Small-Scale Panel Comprising Diverse Gene Family Targets

Subscriber access provided by Macquarie University

Article

Small-scale panel comprising diverse gene family targets to evaluate compound promiscuity Tomoya Sameshima, Tomoya Yukawa, Yoshihiko Hirozane, Masato Yoshikawa, Taisuke Katoh, Hideto Hara, Takatoshi Yogo, Ikuo Miyahisa, Teruaki Okuda, Makoto Miyamoto, and Russell Naven Chem. Res. Toxicol., Just Accepted Manuscript • DOI: 10.1021/acs.chemrestox.9b00128 • Publication Date (Web): 28 Aug 2019 Downloaded from pubs.acs.org on August 29, 2019

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 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 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.

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 29 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

Chemical Research in Toxicology

Small-scale panel comprising diverse gene family targets to evaluate compound promiscuity Tomoya Sameshima┴*, Tomoya Yukawa§, Yoshihiko Hirozane┴, Masato Yoshikawa┴, Taisuke Katoh┴, Hideto Hara┴, Takatoshi Yogo┴, Ikuo Miyahisa┴, Teruaki Okuda┴, Makoto Miyamoto┴ and Russell Naven§, † ┴Research,

Takeda Pharmaceutical Company Limited, Fujisawa, Japan

§Research,

Takeda Pharmaceuticals International, Inc., Cambridge, Massachusetts, USA

†Research,

Takeda California, San Diego, California, USA

KEYWORDS: promiscuity, early safety screening, discovery toxicology, binding assay, highthroughput screening

ACS Paragon Plus Environment

1

Chemical Research in Toxicology 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 29

ABSTRACT

Despite the recent advances in the life sciences and the remarkable investment in drug discovery research, the success rate of small-molecule drug development remains low. Safety is the second most influential factor of drug attrition in clinical studies; thus, the selection of compounds with fewer toxicity concerns is crucial to increase the success rate of drug discovery. Compounds that promiscuously bind to multiple targets are likely to cause unexpected pharmacological activity that may lead to adverse effects. Therefore, avoiding such compounds during early research stages would contribute to identifying the compounds with a higher chance of success in the clinic. To evaluate the interaction profile against a wide variety of targets, we constructed a small-scale promiscuity panel (PP) consisting of eight targets (ROCK1, PDE4D2, GR, PPARγ, 5-HT2B, Adenosine A3, M1 and GABAA) that were selected from diverse gene families. The validity of this panel was confirmed by comparison with the promiscuity index evaluated from larger-scale panels. Analysis of data from the PP revealed that both lipophilicity and basicity are likely to increase promiscuity, while the molecular weight does not significantly contribute. Additionally, the promiscuity assessed using our PP correlated with the occurrence of both in vitro cytotoxicity and in vivo toxicity, suggesting that the PP is useful to identify compounds with fewer toxicity concerns. In summary, this small-scale and cost-effective PP can contribute to the identification of safer compounds that would lead to the reduction in drug attrition due to safety issues.


2

Page 3 of 29 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


INTRODUCTION Despite the recent advances in the life sciences and the large investment in drug discovery research, the success rate of small-molecule drug development remains low.1,2 The success rate in drug discovery projects, from phase I clinical trials to FDA approval, has been reported to be approximately 10%.3,4 Safety accounts for 20–30% of failure in phase II and III studies;5,6 it is the second most influential factor of drug attrition following the lack of efficacy. Therefore, the identification of compounds with few toxicity concerns is considered to be crucial to improve the success rate of drug discovery. According to previous reports, 75% of adverse drug reactions (ADRs) are classified as “type A”,3,7-10 which are dose-dependent and related to the pharmacological action of the compound,3,7-10 i.e., type A ADRs are caused by the interaction of drugs with target proteins. It is widely accepted that compounds that promiscuously bind to multiple targets are likely to cause unexpected pharmacological activity, thus leading to adverse events.9,11-13 The relationship between interaction promiscuity and the in vivo toxicity of compounds has been experimentally investigated by several groups.12,13 In fact, the proportion of promiscuous compounds in the set of marketed drugs is known to be smaller than that of withdrawn drugs or discontinued clinical candidates.9 Considering these facts, avoiding offtarget promiscuity during the drug discovery process should lead to the identification of less toxic clinical candidates and, hopefully, to improving the success rate of drug discovery projects. To assess the promiscuous activity of compounds across gene families, it is necessary to perform experimental evaluations because it is difficult to predict such a property from sequence information alone.14 Therefore, promiscuity is generally measured by assessing the activity across a panel of target-based assays.9 A large-scale panel study is often considered useful because it covers the interaction profile for a wide variety of targets. In addition to promiscuity


