In vivo phenotypic drug discovery: Applying a behavioral assay to the discovery and optimization of novel antipsychotic agents

Article abstract of DOI:10.1039/c6md00128a

Phenotypic drug discovery (PDD) is increasingly being recognized as a viable compliment to target-based drug discovery (TDD). By measuring functional changes, typically at a systems level, PDD can facilitate the identification of compounds having a desirable pharmacology. This capability is particularly important when studying CNS diseases where drug efficacy may require modulation of multiple targets in order to overcome a robust, adaptive biological system. Here, we report the application of a mouse-based high-dimensional behavioral assay to the discovery and optimization of a structurally and mechanistically novel antipsychotic. Lead optimization focused on optimizing complex behavioral features and no explicit effort was made to identify the target (or targets) involved.

Full text of DOI:10.1039/c6md00128a

View Article Online
View Journal
MedChemComm
Accepted Manuscript
This article can be cited before page numbers have been issued, to do this please use: L. Shao, U.
Campbell, Q. K. Fang, N. A. Powell, J. E. Campbell, P. G. Jones, T. Hanania, V. Alexandrov, I. Morganstern,
E. Sabath, H. M. Zhong, T. H. Large and K. L. Spear, Med. Chem. Commun., 2016, DOI:
10.1039/C6MD00128A.
This is an Accepted Manuscript, which has been through the
Royal Society of Chemistry peer review process and has been
accepted for publication.
Accepted Manuscripts are published online shortly after
acceptance, before technical editing, formatting and proof reading.
Using this free service, authors can make their results available
to the community, in citable form, before we publish the edited
article. We will replace this Accepted Manuscript with the edited
and formatted Advance Article as soon as it is available.
You can find more information about Accepted Manuscripts in the
Information for Authors.
Please note that technical editing may introduce minor changes
to the text and/or graphics, which may alter content. The journal’s
standard Terms & Conditions and the Ethical guidelines still
apply. In no event shall the Royal Society of Chemistry be held
responsible for any errors or omissions in this Accepted Manuscript
or any consequences arising from the use of any information it
contains.
www.rsc.org/medchemcomm

Page 1 of 27
MedChemComm
View Article Online
DOI: 10.1039/C6MD00128A
In Vivo Phenotypic Drug Discovery: Applying a Behavioral Assay to the
Discovery and Optimization of Novel Antipsychotic Agents†
Liming Shao,† Una Campbell,* Q. Kevin Fang,* Noel A. Powell,* John E. Campbell,‡
Philip G. Jones,* Taleen Hanania,§ Vadim Alexandrov,§ Irene Morganstern,§ Emily
Sabath,§ Hua M. Zhong,∫ Thomas H. Large,* and Kerry L. Spear*
¤
†The authors declare no competing interests.
*Sunovion Pharmaceuticals Inc., 84 Waterford Dr., Marlborough, MA 01752 USA
†Current address: Fudan University, School of Pharmacy, Medicinal Chemistry Department, R323, South
Chemistry Building, 826 Zhangheng Rd, Pudong Xinqu, Shanghai 201203 PRC
‡Current address: Epizyme, Inc., 400 Technology Sq. 4th Fl., Cambridge, MA 02139 USA
§PsychoGenics Inc, 765 Old Saw Mill River Rd #104, Tarrytown, NY 10591 USA
∫PharmaAdvance, Inc., 6 Dongsheng West Rd., Building D1, Jiangyin, Jiangsu 214437 PRC
¤Corresponding author
Abstract
Phenotypic drug discovery (PDD) is increasingly being recognized as a viable
compliment to targetꢀbased drug discovery (TDD). By measuring functional changes,
typically at a systems level, PDD can facilitate the identification of compounds having a
desirable pharmacology. This capability is particularly important when studying CNS
diseases where drug efficacy may require modulation of multiple targets in order to
overcome a robust, adaptive biological system. Here, we report the application of a
mouseꢀbased highꢀdimensional behavioral assay to the discovery and optimization of a
structurally and mechanistically novel antipsychotic. Lead optimization focused on
optimizing complex behavioral features and no explicit effort was made to identify the
target (or targets) involved.
Introduction
Targetꢀbased drug discovery (TDD) has been the dominant platform for discovering new
drugs for over a quarter century. During this time, development of exquisitely potent and
selective drug candidates has become the key metric of the drug discovery process.
While medicinal chemists have become quite good at achieving these goals, the rate at
which new drugs are approved has stayed largely flat.1 At the same time, R&D
expenditures have increased dramatically.2 This can be attributed, in part, to clinical
attrition rates that remain stubbornly high. Today, on average, only 13.3% of molecules
that enter the clinic achieve NDA approval and success rates in some therapeutic areas
such as CNS (4.7%) are even lower.3 In terms of cost per drug approval, the era of TDD
has been accompanied by a decline in Pharma productivity.4
Since biological systems tend to be complicated and difficult to study, TDD acts by
deconvoluting a system into its constituent parts in order to provide detailed information
about a single molecular target. However, an unintended consequence of this reductionist
approach is that critical information may be lost. In particular, it can be difficult to
understand the full functional role of a molecular target within a biological system or how
1

