- Email: [email protected]
Modeling Organic Anion-Transporting Polypeptide 1B1 Inhibition to Elucidate Interaction Risks in Early Drug Design
Journal of Pharmaceutical Sciences 105 (2016) 3214-3220
Contents lists available at ScienceDirect
Journal of Pharmaceutical Sciences journal homepage: www.jpharmsci.org
Pharmacokinetics, Pharmacodynamics and Drug Transport and Metabolism
Modeling Organic Anion-Transporting Polypeptide 1B1 Inhibition to Elucidate Interaction Risks in Early Drug Design Ismael Zamora 1, Susanne Winiwarter 2, * 1 2
Lead Molecular Design, S.L., Sant Cugat del Vall es, Spain €lndal, Sweden Drug Safety and Metabolism, AstraZeneca R&D Gothenburg, Mo
a r t i c l e i n f o
a b s t r a c t
Article history: Received 10 May 2016 Revised 23 June 2016 Accepted 12 July 2016 Available online 12 August 2016
The importance of transporter proteins for the disposition of drugs has become increasingly apparent during the past decade. A noted drug-drug interaction risk is the inhibition of organic anion-transporting polypeptides (OATPs), key transporters for the liver uptake of the widely used statins. We show here the development of a ligand-based in silico model for interaction with OATP1B1, an important representative of the OATP family. The model is based on a structural overlay of 6 known OATP1B1 inhibitors. A data set of about 150 compounds with published OATP1B1 inhibition data was compared to the resulting “transportophor,” and a similarity threshold was defined to distinguish between active and inactive molecules. In addition, using a statistical model based on physicochemical properties of the compounds as prefilter was found to enhance the overall predictivity of the model (final accuracy 0.73, specificity 074, and sensitivity 0.71, based on 126 compounds). The combined model was validated using an in-house data set (accuracy, specificity, and sensitivity were 0.63, 0.59, and 0.78, respectively; 62 compounds). The model gives also a structural overlay to the most similar template enabling visualization of where a change in a given structure might reduce the interaction with the transporter. © 2016 American Pharmacists Association®. Published by Elsevier Inc. All rights reserved.
Keywords: structure property relationships molecular modeling organic anion-transporting polypeptide transporters drug interactions
Introduction During the past decade, the importance of transporter proteins for the uptake of drugs into various tissues, and the resulting potential for drug-drug interactions, has become more visible.1-6 The continuous effort in drug design to find metabolically stable compounds with high affinity to the target under consideration is likely driving this development. A goal during drug design is to aim for compounds with high lipophilic ligand efficiency,7 that is, compounds that have a high-specific affinity to the target but as low lipophilicity as possible to reduce unspecific binding to other targets including metabolizing enzymes. Metabolically stable compounds
Abbreviations used: 2D, two dimensional; 3D, three dimensional; DDI, drugdrug interaction; HEK cells, human embryonic kidney cells; IC50, inhibition concentration to give 50% inhibition (i.e., half maximal inhibitory concentration); OATP, organic anion-transporting polypeptide; PCA, principal component analysis; PLS, projection to latent structures (also known as partial least squares analysis); r2, coefficient of determination; r2(CV), cross-validated coefficient of determination; TS score, transportophor similarity score. This article contains supplementary material available from the authors by request or via the Internet at http://dx.doi.org/10.1016/j.xphs.2016.07.011. * Correspondence to: Susanne Winiwarter (Telephone: þ46 31 7064913; Fax: þ46 31 7763700). E-mail address: [email protected] (S. Winiwarter).
cannot depend on metabolic clearance for their elimination from the body but require other mechanisms including active transport.8 Moreover, such compounds may also rely on uptake via special transporter proteins for being distributed to the tissue of interest because more hydrophilic compounds are less likely to cross lipophilic membranes through a purely passive mechanism. Organic anion-transporting polypeptides (OATPs) are transporters at the basolateral side of hepatocytes. They have been shown to be of importance for liver uptake and thereby for clearance of endogenous acidic compounds, such as bile acids or bilirubin.9-11 This pathway is also used by certain amphiphilic drugs. For example, 3-hydroxy-3-methylglutaryl-coenzyme A reductase inhibitors, also known as statins, make use of OATPs to reach their target within the hepatocyte. In addition, transport via OATPs ensures clearance from the blood and thereby reduces the risk of toxic side effects.12 Inhibition of OATPs can therefore lead to interaction with a drug's disposition, its safety risks, and action. New pharmaceutics likely to be comedicated with statins need to avoid any interaction with OATPs.13 Successful drug design in the cardiovascular area, where statins are considered standard of care,14 therefore depends on a good understanding of the structure-transporter activity relationship early on the drug discovery process. OATP1B1 inhibition potential can be measured in vitro through determining how much the transport of a known OATP1B1
http://dx.doi.org/10.1016/j.xphs.2016.07.011 0022-3549/© 2016 American Pharmacists Association®. Published by Elsevier Inc. All rights reserved.
I. Zamora, S. Winiwarter / Journal of Pharmaceutical Sciences 105 (2016) 3214-3220
substrate is hindered by addition of the query compound using OATP1B1-transfected cell lines.15,16 Fluorescent marker substrates have been suggested for utilization in high throughput screening.17-20 Knowledge of the substrate used is important to understand which data are comparable because inhibition constants vary depending on the probe substrate used.9,16,21 Specific drugs known to be OATP1B1 substrates or estradiol-17b-glucuronide have been suggested as most relevant probe substrates for drug-drug interaction risk assessment.4,21 In early drug discovery, and to more actively guide compound design, in silico models can be used. Various approaches to model transporter interactions have been reported,22-25 including an increasing number of models for OATP1B1 interactions: In 2005, Chang et al.26 defined a pharmacophore for OATP1B1 substrates consisting of 2 hydrogen bond acceptors, 1 hydrogen bond donor and a central lipophilic region, based on literature data for human and the rat. Inhibitor models for OATP1B1 most often use quantitative structure activityerelationship approaches.15,16,27-31 Lipophilicity and hydrogen bonding were found to be important molecular properties that influence OATP1B1 inhibition in most of these studies.28,30 Using comparative molecular field analysis, Gui et al.32 analyzed 18 diverse inhibitors and proposed an interaction mode with a central core and peripheral polar regions, one of which is acidic, as structural features that promote interactions with the transporter. It was discussed that pharmacophore models might be difficult to obtain for diverse sets of compounds30 since the varying inhibition constants depending on the substrate used,9 as well as biphasic binding kinetics for example in the case of estrone-3sulfate,33,34 indicate the possibility of different binding sites. The aim of the present work was the development of a fast ligand-based in silico model for inhibition of OATP1B1 that can be used to guide drug design in early drug discovery. We used a rapid pharmacophore type approach, alignment of molecular shape,35,36 as primary method, based on a small set of established OATP1B1 inhibitors. Note that these compounds also can be transported by the protein and most likely act as competitive inhibitors. Comparing query structures to this alignment, designated “transportophor,” resulted in both a similarity score and the best structural fit. The similarity score was able to indicate whether a compound had low or high risk for OATP1B1 inhibition, and a
3215
similarity score threshold was defined. The best structural fit showed which parts of a query structure were aligned with the features of a known inhibitor, thereby indicating which part of the molecule may be important for the interaction with the transporter. However, we found that a prefilter based on physicochemical descriptors, using the projection to latent structure (PLS) method, clearly enhanced the prediction quality, as shown in the external validation using in-house compounds. Materials and Methods Compound Sets Three different compound data sets were used in this study. Data set 1 consists of 6 known OATP1B1 inhibitors (estrone-3sulfate, estradiol-17b-glucuronide, taurocholic acid, rosuvastatin, pitavastatin, and pravastatin)15,26 and was used to define the alignment rule for the transportophor generation (see Fig. 1). Three compounds belong to the class of endogenous steroids and the remaining are statins, that is, compounds of particular interest in the cardiovascular research area. The activity for the selected compounds ranges from 60% inhibition for pravastatin and rosuvastatin to >90% for pitavastatin. All 6 compounds have a negatively charged moiety, a lipophilic core and at least 1 hydrogen bond acceptor. The negatively charged moiety is not always a carboxylic acid but can be any acid isostere. Data set 2 comprises 146 compounds reported by Karlgren et al.15 and was used for determination of the similarity threshold to discriminate between active and inactive compounds through comparison to the transportophor. The data set was also used as training set for the PLS model based on physicochemical parameters. Inhibition of OATP1B1-mediated transport had been measured as described in human embryonic kidney cells transfected with OATP1B1, using estradiol-17b-glucuronide (0.5 mM) as substrate. The test compound concentration was 20 mM.15 Data set 3 is a set of 63 AstraZeneca proprietary compounds that was used for validation. OATP1B1 inhibition measurements for these compounds were performed in human embryonic kidney cells transfected with OATP1B1 as described by Soars et al.,16 using estradiol-17b-glucuronide as substrate. The compounds were
Figure 1. Six known OATP1B1 inhibitors: upper row: estrone-3-sulfate, estradiol-17b-glucuronide, taurocholic acid; lower row: rosuvastatin, pitavastatin, pravastatin.
