Discovery and Optimization of 1-Phenoxy-2-aminoindanes as Potent, Selective, and Orally Bioavailable Inhibitors of the Na+/H+ Exchanger Type 3 (NHE3)

Article abstract of DOI:10.1021/acs.jmedchem.6b00624

The design, synthesis, and structure-activity relationship of 1-phenoxy-2-aminoindanes as inhibitors of the Na⁺/H⁺ exchanger type 3 (NHE3) are described based on a hit from high-throughput screening (HTS). The chemical optimization resulted in the discovery of potent, selective, and orally bioavailable NHE3 inhibitors with 13d as best compound, showing high in vitro permeability and lacking CYP2D6 inhibition as main optimization parameters. Aligning 1-phenoxy-2-aminoindanes onto the X-ray structure of 13d then provided 3D-QSAR models for NHE3 inhibition capturing guidelines for optimization. These models showed good correlation coefficients and allowed for activity estimation. In silico ADMET models for Caco-2 permeability and CYP2D6 inhibition were also successfully applied for this series. Moreover, docking into the CYP2D6 X-ray structure provided a reliable alignment for 3D-QSAR models. Finally 13d, renamed as SAR197, was characterized in vitro and by in vivo pharmacokinetic (PK) and pharmacological studies to unveil its potential for reduction of obstructive sleep apneas.

Full text of DOI:10.1021/acs.jmedchem.6b00624

Subscriber access provided by Northern Illinois University
Article
Discovery and Optimization of 1-Phenoxy-2-aminoindanes as Potent, Selective
+
+
and Orally Bioavailable Inhibitors of the Na/H Exchanger Type 3 (NHE3)
Nils Rackelmann, Hans Matter, Heinrich Englert, Markus Follmann, Thomas Maier, John
Weston, Petra Arndt, Winfried Heyse, Katharina Mertsch, Klaus Wirth, and Laurent Bialy
J. Med. Chem., Just Accepted Manuscript • DOI: 10.1021/acs.jmedchem.6b00624 • Publication Date (Web): 08 Sep 2016
Downloaded from http://pubs.acs.org on September 13, 2016
Just Accepted
“Just Accepted” manuscripts have been peer-reviewed and accepted for publication. They are posted
online prior to technical editing, formatting for publication and author proofing. The American Chemical
Society provides “Just Accepted” as a free service to the research community to expedite the
dissemination of scientific material as soon as possible after acceptance. “Just Accepted” manuscripts
appear in full in PDF format accompanied by an HTML abstract. “Just Accepted” manuscripts have been
fully peer reviewed, but should not be considered the official version of record. They are accessible to all
readers and citable by the Digital Object Identifier (DOI®). “Just Accepted” is an optional service offered
to authors. Therefore, the “Just Accepted” Web site may not include all articles that will be published
in the journal. After a manuscript is technically edited and formatted, it will be removed from the “Just
Accepted” Web site and published as an ASAP article. Note that technical editing may introduce minor
changes to the manuscript text and/or graphics which could affect content, and all legal disclaimers
and ethical guidelines that apply to the journal pertain. ACS cannot be held responsible for errors
or consequences arising from the use of information contained in these “Just Accepted” manuscripts.
Journal of Medicinal Chemistry 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 53
Journal of Medicinal Chemistry
1
2
3
4
5
6
7
8
9
Discovery and Optimization of 1-Phenoxy-2-aminoindanes as
+
+
Potent, Selective and Orally Bioavailable Inhibitors of the Na /H
Exchanger Type 3 (NHE3)
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
*
, †
*, †
†
†, #
†
Nils Rackelmann , Hans Matter , Heinrich Englert , Markus Follmann , Thomas Maier ,
†
†
†
†
†
John Weston , Petra Arndt , Winfried Heyse , Katharina Mertsch , Klaus Wirth , Laurent
†
Bialy
†
SanofiꢀAventis Deutschland GmbH, R&D, Dꢀ65926, Frankfurt am Main
#
Author present address: Bayer Pharma AG, Aprather Weg 18A, Dꢀ42113 Wuppertal,
Germany.
Abstract
The design, synthesis and structureꢀactivity relationship of 1ꢀphenoxyꢀ2ꢀaminoindanes as
+
+
inhibitors of the Na /H exchanger type 3 (NHE3) is described based on a hit from highꢀ
throughput screening (HTS). The chemical optimization resulted in the discovery of potent,
selective and orally bioavailable NHE3 inhibitors with 13d as best compound, showing high
in vitro permeability and lacking CYP2D6 inhibition as main optimization parameters.
Aligning 1ꢀphenoxyꢀ2ꢀaminoindanes onto the Xꢀray structure of 13d then provided 3DꢀQSAR
models for NHE3 inhibition capturing guidelines for optimization. These models showed
good correlation coefficients and allowed for activity estimation. In silico ADMET models for
Cacoꢀ2 permeability and CYP2D6 inhibition were also successfully applied for this series.
Moreover, docking into the CYP2D6 Xꢀray structure provided a reliable alignment for 3Dꢀ
QSAR models. Finally 13d, renamed as SAR197, was characterized in vitro and by in vivo
pharmacokinetic (PK) and pharmacological studies to unveil its potential for reduction of
obstructive sleep apneas.
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 2 of 53
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
Introduction.
+
+
Sodium–hydrogen (Na /H ) exchanger type 3, also known as NHE3, sodium–hydrogen
antiporter 3 or solute carrier family 9 member 3 (SLC9A3), is one of nine isoforms of the
1
mammalian NHE gene family. These mammalian NHEs consist of isoforms that occur
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
primarily in the plasma membrane and those that appear to primarily reside in intracellular
2
3
organelles. The human NHE3 transporter is encoded by the SLC9A3 gene.
NHE3 is primarily located in the nephron of the kidney, in the apical membrane of intestinal
enterocytes and in the brain stem cell area. A significant portion of proximal tubule sodium
+
+
4,5
ion reabsorption is mediated by NHE3 and other membrane Na /H exchangers.
Immunochemical studies suggest that NHE3 is the principal exchanger isoform expressed on
proximal tubular brushꢀborder membranes, which might indicate a substantial role of this
transporter for maintaining the sodium ion balance by sodium ion and fluid reabsorption in the
6
proximal tubule with a consequential and significant impact on intracellular pHꢀvalues.
Hypercapnia, a condition of abnormally elevated CO
2
levels in the blood, as well as a lowered
+
intracellular pHꢀvalue are both known to increase the bioelectric activity of CO
2
/H ꢀsensitive
7
+
neurons. Inhibition of NHE3 increases the activity of CO /H ꢀsensitive neurons upon
2
8
8bꢀc,9
intracellular acidification. Inhibition of NHE3 expressed in brain stem areas,
is able to
modulate the ventilation drive and therefore reduce central sleep apneas. In addition NHE3
1
0
inhibition leads to increased muscle tone of the upper airways. Therefore, NHE3 inhibitors
can potentially be used to treat muscle related respiratory impairments such as obstructive
sleep apneas and snoring.
These interesting properties prompted us to initiate a discovery program for novel inhibitors
with a suitable profile for advanced pharmacological evaluations. The focus of our project
was to identify orally bioavailable and potent small molecules without acylguanidine, thus
11,12
complementing earlier NHE3 inhibitors like 1a (S3226).
Guanidines and acylguanidines
are important motifs of numerous active molecules. Depending on the protein environment,
ACS Paragon Plus Environment