3


Page 4 of 29

assessment, it can be useful to identify the potential mechanisms of off-target related toxicity by profiling the activity of compounds for individual targets. However, such studies are costly, time consuming and labor intensive, and thus, they cannot be easily performed at early research stages and are commonly applied only for a limited number of compounds. To successfully select and design less promiscuous compounds, it is necessary to assess the interaction promiscuity using an easily applicable small-scale panel because iterative synthesis and evaluation are required to determine the structure/activity relationship. To date, several small-scale panels have been reported for the investigation of compound promiscuity. For example, an assay panel consisting of targets that were selected based on their relationship with toxicity has been reported.12 A different research group selected targets with a high compound hit rate.11 Although several small scale panels have been reported, they were usually dominated by G-protein-coupled receptor (GPCR) targets, thus making it difficult to evaluate the interaction profile across other gene families. Even though the GPCR family is involved in various pharmacological and toxicological effects, this family only accounts for ~4% of the human protein-coding genome (~800 out of 20,000 in total)15 and is not the only gene family related to toxicity.9 Given that basic compounds tend to have higher interaction promiscuity, particularly towards GPCRs,16 a small scale panel that covers additional physicochemical property space, such as neutral and acidic compounds, should be useful to investigate the interaction promiscuity of a wide variety of gene families. Herein, we constructed a small-scale promiscuity panel (PP) consisting of eight targets selected from diverse gene families to evaluate the interaction of compounds with a wide range of targets. Moreover, we show the usefulness of the PP in the drug discovery process by analyzing the correlation of promiscuity with cytotoxicity and in vivo toxicity. EXPERIMENTAL PROCEDURES


4

Page 5 of 29 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


Materials GST-Rho-associated coiled-coil containing protein kinase 1 (ROCK1) (1-477 amino acids of accession number NP_005397.1) was purchased from Carna Biosciences (Kobe, Japan). The GST-glucocorticoid receptor (GR), LanthaScreen TR-FRET Peroxisome Proliferator-Activated Receptor (PPAR) Gamma Competitive Binding Assay Kit, HEPES solution, Kinase Tracer 236, sodium pyruvate and DMEM high glucose medium (no glutamine, no phenol red, cat# 31053028) were obtained from Thermo Fisher Scientific (Waltham, MA, USA). Staurosporine was from Enzo Life Science (Farmingdale, NY, USA). Dexamethasone, rosiglitazone and flumazenil were internally synthesized. All radioisotope (RI) ligands for the RI binding assay and the membrane fractions of 5-HT2B and adenosine A3 were purchased from PerkinElmer (Waltham, MA, USA). The membrane fraction of GABAA (α1β3γ2) was prepared in Charles River Laboratories (Wilmington, MA, USA). The membrane fraction of M1 was prepared using FreeStyle 293 cells (ThermoFisher Scientific) transiently coexpressing human M1 and PDE3A, as described previously.17 Dithiothreitol (DTT) and GST-PDE4D2 (2-507 amino acids of accession number NM_001197221) were from Sigma Aldrich (St. Louis, MO, USA). IB-MECA was from Adooq Bioscience (Irvine, CA, USA). Tween-20 was purchased from BioRad (Hercules, CA, USA). Brij-35 detergent was obtained from Merck Millipore (Billerica, MA, USA). EGTA was from Dojindo (Kumamoto, Japan). terbium (Tb)-labeled anti-GST antibody (Tb-anti-GST) was purchased from Cisbio (Codolet, France). Sulfo-Cy5-NHS ester was obtained from Lumiprobe (Hunt Valley, MD, USA). Cy5-labeled fluorescent probes for the time-resolved fluorescence resonance energy transfer (TR-FRET) binding assays were synthesized as described in the Supporting Information. All other reagents for the biochemical assays were obtained from Wako (Osaka, Japan).


