Technically Extended MultiParameter Optimization (TEMPO): An

Subscriber access provided by UNIVERSITY OF LEEDS

Article

TEMPO (Technically Extended Multiparameter Optimization): An Advanced Robust Scoring Scheme To Calculate Central Nervous System (CNS) Druggability And Monitor Lead Optimization Arup K. Ghose, Gregory R. Ott, and Robert L Hudkins ACS Chem. Neurosci., Just Accepted Manuscript • DOI: 10.1021/acschemneuro.6b00273 • Publication Date (Web): 14 Oct 2016 Downloaded from http://pubs.acs.org on October 18, 2016

Just Accepted “Just Accepted” manuscripts have been peer-reviewed and accepted for publication. They are posted online prior to technical editing, formatting for publication and author proofing. The American Chemical Society provides “Just Accepted” as a free service to the research community to expedite the dissemination of scientific material as soon as possible after acceptance. “Just Accepted” manuscripts appear in full in PDF format accompanied by an HTML abstract. “Just Accepted” manuscripts have been fully peer reviewed, but should not be considered the official version of record. They are accessible to all readers and citable by the Digital Object Identifier (DOI®). “Just Accepted” is an optional service offered to authors. Therefore, the “Just Accepted” Web site may not include all articles that will be published in the journal. After a manuscript is technically edited and formatted, it will be removed from the “Just Accepted” Web site and published as an ASAP article. Note that technical editing may introduce minor changes to the manuscript text and/or graphics which could affect content, and all legal disclaimers and ethical guidelines that apply to the journal pertain. ACS cannot be held responsible for errors or consequences arising from the use of information contained in these “Just Accepted” manuscripts.

ACS Chemical Neuroscience is published by the American Chemical Society. 1155 Sixteenth Street N.W., Washington, DC 20036 Published by American Chemical Society. Copyright © American Chemical Society. However, no copyright claim is made to original U.S. Government works, or works produced by employees of any Commonwealth realm Crown government in the course of their duties.

Page 1 of 31

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

ACS Chemical Neuroscience

TEMPO (Technically Extended MultiParameter Optimization): An Advanced Robust Scoring Scheme To Calculate Central Nervous System (CNS) Druggability And Monitor Lead Optimization

Arup K. Ghose,* Gregory R. Ott and Robert L. Hudkins

Discovery and Product Development, Teva Branded Pharmaceutical Products R&D, Inc., 145 Brandywine Parkway, West Chester, PA 19380, USA 145 Brandywine Parkway, West Chester, PA 19380, USA

ACS Paragon Plus Environment

1


1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 2 of 31

ABSTRACT

At the discovery stage, it is important to understand the drug design concepts for a CNS drug compared to a non-CNS drug. Previously, we published on ideal CNS drug property space and defined in detail the physicochemical property distribution of CNS versus nonCNS oral drugs, the application of radar charting (a graphical representation of multiple physicochemical properties used during CNS lead optimization) and a recursive partition classification tree to differentiate between CNS- and non-CNS drugs. The objective of the present study was to further understand the differentiation of physicochemical properties between CNS and non-CNS oral drugs by the development and application of a new CNS scoring scheme, TEMPO (Technically Extended MultiParameter Optimization). In this multiparameter method we identified eight key physicochemical properties critical for accurately assessing CNS druggability: (1) number of basic amines, (2) carbon-hetero atom (non-Carbon, non-Hydrogen) ratio, (3) number of aromatic rings, (4) number of chains, (5) number of rotatable bonds, (6) number of H-acceptors, (7) computed octanol/water partition coefficient, AlogP, and (8) number of nonconjugated Catoms in non-aromatic rings. Significant features of the CNS-TEMPO penalty score are the extension of the multiparameter approach to generate an accurate weight factor for each physicochemical property; the use of limits on both sides of the computed property space range during the penalty calculation; and the classification of CNS and non-CNS drug scores. CNS-TEMPO significantly outperformed CNS-MPO and the Schrödinger QikProp CNS parameter (QP_CNS) in evaluating CNS drugs and has been extensively applied in support of CNS lead optimization programs.