Page 3 of 53
Journal of Medicinal Chemistry
1
2
3
4
5
6
7
8
9
acylguanidines are less basic guanidine bioisosteres with potentially improved permeability
and oral bioavailability. Despite lower basicity, they contain many polar atoms with high
polar surface area (PSA), which still might cause lower permeability and therefore insufficient
pharmacokinetic behavior in cases, where acylguanidines are not additionally decorated by
lipophilic groups or cyclized. For interaction with NHE subtypes, unsubstituted Nꢀ
acylguanidines are often required. In contrast, no general adverse effects are reported for this
head group to our knowledge; several molecules were investigated in clinical studies. The
goal of this project was to identify structurally unrelated compounds without polar
acylguanidine to modulate molecular properties, which might also lead to different selectivity
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
profiles among NHE isoforms. Figure 1 summarizes other chemical structures reported as
d,13
NHE3 blockers like the aminobenzimidazole 1b (AVE0657)9
and a series of
14,15
tetrahydroisoquinolines, exemplified by 1c (THiQ).
In addition to earlier acylguanidineꢀ
containing NHE3 inhibitors, blockers of NHE1 containing the same structural motif are
1
2,16
17
described in addition, like 1d (HOE694)
and 1e (HOE642A / Cariporide).
Figure 1. Known inhibitors for NHE3 and / or NHE1.
Here we report the discovery, optimization and structureꢀactivity relationships (SAR) of a
series of 1ꢀphenoxyꢀ2ꢀaminoindanes, which led to potent and selective NHE3 inhibitors. The
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 4 of 53
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
manuscript is organized as follows. First we describe the optimization of a hit series from
highꢀthroughput screening (HTS) towards the identification of 13d as best compound. Next
we describe 3DꢀQSAR (quantitative structureꢀactivity relationship) models for NHE3 and
1
8
CYP2D6 inhibition using comparative molecular field analysis (CoMFA) and comparative
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
9
molecular similarity index analysis (CoMSIA). We then report studies to identify
prerequisites for high in vitro Cacoꢀ2 cell permeability and low CYP2D6 inhibition by in
silico ADMET (absorption, distribution, metabolism, excretion, toxicity) models. Then we
summarize the in vitro characterization of 13d. Finally in vivo pharmacokinetic (PK) and
pharmacological studies are discussed to highlight the potential of 13d for reduction of
obstructive sleep apneas.
Methods.
Biological assays.
The primary assay used to determine NHE3 inhibition for both the highꢀthroughput screening
1
0
campaign and during lead optimization is described in detail in ref. . Briefly, the recovery of
the intracellular pH of LAP1 cells which stably express NHE3 after acidification is
determined. The recover occurs even under bicarbonateꢀfree conditions with functional
NHE3. To this end, the intracellular pH was determined using the pHꢀsensitive fluorescent
dye BCECF (2',7'ꢀBisꢀ(2ꢀcarboxyethyl)ꢀ5ꢀ(andꢀ6)ꢀcarboxyfluorescein, employing the
precursor BCECFꢀAM, from Molecular Probes, Eugene, OR, USA). The cells were initially
4
incubated with BCECFꢀAM in NH Cl buffer. The intracellular acidification was induced by
washing the cells incubated with a NH Clꢀfree buffer. After washing, 90 µL of the buffer was
4
+
left on the cells. The pH recovery was initiated by ading 90 µL of Na ꢀcontaining buffer in a
FLIPR instrument (Fluorometric Imaging Plate Reader, Molecular Devices, Sunnyvale, CA,
USA). The BCECF fluorescence was determined with an excitation wavelength of 498 nm
and an FLIPR emission filter (bandpass from 510 to 570 nm). Changes in fluorescence as a
ACS Paragon Plus Environment

Page 5 of 53
Journal of Medicinal Chemistry
1
2
3
4
5
6
7
8
9
measure for pH recovery were recorded for two minutes. To evaluate the NHE3ꢀinhibition for
each compound, the cells were initially investigated in buffers with which complete or no pH
+
recovery took place. For complete recovery, the cells were incubated in Na ꢀcontaining
+
buffer, for no recovery, the cells were incubated in Na ꢀfree buffer. The recovery of the
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
intracellular pH at each tested concentration of a substance was expressed as percent of the
maximum recovery and used for IC determination.
5
0
Each IC50 value was determined from 10 individual concentrations as duplicates. IC50 values
for 9 more interesting, potent analogs out of 67 compounds described in this study were tested
more than three times. For example, the best compound 13d was tested 10 times with a
standard deviation of 0.011 µM for the IC50 value. For 8 compounds, three tests for IC50
values were performed, for 23 compounds two IC50 tests. For the remaining set of 27
compounds with lower activity mainly, only a single IC50 value was determined. The
geometric mean was used for averaging. The standard deviation for IC50 determination ranges
from 0.00008 for very potent to 3.45 for very weak NHE3 inhibitors. There is a good
correlation between the NHE3 inhibitory potency and standard deviation, as expected.
For statistical model building, NHE3 inhibition (pIC ) is expressed as log(1/IC ).
5
0
50
Permeability assays to assess membrane permeability potential were performed with Cacoꢀ
2
0
2
/TC7 cells. Conditions for apical compartment were pH 6.5; 0.5% bovine serum albumin
(BSA); compound concentration 20 µM; basolateral compartment pH 7.4; 5% BSA; 2 hours
incubation time. Efflux was determined with pH 7.4 in both compartments and 0.5% BSA
with compound concentration of 2 µM. All analyses were performed with LCꢀMS/MS.
2
1
Cytochrome P450 (CYP) inhibition assays were performed with pooled and phenotyped
human hepatic microsomes (0.1 mg/mL for CYP3A4 and CYP2C9; 0.2 mg/mL for CYP2D6).
2
As cofactors, 6 mM MgCl , 0.5 mM EDTA, 1 mg/mL BSA, 1 mM NADPH were used;
compound concentration tested was in the range of 0.001 – 200 µM (in duplicate). Results
were expressed as % inhibition or IC50 values. As substrates, diclofenac was used for
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 6 of 53
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
CYP2C9; dextromethorphan for CYP2D6 and both midazolam and testosterone were
2
2
employed for CYP3A4. In the CYP3A4 inhibition experiments, the control inhibitor
ketoconazole was tested up to 0.25 µM and an IC50 value of 0.0285µM was obtained (5
concentrations). Maximal concentration of 0.25 µM yielded 17% of remaining enzyme
activity compared to control. In CYP2D6 experiments, the control inhibitor quinidine was
tested up to 0.225 µM and resulted in an IC50 value of 0.0413 µM. The remaining activity of
the enzyme was 21% at maximum 0.225 µM quinidine concentration compared to control.
Both IC50 values are in agreement to literature.
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
Lability assays were performed with pooled and phenotyped human liver microsomes taking
21
samples at time 0 and time 20 minutes for LCꢀMS/MS analysis. Clearance in vitro was
ꢀ6
21
determined using primary human hepatocytes at 1.4 * 10 cells/mL. 13d concentration
varied from 0.5 µM to 5 µM. CYP contribution was determined by addition of either
quinidine for CYP2D6 inhibition or ketoconazole for CYP3A4 inhibition to the assay plate.
All bioanalytical analyses were performed by LCꢀMS/MS.
Induction potential for CYP enzymes was tested with primary human hepatocytes and several
batches of cryopreserved human hepatocytes. Incubation medium contained 0.1 µM
dexamethasone; assays were performed on 48ꢀ or 96ꢀwell plates. 13d was tested in
concentrations between 1 ꢀ 30 µM and compared to references βꢀnaphtoflavone F (10 µM)
and omeprazole (30 µM) for CYP1A1/2, and rifampicin (10 µM) for CYP3A4 after 48 hours
22aꢀc
incubation. Analysis of CYP1A1/2, CYP1A2 and CYP3A4 was performed by RTꢀPCR.
Pharmacokinetic parameters in pigs (male Duroc Landrace pigs) were determined after single
intravenous and oral administrations of 13d. Dose, formulation and route were used as
follows: i.v. route with 1 mg/kg in 5% Glucose/water in solution; p.o. route with 3 mg/kg in
agarose with sugar and crushed pellets as gelos formulation; Sampling times: 0.083, 0.25, 0.5,
1, 2, 4, 8 and 24 hours intravenous route; 0.25, 0.5, 1, 2, 4, 6, 8 and 24 hours oral route.
ACS Paragon Plus Environment