3216
I. Zamora, S. Winiwarter / Journal of Pharmaceutical Sciences 105 (2016) 3214-3220
measured at 6 concentrations (ranging typically from 0.3 to 100 mM). Activity classification was based on the % inhibition value at 30 mM.
compound was found, they were aligned with ROCS using estrone3-sulfate as template to give a consistent coordinate space for all solutions. These 6 aligned structures were termed transportophor.
Multivariate Analysis Threshold Definition Principal component analysis (PCA) and PLS analysis were performed using Simca with default options.37 The descriptor set used was an in-house standard descriptor set, consisting of 196 2D- and 3D-physicochemical descriptors (AZ descriptors38). For the PLS model, the % inhibition at 10 mM was transformed into a categorical activity value for the compounds from data set 2 as follows: % inh <30%: 0; 30%-40%: 1; 40%-50%: 2, 50%-60%: 3; 60%-70%: 4; 70%-80%: 5; 80%-90%:6; >90%: 7. This categorical activity was used as y-variable. Transportophor Generation Ligand Preparation 2D representations (SMILES code39) of the 6 selected compounds were used as starting point. The ionization state was adjusted to pH 7.4, using OEleatherface (an in-house adaption of leatherface40 for the OpenEye toolkit41). The structures were converted to 3D via Corina (Molecular Networks) and subjected to energy minimization using Szybki (OpenEye software).41 Finally a conformational analysis was performed using Omega (OpenEye),41 ensuring unique titles for each conformer (option “-warts”). A maximum of 200 conformations per compound was considered. Ligand Superposition ROCS (OpenEye)41 was used for ligand superposition including the ImplicitMillsDean color force field42 as part of the overlay optimization (option “optchem”). All conformations of each of the 6 template molecules were superimposed to all conformations of the other 5 templates. The conformation most often showing the highest similarity score, using the “RefTverskyCombo” score as provided by ROCS, was selected as overall best fitting conformation for the specific template. Once the conformation for each
Data set 2 compounds were classified as active or inactive according to the reported inhibition at 20 mM, using 50% inhibition as cutoff.15 Ligand Preparation The compounds were subjected to the same ligand preparation procedure as the 6 template compounds. For compounds with ambiguous ionic state definitions, all potential ionic states were considered. Compounds that produced any error in this procedure were removed from the automatic analysis. Alignment to Transportophor and Threshold Definition All conformations of each molecule were used for alignment to each of the 6 structures comprising the transportophor using the same procedure as mentioned previously, that is, using ROCS considering the ImplicitMillsDean color force field. For each conformation, the 6 similarity score values (RefTverskyCombo score) obtained from the overlay with each of the 6 template structures were summed up to give the “sum of scores.” The highest value found for any of the conformations, or ionic forms, of a molecule defined the “transportophor similarity score” (TS score) for this molecule. The TS score where 75% of the actives could be identified was selected as primary threshold value. A secondary, lower, value was chosen to define a medium class. Validation Data set 3 was used for validation only. The compounds were classified according to their measured inhibition at 30 mM. Compounds with an inhibition <40% were deemed as noninhibitors, whereas compounds with an inhibition >60% were considered as
Figure 2. PCA score plot based on data set 2 (circles, data set 1 compounds shown as squares). ReddOATP1B1 inhibitors (dark to light correspond to categories 7, >90% inhibition, to 4, 50%-60% inhibition); greendnoninhibitors (dark to light correspond to categories 0, <30% inhibition to 3, 40%-50% inhibition); gray line indicates approximate discrimination line between inhibitors (above) and noninhibitors (below). PC 1 describes mainly molecular size, whereas PC 2 can be related to the lipophilicity of the compounds.
I. Zamora, S. Winiwarter / Journal of Pharmaceutical Sciences 105 (2016) 3214-3220
inhibitors. Compounds with an inhibition between 40% and 60% were regarded as possible inhibitors. The compounds were subjected to ligand preparation and alignment to transportophor as described previously. The resulting TS score was then used to define in which of the 3 categories the compound would fall based on the threshold values defined previously. In addition, the PLS model was used to predict the likelihood of the compound's being an OATP1B1 inhibitor, through estimation of the “activity category,” used as y-variable in the model. Compounds with a predicted activity category below 2 were considered as inactive, whereas compounds with a predicted activity category above 4 were considered as active. Classification statistics were obtained for each of the 2 models but also for the combination model.
3217
Figure 3. Template compound alignment (estrone-3-sulfate [black], estradiol17b-glucuronide [light gray], taurocholic acid [dark gray], rosuvastatin [light green], pitavastatin [green], and pravastatin [dark green]).
Results and Discussion In 2005, Chang et al.15,26 published an analysis of OATP1B1 substrate data and reported a pharmacophore model for both human OATP1B1 and rat oatp1a1 transporter substrates, indicating that interaction with OATP1B1 required several hydrogen-bonding features in the periphery of the molecule and a lipophilic central region. Here, we attempt to generate a pharmacophore type model for OATP1B1 inhibition. Six reasonably diverse OATP1B1 inhibitors15 were selected for this purpose. Figure 2 shows the score plot resulting from a PCA of data set 2 based on physicochemical descriptors. A natural separation between the active and the inactive molecules is noted, which is in line with literature.15 However, the used physicochemical properties do not explain all data because the discrimination is less successful in the central part of the plot. The 6 selected compounds, estrone-3-sulfate, estradiol-17b-glucuronide, taurocholic acid, rosuvastatin, pitavastatin, and pravastatin, can be found in this border area, where the physicochemical properties do not seem to be sufficient to distinguish between actives and inactives. Using a pharmacophore approach should be able to help identify OATP1B1 inhibitors based on their structural similarity instead. Furthermore, the 6 selected compounds represent 2 important compound classes: endogenous steroidal acids and statins, considered as standard of care in treating cardiovascular disease. Transportophor Generation The 6 selected OATP1B1 inhibitors were subjected to conformational analysis as described in Materials and Methods. The ionized form was used for all compounds as the predominant form at pH 7.4. The compounds were flexible except estrone-3-sulfate and estradiol-17b-glucuronide for which only 5 and 16 conformations were generated, respectively. Table 1 shows the number of conformations, the top-ranked conformation and how often it was ranked first, and the averaged RefTverskyCombo similarity score value for the top-ranked conformation for each of the 6
compounds. Alignment of these conformations based on estrone-3sulfate resulted in a depiction of the generated transportophor (see Fig. 3). The main structural features are a peripheral hydrophilic, acidic functionality, and a big lipophilic part with only few heteroatoms, which is consistent with published OATP1B1 pharmacophores.26,32 In Figure 3, this alignment is shown with the acidic substructures on the left-hand side opposite to the lipophilic core of the molecules. Note, that both taurocholic acid (dark gray) and estradiol-17b-glucuronide (light gray) extend the space used by the other 4 compounds in this region. Threshold Definition The similarity score gives an estimation of how similar a compound is to the transportophor. By checking the similarity score for a set of compounds with known inhibition activity, a threshold value could be defined to distinguish between active and inactive molecules. Data set 2 compounds15 were subjected to a similar procedure as the transportophor template molecules: ionization status of the compounds was adjusted to pH 7.4. All conformations, anddif applicabledlikely ionic states at pH 7.4, identified for a compound were aligned to the 6 transportophor structures. The highest “sum of scores” for any of the conformations (forms) defined the TS score for the query molecule. For 18 compounds, no results could be obtained, due to errors in the automatic procedure (see Supporting Information). Removing also the 6 template compounds left 122 compounds for threshold generation
Table 1 Transportophor Generation Results Compound
Number of Selected Conformer Average Score of Selected Conformer Conformers (Number of (RefTverskyCombo) Times Shown as First Option)
Estrone-3-sulfate Estradiol-17bglucuronide Taurocholic acid Rosuvastatin Pitavastatin Pravastatin
5 16
c4 (220) c14 (107)
0.89 1.05
200 200 200 200
c101 c165 c133 c165
1.01 1.10 1.10 1.17
(38) (39) (36) (52)
Figure 4. Cumulative % of active (black; n ¼ 49) and inactive compounds (gray; n ¼ 73) plotted against the TS scores. Vertical lines indicate primary (full) and secondary (dashed) threshold.
