Letter to the Editor BINDING OF DRUGS TO HEPATIC MICROSOMES: COMMENT AND ASSESSMENT OF CURRENT PREDICTION METHODOLOGY WITH RECOMMENDATION FOR IMPROVEMENT

In recent years the phenomenon of hepatic microsomal nonspecific binding of drugs has received considerable attention in this journal, as well as in others. There is general agreement on the need to correct kinetic constants (namely, Km and Ki values) for the fraction of drug concentration unbound (fuinc) to ensure optimal prediction of clearance and inhibition potential in vivo from in vitro hepatic microsomal studies (Tucker et al., 2001; Bjornsson et al., 2003). The ability to predict in silico the extent of microsomal binding from a drug’s physicochemical properties, as proposed by Austin et al. (2002), is an attractive option that has engendered much interest. Recently, the same authors have extended earlier principles from microsomes to intact hepatocytes (Austin et al., 2005) and have shown application to various sets of hepatocyte and microsomal clearance data (Riley et al., 2005). Having demonstrated a relationship between the charge-dependent lipophilicity parameter “logP/D” and binding affinity to microsomes for a range of drugs, Austin et al. (2002) proposed that the fuinc could be predicted using a simple linear relationship. This method was based on a theoretical combination of the above relationship and the ability to project fuinc from one concentration of microsomes to another. As no evaluation of prediction outcome was provided, we have assessed the prediction of fuinc using the Austin equation. A dataset including the 56 drugs used in the original study together with an additional 36 drugs both from other published studies (Obach, 1999; Naritomi et al., 2001; Soars et al., 2002) and from our own unpublished work was used. First, we examined logP values from several sources (including experimental and in silico) for as many drugs for which data were available but found no major influence of source; mean logP values were therefore used, and these were converted to logD values where appropriate. Binding affinity (Ka) values were calculated from experimental microsomal fuinc values, and the latter were converted to values corresponding to 1 mg microsomal protein per ml, as required; when several experimental values were obtained for the same drug, a mean value was used. In demonstrating the relationship between binding affinity and logP/D, Austin et al. (2002) deliberately omitted drugs weakly bound (fu 0.9)—effectively excluding acidic drugs— and justified this based on the relative difficulty in determining minor binding and the fact that this degree of binding was not important. Figure 1A shows the relationship between binding affinity and logP/D for all of the basic and neutral drugs in the extended database (n 64). The fitted linear relationship is similar to that obtained by Austin et al. (2002), and the correlation (r 0.68) is comparable despite the increase in the number of drugs used. The linear equation parameters obtained using this extended database analysis were substituted into the Austin equation, and the resultant predicted fuinc values for all drugs were compared with observed values of fuinc. This prediction calculation typically resulted in underprediction of fuinc across the range of binding (average fold error 1.5); 65% of all fuinc predictions were less than the observed fuinc values. In terms of precision, 28% of the predictions were greater than 2-fold different from observed. The charge-dependent parameter, logP/D, suggested by Austin et al. (2002) does show the most consistent trend between both drug type and individual drugs, including acidic drugs, when compared with logP or logD. It does not seem justified to exclude acidic drugs in quantifying this relationship because they provide limiting values within this inclusive property. Figure 1B, which includes all the drugs, shows that the relationship between binding affinity and logP/D is not linear. Any mechanistic model for microsomal binding would involve a polar interaction as well as a lipophilic/hydrophobic interaction, and so it is not surprising that the overall interaction is more complex than a simple linear relationship. As an alternative empirical approach, we fitted a quadratic relationship to all the data in the extended database and modified the fuinc prediction to that given in the equation below.