Cytochrome P450 Affinity

Cytochrome P450 Affinity

The Cytochromes P450 are a superfamily of metabolic enzyme present in a wide range of organs, and cells. In particular, phase I metabolism by P450s in the liver is a major route of clearance for many drug compounds and, in some cases, may result in bioactivation, forming toxic metabolites. There are a large number of isoforms, each with different substrate specificities, distributions in the body and rates of metabolism.

Why is P450 Affinity Important?
CYP2D6 and CYP2C9 are two of the three most significant drug-metabolising enzymes, along with CYP3A4. Although metabolism of drugs via CYP2D6 and CYP2C9 is not as common as for CYP3A4, interaction with CYP2D6 or CYP2C9 is a significant cause of drug-drug interactions, due to inhibition of clearance of another drug primarily metabolised by the same enzyme. In addition, interaction with CYP2D6 is also considered to be unfavourable as it is the subject of a well-known genetic polymorphism, resulting in approximately 10% of the caucasian population having a 'poor metabolise' phenotype in which activity of CYP2D6 is dramatically reduced. The affinity of a ligand is defined by the Ki, the molar concentration required to occupy half the binding sites available to a competitor ligand, in the absence of radioligand or competitors (if the Ki value is low then the affinity is high). This is commonly reported as a pKi value (i.e. log10(1/Ki) with Ki in M). In this case, the greater the pKi value, the higher the affinity. It should be noted that a high affinity does not necessarily indicate that a compound will be metabolised by the particular P450. Conversely, very low affinity compounds are unlikely to be significantly turned over by the enzyme.

CYP2D6 Affinity Model
The data for this model were generated in-house, due to the high inter-laboratory variation observed in reported P450 affinities in the literature. The data consist of accurate Ki values generated for competitive inhibitors using a multi-point Ki protocol. A total of 213 data points were generated in this data set. Due to an uneven distribution of Ki values in the data set the CYP2D6 affinity data were classified into 4 categories; low (pKi <5), medium (5=< pKi =7). The model for CYP2D6 affinity classifies compounds as 'low', 'medium', 'high' or 'very high' affinity, according to the class boundaries given above. A confidence for each prediction is reported, according to the strength of association of the compound's descriptor values with the predicted classification. Furthermore, the distance of the predicted compound from the chemical space of the training set is calculated to gauge the confidence in the result. The results from the test set show that for every category, between 72% and 78% of compounds were predicted correctly. Importantly, this demonstrates that the model can identify compounds with high/very high affinity for CYP2D6.

Observed vs Predicted 2D6 Affinity Category

CYP2C9 Affinity Model
The data for this model were generated in-house, due to the high inter-laboratory variation observed in reported P450 affinities in the literature. The data consist of accurate Ki values generated for competitive inhibitors using a multi-point Ki protocol. Data for a total of 130 compounds were generated in this data set covering a wide range of chemical diversity. A continuous random forest model for CYP2C9 inhibition was developed. The model predicts a compound's pKi along with an estimate of the RMSE in prediction. The model automatically determines whether or not a compound lies within the chemical space defined by the training set. The R2 for the training set of 105 compounds was 0.92 and the RMSE in fit was 0.33 log units. The R2 value for the independent test set of 25 compounds was 0.64 and the standard error in prediction was 0.60 log units.

Observed vs Predicted 2C9 pKi