3218
I. Zamora, S. Winiwarter / Journal of Pharmaceutical Sciences 105 (2016) 3214-3220
Figure 5. Calculated versus experimental activity using the PLS model, color by experimental activity (black >60% inhibition, i.e., categories 4-7; gray 40%-60% inhibition, i.e., categories 2 and 3; and light gray <40% inhibition, i.e., categories 0 and 1), shape by calculated activity category (stars: high, calculated activity category >4; diamonds: medium, calculated activity categories 2-4; circles: low, calculated activity category <2).
(49 inhibitors and 73 noninhibitors). Previously, both, using only the single highest score and different similarity indices provided by ROCS were tested. As result the “sum of scores” considering the RefTverskyCombo score was found most useful to define a TS score (data not shown). Figure 4 shows the cumulative % of inhibitors and/or noninhibitors versus the TS score, that is, the percentage of active or inactive compounds with a value lower than a given score was plotted against the score. It can be seen that 75% of the inhibitors in this set had a TS score of at least 6.5, which was used as upper threshold. About 30% of the inactive compounds also had a TS score above the threshold (see Fig. 4). An additional threshold of 6.0 was used to define a medium class. As result, only ~10% of the actives are falsely defined as inactive. The model seems to be able to distinguish reasonably well between active and inactive compounds in this data set, with an overall accuracy (true predictions/ all compounds) of 0.58. Although the sensitivity (true predicted inhibitors/all experimental inhibitors) was 0.78, the specificity (true predicted noninhibitors/all experimental noninhibitors) was only 0.45, with about 21% of the noninhibitors classified as medium. Less than 20% of the compounds were classified as medium. PLS Model as Prefilter As observed in the PCA score plot (Fig. 2), there is a natural discrimination between the 2 classes of compounds in physicochemical space. One can reason that a compound simply needs to fulfill certain physicochemical prerequisites to be able to reach the transporter. Only then the structural complementarity will be relevant to understand whether a compound really is an inhibitor or not. Thus, we defined a PLS model based on physicochemical
descriptors, which is suggested to be used before the comparison to the transportophor. Data set 2 compounds were used as training set in the PLS analysis. The compounds were assigned an activity category value based on their inhibition potential as described in the Methods section, essentially enabling a discriminative analysis with eight classes using the PLS technique. This hybrid approach has less emphasis on each single value but uses still more of the available information than a two-class model. The derived PLS model consists of 3 components (r2 ¼ 0.53; r2(CV) ¼ 0.44). Dividing the calculated category values into 3 bins, values <2 considered as low, values between 2 and 4 as medium, and values >4 as high, a classification accuracy of 0.57, sensitivity of 0.52, and specificity of 0.6 were achieved. Note that about 40% of the compounds ended in the middle bin, where no distinction between active or inactive compounds was possible (see Fig. 5). Using the PLS model as prefilter by only considering the transportophor results for compounds that were predicted as high or medium in the PLS model, the total accuracy and specificity of the model could be enhanced to 0.73 and 0.74, respectively. In addition, less than 10% of the compounds were predicted as medium. The approach caused a slight reduction of the sensitivity (0.71 instead of 0.78). Note that these numbers are based on the compounds used as training set in the PLS model. Model Validation An in-house data set of 63 compounds was used for external validation of the models. The compounds had been tested as part of the screening cascade in cardiovascular projects, which may account for the bias in the set: more than 60% of these compounds showed OATP1B1 inhibition and only about 25% were inactive (38 actives, >60% inhibition at 30 mM, 17 inactive, <40% inhibition, and 8 compounds with medium inhibition potential). A similar classification can be obtained based on IC50 values, with 20 and 50 mM as IC50 limits. To remain consistent with the published data used to build the model, % of inhibition was used as criteria for the classification. The models were applied independently and in a combination approach where the PLS model was used as pre-filter (see Table 2). Note that 4 compounds gave no results in the transportophor model. It can be easily seen that the transportophor model showed many false positives, indicating that most of the molecules had the structural features to interact with the transporter. Using the PLS model as prefilter, the overall specificity of the model went from 0.18 to 0.59, whereas the sensitivity was reduced from 0.86 to 0.78. Less than 10% of the compounds were found in the middle class in both cases. The PLS model by itself gave an accuracy of 0.52, and a sensitivity and specificity of 0.53, with almost 40% of the compounds in the middle class. The use of 2 independent computational approaches clearly enhances the model predictivity but also the possibility to understand the interactions with the transporter. The PLS model can explain whether a compound has the right overall physicochemical properties, whereas the 3D model adds information on specific structural features that might be important for the interaction.
Table 2 Model Comparison for 63 In-House Test Set Compounds Model
# Cmpds
Accuracy
Sensitivity
Specificity
Pos Prec
Neg Prec
%Pred Medium
Transportophor PLS Combo
59 63 62
0.56 0.52 0.63
0.86 0.53 0.78
0.18 0.53 0.59
0.59 0.91 0.73
0.75 0.53 0.56
7% 38% 6%
# Cmpds, number of compounds considered in specific model; Accuracy, (TH þ TL)/total number of compounds; Sensitivity, TH/AH; Specificity, TL/AL; Pos Prec, positive Precision (TH/APH); Neg Prec, negative precision (TL/APL); %Pred Medium, % predicted medium (APM/total number of compounds); TH, true high; TL, true low; AH, all experimentally high; AL, all experimentally low; APH, all predicted high; APL, all predicted low; APM, all predicted medium.
I. Zamora, S. Winiwarter / Journal of Pharmaceutical Sciences 105 (2016) 3214-3220
3219
Figure 6. Shape-based superposition of OATP1B1 inhibitor repaglinide (gray) on estrone-3-sulfate (green); repaglinide is an OATP1B1 inhibitor (88% inhibition at 20 mM); 2D structure of repaglinide shown to the right.
For drug design, it is essential to understand how a molecular structure should be altered. In some cases, physicochemical properties can be changed so as to reduce the compounds ability to reach the transporter. In other cases, it may be necessary to modify the compound more based on the direct interaction points with the transporter. Thus, the visualization of the structural overlay with the most similar template provided by the transportophor model will be useful. Figure 6 shows, as example, the overlay of repaglinide, a known OATP1B1 inhibitor included in data set 2, with estrone-3-sulfate to investigate how the molecule best fits the template structure. Such graphical information will help to identify molecular regions where changes could disrupt the interaction with the transporter. Repaglinide is a carboxylic acid with very few polar features in the remaining molecule. The carboxylic acid superimposes nicely with the sulfate group in estrone-3-sulfate, whereas the rest of the molecule fits well with the steroid skeleton. Provided that the carboxylic moiety is required for any new molecule, one could consider changes in the more lipophilic part, for example, addition of an exposed hydrophilic group in the isobutyl side chain. Repaglinide is regarded as inhibitor, both in the PLS and the transportophor model. Conclusion We present here an OATP1B1 inhibition in silico model based on two orthogonal computational methods, considering both the physicochemical properties and the structure similarity to known inhibitors. The actual mechanism on how a molecule inhibits the transporter remains unknown, but the compounds need to reach the site of interaction and interact in a certain manner to be able to inhibit the transport of a substrate compound. To describe the physicochemical behavior of the compound, a PLS model based on this type of descriptors was built in a similar approach as reported by Karlgren et al.15 Moreover, to describe the potential interaction in the cavity, an alignment rule, the transportophor, and a 3D similarity criterion were defined to discriminate which compounds can actually inhibit the transport of another compound. The compounds used to create the transportophor presented were selected to fit the specific needs of drug discovery projects in the cardiovascular research area. The model can be easily adapted to use other templates if the chemical space of interest is elsewhere. The combination of both approaches (PLS and 3D similarity) gave a better discriminative result than the application of just one model to describe the transport inhibition. The ability to visualize the structure based overlay is anticipated to support compound design
by elucidating potentially interacting structural features. The model is available for drug projects within AstraZeneca.