MedChemComm
Page 2 of 27
View Article Online
DOI: 10.1039/C6MD00128A
system adaptation might compensate for the disruption of a single target. The role that
reductionism may have played in the decline in Pharma productivity has been noted5 and
has spurred efforts to identify alternate approaches to drug discovery. One strategy that
has (re)emerged is phenotypic drug discovery (PDD) which focuses on functional rather
than target activity. Since a phenotypic assay probes a compound’s effect on a typically
complex biological response, it may afford a readꢀout that is more physiologically
relevant than that of a targetꢀbased assay. Importantly, no knowledge of the molecular
target (or targets) is required. The process can be driven entirely by measuring changes
in functional endpoints and so might more effectively identify molecules that engage
multiple targets.
Despite the central influence of TDD, phenotypic screening continues to have a
surprisingly significant impact on new drug approvals. Recent reports6, 7 have confirmed
that phenotypic screening remains a relevant approach for drug discovery. Today, PDD,
aided by such critical technological advances as highꢀcontent screening and big data
analysis, is experiencing a renaissance. A number of reasons for this have been
postulated8ꢀ10 but PDD has emerged as an alternative, and complimentary, strategy to
TDD. Phenotypic screens are being used to drive the discovery process in a variety of
therapeutic areas (e.g., infectious diseases,11ꢀ13 oncology,14 CNS15, 16). To date, cellꢀbased
assays, which tend to be more reproducible and amenable to highꢀthroughput screening
than ex vivo or in vivo assays, have predominated. However, there have also been reports
of whole animal assays being used in phenotypic screens. These largely focus on
organisms (e.g., parasites,17, 18 nematodes,19, 20 xenopus embryos,21, 22 zebrafish15) that can
be adapted to automated, highꢀthroughput screens. While in vivo assays can be
challenging (e.g., lower reproducibility, lower throughput, higher cost), there are some
clear advantages associated with in vivo PDD. For example, animalꢀbased screening can
select for compounds having superior ADME (absorption, distribution, metabolism and
excretion) properties and brain exposure.
The complex biology associated with most CNS diseases can explain, at least in part, the
poor clinical success rate in this therapeutic area. Drug efficacy may require modulation
of multiple targets in order to overcome a robust, adaptive biological system. Notably,
most currently marketed CNS drugs act on multiple molecular targets. This is especially
true for drugs that treat psychiatric diseases such as schizophrenia. While all known
drugs that treat schizophrenia block the dopamine D2 receptor,23 activity at other
receptors (e.g., serotonergic, adrenergic, muscarinic, histaminergic) also contributes to
their efficacy.24 Unfortunately, because of the rudimentary understanding of CNS
network topologies, it is exceptionally difficult to select an appropriate target
combination and design a polypharmacological molecule that is both safe and efficacious.
As a consequence, most antipsychotic drugs were discovered serendipitously. This
realization led us to develop a phenotypic strategy for the discovery and optimization of
novel antipsychotics. Here we report the application of a mouseꢀbased in vivo phenotypic
assay for both hit identification and lead optimization resulting in the discovery of a
compound that is structurally and mechanistically novel as well as active in known
animal models relevant to schizophrenia.
2

