A Comparative Assessment Study of Known Small-Molecule Keap1-Nrf2 Protein-Protein Interaction Inhibitors: Chemical Synthesis, Binding Properties, and Cellular Activity

Article abstract of DOI:10.1021/acs.jmedchem.9b00723

Inhibiting the protein-protein interaction (PPI) between the transcription factor Nrf2 and its repressor protein Keap1 has emerged as a promising strategy to target oxidative stress in diseases, including central nervous system (CNS) disorders. Numerous non-covalent small-molecule Keap1-Nrf2 PPI inhibitors have been reported to date, but many feature suboptimal physicochemical properties for permeating the blood-brain barrier, while others contain problematic structural moieties. Here, we present the first side-by-side assessment of all reported Keap1-Nrf2 PPI inhibitor classes using fluorescence polarization, thermal shift assay, and surface plasmon resonance - and further evaluate the compounds in an NQO1 induction cell assay and in counter tests for nonspecific activities. Surprisingly, half of the compounds were inactive or deviated substantially from reported activities, while we confirm the cross-assay activities for others. Through this study, we have identified the most promising Keap1-Nrf2 inhibitors that can serve as pharmacological probes or starting points for developing CNS-active Keap1 inhibitors.

Full text of DOI:10.1021/acs.jmedchem.9b00723

Subscriber access provided by Nottingham Trent University
Article
A Comparative Assessment Study of Known Small#
Molecule Keap1#Nrf2 Protein#Protein Interaction Inhibitors:
Chemical Synthesis, Binding Properties, and Cellular Activity
Kim Tai Tran, Jakob S. Pallesen, Sara Marie Øie Solbak, Dilip Narayanan, Amina Baig,
Jie Zang, Alejandro Aguayo-Orozco, Rosa Carmona, Anthony Garcia, and Anders Bach
J. Med. Chem., Just Accepted Manuscript • DOI: 10.1021/acs.jmedchem.9b00723 • Publication Date (Web): 14 Aug 2019
Downloaded from pubs.acs.org on August 14, 2019
Just Accepted
“Just Accepted” manuscripts have been peer-reviewed and accepted for publication. They are posted
online prior to technical editing, formatting for publication and author proofing. The American Chemical
Society provides “Just Accepted” as a service to the research community to expedite the dissemination
of scientific material as soon as possible after acceptance. “Just Accepted” manuscripts appear in
full in PDF format accompanied by an HTML abstract. “Just Accepted” manuscripts have been fully
peer reviewed, but should not be considered the official version of record. They are citable by the
Digital Object Identifier (DOI®). “Just Accepted” is an optional service offered to authors. Therefore,
the “Just Accepted” Web site may not include all articles that will be published in the journal. After
a manuscript is technically edited and formatted, it will be removed from the “Just Accepted” Web
site and published as an ASAP article. Note that technical editing may introduce minor changes
to the manuscript text and/or graphics which could affect content, and all legal disclaimers and
ethical guidelines that apply to the journal pertain. ACS cannot be held responsible for errors or
consequences arising from the use of information contained in these “Just Accepted” manuscripts.
is published by the American Chemical Society. 1155 Sixteenth Street N.W.,
Washington, DC 20036
Published by American Chemical Society. Copyright © American Chemical Society.
However, no copyright claim is made to original U.S. Government works, or works
produced by employees of any Commonwealth realm Crown government in the course
of their duties.