References 1. Endres CJ, Hsiao P, Chung FS, Unadkat JD. The role of transporters in drug interactions. Eur J Pharm Sci. 2006;27:501-517. 2. Zhang L, Strong JM, Qiu W, Lesko LJ, Huang S. Scientific perspectives on drug transporters and their role in drug interactions. Mol Pharm. 2006;3: 62-69. 3. Zhang L, Zhang Y, Zhao P, Huang S. Predicting drug-drug interactions: an FDA perspective. AAPS J. 2009;11:300-306. 4. Giacomini KM, Huang S, Tweedie DJ. Membrane transporters in drug development. Nat Rev Drug Discov. 2010;9:215-236. 5. Müller F, Fromm MF. Transporter-mediated drug-drug interactions. Pharmacogenomics. 2012;12:1017-1037. 6. Giacomini KM, Huang S. Transporters in drug development and clinical pharmacology. Clin Pharmacol Ther. 2013;94:3-9. 7. Leeson PD, Springthorpe B. The influence of drug-like concepts on decisionmaking in medicinal chemistry. Nat Rev Drug Discov. 2007;6:881-890. 8. Wu C, Benet LZ. Predicting drug disposition via application of BCS: transport/ absorption/elimination interplay and development of a biopharmaceutics drug disposition classification system. Pharm Res. 2005;22:11-23. 9. Niemi M, Pasanen MK, Neuvonen PJ. Organic anion transporting polypeptide 1B1: a genetically polymorphic transporter of major importance for hepatic drug uptake. Pharmacol Rev. 2011;63:157-181. 10. Shitara Y, Maeda K, Ikejiri K, Yoshida K, Horie T, Sugiyama Y. Clinical significance of organic anion transporting polypeptides (OATPs) in drug disposition: their roles in hepatic clearance and intestinal absorption. Biopharm Drug Dispos. 2013;34:45-78. 11. Fenner KS, Jones HM, Ullah M, et al. The evolution of the OATP hepatic uptake transport protein family in DMPK sciences: from obscure liver transporter to key determinants of hepatobiliary clearance. Xenobiotics. 2012;42:28-45. 12. Elsby R, Hilgendorf C, Fenner K. Understanding the critical disposition pathways of statins to assess drug-drug interaction risk during drug development: it's not just about OATP1B1. Clin Pharmacol Ther. 2012;92:584-598. 13. Sharma P, Butters CJ, Smith V, Elsby R, Surry D. Prediction of the in vivo OATP1B1-mediated drug-drug interaction potential of an investigational drug against a range of statins. Eur J Pharm Sci. 2012;47:244-255. 14. Grundy SM, Cleeman JI, Bairey Merz CN, Brewer Jr HB, Clark LT. Implications of recent clinical trials for the National Cholesterol Education Program Adult Treatment Panel III guidelines. Circulation. 2004;110:227-239. €m CAS, Svensson R, Palm J, Artursson P. In vitro 15. Karlgren M, Ahlin G, Bergstro and in silico strategies to identify OATP1B1 inhibitors and predict clinical drugdrug interactions. Pharm Res. 2012;29:411-426. 16. Soars MG, Barton P, Ismair M, Jupp R, Riley RJ. The development, characterization, and application of an OATP1B1 inhibition assay in drug discovery. Drug Metab Dispos. 2012;40:1641-1648. 17. Annaert P, Ye ZW, Stieger B, Augustijns P. Interaction of HIV protease inhibitors with OATP1B1, 1B3, and 2B1. Xenobiotica. 2010;40:163-176. 18. Bednarczyk D. Fluorescence-based assays for the assessment of drug interaction with the human transporters OATP1B1 and OATP1B3. Anal Biochem. 2010;405:50-58. 19. DeBruyn T, Fattah S, Stieger B, Augustijns P, Annaert P. Sodium fluorescein is a probe substrate for hepatic drug transport mediated by OATP1B1 and OATP1B3. J Pharm Sci. 2011;100:5018-5030.
3220
I. Zamora, S. Winiwarter / Journal of Pharmaceutical Sciences 105 (2016) 3214-3220
20. Gui C, Obaidat A, Chaguturu R, Hagenbuch B. Development of a cell-based highthroughput assay to screen for inhibitors of organic anion transporting polypeptides 1B1 and 1B3. Curr Chem Genomics. 2010;4:1-8. 21. Izumi S, Nozaki Y, Maeda K, et al. Investigation of the impact of substrate selection on in vitro organic anion transporting polypeptide 1B1 inhibition profiles for the prediction of drug-drug interactions. Drug Metab Dispos. 2015;43:235-247. 22. Chang C, Ray A, Swaan P. In silico strategies for modeling membrane transporter function. Drug Discov Today. 2005;10:663-671. 23. Winiwarter S, Hilgendorf C. Modeling of drug-transporter interactions using structural information. Curr Opin Drug Discov Devel. 2008;11:95-103. 24. Ekins S, Polli JE, Swaan PW, Wright SH. Computational modeling to accelerate the identification of substrates and inhibitors for transporters that affect drug disposition. Clin Pharmacol Ther. 2012;92:661-665. 25. Artursson P, Matsson P. Computational prospecting for drug-transporter interactions. Clin Pharmacol Ther. 2013;94:30-32. 26. Chang C, Pang KS, Swaan PW, Ekins S. Comparative pharmacophore modeling of organic anion transporting polypeptides: a meta-analysis of rat Oatp1a1 and human OATP1B1. J Pharmacol Exp Ther. 2005;314:533-541. 27. Karlgren M, Vildhede A, Norinder U, et al. Classification of inhibitors of hepatic organic anion transporting polypeptides (OATPs)dinfluence of protein expression on drug-drug interactions. J Med Chem. 2012;55:4740-4763. 28. DeBruyn TM, vanWesten GJP, IJzerman AP, et al. Structure-based identification of OATP1B1/3 inhibitors. Mol Pharmacol. 2013;83:1257-1267. 29. Cao D, Liu S, Fan L, Liang Y. QSAR analysis of the effects of OATP1B1 transporter by structurally diverse natural products using a particle swarm optimizationcombined multiple linear regression approach. Chemometrics Intell Lab Syst. 2014;130:84-90. 30. van de Steeg E, Venhorst J, Nooijen IHG, DeGroot J, Wortelboer HM, Vlaming MLH. Generation of Bayesian prediction models for OATP-mediated drug-drug interactions based on inhibition screen of OATP1B1, OATP1B1*15 and OATP1B3. Eur J Pharm Sci. 2015;70:29-36. €ger W, Ecker GF. Identification of novel 31. Kotsampasakou E, Brenner S, Ja inhibitors of organic anion transporting polypeptides 1B1 and 1B3 (OATP1B1
32.
33.
34.
35.
36. 37. 38.
39.
40.
41. 42.
and OATP1B3) using a consensus vote of six classification models. Mol Pharm. 2015;12:4395-4404. Gui C, Wahlgren B, Lushington GH, Hagenbuch B. Identification, Ki determination and CoMFA analysis of nuclear receptor ligands as competitive inhibitors of OATP1B1-mediated estradiol-17b-glucuronide transport. Mol Pharmacol. 2009;60:50-56. Tamia I, Nozawa T, Koshida M, Nezu J, Sai Y, Tsuji A. Functional characterisation of human organic anion transporting polypeptide b (OATP-b) in comparison with liver-specific OATP-C. Pharm Res. 2001;18:1262-1269. J, Portmann R, Brun M, Funk C. Substrate-dependent drug-drug interactions Noe between gemfibrozil, fluvastatin and other organic anion-transporting peptide (OATP) substrates on OATP1B1, OATP2B1, and OATP1B3. Drug Metab Dispos. 2007;35:1308-1314. Grant JA, Gallardo MA, Pickup BT. A fast method of molecular shape comparison: a simple application of a Gaussian description of molecular shape. J Comput Chem. 1996;17:1653-1666. Hawkins PCD, Skillman AG, Nicholls A. Comparison of shape-matching and docking as virtual screening tools. J Med Chem. 2007;50:74-82. SIMCA-Pþ 12.0.1; Umetrics AB, Umeå, Sweden. Available at: http://umetrics. com/. Accessed August 6, 2016. Wood DJ, Buttar D, Cumming JG, Davis AM, Norinder U, Rodgers SL. Automated QSAR with a hierarchy of global and local models. Mol Inform. 2011;30: 960-972. Weininger D. SMILES, a chemical language and information system. 1. Introduction to methodology and encoding rules. J Chem Inf Comput Sci. 1988;28: 31-36. Kenny PW, Montanari CA, Prokopczyk IM, Sala FA, Rodrigues Sartori G. Automated molecule editing in molecular design. J Comput Aided Mol Des. 2013;27: 655-664. OpenEye Scientific Software, 9 Bisbee Court, Suite D, Santa Fe, NM 87508; Available at: http://www.eyesopen.com/. Accessed August 6, 2016. Mills JEJ, Dean PM. Three-dimensional hydrogen-bond geometry and probability information from a crystal survey. J Comput Aided Mol Des. 1996;10: 607-622.
