1 Introduction
Ulcerative colitis (UC) is a chronic disease of unknown etiology that is characterized by inflammation of the rectum and colon resulting in symptoms such as diarrhea, rectal bleeding, urgency, and tenesmus [
1,
2]. Treatment goals in UC include induction of remission (typically within a 6-week to 12-week time frame) and maintenance of remission in the longer term (assessed over 52 weeks of continuous treatment in clinical trials) [
3]. Only a proportion of patients respond to current small-molecule and biologic therapies, and there is an associated increased risk of infections [
4].
Interleukin-23 (IL-23) is a heterodimeric cytokine that is crucial for maintenance and amplification of T helper 17 cells and stimulation of many innate immune cells [
5,
6], which are both important processes in UC and other chronic inflammatory diseases [
7]. Several monoclonal antibodies that selectively target the p19 subunit of IL-23 are effective in patients with Crohn’s disease [
8‐
10] and psoriasis [
11‐
16]. Selective targeting of IL-23 is more effective than non-selective IL-23 targeting in patients with psoriasis and is a promising strategy for improved treatment of patients with moderately to severely active UC [
17]. The pharmacokinetics of selective anti-IL-23 drugs for the treatment of patients with UC has not been extensively published [
18].
Mirikizumab is a humanized IgG4-variant monoclonal antibody directed against the p19 subunit of IL-23 whose efficacy and safety have been assessed in adults with moderately to severely active UC in three multicenter, randomized, double-blind, placebo-controlled clinical trials: the induction and maintenance phase II AMAC trial (NCT02589665), the induction phase III LUCENT 1 trial (NCT03518086, AMAN), and the maintenance phase III LUCENT 2 trial (NCT03524092, AMBG) [
19‐
22]. In these trials, mirikizumab was well tolerated, and clinical remission rates were higher in patients treated with mirikizumab compared with placebo during induction and maintenance therapy. Pharmacokinetic (PK) data were available from a total of 1362 patients from the three clinical studies. Data from the phase II trial were analyzed separately from the data from the two phase III trials using a non-linear mixed-effects modeling approach. The objectives of the current analyses were to characterize mirikizumab pharmacokinetics in patients with moderately to severely active UC and to evaluate the impact of the relevant patient-specific covariates on mirikizumab systemic exposures to inform dosing recommendations.
2 Methods
2.1 Patients and Study Designs
The three clinical trials were designed to evaluate the efficacy, safety, and pharmacokinetics of mirikizumab in inducing (AMAC and LUCENT 1) and maintaining (AMAC and LUCENT 2) remission in patients aged 18–75 years (AMAC) or 18–80 years (LUCENT 1+2) with moderately to severely active UC. Details of the three trials have been reported elsewhere [
19‐
22].
The AMAC (NCT02589665) trial was conducted at 85 study sites in 14 countries between January 2016 and September 2017. LUCENT 1 (NCT03518086, AMAN) was conducted at 383 study sites in 34 countries from 18 June, 2018 to 21 January, 2021. LUCENT 2 (NCT03524092, AMBG) was conducted at 367 study sites in 34 countries from 19 October, 2018 to 3 November, 2021. These three studies were approved by the applicable ethics review boards and performed in accordance with the Declaration of Helsinki, Council for International Organizations of Medical Sciences International Ethical Guidelines, and the International Council for Harmonisation Guideline for Good Clinical Practice. All patients provided written informed consent [
23].
Participating patients had a Mayo endoscopic subscore of ≥2 and a total Mayo score of 6–12 (AMAC) or a modified Mayo score 4–9 (LUCENT 1), and had not responded or were intolerant to conventional UC therapy (AMAC and LUCENT 1), biologic therapy with a target other than IL-23 (LUCENT 1), and/or Janus kinase inhibitors for UC treatment (LUCENT 1). Patients were excluded if they had previous exposure to any other biologic therapy targeting IL-23, with the LUCENT 1 trial also excluding patients who had not responded to three or more different biologic therapies. The LUCENT 2 trial accepted patients who had completed the LUCENT 1 trial.
