Structure-Based Optimization of a Novel Class of Aldehyde Dehydrogenase 1A (ALDH1A) Subfamily-Selective Inhibitors as Potential Adjuncts to Ovarian Cancer Chemotherapy

Article abstract of DOI:10.1021/acs.jmedchem.8b00930

Aldehyde dehydrogenase (ALDH) activity is commonly used as a marker to identify cancer stem-like cells. The three ALDH1A isoforms have all been individually implicated in cancer stem-like cells and in chemoresistance; however, which isoform is preferentially expressed varies between cell lines. We sought to explore the structural determinants of ALDH1A isoform selectivity in a series of small-molecule inhibitors in support of research into the role of ALDH1A in cancer stem cells. An SAR campaign guided by a cocrystal structure of the HTS hit CM39 (7) with ALDH1A1 afforded first-in-class inhibitors of the ALDH1A subfamily with excellent selectivity over the homologous ALDH2 isoform. We also discovered the first reported modestly selective single isoform 1A2 and 1A3 inhibitors. Two compounds, 13g and 13h, depleted the CD133⁺ putative cancer stem cell pool, synergized with cisplatin, and achieved efficacious concentrations in vivo following IP administration. Compound 13h additionally synergized with cisplatin in a patient-derived ovarian cancer spheroid model.

Full text of DOI:10.1021/acs.jmedchem.8b00930

Subscriber access provided by Kaohsiung Medical University
Article
Structure-Based Optimization of a Novel Class of Aldehyde
Dehydrogenase 1A (ALDH1A) Subfamily-Selective Inhibitors
as Potential Adjuncts to Ovarian Cancer Chemotherapy
Brandt C Huddle, Edward Grimley, Cameron D Buchman, Mikhail Chtcherbinine, Bikash Debnath,
Pooja Mehta, Kun Yang, Cynthia A Morgan, Siwei Li, Jeremy Antonio Felton, Duxin Sun,
Geeta Metha, Nouri Neamati, Ronald J Buckanovich, Thomas D Hurley, and Scott D Larsen
J. Med. Chem., Just Accepted Manuscript • DOI: 10.1021/acs.jmedchem.8b00930 • Publication Date (Web): 17 Sep 2018
Downloaded from http://pubs.acs.org on September 17, 2018
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 61
Journal of Medicinal Chemistry
StructureꢀBased Optimization of a Novel Class of Aldehyde
1
2
3
4
5
6
7
8
9
Dehydrogenase 1A (ALDH1A) SubfamilyꢀSelective Inhibitors as
Potential Adjuncts to Ovarian Cancer Chemotherapy
1
1
1
1
1
1
1
1
1
1
2
2
2
2
2
2
2
2
2
2
3
3
3
3
3
3
3
3
3
3
4
4
4
4
4
4
4
4
4
4
5
5
5
5
5
5
5
5
5
5
6
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
†
&
ǂ&
‡
‡
†
Brandt C. Huddle, Edward Grimley, Cameron D. Buchman, Mikhail Chtcherbinine, Bikash Debnath,
§
+
‡
#
#
#
§, ^, %
Pooja Mehta, Kun Yang, Cynthia A. Morgan, Siwei Li, Jeremy Felton, Duxin Sun, Geeta Mehta,
†
,+,∫
‡
,†,¶
Nouri Neamati, Ronald J. Buckanovich,* Thomas D. Hurley, and Scott D. Larsen*
¶
†
Vahlteich Medicinal Chemistry Core and Department of Medicinal Chemistry, College of Pharmacy;
University of Michigan, Ann Arbor, Michigan 48109
‡
Department of Biochemistry and Molecular Biology; Indiana University School of Medicine, Indianapolis,
Indiana 46202
+
;
Division of Hematology Oncology, Department of Internal Medicine University of Michigan, Ann Arbor,
Michigan 48109
∫
ǂ
Division of HematologyꢀOncology, Department of Internal Medicine, Department of Obstetrics, Gynecology,
and Reproductive Sciences, University of Pittsburgh Medical Center and the Magee Womens Research
Institute, Pittsburgh PA 15213
§
Department of Materials Science Engineering, University of Michigan, Ann Arbor, Michigan 48109
^
Department of Biomedical Engineering, University of Michigan, Ann Arbor, Michigan 48109
%
Macromolecular Science and Engineering, University of Michigan, Ann Arbor, Michigan 48109
#
Department of Pharmaceutical Sciences, College of Pharmacy; University of Michigan, Ann Arbor,
Michigan 48109
&
Indicates equal contribution
1
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 2 of 61
Abstract: Aldehyde dehydrogenase (ALDH) activity is commonly used as a marker to identify cancer stemꢀlike cells.
The three ALDH1A isoforms have all been individually implicated in cancer stemꢀlike cells and in chemoresistance;
however, which isoform is preferentially expressed varies between cell lines. We sought to explore the structural
determinants of ALDH1A isoform selectivity in a series of small molecule inhibitors in support of research into the role of
ALDH1A in cancer stem cells. An SAR campaign guided by a coꢀcrystal structure of the HTS hit CM39 (7) with
ALDH1A1 afforded firstꢀinꢀclass inhibitors of the ALDH1A subfamily with excellent selectivity over the homologous
ALDH2 isoform. We also discovered the first reported modestly selective single isoform 1A2 and 1A3 inhibitors. Two
1
2
3
4
5
6
7
8
9
1
1
1
1
1
1
1
1
1
1
2
2
2
2
2
2
2
2
2
2
3
3
3
3
3
3
3
3
3
3
4
4
4
4
4
4
4
4
4
4
5
5
5
5
5
5
5
5
5
5
6
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
+
compounds, 13g and 13h, depleted the CD133 putative cancer stem cell pool, synergized with cisplatin, and achieved
efficacious concentrations in vivo following IP administration. 13h additionally synergized with cisplatin in a patient
derived ovarian cancer spheroid model.
2
ACS Paragon Plus Environment