Page 7 of 53
Journal of Medicinal Chemistry
1
2
3
4
5
6
7
8
9
Analysis was performed with ESIꢀLC/MSꢀMS assay of 13d in plasma with a limit of
2
3
quantitation (LOQ) of 5 ng/mL.
All studies in animals were conducted in accordance with German laws for protection of
animals. Furthermore, the investigation conforms to the guidance for the Care and Use of
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
2
4
laboratory Animals published by the US National Institutes of health. 24
X-Ray structure determination.
The molecular and crystal structure of 13d was determined by single crystal Xꢀray structure
3
analysis. A single crystal (size: 0.50*0.05*0.02 mm ) obtained from dichloromethane / diethyl
ether was sealed in a Lindemannꢀglass capillary. Xꢀray diffraction data were collected on a
Bruker/AXS three circle diffractometer, equipped with a SMART APEX areaꢀdetector, a low
temperature device and a molybdenumꢀKα rotating anode generator, operated at 50 kV/120
mA and adjusted to a fineꢀfocus of 0.5 * 5 mm². Data frames were collected using the
2
5
program SMART V 5.628, applying ω−scans with step widths of 0.3° and an exposure time
2
6
of 60 seconds. Data processing with the program SAINT + Release 6.45 yielded 10173
reflections (ϑmin = 1.98, ϑmax = 21.97; ꢀ18 < h < 20, ꢀ7< k < 7, ꢀ21< l < 22) of which 6343
reflections were unique (Rint = 0.0819, Rσ = 0.2271). Refinement of cell parameters was
performed using 1101 reflections. The phase problem was solved with direct methods by the
2
7
XS module of SHELXTL 6.14.
2
2 2
The structure was refined by leastꢀsquares methods (minimization of (F
o
c
ꢀ F ) ) using the XL
module of SHELXTL 6.14 (Bruker AXS, 2000). The positions of all H atoms were calculated,
except those bonded to the water molecules O1 and O4 to O7 which were experimentally
determined from a difference Fourier synthesis map. Hydrogen atoms for oxygen atoms O2
and O3 could not be determined. Sgoodness of fit = 0.850, Rall data = 0.1912 (Robs. data = 0.0607 for
2
464 reflections with |Fobs| > 4σ, wR2all data = 0.1506, wR2obs. data = 0.1161). The largest
3
unassigned peaks in the difference map correspond to ꢀ0.248 versus +0.344 electrons per Å .
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 8 of 53
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
The average estimated standard deviation (e.s.d.) of a CꢀC bond is 0.014 Å, that of an OꢀC
bond 0.011 Å, that of an NꢀC bond 0.011 Å. The average e.s.d. of CꢀCꢀC bond angles is 1.1
and that of CꢀCꢀCꢀC torsion angles 1.4°. The Flack parameter has been determined to 0.0566
(e.s.d. 0.1296), reference values are 0 (within 3 e.s.d.'s) for the correct and +1 for inverted
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
absolute structure.
Computational procedures.
The Xꢀray structure of 13d was used as template for aligning 1ꢀphenoxyꢀ2ꢀaminoindanes. The
2
8
ligandꢀbased alignment was performed in a pairwise manner using tfit in QXP with a
2
9
30
modified AMBER force field. The CYP2D6 Xꢀray structure 2f9q with a resolution of 3.0
3
1
Å from RCSB was used for docking. Molecules were minimized or automatically docked
28
using QXP. Selection of poses was guided by consistent alignment of the indane. Selected
side chains lining the substrate pocket were treated as flexible. The alignment for 27
molecules from tables 3ꢀ5 with CYP2D6 inhibition served as basis for 3DꢀQSAR.
3
2
33
The program MOLCAD was used to visualize lipophilicity, hydrogenꢀbonding potential
34
and cavity depth on solvent accessible protein surfaces.
35
Default settings were used for CoMFA and CoMSIA in Sybyl. For CoMFA steric and
electrostatic energies are calculated at grid points with 2 Å spacing, a positively charged
3
6
carbon atom and a distanceꢀdependent dielectric constant with MMFF94 charges. The same
alignments were used for CoMSIA steric, electrostatic and hydrophobic similarity index
19
fields using a probe with charge +1, a radius of +1, a hydrophobicity of +1 and 0.3 as
attenuation factor
α for Gaussianꢀtype distance dependence.
3
7
38
Crossvalidated analyses were run using SAMPLS or two and five crossvalidation groups
3
9
with random selection of members. PLS (partial least squares) analyses using two or five
random crossvalidation groups were averaged over 100 runs. For validation all affinities were
4
0
2
randomized 100 times, subjected to PLS and the mean crossvalidated r was calculated. 2D
ACS Paragon Plus Environment

Page 9 of 53
Journal of Medicinal Chemistry
1
2
3
4
5
6
7
8
9
4
1
fingerprints were generated using UNITY for selection of training and test sets; their
4
2
similarity is given as Tanimoto coefficient.
43
In silico ADMET models were generated using data from multiple internal series. Model
building for endpoints like human microsomal lability and PXR activation is described
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
4
4
elsewhere. Preꢀtreatment included removal of counterꢀions and smaller fragments,
neutralization plus canonization of structures. Canonical 3D geometries including hydrogen
4
5
positions were generated using Corina. The 2DꢀQSAR models were based on MOE
46
47
descriptors in combination with Cubist, which analyses continuous variables by a
regression tree. The principle was to construct a ruleꢀbased decision tree, where each rule was
characterized by a multiꢀlinear regression (MLR) model describing the SAR in this node. A
4
4
GAꢀbased approach for variable selection maximized the sum of regression coefficients for
training and test sets. After validation, the best models were consensus models from five
Cubist regression trees. The best model is selected from prediction and classification of
internal and external test sets. Preliminary versions with smaller training sets were used and
regularly updated during optimization. Therefore, the final correlation with the current model
version herein included this series in training, but serves to illustrate the workflow.
Results and Discussion.
Discovery of NHE3 inhibitors.
The 1ꢀphenoxyꢀ2ꢀaminoindane series was identified following the analysis of a HTS
campaign using the Sanofi screening library. The 1ꢀphenoxyꢀ2ꢀaminoindane hit series
originally resulted from a parallel synthesis library targeted towards generic ion channel
pharmacophore motifs. The series was identified by using a FLIPR assay. Three
representative molecules 2a-c as hits from this campaign are shown in Table 1. These
compounds exhibit NHE3 inhibitory activity ranging from 0.330 to 0.650 µM, respectively.
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 10 of 53
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
Hit structures 2a-c from the HTS, were obtained as mixtures of diastereomeres with defined
transꢀconfiguration at the indane C1 and C2 atoms. Further exploration then led to the
observation that compounds with S,Sꢀchirality at both result in more potent NHE3 inhibitors.
Due to the presence of basic aminoꢀgroups plus additional polar functionality, the in vitro
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
ꢀ7
membrane permeability, expressed as Cacoꢀ2 permeability (PTotal flux in 10 cm/sec) was
low, therefore suggesting a low in vivo intestinal absorption.
Table 1. Representative NHE3 inhibitors identified from HTS.
Caco2
PTotal (10- Inhibition
7cm/sec)
CYP2D6
NHE3_IC50_
uM
Structure
ID
IC50 uM
O
NH
O
2a
0.440
0.330
0.650
0,30
5,100
N
NH
2
A003499300A
O
NH
NH
2
O
2b
0,40
8,600
N
NH
2
A003499301A
H
2
N
O
Cl
S
O
Cl
O
2c
0,80
<0.300
N
NH
2
A003498896A
Chemistry.
To explore the SAR around the HTS hits 2a-c, a general, a robust and enantioselective
synthesis was established, as described below for representative members of this series. For
example, the finally selected compound 13d (c.f. table 3) was synthesized in large scale
starting from the propionic acid derivative 3a, which was converted in excellent yields into
the corresponding indanone 4b following a intramolecular Friedel-Crafts acylation protocol,
as outlined in Scheme 1.
a
Scheme 1. Synthesis of indenes.
ACS Paragon Plus Environment