Page 1 of 77
Journal of Medicinal Chemistry
1
2
3
4
5
6
7
8
9
A Comparative Assessment Study of Known Small-Molecule Keap1-Nrf2
Protein-Protein Interaction Inhibitors: Chemical Synthesis, Binding
Properties, and Cellular Activity
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
†,‖
†,‖
†
†
†
†
Kim T. Tran, Jakob S. Pallesen, Sara M. Ø. Solbak, Dilip Narayanan, Amina Baig, Jie Zang, Alejandro
†,#
†
†,◊,‡
Anders Bach†,*
Aguayo-Orozco, Rosa M. C. Carmona, Anthony D. Garcia,
†Department of Drug Design and Pharmacology, Faculty of Health and Medical Sciences, University of
Copenhagen, Universitetsparken 2, DK-2100 Copenhagen, Denmark
◊École Nationale Supérieure de Chimie de Rennes, 11 Allée de Beaulieu, CS 50837, Rennes Cedex 7 35708,
France
KEYWORDS: Kelch-like ECH-associated protein 1 (Keap1); nuclear factor erythroid 2-related factor 2 (Nrf2);
small-molecule inhibitors; protein-protein interactions (PPIs); validation study
ABSTRACT
Inhibiting the protein-protein interaction (PPI) between the transcription factor Nrf2 and its repressor protein
Keap1 has emerged as a promising strategy to target oxidative stress in diseases, including CNS disorders.
Numerous non-covalent small-molecule Keap1-Nrf2 PPI inhibitors have been reported to date, but many
feature suboptimal physicochemical properties for permeating the blood-brain barrier, while others contain
problematic structural moieties. Here, we present the first side-by-side assessment of all reported Keap1-Nrf2
PPI inhibitor classes using fluorescence polarization (FP), thermal shift assay (TSA), and surface plasmon
resonance (SPR)—and further evaluate the compounds in an NQO1 induction cell assay and in counter tests for
1
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 2 of 77
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
non-specific activities. Surprisingly, half of the compounds were inactive or deviated substantially from
reported activities, while we confirm the cross-assay activities for others. Through this study, we have
identified the most promising Keap1-Nrf2 inhibitors that can serve as pharmacological probes or starting points
for developing CNS active Keap1 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
2
ACS Paragon Plus Environment

Page 3 of 77
Journal of Medicinal Chemistry
1
2
3
4
5
6
7
8
9
INTRODUCTION
The nuclear factor erythroid 2-related factor 2 (Nrf2) transcription factor is a principal regulator of the cellular
defense system against oxidative stress, promoting the transcription of an expansive set of antioxidant and
cytoprotective enzymes, such as NAD(P)H:quinone oxidoreductase 1 (NQO1), heme oxygenase 1 (HO-1), and
glutathione S-transferase for combating reactive oxidants and electrophilic species.1-3 Under quiescent
conditions, Nrf2 is kept at low concentrations by the cytosolic repressor protein Kelch-like ECH-associated
protein 1 (Keap1), which associates with the Nrf2-ECH homology 2 (Neh2) domain of Nrf2 to facilitate cullin 3
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
(
Cul3) mediated polyubiquitination and proteasomal degradation of Nrf2.2, 4 Structurally, Keap1 consists of a
broad complex, tramtrack, and bric-à-brac (BTB) domain, an intervening region (IVR), and a Kelch domain. It
has been proposed that a Keap1 homodimer, via its two Kelch domains, binds a single Nrf2 peptide at a high-
affinity ETGE-site and low-affinity DLG-site contained in the Neh2 domain. This creates a fixed spatial alignment
of Nrf2 that allows polyubiquitination by the BTB domain-docked Cul3 to take place.5-8 Under oxidative or
electrophilic assault, specific sensor cysteine residues on the IVR and/or BTB domains of Keap1 are modified to
disrupt this oligomeric Nrf2-Keap1-Cul3 complex, thus prohibiting polyubiquitination.5, 9, 10 In combination with
de novo synthesis, Nrf2 then accumulates and translocates into the nucleus to transactivate a large assembly of
antioxidant response element (ARE)-dependent enzymes that can reestablish redox homeostasis.1, 11, 12
Nrf2 upregulation by Keap1 inhibition has been proposed as a potential therapeutic strategy to target
oxidative stress-involved pathologies.10, 13-15 Importantly, Nrf2 activation by genetic knock-out studies and
pharmacological Keap1 inhibition has shown encouraging neuroprotective effects in animal models of cerebral
ischemia, hemorrhagic stroke, traumatic brain injury, and chronic neurodegenerative disorders.16-20
An increasingly prevalent strategy to upregulate Nrf2 is to competitively disrupt the protein-protein
interaction (PPI) between the Keap1 Kelch domain and the Nrf2 Neh2 peptide motifs.21, 22 Early research
focused on developing truncated peptide inhibitors containing the high-affinity ETGE-sequence of the Neh2
domain e.g. 9mer peptides 1 and 2 with submicromolar binding affinities to the Keap1 Kelch domain (Figure
24
1
A).23,
Although recent research has attempted to address their inherent shortcomings by e.g.
3
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 4 of 77
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
5
macrocyclization, peptide inhibitors generally suffer from low cellular permeability, stability, and activity. The
Keap1-Nrf2 interaction stands out from classical PPIs by having a relatively small interaction surface (550–780
2
7
Å ) and concave binding pocket. Furthermore, the pocket features five interconnected epitopes (P1–P5), of
which at least P1, P4, and P5 are considered hot spots for ligand-binding (Figure 1B).22, 26, 27
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
N
A
HO
NH2-LDEETGEFL-OH
R
S
R
S
HN
S
N
O
O
O
1
2
Ki = 0.8-0.9 M (FP)
N
N
HN
S
N
(S)
S
O
O
O
OH
O
O
Kd = 352 nM (comp. SPR)
N
O
OH
O
O
O
O
O
O
(S)
(S)
N
(R)
O
O
O
O
Ac-LDEETGEFL-OH
Ki = 0.05-0.2 M (FP)
3
IC50 = 2.3 M (FP)
Kd = 1.0/0.8 M (comp.
4
R = -H
IC50 = 0.94-1.5 M (FP)
Kd = 1.7 M (BLI/SPR)
7
IC50 = 0.14-0.2 M (FP)
Kd = 6 M (ITC)
8
IC50 = 0.2 M (FP)
Kd = 23 nM (comp. SPR)
SPR/SPR)
Tm (100 M) = +1 °C (TSA)

