Improving Metabolic Stability
This project's goal was to employ a 'fast-follower' strategy to improve on a competitor compound that had problems with poor oral bioavailability and short and variable half-life in man. These issues were caused by rapid metabolism by Cytochrome P450 CYP3A4; in agreement with StarDrop's identification of highly labile sites and a high predicted logP. The final compound had to be suitable for oral dosing and the target was in the CNS.
In the initial phase of the project, the goals were to identify compounds which improved the in vitro half-life with respect to CYP3A4 metabolism by a factor of 2 while maintaining receptor IC50 of less than 200nM.
A virtual library of 13,000 novel compounds was enumerated, based on the structure of the competitor compound. From this, a set of 100 compounds was chosen for synthesis and testing in vitro with a range of chemical diversity and physicochemical properties. The objective of this phase was to understand the relationship between potency and stability SAR.
The synthesised compounds were tested for potency and in vitro stability against recombinant CYP3A4. The results: 70% met the criterion for potency, 11% met the requirement for stability, but only 3% met both criteria.
From these data, local models were built for both CYP3A4 stability and potency, which were used in Phase II to select compounds likely to meet both criteria.
All 13,000 virtual molecules were assessed with the local models created in Phase I. From these, only 40 were predicted to be both stable and potent. All of these were also predicted to be soluble, absorbable and to cross the blood-brain barrier.
These compounds were then synthesised and tested in vitro for potency and CYP3A4 stability.
Only 140 compounds were synthesised in this project and this resulted in 4 lead series, each containing multiple compounds that were both potent and more than twice as stable as the competitor with respect to CYP3A4 metabolism. This demonstrates how StarDrop can guide decisions to rapidly overcome liabilities and focus on chemistries with a good balance of properties.
Building a Local Model of CYP3A4 Rate
The CYP3A4 stability data generated in Phase I of this project were expressed as the log of the rate of metabolism (logk3A4). A simple set of descriptors was chosen based on the primary factors affecting the rate of CYP3A4 metabolism. The affinity of CYP3A4 tends to be higher (lower Km) for neutral, lipophilic compounds. Therefore, as all of the compounds in this project had a basic site, the log of the fraction of compound that will be neutral at pH 7.4 (based on pKa calculated with ACD Labs) was used as a descriptor, along with logP calculated with StarDrop. The efficiency of metabolism is expected to be higher (leading to higher Vmax) for compounds with labile sites, therefore the composite site lability (CSL) calculated by StarDrop's P450 models was also used as a descriptor.
StarDrop's Auto-Modeller was used to build a model of these data. The data set was split into training, validation and test subsets using Y-based sampling, as the data set was small and all compounds were relatively similar to the competitor compound. Nine models were automatically built based on the training set and their performance compared on the validation set. The output of the Auto-Modeller is shown to the right; the best model, generated with the Gaussian Process Nested Sampling method had an R2 of 0.66 (r2corr=0.72) on the combined independent validation and test sets and had a high specificity at the required cut-off (logk3A4<-1.5).