Plan de Capacitación de Calidad

The dataset of 225 compounds collated, or experimentally determined, by Karlgren and co-workers (2012a) were used in this study. The OATP subfamilies,

158

OATP1B1, OATP1B3 and OATP2B1 were included in the dataset. A total of 142 compounds in this dataset was from an earlier investigation (Karlgren et al 2012b), which was then expanded to include compounds known to interact with OATPs or CYP enzymes (Karlgren et al., 2012a). The compounds were from the chemical space of oral drugs (Karlgren et al., 2012a). Data consisted of percentage OATP inhibition by the compounds.

The experimental measurements were performed using the human embryotic kidney 293 (HEK293) cells stably transfected with OATP1B1, OATP1B3 or OATP2B1. In the screening experiments to measure interaction of the 225 compounds with each individual OATP, a concentration of 20 µM of the compounds was used. The substrates used in the inhibition studies were estradiol- 17β-glucuronide for OATP1B1 and OATP1B3, and estrone-3-sulfate for OATP2B1. The substrate concentration was 0.52 µM in the inhibition of OATP1B1 mediated estradiol-17β-glucuronide uptake. In the inhibition of OATP1B3 mediated estradiol-17β-glucuronide uptake, the substrate concentration was 1.04 µM and in the inhibition of OATP2B1 mediated estrone-3-sulfate uptake, the substrate concentration was 1.02 µM.

The PCA of the dataset indicates that compounds are well distributed in the oral drug space with 95% confidence interval. The dataset included 43% neutral compounds, 29% negatively charged, 22% positively charged and 6% zwitterionic compounds at pH 7.4 (Karlgren et al., 2012a).

For development of QSAR models for OATP interaction, both classification and prediction (regression based) methods were used. The continuous (numerical) percentage inhibition data were used for regression based analyses. For classification methods, compounds were considered as inhibitors if they significantly decreased the uptake of the substrate by at least 50%. In this case, 78 compounds (out of 225 compounds) were OATP1B1 inhibitors, while 46 and 45 compounds (out of 225) were OATP1B3 and OATP2B1 inhibitors, respectively. In the dataset, a few compounds stimulated OATP mediated transporter (instead of inhibition). Clotrimazole, fendiline, progesterone and testosterone are the example of stimulators (Karlgren et al., 2012a). In this investigation all such compounds were considered as non-inhibitors in classification studies.

159

A total of 387 2D and 3D molecular descriptors were calculated for OATP dataset using the same methods and software as explained in Chapter 4.

6.2.2. QSAR Model Development and Validation

6.2.2.1. OATP Models

Both regression-based and classification models were developed for OATP interaction. The regression based models were linear and non-linear methods of stepwise regression analysis, C&RT, BT, RF and MARS. The classification method was C&RT. All statistical analyses were performed using STATISTICA Data Miner v11 (StatSoft Ltd.).

The compounds were divided into external validation set and training data. Models were developed using training set compounds and assessed using external validation sets. To divide the compounds, they were ordered according to their inhibition percentage and from every set of five compounds, four were allocated into the training and one into the external validation set by random. In this way, training data consisted of 180 compounds and external validation set consisted of 45 compounds. For the analytical methods that required parameter optimization, a fraction of training set compounds were randomly assigned into internal validation set, or alternatively cross validation was used if the option was available in the statistical software. For the internal validation set, where applicable, the risk estimate and standard error were calculated in STATISTICA software and used as the performance indicators.

In OATP modelling using boosted trees, the default values for learning rate, the number of additive terms (number of trees), random test data proportion (fraction of data points in testing pool) and subsample proportion were 0.1, 200, 0.2 and 0.5, respectively. In addition to the default values, various subsample proportions of 0.4, 0.45, 0.50, 0.55 and 0.60 were examined in combination with the learning rates of 0.1 and 0.05. The best OATP models were selected based on the performance indicators for the internal validation set.

160

6.2.2.2 Biliary Excretion Models

QSAR models were developed for biliary excretion using the dataset and methods explained in Chapter 4. In addition to the molecular descriptors, the OATP effects predicted by the selected models from section 6.2.2.1 were used as the independent variables of the analyses. To this end, the selected OATP models from section 6.2.2.1 were used to predict OATP interaction (percentage inhibition values or inhibitor/non-inhibitor classes) for the compounds in biliary excretion dataset (n = 217). In addition to C&RT method, interactive C&RT was used in which the predicted OATP effects were manually incorporated in the models, when they were not picked by C&RT feature selection automatically.

6.3. Results

It has been cited in the literature that presence of OATPs in the hepatocytes may indicate their significance in biliary excretion process (Matsushima et al., 2005; Pfeifer et al., 2014; Shitara et al., 2013). Binding of 225 compounds to three major sub-family members of hepatic organic anion transporting polypeptides (OATP transporters) were available for this analysis. These sub-families were OATP1B1, OATP1B3 and OATP2B1. The ratios of inhibitors to non-inhibitors were different for each of these three proteins, as can be seen in Table 6.1. A total of 387 molecular descriptors were used for the QSAR model development for the training set consisting 180 compounds. The method of data allocation into training and test sets outlined in the methods section ensured that these sets contained similar ranges of percentage inhibition values. The lipophilicity (LogP by ACD software) was between -4.73 and 8.51 for the training set, and -3.26 and 7.28 for the validation set with similar mean values of 2.43 and 2.58 respectively. Molecular weights of the compounds were between 129-1214 Da for the training set and 94-1202 Da for the validation set, with mean values of 405 and 392 respectively.

161

Table 6.1. Number of inhibitor/non-inhibitor compounds based in 50% inhibition for each OATP sub-family members

Transporter Inhibitor Non-inhibitor Total

OATP1B1 78 147 225

OATP1B3 46 179 225

OATP2B1 45 180 225

Several QSAR models were developed for each sub-family of OATP transporter using the training set compounds. Based on the prediction error for the validation sets, two QSAR models were selected for the prediction of binding to each OATP for the biliary excretion dataset. Section 6.3.1 gives a brief description of the regression based models, while section 6.3.2 gives description of classification models for OATP interaction. The results of using the predicted OATP effects as the independent variables (descriptors) of biliary excretion models have been presented in section 6.3.3.