5


Page 6 of 29

TR-FRET binding assay TR-FRET binding assays were performed using 384-well white flat-bottomed plates (product no. 784075, Greiner Bio-One, Frickenhausen, Germany). Unless otherwise indicated, the Tbanti-GST, GST-tagged proteins and fluorescent probes used in this work were preincubated over 60 min at 4 °C in the dark before addition to the assay plates. In each well of the assay plate, a 2fold concentration of compounds (2 μl) and 2-fold concentration of Tb-protein-fluorescent probe premix (2 μl) were mixed and incubated for 60 min at either 4 ºC (for GR) or room temperature (for the other targets). Subsequently, the TR-FRET signal was measured using an EnVision microplate reader (PerkinElmer). The solution in each well was excited with a 337 nm laser reflected by a dichroic mirror, and the fluorescence signals from Tb and the fluorescence probe were detected by using the corresponding emission filters. All of the optical parts for the EnVision plate reader were purchased from PerkinElmer. The combination of optical parts for each assay is described in Table S1. The signal of TR-FRET was expressed as the probe/Tb fluorescence ratio. The percentage of inhibition was calculated from equation (1):

  T   Percentage of inhibition  100   H    L   H

(1)

where T is the signal of the wells containing test compounds and μH and μL are the mean signals of the 0% and 100% inhibition control wells, respectively. The 0% and 100% inhibition control signals were obtained in the absence and presence of competitive inhibitors for each target. The buffer conditions, names and concentrations of the protein, fluorescent ligand and competitive inhibitor for each assay are summarized in Table S1. RI binding assay RI binding assays were performed in 96-well plates with a final volume of 0.1 ml. For the competition experiments, cell membranes were incubated with a specific concentration of the


6

Page 7 of 29 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


radio ligand and test compounds for 2 h at room temperature in the assay buffer. The reaction was terminated by rapid filtration through polyethyleneimine-coated GF/C filter plates (PerkinElmer) using a cell harvester (PerkinElmer). The filter plates were washed five times with 50 mM Tris-HCl and dried at 42 °C. Microscint 0 (PerkinElmer) was added to each well, and the radioactivity was measured using a TopCount microplate scintillation counter (PerkinElmer). The percentage of inhibition was calculated using equation (1). The 0% and 100% inhibition control signals were obtained in the absence and presence of the indicated inhibitors for each target. The buffer conditions, names and concentrations of the protein, fluorescent ligand and competitive inhibitor for each assay are summarized in Table S2. Cytotoxicity assay HepG2 cells were cultured in DMEM high glucose medium containing 2 mM L-glutamine, 1 mM sodium pyruvate, 5 mM HEPES, 10% (v/v) fetal bovine serum, 100 U ml−1 penicillin, and 100 μg ml−1 streptomycin. The cells were seeded at 2500 cells/20 μl/well on 384-well plates (Greiner) for nearly 24 h at 37 °C/5% CO2. Subsequently, 20 μl of the medium containing the test compounds was added to the plate, followed by 72 h of incubation at 37 °C/5% CO2. Subsequently, 20 μl of the Cell Titer Glo (Promega) reagent was added to each well, and the luminescence was measured using the EnVision plate reader. The values of the 0 and 100% inhibition controls were obtained from wells containing DMSO-treated cells and the medium only, respectively. The percentage of inhibition was calculated using equation (1). IC50 values were calculated by fitting a sigmoidal dose-response curve to the plot of the percentage of inhibition as a function of inhibitor concentration. Fittings were performed using the XLfit software (IDBS, Guildford, UK). Calculation of the physicochemical properties