2

Page 3 of 31

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


Table of Content Figure

KEYWORDS: central nervous system (CNS), CNS-TEMPO, non-CNS, CNS drug discovery, CNS lead optimization, non-CNS drug discovery, non-CNS lead optimization, multiparameter optimization INTRODUCTION

The central nervous system (CNS), consisting of the brain and the spinal cord, is protected from injury by the skull, meninges, cerebrospinal fluid and the blood-brain barrier (BBB). Apart from these physical barriers, several additional mechanisms are in place to further protect the CNS. A detailed description of the BBB components was previously summarized in our work1 and in a review by Rankovic.2 Briefly, the BBB consists of three major components: endothelial cell tight junctions, an enzymatic barrier, and an active efflux barrier. Despite all these protections, the CNS may malfunction. The loss of CNS function increases considerably with age and disorders of the brain remain some of the most prevalent, devastating and yet poorly treated diseases. It was estimated that by 2050, the world population over the age of 65 will triple3 and health care expenditures will increase to about 29% of the gross domestic product.4 With an aging population costs for the treatment of neurodegenerative diseases, brain cancer, and stroke will escalate to the trillions of dollars. The incidence of Alzheimer’s disease (AD) alone


3


1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 4 of 31

has been projected to increase over 100% in the United States and Western Europe, and 300% in China, Latin America and India.5 Because of the projected increase in incidence and associated socioeconomic costs of CNS disorders, the development of innovative and effective CNS drugs remains an area of high interest for the pharmaceutical industry, and has the potential to not only provide significant improvements in quality of life, but also reduce the future economic burden on the healthcare system.2

Drug discovery in general has always faced multiple challenging hurdles, such as long development times, high costs, validation of novel drug targets with unproven therapeutic potential, lack of acceptable biomarkers and preclinical models, and the demonstration of clinical value. CNS drug discovery has the added hurdle of penetrating the blood brain barrier. As a result, reductions or elimination of CNS programs in the industry contributed to the limited progress in CNS drug development.6,7,8 Reducing attrition in the clinic by designing high-quality drug candidates that are able to reach and effectively modulate targets in the brain makes CNS research one of the more daunting challenges in drug discovery. At the discovery stage, it is critical to understand CNS vs. non-CNS drug design concepts. A number of researchers, including our group, have published on ideal CNS physicochemical property space in high-quality CNS drug candidates.1 We previously described in detail the physicochemical property space distribution of CNS and non-CNS oral drugs, and the application of radar charting, a graphical representation of multiple physicochemical properties used during CNS lead optimization.1 We also provided a recursive partition classification tree to differentiate between CNS and nonCNS drugs.


4

Page 5 of 31

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


In the current multiparameter method we identified eight physicochemical properties important for accurately assessing CNS druggability: (1) number of basic amines, (2) carbon-hetero atom (non-C, non-H) ratio, (3) number of aromatic rings, (4) number of chains, (5) number of rotatable bonds, (6) number of H-acceptors, (7) computed octanol/water partition coefficient, AlogP, and (8) number of nonconjugated C-atoms in non-aromatic rings. The objective of the present study was to increase our understanding of the differences between CNS and non-CNS oral drugs by the development and application of an effective CNS scoring scheme, TEMPO (Technically Extended MultiParameter Optimization).

A significant feature of the CNS-TEMPO method

compared with CNS-MPO9 was the extension of the multiparameter approach to generate positive weight factors and apply penalty limits on both sides of a computed property space (see Figure 1 and Figure 2). Wager et al.10 recently reevaluated MPO scoring and reported minor changes using the newer version of the ACDLabs logD and pKa calculator. Several weaknesses of the CNS-MPO scoring function were the application of limits on only the upper limits of several properties (e.g. molecular weight, clogP and clogD) that contributed to incorrectly scoring small molecules, computing clogP and clogD using different software (clogD is dependent on clogP and pKa), and not differentiating CNS from non-CNS drugs. Recently, a probabilistic approach to MPO (pMPO) was reported comparing a small set of CNS and non-CNS drugs using five parameters (topological polar surface area, hydrogen bond donor, molecular weight, clogD and basic pKa).11 In this paper we demonstrate that CNS-TEMPO significantly outperformed the CNS-MPO and Schrodinger QikProp (QP_CNS) methods in assessing CNS druggability space and discriminating CNS drugs from non-CNS drugs.