Page 3 of 27
MedChemComm
View Article Online
DOI: 10.1039/C6MD00128A
Results and Discussion
At the outset of our program, we selected the SmartCube® system, a mouseꢀbased, highꢀ
throughput automated behavioral platform developed by PsychoGenics Inc as our
screening assay. The system presents a sequence of challenges to a mouse through its
customized hardware, extracts more than 2000 features per session, and, using
proprietary bioinformatics tools, predicts potential therapeutic efficacy of compounds
from the changes that the drug induces in those behavioral features. A thorough
description of the platform and its validation has been described elsewhere.25
Additionally, the system had previously been validated using a collection of marketed
antipsychotic drugs26 as well as in other drug discovery programs. 27 While analysis of
these data can be complex, two basic analytical measures were used. The first was signal
strength (shown as a bar with color bands and normalized to a scale of 0 to 100%), which
is a measure of the probability of behavioral activity differing from vehicle (i.e., a crossꢀ
validated accuracy of a binary classifier of discriminating between the vehicle and drug
mice based on their behavioral features is a measure of activity, where 0 corresponds to a
chance level in which drug and vehicle are indistinguishable and 100% accuracy
representing perfect discrimination or clear differentiation from vehicle). Signal strength
is a measure of efficacy with a minimum efficacious dose (MED) defined for this
program as the lowest dose administered (i.p.) in which a compound’s nonꢀvehicle signal
remains above 50%. The second measurement was class signature (represented as one or
more of 15 colors; Figure 1A) which is a probabilistic outcome of a multiꢀclass support
vector machines (SVM) based classifier mapping behavioral features to the class labels
defined by know reference drugs (CNS indications). Both SVM28 and probabilistic
neural network based methods29 performed very similarly on all crossꢀvalidation tests. A
SVM based method was selected for speed considerations. The size of a given color
band in the signal bar indicates the probability that the observed behavior matches that of
a set of reference drugs for each therapeutic indication (e.g., Figure 1B). For example, an
antipsychotic signature is represented by a purple color. Since, in our hands, we
encountered no instance where a single molecule was found to have two different class
colors that are both greater than 30% at a single dose, we considered a class signature of
30% or more at a specific dose as representing a high probability that the test compound
will be behaviorally similar to known drugs from that class.
3

MedChemComm
Page 4 of 27
View Article Online
DOI: 10.1039/C6MD00128A
A
B
Figure 1.
A. SmartCube® class signature color key (15 classes represented including highꢀdose antidepressant, highꢀ
dose antipsychotic and vehicle); *includes antidepressant (green) and highꢀdose antidepressant (dark
green); **includes antipsychotic (purple) and highꢀdose antipsychotic (dark purple)
B. Structure and SmartCube® class signature of compound 1 (administered i.p. 15 min prior to testing).
The predominant purple color is characteristic of an antipsychotic signature.
The initial screening campaign had two primary goals. The first goal was to identify hits
having a predominant antipsychotic signature. In order to ensure that the process did not
simply identify another compound acting via a D2 mechanism, a second goal was to
avoid any D2 receptor activity. Screening of a 1000ꢀcompound library led to the
identification of multiple hits, including compound 1 (Figure 1B).30 This hit was
commercially acquired as the HCl salt and was screened as the racemate. As shown in
Figure 1B, racemic 1 displayed a strong signal (>95%) at a dose of 30 mg/kg (i.p.) as
well as a predominant antipsychotic signature (~40% purple). Given the unusual
dihydropyrimidyl (DHP) moiety, initial hit followꢀup focused on confirming compound
stability. When stored and used as the HCl salt, 1 was found to be sufficiently stable for
in vivo studies, although it proved to be less stable as the freebase, decomposing over the
course of days. Additionally, 1 was found to have a favorable in vitro ADME profile31
(see Table 1). Further profiling of this hit in a receptor panel screen32 confirmed that it
had no D2 binding at 10 µM. Encouraged by these data, we decided to bypass the hitꢀtoꢀ
lead phase and directly initiate lead optimization using the SmartCube® system as the
primary assay. Several goals were identified, most notably maximization of the
antipsychotic signature and improvement of the MED, while avoiding D2 receptor
activity.
4

Page 5 of 27
MedChemComm
View Article Online
DOI: 10.1039/C6MD00128A
Table 1. In vitro ADME profile of compound 1. All assays are commercially available from Cyprotex.31
Scheme 1 depicts the general route for the synthesis of a variety of 2ꢀ and 3ꢀsubstituted
dihydropyrimidinothiophenes. Addition of a pyrimidine moiety to the thiophene
substrates was generally straightforward and afforded the desired compounds in one step
in fair to good yields. To facilitate product isolation, in some instances, the unpurified
reaction mixture was treated directly with Boc anhydride to give a Boc protected
precursor, which after flash chromatography, was deꢀprotected with HCl to afford the
final product as the salt in 10 to 83% overall yields (see supplemental material). We also
discovered that, if the thiophene was substituted with an electronꢀwithdrawing group, the
reaction could be assisted with microwave irradiation or ultrasonication. Addition
occurred preferentially at the 2 or 2’ position of the thiophene. When both of these
positions were blocked, then addition occurred at the 3 or 3’ positions to give 3ꢀ
substituted dihydropyrimidinothiophenes as racemates.33, 34
Scheme 1. General synthetic route to DHPꢀsubstituted thiophenes. Reagents and conditions: (a)
pyrimidines, TFA, MW or ultrasonication; (b) Boc2O, Na2CO3; (c) HCl.
In the interest of expediency, initial exploration of the thiophene scaffold was limited to
racemates. With a DHP group at the 3ꢀposition, we found that structural modifications
that included large functional groups or functional group changes at multiple positions
frequently resulted in substantial transformations in the class signature (see Figure 2,
compounds 5-9). Conservative structural changes were better tolerated, but, even then,
sometimes dramatic class signature changes could be observed. For example,
compounds 2, 3 and 4 differ from 1 by only a single atom or methyl group and yet the
predominant class signature for each of these compounds is different (i.e., anxiolytic,
antidepressant and analgesic classes respectively). While the reason for this is not
known, these results may be consistent with a polypharmacological mechanism. If the
assay is measuring how modulation of multiple molecular targets affects an intact
biological system, then it is conceivable that, as the structure changes, the targets
modulated may also change in ways that lead to different behavioral outcomes. Based on
these findings, we decided to mostly limit SAR studies to conservative changes.
5

