The final PopPK dataset comprised 116 participants (82 healthy volunteers, 24 patients with PNH, and 10 patients with CHAPLE disease) contributing a total of 2795 concentration samples (total pozelimab n = 1640; total C5 n = 1155) (Supplementary Table S2). Of the 2795 concentration samples, 2695 (96.4%) were quantifiable; 100 (3.6%) post-dose samples were BLQ and were thus excluded from the PopPK analysis given the low proportion (< 5%).

Demographic characteristics and baseline values for relevant categorical and continuous covariates in the PopPK model are presented in Table 1.

For adult participants (excluding patients with CHAPLE disease), median age was 37.5 years (range: 19–76), median baseline body weight was 68 kg (range: 45–108), median baseline body mass index (BMI) was 24 kg/m2 (range: 17.6–34.1), and median baseline albumin level was 44 g/L (range: 37–51).

For patients with CHAPLE disease, the median age was 8.5 years (range: 3–19), median baseline body weight was 25.0 kg (range: 11.0–53.8), median baseline BMI was 15.5 kg/m2 (range: 12.2–24.5), and median baseline albumin level was 23.0 g/L (range:11.0–29.0 g/L).

Baseline C5 levels were comparable between the predominantly pediatric patients with CHAPLE disease, and adult patients with PNH or healthy adult volunteers (Supplementary Figure S2). Moreover, there was no trend observed between baseline C5 levels and baseline covariates of age, body weight, and albumin levels across the analysis population (Supplementary Figure S3).

The PopPK dataset used to develop the initial PopPK model was pooled from three clinical studies comprised of healthy adult volunteers or patients with PNH. Using this PopPK dataset, the PopPK model developed was a two-compartment TMDD model with two binding sites based on the QE approximation [16] and defined by the following parameters: CL, Vc, absorption rate constant (ka), intercompartmental clearance (Q), Vp, bioavailability, synthesis rate of free C5 (ksyn), degradation rate constant of free C5 (kdeg), equilibrium dissociation constant (kD), internalization rate constant for the pozelimab-C5 complexes (kint1), and an internalization rate constant for the pozelimab-C5-C5 complexes (kint2). The kD value was fixed to 0.189 nM (0.03591 mg/L), which was obtained from SPR-Biacore technology. IIV random effects were included on CL, Vc, ka and ksyn. Different error models were evaluated for total pozelimab and total C5. An additive error model with log-transformed data best described the total pozelimab data, while a proportional error model was used for the total C5 data. A correction factor was added to baseline free C5 as a linear function described by the equation below, to provide model flexibility and account for the initial drop in total C5 concentrations between baseline (time 0) and the sampling time at 1 h after drug administration. Inclusion of the correction factor resulted in a decrease in OFV of ~ 113 units, which was statistically significant. The impact of baseline body weight on CL and Vc was included in the base model.

R0 is the baseline free C5 and θ13 is the correction factor.

In addition to baseline body weight, which was included as a covariate in the base structural model, PNH patient status on Vc was the only statistically significant covariate identified following covariate analysis.

After the TMDD PopPK model was updated to include data from the phase 2/3 clinical study of pozelimab of patients with CHAPLE disease, model refinements were undertaken. These comprised of inclusion of the impact of baseline body weight on Vp, the inclusion of IIV on Vp and kint1, and the removal of IIV on ka, which led to improvements in model fitting, especially for patients with CHAPLE disease. The effect of baseline body weight on Q was evaluated, however, this resulted in unstable model performance and no visible improvement in model fitting. As a result, the TMDD model with the effect of IIV estimated on CL, Vc, Vp, ksyn, and kint1, with residual error separated by adults and children with CHAPLE disease for both pozelimab and total C5, baseline body weight effect on CL, Vc, and Vp, and PNH patient status on Vc, was declared the final TMDD PopPK model for pozelimab.

The PK parameter estimates for the final PopPK model are presented in Table 2. Structural PK parameters were well estimated in the final PopPK model, with percent relative standard error values ≤ 17% for structural parameters. The model was stable with a condition number of 138.3, well below the reference threshold value of 1000. The empirical Bayes estimate for IIV shrinkage was high for linear CL (53.3%) and kint1 (38.5%), and moderate for Vc (19.9%), ksyn (21.7%), and Vp (24.0%). Inspection of diagnostic plots demonstrated alignment between observed, individual predicted, and population predicted concentrations of total pozelimab and total C5 in all participants (Supplementary Figures S4 and S5). Similarly, inspection of diagnostic plots demonstrated alignment between observed data and model-predicted data for total pozelimab and total C5 in patients with CHAPLE disease (Supplementary Figures S6 and S7).

