameter for the j-th individual, P is the population typical value for the PK parameter, ηj is a random interindividual effect and it is assumed to be a random Gaussian variable with zero mean and variance (ω2) that distinguished the j-th individual's PK parameter from the population typical value (P) as estimated by the regression model. The magnitude of interindividual variability in the PK parameters was expressed as a %CV.
      Residual variability was evaluated by using an additive error model after natural logarithmic transformation of the measured concentrations and model predictions according to Equation 2:
      LnCobs=LnCpred+ε


      where Cobs is the observed ABP 501, adalimumab US, or adalimumab EU concentrations; Cpred is the corresponding model-predicted ABP 501, adalimumab US, or adalimumab EU concentrations; and Ɛ was the residual departure of the natural logarithm of the observed concentration from the predicted concentration in serum, which was assumed to follow an independent Gaussian distribution with mean zero and variances σ2.

      Model Assessment

      The improvement of the fit obtained was assessed by examination of several diagnostics. Change in the minimum value of the objective function (MVOF) was examined and used as a criterion for the inclusion of covariates in the final model. For the addition of one fixed effect, a change in the MVOF (ΔMVOF) of ≥10.8 is required to reach statistical significance (α = 0.001).
      • Hutmacher MM
      • Kowalski KG.
      Covariate selection in pharmacometric analyses: a review of methods.
      The goodness-of-fit of NONMEM analyses were also assessed by examination of scatterplots of observed concentrations versus population predicted concentrations and versus individual predicted concentrations.
      Residual-based diagnostics (conditional weighted residuals and numerical prediction distribution error) were not evaluated because of the prevalence of BQL observations and because of the use of the M3 method in the analysis. The shrinkage of random effects was assessed as described by Karlsson et al.
      • Karlsson MO
      • Savic RM.
      Diagnosing model diagnostics.
      The model was fitted to the analysis data, and the resulting MVOF was used as a starting value to test the significance of the covariates. Body weight, plasma albumin, and ADA status are known to affect the adalimumab PK profile.

      European Medicines Agency. Humira® (adalimumab). European Public Assessment Report: Scientific discussion. Procedure No. EMEA/H/C/000481/II/0082, EMA/217675/2012. London, UK: 15 March 2012. Available at: https://www.ema.europa.eu/en/documents/variation-report/humira-h-c-481-ii-0082-epar-assessment-report-variation_en.pdf. Accessed November 17, 2020.

      Because the analysis used PK data from healthy subjects, plasma albumin values were within the normal range, and the effect of ADA was included in the structure model, body weight was the only demographic factor evaluated as a potential covariate. The inclusion of body weight in the population model was considered if a decrease in the MVOF >7.88 (χ2 test; df, 1; P < 0.005) was obtained. Noncompartmental analysis of the unadjusted PK parameters did not show significant differences in exposure by treatment, and thus treatment effect was not prospectively tested as a covariate. The effect of treatment on PK parameters was evaluated by graphical and statistical comparisons of the empirical Bayes estimates of the individual PK model parameters.
      Body weight was included through the centering on the median and power equations. The following general form was used to constrain the typical value of this parameter to be positive through the convergence process and was also in line with the assumed log-normal distribution (Equation 3):
      TVP=θn·m=1M(covmrefm)θ(n+m)·p=1Pθ(n+M=p)covM+p


      where the typical value of a model parameter was described as a function of M, individual continuous covariates (covm, m = 1,…,M) and P individual categorical (0 or 1) covariates (covM+p), such that θn is an estimated parameter describing the typical value of a model parameter for an individual with covariates equal to the reference covariate values (covm-refm, covM = p = 0). θ(n+m) and θ(n+M+p) are estimated parameters that describe the magnitude of the covariate–parameter relationships.
      The resulting random effects accounting for interindividual differences in model parameters were examined graphically by plotting each one against all others and by fitting the final model with all non-diagonal random effects implemented ($OMEGA BLOCK option in NONMEM). When implementation of a correlation significantly improved the fit, the non-diagonal element of the corresponding covariance between random effects was introduced into the interindividual random effect matrix and was kept in the model, and the process was repeated until no further improvement of the fit could be achieved.
      The covariance step was examined, and the asymptotic SEs of fixed and random effects produced by NONMEM were used to construct the asymptotic 95% CIs. Correlations between population parameters, the eigenvalues, and the condition number were evaluated to ensure that the model is not ill-conditioned and thus would not provide the final PK model for the study drugs.