MedChemComm
Page 6 of 27
View Article Online
DOI: 10.1039/C6MD00128A
Figure 2. Structure and SmartCube® class signatures of representative substituted 3ꢀ
dihydropyrimidinothiophenes (administered i.p. 15 min prior to testing).
Moving the DHP moiety to the 2ꢀposition of the thiophene maintained or enhanced the
antipsychotic signature at a dose of 30 mg/kg (Figure 3). Additionally, the size of the
thiophene substituents was found to be important for signal strength. Small substituents
tended to maintain a strong antipsychotic signature and often improved the MED (e.g.,
10, 12). As previously noted, larger substituents (e.g., 14, 15) usually altered the
predominant class signature. In this series, compound 12 was found to have the strongest
antipsychotic signature, the best MED and little (if any) binding to the D2 receptor (18%
at 30 µM) (see supplemental material).
Figure 3. Structure and SmartCube® class signatures of representative substituted 2ꢀ
dihydropyrimidinothiophenes (administered i.p. 15 min prior to testing)
.
12 was submitted to chiral chromatography to obtain the pure enantiomers (12a, 12b).
For the purposes of this study, no effort was made to determine the absolute
6

Page 7 of 27
MedChemComm
View Article Online
DOI: 10.1039/C6MD00128A
stereochemistry and all stereochemical assignments were arbitrary. The full doseꢀ
response signatures of these compounds are shown in Figure 4A. Most of the activity
was found to reside with one enantiomer, 12a, having a MED of 3 mg/kg, the same as the
racemate. Additionally, 12a was found to have a strong antipsychotic signature at three
doses (3, 10, 30 mg/kg). While the other enantiomer (12b) showed a similar signature, it
had a much weaker activity (MED=30 mg/kg).
During the isolation of the enantiomers, we noticed that concentration of the free bases at
elevated temperatures resulted in modest decomposition but no changes in enantiomeric
purities. To explore the extent of this decomposition, a stability study of 12a was
conducted. At pH 2 the compound was found to be stable in aqueous solution at room
temperature. However, when the pH was adjusted to 7.1, a slow decomposition was
observed (see supplemental material), affording a complex mixture. We anticipated that
reꢀaromatization of the DHP ring by oxidation could be a route of decomposition so an
authentic standard of the presumed aromatized decomposition product 16 was prepared
by H2O2 oxidation of 12 (Figure 4B). While 16 did display some SmartCube activity, we
were unable to detect the formation of 16 during either chemical stability studies or the
mouse and human liver microsomal stability assays of 12a.
A
B
Figure 4.
A. Structure and SmartCube® class signatures of 12 (racemate), 12a (enatiomer 1) and 12b (enatiomer 2) at
multiple doses (administered i.p. 15 min prior to testing). Absolute stereochemistry was not
determined.
B. Structure and SmartCube® class signatures of 16 at multiple doses (administered i.p. 15 min prior to
testing).
Since the observed degradation appeared to be a consequence of the thiophene ring
instead of the DHP moiety, we expanded the scope of our SAR studies to include
dihydropyrimidinobenzothiophenes, which could be prepared by the same general
synthetic route shown in Scheme 1 (see supplemental material). In order to probe
whether the SAR developed for the thiophene series translated to the benzothiophenes,
we prepared a series of 2ꢀsubstituted benzothiophenes as shown in Figure 5. To our
surprise, these compounds (17-22) exhibited little of the desired antipsychotic signature,
7