7


Page 8 of 29

The partition coefficient, clogP, was calculated using the Daylight version 4.95 software (Daylight Chemical Information Systems, Inc., Laguna Niguel, CA, USA). The pKa was calculated with ACD/Labs version 2017.1.3 (Advanced Chemistry Development Inc., Toronto, Canada). The molecular weight was calculated using the JChem version 5.10.2 software (ChemAxon, Budapest, Hungary). Definition of hit criteria and calculation of the promiscuity index in large-scale panels An inhibition equal to or greater than 50% at a concentration of 10 µM was set as the hit criterion of the Panlabs panel (Eurofins Panlabs, St Charles, MO, USA). The ratio of hit assays was adopted as the promiscuity index. The ratio of hit assays for each compound was calculated by dividing the number of hit targets by the total number of assays for the compound. Since the ratio of hit assays cannot be accurately assessed for the compounds tested in a small number of assays, only the compounds tested in 50 or more assays (enzymatic or binding) at 10 µM were selected for the data analysis. For the SAFETYscan (Eurofins DiscoverX, Fremont, CA, USA), compounds showing an EC50/IC50 under 10 μM were determined as hits. SAFETYscan includes targets for which both agonist and antagonist assays are performed. When counting the hit assays in the SAFETYscan, agonist and antagonist assays were considered as two independent assays even though they correspond to the same target. Annotation of toxicity from in vivo toxicity studies For the data analysis, we selected 221 internal compounds provided by the Takeda Pharmaceutical Company Ltd., for which in vivo toxicity studies were performed in rats or mice with oral administration and a duration from 3 days to 28 days and for which the pharmacokinetic exposure data (Cmax) could be found. All animal studies were approved by the Takeda ethical committee. No distinction was made regarding the sex of the animals. According


8

Page 9 of 29 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


to a previous report,13 compounds showing no significant toxicity at greater than or equal to 10 µM total Cmax were annotated as “Clean” and those showing significant toxicity below a total 10 µM Cmax were defined as “Tox.” Compounds annotated as neither “Tox” nor “Clean” were excluded from the data analysis. RESULTS AND DISCUSSION Development of the PP and overall assay results To evaluate the interaction promiscuity of compounds during the early stage of drug discovery, we intended to construct a small-scale PP, which consists of a limited number of targets. Since the purpose of this PP is to estimate interaction promiscuity and not to identify off-target(s) for small molecule compounds, targets for the PP were selected based on their gene family diversity and coverage of physicochemical property space rather than their toxicological significance. For the components of the PP, it is reasonable to select targets with which small molecules are likely to interact considering the aim of the PP. For this reason, gene families whose small molecule ligands are hard to identify, such class B/C GPCR18,19 and protease,19 were not selected for the components of the PP. From the analysis of the public and internal assay datasets (Supporting Information), eight targets were selected as the PP components (Table 1). Table 1. Selected Targets for the Promiscuity Panel.

Target

Gene name

Class

Hit (%)*

ROCK1

ROCK1

Kinase

11.7

PDE4D2**

PDE4D

PDE

7.1

GR

NR3C1

Nuclear receptor

0.4

PPARγ

PPARG

Nuclear receptor

3.4

5-HT2B

HTR2B

GPCR

18.8

ratio


9


Page 10 of 29

Adenosine A3

ADORA3

GPCR

28.9

M1

CHRM1

GPCR

16.4

Ion channel

3.8

GABRA1 GABAA (α1β3γ2)***

GABRB3 GABRB2

*

Hit ratio was calculated by compounds showing ≥50% inhibition at 10 μM divided by the total number of compounds. **

PDE4D2 is a splice valiant of PDE4D.