      Model Validation

      Visual predictive checks were performed as an internal evaluation method for the model developed to describe the concentration–time profile on linear and log scale. Separate visual predictive checks were used for the absorption and elimination phases and were conducted by using previously published techniques.
      • Yano Y
      • Beal SL
      • Sheiner LB.
      Evaluating pharmacokinetic/pharmacodynamic models using the posterior predictive check.
      • Wilkins JJ
      • Karlsson MO
      • Jonsson EN.
      Patterns and power for the visual predictive check.
      • Post TM
      • Freijer JI
      • Ploeger BA
      • Danhof M.
      Extensions to the visual predictive check to facilitate model performance evaluation.
      Based on the parameter estimates of the model, 500 replicates of the serum concentration profiles were simulated. The median and 95% prediction interval were computed for each replicate and summarized across replicates. Distributions of the simulated prediction intervals were compared with the distributions of the observed data points in log scale, linear scale, and during the absorption phase (until day 9) stratified according to any relevant covariates. The proportion of BQL over time was summarized for each replicate and across replicates and compared with the observed proportion BQL from day 21 to day 63 stratified according to any relevant covariates.

      Model-based Simulations and Statistical Analysis of Bioequivalence

      Model-based simulations were performed to evaluate the differences in distribution of AUC0–∞ according to treatment group. After model evaluation, the individual post hoc PK parameter estimates from the final population PK model were used to estimate study drug exposures (AUC0–∞) for each subject based on their respective individual predicted PK parameters and study drug received. For evaluation of PK similarity, point estimates of 90% CIs of the geometric mean ratio for the model- estimated AUC0–∞, with and without adjustment for differences in protein content for comparisons of ABP 501 versus adalimumab US, ABP 501 versus adalimumab EU, and adalimumab US versus adalimumab EU, were estimated by using an ANCOVA model. The model was adjusted for region and weight with SAS version 9.4.

      Results

      Dataset

      Demographic data for the subjects in the study have been published previously.
      • Kaur P
      • Chow V
      • Zhang N
      • Moxness M
      • Kaliyaperumal A
      • Markus R.
      A randomized, single-blind, single-dose, three-arm, parallel-group study in healthy subjects to demonstrate pharmacokinetic equivalence of ABP 501 and adalimumab.
      The modeling dataset included 4408 serum samples from 203 healthy subjects who received a single 40-mg SC injection of ABP 501 (38.32 mg based on protein content; n = 67), adalimumab US (39.84 mg protein; n = 69), or adalimumab EU (42.64 mg protein; n = 67). The difference from the nominal dose of 40 mg due to protein content of each treatment was incorporated into the modeling exercise by adjusting the dose based on the actual protein content.