In the AMAC trial (Fig. 1a of the Electronic Supplementary Material [ESM]), patients were randomized to placebo or mirikizumab 600-mg, 200-mg, or 50-mg induction dose regimens administered intravenously every 4 weeks (Q4W) for 12 weeks. In the mirikizumab 200-mg and 50-mg groups, doses after the first one were adjusted based on exposure. Serum concentrations of mirikizumab were assessed at weeks 2 and 6, and the dose was increased at weeks 4 and 8 according to a prespecified algorithm [
21]. The dose increase ranged from 2-fold to 12-fold for patients in the 50-mg arm and from 1.5-fold to 3-fold for patients in the 200-mg arm. The 600-mg arm remained at a fixed dose during the first 12 weeks. No patient was given a dose of > 600 mg in the induction period. The dose adjustments in the 200-mg and 50-mg groups resulted in an average group dose of 250 mg and 100 mg, respectively.
Responders to induction regimens were randomized to receive 200 mg subcutaneously either Q4W or every 12 weeks as a maintenance regimen through 92 weeks, terminating with a final 16-week follow-up period. Non-responders to induction regimens had the option to discontinue from the study or enter the open-label extension period, which consisted of a 12-week induction phase with a fixed intravenous (IV) dose of mirikizumab 600 mg or 1000 mg Q4W (three doses in total), and if they responded to the induction phase, a subsequent 92-week maintenance phase with subcutaneous (SC) mirikizumab 200 mg Q4W. Patients who did not achieve a clinical response at week 12 were discontinued from the study.In the LUCENT 1 trial (Fig. 1b of the ESM), patients were randomized (3:1) to mirikizumab 300 mg or IV placebo Q4W. Once they completed the 12-week induction phase, patients could enter the LUCENT 2 trial and receive treatment based on their response in LUCENT 1.
In the LUCENT 2 trial (Fig. 1b of the ESM), mirikizumab-treated responders from LUCENT 1 (a decrease in the modified Mayo score of ≥ 2 points and a ≥ 30% decrease from baseline, and a decrease of ≥ 1 point in the rectal bleeding subscore from baseline or a rectal bleeding subscore of 0 or 1) were randomized (2:1) to mirikizumab 200 mg or SC placebo Q4W, whereas placebo-treated responders continued to receive placebo Q4W, but with SC administration. Patients who maintained clinical response continued the same regimen until week 40, whereas patients who had a loss of response were given IV mirikizumab 300 mg Q4W for three doses followed by a final assessment. LUCENT 1 trial non-responders received IV mirikizumab 300 mg Q4W in the LUCENT 2 trial. After 12 weeks, if they responded, their dose was changed to SC mirikizumab 200 mg Q4W until week 40; if they did not respond, they discontinued the trial.
Intravenous infusion of mirikizumab or placebo occurred over at least 30 minutes. Subcutaneous administration of mirikizumab or placebo was given in up to four injections (AMAC) or in two injections (LUCENT 2), with a maximum volume of 1.5 mL per injection.
2.2 Sample Collection and Analyses
For the AMAC primary and extension studies, blood samples for the PK analysis were collected every second week during the induction phase, every 4 weeks in the first part of the maintenance phase, then every 8 weeks until the end of the study, including the follow-up period. A single sample was collected prior to study drug administration (if occurring on a dosing day), and samples were taken before each IV infusion (trough) and at the end of each IV infusion (maximum plasma drug concentration [Cmax]) at weeks 0, 4, and 8 (Table 1 of the ESM).
For the LUCENT trials, a PK sample collection was performed with a total of six samples per patient in LUCENT 1 and four to six samples per patient in LUCENT 2. Blood samples were collected prior to mirikizumab dosing at weeks 0, 4, and 8 and after mirikizumab dosing at weeks 0 and 4. Post-dosing samples were collected up to 2 h after dosing, with the actual time of all dose administrations recorded in the patient’s electronic case report form. Additional samples were collected at week 12, at any unscheduled visits, at early termination visits, and 16 weeks after the last visit (Table 1 of the ESM).
Serum mirikizumab concentrations were measured in blood samples using a validated drug-tolerant quantitative enzyme immunoassay method at ICON Laboratory Services, Inc. (Whitesboro, NY, USA). The lower limit of quantification was 100 ng/mL and the upper limit of quantification was 10,000 ng/mL. Samples above the limit of quantification were diluted to yield results within the calibrated range. The inter-assay accuracy (% relative error) during validation ranged from − 9.09% to 2.30%. The inter-assay precision (% coefficient of variation [CV%]) during validation ranged from 2.18% to 6.57%. Samples were stored at − 70 °C.
2.3 Model Development
A population PK model was initially developed using the data from the phase II AMAC trial. A two-compartment model with first-order absorption for the SC maintenance doses was found to best describe the pharmacokinetics of mirikizumab. The final population PK model was developed using the combined data from LUCENT 1+2 using the same model structure as the phase II study.