MedChemComm
Page 8 of 27
View Article Online
DOI: 10.1039/C6MD00128A
although most maintained a strong signal at 30 mg/kg. Because of the generally
undesired class signatures, the MED for these compounds was not determined.
Figure 5. Structure and SmartCube® class signatures of representative substituted 2ꢀ
dihydropyrimidinobenzothiophenes (administered i.p. 15 min prior to testing).
In contrast, when the DHP group was moved to the 3ꢀposition, the antipsychotic
signature tended to be maintained when dosed at 30 mg/kg. Additionally, improved
MEDs were more frequently found (Figure 6). While these results were unanticipated by
the earlier thiophene SAR, they did offer an alternative to the 2ꢀDHP substituted
thiophenes. The SAR around the phenyl ring was quickly established. Small functional
groups at the 4 or 5 positon of the benzothiophene ring were preferred (e.g., 23, 24, 25)
for both antipsychotic signature and improved MEDs. Substitutions at positions 6 or 7
(e.g., 26, 27, 28) were less tolerated, exhibiting weaker antipsychotic signatures and/or
weaker MEDs. Finally, replacement of the phenyl ring with saturated carbocyclic rings
(e.g., 29, 30) was also acceptable.
Figure 6. Structure and SmartCube® class signatures of representative substituted 3ꢀ
dihydropyrimidinobenzothiophenes (administered i.p. 15 min prior to testing).
We also studied the effects of DHP group substitutions on activity. As illustrated in
Figure 7, while some modifications of the DHP moiety maintained the antipsychotic
8

Page 9 of 27
MedChemComm
View Article Online
DOI: 10.1039/C6MD00128A
signature (e.g., 31), the trend was for less active compounds or compounds having
undesired class signatures such as sedative (orange bar) or adverse effects (brown bar)
(e.g., 32-36). In this series all measured MEDs were greater than 10 mg/kg.
Figure 7. Structure and SmartCube® class signatures of representative DHPꢀsubstitutions of 2ꢀ
dihydropyrimidinobenzothiophenes (administered i.p. 15 min prior to testing). All signatures shown are at
a 30 mg/kg dose.
Based on its in vitro profile, compound 23 was selected from the benzothiophene series
for further evaluation. After chiral separation, it was determined that the antipsychotic
signature largely resided with enantiomer 23a which exhibited a highꢀdose antipsychotic
(dark purple) signature at 30 mg/kg and the more commonly encountered antipsychotic
(purple) signature at 10 mg/kg (Figure 8). The class signature changes that were
observed for 23a in going from a 10 mg/kg dose to a 3 mg/kg dose was not uncommon in
our experience. The reasons for this are not known but may be attributable to the nature
of in vivo assays. Compound 23a is an example of how a signature can become unstable
near the MED. As the signal becomes weaker, the test animals’ behavior becomes more
vehicleꢀlike as indicated by the white bar. We believe that this could lead to signatures
becoming less reproducible.
9

MedChemComm
Page 10 of 27
View Article Online
DOI: 10.1039/C6MD00128A
Figure 8. Structure and SmartCube® class signatures of 23 (racemate), 23a (enatiomer 1) and 23b
(enatiomer 2) at multiple doses (administered i.p. 15 min prior to testing). Absolute stereochemistry was
not determined.
We were pleased to find that, in contrast to 12a, 23a (free base or HCl salt) was
determined to be stable at room temperature (see supplemental material) with no
detectable amount of the aromatized pyrimidine 37 observed in the chemical stability
studies. Formation of 37 was also not observed in mouse, rat or human liver microsomal
assays. This improved chemical stability led to the selection of 23a for additional
profiling and the results are summarized in Table 2. This is a very small molecule
(MW=228) with excellent physicochemical properties and an acceptable in vitro ADME
profile. Receptor panel profiling confirmed that there was no detectable D2 receptor
binding at 10 µM, although binding to other receptors was observed. Several of these
have been associated with molecular mechanismsꢀofꢀaction (MMOA) of known
antipsychotics. 24 Most significantly, it was determined that 23a was a 5ꢀHT2A antagonist
(IC50=356 nM).
10

