Patients were enrolled at the First Affiliated Hospital of Zhengzhou University during August 2019 and June 2021, and all provided written informed consent. This study was approved by the institutional research ethics committee of First Affiliated Hospital of Zhengzhou University (No. SS-2019–058).

Adult (aged over 18 years) renal transplant patients receiving MMF (MMF dispersible tablets, Hangzhou Zhongmei Huadong Pharmaceutical Co. Ltd., Hangzhou, China) were enrolled. Patients who monitored MPA concentration within 1 week after renal transplantation were assigned to the early post-transplant stage group, and those who monitored MPA concentration during at least 5 years after renal transplantation were assigned to the stable state group. Exclusion criteria were that patients 1) had liver dysfunction before surgery, 2) received a combined organ transplantation, 3) had organ rejection or gastrointestinal disease, and 4) received co-medications that potentially affect MPA PK parameters, such as cholestyramine and rifampicin, according to the drug label. Clinical data, including age, gender, weight, post-transplantation time, medication, tacrolimus concentration, and laboratory results were collected from electronic medical records.

Patients were received triple immunosuppressive therapy comprising MMF, tacrolimus, and corticosteroids. MMF was administered orally at a dose of 1.0–3.0 g/d (twice daily, bid) in the early stage and 0.5–1.5 g/d in the stable stage under fasting conditions. Triple immunosuppressive dosages for individual depended on the medical teams.

Blood samples were obtained at least 3 days after the same dose. For each patient, one blood sample (2 ml, C0h) was collected immediately at pre-dose, eight to ten blood samples (mainly C0.5h, C1h, C1.5h, C2h, C3h, C4h, C6h, C8h, and C12h) were collected at pre-next dose into EDTA tubes. All samples were centrifuged at 3,500×g for 10 min. The supernatant was collected and stored at −80°C until analysis.

MPA plasma concentration was determined by ultra-performance liquid chromatography-diode array detection method as previously described (Zhang et al., 2020). Briefly, 100 μL of plasma sample was mixed with 10 μL of internal standard solution and 200 μL of acetonitrile. After a thorough vortex for 1 min, the mixture was centrifuged at 13,000 rpm for 10 min. Finally, 100 μL of supernatant was transferred to an autosampler vial and 10 μL was injected for quantification. The calibration curves showed an acceptable linearity from 0.04 to 40.0 mg/L, with a lower limit of detection of 0.06 mg/L.

The PK parameters of MPA were analyzed by non-compartmental analysis (NCA) using Phoenix^® WinNonlin software (v8.3, Pharsight, Mountain View, CA, United States). Peak concentration (C_max) and trough concentration (C_min) were obtained directly from the raw data. The area under the concentration at steady state for 12 h (AUC_ss,12h) was calculated by the linear trapezoidal linear interpolation method.

The population PK parameters were estimated by the first-order conditional estimation method (FOCE ELS) using Phoenix NLME^® software. For the basic modeling, the data were fitted to one or two compartment disposition models with first order absorption kinetics and with or without absorption lag-time (Tlag). The EHC process was also estimated. Parameters for the model included central clearance (CL/F), central distribution volume (V/F), intercompartmental clearance (Q/F), peripheral distribution volume (V2/F), and absorption rate constant (ka). The initial values for these parameters were obtained from NCA results. PK model was selected based on the precision of parameter estimates (standard error), goodness-of-fit plots, and likelihood ratio test (−2LL). Subsequently, inter-individual variability (residual error) was modelled with additive (C_obs = C_pred + ε), proportional [C_obs = C_pred × (1+ ε)] or mixed (additive + proportional) error models, where C_obs and C_pred were the observed and predicted concentrations, and ε was the error variable with a mean of 0 and variance of σ ².

For covariate modeling, candidate covariates, including age, sex, body weight, alanine aminotransferase (ALT), aspartate aminotransferase (AST), total bilirubin (TBIL), white blood cells (WBC), erythrocyte, haemoglobin, platelet, serum proteins, albumin, serum creatinine (Scr), creatinine clearance (CrCL), glomerular filtration rate (GFR), urea, and post-transplant stages (the early stage as 0 and the stable stage as 1), evaluated using a stepwise process. Continuous covariates were normalized by median values (of observed values) and categorical covariates were reflected as index variables in the model. By comparison with the initial model, the inclusion criteria for covariates were a drop >6.63 (p = 0.01) of objective function value (OFV; −2LL) for forward inclusion and an increase of OFV >10.83 (p = 0.001) for backward elimination.

Finally, goodness-of-fit plots were used to evaluate the adequacy of the final model. Model evaluation was carried out using prediction-corrected visual predictive check (pc-VPC) and bootstrap analyses. For pc-VPC, the prediction-corrected observations were plotted against the 5th, 50th, and 95th percentiles of the time-simulated MPA concentrations (n = 1,000). A nonparametric bootstrap procedure (n = 1,000) was conducted to compare the parameters with the final model parameters estimated from the original data set.