The covariate effect of time-varying body weight was evaluated using the final TMDD PopPK model by replacing baseline body weight with time-varying body weight, due to rapid weight gain in patients with CHAPLE disease following pozelimab treatment. Time-varying body weight was not available in healthy volunteers, therefore this was imputed using baseline body weight for this group. A comparison of PK parameter estimates for the models using time-varying and baseline body weight is presented in Supplementary Table S3. Using time-varying body weight resulted in similar (< 5% difference) PK parameter estimates and OFV, except for the exponent relating body weight to CL, which increased from 0.9989 to 1.108 with time-varying body weight.

A sensitivity analysis of exponents relating time-varying body weight to CL, Vc, and Vp was conducted by fixing body weight exponents to classical allometric values (1 for Vc or Vp and 0.75 for CL). Fixing body weight exponents to allometric values resulted in an increase in OFV by 11.2 points (Supplementary Table S4) and diagnostic plots that indicated poorer fitting in patients with CHAPLE disease, compared to the model with estimated exponents (Supplementary Figures S8 and S9).

Internal VPCs stratified by dose showed that most of the observed concentrations of pozelimab were within the range of the 5th to 95th predicted percentiles (Fig. 2). Similarly, the internal VPC for total C5 showed that the observed concentrations of total C5 were generally contained within the range of the 5th to 95th predicted percentiles (Fig. 3). The VPCs indicate that the final PopPK model allowed for reliable predictions for individual exposures.

Visual predictive check of the final TMDD PopPK model for total pozelimab stratified by dose. Red solid line: simulated 50th percentile; black solid line: observed 50th percentile; blue dashed lines: simulated 5th and 95th percentiles; symbols: observed data; shaded area: 90% simulation-based confidence interval around median (pink) and upper/lower percentiles (blue). PopPK, population pharmacokinetic; IV, intravenous; QW, once weekly; SC, subcutaneous; TMDD, target-mediated drug disposition

Visual predictive check of the final TMDD PopPK model for total C5. Red solid line: simulated 50th percentile; black solid line: observed 50th percentile; blue dashed lines: simulated 5th and 95th percentiles; symbols: observed data; shaded area: 90% simulation-based confidence interval around median (pink) and upper/lower percentiles (blue). C5, complement component 5; PopPK, population pharmacokinetic; TMDD, target-mediated drug disposition

Simulated concentration–time profiles for total pozelimab, free pozelimab (i.e., pozelimab not bound to any target), and free C5 following a single IV dose of 1, 3, 10 and 30 mg/kg of pozelimab are shown in Fig. 4, which demonstrated free pozelimab marked nonlinearity at lower concentrations. Following a single IV dose of 30 mg/kg pozelimab, free C5 was predicted to be suppressed for approximately 12 weeks before returning to baseline.

Simulated median concentration–time profiles of total pozelimab (green dashed lines), free pozelimab (blue dashed lines), and free C5 (solid red lines) following a single IV dose of 1, 3, 10, and 30 mg/kg. C5, complement component 5; IV, intravenous

Simulated exposure metrics for total pozelimab based on the final TMDD PopPK model following 10 mg/kg weight-based dose or weight-tiered doses are shown in Fig. 5 and Supplementary Table S5. Similar simulation results based on the PopPK model with time-varying body weight effect are illustrated in Supplementary Figure S10. Simulation results based on either the final TMDD PopPK model or the model with time-varying body weight effect suggest that pozelimab PK exposure metrics at steady-state were generally comparable for patients with CHAPLE disease weighing > 20 kg following either a 10 mg/kg weight-based dose or weight-tiered doses. In addition, similar pozelimab PK exposures are expected across the different weight groups.

Simulated total pozelimab Cmin, ss and AUCtau, ss for patients with CHAPLE disease following a 10 mg/kg or weight-tiered SC QW maintenance dose. Weight-based dose: 10 mg/kg, not exceeding 800 mg SC QW; weight-tiered dose: 125 mg for < 10 kg, 200 mg for 10–<20 kg, 350 mg for 20–<40 kg, 500 mg for 40–<60 kg, and 800 mg for ≥ 60 kg. AUCtau, ss, steady-state area under the concentration time curve for a dosing interval; CHAPLE, complement hyperactivation, angiopathic, thrombosis, protein-losing enteropathy; Cmin, ss, trough concentration at steady-state; QW, weekly; SC, subcutaneous

Simulations for free C5 following the 10 mg/kg weight-based dose or weight-tiered doses are shown in Supplementary Figure S11. Free C5 concentrations are predicted to be extremely low (< 0.05 mg/L), indicating complete inhibition of complement activity.