Page 3 of 61
Journal of Medicinal Chemistry
INTRODUCTION
1
2
3
4
5
6
7
8
9
1
1
1
1
1
1
1
1
1
1
2
2
2
2
2
2
2
2
2
2
3
3
3
3
3
3
3
3
3
3
4
4
4
4
4
4
4
4
4
4
5
5
5
5
5
5
5
5
5
5
6
Currently, more than half of the 21,400 U.S. women diagnosed with epithelial ovarian cancer (EOC)
1
annually are expected to succumb to the disease within 5 years. The first line therapy for the majority of EOC
cases is surgical debulking of the primary tumor with adjuvant platinumꢀ and taxaneꢀbased chemotherapeutics
2ꢀ3
to treat the residual disease. Approximately 70% of EOC patients are initially responsive to
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
4
chemotherapeutics; however, most relapse and ultimately become unresponsive to further chemotherapy.
EOC tumors contain a hierarchy of heterogeneous cells consistent with the Cancer Stem Cell (CSC)
5ꢀ8
hypothesis. New therapies which target these CSCs may improve patient outcomes alone or when combined
9
5-7, 10-12
with chemotherapy. In EOC, elevated aldehyde dehydrogenase (ALDH) activity is a marker of CSCs.
bright
ALDH
cells (the subꢀpopulation of cells with the greatest activity in the ALDEFLUOR assay) are more
dim
tumorigenic and chemoresistant than ALDH cells, are more prevalent in chemoresistant tumors following
6
ꢀ8, 12
chemotherapy, and their presence within a tumor is predictive of poorer patient outcomes.
There are 19 distinct genes for ALDH superꢀfamily members in humans. The primary function of
ALDH is to oxidize endogenous aldehydes generated through various cellular processes to the corresponding
carboxylic acids. In addition to neutralization of these reactive species, the 3 members of the ALDH1A
subfamily also function in cellular signaling by generating the nuclear hormone allꢀtrans retinoic acid (ATRA)
1
3
from retinal. In some solid tumors, ATRA has been shown to activate transcription of oncogenes such as cꢀ
14
MYC, PDKꢀ1, and cyclin D1. Although the strongest body of evidence supports the role of ALDH1A1 (1A1)
1
5
in CSCs, other isoforms of the ALDH1A family are often simultaneously expressed.
Given that ALDH plays a potentially critical role in CSCs, inhibition of ALDH is a potential strategy to
target CSC and reverse resistance to chemotherapy. Indeed, knockdown or inhibition of 1A1 increases
5
, 16ꢀ21
chemosensitivity in ovarian and other cancers.
ALDH1A2 (1A2) and ALDH1A3 (1A3) have similarly
2
2ꢀ23
been implicated in chemoresistance in other tumor types.
Some ALDH isoforms, including 1A1, are able to
divert cyclophosphamide metabolism, preventing generation of the active phosphoramide mustard; however,
3
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 4 of 61
their ability to attenuate the effects of other chemotherapeutics such as cisplatin and paclitaxel is poorly
1
2
3
4
5
6
7
8
9
1
1
1
1
1
1
1
1
1
1
2
2
2
2
2
2
2
2
2
2
3
3
3
3
3
3
3
3
3
3
4
4
4
4
4
4
4
4
4
4
5
5
5
5
5
5
5
5
5
5
6
1
5
understood.
A panel of cellꢀpermeable, single or dual isoformꢀselective inhibitors for 1A1, 1A2, and 1A3 could have
utility as probes for dissecting the role of the various isoforms in any number of applications which currently
rely on siRNA knockdown. Various tumors and cancer cell lines differ in which ALDH1A isoforms are highly
expressed. Importantly, high grade serous ovarian tumors, the histologic subtype responsible for 70ꢀ80% of
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
EOC deaths, show strong elevation of 1A3 expression but only modest elevation in 1A1 expression. Other
24ꢀ25
histologic subtypes show significant elevation of both isoforms.
The inability of existing 1A1 selective
2
6
inhibitors to inhibit ALDEFLUOR in 1A3 high cell lines has previously been demonstrated.
Figure 1. Representative ALDH inhibitors 1 - 6 and lead compound 7 (CM39)
4
ACS Paragon Plus Environment

Page 5 of 61
Journal of Medicinal Chemistry
1
3, 15
There are a number of small molecule ALDH inhibitors reported in the literature (Figure 1).
Many
1
2
bear an electrophilic warhead and rely on reversible or irreversible covalent interaction with the ALDH catalytic
cysteine to achieve potency. Design of highly isoform selective compounds employing these warheads is
complicated by the presence of this critical cysteine throughout the ALDH family. One of most widely studied
ALDH inhibitors, DEAB (1), inhibits at least 6 isoforms of ALDH with an IC <15 µM and is a substrate of at
3
4
5
6
7
8
9
1
1
1
1
1
1
1
1
1
1
2
2
2
2
2
2
2
2
2
2
3
3
3
3
3
3
3
3
3
3
4
4
4
4
4
4
4
4
4
4
5
5
5
5
5
5
5
5
5
5
6
5
0
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
2
7
least 5 isoforms, including 1A1. Compound 1 requires concentrations of ~100 µM to induce
28
chemosensitization of CSCs. Disulfiram (2), a nonꢀselective covalent ALDH inhibitor approved for treatment
13
of alcoholism, is rapidly metabolized into several covalent ALDH inhibitors with variable isoform selectivity.
While 2 is also reported to deplete CSCs in combination with chemotherapy, this effect likely does not result
2
9ꢀ31
from its action on ALDH.
Win 18,446 (3) is a potent inhibitor of the ALDH1A subfamily (1A1, 1A2, 1A3
32
K = 285, 56, and 261 nM respectively). Unfortunately, 3 also inhibits ALDH2 and is a known teratogen due to
i
3
3
its ability to chelate zinc, making it an imperfect tool for the study of ALDH1A biology. Hurley and
colleagues disclosed the nonꢀcovalent 1A1 inhibitor high throughput screening (HTS) hits 4 and 5a along with
3
4ꢀ35
36
coꢀcrystal structures.
Soon thereafter, compound 5b was disclosed elsewhere. Recently, a scaffoldꢀ
hopping approach to improve the cell permeability of 5b yielded compound 6 and other similar analogs.
Compound 6 exhibits potent and highly selective 1A1 inhibition (1A1 IC = 7 nM; 1A2, 1A3, ALDH2,
50
ALDH3A1 IC50s > 1µM ) and potent inhibition of ALDEFLUOR in MIA PaCaꢀ2 cells, which preferentially
express 1A1. Furthermore, this series was able to sensitize a paclitaxel resistant ovarian cancer cell line to
3
7
13, 15
treatment with paclitaxel. To our knowledge there are no selective inhibitors of 1A2 or 1A3.
The high throughput screen which identified 4 and 5a also yielded the pyrazolopyrimidinone CM39 (7).
This compound was an attractive lead for the development of ALDH1A subfamily inhibitors because it
exhibited reasonably ligandꢀefficient inhibition of 1A1 (IC = 0.9 µM, LE = 0.32) and good selectivity (<20%
5
0
3
4
inhibition of ALDH2 and 3A1 at 20uM). A further screen of selectivity showed ~50% inhibition of 1A2 and
A3 at 20 µM and little activity against other ALDH isoforms (vide infra). Following our successful acquisition
1
of an xꢀray crystal structure of 7 bound in the 1A1 active site (vide infra), we undertook the rational design and
5
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 6 of 61
synthesis of analogs. Herein we report the ability of these analogs to inhibit ALDH1A with widely varying
1
2
3
4
5
6
7
8
9
1
1
1
1
1
1
1
1
1
1
2
2
2
2
2
2
2
2
2
2
3
3
3
3
3
3
3
3
3
3
4
4
4
4
4
4
4
4
4
4
5
5
5
5
5
5
5
5
5
5
6
selectivity profiles. We also examine the ability of selected compounds to inhibit cellular ALDH function using
+
ALDEFLUOR, to deplete the CD133 putative cancer stem cell pool, and to synergize with cisplatin in two
ovarian cancer cell lines and in patient derived ovarian cancer spheroids. Aqueous solubility and metabolic
stability were assessed for select compounds to guide future optimization of in vivo probes.
RESULTS and DISCUSSION
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
Chemistry: As shown in Scheme 1, synthesis of compounds 7, 11, 12aꢀd,f-h, and 13aꢀh began with the
38
commercially available pyrazole 8a. Treatment of 8a with the appropriate aryl isothiocyanate in refluxing
toluene afforded the thioureas 9a and 9b which precipitated from the reaction mixture upon cooling. Ring
closure to generate thiopyrimidinones 10a and 10b proceeded in aqueous NaOH at reflux. Selective Sꢀ
alkylation with 3ꢀfluorobenzyl bromide yielded compounds 7 and 11. Subsequent Nꢀalkylation resulted in a
mixture of regioisomers 12aꢀd,f-h and 13aꢀh which were easily separated by flash chromatography on silica.
Regioisomeric assignment was made by NOE; in all cases the 1ꢀalkylꢀpyrazoles 12 eluted before the
corresponding 2ꢀalkylꢀpyrazoles 13.
a
Scheme 1. Synthesis of compounds 7, 11, 12a-d,f-h, and 13a-h.
a
Reagents and conditions: (a) ArNCS, toluene, reflux; (b) 1N NaOH, reflux; (c) 3ꢀFꢀPhCH Br, K CO ,
2
2
3
1
DMF, RT; (d) R ꢀX, K CO , DMF, 50°C X = Br, OMs
2
3
6
ACS Paragon Plus Environment