Contents lists available at ScienceDirect
Journal of Pharmaceutical Sciences journal homepage: www.jpharmsci.org
Pharmacokinetics, Pharmacodynamics and Drug Transport and Metabolism
Modeling Organic Anion-Transporting Polypeptide 1B1 Inhibition to Elucidate Interaction Risks in Early Drug Design Ismael Zamora 1, Susanne Winiwarter 2, * 1 2
Lead Molecular Design, S.L., Sant Cugat del Vall es, Spain €lndal, Sweden Drug Safety and Metabolism, AstraZeneca R&D Gothenburg, Mo
a r t i c l e i n f o
a b s t r a c t
Article history: Received 10 May 2016 Revised 23 June 2016 Accepted 12 July 2016 Available online 12 August 2016
The importance of transporter proteins for the disposition of drugs has become increasingly apparent during the past decade. A noted drug-drug interaction risk is the inhibition of organic anion-transporting polypeptides (OATPs), key transporters for the liver uptake of the widely used statins. We show here the development of a ligand-based in silico model for interaction with OATP1B1, an important representative of the OATP family. The model is based on a structural overlay of 6 known OATP1B1 inhibitors. A data set of about 150 compounds with published OATP1B1 inhibition data was compared to the resulting “transportophor,” and a similarity threshold was defined to distinguish between active and inactive molecules. In addition, using a statistical model based on physicochemical properties of the compounds as prefilter was found to enhance the overall predictivity of the model (final accuracy 0.73, specificity 074, and sensitivity 0.71, based on 126 compounds). The combined model was validated using an in-house data set (accuracy, specificity, and sensitivity were 0.63, 0.59, and 0.78, respectively; 62 compounds). The model gives also a structural overlay to the most similar template enabling visualization of where a change in a given structure might reduce the interaction with the transporter. © 2016 American Pharmacists Association®. Published by Elsevier Inc. All rights reserved.
Keywords: structure property relationships molecular modeling organic anion-transporting polypeptide transporters drug interactions
Introduction During the past decade, the importance of transporter proteins for the uptake of drugs into various tissues, and the resulting potential for drug-drug interactions, has become more visible.1-6 The continuous effort in drug design to find metabolically stable compounds with high affinity to the target under consideration is likely driving this development. A goal during drug design is to aim for compounds with high lipophilic ligand efficiency,7 that is, compounds that have a high-specific affinity to the target but as low lipophilicity as possible to reduce unspecific binding to other targets including metabolizing enzymes. Metabolically stable compounds
Abbreviations used: 2D, two dimensional; 3D, three dimensional; DDI, drugdrug interaction; HEK cells, human embryonic kidney cells; IC50, inhibition concentration to give 50% inhibition (i.e., half maximal inhibitory concentration); OATP, organic anion-transporting polypeptide; PCA, principal component analysis; PLS, projection to latent structures (also known as partial least squares analysis); r2, coefficient of determination; r2(CV), cross-validated coefficient of determination; TS score, transportophor similarity score. This article contains supplementary material available from the authors by request or via the Internet at http://dx.doi.org/10.1016/j.xphs.2016.07.011. * Correspondence to: Susanne Winiwarter (Telephone: þ46 31 7064913; Fax: þ46 31 7763700). E-mail address: [email protected] (S. Winiwarter).
cannot depend on metabolic clearance for their elimination from the body but require other mechanisms including active transport.8 Moreover, such compounds may also rely on uptake via special transporter proteins for being distributed to the tissue of interest because more hydrophilic compounds are less likely to cross lipophilic membranes through a purely passive mechanism. Organic anion-transporting polypeptides (OATPs) are transporters at the basolateral side of hepatocytes. They have been shown to be of importance for liver uptake and thereby for clearance of endogenous acidic compounds, such as bile acids or bilirubin.9-11 This pathway is also used by certain amphiphilic drugs. For example, 3-hydroxy-3-methylglutaryl-coenzyme A reductase inhibitors, also known as statins, make use of OATPs to reach their target within the hepatocyte. In addition, transport via OATPs ensures clearance from the blood and thereby reduces the risk of toxic side effects.12 Inhibition of OATPs can therefore lead to interaction with a drug's disposition, its safety risks, and action. New pharmaceutics likely to be comedicated with statins need to avoid any interaction with OATPs.13 Successful drug design in the cardiovascular area, where statins are considered standard of care,14 therefore depends on a good understanding of the structure-transporter activity relationship early on the drug discovery process. OATP1B1 inhibition potential can be measured in vitro through determining how much the transport of a known OATP1B1
http://dx.doi.org/10.1016/j.xphs.2016.07.011 0022-3549/© 2016 American Pharmacists Association®. Published by Elsevier Inc. All rights reserved.
I. Zamora, S. Winiwarter / Journal of Pharmaceutical Sciences 105 (2016) 3214-3220
substrate is hindered by addition of the query compound using OATP1B1-transfected cell lines.15,16 Fluorescent marker substrates have been suggested for utilization in high throughput screening.17-20 Knowledge of the substrate used is important to understand which data are comparable because inhibition constants vary depending on the probe substrate used.9,16,21 Specific drugs known to be OATP1B1 substrates or estradiol-17b-glucuronide have been suggested as most relevant probe substrates for drug-drug interaction risk assessment.4,21 In early drug discovery, and to more actively guide compound design, in silico models can be used. Various approaches to model transporter interactions have been reported,22-25 including an increasing number of models for OATP1B1 interactions: In 2005, Chang et al.26 defined a pharmacophore for OATP1B1 substrates consisting of 2 hydrogen bond acceptors, 1 hydrogen bond donor and a central lipophilic region, based on literature data for human and the rat. Inhibitor models for OATP1B1 most often use quantitative structure activityerelationship approaches.15,16,27-31 Lipophilicity and hydrogen bonding were found to be important molecular properties that influence OATP1B1 inhibition in most of these studies.28,30 Using comparative molecular field analysis, Gui et al.32 analyzed 18 diverse inhibitors and proposed an interaction mode with a central core and peripheral polar regions, one of which is acidic, as structural features that promote interactions with the transporter. It was discussed that pharmacophore models might be difficult to obtain for diverse sets of compounds30 since the varying inhibition constants depending on the substrate used,9 as well as biphasic binding kinetics for example in the case of estrone-3sulfate,33,34 indicate the possibility of different binding sites. The aim of the present work was the development of a fast ligand-based in silico model for inhibition of OATP1B1 that can be used to guide drug design in early drug discovery. We used a rapid pharmacophore type approach, alignment of molecular shape,35,36 as primary method, based on a small set of established OATP1B1 inhibitors. Note that these compounds also can be transported by the protein and most likely act as competitive inhibitors. Comparing query structures to this alignment, designated “transportophor,” resulted in both a similarity score and the best structural fit. The similarity score was able to indicate whether a compound had low or high risk for OATP1B1 inhibition, and a
3215
similarity score threshold was defined. The best structural fit showed which parts of a query structure were aligned with the features of a known inhibitor, thereby indicating which part of the molecule may be important for the interaction with the transporter. However, we found that a prefilter based on physicochemical descriptors, using the projection to latent structure (PLS) method, clearly enhanced the prediction quality, as shown in the external validation using in-house compounds. Materials and Methods Compound Sets Three different compound data sets were used in this study. Data set 1 consists of 6 known OATP1B1 inhibitors (estrone-3sulfate, estradiol-17b-glucuronide, taurocholic acid, rosuvastatin, pitavastatin, and pravastatin)15,26 and was used to define the alignment rule for the transportophor generation (see Fig. 1). Three compounds belong to the class of endogenous steroids and the remaining are statins, that is, compounds of particular interest in the cardiovascular research area. The activity for the selected compounds ranges from 60% inhibition for pravastatin and rosuvastatin to >90% for pitavastatin. All 6 compounds have a negatively charged moiety, a lipophilic core and at least 1 hydrogen bond acceptor. The negatively charged moiety is not always a carboxylic acid but can be any acid isostere. Data set 2 comprises 146 compounds reported by Karlgren et al.15 and was used for determination of the similarity threshold to discriminate between active and inactive compounds through comparison to the transportophor. The data set was also used as training set for the PLS model based on physicochemical parameters. Inhibition of OATP1B1-mediated transport had been measured as described in human embryonic kidney cells transfected with OATP1B1, using estradiol-17b-glucuronide (0.5 mM) as substrate. The test compound concentration was 20 mM.15 Data set 3 is a set of 63 AstraZeneca proprietary compounds that was used for validation. OATP1B1 inhibition measurements for these compounds were performed in human embryonic kidney cells transfected with OATP1B1 as described by Soars et al.,16 using estradiol-17b-glucuronide as substrate. The compounds were
Figure 1. Six known OATP1B1 inhibitors: upper row: estrone-3-sulfate, estradiol-17b-glucuronide, taurocholic acid; lower row: rosuvastatin, pitavastatin, pravastatin.