      Model Development

      Several linear one-compartment models with different parameterizations of clearance pathways were evaluated (Table I). The reference model was the one-compartment distribution and first-order SC absorption model with BQL observations analyzed with the M3 method.
      • Beal SL.
      Ways to fit a PK model with some data below the quantification limit.
      ,
      • Ahn JE
      • Karlsson MO
      • Dunne A
      • Ludden TM.
      Likelihood based approaches to handling data below the quantification limit using NONMEM VI.
      After assessing the relationship between body weight and individual clearance and volume estimates, the effect of body weight was added to clearance and volume (ΔMVOF, –62.17). The body weight effect was centered at the median study body weight (72.7 kg), and the exponent was fixed to 0.75 for clearance and 1.0 for volume. To further evaluate the effect of potential nonlinear kinetics, Michaelis-Menten clearance was evaluated instead of linear clearance and in parallel with linear clearance. A model with Michaelis-Menten clearance was evaluated, and the model fit showed improvement over the base model (Δ, –1364.291) (Table 1). A model with parallel linear and Michaelis-Menten clearance was also evaluated. Although this model resulted in a lower objection function relative to a model with Michaelis-Menten clearance alone, the Km estimate was forced to the lower bound (0.01 ng/mL), which is >2 orders of magnitude lower than the concentrations observed in the study (50–5800 ng/mL), and therefore this model was not considered appropriate. When the additional clearance pathway was limited to ADA-positive subjects, incorporation of Michaelis-Menten clearance resulted in a more robust model fit with demonstrated improvement over the reference model (Δ, –1541.96).
      Table IComparison of the various structural population pharmacokinetic/pharmacodynamic models evaluated for ABP 501 and the adalimumab reference product sourced from the United States or the European Union.
      Pharmacokinetic Model DescriptionNONMEM MVOFΔMVOFAIC
      Reference one-compartment with M3 method–3755.960 (reference)–3741.96
      Add effect of weight to clearance and volume–3818.13–62.17–3804.13
      Test Michaelis-Menten clearance instead of linear clearance–5120.251–1364.291–5104.25
      Test parallel Michaelis-Menten clearance and linear clearance–5469.129–1713.169–5449.13
      Test linear clearance for ADA-negative subjects and parallel linear and nonlinear clearance for ADA-positive subjects–5297.92–1541.96–5275.92
      Test linear clearance for ADA-negative subjects and additional linear clearance for ADA-positive subjects–3845.52–89.56–3827.52
      Test linear clearance for ADA-negative subjects and time-dependent additional linear clearance for ADA-positive subjects–5347.88–1591.92–5325.88
      ADA = antidrug antibody; AIC = Akaike information criterion; MVOF = minimal value of objective function; NONMEM = nonlinear mixed effects modeling.
      Although there was obvious benefit of inclusion of Michaelis-Menten clearance for ADA-positive subjects, the observed concentration profiles of ADA-positive subjects did not show that the more rapid clearance in these subjects was concentration dependent. To assess the potential of a concentration-independent clearance pathway, a second additive linear clearance for ADA-positive subjects was added instead of Michaelis-Menten clearance. Including the second linear clearance for ADA-positive subjects improved the model fit slightly compared with the reference model (ΔMVOF, –27.39) but not as drastically as adding the Michaelis-Menten clearance (Table I). Because it was not evident that the increased clearance was concentration dependent and any ADA effect is likely time dependent, a time-dependent onset was added to a second additive linear clearance pathway for ADA-positive subjects. This addition resulted in a larger drop in MVOF than inclusion of the Michaelis-Menten clearance (ΔMVOF, –1591.92 vs –1541.96).
      Graphical and statistical comparisons of empirical Bayes estimates per treatment revealed no difference in any PK model parameter according to treatment. As shown in the box plots (Figure 1), there was no apparent association between interindividual random effects (ANOVA testing) for each PK model parameter and study treatment. Because evidence for an effect of treatment on PK parameters was not found, treatment effect was not tested in the PK model.
      Figure 1
      Figure 1Relationship between the individual random effects and treatment. ADA = antidrug antibody; CL = apparent linear clearance; EU = European Union; US = United States.
      The PK profile of ABP 501, adalimumab US, and adalimumab EU following a single subcutaneous administration in ADA-negative healthy subjects was adequately described by a one-compartment model with first-order absorption and linear clearance. For ADA-positive subjects, the PK variables were adequately described by using a one-compartment model with an additional additive time-dependent linear clearance component.