Page 7 of 61
Journal of Medicinal Chemistry
1
2
3
4
In a departure from our earlier synthetic strategy, we envisioned an alternative route to 12e and 15aꢀm
3
5
6
7
8
9
1
1
1
1
1
1
1
1
1
1
2
2
2
2
2
2
2
2
2
2
3
3
3
3
3
3
3
3
3
3
4
4
4
4
4
4
4
4
4
4
5
5
5
5
5
5
5
5
5
5
6
that would use readily available amines rather than the corresponding isothiocyanates to introduce diverse R
substituents as shown in Scheme 2. The synthesis of 14 has been previously accomplished by treating
commercially available pyrazole 8b with thiophosgene under SchottenꢀBaumann conditions; however, we
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
3
9
found the low yield, difficult purification, and use of noxious reagents to be unsatisfactory. After screening
many reported isothiocyanateꢀforming conditions employing carbon disulfide as the thiocarbonyl source, we
found that treatment of 8b with NaH and CS followed by the slow addition of I afforded 14 in good yield
2
2
4
0
following a simple purification. Isothiocyanate 14 reacted with amines rapidly under mild conditions to form
thioureas. Subsequent addition of NaH in the same pot facilitated cyclization to the thiopyrimidinone. For the
majority of arylamines, adding 3ꢀfluorobenzyl bromide at this point completed the transformation of 14 to 12e,
1
5aꢀi in one pot. For alkylamines and 2ꢀaminopyridine (15j-l), an aqueous workup prior to Sꢀalkylation with 3ꢀ
fluorobenzyl bromide reduced the formation of several unidentified side products. The thiourea resulting from
treatment of 14 with 8N ethanolic methylamine spontaneously cyclized to form the thiopyrimidinone.
Following removal of volatiles, Sꢀalkylation of this intermediate with 3ꢀfluorobenzyl bromide under basic
conditions afforded 15m.
a
Scheme 2. Synthesis of analogs 12e, 15aꢀm.
a
3
Reagents and conditions: (a) NaH, CS , I , THF, 0ꢀ40°C; (b) R ꢀNH , 3ꢀFꢀPhCH Br,
2
2
2
2
3
NaH, DMF, 0ꢀ20°C; (c) (i) R ꢀNH , NaH, DMF, 0ꢀ20°C; (ii) 3ꢀFꢀPhCH Br, K CO ,
2
2
2
3
DMF, RT; (d) (i) MeNH , EtOH; (ii) 3ꢀFꢀPhCH Br, K CO , DMF, RT
2
2
2
3
Characterization of lead 7: To establish a benchmark and guide design of new analogs, we further
characterized compound 7. As shown in Figure 2A, 7 has excellent selectivity for 1A1 vs. 2, 1B1, 3A1, 4A1,
7
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 8 of 61
5
A1, and 1L1 yet modestly inhibits 1A2, 1A3. Despite the presence of a potentially electrophilic isothiourea, 7ꢀ
1
2
3
4
5
6
7
8
9
1
1
1
1
1
1
1
1
1
1
2
2
2
2
2
2
2
2
2
2
3
3
3
3
3
3
3
3
3
3
4
4
4
4
4
4
4
4
4
4
5
5
5
5
5
5
5
5
5
5
6
mediated inhibition of 1A1 was fully reversed following gel filtration, indicating it is nonꢀcovalent
(Supplemental Figure S1). Similar to other nonꢀcovalent inhibitors of 1A1, 7 was found to be nonꢀcompetitive
with respect to the substrate acetaldehyde with a K = 0.38 ± 0.05 µM and uncompetitive with respect to
i
+ 35
NAD . Liabilities of the lead with regard to its potential as a therapeutic included a halfꢀlife of 9 minutes in
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
the presence of MLMs and poor aqueous solubility (2 µM), both of which could be attributed to high
4
1
lipophilicity (cLogP of 5.2). As shown in Figure 2B, compound 7 was inferior to the widelyꢀused
Bright
ALDEFLUOR control compound DEAB (1) at reducing ALDH
PEO1 cells at concentrations up to 30 µM.
This result suggested that ALDH1A1 selective inhibition is not sufficient to inhibit ALDEFLUOR in this 1A3
high cell line. It is important to note that while DEAB is a 57nM 1A1 selective inhibitor, it only inhibits
27
ALDEFLUOR above its 1A3 IC (3 µM). More potent ALDEFLUOR inhibition for DEAB in two 1A1 high
5
0
2
6
cell lines, MIA PaCaꢀ2 (IC 3.4 µM) and HTꢀ29 (IC 1.7 µM), has previously been reported.
50
50
A
B
Figure 2. (A) Fraction of enzyme activity remaining for 9 ALDH isoforms in the presence of 20 µM 7. (B)
Percent control ALDH
DEAB (1) or compound 7 (N=2).
Bright
cells remaining following treatment of PEO1 cells at the specified concentration with
Crystal structure of ALDH1A1 complexed with 7: To guide the design of new analogs, an Xꢀray crystal
structure of 7 bound to N121S ALDH1A1 was obtained with a resolution of 2.1Å (PDB code 5TEI, Figure 3A,
Table S1, Figure S2). The known single nucleotide polymorphism (NCBI rs1049981) resulting in the N121S
8
ACS Paragon Plus Environment

Page 9 of 61
Journal of Medicinal Chemistry
missense mutation has been found in a small percentage of the HapMapꢀCEU population. There is no known
1
2
3
4
5
6
7
8
9
1
1
1
1
1
1
1
1
1
1
2
2
2
2
2
2
2
2
2
2
3
3
3
3
3
3
3
3
3
3
4
4
4
4
4
4
4
4
4
4
5
5
5
5
5
5
5
5
5
5
6
3
4ꢀ35
clinical significance to the mutation and its structure is highly similar to the wildꢀtype.
The N121S enzyme
crystallizes under identical conditions to the WT enzyme, but complexes with inhibitors are frequently easier to
obtain, presumably due to dynamic fluctuations within the active site that are not detectable by crystallography
and have little impact on observed potency. Furthermore, the IC of 7 was comparable for the WT and N121S
5
0
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
(0.9 and 0.7 µM respectively, Table 1, Supplementary Figure S1) Similar to the complex with 5a (PDB code
4
WP7), the bicyclic ring binds across the exterior 1/3 of the substrate binding pocket. The 3ꢀfluorobenzyl group
extends in towards the catalytic nucleophile (C303) and makes aromatic contacts with F171, W178 and F466.
This binding mode is orthogonal to that for 4 (PDB code 4X4L), in which its tricyclic ring system extends
3
5
toward both C303 and W178 (Figure 3B). The core heterocycle of 7 binds similarly to the substituted xanthine
ring of 5a, binding beneath Y297 and occupying the additional active site volume afforded by the 1A1ꢀspecific
residue, G458. Other ALDH1/2 isoenzymes have either an Asn or Asp residue at position 458 which may drive
selectivity by prohibiting a similar binding mode in these isoforms. There appears to be room for small
substitutions on the pyrazole at Nꢀ1 but not Nꢀ2, suggesting that larger substituents at these positions will not be
tolerated without alteration of the binding pose. The lipophilic 2ꢀmethylphenyl ring binds near a relatively polar
region of the 1A1 active site, suggesting that appropriate polar substituents could improve or maintain binding
affinity while lowering cLogP. The 3ꢀfluorobenzyl ring attached to the sulfur projects into the deepest section
of the active site and maximizes contacts with the aromatic residues near the catalytic nucleophile, C303. The
apparent goodness of fit of the 3ꢀfluorobenzyl substituent, as well as its expected metabolic stability,
encouraged us to retain it for the initial set of analogs based on 7.
9
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 10 of 61
A
B
1
2
3
4
5
6
7
8
9
1
1
1
1
1
1
1
1
1
1
2
2
2
2
2
2
2
2
2
2
3
3
3
3
3
3
3
3
3
3
4
4
4
4
4
4
4
4
4
4
5
5
5
5
5
5
5
5
5
5
6
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
Figure 3. Structure of ALDH1A1 N121S with 7 (2.10 Å, PDB 5TEI). (A) The fluorophenyl of compound
projects into the lipophilic pocket towards the catalytic cysteine (C303). Key residues are labelled. (B)
7
Overlay of Compounds 4 (Cyan, PDB entry 4WP7), 5a (Grey, PDB Entry 4X4L), and 7(Orange) bound in
ALDH1A1.
10
ACS Paragon Plus Environment

