How the body processes efpeglenatide in people with obesity and type 2 diabetes

The population PK analysis was performed based on pooled data from one phase 1 and five phase 2 studies of efpeglenatide. The phase 1 study (HM-EXC-102) was an exploratory, randomized, partially single-blind trial. The Phase 2 studies (HM-EXC-201, -202, −203, −204, and −205) were randomized, double-blind, placebo-controlled or open-label designs. Liraglutide comparator groups in HM-EXC-102 and HM-EXC-203 were excluded from the analysis. Dosing regimens ranged from once weekly (QW) to every other week (Q2W) and once monthly (QM) subcutaneous administration. Key study characteristics are presented in Table 1. The key demographic and clinical characteristics of the study populations are summarized in Table 2.

All clinical studies were conducted in accordance with the Declaration of Helsinki, the International Council for Harmonisation (ICH) E6 (R2) Good Clinical Practice guidelines, and all applicable local regulatory requirements. Study protocols were reviewed and approved by the Institutional Review Board or Independent Ethics Committee at each participating center, and written informed consent was obtained from all participants prior to enrollment.

The pooled dataset comprised six clinical studies (five studies in patients with T2DM and one study in obesity). A total of 3,596 plasma efpeglenatide concentration measurement were collected from 498 subjects across six clinical studies at prespecified time points.

All samples were analyzed at certified laboratories using validated analytical method. Serum concentrations of efpeglenatide were quantified using a validated Enzyme-Linked Immunosorbent Assay (ELISA) method.

For the PK analysis, 3,316 post-dose plasma concentrations above the lower limit of quantification (LLOQ) were used in the model estimation. All pre-dose concentrations before the first administration were excluded from the dataset. Post-dose samples below the limit of quantification (BLQ) were treated as missing (i.e., not included in the analysis). In total, 117 pre-dose samples (3.25%) and 153 post-dose samples (4.0%) were BLQ.

The base PK model evaluation was begun with a one-compartment disposition model with first-order elimination, with additional refinement to better characterize the absorption phase. To describe the double peaks observed during the absorption phase, an empirical absorption model incorporating two divided subcutaneous administrations was implemented. The first one of the two divided doses was described as the first order absorption after bolus doses into depot compartment, capturing the first peak during the absorption phase. The second part of the dose was introduced to describe the second peak with time delay during absorption phase, which was described using a transit compartment model (Savic et al., 2007). This dual dose approach successfully described the double-peak observed in the PK data, and significantly improving model fit. After the absorption model was established, various compartment models were evaluated. Ultimately, a two-compartment disposition model with first-order elimination was selected as the base PK model (Figure 1).

The transit compartment model describes the delay in absorption as the sequential passage of the drug through a chain of transit compartments, each connected by a first-order transit rate constant (K_tr). The number of transit compartments (n) and the mean transit time (MTT) were estimated from the data, with K_tr calculated as (n+1)/MTT. Drug exiting the final transit compartment entered the absorption compartment and was transferred to the central compartment via first-order absorption, as described by Equations 1–3. (Savic et al., 2007).dandt=Ktr·an−1−Ktr·an(1)ant=Dose·F·Ktr·tnn!·e−Ktr·t(2)n!≈2π·nn+0.5·e−n(3)

The amount of drug administered subcutaneously (A_sc) in the depot, central compartment (A_c) and peripheral compartment (A_p) were described using the following differential Equations 4–6:dAscdt=Dose·BIOA1+Dose·BIOA2·Ktr·tn·e−Ktr·t2π·nn+0.5·e−n−ka·Asc(4)dAcdt=ka·Asc−AcVc·CL−AcVc·Q+ApVp·Q(5)dApdt=AcVc·Q−ApVp·Q(6)

To confine the estimated fraction values to the range between 0 and 1 and to allow implementation of interindividual variability, a logit transformation was applied. BIO_A1 and BIO_A2 represent the fractions of the administered dose entering the depot compartment via the bolus administration route and the delayed absorption route, respectively. The transformation was defined as shown in Equation 7 (Adedokun et al., 2020):BIOA1=eBIOF1+eBIOF,BIOA2=11+eBIOF(7)where BIOF is an unbounded parameter (−∞∼∞) representing the logit-transformed value of BIOA1, and BIO_A2 = 1-BIO_A1

The population PK modeling analysis was performed using NONMEM version 7.5 (ICON Development Solutions, Hanover, MD, United States). The first-order conditional estimation method with interaction (FOCEI) was used to approximate the marginal likelihood and facilitate estimation of population model parameters. Pre- and post-processing of the data and analysis of modeling results were performed using R software version 4.3.1 or other (R Core Team, 2023).