Page 11 of 27
MedChemComm
View Article Online
DOI: 10.1039/C6MD00128A
Table 2. In vitro profile of 23a. Assays are commercially available from CEREP S.A.32 (in vitro
pharmacology), or Cyprotex31 (in vitro ADMET and physicochemical); (h) = human, (r) = rat, (m) =
mouse.
Compound 23a was also determined to be orally active in two rodent tests associated
with antipsychotic behavior. The reversal of stimulantꢀinduced hyperlocomotor activity
is commonly used to assess antipsychotic activity in preclinical models.35 Similar to the
clozapine positive control, 23a (3, 10 and 30 mg/kg, p.o.) reversed phencyclidine (PCP)ꢀ
induced hyperlocomotor activity in mice (Figure 9A). Although, baseline locomotor
activity was also decreased at the two highest doses of 23a (10 and 30 mg/kg), there was
a separation of dose for efficacy (3 mg/kg) that did not alter baseline activity, an effect
also reported for some other antipsychotics.35 Prepulse inhibition of acoustic startle (PPI)
can be used to assess sensory gating mechanisms36 and PPI deficits are widely seen in
schizophrenia.37 Treatment of mice with 23a (3 and 10 mg/kg, p.o.) significantly
increased PPI compared to vehicle (Figure 9B).
A
B
Figure 9.
A. Effects of 23a (p.o. administration) in mice on baseline locomotor activity and after a PCP challenge (5
mg/kg). Clozapine was used as a positive control.
B. Effects of compound 23a (p.o. administration) on preꢀpulse inhibition in mice. Haloperidol was used as
a positive control.
Conclusions
A mouseꢀbased behavioral platform, the SmartCube® system, was successfully used in a
phenotypic screening campaign leading to the identification of a hit (compound 1). Lead
optimization of 1 was then efficiently conducted using the same phenotypic assay to
afford compound 23 having a strong antipsychotic signature. The more active
enantiomer, 23a, was found to have good chemical stability, good in vitro ADME and
DMPK profiles and excellent CNS drugꢀlike properties. Furthermore, this compound
exhibited potent oral activity in two rodent models relevant to schizophrenia despite
having no D2 receptor activity. The absence of any D2 activity is important considering
that the MMOA of all approved drugs for schizophrenia includes inhibition of the D2
receptor, making compound 23a a potentially useful research tool. Throughout the
course of this program, no effort was made to understand the MMOA associated with 23a
11

MedChemComm
Page 12 of 27
View Article Online
DOI: 10.1039/C6MD00128A
but we believe that it may likely be polypharmacological in nature and possibly includes
5ꢀHT2A receptor antagonism. These results support the utility of the SmartCube® system
as a tool for PDD and demonstrate that a molecular target need not be a required part of
the drug discovery process.
Experimental Section: In vivo Pharmacology
All housing and testing was in accordance with the Principles of Laboratory Animal
Care and the approval of the PsychoGenics, Inc., Institutional Animal Care and Use
Committee in AAALACꢀaccredited facilities. Experimental details are described in the
supplementary.
Supplementary Data
Supplementary data associated with this article can be found in the online version.
12