Page 11 of 53
Journal of Medicinal Chemistry
1
2
3
4
5
6
7
8
9
O
OH
4
1
1
3
CO H
2
a, b
c
d
2
2
R
1
R
1
R
1
R
1
1
4
4
3
4
5
6
3
3
a R1=2-Cl, 4-Cl
b R1=3-Cl, 4-Cl
4a R1=6-Cl; commercial
b R1=4-Cl, 6-Cl; commercial
4c R1=5-Cl, 7-Cl; commercial
4d R1=5-Cl, 6-Cl; major from 3b
5a R1=6-Cl
5b R1=4-Cl, 6-Cl
6a R1=5-Cl
6b R1=5-Cl, 7-Cl
6c R1=4-Cl, 6-Cl
6d R1=5-Cl, 6-Cl
6e R1=4-Cl, 5-Cl
6f R1=5-F
4
5
5
5
c R1=5-Cl, 7-Cl
d R1=5-Cl, 6-Cl
e R1=6-Cl, 7-Cl
5f R1=6-F
g R1=6-F,4-Cl
5h R1=6-Cl, 4-F
4
4
4
e R1=6-Cl, 7-Cl; minor from 3b
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
5
6g R1=5-F,7-Cl
6h R1=5-Cl,7-F
f R1=6-F; commercial
g R1=6-F, 4-Cl; b
b
4h R1=6-Cl, 4-F;
a
b
(
a) (COCl) , CH Cl ; (b) AlCl , CH Cl ; (c) NaBH , MeOH (d) pꢀTsOH, toluene
2 2 2 3 2 2 4
48
ref.
Alternatively, indanones 4 with other substitutions of the central core were commercially
4
8
available or were synthesized as previous described. After reduction of the indanones 4 with
NaBH
4
and subsequent elimination of water in presence of catalytic amounts of pꢀTsOH, the
49
corresponding 1
H
indenes 6 were isolated in good yields. The detailed synthesis of 5,7ꢀ
dichloroꢀ1ꢀHꢀindene will serve as representative example for this synthesis sequence (c.f.
experimental section).
a
Scheme 2. Synthesis of indane oxides.
OH
1
3
O
a
b
c
O
R1
R1
R1
Br
R1
2
1
1
S, 2R-7
6
8
rac-7
7g R =6-F
7h R =6-Cl, 7-Cl
7
a R =H
7b R =6-Cl
c R =4-Cl, 6-Cl
d R =4-F, 6-Cl
8a R =6-F
1
1
1
8b R =6-Cl, 7-Cl
8c R =5-Cl, 6-Cl
1
1
1
d
7
7
7i R =5-Cl, 6-Cl
1
1
1
1
O
R1
7
7
1
e R =4-Cl, 6-F
f R =5-Cl, 7-Cl
1
rac-7
a
(
a) Jacobsen Salen complex; (b) NBS, DMSO/H O; (c) NaOH, THF; (d) mꢀCPBA,
2
2 2
CH Cl
a
Scheme 3. Synthesis of NHE3ꢀinhibitors 10-15.
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 12 of 53
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
OAr
NR R
2 3
OAr
1
O
a
b
c
R1
R1
OH
R
1
NR
2
R
3
R1
2 3
NR R
2
7
9
10 R
2
R
R
3
containing no pg
containing BudiPhSi-O
2 R R3 containing NBoc
2 3
13 R R containing OH
t
1
1
1 R
2
3
2
d
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
OAr
OAr
e
R1
NR R
R1
NR R
2 3
2
3
1
5 R R containing NR
14 R R containing NH
2 3
2
3
4
a
(a) HNR R , CH CN; (b) HOAr, DIAD, PPh , THF; (c) TBAF, THF; (d) 4M HCl,
2
3
3
3
dioxane; (e) R
4
2 3 3
Br, K CO , CH CN.
Using Jacobsen’s Salen complex, indane oxides 7a-d were obtained in good yields with high
5
0,51
enantioselectivity.
The observed enantioselectivity was further increased by
recrystallization of building blocks 7a-d from nꢀheptane. Racemic indaneꢀoxides 7g-f were
either obtained by direct oxidation of the indenes 6 with mꢀCPBA or using a stepwise protocol
outlined in Scheme 2.
Indane oxides 7 were regioselectively opened at position C1, with secondary amines in
5
2
acetonitrile as solvent. Utilizing a rearrangement of the aminoꢀalcohol 9 in presence of a
phenol under Mitsunobuꢀconditions, the trans configured 1ꢀphenoxyꢀ2ꢀamino indanes 10-12
53
were obtained in good yields and excellent selectivity, respectively. The syntheses of
corresponding nonꢀcommercial phenols and secondary amines were performed as earlier
1
0
described. In some cases, the final compounds 10-12 contained impurities after column
chromatography (reagents from the Mitsunobuꢀreaction; c.f. Supporting information). In those
cases, the corresponding NHE3 inhibitors 10-12 were subjected to an additional preparative
HPLC purification step.
Starting from 11, NHE3 inhibitors containing an alcohol moiety 13 were isolated after using a
standard TBAF deprotection protocol. Secondary amines 14, were obtained after treatment of
Boc protected compounds 12 with HCl in dioxane. To further explore the SAR of secondary
ACS Paragon Plus Environment