Based on the final population PK model, six fixed dosing regimens, including 0.25 g bid, 0.50 g bid, 0.75 g bid, 1.0 g bid, 1.25 g bid, and 1.50 g bid in the early and stable stages, were simulated by stochastic simulations (n = 1,000) to predict concentrations and AUC_ss,12h on day 4. The exposure threshold (AUC_ss,12h) of MPA was set at 30–60 mg h/L (van Gelder et al., 2006; Jeong and Kaplan 2007; Tett et al., 2011). The dosage regimen with the higher percentage of therapeutic level achieved was selected.

LSSs adopted 3 to 4 timed samples to calculate AUC by both Bayesian and multiple regression analysis. The Bayesian approach was performed on Phoenix^® NLME software. Predicted MPA concentrations were estimated from limited sample points based on the final population PK model and then were used to calculate predicted AUC by NCA. Pearson correlation coefficient (r) was used to compare the observed and predicted AUCs. The linear regression analysis was performed on IBM SPSS Statistics (v26.0, SPSS Inc., Chicago, United States). The stepwise forward method was used to deduce the AUC formula based on MPA concentrations at different time points. And then MPA concentrations at different time points were brought into the formula to calculate the predicted AUC. The determination coefficient (r) was used to assess the regression level of the formula.

The predicted AUCs of the two approaches were compared with the observed AUCs. Prediction performance was assessed in terms of accuracy and precision, expressed as prediction error (PE, Eq. 1) and root means square error (RMSE, Eq. 2). PE ( % ) =   ( AUC pred - AUC obs ) × 100 / AUC obs (1) RMSE = 1 / n ∑ ( PE % ) 2 (2)

The study recruited 51 patients (561 plasma samples) in the early post-transplant stage and 48 patients (518 plasma samples) in the stable state. Blood samples of the two groups were collected on day 4–8 or at least 5 years after renal transplantation, respectively. Table 1 summarized the demographic and clinical information of the patients. There was a significant difference in renal function, concomitant medication dosage, and haemoglobin between the two groups (p < 0.001). Patients in early post-transplant stage were on concomitant pantoprazole.

Plasma concentration time profiles and PK parameters were shown in Figure 1 and Table 2, respectively. PK parameters were significantly different between the two groups (p = 0.05). Preliminary MPA concentration showed large individual variability, ranging from below the quantification limit (0.1 mg/L) to 34.0 mg/L. Delayed absorption at 2–4 h post-dosing and EHC at 6–12 h post-dosing were observed in some patients. The AUC reached the target exposure in 51 patients (22 in the early stage and 29 in the stable stage). After adjusted AUC with dosage (AUC/dose), the mean normalized AUC_ss,12h of MPA (40.83 ± 22.26 mg h/L) in the early post-transplant period group was significantly lower than that (77.86 ± 21.34 mg h/L; p < 0.001) in the stable post-transplant period group (Figure 2).

The raw concentration data were adequately described by a two-compartmental model with first order absorption with lag-time and linear elimination. Inter-individual variability was described by the proportional error model. Additionally, the inclusion of the EHC process did not significantly decrease OFV value (ΔOFV = 7.58; Supplementary Table S1 and Supplementary Figure S1). In the covariate analysis, post-transplant stages were the covariate for V (ΔOFV = 70.72, p < 0.001) and CL (ΔOFV = 57.41, p < 0.001). Other covariates, including age, gender, body weight, ALT, AST, WBC, haemoglobin, serum proteins, albumin, TBIL, Scr, CrCL, GFR, and urea, showed no significant effect on PK parameters (Supplementary Figure S2). Finally, the final population PK model parameters were shown in Table 3. Typical PK parameters for patients in the early post-transplant stage, including CL/F (23.36 L/h vs. 10.25 L/h) and V/F (78.07 vs. 16.24 L), were considerably different from the values for patients in the stable state.

In the goodness-of-fit plots (Figure 3), the plots were normally distributed with no obvious systematic bias. In the pc-VPC plots (Figure 4), most of observed data were within the 5th-95th prediction percentile. Consistently, the bootstrap results closely matched the mean estimates from the population PK models, confirming the stability of the model.

Figure 5 presented the AUC_ss,12h for various MPA dosages. In the early post-transplant stage, the probability of achieving target exposure (30–60 mg h/L) was 0.1, 9.0, 41.2, 59.4, 62.0, and 51.2% for six fixed regimens (0.25, 0.50, 0.75, 1.0, 1.25, and 1.50 g bid), respectively. As for the stable post-transplant period, the likelihood of achieving efficacious exposure for six fixed regimens (0.25, 0.50, 0.75, 1.0, 1.25, and 1.50 g bid) were 7.3, 60.1, 53.3, 31.7, 15.0, and 7.5%, respectively.

Based on the literature (Filler 2004; Kiang and Ensom 2016), seven 3 or 4-point samples models were selected, while the whole-point model was provided as a reference. The predictive performance of Bayesian and multiple regression analyses was presented in Table 4, and the regression equations were shown in Supplementary Table S2. For both analyses, the predictive ability of 4-time point schemes was slightly higher than that of 3-time point schemes, but the improvement of models was limited. Two strategies (C0-C1-C2-C4 and C1-C2-C3-C6) showed good correlations with the observed AUC in the early staged and stable stage, respectively.