Model development consisted of structural model selection, a covariate search, and a final model evaluation (Fig. 2 of the ESM). Key selection criteria were convergence of the estimation and covariance routines, reasonable parameter and variance estimates based on the known pharmacokinetics of mirikizumab, good precision of the parameter and variance estimates (relative standard error < 50%), and graphical methods to ensure the presence of reasonable characterization of data with no obvious model misspecification. For the final model development, an assessment of covariate effects was conducted. Several covariates were investigated (Table 2 of the ESM), including body weight at study entry, age, sex, race, disease severity, prior biologic therapy, concomitant use of corticosteroids or immunomodulators (tested as single and combined covariates), and an antidrug antibody (ADA) titer over time. In particular, age was evaluated as a continuous covariate, and sex, race (White, Asian, and Other [American Indian or Alaska Native, Native Hawaiian or Other Pacific Islander, multiple, and missing]), ethnicity, and prior and concomitant therapies were evaluated as categorical covariates. Baseline disease state was evaluated using serum albumin (continuous), modified Mayo score (continuous/categorical), fecal calprotectin (continuous), and C-reactive protein (continuous). Key covariate selection criteria were as per structural model selection. In addition, the criterion for a statistically significant difference in objective function value (OFV) was a ≥ 6.635-point drop in OFV (p < 0.01) for forward inclusion and a ≥ 10.828-point drop in OFV (p < 0.001) for backward elimination. A bootstrap analysis was performed to assess the precision of the final parameter estimates of the final model using the bootstrap routine in Perl-speaks-NONMEM software (Version 4.8.1, ©2018–2019 by Mats Karlsson, Rikard Nordgren, Svetlana Freiberga, Sebastian Ueckert, and Gunnar Yngman), and a visual predictive check was performed to ensure that the model maintained fidelity with the observed data using the visual predictive check algorithm in Perl-speaks-NONMEM.
Missing values of independent variables (patient characteristic data) were imputed within patients using last observation carried forward. Approximately 4% and 2% of the PK samples collected for the AMAC and LUCENT 1+2 studies, respectively, were below the quantifiable limit. Models with and without these samples were compared using a first-order conditional estimation with interaction method and a Laplacian conditional estimation with interaction method when values below the quantifiable limit were included. There were no significant changes on the estimated PK parameters, and no increased uncertainty was noted with the additional samples, so the final analysis excluded these samples.
All analyses were conducted using non-linear mixed-effects modeling [NONMEM] (Version 7.4.2; ICON Development Systems, Gaithersburg, MD, USA), R (Version 4.0.3; R Foundation for Statistical Computing, Vienna, Austria), and Perl-speaks-NONMEM.
4 Discussion
The population PK models developed in this analysis were used to characterize the PK properties of mirikizumab in patients with UC and to evaluate the impact of demographic and disease characteristic covariates on mirikizumab pharmacokinetics during induction and maintenance therapy. Data from AMAC were not pooled with data from LUCENT 1+2 as the latter were sufficient to inform the model, and the approach adopted in this analysis allowed a comparison of model parameters between the phase II and III studies. Overall, the mirikizumab pharmacokinetics in patients with UC was typical of a monoclonal antibody drug [
24‐
26] and consistent with previous studies of mirikizumab in healthy volunteers (unpublished data). Clearance increased with a higher body weight and a lower serum albumin level, and bioavailability decreased with a higher BMI. However, changes in mirikizumab exposure related to changes in patient covariates were small relative to the overall variability in exposure and therefore are not considered clinically relevant.
The concentration–time data from the phase III studies were sufficient to inform the population PK model development; therefore, data from phase II and phase III studies were not combined. The population-typical CL and volume of distribution estimated by the PK model were similar between the AMAC and LUCENT 1+2 studies and to those of phase I studies of mirikizumab in patients with psoriasis and healthy volunteers (estimated CL: 0.022 L/h, NCT01947933; estimated volume of distribution: 5.74–6.91 L, NCT01947933; 4.35–5.22 L, NCT02568423; 3.98–4.43 L, NCT04137380). This indicates that the pharmacokinetics behaved consistently among different clinical trials, and there was no difference in the pharmacokinetics of mirikizumab between patient populations. In addition, the other PK parameters and inter-patient variability terms were similar between the AMAC and LUCENT 1+2 studies. Of note, the estimated bioavailability of mirikizumab resulting from the SC regimens in the LUCENT 1+2 study was slightly higher (median and geometric mean of 48% and 44%, respectively) than the bioavailability estimated in the AMAC study (median and geometric mean of 43% and 42%, respectively). This difference in bioavailability may be due to variability between the studies and to differences in the mirikizumab formulations and administration. In fact, a lyophilized formulation of mirikizumab was manually withdrawn from a vial into a syringe for administration in the AMAC trial, whereas a syringe prefilled with a mirikizumab solution formulation (planned for commercialization) was used in the LUCENT trials.