      Model Evaluation

      Final parameter estimates for ABP 501, adalimumab US, and adalimumab EU from the final model are presented in Table II. The population estimates and interindividual variability of clearance and volume were 0.399 (33.6%) L/d and 8.94 (23.5%) L, respectively, for a 72.7 kg subject (median body weight of subjects in this analysis). The time to onset and magnitude of the additive ADA-related linear clearance was 33.7 (54.4%) days and 1.15 (52.6%) L/d.
      Table IIParameter estimates from the final population pharmacokinetic model.
      Parameter
      Exponent for effect of body weight on clearance and volume fixed to 0.75 and 1, respectively.
      Estimate% RSEUnits
      Absorption rate0.3954.20Day–1
      Clearance0.3992.61L/d/72.7 kg
      Volume8.942.16L/72.7 kg
      ADA-related clearance1.156.39L/d
      Onset of ADA-related clearance33.75.37Day
      Intersubject variability
       Absorption rate55.610.6%CV
       Clearance33.611.5%CV
       Volume23.518.7%CV
       ADA-related clearance52.628.0%CV
       Onset of ADA-related clearance54.413.7%CV
      Residual error
       Proportional error22.12.88%CV
      ADA = antidrug antibody; RSE = relative SE ([SE/parameter estimate] × 100).
      low asterisk Exponent for effect of body weight on clearance and volume fixed to 0.75 and 1, respectively.
      Diagnostic plots exhibited good concordance between observed and population or individual predicted concentrations without any bias (Figure 2). Visual predictive checks showed that the model clearly described the absorption phase, concentrations over the duration of the study (Figure 3), and the proportion of observation BQL after a single dose of ABP 501, adalimumab US, and adalimumab EU. Plots of observed concentration profiles overlaid with the individual predicted profile showed that the model described each individual subject well and can be used to estimate accurately each individual's AUC0–∞ for all 203 subjects who completed the study (Figure 4). In contrast, a noncompartmental approach model quantified the AUC0–∞ for only 176 subjects.
      Figure 2
      Figure 2Diagnostic plots for the final model. The blue line is the lowest (local regression smoother) trend line.
      Figure 3
      Figure 3Visual predictive check for the final model complete profile, stratified according to treatment, antidrug antibody (ADA) status, and type in log scale. Open circles and green lines represent the observed concentrations and the 2.5th, 50th, and 97.5th percentiles of observed concentrations, respectively. Red lines and blue shaded areas represent the 2.5th, 50th, and 97.5th prediction intervals and the 95% prediction interval of each red line, across 500 simulation replicates.
      Figure 4
      Figure 4Overlaid observed and predicted concentration–time curves for healthy subjects who received ABP 501 and the adalimumab reference product sourced from the United States (adalimumab US) or the European Union (adalimumab EU). ABP 501, n = 67; adalimumab EU, n = 67; adalimumab US, n = 69.

      Bioequivalence

      Comparisons of the individual estimated model based on AUC0–∞ for all 203 subjects according to treatment group, with and without protein content adjustment, found that the 90% CIs of the geometric mean ratios of AUC0–∞ all lie within the conventional bioequivalence criteria of 80% to 125% (Table III). This finding confirms the PK similarity of ABP 501 and adalimumab RP.
      Table IIISummary of the statistical assessment of adjusted AUC0–∞ derived from the final pharmacokinetic model parameters.
      ComparisonTest (n)Geometric LS MeanRatio of Geometric LS Means (90% CI)
      Reference (n)TestReference
      ABP 501 vs adalimumab US
      ABP 501 was the test treatment, and adalimumab was the reference treatment.
      67692146.271986.301.10 (0.9784–1.2269)
      ABP 501 vs adalimumab EU
      ABP 501 was the test treatment, and adalimumab was the reference treatment.
      67672176.271977.641.10 (0.9813–1.2341)
      Adalimumab US vs adalimumab EU
      Adalimumab reference product sourced from the United States (adalimumab US) was the test treatment and adalimumab sourced from the European Union (adalimumab EU) was the reference treatment.
      69671986.301977.641.00 (0.8964–1.1253)
      LS = least squares.
      low asterisk ABP 501 was the test treatment, and adalimumab was the reference treatment.
      Adalimumab reference product sourced from the United States (adalimumab US) was the test treatment and adalimumab sourced from the European Union (adalimumab EU) was the reference treatment.