Page 13 of 53
Journal of Medicinal Chemistry
1
2
3
4
5
6
7
8
9
and tertiary amines, respectively, compounds 14 were reacted with alkylating reagents R
4
X in
acetonitrile in the presence of potassium carbonate.
Table 2. NHE3 inhibitors 10 from the optimization program.
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
Caco2
PTotal (10- Inhibition
cm/sec)
CYP2D6
Caco2
CYP2D6
NHE3_IC50_
uM
NHE3_IC50_
PTotal (10- Inhibition
uM
JMC ID
OAr
NR2R3
R1
JMC ID
OAr
NR2R3
R1
7
IC50 uM
7cm/sec)
IC50 uM
O
O
X
N
S
O
X
N
10a
X
H
0.438
0.123
91
0.640
10r
O
X
N
H
1,530
2,070
NHAc
Chemistry 1
Chemistry 2
Chemistry 20
Chemistry 2
O
S
O
O
NH
2
10b
X
N
H
132
0.400
10s
O
X
X
N
H
X
O
NEt2
Cl
Cl
Chemistry 3
Chemistry 2
Chemistry 21
Chemistry 2
O
X
NHAc
X
N
O
10c
H
H
H
2,770
0.500
0.281
10t
10u
10v
O
N
X
N
H
H
H
3,520
1,710
0.078
X
Chemistry 4
Chemistry 2
Chemistry 22
Chemistry 2
O
S
N
S
10d
O
X
NH
2
X
N
O
X
N
O
X
Chemistry 5
Chemistry 2
Chemistry 23
Chemistry 2
NH
Me
N
N
N
O
X
N
98
3,800
O
N
X
N
67
0.730
10e
O
X
X
Cl
Me
Chemistry 6
Chemistry 2
Chemistry 18
Chemistry 2
O
Me
O
S
NMe
2
N
N
10f
X
N
H
H
H
0.475
0.572
2,960
10w
10x
10y
O
X
N
X
N
H
H
H
0.118
0.170
0.272
X
O
Cl
Cl
F
3
C
Me
Chemistry 7
Chemistry 2
Chemistry 24
Chemistry 2
O
S
Me
O
X
Me
X
N
N
N
10g
215
<0.300
O
X
N
X
N
O
1-(4-methanesulfonyl-
Me
Me Me
phenoxy)
Chemistry 2
Chemistry 25
Chemistry 2
O
Me
N
O
X
N
N
N
O
X
N
X
N
10h
O
X
Chemistry 9
Chemistry 2
Chemistry 26
Chemistry 2
O
Me
Me
O
S
Me
N
O
10i
X
N
H
H
H
0.540
4,530
4,500
192
10z
O
X
X
N
H
0.411
0.737
0.094
X
O
Cl
Cl
Chemistry 10
Chemistry 2
Chemistry 27
Chemistry 2
O
O
X
N
10j
X
N
10aa
10ab
O
X
S
Me
X
N
4-F, 6-Cl
283
162
<0.300
2,500
NHAc
1-(4-methanesulfonyl-
O
Chemistry 11
Chemistry 2
phenoxy)
Chemistry 2
Me
O
X
CN
X
N
N
10k
O
N
X
N
6-Cl
N
X
Cl
Me
Chemistry 12
Chemistry 2
Chemistry 18
Chemistry 2
Me
N
O
S
N
10l
O
X
N
H
H
H
0.197
0.345
0.640
98
<0.300
10ac
10ad
10ae
O
X
Me
X
N
4-Cl, 6-Cl
4-Cl, 6-Cl
4-Cl, 6-Cl
0.093
0.017
4,040
64
<0.300
N
O
O
Me
1
-(4-methanesulfonyl-
X
Chemistry 13
Chemistry 2
phenoxy)
Chemistry 28
Me
Me
N
N
N
>10 (22%
10 uM)
@
10m
O
X
N
X
N
O
X
N
X
N
2
N
N
Me
Me
Me
Chemistry 14
Chemistry 2
Chemistry 14
Chemistry 28
Me
O
S
10n
N
X
N
O
X
Me
X
N
N
Me
O
N
N
1-(4-methanesulfonyl-
phenoxy)
O
X
Chemistry 15
Chemistry 2
Chemistry 29
O
O
Me
10o
O
X
N
H
H
H
2,440
9,520
0.955
10af
10ag
10ah
O
S
Me
X
N
NMe
2
4-Cl, 6-Cl
4-Cl, 6-Cl
4-Cl, 6-Cl
0.020
0.048
0.166
80
<0.300
X
X
O
O
1-(4-methanesulfonyl-
phenoxy)
Chemistry 16
Chemistry 2
Chemistry 30
O
Me
10p
X
N
O
S
Me
X
N
NMe
2
1-(4-methanesulfonyl-
X
O
X
O
N
Chemistry 17
Chemistry 2
phenoxy)
Chemistry 31
O
O
O
X
N
X
N
10q
X
N
O
S
Me
O
1-(4
X
-methanesulfo
O
nyl-
F
Chemistry 19
Chemistry 2
phenoxy)
Chemistry 32
O
S
X
N
10ai
O
Me
4-Cl, 6-Cl
0.883
1
-(4
X
-methanesulfo
O
nyl-
F
phenoxy)
Chemistry 32
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 14 of 53
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
Tables 3-5. NHE3 inhibitors 13-15 from the optimization program.
Caco2
PTotal (10- Inhibition
cm/sec)
CYP2D6
Caco2
PTotal (10- Inhibition
7cm/sec)
CYP2D6
NHE3_IC50_
uM
NHE3_IC50_
uM
JMC ID
OAr
NR2R3
R1
JMC ID
OAr
NR2R3
R1
7
IC50 uM
IC50 uM
Me
O
N
X
N
1-(4
-methanesul
fo
n
yl-
13a
O
Chemistry 17N
N
Chemistry 2
6-Cl
0.090
0.588
5
9,500
14j
O
S
M
e
Chemistry 2
X
N
NH
4-Cl, 6-Cl
0.022
0.028
30
<0.300
X
OH
X
phenoxy)O
Cl
Me
O
H
O
1
X
-(
O4-methanesu
S
lfo
M
n
e
yl-
X
N
1-(4-methanesulfonyl-
13b
Chemi
O
s
H
try 2
4-Cl, 6-Cl
4-Cl, 6-Cl
4-Cl, 6-Cl
286
2,400
14k
14l
O
S
Me
X
Ch
N
emistry 1
N
0
H
4-Cl, 6-Cl
4-Cl, 6-Cl
4-Cl, 6-Cl
0.7
0.900
phenoxy
O
)
X
phenoxy)O
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
H
O
O
N
X
N
X
1
X
-(
O
4-methanesu
S
lfon
M
e
yl-
1-(4-methanesulfonyl-
13c
Chemi
O
s
H
try 2
0.384
0.034
1,640
O
S
Me
Chemistry 11
0.011
0.020
0.168
phenoxy
O
)
X
phenoxy)O
N
H
Me
O
N
X
N
>10 (17%
10 uM)
@
1-(4
-methanesul
fo
n
yl-
X
N
NH
13d
O
Chemistry 18N
N
Chemistry 2
28
14m
O
S
M
e
Chemistry 12
51
X
OH
X
phenoxy)O
F
Me
O
O
1
-(
O
4-methanesu
S
lfo
M
n
e
yl-
X
N
OH
1-(4-methanesulfonyl-
13e
Chemistry 5
4-Cl, 6-Cl
14n
14o
14p
14q
O
S
Me
C
X
he
N
mistryNH
2
4-F, 6-Cl
4-Cl, 6-F
5-Cl, 7-Cl
6-F
X
phenoxy
O
)
phenoxy)O
X
O
O
1
-(4-methanesulfonyl-
1-(4-methanesulfonyl-
14a
O
S
Me
C
X
he
N
mistryNH
2
H
6,680
0.540
O
S
Me
C
X
he
N
mistryNH
2
0.123
0.135
X
phenoxy)
O
X
phenoxy)O
O
O
1-(4-methanesulfonyl-
1-(4-methanesulfonyl-
14b
O
S
Me
C
X
he
N
mistryNH
2
6-Cl
O
S
Me
Che 2
X N
mistryNH
X
phenoxy)
O
X
phenoxy)O
Cl
Cl
O
O
X
N
1
-(4-methanesulfonyl-
14c
O
Chemistry
S
3
NH2
Chemistry 4
NH
4-Cl, 6-Cl
0,0002
6
1,500
O S Me
phenoxy)
C
X
he
N
mistryNH
2
3,490
X
O
X
O
2
N
X
N
O
1
-(4-methanesulfonyl-
14d
O
Chemistry
N
5
N
Chemistry 4
NH
4-Cl, 6-Cl
4-Cl, 6-Cl
4-Cl, 6-Cl
4-Cl, 6-Cl
4-Cl, 6-Cl
4-Cl, 6-Cl
0.030
0.023
0.004
0.006
0.023
24
11
<0.300
<0.300
14r
O S Me
phenoxy)
Che
N
mistry
NH
7
6-Cl, 7-Cl
5-Cl, 6-Cl
4-Cl, 6-Cl
4-Cl, 6-Cl
4-Cl, 6-Cl
4-Cl, 6-Cl
0.043
6,055
0.056
0.004
0.241
X
X
X
O
2
N
X
N
O
O
N
1-(4-methanesulfonyl-
O S Me
14e
Chemistry 6
Chemistry 4
14s
X
N
NH
Chemistry 7
X
X
phenoxy)
NH
2
O
O
O
X
N
X
N
F
1
-(4-methanesulfonyl-
14f
O
S
M
e
Chemistry 4
15a
O
S
Me
105
<0.300
<0.300
<0.300
X
phenoxy)
O
NH
1-(4-methanesulfonyl-
phenoxy)
X
O
2
HN
Chemistry 2
O
O
1-(4-methanesulfonyl-
X
N
HO
14g
O
S
Me
Che
N
mistry
N
7
H
16
3
<0.300
15b
15c
15d
O
S
Me
phenoxy)
X
X
X
O
1
-(4-methanesulfonyl-
O
HN
phenoxy)
Chemistry 3
Me
H
H
O
S
N
>10 (25%
10 uM)
@
14h
O
Chemistry
N
14 N
Che
N
mistry
N
7
H
O
X
Me
X
N
N
Me
60
X
X
1-(4-methanesulfo
O
nyl-
Me
phenoxy)
Chemistry 4
O
O
1
-(4-methanesul
O
S
f2o
M
e
nyl-
N
N
14i
Che
X
N
mistry
N
7
H
0.002
O
S
Me
X
0.030
0.064
X
phenoxy)
O
X
O
Cl
1-(4-methanesulfonyl-
phenoxy)
OH
Chemistry 5
O
15e
O
S
Me
X
N
N
4-Cl, 6-Cl
110
1-(4
X
-methanesulfo
O
nyl-
OMe
phenoxy)
Chemistry 6
SAR of 1-phenoxy-2-aminoindanes.
This section describes the optimization of the HTS hits to identify favourable substitutions at
the indane scaffold. First, a pyrrolidine was introduced into the 2ꢀaminoꢀposition and
combined with phenoxy side chains at 1ꢀposition, as summarized in table 2. With 3ꢀ
acetylaminophenyl at O1, the derivative 10a with an IC50 of 0.438 µM for NHE3 was
obtained. Replacing the 3ꢀacetylamino substituent by a 3ꢀ(1,3ꢀdihydroꢀ2ꢀoxoimidazolꢀ1ꢀyl)
substituent further improves inhibition (10e, IC50 0.281 µM). Adding a 3ꢀmethyl group to the
ACS Paragon Plus Environment