Page 13 of 27
MedChemComm
View Article Online
DOI: 10.1039/C6MD00128A
1. K.I. Kaitin, Clin. Pharmacol. Ther., 2010, 87, 356ꢀ361.
2. J. A. DiMasi, H. G. Grabowski, and R. W. Hansen, Tufts Center for the Study of Drug
Development, 2014, November 18.
(http://csdd.tufts.edu/news/complete_story/cost_study_press _event_webcast).
3. J. A. DiMasi, Tufts Center for the Study of Drug Development Impact Report 2014,
16(6). (http://csdd.tufts.edu/files/uploads/Summary__NovDecImpactIR.pdf).
4. S. M Paul, D.S. Mytelka, C. T. Dunwiddie, C.C. Persinger, B.H. Munos, S. R.
Lindborg, and A. L. Schacht, Nature Rev. Drug Discovery, 2010, 9, 203ꢀ214.
5. G. M. Maggiora, J. Comput. Aided Mol. Des., 2011, 25, 699ꢀ708.
6. D. C. Swinney, and J. Anthony, Nature Rev. Drug Discovery, 2011, 10, 507ꢀ519.
7. J. Eder, R. Sedrani, and C. Wiesmann, Nature Rev., Drug Discovery 2014, 13, 577ꢀ
587.
8. J. A. Lee, M. T. Uhlik, C. M. Moxham , D. Tomandl, and D. J. Sall, J. Med. Chem.,
2012, 55, 4527ꢀ4538.
9. W. Zheng, N. Thorne, and J. C. McKew, Drug Discov. Today, 2013, 18, 1067ꢀ1073.
10. D. F. Tardiff, and S. Lindquist, Drug Discov. Today Technol., 2013, 10, e121ꢀe128.
11. T. H. Keller, PꢀY. Shi, and QꢀY. Wang, Curr. Opin. Chem. Biol., 2011, 15, 529ꢀ533.
12. P. Brodin, and T. Christophe, Curr. Opin. Chem. Biol., 2011, 15, 534ꢀ539.
13. M. L. Sykes, and V. M. Avery, J. Med. Chem., 2013, 56, 7727ꢀ7740.
14. J. G. Moffat, J. Rudolph, and D. Bailey, Nature Rev. Drug Discovery, 2014, 13, 588ꢀ
602.
15. D. Kokel, J. Bryan, C. Laggner, R. White, C. Y. J. Cheung, R. Mateus, D. Healey, S.
Kim, A. A. Werdich, S. J. Haggarty, C. A. MacRae, B. Schoichet, and R. T. Peterson,
Nature Chem. Biol., 2010, 6, 231ꢀ237.
16. M. Prior, C. Chiruta, A. Curraia, J. Goldberg, J. Ramsey, R. Dargusch, P. A. Maher,
and D. Schubert, ACS Chem. Neurosci., 2014, 5, 503ꢀ513.
17. R. A. Paveley, N. R. Mansour, I. Hallyburton, L. S. Bleicher, A. E. Benn, I. Mikic, A.
Guidi, I. H. Gilbert, A. L. Hopkins, and Q. D. Bickle, PLoS Negl. Trop. Dis., 2012, 6,
e1762.
18. S. Dandapani, A. R. Germain, I. Jewett, S. le Quement, J. C. Marie, G. Muncipinto, J.
R. Duvall, L. C. Carmody, J. R. Perez, J. C. Engel, J. Gut, D. Kellar, J. L. Siqueiraꢀ
Neto, J. H. McKerrow, M. Kaiser, A. Rodriguez, M. A. Palmer, M. Foley, S. L.
Schreiber, and B. Munoz, ACS Med. Chem. Lett. 2013, 5, 149ꢀ153.
19. L. P. O’Reilly, C. J. Luke, D. H. Perlmutter, G. A. Silverman, and S. C. Pak, Adv.
Drug Deliv. Rev., 2014, 69-70, 247ꢀ253.
20. A. L. Lublin, and C. D. Link, Drug Discov. Today Technol., 2013, 10, e115ꢀe119.
21. R. E. Kalin, N. E. BanzingerꢀTobler, M. Detmar, and A. W. Brandli, Blood, 2009,
114, 1110ꢀ1122.
22. S. M. Schmitt, M. Gull, and A. W. Brandli, Adv. Drug Deliv. Rev., 2014, 69-70, 225ꢀ
246.
23. N. Ginovart, and S. Kapur, Handb. Exp. Pharmacol., 2012, 212, 27ꢀ52.
24. S. Miyamoto, G. E. Duncan, C. E. Marx, and J. A. Lieberman, Mol. Psychiatry 2005,
10, 79ꢀ104.
25. V. Alexandrov, D. Brunner, T. Hanania, and E. Leahy, Eur. J. Pharmacol., 2015,
750, 82ꢀ89.
13

MedChemComm
Page 14 of 27
View Article Online
DOI: 10.1039/C6MD00128A
26. S. L. Roberds, I. Filippov, V. Alexandrov, T. Hanania, and D. Brunner, Front.
Neurosci., 2011, 5, article 103.
27. D. Brunner, V. Alexandrov, B. Caldarone, T. Hanania, D. Lowe, J. Schneider and J.
Chandrasekhar, Behavior-Based Screening as an Approach to Polypharmacological
Ligands, in Polypharmacology in Drug Discovery (ed J.-U. Peters), John Wiley &
Sons, Inc., Hoboken, NJ, USA, 2012, chapter 14, doi: 10.1002/9781118098141.
28. C.ꢀC. Chang and C.ꢀJ. Lin, ACM Transactions on Intelligent Systems and
Technology, 2011, 2 (3), article 27, doi: 10.1145/1961189.1961199.
29. D. F. Specht, Neural Networks, 1990, 3, 109ꢀ118.
30. For all SmartCube® studies, compounds were formulated in a cocktail consisting of
5% 1ꢀmethylꢀ2ꢀpyrrolidone, 30% 1:1:1 solution of PEG200, PEG400, propylene
glycol, and 65% saline. Additionally, PsychoGenics’ standard operating procedures
include the routine testing of reference compounds to confirm signature. For example,
haloperidol has been run multiple times. The antipsychotic signature was consistent
and no drift was seen.
31. In vitro ADME, toxicology and physicochemical profiling assays were all selected
from the Cyprotex (Watertown, MA, USA) services catalog
(http://www.cyprotex.com/services) and used without modification.
32. A customized Cerep S. A. (Poitiers, France) panel was used that included the
“ExpresSProfile” of 55 targets
(http://www.cerep.fr/cerep/users/pages/catalog/profiles/catalog.asp). Screening
concentrations of 10 µM or 30 µM were used. See the supplementary material for the
full 52 assay list.
33. (a) E. V. Verbitskiy, E. M. Cheprakova, P. A. Slepukhin, M. I. Kodess, M. A.
Ezhikova, M. G. Pervova, G. L. Rusinov, O. N. Chupakhin, and V. N. Charushin,
Tetrahedron 2012, 68, 5445–5452. (b)W. P. K. Girke, Chem. Ber., 1979, 112, 1ꢀ15.
(c) W. P. K. Girke, Chem. Ber., 1979, 112, 1348ꢀ1358.
34. J. E. Campbell, U. Campbell, T. G. Hanania, and L. Shao, WO 2013119895 A1.
35. M. A. Geyer, and B. Ellenbroek, Prog. Neuro-PyschopharmBiol. Psych., 2003, 27,
1071ꢀ1079.
36. M. A. Geyer, K. KrebsꢀThomson, D. L. Braff, and N. Swerdlow, Psychopharmaol.
2001, 156, 117ꢀ154.
37. N. R. Swerdlow, G. A. Light, K. S. Cadenhead, J. Sprock, M. Hsieh, and D. L. Braff,
Arch. Gen. Psychiatry, 2006, 63, 1325ꢀ1335.
14