      Discussion

      The use of modeling and simulation in the development of biosimilar therapeutic proteins has been reviewed and identified as invaluable for optimizing the design of comparative PK and pharmacodynamic studies, including the selection of doses to best show PK and pharmacodynamic comparability.
      • Dodds M
      • Chow V
      • Markus R
      • Pérez-Ruixo J
      • Shen D
      • Gibbs M.
      The use of pharmacometrics to optimize biosimilar development.
      ,
      • Wang Y-M
      • Wang Y
      • Schrieber SJ
      • et al.
      Role of modeling and simulation in the development of novel and biosimilar therapeutic proteins.
      Further use of modeling and simulation to show no clinically meaningful differences between the RP and biosimilar in biosimilar development is also a topic of interest.
      • Zhu P
      • Ji P
      • Wang Y
      Using clinical PK/PD studies to support no clinically meaningful differences between a proposed biosimilar and the reference product.
      Previously, population PK modeling was applied to assess PK similarity for another adalimumab biosimilar and a few other biosimilars.
      • Kang J
      • Eudy-Byrne R
      • Mondick J
      • Knebel W
      • Jayadeva G
      • Liesenfeld K-H.
      Population pharmacokinetics of adalimumab biosimilar adalimumab-adbm and reference product in healthy subjects and patients with rheumatoid arthritis to assess pharmacokinetic similarity.
      • Chen X
      • Li C
      • Ewesuedo R
      • Yin D.
      Population pharmacokinetics of PF-05280014 (a trastuzumab biosimilar) and reference trastuzumab (Herceptin®) in patients with HER2-positive metastatic breast cancer.
      • Candelaria M
      • Gonzalez D
      • Torresan M
      • et al.
      Results of the population PK modelling of RTXMB83, a rituximab biosimilar candidate in patients with diffuse large B-cell lymphoma (DLBCL).
      This analysis used data from a previously published PK study comparing ABP 501 and adalimumab RP to develop a population PK model describing ABP 501 and adalimumab RP following a single SC administration to healthy adult subjects and to further assess the impact of ADAs on the PK similarity of ABP 501 and adalimumab RP. The population PK profile for ABP 501 and adalimumab RP after SC administration was linear and adequately described by a one-compartment model with first-order absorption and linear clearance from the central compartment in subjects with negative ADA status. However, as shown here, a linear model with additive time-dependent clearance more appropriately described the PK variables of ABP 501 and adalimumab RP in healthy subjects with a positive ADA status than a standard concentration-dependent Michaelis-Menten clearance model. Individual post hoc PK parameters provided robust model-based estimates of individual AUC0–∞ for all subjects in the study. PK parameter estimates obtained from the population PK model confirmed the similarity of PK parameters for ABP 501 and adalimumab RP within the subpopulations of ADA-negative and ADA-positive subjects.
      Development of ADAs to several therapeutic anti-TNFα monoclonal antibodies, including adalimumab RP, has been shown to decrease antibody concentration through more rapid clearance.
      • Ternant D
      • Bejan-Angoulvant T
      • Passot C
      • Mulleman D
      • Paintaud G.
      Clinical pharmacokinetics and pharmacodynamics of monoclonal antibodies approved to treat rheumatoid arthritis.
      Although previous models have described ADA-mediated drug clearance via a nonlinear concentration-dependent mechanism,
      • Ng CM
      • Loyet KM
      • Iyer S
      • Fielder PJ
      • Deng R.
      Modeling approach to investigate the effect of neonatal Fc receptor binding affinity and anti-therapeutic antibody on the pharmacokinetic of humanized monoclonal anti-tumor necrosis factor-α IgG antibody in cynomolgus monkey.
      the present analysis found that inclusion of an additional time-dependent linear clearance resulted in more plausible PK parameter estimates and an acceptable description of the observed PK data. The results of the parallel linear clearance and ADA-mediated nonlinear clearance model revealed a very low population estimate of Km (0.01 ng/mL), which was much lower than the observed concentrations (50–5800 ng/mL). The very low estimate of Km also indicated that the nonlinear clearance was an added linear clearance from time of dosing to the end of study, which is not consistent with development of ADAs postadministration. A time-dependent ADA clearance mechanism may be more appropriate as ADA-positive subjects have a common linear clearance with ADA-negative subjects and an additional linear clearance corresponding to individual onset of ADA development. In addition, the time-dependent ADA clearance is concentration independent with the same additional clearance rate regardless of drug concentration, which corresponds conceptually to the Km value estimated for the nonlinear clearance model.
      Data showing similarity in PK parameters between a proposed biosimilar and its RP is a fundamental component of the totality of evidence of similarity of the biosimilar.
      • Markus R
      • Liu J
      • Ramchandani M
      • Landa D
      • Born T
      • Kaur P.
      Developing the totality of evidence for biosimilars: regulatory considerations and building confidence for the healthcare community.
      The analysis population consisted of healthy subjects chosen for the original comparative PK study to provide a population that would be homogeneous and highly sensitive with the absence of underlying disease and lack of use of immunosuppressive drugs.
      • Kaur P
      • Chow V
      • Zhang N
      • Moxness M
      • Kaliyaperumal A
      • Markus R.
      A randomized, single-blind, single-dose, three-arm, parallel-group study in healthy subjects to demonstrate pharmacokinetic equivalence of ABP 501 and adalimumab.
      The analysis plan did not include assessment of relationship of ADA status with therapeutic response or adverse events. Instead, per its objective, this analysis provided further confirmation of the similarity of the PK properties of ABP 501 and adalimumab RP in the presence or absence of ADAs. The similarity of efficacy and safety of ABP 501 and adalimumab RP has been shown in comparative clinical studies in patients with rheumatoid arthritis and plaque psoriasis.
      • Papp K
      • Bachelez H
      • Costazo A
      • et al.
      Clinical similarity of biosimilar ABP 501 to adalimumab in the treatment of patients with moderate to severe plaque psoriasis: a randomized, double-blind, multicenter, phase III study.
      ,
      • Cohen S
      • Genovese MC
      • Choy E
      • et al.
      Efficacy and safety of the biosimilar ABP 501 compared with adalimumab in patients with moderate to severe rheumatoid arthritis: a randomized, double-blind, phase III equivalence study.