3216
I. Zamora, S. Winiwarter / Journal of Pharmaceutical Sciences 105 (2016) 3214-3220
measured at 6 concentrations (ranging typically from 0.3 to 100 mM). Activity classification was based on the % inhibition value at 30 mM.
compound was found, they were aligned with ROCS using estrone3-sulfate as template to give a consistent coordinate space for all solutions. These 6 aligned structures were termed transportophor.
Multivariate Analysis Threshold Definition Principal component analysis (PCA) and PLS analysis were performed using Simca with default options.37 The descriptor set used was an in-house standard descriptor set, consisting of 196 2D- and 3D-physicochemical descriptors (AZ descriptors38). For the PLS model, the % inhibition at 10 mM was transformed into a categorical activity value for the compounds from data set 2 as follows: % inh <30%: 0; 30%-40%: 1; 40%-50%: 2, 50%-60%: 3; 60%-70%: 4; 70%-80%: 5; 80%-90%:6; >90%: 7. This categorical activity was used as y-variable. Transportophor Generation Ligand Preparation 2D representations (SMILES code39) of the 6 selected compounds were used as starting point. The ionization state was adjusted to pH 7.4, using OEleatherface (an in-house adaption of leatherface40 for the OpenEye toolkit41). The structures were converted to 3D via Corina (Molecular Networks) and subjected to energy minimization using Szybki (OpenEye software).41 Finally a conformational analysis was performed using Omega (OpenEye),41 ensuring unique titles for each conformer (option “-warts”). A maximum of 200 conformations per compound was considered. Ligand Superposition ROCS (OpenEye)41 was used for ligand superposition including the ImplicitMillsDean color force field42 as part of the overlay optimization (option “optchem”). All conformations of each of the 6 template molecules were superimposed to all conformations of the other 5 templates. The conformation most often showing the highest similarity score, using the “RefTverskyCombo” score as provided by ROCS, was selected as overall best fitting conformation for the specific template. Once the conformation for each
Data set 2 compounds were classified as active or inactive according to the reported inhibition at 20 mM, using 50% inhibition as cutoff.15 Ligand Preparation The compounds were subjected to the same ligand preparation procedure as the 6 template compounds. For compounds with ambiguous ionic state definitions, all potential ionic states were considered. Compounds that produced any error in this procedure were removed from the automatic analysis. Alignment to Transportophor and Threshold Definition All conformations of each molecule were used for alignment to each of the 6 structures comprising the transportophor using the same procedure as mentioned previously, that is, using ROCS considering the ImplicitMillsDean color force field. For each conformation, the 6 similarity score values (RefTverskyCombo score) obtained from the overlay with each of the 6 template structures were summed up to give the “sum of scores.” The highest value found for any of the conformations, or ionic forms, of a molecule defined the “transportophor similarity score” (TS score) for this molecule. The TS score where 75% of the actives could be identified was selected as primary threshold value. A secondary, lower, value was chosen to define a medium class. Validation Data set 3 was used for validation only. The compounds were classified according to their measured inhibition at 30 mM. Compounds with an inhibition <40% were deemed as noninhibitors, whereas compounds with an inhibition >60% were considered as
Figure 2. PCA score plot based on data set 2 (circles, data set 1 compounds shown as squares). ReddOATP1B1 inhibitors (dark to light correspond to categories 7, >90% inhibition, to 4, 50%-60% inhibition); greendnoninhibitors (dark to light correspond to categories 0, <30% inhibition to 3, 40%-50% inhibition); gray line indicates approximate discrimination line between inhibitors (above) and noninhibitors (below). PC 1 describes mainly molecular size, whereas PC 2 can be related to the lipophilicity of the compounds.
I. Zamora, S. Winiwarter / Journal of Pharmaceutical Sciences 105 (2016) 3214-3220
inhibitors. Compounds with an inhibition between 40% and 60% were regarded as possible inhibitors. The compounds were subjected to ligand preparation and alignment to transportophor as described previously. The resulting TS score was then used to define in which of the 3 categories the compound would fall based on the threshold values defined previously. In addition, the PLS model was used to predict the likelihood of the compound's being an OATP1B1 inhibitor, through estimation of the “activity category,” used as y-variable in the model. Compounds with a predicted activity category below 2 were considered as inactive, whereas compounds with a predicted activity category above 4 were considered as active. Classification statistics were obtained for each of the 2 models but also for the combination model.
3217
Figure 3. Template compound alignment (estrone-3-sulfate [black], estradiol17b-glucuronide [light gray], taurocholic acid [dark gray], rosuvastatin [light green], pitavastatin [green], and pravastatin [dark green]).
Results and Discussion In 2005, Chang et al.15,26 published an analysis of OATP1B1 substrate data and reported a pharmacophore model for both human OATP1B1 and rat oatp1a1 transporter substrates, indicating that interaction with OATP1B1 required several hydrogen-bonding features in the periphery of the molecule and a lipophilic central region. Here, we attempt to generate a pharmacophore type model for OATP1B1 inhibition. Six reasonably diverse OATP1B1 inhibitors15 were selected for this purpose. Figure 2 shows the score plot resulting from a PCA of data set 2 based on physicochemical descriptors. A natural separation between the active and the inactive molecules is noted, which is in line with literature.15 However, the used physicochemical properties do not explain all data because the discrimination is less successful in the central part of the plot. The 6 selected compounds, estrone-3-sulfate, estradiol-17b-glucuronide, taurocholic acid, rosuvastatin, pitavastatin, and pravastatin, can be found in this border area, where the physicochemical properties do not seem to be sufficient to distinguish between actives and inactives. Using a pharmacophore approach should be able to help identify OATP1B1 inhibitors based on their structural similarity instead. Furthermore, the 6 selected compounds represent 2 important compound classes: endogenous steroidal acids and statins, considered as standard of care in treating cardiovascular disease. Transportophor Generation The 6 selected OATP1B1 inhibitors were subjected to conformational analysis as described in Materials and Methods. The ionized form was used for all compounds as the predominant form at pH 7.4. The compounds were flexible except estrone-3-sulfate and estradiol-17b-glucuronide for which only 5 and 16 conformations were generated, respectively. Table 1 shows the number of conformations, the top-ranked conformation and how often it was ranked first, and the averaged RefTverskyCombo similarity score value for the top-ranked conformation for each of the 6
compounds. Alignment of these conformations based on estrone-3sulfate resulted in a depiction of the generated transportophor (see Fig. 3). The main structural features are a peripheral hydrophilic, acidic functionality, and a big lipophilic part with only few heteroatoms, which is consistent with published OATP1B1 pharmacophores.26,32 In Figure 3, this alignment is shown with the acidic substructures on the left-hand side opposite to the lipophilic core of the molecules. Note, that both taurocholic acid (dark gray) and estradiol-17b-glucuronide (light gray) extend the space used by the other 4 compounds in this region. Threshold Definition The similarity score gives an estimation of how similar a compound is to the transportophor. By checking the similarity score for a set of compounds with known inhibition activity, a threshold value could be defined to distinguish between active and inactive molecules. Data set 2 compounds15 were subjected to a similar procedure as the transportophor template molecules: ionization status of the compounds was adjusted to pH 7.4. All conformations, anddif applicabledlikely ionic states at pH 7.4, identified for a compound were aligned to the 6 transportophor structures. The highest “sum of scores” for any of the conformations (forms) defined the TS score for the query molecule. For 18 compounds, no results could be obtained, due to errors in the automatic procedure (see Supporting Information). Removing also the 6 template compounds left 122 compounds for threshold generation
Table 1 Transportophor Generation Results Compound
Number of Selected Conformer Average Score of Selected Conformer Conformers (Number of (RefTverskyCombo) Times Shown as First Option)
Estrone-3-sulfate Estradiol-17bglucuronide Taurocholic acid Rosuvastatin Pitavastatin Pravastatin
5 16
c4 (220) c14 (107)
0.89 1.05
200 200 200 200
c101 c165 c133 c165
1.01 1.10 1.10 1.17
(38) (39) (36) (52)
Figure 4. Cumulative % of active (black; n ¼ 49) and inactive compounds (gray; n ¼ 73) plotted against the TS scores. Vertical lines indicate primary (full) and secondary (dashed) threshold.