Overall, the changes in mirikizumab pharmacokinetics and exposures relative to patient factors were consistent with expectations for monoclonal antibodies. Body weight was a statistically significant covariate on CL and volume of distribution in the population PK analyses. As a result, there was an overall trend for serum concentrations to decrease as body weight increased. The effect of body weight on the pharmacokinetics of monoclonal antibodies is well established [
27,
28], and simulation studies have shown that dosing by body weight is not necessarily beneficial compared with flat dosing [
27,
29]. A large degree of overlap in exposures was noted when patients were stratified by body weight quartiles in the population PK analysis for mirikizumab. The result indicates that body weight does not have a clinically relevant impact on the exposure of mirikizumab, and no dose adjustment on the basis of body weight is needed.
Serum albumin level and CL of monoclonal antibodies have frequently been reported to be inversely correlated in PK studies of monoclonal antibodies [
30‐
33]. This relationship can be explained by metabolic pathways shared by albumin and monoclonal antibodies, such as the neonatal Fc receptor rescue pathway, which plays an important role in albumin and monoclonal antibody homeostasis [
34]. Lower CL with higher baseline albumin levels was observed for mirikizumab; however, the magnitude of effect was within the PK variability of mirikizumab, and therefore the effect was not considered clinically relevant.
Body mass index was identified as a statistically significant covariate affecting SC bioavailability of mirikizumab, with bioavailability decreasing as BMI increased. The bioavailability of monoclonal antibodies is thought to be influenced by several factors such as skin thickness, subcutaneous adipose tissue, and lymphatic circulation [
35]. As such, the relationship between mirikizumab bioavailability and BMI is likely the result of the relationship between BMI and these physiologic factors. Body weight and BMI are highly correlated. Accounting for this correlation and the effect of both covariates on mirikizumab pharmacokinetics, i.e., the effect of weight on CL and BMI on bioavailability, a large degree of overlap in exposures was noted when patients were stratified by body weight/BMI quartiles (Fig.
4e, f). The result indicates that BMI does not have a clinically relevant impact on the exposure of mirikizumab, and no dose adjustment on the basis of BMI is needed.
Immunogenicity was not identified as a significant factor affecting mirikizumab pharmacokinetics in either the AMAC or LUCENT 1+2 studies. However, the lack of statistical significance in the models may be the result of the low number of patients who developed a high ADA titer response. Figure
5 shows boxplots of the post hoc CL values for different ADA titer levels and does not show any significant change in CL with an ADA titer. As immunogenicity and the ADA titer were not covariates in the final model, the post hoc CL value for a given patient would only increase with higher ADA titers if the patient’s ADA titer was consistently high and serum mirikizumab concentrations were consistently low throughout the study. This way, the effect of high ADA/low serum mirikizumab concentration is significant enough to influence the random-effect eta estimate for CL in those particular patients. Overall, the changes in mirikizumab pharmacokinetics and exposures relative to patient factors are consistent with expectations for a monoclonal antibody. The magnitude of the changes in exposure related to patient factors also was small relative to the overall range of exposures caused by random unexplained variability. Furthermore, there was no apparent impact on mirikizumab pharmacokinetics of other demographic factors of interest, such as age, sex, and race in the population tested.
For the modeling, the allometric exponents were not fixed to values typically reported in the literature (i.e., 0.8 and 1 for CL and volume of distribution, respectively) as this resulted in a poorer model fit (increased OFV). This may be due to the typically reported values for allometric exponents only applying to changes in body weight with a consistent body composition, without accounting for changing in body composition (i.e., fat).
The LUCENT studies had sparse PK sample collection schemes, and most PK samples were collected prior to dosing when mirikizumab concentrations are expected to be at trough. Although this is typical of phase III trials, it may not be sufficient to fully characterize the pharmacokinetics of the compound. Nevertheless, the PK parameters found in this analysis were similar to those in studies with denser sample collection schemes, such as the phase II AMAC study.