      Conclusions

      A population PK model was developed that described observed data for healthy subjects who received a single 40-mg SC dose of ABP 501 or adalimumab RP. The PK parameter estimates obtained from the population PK model confirmed the similarity of PK parameters for ABP 501 and adalimumab RP in the presence and absence of ADAs. The population PK profile for ABP 501 and adalimumab RP was linear and was adequately described by a one-compartment model with first-order absorption and linear clearance from the central compartment in ADA-negative healthy subjects. Inclusion of an additive time-dependent linear clearance allowed for appropriate characterization of ABP 501 and RP serum concentrations for ADA-positive subjects in this study.

      Acknowledgments

      Funding for the Phase I pharmacokinetic study (EudraCT number 2012-000785-37) and for this analysis was provided by Amgen Inc. Medical writing assistance was provided by Annette F. Skorupa, PhD, Innovation Communications Group, under the guidance of Monica Ramchandani, PhD, Amgen Inc.
      Mr Doshi was responsible for conceptualization, methodology, formal analysis, writing, and visualization; Dr Wang reviewed study data and was responsible for statistical methodology and analysis; and Dr Chow contributed to design, evaluation, and interpretation of study and results.

      Disclosures

      All authors are employees/former employees and stockholders of Amgen Inc. The authors have indicated that they have no other conflicts of interest regarding the content of this article.