3218
I. Zamora, S. Winiwarter / Journal of Pharmaceutical Sciences 105 (2016) 3214-3220
Figure 5. Calculated versus experimental activity using the PLS model, color by experimental activity (black >60% inhibition, i.e., categories 4-7; gray 40%-60% inhibition, i.e., categories 2 and 3; and light gray <40% inhibition, i.e., categories 0 and 1), shape by calculated activity category (stars: high, calculated activity category >4; diamonds: medium, calculated activity categories 2-4; circles: low, calculated activity category <2).
(49 inhibitors and 73 noninhibitors). Previously, both, using only the single highest score and different similarity indices provided by ROCS were tested. As result the “sum of scores” considering the RefTverskyCombo score was found most useful to define a TS score (data not shown). Figure 4 shows the cumulative % of inhibitors and/or noninhibitors versus the TS score, that is, the percentage of active or inactive compounds with a value lower than a given score was plotted against the score. It can be seen that 75% of the inhibitors in this set had a TS score of at least 6.5, which was used as upper threshold. About 30% of the inactive compounds also had a TS score above the threshold (see Fig. 4). An additional threshold of 6.0 was used to define a medium class. As result, only ~10% of the actives are falsely defined as inactive. The model seems to be able to distinguish reasonably well between active and inactive compounds in this data set, with an overall accuracy (true predictions/ all compounds) of 0.58. Although the sensitivity (true predicted inhibitors/all experimental inhibitors) was 0.78, the specificity (true predicted noninhibitors/all experimental noninhibitors) was only 0.45, with about 21% of the noninhibitors classified as medium. Less than 20% of the compounds were classified as medium. PLS Model as Prefilter As observed in the PCA score plot (Fig. 2), there is a natural discrimination between the 2 classes of compounds in physicochemical space. One can reason that a compound simply needs to fulfill certain physicochemical prerequisites to be able to reach the transporter. Only then the structural complementarity will be relevant to understand whether a compound really is an inhibitor or not. Thus, we defined a PLS model based on physicochemical
descriptors, which is suggested to be used before the comparison to the transportophor. Data set 2 compounds were used as training set in the PLS analysis. The compounds were assigned an activity category value based on their inhibition potential as described in the Methods section, essentially enabling a discriminative analysis with eight classes using the PLS technique. This hybrid approach has less emphasis on each single value but uses still more of the available information than a two-class model. The derived PLS model consists of 3 components (r2 ¼ 0.53; r2(CV) ¼ 0.44). Dividing the calculated category values into 3 bins, values <2 considered as low, values between 2 and 4 as medium, and values >4 as high, a classification accuracy of 0.57, sensitivity of 0.52, and specificity of 0.6 were achieved. Note that about 40% of the compounds ended in the middle bin, where no distinction between active or inactive compounds was possible (see Fig. 5). Using the PLS model as prefilter by only considering the transportophor results for compounds that were predicted as high or medium in the PLS model, the total accuracy and specificity of the model could be enhanced to 0.73 and 0.74, respectively. In addition, less than 10% of the compounds were predicted as medium. The approach caused a slight reduction of the sensitivity (0.71 instead of 0.78). Note that these numbers are based on the compounds used as training set in the PLS model. Model Validation An in-house data set of 63 compounds was used for external validation of the models. The compounds had been tested as part of the screening cascade in cardiovascular projects, which may account for the bias in the set: more than 60% of these compounds showed OATP1B1 inhibition and only about 25% were inactive (38 actives, >60% inhibition at 30 mM, 17 inactive, <40% inhibition, and 8 compounds with medium inhibition potential). A similar classification can be obtained based on IC50 values, with 20 and 50 mM as IC50 limits. To remain consistent with the published data used to build the model, % of inhibition was used as criteria for the classification. The models were applied independently and in a combination approach where the PLS model was used as pre-filter (see Table 2). Note that 4 compounds gave no results in the transportophor model. It can be easily seen that the transportophor model showed many false positives, indicating that most of the molecules had the structural features to interact with the transporter. Using the PLS model as prefilter, the overall specificity of the model went from 0.18 to 0.59, whereas the sensitivity was reduced from 0.86 to 0.78. Less than 10% of the compounds were found in the middle class in both cases. The PLS model by itself gave an accuracy of 0.52, and a sensitivity and specificity of 0.53, with almost 40% of the compounds in the middle class. The use of 2 independent computational approaches clearly enhances the model predictivity but also the possibility to understand the interactions with the transporter. The PLS model can explain whether a compound has the right overall physicochemical properties, whereas the 3D model adds information on specific structural features that might be important for the interaction.
Table 2 Model Comparison for 63 In-House Test Set Compounds Model
# Cmpds
Accuracy
Sensitivity
Specificity
Pos Prec
Neg Prec
%Pred Medium
Transportophor PLS Combo
59 63 62
0.56 0.52 0.63
0.86 0.53 0.78
0.18 0.53 0.59
0.59 0.91 0.73
0.75 0.53 0.56
7% 38% 6%
# Cmpds, number of compounds considered in specific model; Accuracy, (TH þ TL)/total number of compounds; Sensitivity, TH/AH; Specificity, TL/AL; Pos Prec, positive Precision (TH/APH); Neg Prec, negative precision (TL/APL); %Pred Medium, % predicted medium (APM/total number of compounds); TH, true high; TL, true low; AH, all experimentally high; AL, all experimentally low; APH, all predicted high; APL, all predicted low; APM, all predicted medium.
I. Zamora, S. Winiwarter / Journal of Pharmaceutical Sciences 105 (2016) 3214-3220
3219
Figure 6. Shape-based superposition of OATP1B1 inhibitor repaglinide (gray) on estrone-3-sulfate (green); repaglinide is an OATP1B1 inhibitor (88% inhibition at 20 mM); 2D structure of repaglinide shown to the right.
For drug design, it is essential to understand how a molecular structure should be altered. In some cases, physicochemical properties can be changed so as to reduce the compounds ability to reach the transporter. In other cases, it may be necessary to modify the compound more based on the direct interaction points with the transporter. Thus, the visualization of the structural overlay with the most similar template provided by the transportophor model will be useful. Figure 6 shows, as example, the overlay of repaglinide, a known OATP1B1 inhibitor included in data set 2, with estrone-3-sulfate to investigate how the molecule best fits the template structure. Such graphical information will help to identify molecular regions where changes could disrupt the interaction with the transporter. Repaglinide is a carboxylic acid with very few polar features in the remaining molecule. The carboxylic acid superimposes nicely with the sulfate group in estrone-3-sulfate, whereas the rest of the molecule fits well with the steroid skeleton. Provided that the carboxylic moiety is required for any new molecule, one could consider changes in the more lipophilic part, for example, addition of an exposed hydrophilic group in the isobutyl side chain. Repaglinide is regarded as inhibitor, both in the PLS and the transportophor model. Conclusion We present here an OATP1B1 inhibition in silico model based on two orthogonal computational methods, considering both the physicochemical properties and the structure similarity to known inhibitors. The actual mechanism on how a molecule inhibits the transporter remains unknown, but the compounds need to reach the site of interaction and interact in a certain manner to be able to inhibit the transport of a substrate compound. To describe the physicochemical behavior of the compound, a PLS model based on this type of descriptors was built in a similar approach as reported by Karlgren et al.15 Moreover, to describe the potential interaction in the cavity, an alignment rule, the transportophor, and a 3D similarity criterion were defined to discriminate which compounds can actually inhibit the transport of another compound. The compounds used to create the transportophor presented were selected to fit the specific needs of drug discovery projects in the cardiovascular research area. The model can be easily adapted to use other templates if the chemical space of interest is elsewhere. The combination of both approaches (PLS and 3D similarity) gave a better discriminative result than the application of just one model to describe the transport inhibition. The ability to visualize the structure based overlay is anticipated to support compound design
by elucidating potentially interacting structural features. The model is available for drug projects within AstraZeneca.