Page 15 of 53
Journal of Medicinal Chemistry
1
2
3
4
5
6
7
8
9
3ꢀ(1,3ꢀdihydroꢀ2ꢀoxoimidazolꢀ1ꢀyl) substituent results in the potent compound 10l with an
IC50 value of 0.197 µM.
Testing selected compounds in the internal Cacoꢀ2 cell assay suggests reasonable
ꢀ7
permeability, as indicated by PTotal values >20 * 10 cm/sec as threshold. However, testing
for cytochrome P450 (CYP) inhibition reveals that only the less potent 10e showed low
CYP2D6 inhibition (IC 3.80 µM), which subsequently was one key optimization parameter.
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
5
0
No issues were observed for other cytochrome P450s tested.
Introducing the 4ꢀaminosulphonylphenylꢀ2,3ꢀdiꢀchlorophenoxy side chain led to the potent
compound 10b (IC50 0.123 µM), which still exhibited CYP2D6 inhibition (IC50 0.40 µM).
Other substituents at the phenoxyꢀgroup, heteroꢀbicyclic systems or a second ring in 3ꢀ or 4ꢀ
position of the phenoxyꢀmoiety did not improve NHE3 inhibition.
Introducing a 3,5ꢀdimethylꢀ4Hꢀ1,2,4ꢀtriazole ring into 4ꢀposition of the phenoxy chain
resulted in 10m with good NHE3 inhibition (IC50 0.345 µM). Introducing additional
substituents to the phenoxy further improves NHE3 inhibition, e.g. 10v (IC50 0.078 µM).
While it also showed favourable permeability, it inhibited CYP2D6 with an IC50 of 0.730 µM.
Replacing the chlorine of 10v by trifluoromethyl lowered potency (10w, IC 0.118 µM), as
5
0
was also observed with two methyl substituents in positions 2 and 3 of the phenoxyꢀside chain
10x, IC50 0.170 µM). Modulating the nature of both hydrogen bond acceptor atoms by
(
replacing the 1,2,4ꢀtriazolꢀ4ꢀyl ring with a dimethylꢀisoxazole ring with corresponding
acceptor atom topology slightly lowers activity (c.f. 10z, IC50 0.411 µM compared to 10m).
This is partially attributed to unique electronic features of the 1,2,4ꢀtriazolꢀ4ꢀyl ring system.
Next we varied halogen substituents on the indane. Introducing an indaneꢀ6ꢀchloro substituent
into 10v resulted in 10ab with an IC of 0.094. Interestingly CYP2D6 inhibition was reduced
50
by this variation (IC50 2.500 µM) compared to 10v.
Then we replaced the monobasic N2ꢀpyrrolidine on the indane with dibasic moieties. Of
particular interest is 10ac with the Nꢀmethylꢀ1,4ꢀdiazepane, which showed an IC50 of 0.093
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 16 of 53
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
µM. While this dibasic molecule also revealed a remarkable permeability in the Cacoꢀ2 assay,
it is a strong CYP2D6 inhibitor (IC50 < 0.300 µM).
Combining the Nꢀmethylꢀ1,4ꢀdiazepane with a 4ꢀ(3,5ꢀdimethylꢀ1,2,4ꢀtriazolꢀ4ꢀyl)ꢀphenoxy
substituent then led us to identify the very potent compound 10ad (IC 0.017 µM), again with
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
a 4,6ꢀdichloroꢀindane scaffold. While this compound showed low CYP2D6 inhibition, it was
found poorly permeable, suggesting that low permeability and lack of CYP2D6 inhibition are
related and linked to physicochemical properties.
Molecules with linear dibasic side chains also led to high NHE3 inhibition. One example was
compound 10af with a Nꢀmethylꢀ(Nꢀ2ꢀdimethylaminoethyl)ꢀamino side chain (IC50 0.020
µM), again with CYP2D6 inhibition (IC50 < 0.300 µM). This led us to focus on less basic
derivatives like 3ꢀfluoroꢀpyrrolidyl (e.g. 10ah, IC 0.166 µM). It became clear that basicity,
5
0
although favourable for NHE3, should be lowered for permeability and CYP2D6 inhibition.
We then pursued structureꢀactivity studies with indanes carrying chlorine atoms in position 4
and 6 together with 4ꢀmethylsulfonylphenoxy or 4ꢀ(3,5ꢀdimethylꢀ1,2,4ꢀtriazolꢀ4ꢀyl)ꢀphenoxyꢀ
side chains. Those were combined with less basic amines (c.f. table 3), which led to the
identification of 13a (IC 0.090 µM). While only weak inhibition of CYP2D6 was observed
5
0
(IC50 9.500 µM), this molecule still showed low permeability. Interestingly replacing the
chlorine atom in 13a in the 4ꢀ(3,5ꢀdimethylꢀ1,2,4ꢀtriazolꢀ4ꢀyl)ꢀphenoxy side chain by a
smaller, lipophilic fluorine atom and adding another chlorine atom to the indane in 4ꢀposition
resulted in improved NHE3 potency for 13d (IC50 0.034 µM) without significant CYP2D6
inhibition. On the other hand this increased lipophilicity of 13d with additional halogen
ꢀ7
decoration led to acceptable permeability (28 * 10 cm/sec). Based on the entire in vitro
profile, compound 13d was perceived as best compound and further investigated, as described
below. This compound was also named SAR197.
Replacing the 2ꢀfluoroꢀ4ꢀ(3,5ꢀdimethylꢀ1,2,4ꢀtriazolꢀ4ꢀyl)ꢀphenoxy side chain from 13d by
the 4ꢀmethylsulfonylphenoxy side chain resulted in compound 13b with lower activity (IC50
ACS Paragon Plus Environment

Page 17 of 53
Journal of Medicinal Chemistry
1
2
3
4
5
6
7
8
9
0.588 µM), higher CYP2D6 inhibition of 2.400 µM, but higher permeability. Therefore
increased lipophilicity and less basic amino side chains produced interesting NHE3 inhibitors
with 13d as best analogue.
Further investigations were directed towards dibasic amines combined with lipophilic
substitutions (c.f. table 4). As aminoꢀside chains, we have employed piperazine, 3ꢀaminoꢀ
piperidine, 1,4ꢀdiazepane plus bicyclic and spirocyclic systems (14a-14s). The unsubstituted
indane with piperazine and 4ꢀmethylsulfonylphenoxy side chains only led to moderate NHE3
inhibition (14a, IC50 6.680 µM). Adding a chlorine in 6ꢀposition increases activity (14b, IC50
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
0.540 µM), while two chlorines in 4ꢀ and 6ꢀpositions resulted in the potent 14j (IC50 0.022
µM). Although permeability was acceptable, it inhibited CYP2D6 with an IC50 value < 0.300
µM. The replacement of the 4ꢀchloro substituent in 14j by a fluorine resulted in a slightly
lower NHE3 inhibition (14n, IC50 0.168 µM). Replacing the other 6ꢀchloro atom in 14j by a
fluorine has the same detrimental effect on activity, as exemplified by 14o (IC50 0.123 µM).
Adding a methyl group in 3ꢀposition to the piperazine in 14j led to the equipotent 14m (IC50
0.020 µM). Replacing piperazine by octahydroꢀpyrroloꢀpyrrole did slightly affect NHE3
inhibition (14k, IC50 0.028 µM), but had a detrimental effect to permeability and CYP2D6
(IC50 0.900 µM). In contrast, replacing the piperazine by the spirocyclic diazaspiroꢀnonane
side chain slightly increases NHE3 inhibition (14l, IC 0.011 µM).
50
Some active molecules were obtained with 3ꢀaminopiperidine side chains. Here the 3ꢀ
aminopiperidine 14f (IC50 0.004 µM) was 5 times more active than the piperazine 14j. The
activity of the 1,4ꢀdiazepane side chain in 14g with an IC50 of 0.006 µM was comparable to
the 3ꢀaminopiperidine 14f, although showing low permeability and CYP2D6 inhibition (IC50
< 0.300 µM). Adding a chlorine to the 4ꢀmethylsulfonylphenoxy chain of 14g resulted in 14i
with better NHE3 inhibition (IC50 0.002 µM).
Preparing the analogue with a 4ꢀ(3,5ꢀdimethylꢀ1,2,4ꢀtriazolꢀ4ꢀyl)ꢀphenoxy side chain with the
1,4ꢀdiazepane side chain led to a molecule with reduced NHE3 activity (14h, IC50 0.023 µM),
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 18 of 53
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
low CYP2D6 inhibition, but also low permeability. Exploring other 4ꢀheteroarylꢀphenoxy side
chains in combination with 3ꢀaminoꢀpiperidine maintained NHE3 activity (14d, IC50 0.030
µM and 14e IC50 0.023 µM). However, both compounds are inhibitors of CYP2D6, which is
typical for unsubstituted imidazoles and triazoles. Most active was 14c with an IC of 0.0002
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
µM, but also low permeability and significant CYP2D6 inhibition.
In a final step, the dibasic aminoꢀsubstituent was substituted in order to modulate accessibility
and pKa of the basic center (c.f. table 5). Although permeability was acceptable, none of the
substitutions improved CYP2D6 inhibition. In summary, increasing lipophilicity and bulk in
specific indane regions and introducing the favourable 4ꢀ(3,5ꢀdimethylꢀ1,2,4ꢀtriazolꢀ4ꢀyl)ꢀ
phenoxy side chain combined with a reduced basicity of the pyrrolidine produced an
interesting NHE3 inhibitor with good activity and permeability, but without CYP2D6
inhibition. Based on the multidimensional SAR analysis of key optimization parameters, 13d
(table 3) was the most interesting candidate for in vitro and in vivo profiling. Therefore further
investigations for this compound are reported below.
3D-QSAR models and interpretation for NHE3 inhibition.
Figure 2A summarizes the flexible alignment of 13d (grey carbons) onto its Xꢀray crystal
structure (orange carbons). Crystal data and model quality from Xꢀray structure determination
are summarized in the Supporting Information. The Xꢀray structure reveals a planar
orientation of the indane to the pyrrolidine ring with its hydroxyꢀgroup in pseudoꢀequatorial
position. It also confirmed the transꢀS,Sꢀconfiguration of substituents in 1,2ꢀposition of the
indane scaffold. The phenoxy side chain adopts an extended orientation to the indane with its
central phenyl oriented in an orthogonal manner and the triazole in a plane similar to the
indane. The additional fluorine at the phenoxyꢀside chain is pointing downwards in figure 2A.
All features of both 13d structures are superimposed, while small differences in side chain
orientations are obvious. For consistency, we used the same pairwise alignment for all
ACS Paragon Plus Environment