***

GABAA was expressed as a heteromer of α1, β3 and γ2 subunits.

Considering the assay throughput and suitability, the binding assay format, in which the affinity of compounds is determined by their ability to displace tracer ligands, was selected to construct the PP. An inhibition equal or greater than 50% at a concentration of 10 µM was set as the hit criterion for the PP. Using this panel, the activity of 1713 compounds was evaluated; the hit rate for each target is summarized in Table 1. Adenosine A3 showed the highest hit rate (29%), and GR showed the lowest (0.4%). The distribution of the number of hits across the panel is shown in Figure 1A. Among the test compounds, 55% hit at least one target in the PP (1 hit 30%, 2 hits 18%, 3 hits 5.9%, 4 hits 1.5%, and 5 hits 0.23%).


10

Page 11 of 29 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


Figure 1. Overview of the assay data in the PP. A. Distribution of the target hit number. B. PP heatmap. Red and blue correspond to 100% and 0% inhibition, respectively. The overall results of the PP are presented in Figure 1B. As shown in the heatmap, there was not a significant overlap in interaction profile among the eight targets. In fact, the highest correlation coefficient of a pair in the PP was only 0.34 (5-HT2B and M1, Table S3), indicating a low level of target redundancy within the panel. To further investigate the differences in the physicochemical properties of compounds interacting with each target in the PP, correlation between the inhibitory activity and most basic pKa (MB-pKa) for basic and neutral compounds was examined. Of the three GPCRs in the PP, inhibition of 5-HT2B and M1 correlated with MBpKa, while that of adenosine A3 did not (Figure S5). In addition, MB-pKa was slightly


11


Page 12 of 29

negatively correlated with the inhibition of adenosine A3, PDE4D2 and PPARγ, while that did not show significant correlation with the inhibition of ROCK1 and GABAA (Figure S5). These results also indicated that our PP consists of targets with diverse properties and it is useful to detect the promiscuity of compounds with a wide range of chemical properties. A biased distribution was not observed in the compound set for clogP or the molecular weight (MW) (Figure S6). The mean ± SD values of clogP and MW were 3.3 ± 1.5 and 410 ± 90, respectively (Figure S6), suggesting that the compound set did not deviate from the compounds generally used for small-molecule drug discovery.20,21 Target hit rate correlation between the PP and large-scale panels To assess the validity of the promiscuity evaluation using the PP, we examined the target hit rate correlation between the PP and other large-scale panels. We previously selected over 50 in vitro assays provided by Eurofins Panlabs (mainly composed of binding assays, based on their pharmacological and toxicological properties)22 and utilized them as a secondary pharmacology panel (Panlabs panel). The compounds that were evaluated in both the Panlabs panel and the PP were selected (128 compounds), and the target hit ratio of each panel was compared. The ratio of hit assays was selected as an evaluation index of the Panlabs panel because the composition of assays in this panel was different depending on the assay date. Compounds were grouped based on the number of hit targets in the PP and compared against the distribution of the target hit ratios in the Panlabs panel for each compound. The target hit ratio of the Panlabs panel increased with an increasing number of hit targets in the PP (Figure 2A). Although the number of assays in the Panlabs panel is larger than in the PP, there was only one compound with a hit in the PP and not in the Panlabs panel. This is because the inhibitory activity of this compound against adenosine A3, which was its hit target in the PP, was not evaluated in the Panlabs panel.


12

Page 13 of 29 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


Next, we compared the number of hit targets in the PP with that of the SAFETYscan (Eurofins DiscoverX),23 which is a panel composed of functional assays selected according to toxicity concerns.9 Compounds submitted to both panels (273 in total) were selected and the number of hit assays in the SAFETYscan was compared with the number of hit targets in the PP. Similar to the Panlabs panel, there was a correlation between the PP and SAFETYscan (Figure 2B). Interestingly, compounds hitting ≥2 targets in the PP were active in at least one assay in the SAFETYscan (Figure S7).