Page 11 of 61
Journal of Medicinal Chemistry
Table 1. Characterization of N-1 alkylated analogs
1
2
3
4
5
6
7
8
9
1
1
1
1
1
1
1
1
1
1
2
2
2
2
2
2
2
2
2
2
3
3
3
3
3
3
3
3
3
3
4
4
4
4
4
4
4
4
4
4
5
5
5
5
5
5
5
5
5
5
6
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
a
b
c
Cmpd
No.
R
R
ALDH IC50 (µM) or % control at 20 µM
cLogP Aq. Sol.
(µM)
MLM
t1/2
(min)
1
A1
1A2
53%
1A3
53%
2
7
H
H
Me
H
0.9 ± 0.2
85%
94%
94%
100%
96%
97%
5.2
4.7
4.8
5.6
4.7
4.7
4.3
2
9
11
0.45 ± 0.03 0.67 ± 0.08 0.39 ± 0.02
1
2a
2b
Me
Me
Me
Me
Me
H
0.66 ± 0.06
0.18 ± 0.01
0.26± 0.03
0.42 ± 0.01
52%
>10
4 ± 1
>10
38
1
cꢀPrCH
2
1
2c
2d
2e
2f
3ꢀoxetanyl
3ꢀoxetanylꢀCH
Me
6 ± 1
45%
60%
58%
1
2
1
0.08 ± 0.01 0.15 ± 0.01 0.09 ± 0.01 100%
<0.7 d
BLQ)
8
(
1
cꢀPrCH
2
H
H
H
0.56 ± 0.03 1.26 ± 0.08 0.55 ± 0.02
0.13 ± 0.01 0.23 ± 0.01 0.17 ± 0.01
83%
94%
96%
5.1
4.1
4.1
1
2g
2h
3ꢀoxetanyl
8
1
3ꢀoxetanylꢀCH
2
0.34 ± 0.07
1.1 ± 0.1
0.61 ± 0.09
a
b
c
Values are expressed as Mean ± SEM (n=3), Mean (n=3); Thermodynamic solubility analysis was performed by
d
Analiza Inc. using quantitative nitrogen detection. (www.analiza.com); BLQ = below limit of quantitation.
Structure-Activity Relationships (SAR): All new compounds were tested for inhibition of ALDH1A1, 1A2,
1
A3, and 2. Selected compounds were also tested for thermodynamic aqueous solubility and stability in the
presence of mouse liver microsomes (MLM).
We began our SAR campaign by generating Nꢀalkylated analogs to fill the small lipophilic pocket
4
2
adjacent to Ile304. We also hypothesized that eliminating the moderately acidic (pKa = 8.1 ) pyrazole NꢀH
might improve cell permeability. Encouragingly, methylation at Nꢀ1 (Table 1, 12a) resulted in a modest
4
2
improvement in 1A1 inhibition compared to 7. Consistent with a reduction in cLogP aqueous solubility was
also improved. Installing the larger cyclopropylꢀmethyl substituent (12b) improved the 1A1 IC but did not
5
0
11
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 12 of 61
improve potency against 1A2 or 1A3. The unfavorable cLogP (5.6) of 12b encouraged us to explore less
1
2
3
4
5
6
7
8
9
1
1
1
1
1
1
1
1
1
1
2
2
2
2
2
2
2
2
2
2
3
3
3
3
3
3
3
3
3
3
4
4
4
4
4
4
4
4
4
4
5
5
5
5
5
5
5
5
5
5
6
lipophilic oxetanyl analogs 12c-d. 12c maintained comparable potency against 1A1 with substantially lower
cLogP (4.7). The homologated oxetane 12d was 2ꢀfold less potent against 1A1, suggesting that larger
substituents would not be tolerated.
In an effort to pare down the CM39 scaffold to the key pharmacophore, we discovered that removing
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
2
the R methyl markedly improved inhibition of 1A2 and 1A3 affording compound 11, which potently inhibited
all three ALDH1A isoforms but retained excellent selectivity over ALDH2. We revisited the panel of Nꢀ1
substituents (12e-h) and observed different SAR trends. The small Nꢀmethyl substituent of 12e afforded the
most potent of the four analogs tested but exhibited very poor solubility. Given the lower cLogP of 12e
compared to 7 or 12a, we believe the poor solubility is best explained by the increased planarity of 12e. The
MLM stability of 12e was unchanged relative to 7. The bulkier cyclopropylꢀmethyl and oxetanylꢀmethyl
substituents of 12f and 12h both resulted in 4ꢀfold or greater reduction in potency against the ALDH1A family
relative to 12e. The smaller oxetanyl substituent of 12g was only slightly less potent than 12e and afforded a
modest improvement in aqueous solubility. The divergent SAR trends for 12a-d compared to 12eꢀh suggested
that, upon removal of the oꢀmethyl, a new binding mode was accessible.
12
ACS Paragon Plus Environment

Page 13 of 61
Journal of Medicinal Chemistry
1
2
Table 2. Characterization of N-2 alkylated analogs
3
4
5
6
7
8
9
1
1
1
1
1
1
1
1
1
1
2
2
2
2
2
2
2
2
2
2
3
3
3
3
3
3
3
3
3
3
4
4
4
4
4
4
4
4
4
4
5
5
5
5
5
5
5
5
5
5
6
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
a
b
Cmpd
No.
R
R
ALDH IC50 (µM) or % control at 20 µM
cLogP Aq.
MLM
t1/2
c
Sol.
(µM) (min)
1
A1
1A2
1A3
2
1
3a
3b
Me
Me
Me
Me
Me
H
0.67 ± 0.07
0.10 ± 0.01
0.44 ± 0.06
0.20 ± 0.03
6 ± 1
2.2 ± 0.4
100%
100%
100%
98%
5.3
6.1
5.2
5.2
4.8
5.6
4.7
4.7
1
cꢀPrꢀCH
2
>10
8 ± 1
1.0 ± 0.1
3.9 ± 0.4
2.6 ± 0.3
13c
3d
3e
3f
3g
13h
3ꢀoxetanyl
3ꢀoxetanylꢀCH
Me
1
2
3.9 ± 0.3
1
0.25 ± 0.01 0.34 ± 0.05 0.13 ± 0.01 94%
0.13 ± 0.03 1.1 ± 0.2 0.17 ± 0.03 87%
69
13
1
cꢀPrꢀCH
2
H
1
3ꢀoxetanyl
H
0.08 ± 0.01 0.25 ± 0.04 0.12 ± 0.02 97%
0.27 ± 0.01 0.48 ± 0.04 0.13 ± 0.01 99%
5
23
27
3ꢀoxetanylꢀCH
H
53
2
a
b
c
Values are expressed as Mean ± SEM (n=3), Mean (n=3); Thermodynamic solubility analysis was performed
by Analiza Inc. using quantitative nitrogen detection. (www.analiza.com);
Nꢀ2 alkylated analogs were also tested to further probe the flexibility of the binding mode depicted in
the CM39 crystal structure (Table 2). 13a exhibited a similar ALDH1A inhibition profile to 12a. Extending to
bulkier substituents (13bꢀd) resulted in improved 1A1 potency with the cyclopropylꢀmethyl being optimal,
providing strong evidence for an alternative binding mode. 13a-d were more potent against 1A3 than the
corresponding Nꢀ1 regioisomers (12aꢀd), as exemplified by the modestly 1A1/1A3 selective cyclopropylꢀ
methyl analog 13b. As we observed previously, removal of the oꢀmethyl significantly improved the inhibition
of 1A2 and 1A3 in analogs 13e-h. Oxetane was the optimal Nꢀ2 substituent to achieve inhibition of all three
ALDH1A isoforms (13g), while cyclopropylꢀmethyl (13f) afforded a potent 1A1/1A3 inhibitor with >5ꢀfold
13
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 14 of 61
selectivity over 1A2. Despite having higher values for cLogP than the corresponding Nꢀ1 regioisomers, Nꢀ2
analogs were observed to be more polar by normal and reverse phase chromatography. Consistent with this
observation, the solubility of Nꢀ2 methyl analog 13e was substantially improved over regioisomer 12e. 13e was
also moderately more stable in the presence of MLMs, leading us to prioritize Nꢀ2 substituted analogs for
further pharmacokinetic studies. Curiously, appending the oxetanyl substituent at Nꢀ2 (13g) was not favorable
for solubility as was observed at Nꢀ1 (12g). Reasonable aqueous solubility was restored for the homologated
analog 13h. Perhaps resulting from steric blocking of Nꢀdealkylation, 13g and 13h were about 2ꢀfold more
stable in the MLM assay than 13e. Exquisite selectivity over the closely related ALDH2 isoform was
maintained for all tested analogs, positioning the series as a useful set of tool compounds to probe the biological
effect of ALDH1A inhibition.
1
2
3
4
5
6
7
8
9
1
1
1
1
1
1
1
1
1
1
2
2
2
2
2
2
2
2
2
2
3
3
3
3
3
3
3
3
3
3
4
4
4
4
4
4
4
4
4
4
5
5
5
5
5
5
5
5
5
5
6
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
14
ACS Paragon Plus Environment