      References

        • Liu J
        • Eris T
        • Li C
        • Cao S
        • Kuhns S.
        Assessing analytical similarity of proposed Amgen biosimilar ABP 501 to adalimumab.
        BioDrugs. 2016; 30: 321-338
        • Velayudhan J
        • Chen Y-F
        • Rohrbach A
        • Pastula C
        • Maher G
        • Thomas H
        • Brown R
        • Born TL.
        Demonstration of functional similarity of proposed biosimilar ABP 501 to adalimumab.
        BioDrugs. 2016; 30: 339-351
        • Kaur P
        • Chow V
        • Zhang N
        • Moxness M
        • Kaliyaperumal A
        • Markus R.
        A randomized, single-blind, single-dose, three-arm, parallel-group study in healthy subjects to demonstrate pharmacokinetic equivalence of ABP 501 and adalimumab.
        Ann Rheum Dis. 2017; 76: 526-533
        • Papp K
        • Bachelez H
        • Costazo A
        • et al.
        Clinical similarity of the biosimilar ABP 501 compared with adalimumab after single transition: long-term results from a randomized controlled, double-blind, 52-week, phase III trial in patients with moderate-to-severe plaque psoriasis.
        Br J Dermatol. 2017; 177: 1562-1574
        • Papp K
        • Bachelez H
        • Costazo A
        • et al.
        Clinical similarity of biosimilar ABP 501 to adalimumab in the treatment of patients with moderate to severe plaque psoriasis: a randomized, double-blind, multicenter, phase III study.
        J Acad Dermatol. 2017; 76: 1093-1102
        • Cohen S
        • Genovese MC
        • Choy E
        • et al.
        Efficacy and safety of the biosimilar ABP 501 compared with adalimumab in patients with moderate to severe rheumatoid arthritis: a randomized, double-blind, phase III equivalence study.
        Ann Rheum Dis. 2017; 0: 1-9
        • Cohen S
        • Pablos JL
        • Pavelka K
        • et al.
        An open-label extension study to demonstrate long-term safety and efficacy of ABP 501 in patients with rheumatoid arthritis.
        Arthritis Res Ther. 2019; 21: 84-93
        • Markus R
        • McBride HJ
        • Ramchandani M
        • et al.
        A review of the totality of evidence supporting the development of the first adalimumab biosimilar.
        Adv Ther. 2019; 36: 1833-1850
        • Gorovits B
        • Baltrukonis DJ
        • 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; 19: 348-365
        • Ternant D
        • Bejan-Angoulvant T
        • Passot C
        • Mulleman D
        • Paintaud G.
        Clinical pharmacokinetics and pharmacodynamics of monoclonal antibodies approved to treat rheumatoid arthritis.
        Clin Pharmacokinet. 2015; 54: 1107-1123
        • Atiqi S
        • Hooijberg F
        • Loeff FC
        • Rispens T
        • Wolbink GJ.
        Immunogenicity of TNF-inhibitors.
        Frontiers Immunol. 2020; 11: 213
        • US Department of Health and Human Services, Food and Drug Administration, Center for Drug Evaluation and Research, Center for Biologics Evaluation and Research
        Quality considerations in demonstrating biosimilarity of a therapeutic protein product to a reference product.
        Guidance for industry, 2015 (Available at:) (Accessed September 24, 2020)
      1. European Medicines Agency. Guideline on similar biological medicinal products (revision 1). 2014. Available at: https://www.ema.europa.eu/en/documents/scientific-guideline/guideline-similar-biological-medicinal-products-rev1_en.pdf. Accessed September 24, 2020.