Page 15 of 27
MedChemComm
Figure 1
100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%
Compound 1
10
30
Dose
(mg/kg)

MedChemComm
Page 16 of 27
Table 1
CYP450 Inhibition IC ( M)
Plasma Protein Binding
Fraction bound (%)
Microsomal Stability
t1/2 min
hERG
Inhibition
m
50
CYP1A CYP2C19CYP2C9CYP2D6 CYP3A4
IC ( M)
m
50
Human Mouse
100 76
Rat
Human Mouse
48 155
Rat
>25
>25
12.7
>25
6.3
19.8
70
39

Page 17 of 27
MedChemComm
Scheme 1

MedChemComm
Page 18 of 27
Fig 2
2
3
4 5
6
7
8
9
100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%
Compound
Structure
Compound
Structure
2
6
3
7
4
8
5
9
30 30 25 30 25 30 30 30
Dose (mg/kg)

Page 19 of 27
MedChemComm
Figure 3
Compound
Structure
10
11
12
100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%
MED
10
30
3
Compound
13
14
15
Structure
MED
30 30 30 30 30 30
Dose (mg/kg)
30
3
10

MedChemComm
Page 20 of 27
Figure 4
12
12a
12b
16
12(a, b)
16
Dose (mg/kg)

Page 21 of 27
MedChemComm
fig 5
17 18 19 20 21 22
100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%
Compound
17
18
19
20
21
22
Structure
30 30 30 30 30 30
Dose (mg/kg)

MedChemComm
Page 22 of 27
Fig. 6
Compound
Structure
23
24
25
26
23 24 25 26 27 28 29 30
100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%
MED
3
3
3
10
Compound
27
28
29
30
Structure
MED
25 30 30 30 30 30 30 30
Dose (mg/kg)
30
30
10
10

Page 23 of 27
MedChemComm
Fig 7
Compound
Structure
31
32
33
31 32 33 34 35 36
100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%
MED
30
30
30
Compound
34
35
36
Structure
MED
30
30
25
30
30
30
Dose (mg/kg)
30
30
30

MedChemComm
Page 24 of 27
Fig 8
23
23a
23b
R =
23 and 23(a, b)
R =
Dose (mg/kg)
24

Page 25 of 27
MedChemComm
Table 2
IN vitro Pharmacology
Found for Compound 23a
Binding (%)
D2 (0), a1 (88); a2 (97); D1 (77); 5-HT3 (64); 5-HT6
(61); 5-HT7 (94); H1 (86); 5-HT trans.(94); DA trans.
(78); NE trans. (51); 5-HT2B (81); 5-HT2A (99),
Anti-Target Profile
Functional Activity (%)
5HT2A: IC50 = 356 nM
ADMET/PK
CYP profile
5 isozymes (mM)
2C19 (25); 2D6 (3.1); 2C9 (6.9);1A (6.9); 3A4 (11)
LM profile
hERG
t1/2 (min)
137 (h), 29(r), 242(m)
(mM)
>25
Physicochemical
Solubility
cLogP
pH 7.4
>100 mg/mL
1.9
Protein Binding
30 (h), 18 (m)
% Fu

MedChemComm
Page 26 of 27
Fig 9

Page 27 of 27
MedChemComm
View Article Online
DOI: 10.1039/C6MD00128A
A mouse-based high-dimensional assay, the SmartCube, was used for both screening and lead
optimization of a novel antipsychotic.