**4. Discussion**

PBPK models for tramadol and M1 were developed. Tramadol plasma concentration–time profiles were well predicted from the proposed model. Prediction results for M1 included values in the 5th to 95th percentiles of most observed plasma concentration–time values, and the predicted mean plasma concentration was also similar to the observed concentration–time profile. However, geometric mean Cmax,ss and AUClast,ss ratios were under-predicted (0.63 and 0.67 for Cmax,ss and AUClast,ss, respectively). To predict concentration-dependent toxicities, the therapeutic range (100–800 ng/mL for tramadol, and up to 200 ng/mL for M1) and the tramadol toxic range and lethal concentration (>1000 ng/mL, and >2000 ng/mL, respectively) were obtained from the literatures [34,35]. In general, the recommended dose of tramadol is up to 400 mg per day for immediate-release formulations and 300 mg per day for ER formulations [2]. Simulations were performed for 100 and 200 mg with 12-h intervals (5 times) according to CYP2D6 genotypes. After administration of 100 mg of tramadol, the predicted Cmax,ss of tramadol reached to toxic range in CYP2D6 PMs and exceeded therapeutic range in some IMs, and the predicted Cmax,ss of M1 exceeded therapeutic margin in CYP2D6 UMs. After tramadol 200 mg administrations, the predicted tramadol Cmax,ss reached to toxic ranges in all CYP2D6 metabolizer groups, even in some EMs and the predicted M1 Cmax,ss exceeded the therapeutic margin in CYP2D6 IMs, EMs, and UMs. The concentrations exceeded the therapeutic margins or reached to the toxic range might be related to potential toxicities after tramadol administrations, even though recommended doses of tramadol were administered.

PBPK modeling is useful for predicting PK profiles for rare genotypes in the population. The frequency of CYP2D6 UM in the Korean population has been reported as approximately 1.25% [36]. In the clinical study used for our PBPK model development, there was only one PM subject, and no UM subject was found. The model developed in this study could predict the plasma concentration–time

profiles of tramadol and M1 for these two groups. Using the developed model, plasma tramadol/M1 concentration–time profiles for CYP2D6 UM, a very rare genotype in Koreans, were also predicted.

Tramadol inhibits reuptake of 5-HT and norepinephrine. M1 binds to μ-opioid receptors and exhibits analgesic e ffects. Due to these actions, the side e ffects of tramadol di ffer depending on CYP2D6 genotype. In the PM group, a high risk of side e ffects due to tramadol, such as serotonin syndrome, can be expected; and in the UM group, a high risk of μ-opioid receptor-related side e ffects, such as respiratory depression, can be expected relative to other CYP2D6 genotypes [16]. In our simulation, the plasma concentrations of tramadol and M1 exceeded the therapeutic concentration range, even after administration of recommended doses. These results sugges<sup>t</sup> that the frequency of concentration-related adverse drug reactions may be reduced by optimizing the dosing regimen according to CYP2D6 genotype of the patient or population.

Tramadol and M1 distribution in each tissue were estimated using the PBPK model, and tramadol and M1 were distributed most to the liver. In cases of fatal intoxication due to tramadol, the highest concentration of tramadol was evident in the liver, after the blood and urine. These distribution characteristics are considered due to the hepatic metabolism of tramadol and its metabolites [37]. The distribution of tramadol to adipose tissue di ffered from that for M1. Indeed, tramadol is considered to distribute widely to lipid-rich tissues because of its higher a ffinity for lipids than M1 (logD for tramadol and M1: 1.13 and 0.4, respectively) [38]. Further research is needed about the distribution characteristics of tramadol and M1 to each tissue.

The predicted plasma M1 concentration–time profiles were under-predicted due to a lack of information about distribution and elimination properties. Since M1 is produced by tramadol metabolism, elimination profiles (intrinsic clearance by CYP, renal clearance and additional clearance) of tramadol were adjusted to improve the M1 model; however, there were no significant changes in M1 concentration–time profiles. This might be due to poor distribution of M1 from liver to plasma, or to exaggeration of elimination. For improvement, the M1 model was built using parameter estimation by observed plasma concentration–time profiles as distribution and elimination profiles (tissue–plasma partition coe fficient, additional clearance, renal clearance, bile clearance). When estimating several parameters, the predicted plasma M1 concentration–time profiles changed significantly when values of the unbound fraction in incubated microsomes (fumic) and active hepatic scalar were changed. Thus, the plasma M1 concentration–time profile might be greatly influenced by metabolism. More detailed information and parameters for M1 metabolism are needed for more accurate predictions of plasma M1 concentration–time profiles.

Regarding limitations of our study, tramadol is metabolized not only to M1, but also to *N*-desmethyltramadol (M2) by CYP2D6, CYP2B6, and CYP3A4. In accordance with the literature, the toxicity of tramadol and M1 can be determined using M1/M2 ratio [34]. Therefore, an M2 model could improve the predictability of concentration-related adverse drug reactions after tramadol administration. Moreover, organic cation transporter 1 (OCT1) and multidrug resistance protein 1 (MDR1) influence the disposition of tramadol and M1. Significant di fferences in drug disposition according to OCT1 and MDR1 genotypes have been shown, even in same CYP2D6 phenotype [39–41]. Due to lack of information of transporter kinetic parameter for each organ, the transport kinetic parameters for M1 were excluded for the model. For elaborate model prediction, OCT1 and MDR1 genotypes (*OCT*\*1, \*2, \*3, \*4, \*5, and *MDR1* C3435T) could be incorporated.