Page 15 of 61
Journal of Medicinal Chemistry
1
2
Table 3. Characterization of N-phenyl modifications
3
4
5
6
7
8
9
1
1
1
1
1
1
1
1
1
1
2
2
2
2
2
2
2
2
2
2
3
3
3
3
3
3
3
3
3
3
4
4
4
4
4
4
4
4
4
4
5
5
5
5
5
5
5
5
5
5
6
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
3
a
b
Cmpd
No.
R
IC50 (µM) or % control at 20 µM
Aq.d
Sol.
ALDH
(
µM)
1
A1
1A2
1A3
4 ± 1
2
94%
1
1
1
1
2a
2e
5a
5b
2ꢀMeꢀPh 0.66 ± 0.06 52%
38
Ph
0.08 ± 0.01 0.15 ± 0.01
0.25 ±0.04 75%
0.09 ± 0.01
10 ± 2
100% <0.7
BLQ)
e
(
2ꢀClꢀPh
3ꢀClꢀPh
4ꢀClꢀPh
2ꢀFꢀPh
100%
4.1 ± 0.6
69%
0.56 ± 0.06 2.6 ± 0.3
100% 55
100%
90%
15c
1.1 ± 0.1
0.75 ± 0.05
c
c
1
1
1
1
1
5d
5e
5f
0.11 ± 0.01 8%
1%
2ꢀOMeꢀ
Ph
3ꢀOMeꢀ
Ph
4ꢀOMeꢀ
Ph
3ꢀPyridyl 3.83 ± 0.03 45%
0.97 ± 0.10 >10
>10
100%
82%
41%
1.3 ± 0.1
1.77 ± 0.09
44%
5g
5h
8.2 ± 0.7
0.78 ± 0.09
117%
60%
86%
99%
15i
15j
4ꢀPyridyl 52%
52%
64%
97%
2ꢀPyridyl 2.4 ± 0.4
67%
99%
1
1
1
5k
5l
Bn
0.89 ± 0.13 0.72 ± 0.04
>10
100%
87%
95%
5
3
PhꢀCH
CH2
Me
ꢀ
1.4 ± 0.15
0.9 ± 0.1
79%
0.18 ± 0.04
2
5m
4.1 ± 0.37
66%
a
b
c
Values are expressed as Mean ± SEM (n=3), Mean (n=3); Dose response curves
were not satisfactory for IC50 determination; Thermodynamic solubility analysis
was performed by Analiza Inc. using quantitative nitrogen detection.
www.analiza.com); BLQ = below limit of quantitation.
d
e
(
The substantial increase in inhibition across the ALDH1A subfamily upon removal of the oꢀmethyl in
3
R (12e) prompted us to further probe the effect of varied substitutions at this position as shown in Table 3. We
15
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 16 of 61
retained the Nꢀ1 methyl substituent of 12e for synthetic ease. Although we had evidence to suggest the binding
1
2
3
4
5
6
7
8
9
1
1
1
1
1
1
1
1
1
1
2
2
2
2
2
2
2
2
2
2
3
3
3
3
3
3
3
3
3
3
4
4
4
4
4
4
4
4
4
4
5
5
5
5
5
5
5
5
5
5
6
3
mode of 7 depicted in the crystal structure was flexible, we began by designing R analogs according to the
3
crystal structure. The R region of 7 is flanked by a region of exposed peptide backbone (residues 458ꢀ460) and
the sidechain of N121 (S121 in WT), suggesting that polar interactions in this region might improve binding
3
over the lipophilic 2ꢀmethylphenyl. Additionally, as the R methylꢀphenyl is partially solvent exposed, polar
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
solubilizing moieties were expected to improve pharmacokinetic properties without substantial loss of potency.
Unfortunately, all attempts to alter the phenyl present in 12e resulted in significant loss of activity across the
ALDH1A subfamily. Nevertheless, interesting selectivity trends did emerge. Substituents other than methyl at
the ortho position (Cl, OMe) also rendered compounds 1A1 selective, but decreased 1A1 potency (15a and 15e)
relative to proteo analog 12e. Chloro analog 15a showed a twofold improvement in potency (1A1 IC50 = 250
nM) compared to 12a. Because of the comparable size of Cl and CH , this difference suggested a preference
3
for electron withdrawing groups; however, the electronꢀdeficient fluorinated analog 15d was merely equipotent
with 12e despite the similar size of the substituents. Electronꢀrich methoxy analog 15e was fourꢀfold less potent
against 1A1 than 15a. Meta substituents rendered analogs 15b and 15f modestly 1A2 selective. Consistent with
our hypothesis that disruption of planarity is favorable for solubility in this series, 15b was considerably more
soluble than 12e. Addition of a chlorine in the para position resulted in a 1A2/1A3ꢀselective inhibitor vs 1A1
(15c), while para methoxy analog 15g exhibited very modest selectivity for 1A3. Pyridyl analogs 15hꢀj were
substantially less potent across the ALDH1A family, perhaps reflecting a desolvation penalty without formation
of productive hydrogen bonds within the active site. Benzyl analog 15k inhibited 1A1 and 1A2 with good
selectivity over 1A3. Homologation to phenylethyl analog 15l resulted in potent inhibition of 1A3 (IC = 180
5
0
nM) with more than 5ꢀfold selectivity over 1A1 and 1A2 and greater selectivity over other ALDH isoforms as
shown in Figure 4. Replacing the phenyl of 12e with a methyl group led to a large decrease in potency (15m).
3
The steep SAR at R , inconsistent with partial solvent exposure, was further evidence that the binding mode
depicted in the crystal structure of 7 was no longer relevant.
16
ACS Paragon Plus Environment