Figure 2. Comparison of the target hit ratio of the PP and large-scale panels: (A) Panlabs and (B) SAFETYscan. Data are shown in box-and-whisker plots.


13


Page 14 of 29

For almost all of the compounds utilized in this study, a large number of targets in SAFETYscan were also included (70–80%) in the Panlabs panel (Table S4). However, the compound property detected in the SAFETYscan is different from that detected in the Panlabs panel due to the following reasons: 1) SAFETYscan is composed of functional assays and it can detect allosterically interacting compounds as well as orthosterically interacting ones and 2) almost all compounds (127 out of 128) are tested in over 90 assays in the Panlabs panel and evaluated for a large number of targets that are not included in SAFETYscan. Considering these facts, the interaction physicochemical space of these two large panels is qualitatively different. Correlation between PP and these two different large-scale panels indicated the validity of the estimation of compound promiscuity using the small-scale PP. It should be noted that the number of hit targets in the PP was correlated with these large scale panels even if GPCR targets (Figure S8) or targets in the PP (Figure S9) were excluded. These results indicated that the correlation of the target hit ratio in the PP with those in the larger panels was not derived from target overlapping or a specific gene family, that is GPCRs. These finding also supported the validity of promiscuity assessment using the PP. Physicochemical properties that affect compound promiscuity It has been reported that interaction promiscuity depends on the physicochemical properties of the compounds.24,25 The consensus is that interaction promiscuity is positively correlated with the clogP24,25 and pKa (for basic molecules).16,24,25 To confirm whether this tendency is observed in our PP, we examined the relationship between the number of hit targets in the PP and the physicochemical properties of the test compounds. Initially, we investigated the influence of clopP on the number of hit targets. Similar to the previous reports,24,25 both the median and average of the number of hit targets (i.e., promiscuity) in the PP increased with the clogP value


14

Page 15 of 29 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


(Figure 3A). This relationship between promiscuity and lipophilicity may be explained because orientation is not crucial for hydrophobic interactions (entropy driven), unlike hydrogen bonds and ionic interactions (enthalpy driven), thus leading to nonspecific interactions with various targets.26 Next, we analyzed the correlation between the MB-pKa and interaction promiscuity. Since the relationship between the pKa and promiscuity was reported for basic molecules,25 not for zwitterions, compounds with the most acidic pKa under 7 were excluded from the analysis. On average, when both the median and average were utilized, compounds with a high hit rate in the PP tended to have large MB-pKa values (Figure 3B), suggesting that compounds with basic functional groups are likely to show a promiscuous interaction behavior. Since we confirmed the similar relationship between promiscuity and physicochemical properties with the previous reports,16,24,25 the miniaturized PP is considered to reflect the results obtained from large-scale panels.


15


Page 16 of 29

Figure 3. Correlation between the number of hit target in the PP and the physicochemical properties. The median (closed diamonds) and average (open circles) of each parameter (A. clogP, B. MB-pKa, C. MW) is plotted against the number of hit targets in the PP. Conversely, no correlation was found between the MW and the number of hit targets in the PP (Figure 3C). Unlike lipophilicity and basicity, the relationship between the MW and promiscuity is controversial. Within Pfizer's dataset, compounds with a small MW showed high promiscuity, whereas in the dataset of Novartis and Roche, the average MW was higher for promiscuous compounds.25 The reason for the apparent discrepancy among these reports may be the difference in the interaction property utilized in each dataset and the composition of the targets in the panel study.