Page 19 of 53
Journal of Medicinal Chemistry
1
2
3
4
5
6
7
8
9
molecules, which resulted in the superposition of 67 molecules (figure 2B), producing
consistent 3DꢀQSAR models. In earlier project phases, models with fewer compounds were
regularly built and updated. Conclusions from these models after validation were applied in
combination with SAR consideration to support the next optimization cycles. As the goal of
the optimization was to balance NHE3 with CYP2D6 inhibition and permeability, novel ideas
were ranked in these models plus in silico ADMET models described below. These models
were updated using novel structures and experimental data to adapt to the chemical space to
be explored during optimization. Here we present a final version of the model with broader
variations around the 1ꢀphenoxyꢀ2ꢀaminoindanes.
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
2
Using a 2 Å grid spacing, a CoMFA model with an r (cv) value of 0.718 for 6 PLS
2
components and a conventional r of 0.945 with a standard error of 0.235 was obtained (figure
3
). The steric field descriptors explain 63% of the variance, while the electrostatic field
accounts for 37% from the normalized sum of standard deviations after CoMFA_STD scaling
multiplied with PLS coefficients.
Figure 2. A) Alignment for 13d (grey carbons) onto its Xꢀray structure (orange carbons)
using QXP tfit. B) Superposition of 67 NHE3 inhibitors for 3DꢀQSAR by pairwise alignment
onto the Xꢀray structure.
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 20 of 53
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. Experimental versus fitted NHE3 inhibition (pIC50) for the CoMFA model using
6
7 compounds.
Although crossꢀvalidation evaluates the predictive power of a statistical model, the leaveꢀoneꢀ
2
out method might produce too optimistic r (cv) values. Thus, repeated PLS analyses were run
1
00 times with two and five randomly chosen crossꢀvalidation groups containing randomly
2
selected compounds for prediction. The mean r (cv) value of 0.698 for 6 PLS components and
5 crossvalidation groups (SD: 0.03, high: 0.771, low: 0.605) is slightly lower than using the
2
leaveꢀoneꢀout method. Using only two crossvalidation groups a lower mean r (cv) value of
0
.636 was observed, (SD: 0.08, high: 0.768, low: 0.372). 100 randomizations of biological
2
activity resulted in a mean r (cv) of ꢀ0.13 (SD: 0.14, high: 0.04, low: –0.74), showing that the
true model is significantly better than a random model.
A statistical PLS model of slightly lower quality was obtained using CoMSIA steric,
electrostatic and hydrophobic fields. Using a 2 Å grid spacing, a CoMSIA model with an
2
2
r (cv) value of 0.736 for 5 PLS components and a conventional r of 0.895 with a standard
ACS Paragon Plus Environment

Page 21 of 53
Journal of Medicinal Chemistry
1
2
3
4
5
6
7
8
9
error of 0.321 was obtained. The steric field descriptors explain 16 % of the variance, the
electrostatic descriptors 37 % and the hydrophobic field explains the remaining 47 %.
Further validation also supported the finding of a stable and predictive model. When
2
averaging over 100 PLS analyses with two random crossꢀvalidation groups, a mean r (cv) of
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
0
.631 (SD: 0.07, high: 0.756, low: 0.420) results. This value is increased for 5 crossvalidation
2
groups: Here a mean r (cv) value of 0.702 for 5 PLS components results (SD: 0.03, high:
2
0
0
.761, low: 0.622). When randomizing the biological activities, a mean r (cv) of ꢀ0.15 (SD:
.10, high: 0.04, low: ꢀ0.46) is obtained from 100 individual runs.
To evaluate the predictivity of both models for new compounds, the dataset was subdivided
into a training set of 59 and a test set of 8 compounds by statistical design using hierarchical
2
clustering with UNITY fingerprints. For CoMFA, a model with an r (cv) value of 0.672 for 4
2
PLS components and an r value of 0.885 with a standard error of 0.333 resulted, while for
2
2
CoMSIA, a model with an r (cv) value of 0.720 for 5 PLS components and an r of 0.897 with
a standard error of 0.318 was obtained. Both models were similar in interpretation to models
from the entire dataset. Applying both model to the test set of 8 compounds after alignment
2
results in predictive r values of 0.830 and 0.694 for CoMFA and CoMSIA, suggesting that
the quality of external predictions is acceptable for to allow for synthesis prioritization.
Affinity predictions for these models are given in the Supporting Information.
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 22 of 53
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. CoMFA and CoMSIA std*coeff contour maps with 13d (34 nM). A) CoMFA steric
contour map: Green contours (>80 % contribution) refer to regions, where steric bulk is
favorable for affinity, while yellow contours (<20 % contribution) indicate disfavored areas.
B) CoMFA electrostatic contour map: Blue contours (>80 % contribution) refer to regions,
where negatively charged substituents are unfavorable for affinity; red contours (<20 %
contribution) indicate regions, where negatively charged substituents are favorable. C)
CoMSIA steric contour map: Green contours (>80 % contribution) refer to regions, where
steric bulk is favorable, yellow contours (<20 % contribution) indicate disfavored areas. D)
CoMSIA electrostatic contour map: Blue contours (>80 % contribution) refer to regions,
where negatively charged substituents are unfavorable, red contours (<20 % contribution)
indicate regions, where negatively charged substituents are favorable. E) CoMSIA
hydrophobic contour map: White contours (>80 % contribution) refer to regions, where
hydrophilic substituents are favorable, yellow contours (<20 % contribution) indicate regions,
where hydrophobic substituents are favorable.
The steric and electrostatic std*coeff fields from CoMFA are shown in figure 4 with 13d (IC50
0
.034 µM). Green contours in figure 4A indicate regions, where steric bulk is favorable for
NHE3 activity, while yellow contours highlight regions, where bulky substituents are
ACS Paragon Plus Environment

Page 23 of 53
Journal of Medicinal Chemistry
1
2
3
4
5
6
7
8
9
detrimental. A similar analysis is provided for the electrostatic SD*coeff field (figure 4B).
Blue contours refer to regions, where an increase of positive charge (or less negative charge)
is favored to enhance inhibition, while red contours indicate regions, where an increase of
negative charge is favorable for inhibition.
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
Steric bulk at 6ꢀposition of the indane is favorable for activity, as indicated by a chlorine at
this position. In contrast, larger substituents or extending bulk more towards the indaneꢀ7
position is unfavorable, as exemplified by a nearby yellow contour region between C6 and
C7. Other important regions for NHE3 activity are located close to the pyrrolidine 4ꢀposition
and the attached 3ꢀhydroxyl group, while other substitutions on the fiveꢀmembered system are
detrimental, as indicated by yellow contour regions above and below this ring on the right side
of this figure. Finally, one of the methylꢀgroups on the triazole ring is favorable for NHE3
activity, while further substitution on the opposite site is less favorable. The optimal
orientation of the methyl group is indicated by a yellow contour next to this substituent.
The conclusions from CoMSIA and CoMFA are similar; the contour map for the steric field is
displayed in figure 4C. As above, green contours indicate regions, where steric bulk is
favorable, while yellow contours highlight regions, where steric bulk is detrimental for
affinity. In addition to CoMFA, the positive influence of methyl substitution at the triazole on
affinity is highlighted by a green region surrounding the methyl group on one side of the
triazole ring. This differing interpretation results from the complementary information
captured in both field types. In CoMFA forceꢀfield derived molecular interaction fields,
interactions of the ligand to binding site areas are indicated outside the volume of the ligand.
In contrast, CoMSIA works by identifying similarities between ligand atoms and analogue
substructures within the volume of the ligand.
The interpretation of electrostatic contour regions for CoMFA and CoMSIA (figures 3B,D)
shows that an increase in positive charge is preferable at the 3ꢀhydroxyꢀposition of the
pyrrolidine ring (blue contour), which is in agreement with the experimental preference for a
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 24 of 53
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
basic amine. The less basic hydroxylꢀgroup was studied in our efforts to balance membrane
permeability with NHE3 activity. What is more, the CoMFA contours highlight the impact of
more negative charge at the opposite side of the pyrrolidine ring system. Interestingly the
positive effect of the chlorine atom attached at C4 of the indane is due to the preference of
negative charge in this area, as obvious from a red contour region in this area. This is related
to the indication of more positive charge in two blue contour regions parallel either side of the
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
plane of the aromatic ring, which suggests that a less negative aromatic
πꢀsystem is favorable,
primarily caused by substitution of the aromatic ring with halogen atoms.
The positive effect of the 2,5ꢀdimethylꢀtriazolylꢀsubstituent attached in 4ꢀposition to the
phenoxy group can be discerned from inspection of the electrostatic contour regions
surrounding the aromatic triazole system. The favorable effect of the acceptor nitrogen atoms
in this aromatic ring are indicated by a red contour region surrounding these atoms.
Furthermore, a more negative πꢀsystem of this aromatic ring is favorable for NHE3 activity,
as indicated by a red contour region in CoMFA. The favorable electronegative character of
this ring with acceptor nitrogen atoms can also be seen from CoMSIA red contour regions.
The interpretation of hydrophobic CoMSIA fields is given in figure 4E. Grey contours
indicate regions, where hydrophilic interactions are favorable for affinity (>80 %
contribution). Those regions are located close by the 2,5ꢀdimethylꢀtriazolylꢀmoiety, thus
showing the influence of the polar nitrogen atoms in this ring. In addition, polarity is also
favored around the hydroxyꢀgroup of the pyrrolidine ring. The effect of favorable polarity
appears to be correlated with electrostatic properties for these substituents and can be deduced
by CoMSIA with its additional hydrophobic field. Additionally regions surrounding the
aromatic triazole appear to be preferable for polar interactions.
Preferred hydrophobic interactions are indicated by yellow contour regions (<20 %
contribution); these are primarily located at both chlorine atoms at C4 and C6 of the indane.
The CoMSIA analysis attributes the favorable effect of this diꢀsubstitution substitution
ACS Paragon Plus Environment