Page 17 of 61
Journal of Medicinal Chemistry
1
2
3
4
5
6
7
8
9
1
1
1
1
1
1
1
1
1
1
2
2
2
2
2
2
2
2
2
2
3
3
3
3
3
3
3
3
3
3
4
4
4
4
4
4
4
4
4
4
5
5
5
5
5
5
5
5
5
5
6
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
Figure 4. Fraction of enzyme activity remaining for 9 ALDH
isoforms in the presence of 20µM 15l.
Crystal structure of ALDH1A1 complexed with compound 13g: To elucidate the change in binding mode
we suspected, we obtained a coꢀcrystal structure of N121S ALDH1A1 with compound 13g resolved to 2.0Å
(PDB code 6DUM) (Figure 5, Table S1, Figure S3). The structure revealed a dramatically different binding
3
pose for 13g in which the R Nꢀphenyl is situated near the catalytic cysteine. The Nꢀphenyl is flanked by
Phe171 and Phe466, potentially explaining why steric bulk and polarity are not tolerated at the ortho and meta
positions (Table 3). It is more difficult to justify why para substituents are so poorly tolerated. In contrast to
the binding pose of compound 7, there is room for fairly large Nꢀ2 pyrazole substituents to extend across the
mouth of the active site. Consistent with this observation, larger substituents at Nꢀ2 improve 1A1 potency (13f
and 13g) relative to methyl analog 13e. As expected, given that large substituents at Nꢀ1 would project into
solvent, compounds 12f-h do not improve potency over methyl analog 12e.
17
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 18 of 61
1
2
3
4
5
6
7
8
9
1
1
1
1
1
1
1
1
1
1
2
2
2
2
2
2
2
2
2
2
3
3
3
3
3
3
3
3
3
3
4
4
4
4
4
4
4
4
4
4
5
5
5
5
5
5
5
5
5
5
6
A
B
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
Figure 5. (A) XꢀRay crystal structure of 13g bound to ALDH1A1 (PDB code 6DUM). (B) Overlay of compounds 7
(PDB code 5TEI) (Orange) and 13g (Green).
Homology Modeling: In order to better understand the SAR of these compounds, we performed a molecular
docking study of select compounds against all isoforms of the ALDH1A class. At the time this manuscript was
drafted, no high quality crystal structures were available for 1A2 and 1A3. Recently, Chen et al. disclosed high
4
3
quality structures of 1A2 with small molecules bound in the active site (e.g. PDB code 6B5H). We opted for
building our own homology models of 1A2 and 1A3 isoforms based on the PDB structure of ALDH1A1
reported here. Homology models were developed on SwissꢀModel Workspace using ProMod3. The quality of
the models was assessed by QMEAN scores as well as Ramachandran plots (Supplementary Figure S4,5).
Furthermore, when comparing all residues in our 1A2 homology model with PDB:6B5H, the average RMSD
was 0.76Å, indicative of an excellent homology model (Supplementary Figure S4). In order to prove
robustness of our docking methods, we docked coꢀcrystal ligands 7 and 13g into the binding site of ALDH1A1.
The docked poses for both ligands were in good agreement with the crystal structures with RMSD values of 0.8
and 1.0 Å for 7 and 13g, respectively (Supplementary Figure S6).
18
ACS Paragon Plus Environment

Page 19 of 61
Journal of Medicinal Chemistry
1
A
B
C
2
3
4
5
6
7
8
9
1
1
1
1
1
1
1
1
1
1
2
2
2
2
2
2
2
2
2
2
3
3
3
3
3
3
3
3
3
3
4
4
4
4
4
4
4
4
4
4
5
5
5
5
5
5
5
5
5
5
6
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
Fig 6. ALDH1A2 and 1A3 homology models provide rationale for isoform selectivity of Compound 7. (A) Val460 does
not clash with oꢀmethyl substituent of NꢀPhenyl in ALDH1A1ꢀ Compound 7 structure. (B, C) Docking poses of 7 in
ALDH1A2 and ALDH1A3, respectively. Red arrow indicates that in both ALDH1A2 and A3, 2ꢀmethyl substituent at
phenyl clashes with Leu477/471.
The molecular docking study of compound 7 in ALDH1A2 (Figure 6B) and 1A3 (Figure 6C) shows that
the 2ꢀmethyl substituent on the phenyl ring would clash with nonꢀconserved residues Leu477 (1A2) or Leu471
(
1A3) if it adopted the same binding pose as in 1A1. The corresponding residue Val460 in ALDH1A1 (Figure
A) avoids this steric clash with 2ꢀmethyl substituent. These observations agree with the observed 1A1
selectivity of 7.
6
19
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 20 of 61
1
2
3
4
5
6
7
8
9
1
1
1
1
1
1
1
1
1
1
2
2
2
2
2
2
2
2
2
2
3
3
3
3
3
3
3
3
3
3
4
4
4
4
4
4
4
4
4
4
5
5
5
5
5
5
5
5
5
5
6
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
Figure 7. Compound 11 docks without clashes into the active sites of ALDH1A1, 1A2 and 1A3. (A) Structure and ligand
interaction plot of Compound 11 docked into ALDH1A1 crystal structure. (B,C) Structure and ligand interaction plot of
compound 11 docked into ALDH1A2, 1A3 homology models respectively.
Desꢀmethyl compound 11, which is active against all ALDH1A isoforms, docks well into the active site
of all three ALDH1A proteins (Figure 7). For ALDH1A1, the 3ꢀfluorobenzyl ring engages in piꢀpi interactions
with both Phe171 and Trp178. The bicyclic ring makes a piꢀpi interaction with Tyr297 and van der Waals
20
ACS Paragon Plus Environment

Page 21 of 61
Journal of Medicinal Chemistry
interactions with Gly458 and Tyr297, while the phenyl ring participates in van der Waals interactions with
Val460 and Asn121. In the active site of ALDH1A2, the bicyclic ring forms piꢀpi interaction with Phe314 and a
van der Waals interaction with Asn475. The 3ꢀfluorobenzyl ring makes a piꢀpi interaction with Phe188 and van
der Waals hydrophobic interactions with Trp195, while the phenyl ring engages in van der Waals hydrophobic
interactions with Val138 and Leu477. In the case of ALDH1A3, the bicyclic ring of compound 11 makes a piꢀpi
interaction with Phe308 and a van der Waals interaction with Asn469. The 3ꢀfluorobenzyl ring forms a piꢀpi
interaction with Phe182 and a van der Waals hydrophobic interaction with Trp189, while the phenyl ring makes
van der Waals hydrophobic interactions with Ile132, Phe182 and Leu477.
1
2
3
4
5
6
7
8
9
1
1
1
1
1
1
1
1
1
1
2
2
2
2
2
2
2
2
2
2
3
3
3
3
3
3
3
3
3
3
4
4
4
4
4
4
4
4
4
4
5
5
5
5
5
5
5
5
5
5
6
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
21
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 22 of 61
1
2
3
4
5
6
7
8
9
1
1
1
1
1
1
1
1
1
1
2
2
2
2
2
2
2
2
2
2
3
3
3
3
3
3
3
3
3
3
4
4
4
4
4
4
4
4
4
4
5
5
5
5
5
5
5
5
5
5
6
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
Figure 8. Docking of compound 12g predicts a similar binding mode to the 13g crystal structure for ALDH1A1, 1A2, and
1A3. (A) Structure and ligand interaction plot of Compound 12g docked into ALDH1A1 crystal structure. (B,C) Structure
and ligand interaction plot of compound 12g docked into ALDH1A2, 1A3 homology models respectively.
When inhibitor 12g was docked, it adopted a similar pose to the crystal structure of the regioisomer 13g.
The similar potency and isoform selectivity for these compounds is consistent with the relatively solvent
22
ACS Paragon Plus Environment