Interindividual variability (IIV) in most PK parameters was modeled using an exponential distribution (Equation 8). For the ith individual, η_i, is a random variable assumed to be independently selected from normal distribution with a mean of zero and variance of ω².Pi=PTV×exp⁡ηi(8), where PTV is the typical value (population value) of the fixed effect parameter. The approximate percent coefficient of variation (%CV) was calculated as shown in Equation 9:%CVIIV=expω2−1×100(9)

Residual variability (RV), representing within-subject unexplained variability and contributions from assay error, sampling deviations, and model misspecification, was evaluated using additive, proportional, and combined residual error models. Model selection was guided by the likelihood ratio test (LRT) and inspection of goodness-of-fit (GOF) plots, and the proportional error model was selected as the final residual error structure (Equation 10). The RV for the jth observed value in the ith individual and ε_prop,ij; represents the proportional component, assumed to be independent and normally distributed, with a mean of zero and a variance of σ².Yij=IPREDij×1+εprop,ij(10)where y_ij: jth observed value in the ith individual; IPRED_ij: jth model predicted value for the ith individual; ε_prop,ij is the proportional error.

Model selection and evaluation were performed using both statistical and graphical approaches. For statistical evaluation, the LRT was used to compare hierarchical models, with a p-value <0.05 (corresponding to a decrease in objective function value [OFV] of 3.84 points) considered statistically significant, assuming that the difference in −2 log-likelihood (-2LL) follows an approximate chi-square distribution (Karlsson et al., 1998). In addition, the Wald test was used to assess parameter significance and precision: 95% confidence intervals (CIs) for each parameter estimate were constructed using the point estimate and standard error derived from the model outcomes, and parameters were considered statistically significant if the 95% CI excluded zero (Hooker et al., 2007).

Graphical evaluation included the inspection of standard GOF plots and visual predictive checks (VPCs) to compare observed versus model-predicted values, with predictive performance assessed in terms of both central tendency and variability (Savic et al., 2007). VPCs were performed by generating 1000 replicates using Monte Carlo simulations in NONMEM, and the resulting simulated prediction intervals were overlaid on the original PK data in plots generated using R. In addition, prediction-corrected VPCs (pcVPCs) stratified by key covariates (body weight and disease status) were performed to assess model performance across subgroups (Karlsson and Savic, 2007; Beal et al., 2009).

The stability of parameter estimates for all fitted models was assessed by examining pairwise correlations (with ρ > 0.95 indicating potential collinearity) in the correlation matrix and evaluating the condition number (ratio of the largest to smallest eigenvalues), with values below 1000 considered acceptable (Jonsson and Karlsson, 1998).

Covariate selection was performed using the stepwise covariate modeling (SCM) approach guided by the LRT. In the forward inclusion step, each covariate–parameter relationship was tested individually, and the covariate was retained if the decrease in the OFV (ΔOFV) was ≥7.87 (p < 0.005, χ² distribution with 1 degree of freedom). In the subsequent backward elimination step, starting from the full model, covariates were sequentially removed if the increase in OFV upon removal was <10.8 (p < 0.001). The covariates evaluated included body weight, age, sex, race (Caucasian, Black, Asian, Native Hawaiian or Pacific Islander, and Others), body mass index (BMI), lean body weight (LBW), and obesity status. All covariates were tested for potential influence on all PK parameters (Jonsson and Karlsson, 1998).

