Which unbound fraction should we use in the well-stirred model for more accurately predicting hepatic clearance of drugs for humans?

As the hepatic clearance (CL) of drugs becomes independent of hepatic blood flow rate, CL depends primarily on intrinsic clearance (CL). Incubations of microsomes or hepatocytes are commonly used to generate CL. Therefore, CL estimates corrected for binding to the in vitro systems and scaled to whole liver, are applied to a well-stirred liver model to obtain CL predictions for drugs. In other words, CL is extrapolated with the ratio of unbound fraction between the plasma (fu) and incubation medium (fu). However, this binding correction resulted to an important underprediction bias of CL. Therefore, the approach considering fu and fu needs to be better understood for more precisely scaling CL. The objective of this study was to explain the underprediction bias of CL based on physicochemical properties of drugs. Analysis-ready datasets have been collected so that evaluation of the data generates a mechanistic understanding of the impact of unbound fraction on the prediction of CL of human for 128 drugs. Experimental values of fu for liver are quantifying solely the binding to lipids in microsomes or hepatocytes in the absence of plasma proteins in the incubations. However, the experimental values of fu for plasma can estimate the binding to lipids and plasma proteins. Therefore, drugs binding primarily to lipids in the liver and plasma showed a less pronounced underprediction bias of CL by using the conventional ratios of fu/fu in the well-stirred model. In contrast, drugs binding primarily to plasma proteins in the liver and plasma showed a larger underprediction bias of CL. Furthermore, for the ionized drugs, values of fu and fu are not covering the pH gradient effect between plasma and hepatocytes, which also impacted the CL predictions. For these reasons, a mechanistic approach was proposed to replace the conventional fu value with an adjusted fu value (fu) that accounts for differences in proteins/lipids binding and pH gradient effect between the liver and plasma. Hence, replacing fu with fu did provide a universal and mechanism-based approach removing the underprediction bias for all datasets of drugs. Overall, this study indicates which drug properties generated the largest underprediction bias of CL and suggests that the Poulin et al. method referring to fu could be the most proper unbound fraction to remove that bias with the well-stirred model.