Page 23 of 61
Journal of Medicinal Chemistry
exposed nature of the oxetane in both cases. The phenyl ring docks deep in the cavity and forms piꢀpi
interactions with Phe171 and Trp178 (1A1), Phe188(1A2), Phe182(1A3) and van der Waals interactions with
Trp195(1A2), Trp189(1A3) as shown in Figure 8. The 3ꢀfluorobenzyl ring makes van der Waals interactions
with Val460 and Asn121 (1A1), Leu477 and Val138 (1A2), and Leu471 and Ile132(1A3). The bicyclic ring of
1
2
3
4
5
6
7
8
9
1
1
1
1
1
1
1
1
1
1
2
2
2
2
2
2
2
2
2
2
3
3
3
3
3
3
3
3
3
3
4
4
4
4
4
4
4
4
4
4
5
5
5
5
5
5
5
5
5
5
6
1
2g engages the same residues as 11 for the three isoforms. The predicted binding mode of Compound 13h
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
(Supplementary Figure S7), which is also active against all three isoforms, is similar to that of 12g.
The modestly ALDH1A3 selective compound 15g, which has a methoxy group at the 4ꢀposition of the
phenyl ring, is predicted to bind ALDH1A3 in a similar fashion to 7 (Figure 9). The positioning of the methoxy
group near the ALDH1A3 specific residue Glu135, may explain the selectivity of 7 towards ALDH1A3. In
contrast, ALDH1A1 bears Ala124 at this position which could not form a favorable polar contact with the
methoxy.
Figure 9. ALDH1A3 specific residue Glu135 may explain the observed selectivity of compound 15g. (A) 15g docked
into the ALDH1A3 homology model. (B) Overlay of ALDH1A1 (pink) crystal structure with 15g docked in the 1A3
homology model (cyan).
23
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 24 of 61
Confirming activity in live cells: We next performed the ALDEFLUOR assay (Stem Cell Technologies) in
PEO1 and OVCAR5 cells, to confirm ALDH inhibition in live cells (Figure 10A, B). Given the association
between ALDH1A3 expression and the high grade serous histology we chose cell lines with high 1A3 activity
1
2
3
4
5
6
7
8
9
1
1
1
1
1
1
1
1
1
1
2
2
2
2
2
2
2
2
2
2
3
3
3
3
3
3
3
3
3
3
4
4
4
4
4
4
4
4
4
4
5
5
5
5
5
5
5
5
5
5
6
2
4
44
(
Supplementary Figure S8). PEOꢀ1 is classified as a high grade serous ovarian cancer cell line. OVCAR5, a
45
commonly used cell line originally classified as high grade serous ovarian, may be gastrointestinal in origin.
The ALDH1A subfamily inhibitors 12g, 13g, and 13h all exhibited strong inhibition of ALDEFLUOR signal
with significant activity in both cell lines at 1 µM and reductions of greater than 50% at 10 µM. While both
compounds inhibited ALDEFLUOR to some extent, the 1A3 selective compound 15l was significantly better
than 1A1 selective DEAB. The 1A2 selective inhibitor 15b showed little to no inhibition at 30 µM. Compound
15b is reasonably soluble and has a lower TPSA and molecular weight than the efficacious inhibitors,
suggesting that this effect is indeed due to its isoform selectivity profile and not simply an inability to enter the
cell. Despite having biochemical 1A3 IC s (0.12ꢀ0.17 µM) that are very similar to 15l (0.18 µM), the inhibitors
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
5
0
of all 3 ALDH1A isoforms, 12g, 13g, and 13h, were somewhat superior in the ALDEFLUOR assay
statistically significant differences at 3 ꢀM are highlighted in Figures 10A and 10B). Although we cannot rule
(
out variable cell penetration for the compounds as a potential cause, this may indicate that low levels of 1A1
and/or 1A2 expression in these cell lines slightly diminish the effective ALDEFLUOR inhibition of 1A3ꢀ
selective 15l. Low levels of 1A1 and 1A2 mRNA were in fact detected in OVCAR5 but not PEO1 cells
(Supplementary Figure S8).
In addition to ALDH, the surface glycoprotein CD133 has also been used as a marker of stemness in
many cancers including ovarian cancer, and its expression is correlated with poorer patient outcomes in ovarian
7, 46
cancer.
Interestingly, treating PEO1 and OVCAR5 cells with inhibitors 12g, 13g, and 13h also elicited a
+
dose dependent preferential reduction in the percentage of CD133 cells (Figure 10C, D). The 1A1 selective
DEAB was not effective in this assay, and the ALDH1A2 selective 15b had only a modest doseꢀindependent
+
effect. The 1A3 selective compound 15l depleted CD133 cells at 30 µM, but had no effect at 10 µM.
24
ACS Paragon Plus Environment

Page 25 of 61
Journal of Medicinal Chemistry
1
A
B
2
3
4
5
6
7
8
9
1
1
1
1
1
1
1
1
1
1
2
2
2
2
2
2
2
2
2
2
3
3
3
3
3
3
3
3
3
3
4
4
4
4
4
4
4
4
4
4
5
5
5
5
5
5
5
5
5
5
6
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
C
D
Figure 10. Compounds 15l, 12g, 13g, and 13h inhibit ALDH in cells and deplete cells bearing the putative CSC marker
CD133. (A, B) Graphs showing changes in ALDEFLUOR activity in PEO1 (A) and OVCAR5 (B) cells following the
specified treatment (N=3). (C, D) Graphs showing changes in the percent of CD133+ cells as determined by flow
cytometry following 48 h drug treatment at the specified concentration (N=3ꢀ6) *p≤0.05, **p≤0.01, ***p≤0.0001.
Evaluation of synergy with Cisplatin: ALDH1A family members are known to play a role in resistance to
4
7ꢀ50
16ꢀ22, 51ꢀ
multiple chemotherapy agents
and a reduction in their activity has been shown to restore sensitivity.
5
3
Therefore, we investigated the ability of our inhibitors with the best ALDEFLUOR activity to increase
sensitivity to and potentially synergize with cisplatin in PEO1 and OVCAR5 cells. Using the ChouꢀTalalay
method, we determined that 12g, 13g, and 13h are highly synergistic (Combination Index values <1 indicate
synergy with smaller numbers indicating stronger synergy) with cisplatin (Fig. 11A, B, Supplemental Fig. S9).
The observed synergy for 12g and 13g at 1 µM was already strong but it was improved by increasing the
5
4
concentration of these compounds to 10 µM. While 12g and 13g alone did not show cytotoxicity at these doses,
13h did and, therefore, had to be tested at lower doses. Nonetheless, synergy with 13h was observed at
concentrations as low as 0.3 µM. As a control, we included the ALDEFLUOR inactive 15b, which was
25
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 26 of 61
somewhat synergistic but was overall the least effective compound, especially at the lower concentrations of
cisplatin.
1
2
3
4
5
6
7
8
9
1
1
1
1
1
1
1
1
1
1
2
2
2
2
2
2
2
2
2
2
3
3
3
3
3
3
3
3
3
3
4
4
4
4
4
4
4
4
4
4
5
5
5
5
5
5
5
5
5
5
6
We similarly evaluated synergy using primary patient derived samples in a 3ꢀD tumor spheroid
5
5ꢀ56
assay; a platform which mimics in vivo growth (Fig 11C).
3g at 3ꢀfold lower concentrations in the initial synergy studies, we tested this compound. Ovarian cancer
spheroids, compared to 2D monolayer culture, are more resistant to standard therapies and express higher levels
As 13h showed comparable synergy to 12g and
1
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
2
0, 57
of ALDH1A1.
The sphereoids employed in this experiment express both 1A1 and 1A3 as determined by
qRTꢀPCR (Supplementary Figure S10). Treatment of patient derived ovarian cancer spheroids with 0.3 µM 13h
did not significantly reduce viability as monotherapy. However it was highly synergistic with all three
5
4
concentrations of cisplatin tested (Fig 11C, Combination indices 0.08ꢀ0.27). Similar to our observations,
Condello et al. report that concentrations of 4 below its ALDEFLUOR IC are able to potentiate the effects of
5
0
2
0
cisplatin against ovarian cancer spheroids.
26
ACS Paragon Plus Environment