5


1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 6 of 31

RESULTS AND DISCUSSION

Preferred and qualified physicochemical property range. The steps involved in the model building and selection of the CNS and non-CNS drug sets were described in the Methods section. The physicochemical properties of each database were computed (using Pipeline Pilot (http://accelrys.com/products/pipeline-pilot/) and Schrödinger QikProp method (https://www.schrodinger.com/qikprop), followed by analysis and distribution of the various property spaces. Shown in Figure 1, for example, was the distribution of the computed octanol/water partition coefficient AlogP12 and hydrogen bond acceptor properties. The calculated properties displayed a distorted Gaussian type distribution for both CNS and non-CNS drugs, except in the case of the number of hydrogen bond donors. For approved CNS drugs, approximately 40% contained zero, 35% had one, and 20% had two hydrogen bond donors.1 The concept of preferred and qualifying ranges developed from the observation that the 50th percentile range or ‘preferred range’ was more densely populated and had a higher probability of identifying a CNS drug compared with the 95th percentile or ‘qualifying range’ (Figure 1).13,1 The 95% qualifying range was selected over the 100 percentile range since it avoided outliers.

Penalty calculation from the physicochemical property value. The model described here was based on computed penalties using the location of the property value relative to the preferred and qualifying ranges. No penalty was calculated if a property value was within the preferred range (preferred lower to preferred upper; PL-PU). The penalty score increased linearly to unity at the end of the qualifying range (qualifying lower to


6

Page 7 of 31

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


qualifying upper; QL-QU) and beyond as shown in Figure 2. To differentiate CNS and non-CNS drugs numerically, we introduced a ‘class variable’, with a score of 0 for CNS drugs and 10 for non-CNS drugs. Compounds that had scores ≤5 were more CNS-like and those with scores >5 were non-CNS. The physicochemical property-based penalties that were used to assign the ‘class’ were given in Table 1, along with their weight factor C (see equation 1 in the Methods section). The penalties for five randomly selected molecules along with the computed property values are shown in Table 2. Briefly, the class determination steps are to: (1) compute the eight properties and penalties (using PL, PU, QL and QU of each property; Table 1), and the function(s) given in Figure 2; and (2) compute the class using equation (1) where the coefficient for each property is derived from Table 1. Special provisions occur where PL = PU or, PL = QL, or PU = QU. In the first case the penalty will be zero if the property is equal to PL. In the second case, the penalty is zero if the property is less than or equal to PL, and third, the penalty will be zero if the property is greater than or equal to PU.

Model development. Based on the ease of automation, Pipeline Pilot was used for many of the topology-based property calculations. With the inconsistencies of computed pKa and logD values using different packages, these properties were removed from model building. The Lemke algorithm14,15 of quadratic programming16 was applied for the variable reduction and development of an interpretable model. A key feature of the current model was the decreasing contribution from the topological polar surface area. The importance of low polar surface area in a CNS drug has been noted previously by us and others.1,2 The decreasing contribution from PSA (coefficient = 0.001) in TEMPO


7


1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 8 of 31

resulted from the utilization of the number of hydrogen bond acceptors in the calculation, which highly correlated with the PSA value (Table 3).

TEMPO performance analysis. The CNS TEMPO scoring scheme was developed using both the CNS and non-CNS training sets. The validity of the model was evaluated by its performance on a randomly picked test set. The performance of the CNS-TEMPO score is shown in Table 4 for the classification of four compound collections: (i) a CNS training set of 310 approved CNS drugs; (ii) a non-CNS training set of 738 approved non-CNS drugs; (iii) a CNS test set of 70 approved CNS drugs; and (iv) a non-CNS test set containing 232 approved non-CNS drugs. The comparison of the other methods in the same format was not warranted, since the validation of those models used different training and/or test sets. For a qualitative direct comparison of the three methods, CNSMPO was implemented using ACDLabs (http://www.acdlabs.com/) consensus LogP, strongest basic pKa and LogD (at pH 7.4) in Pipeline Pilot. The training and test sets were combined and the scores of CNS and non-CNS drugs in the three scoring schemes were binned with a consensus score (the number of methods classifying a test molecule as CNS-like; Figure 4). A consensus score of 2 or more should be considered as ‘good’ for CNS druggability. The use of a consensus score was derived from the observation that the methods agreed to less than 10% for their CNS druggability for a randomly picked large MDDR dataset. For QP_CNS, a molecule was considered to be CNS if the score was 1 or 2, non-CNS if the score was –1 or –2, and equivocal when the score was 0. The largest separation of the distribution maxima of CNS and non-CNS drug should be considered as the measure of success of a method.


8

Page 9 of 31

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


Performance of the methods in different molecular weight ranges. Three random diverse sets of MDDR compounds separated into three molecular weight ranges (

Technically Extended MultiParameter Optimization (TEMPO): An

Recommend Documents