16

Page 17 of 29 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


Knowing that lipophilic compounds with stronger basic functional groups are likely to be promiscuous is very useful for the design of less promiscuous compounds. Although an overall tendency was obtained, a broad distribution of physicochemical properties was observed in each group classified by the target hit number in the PP (Figure S10). This indicates that the physicochemical properties can be an index for promiscuity improvement but cannot perfectly reflect the interaction promiscuity of individual compounds. To date, there have been several reports on an in silico approach to predict the interaction with off-targets.27,28 It is true that such approaches are helpful to assess the interaction promiscuity of individual compounds, but our PP can be performed in a high-throughput manner that enables us to experimentally evaluate offtarget promiscuity. Correlation between the cytotoxicity and compound promiscuity The suitability of promiscuity assessment using our miniaturized PP was supported by its correlation with the promiscuity measured in large-scale panels (Figure 2A and B) and by comparison with the promiscuity and physicochemical properties relationship reported previously (Figure 3A and B). Encouraged by these results, we used the PP to investigate the correlation between promiscuity and the biological effects of compounds. Since promiscuous compounds have the potential to act on unexpected targets, they may cause cytotoxicity by affecting targets involved in cell survival. To assess this hypothesis, we analyzed the relationship between promiscuity and cytotoxicity. The compounds evaluated in both cytotoxicity assays and the PP were extracted (1050 in total), and the ratio of cytotoxic compounds in each group, classified by the number of hit targets in the PP, was examined. Table S5 summarizes the number of cytotoxic and noncytotoxic compounds in each number of hit targets considering an IC50 under 100 μM as a hit criterion for cytotoxicity. As expected, the cytotoxicity increased as


17


Page 18 of 29

the number of hit targets increased (Figure 4). A similar tendency could be observed when lower IC50 values were applied as criteria for cytotoxicity (Figure S11). From these results, we confirmed that promiscuous compounds tend to be cytotoxic.

Figure 4. Ratio of cytotoxic compounds in each group classified by the number of hit targets in the PP. Compounds showing IC50 values under 100 μM were classified as cytotoxic. The dashed line represents 100%. Although the relationship between the number of hit targets in the PP and cytotoxicity was observed, approximately 27% of the compounds showed cytotoxicity despite having no hits in the PP (123 out of 452, Table S5) when IC50 < 100 μM was utilized as the criterion for cytotoxicity. This can be explained by the fact that there are some targets that are indispensable for cell survival are not covered by the eight targets of the PP. In contrast, approximately 10% of the compounds that hit two or more targets in the PP (25 out of 277, Table S5) did not show cytotoxicity even at the concentration of 100 μM, suggesting that they do not inhibit targets important for cell survival or there are factors, such as physicochemical properties, that mitigate a cytotoxic response. It is worth noting that the targets for the PP were selected considering the coverage of interaction profiles rather than their cellular toxicity properties, as such. Despite


18

Page 19 of 29 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


having a high target hit rate in the PP, it can also be assumed that these compounds are not necessarily cytotoxic. Relationship of compound promiscuity with the in vivo toxicity We investigated whether compound promiscuity, as measured from the PP, correlates with the in vivo toxicity. From the compound set, those that had been evaluated in in vivo animal toxicity studies were extracted (221 in total) and annotated as “Tox” and “Clean” based on the in vivo toxicity studies as indicated in the Experimental Procedures section. Among them, 82 were “Clean” and 97 were “Tox” (Table S6). The remaining 42 compounds, which were neither “Clean” nor “Tox,” were excluded from the data analysis. The probability of toxicological findings for more promiscuous compounds (≥2 hit in the PP) is higher than that of less promiscuous ones (≤1 hit in the PP) (Figure 5 and Table S6). The list of targets for the highly promiscuous compounds (≥3 hit in the PP) annotated as “Tox” is summarized in Table S7. Except for one compound (D in Table S7), these compounds interact with multiple targets across gene families. Their promiscuous activity on targets across gene families may be the reason for the adverse events from unexpected targets.


19


Page 20 of 29

Figure 5. Relationship between the number of hit targets in the PP and the ratio of "Tox” compounds. (* p

Small-Scale Panel Comprising Diverse Gene Family Targets

Recommend Documents