Page 25 of 53
Journal of Medicinal Chemistry
1
2
3
4
5
6
7
8
9
primarily to lipophilicity, while the CoMFA analysis predicts a steric effect for the chlorine at
C6 and negative electrostatics for the C4 position. This highlights that both QSAR analyses in
the absence of a 3Dꢀreceptor binding site might offer differing explanations related to
physicochemical trends and are thus complement each other. These interpretations could lead
to investigations to clarify the likely causes of the positive effects in ligand SAR.
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
In silico ADMET models for Caco-2 permeability and CYP2D6 inhibition.
To further characterize the 1ꢀphenoxyꢀ2ꢀaminoindanes and balance ADMET properties versus
NHE3 inhibition during optimization, we have applied in silico ADMET models to compare
predictions for Cacoꢀ2 permeability and CYP2D6 inhibition for molecules in tables 2ꢀ5 with
43
experimental data. These models were routinely generated from internal data for multiple
46
47
series using MOE descriptors and Cubist regression trees. Preliminary models were
applied for prediction and regularly updated during the optimization process. In general,
satisfactory agreement of models for Cacoꢀ2 permeability and CYP2D6 inhibition with
experimental data was observed.
The current Cacoꢀ2 model was built for logPTotal values reflecting flux from in vitro
43
permeability using a training set of 31495 and a test set of 7769 compounds. 66 MOE
descriptors and 30 rules are present in the final consensus model of five regression trees with
2
2
2
a r (cv) value of 0.689 using 10ꢀfold crossvalidation, an r of 0.706 and a predictive r of
43
0
.689 for the test set. For 29 out of the 67 compounds discussed here, experimental Cacoꢀ2
2
data were compared with this in silico model. An r value of 0.781 with a standard error of
estimate of 0.429 was obtained, as shown in figure 5A. Predictions for these in silico ADMET
models are given in the Supporting Information.
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 26 of 53
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 5. Experimental versus predicted Cacoꢀ2 permeability (A) and CYP2D6 inhibition (B)
for 29 or 27 1ꢀphenoxyꢀ2ꢀaminoindanes from prediction using in silico ADMET models.
One outlier for the Cacoꢀ2 model in figure 5A is the methylsulfone 14k with a dibasic cyclic
substituent at the indane N2 position (NHE3 0.028 µM, Cacoꢀ2 permeability 0.7 * 10ꢀ7
cm/sec). It is predicted too optimistic by the in silico Cacoꢀ2 permeability model, although
there is no unusual feature in the chemical structure. Quality control by MS (mass
spectroscopy) indicates high purity (~100%) and 14k was stable in human liver microsomes
(
~0% metabolism). However, solubility of this compound at pH 7.4 is unexpectedly only low
to moderate with 0.227 mg/mL in the internal nephelometric solubility assay. Moreover, in
the Cacoꢀ2 experiment this compound showed a low recovery rate of only 63% after transport.
These latter two factors might limit the experimental data quality in the Cacoꢀ2 assay for this
molecule, while in silico model insufficiencies for a global 2DꢀQSAR model based on many
diverse training data cannot be excluded. This topic was not evaluated further.
The CYP2D6 inhibition model was built for pIC50 values using a training set of 1646 and a
4
3
test set of 398 compounds. 52 MOE descriptors and 7 rules are present in the consensus
2
2
model with a r (cv) value of 0.563 using 10ꢀfold crossvalidation, an r of 0.624 and a
2
43
predictive r of 0.518 for the test set. For 27 1ꢀphenoxyꢀ2ꢀaminoindanes, the CYP2D6 pIC50
ACS Paragon Plus Environment

Page 27 of 53
Journal of Medicinal Chemistry
1
2
3
4
5
6
7
8
9
2
values were compared with predictions from the model. A correlation coefficient r of 0.862
with a standard error of estimate of 0.232 was obtained (figure 5B). It should be mentioned
that the experimental data range for the CYP2D6 assay covers only 1.5 log units, as the assay
is intended to provide qualitative information for potent and inactive molecules. Nevertheless,
the data quality and spread was sufficient for model building. Resulting models are
prospectively applied in a qualitative manner. This performance suggests both in silico models
are predictive for this chemical series, which was further supported during lead optimization.
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
3
D-QSAR models and interpretation for CYP2D6 inhibition.
The postulated binding mode of 10aa (NHE3 IC50 0.737 µM, CYP2D6 IC50 < 0.3 µM) in the
30
site of the apoꢀCYP2D6 Xꢀray structure 2f9q is displayed in figure 6A. Substrates and
inhibitors of CYP2D6 often contain a basic nitrogen and an aromatic ring, which allow
potential interactions with carboxylic acids Glu216 and Asp301 (yellow carbon atoms) and
30
the aromatic ring of Phe120 at the back of the binding site. The Phe120 contacts the indane
system, thereby orienting it within the site, while potentially directing the 4ꢀchlorine towards
heme and the 6ꢀchlorine towards a small lipophilic area near Leu484 and Val370 side chains.
The basic pyrrolidine nitrogen atom appears to interact with the Glu216 side chain by a salt
bridge, while additional basic centers at the pyrrolidine might interact with Asp301. The 4ꢀ
methylsulfonylphenoxy side chain is situated in a larger pocket near Phe483 and Leu213. The
2
terminal SO group is possibly interacting with Ser217ꢀOH and the Phe483ꢀNH, while the
methyl group is oriented towards a lipophilic area above the Leu484 side chain.
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 28 of 53
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 6. A) Postulated binding mode of 10aa (CYP2D6 IC50 <0.3 µM, grey carbons) from
30
docking into the apoꢀCYP2D6 binding site from the Xꢀray structure 2f9q (resolution 3.0 Å).
The binding site is displayed with solvent accessible surface plus mapped hydrogenꢀbonding
properties. B) Superposition of 27 1ꢀphenoxyꢀ2ꢀaminoindanes as CYP2D6 inhibitors for 3Dꢀ
QSAR by docking into the CYP2D6 Xꢀray structure.
ACS Paragon Plus Environment

Page 29 of 53
Journal of Medicinal Chemistry
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 7. CoMFA std*coeff contour maps with 10aa (CYP2D6 IC50 <0.3 µM) plus selected
3
0
residues from the CYP2D6 binding site from PDB 2f9q. For colours, contour levels and
interpretation see figure 4. A) Steric contour map. B) Electrostatic contour map. C) Steric
contour map with CYP2D6. D) Electrostatic contour map with CYP2D6.
Assuming that all 27 phenoxyindanes bind in a similar way results in a significant 3DꢀQSAR
2
model (figure 6B). A CoMFA model with an r (cv) value of 0.584 for 4 PLS components and
2
a conventional r of 0.928 with a standard error of 0.178 resulted. The steric field descriptors
explain 77% of the variance, while the electrostatic field accounts for 23%. For CoMSIA, a
2
2
less significant model with an r (cv) value of 0.462 for 3 PLS components and an r value of
0
.869 with a standard error of 0.236 was obtained. The steric field explains 23% of the
variance, the electrostatic field 28% and the hydrophobic field explains the remaining 49%.
2
In both cases, leaveꢀoneꢀout r (cv) values are lower compared to the NHE3 inhibition models,
probably due to the smaller size of the training set plus the inadequate spread of pIC50 values
with experimental cutoffs at both ends of the scale. Nevertheless, the model interpretation
(figure 7) agrees with SAR trends above for structural optimization.
ACS Paragon Plus Environment