Contribution of molecular properties to extrapolation of the volume of distribution in human from preclinical animal species data

This paper quantitatively investigated the contribution of molecular properties to the volume of distribution in human (V(d human) ) when extrapolated from preclinical animal data, and identified which molecular descriptors and animal species were essential or better for acquiring the optimal accuracy of extrapolation. First, several two-dimensional molecular descriptors which can potentially contribute to V(d) were selected to establish the model to predict V(d) in humans. Then, several linear predictive models were constructed by partial least squares (PLS) using three or one animal data in combination with or without the molecular descriptors. Stepwise regression was performed to determine the important molecular descriptors. For comparison, the allometric method was also performed to estimate the V(d human) from preclinical species (rat, dog and monkey) data. The predictive accuracy of PLS models was better than that obtained by the allometric method. Moreover, the predictive accuracy of these models was improved when the molecular descriptors were introduced. Interestingly, the common contributors selected were molecular weight (MW) and the negatively charged fraction value (f(i-) ) for any stepwise regression model concerning molecular descriptors. Additionally, the dog was a more suitable species for predicting V(d human) during drug-lead pharmacokinetic optimization. More importantly, the introduction of molecular properties could improve the predictive accuracy for the prediction of V(d human) from the preclinical species, among which MW and f(i-) were the significant parameters and the latter was considered in the extrapolation process for the first time.