Page 27 of 61
Journal of Medicinal Chemistry
1
2
C
3
4
5
6
7
8
9
1
1
1
1
1
1
1
1
1
1
2
2
2
2
2
2
2
2
2
2
3
3
3
3
3
3
3
3
3
3
4
4
4
4
4
4
4
4
4
4
5
5
5
5
5
5
5
5
5
5
6
A
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
B
Figure 11. Compounds 12g, 13g, and 13h synergistically enhance the effect of cisplatin on ovarian cancer cell lines and
patient derived spheroids. (A, B) Representative heatmaps of combination indices for PEO1 (A) and OVCAR5 (B) cells
cultured for 72 h in the presence of cisplatin and the indicated ALDH inhibitor at the concentrations shown. (N=3) (C)
Normalized average viability as determined using Alamarblue dye following exposure to the specified compounds for 72
hours. (N=5ꢀ8) *p≤0.05, ****p≤0.0001 one way ANOVA compared to Control. Combination indices for the specified
5
4
drug combinations calculated using CompuSyn software. (CI < 1 indicates synergy) (http://www.combosyn.com).
Table 6. In Vivo exposure following IP injection in mice.
obs
Compound
Plasma Concentration (µM)
AUC0ꢀ7h
(hrꢁµM)
0
.5 h
2 h
4 h
7 h
1
1
3g
3h
7.6 ± 0.8
5.6 ± 0.2
5.2 ± 1.0
2.6 ± 0.6
3.0 ± 1.3
0.6 ± 0.2
0.9 ± 0.7
0.1 ± 0.1
27.8
11
Compounds at the specified doses were administered by a single intraperitoneal injection of 10mg/kg to CDꢀ1 mice.
Values expressed as Mean ± S.D for 3 mice at each timepoint.
In Vivo Murine Metabolic Stability: Given that 13g and 13h showed promising cellular activity and were the
most stable in the presence of MLM among compounds tested, we assessed in vivo exposure following IP
injection in mice (Table 6). Despite the two compounds having very similar MLM stability, the AUC for 13g
27
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 28 of 61
was more than double that of 13h. Plasma concentration of 13g exceeded 1 µM, the concentration needed for
0% inhibition of ALDEFLUOR (Figure 10A,B), for nearly 7 hours. 13h exceeded 0.3 µM, an efficacious dose
1
2
3
4
5
6
7
8
9
1
1
1
1
1
1
1
1
1
1
2
2
2
2
2
2
2
2
2
2
3
3
3
3
3
3
3
3
3
3
4
4
4
4
4
4
4
4
4
4
5
5
5
5
5
5
5
5
5
5
6
5
in the spheroid cisplatin synergy assay (Figure 11C), for almost 4 hours.
CONCLUSIONS
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
In order to explore the structural determinants of ALDH1A isoform selectivity, we generated a number of
analogs of lead compound 7 guided by a coꢀcrystal structure with 1A1. Expedient synthetic routes were
developed to allow lateꢀstage diversification at multiple sites. The observation that the orthoꢀmethyl group on
the Nꢀphenyl was disfavorable to binding across the ALDH1A subfamily led to the first potent inhibitors of the
three ALDH1A isoforms with excellent selectivity over ALDH2 (e.g. 12e). Small structural changes led to a
new binding mode for compound 13g, explaining SAR trends which did not agree with the 1A1 crystal structure
for lead compound 7. We demonstrated that remarkable changes in ALDH1A isoformꢀselectivity can be
achieved by minor structural changes to the Nꢀphenyl substituent, which binds in the narrow lipophilic region of
the active site near the catalytic cysteine according to the 13g-1A1 structure. In the course of exploring SAR at
this position we discovered the 1A2 selective compounds 15b and 15f, and the 1A3 selective compounds 15g
and 15l. Although selectivity within the ALDH1A subfamily is modest, to our knowledge, these are the first
examples of ALDH1A2 and ALDH1A3 selective inhibitors with excellent selectivity against ALDH2. Further
efforts to create more potent and selective 1A2 and 1A3 inhibitors are underway.
Our preliminary data lead us to hypothesize that ALDEFLUOR activity will be improved by inhibiting
multiple ALDH1A isoforms simultaneously as compared to an equipotent 1A3 selective compound (12g, 13g,
1
3h vs. 15l), even in cell lines which only express detectable levels of 1A3. This hypothesis will require more
extensive cell testing to confirm our preliminary findings. Testing optimized compounds in an expanded panel
of cell lines which express various ALDH1A isoforms will be necessary and is underway.
+
Interestingly 12g, 13g, and 13h depleted the CD133 putative stem cell pool in a dose dependent
manner, and synergized with cisplatin at a range of concentrations. 13h also showed synergy with cisplatin in
patient derived ovarian cancer spheroids. Compound 13g furthermore shows promising PK properties for in
28
ACS Paragon Plus Environment

Page 29 of 61
Journal of Medicinal Chemistry
vivo studies, exceeding its efficacious cellular concentration in plasma for nearly 7 hours following 10mg/kg IP
administration in a preliminary study. Xenograft studies on optimal compounds will be reported in due course.
Future medicinal chemistry development of this series will focus on further improving enzymatic and cellular
potency and pharmacokinetic properties, as well as tuning the selectivity profile among the ALDH1A isoforms
in order to generate diverse probes for establishing the contribution of individual isoforms to cellular activity.
1
2
3
4
5
6
7
8
9
1
1
1
1
1
1
1
1
1
1
2
2
2
2
2
2
2
2
2
2
3
3
3
3
3
3
3
3
3
3
4
4
4
4
4
4
4
4
4
4
5
5
5
5
5
5
5
5
5
5
6
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
EXPERIMENTAL SECTION
Crystal Structure Determination. Crystals of ALDH1A1 N121S were grown by equilibrating 4ꢀ8 mg/mL
ALDH1A1 N121S against 100 mM sodium BisTris, pH 6.2ꢀ6.5, 6ꢀ11% PEG3350, 200 mM NaCl, and 5ꢀ10 mM
YbCl . The crystal of ALDH1A1 N121S in complex with 7 was prepared by soaking approximately 2ꢀweek old
3
apo crystals for 6 hr in crystallization solution containing 500 µM 7 with 1% (v/v) DMSO and 1 mM NAD. The
crystal of ALDH1A1 N121S in complex with 13g was prepared by soaking apo crystals for 1 hr in
crystallization solution containing 1 mM NAD followed by soaking overnight in crystallization solution
containing 500 µM 13g with 2% (v/v) DMSO and 1 mM NAD. Cryoprotection for flashꢀfreezing was 20%
(v/v) ethylene glycol in ligand soaking solution. Diffraction data for 7 in complex with ALDH1A1 N121S was
collected at Beamline 23ꢀIDꢀD (GM/CA), National Institute of General Medicine Sciences and National Cancer
Institute of the NIH at APS. Diffraction data for 13g in complex with ALDH1A1 N121S was collected at
Beamline 19ꢀID operated by the Structural Biology Consortium at the Advance Photon Source (APS), Argonne
5
8
National Laboratory. Diffraction data were indexed, integrated, and scaled using HKL3000. The CCP4
59
program suite was used for molecular replacement and refinement. The Coot molecular graphic application
6
0
was used for model building. The TLSMD (translation/libration/screw motion determination) server was used
61ꢀ62
to determine dynamic properties of ALDH1A1 N121S.
Protein Purification and Enzymatic Assays
29
ACS Paragon Plus Environment