References 1. Endres CJ, Hsiao P, Chung FS, Unadkat JD. The role of transporters in drug interactions. Eur J Pharm Sci. 2006;27:501-517. 2. Zhang L, Strong JM, Qiu W, Lesko LJ, Huang S. Scientific perspectives on drug transporters and their role in drug interactions. Mol Pharm. 2006;3: 62-69. 3. Zhang L, Zhang Y, Zhao P, Huang S. Predicting drug-drug interactions: an FDA perspective. AAPS J. 2009;11:300-306. 4. Giacomini KM, Huang S, Tweedie DJ. Membrane transporters in drug development. Nat Rev Drug Discov. 2010;9:215-236. 5. Müller F, Fromm MF. Transporter-mediated drug-drug interactions. Pharmacogenomics. 2012;12:1017-1037. 6. Giacomini KM, Huang S. Transporters in drug development and clinical pharmacology. Clin Pharmacol Ther. 2013;94:3-9. 7. Leeson PD, Springthorpe B. The influence of drug-like concepts on decisionmaking in medicinal chemistry. Nat Rev Drug Discov. 2007;6:881-890. 8. Wu C, Benet LZ. Predicting drug disposition via application of BCS: transport/ absorption/elimination interplay and development of a biopharmaceutics drug disposition classification system. Pharm Res. 2005;22:11-23. 9. Niemi M, Pasanen MK, Neuvonen PJ. Organic anion transporting polypeptide 1B1: a genetically polymorphic transporter of major importance for hepatic drug uptake. Pharmacol Rev. 2011;63:157-181. 10. Shitara Y, Maeda K, Ikejiri K, Yoshida K, Horie T, Sugiyama Y. Clinical significance of organic anion transporting polypeptides (OATPs) in drug disposition: their roles in hepatic clearance and intestinal absorption. Biopharm Drug Dispos. 2013;34:45-78. 11. Fenner KS, Jones HM, Ullah M, et al. The evolution of the OATP hepatic uptake transport protein family in DMPK sciences: from obscure liver transporter to key determinants of hepatobiliary clearance. Xenobiotics. 2012;42:28-45. 12. Elsby R, Hilgendorf C, Fenner K. Understanding the critical disposition pathways of statins to assess drug-drug interaction risk during drug development: it's not just about OATP1B1. Clin Pharmacol Ther. 2012;92:584-598. 13. Sharma P, Butters CJ, Smith V, Elsby R, Surry D. Prediction of the in vivo OATP1B1-mediated drug-drug interaction potential of an investigational drug against a range of statins. Eur J Pharm Sci. 2012;47:244-255. 14. Grundy SM, Cleeman JI, Bairey Merz CN, Brewer Jr HB, Clark LT. Implications of recent clinical trials for the National Cholesterol Education Program Adult Treatment Panel III guidelines. Circulation. 2004;110:227-239. €m CAS, Svensson R, Palm J, Artursson P. In vitro 15. Karlgren M, Ahlin G, Bergstro and in silico strategies to identify OATP1B1 inhibitors and predict clinical drugdrug interactions. Pharm Res. 2012;29:411-426. 16. Soars MG, Barton P, Ismair M, Jupp R, Riley RJ. The development, characterization, and application of an OATP1B1 inhibition assay in drug discovery. Drug Metab Dispos. 2012;40:1641-1648. 17. Annaert P, Ye ZW, Stieger B, Augustijns P. Interaction of HIV protease inhibitors with OATP1B1, 1B3, and 2B1. Xenobiotica. 2010;40:163-176. 18. Bednarczyk D. Fluorescence-based assays for the assessment of drug interaction with the human transporters OATP1B1 and OATP1B3. Anal Biochem. 2010;405:50-58. 19. DeBruyn T, Fattah S, Stieger B, Augustijns P, Annaert P. Sodium fluorescein is a probe substrate for hepatic drug transport mediated by OATP1B1 and OATP1B3. J Pharm Sci. 2011;100:5018-5030.
3220
I. Zamora, S. Winiwarter / Journal of Pharmaceutical Sciences 105 (2016) 3214-3220
20. Gui C, Obaidat A, Chaguturu R, Hagenbuch B. Development of a cell-based highthroughput assay to screen for inhibitors of organic anion transporting polypeptides 1B1 and 1B3. Curr Chem Genomics. 2010;4:1-8. 21. Izumi S, Nozaki Y, Maeda K, et al. Investigation of the impact of substrate selection on in vitro organic anion transporting polypeptide 1B1 inhibition profiles for the prediction of drug-drug interactions. Drug Metab Dispos. 2015;43:235-247. 22. Chang C, Ray A, Swaan P. In silico strategies for modeling membrane transporter function. Drug Discov Today. 2005;10:663-671. 23. Winiwarter S, Hilgendorf C. Modeling of drug-transporter interactions using structural information. Curr Opin Drug Discov Devel. 2008;11:95-103. 24. Ekins S, Polli JE, Swaan PW, Wright SH. Computational modeling to accelerate the identification of substrates and inhibitors for transporters that affect drug disposition. Clin Pharmacol Ther. 2012;92:661-665. 25. Artursson P, Matsson P. Computational prospecting for drug-transporter interactions. Clin Pharmacol Ther. 2013;94:30-32. 26. Chang C, Pang KS, Swaan PW, Ekins S. Comparative pharmacophore modeling of organic anion transporting polypeptides: a meta-analysis of rat Oatp1a1 and human OATP1B1. J Pharmacol Exp Ther. 2005;314:533-541. 27. Karlgren M, Vildhede A, Norinder U, et al. Classification of inhibitors of hepatic organic anion transporting polypeptides (OATPs)dinfluence of protein expression on drug-drug interactions. J Med Chem. 2012;55:4740-4763. 28. DeBruyn TM, vanWesten GJP, IJzerman AP, et al. Structure-based identification of OATP1B1/3 inhibitors. Mol Pharmacol. 2013;83:1257-1267. 29. Cao D, Liu S, Fan L, Liang Y. QSAR analysis of the effects of OATP1B1 transporter by structurally diverse natural products using a particle swarm optimizationcombined multiple linear regression approach. Chemometrics Intell Lab Syst. 2014;130:84-90. 30. van de Steeg E, Venhorst J, Nooijen IHG, DeGroot J, Wortelboer HM, Vlaming MLH. Generation of Bayesian prediction models for OATP-mediated drug-drug interactions based on inhibition screen of OATP1B1, OATP1B1*15 and OATP1B3. Eur J Pharm Sci. 2015;70:29-36. €ger W, Ecker GF. Identification of novel 31. Kotsampasakou E, Brenner S, Ja inhibitors of organic anion transporting polypeptides 1B1 and 1B3 (OATP1B1
32.
33.
34.
35.
36. 37. 38.
39.
40.
41. 42.
and OATP1B3) using a consensus vote of six classification models. Mol Pharm. 2015;12:4395-4404. Gui C, Wahlgren B, Lushington GH, Hagenbuch B. Identification, Ki determination and CoMFA analysis of nuclear receptor ligands as competitive inhibitors of OATP1B1-mediated estradiol-17b-glucuronide transport. Mol Pharmacol. 2009;60:50-56. Tamia I, Nozawa T, Koshida M, Nezu J, Sai Y, Tsuji A. Functional characterisation of human organic anion transporting polypeptide b (OATP-b) in comparison with liver-specific OATP-C. Pharm Res. 2001;18:1262-1269. J, Portmann R, Brun M, Funk C. Substrate-dependent drug-drug interactions Noe between gemfibrozil, fluvastatin and other organic anion-transporting peptide (OATP) substrates on OATP1B1, OATP2B1, and OATP1B3. Drug Metab Dispos. 2007;35:1308-1314. Grant JA, Gallardo MA, Pickup BT. A fast method of molecular shape comparison: a simple application of a Gaussian description of molecular shape. J Comput Chem. 1996;17:1653-1666. Hawkins PCD, Skillman AG, Nicholls A. Comparison of shape-matching and docking as virtual screening tools. J Med Chem. 2007;50:74-82. SIMCA-Pþ 12.0.1; Umetrics AB, Umeå, Sweden. Available at: http://umetrics. com/. Accessed August 6, 2016. Wood DJ, Buttar D, Cumming JG, Davis AM, Norinder U, Rodgers SL. Automated QSAR with a hierarchy of global and local models. Mol Inform. 2011;30: 960-972. Weininger D. SMILES, a chemical language and information system. 1. Introduction to methodology and encoding rules. J Chem Inf Comput Sci. 1988;28: 31-36. Kenny PW, Montanari CA, Prokopczyk IM, Sala FA, Rodrigues Sartori G. Automated molecule editing in molecular design. J Comput Aided Mol Des. 2013;27: 655-664. OpenEye Scientific Software, 9 Bisbee Court, Suite D, Santa Fe, NM 87508; Available at: http://www.eyesopen.com/. Accessed August 6, 2016. Mills JEJ, Dean PM. Three-dimensional hydrogen-bond geometry and probability information from a crystal survey. J Comput Aided Mol Des. 1996;10: 607-622.