Continuous covariates were incorporated using a power function normalized to the reference (median) value, as shown in:whereand θ are fixed-effect parameters andis a reference value of covariate. For this analysis, the approximate median of the population will be used for. Equation 11 T V P i T V P R E F = C O V i C O V R E F θ (11) TVP REF COV REF COV i COV REF

Categorical covariates were modeled using an exponential/indicator function, as shown in:whereandare fixed-effect parameters without and with the covariate, respectively.is an indicator variable that takes a value of 0 or 1 depending on whether the subject has the covariate or not. Equation 12 TVP i TVP REF TVP C a t Cat i = · + · 1 − Cat i (12) TVP REF TVP C a t Cat i

After final model selection, the numerical stability of parameter estimates (e.g., standard errors [SE] and relative standard errors [RSE]) was evaluated, and covariates associated with unstable estimates were excluded.

Inter-occasional variability (IOV) in PK was assessed by introducing a random effect parameter to selected fixed effect PK parameters, allowing different random value to assigned for each dosing occasion within the same individual.

Monte Carlo simulations of plasma efpeglenatide concentration–time profiles were performed using NONMEM using the final population PK model. Each scenario consisted of 1,000 virtual subjects receiving the specified dosing regimen, and was designed to evaluate steady-state exposure and the impact of stepwise dose escalation on PK profiles. IIV and RV were incorporated to reflect population variability. Covariates were fixed based on predefined subgroup scenarios rather than sampled from distributions.

The PK of efpeglenatide was well-described by a two-compartment linear model with dual absorption pathways, which characterized the pooled data from six completed clinical studies (Table 3). IIV was estimated for all parameters except Q/F and N, with no statistically significant covariance identified between IIV terms. IOV was evaluated but not found to be significant and was therefore not included in the final model.

All fixed-effect PK parameters and random effect parameters for IIV were estimated with acceptable precision, with relative standard errors (RSEs) < 30% for all fixed effects. Although the IIV of Vc/F (CV: 68.31%, RSE: 18.28%), BIOA1 (CV: 66.93%, RSE: 14.6%), and MTT (CV: 46.27%, RSE: 35.88%) was relatively large, the corresponding RSE values suggested that these estimates were supported by the data. There was no sign of over-parameterization.

Covariate analysis revealed that body weight influenced both k_a and CL/F through a power model, with estimated exponents of −0.927 and 0.964, respectively. The disease status (T2DM vs. obese) also appeared to influence CL/F, with estimated values of 0.044 and 0.032, respectively.

Goodness-of-fit (GOF) plots for the PK model showed good agreement between observed and predicted plasma concentration (Figure 2A). Both population predicted and individual predictions versus observed concentrations showed that the data were randomly distributed around the line of identity, indicating no significant bias (Figure 2B). Residual diagnostics further supported model adequacy, as conditional weighted residuals were randomly distributed around zero across both the time and predicted concentration axes, no apparent trends or patterns observed (Figures 2C–E).

The pcVPC for the final PK model showed good agreement between observed and predicted efpeglenatide concentration across dosing regimen. Approximately 95% of the observed values fell within the 95% prediction intervals, indicating that the model described an adequate description of the data (Figure 3).

Model-predicted steady-state AUCs of efpeglenatide appeared to increase in a dose-proportional manner across the 2–18 mg once-weekly dosing range (Figure 4A). In subgroups analyses, obese subjects exhibited approximately 21% higher steady-state AUC compared to subjects with T2DM (Figure 4B).

In the simulated concentration–time profiles of the stepwise dose-escalation regimen (2 mg–12 mg once weekly in 2 mg increments every 4 weeks), steady-state concentrations were generally achieved within 4 weeks at each dose level, and exposures increased progressively during the titration period (Figure 5).

Compared with the obese median weight individual (96 kg), the 10th percentile individual (79 kg) exhibited approximately 18% lower AUC and C_max (GMR 0.82; 90% CI: 0.81–0.83), whereas the 90th percentile individual (120 kg) demonstrated about 20% higher exposure (GMR 1.22; 90% CI: 1.20–1.24). These results are summarized in Figure 6. Across most simulated subgroups, exposure differences remained within the commonly referenced bioequivalence range (0.80–1.25). Similarly, simulated exposures across age, sex, and race subgroups showed minimal variation from the reference, all of which were also within this range.