        • Beal SL
        • Sheiner LB
        • Boeckmann R
        • Bauer RJ.
        NONMEM user's guides (1989-2009).
        Icon Development Solutions, Ellicott City, MD, USA2009
        • Beal SL.
        Population pharmacokinetic data and parameter estimation based on their first two statistical moments.
        Drug Metab Rev. 1984; 15: 173-193
        • Doshi S
        • Krishnan E
        • Wang H
        • Zhang N
        • Chow V.
        Establishing pharmacokinetic equivalence between adalimumab and ABP 501 in the presence of anti-drug antibodies using population PK modeling.
        European Crohn's and Colitis Organisation, 2018 (Available at:) (Accessed November 17, 2020)
        • Beal SL.
        Ways to fit a PK model with some data below the quantification limit.
        J Pharmacokinet Pharmacodyn. 2001; 28: 481-504
        • Ahn JE
        • Karlsson MO
        • Dunne A
        • Ludden TM.
        Likelihood based approaches to handling data below the quantification limit using NONMEM VI.
        J Pharmacokinet Pharmacodyn. 2008; 35: 401-421
        • Hutmacher MM
        • Kowalski KG.
        Covariate selection in pharmacometric analyses: a review of methods.
        Br J Clin Pharmacol. 2015; 79: 132-147
        • Karlsson MO
        • Savic RM.
        Diagnosing model diagnostics.
        Clin Pharmacol Ther. 2007; 82: 17-20
      2. European Medicines Agency. Humira® (adalimumab). European Public Assessment Report: Scientific discussion. Procedure No. EMEA/H/C/000481/II/0082, EMA/217675/2012. London, UK: 15 March 2012. Available at: https://www.ema.europa.eu/en/documents/variation-report/humira-h-c-481-ii-0082-epar-assessment-report-variation_en.pdf. Accessed November 17, 2020.

        • Yano Y
        • Beal SL
        • Sheiner LB.
        Evaluating pharmacokinetic/pharmacodynamic models using the posterior predictive check.
        J Pharmacokinet Pharmacodyn. 2001; 28: 171-192
        • Wilkins JJ
        • Karlsson MO
        • Jonsson EN.
        Patterns and power for the visual predictive check.
        PAGE. 2006; (Abstract 1029, p 15. Available at:) (Accessed November 20, 2020)
        • Post TM
        • Freijer JI
        • Ploeger BA
        • Danhof M.
        Extensions to the visual predictive check to facilitate model performance evaluation.
        J Pharmacokinet Pharmacodyn. 2008; 35: 185-202
        • Dodds M
        • Chow V
        • Markus R
        • Pérez-Ruixo J
        • Shen D
        • Gibbs M.
        The use of pharmacometrics to optimize biosimilar development.
        J Pharm Sci. 2013; 102: 3908-3914
        • Wang Y-M
        • Wang Y
        • Schrieber SJ
        • et al.
        Role of modeling and simulation in the development of novel and biosimilar therapeutic proteins.
        J Pharm Sci. 2019; 108: 73-77
        • Zhu P
        • Ji P
        • Wang Y
        Using clinical PK/PD studies to support no clinically meaningful differences between a proposed biosimilar and the reference product.
        AAPS J. 2018; 20: 89
        • Kang J
        • Eudy-Byrne R
        • Mondick J
        • Knebel W
        • Jayadeva G
        • Liesenfeld K-H.
        Population pharmacokinetics of adalimumab biosimilar adalimumab-adbm and reference product in healthy subjects and patients with rheumatoid arthritis to assess pharmacokinetic similarity.
        Br J Clin Pharmacol. 2020; 86: 2274-2285
        • Chen X
        • Li C
        • Ewesuedo R
        • Yin D.
        Population pharmacokinetics of PF-05280014 (a trastuzumab biosimilar) and reference trastuzumab (Herceptin®) in patients with HER2-positive metastatic breast cancer.
        Cancer Chemother Pharmacol. 2019; 84: 83-92
        • Candelaria M
        • Gonzalez D
        • Torresan M
        • et al.
        Results of the population PK modelling of RTXMB83, a rituximab biosimilar candidate in patients with diffuse large B-cell lymphoma (DLBCL).
        Clin Ther. 2017; 39: E6-E7
        • Ng CM
        • Loyet KM
        • Iyer S
        • Fielder PJ
        • Deng R.
        Modeling approach to investigate the effect of neonatal Fc receptor binding affinity and anti-therapeutic antibody on the pharmacokinetic of humanized monoclonal anti-tumor necrosis factor-α IgG antibody in cynomolgus monkey.
        Eur J Pharm Sci. 2014; 51: 51-58
        • Markus R
        • Liu J
        • Ramchandani M
        • Landa D
        • Born T
        • Kaur P.
        Developing the totality of evidence for biosimilars: regulatory considerations and building confidence for the healthcare community.
        BioDrugs. 2017; 31: 175-187