Tm (38 M) = +1.3 °C (TSA)
5
6
R = -CH2CO2H IC50 = 24-29 nM (FP)
Kd = 3.6/5-20 nM (BLI/SPR)
R = -CH2CONH2 IC50 = 63 nM (FP)
Kd = 44 nM (SPR)
OH
O
N
N
(R)
N
I
S
NO2
O
O
O
N
N
O
S
(cis)
S
N
O
N
N
N
O2N
N N
O
(
S)
N
H
O
N
N
CF3
HN
N
S
N
O
O
O
OH
O
9
IC50 < 15 nM (FP)
10 IC50 = 63 nM (FP)
11 IC50 = 118 M (2D-FIDA)
12 IC50 = 7.1 M (FP)
13 Kd = 22.8 M (SPR)
Kd = 1.3 nM (ITC)
O
N
N
N
N
O
O
H
N
HO
OH
F
X
R
N
N
N
O
O
N
N
N
O
O
N
N
O
H
H
O
H
H
O
N
N
14 X = -CH, R = -OCH2CO2H
Kd = 0.1-1 mM (SPR)
16 CD = 2.5 M (ARE-induction assay)
17 CD = 0.07 M (NQO1-induction assay)
18 IC50 = 9.8 M (FP)
15
X = -N, R = -H
HO
O
Conc.-depend. binding (SPR)
O
O
S
HN
O
N
O
NH
O
N
O
O
S
OH
S
N
HO
Cl
N
N
N
O
H
Cl
NH
O
19
Ki = 15.2 M (FP)
20 Ki = 2.9-7.2 M (FP)
21 Ki = 10.4 M (FP)

Tm (100 M) = -2 °C (TSA)
Tm (100 M) = -5 °C (TSA)
Tm (100 M) = -1 °C (TSA)
B
P1
P4
P2
P5
P3
4
ACS Paragon Plus Environment

Page 5 of 77
Journal of Medicinal Chemistry
1
2
3
4
5
6
7
8
Figure 1. (A) Structure and activities of all known non-covalent small-molecule Keap1-Nrf2 PPI inhibitors (for
inhibitors from larger SAR-studies only representative compounds are shown) and representative peptide
2
8
24
29
30
31
32
inhibitors. 1 and 2: Inoyama et al. , Hancock et al. ; 3: Hu et al. , Jnoff et al. , Bresciani et al. , PubChem ;
3
3
22
34
35
22
34
31
4
: Marcotte et al. , Jiang et al. , Jain et al. , Zhuang et al. ; 5: Jiang et al. ; Jain et al. ; Bresciani et al. ; 6:
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
4
36
37
38
27
39
Jain et al. ; 7: Winkel et al. , Szill et al. ; 8: Yasuda et al. ; 9: Davies et al. ; 10: Callahan et al. ; 11:
33
40
41
42
43
Marcotte et al. ; 12: Bertrand et al. ; 13: Kazantsev et al. ; 14: Satoh et al. ; 15: Shimozono et al. ; 16: Xu et
4
4
45
46
35
al. ; 17: Ghorab et al. ; 18: Sun et al. ; 19-21: Zhuang et al. Comp. SPR: Competitive SPR. BLI: Bio-layer
interferometry. ITC: Isothermal titration calorimetry. 2D-FIDA: Two-dimensional fluorescence intensity
distribution analysis. CD: Concentration inducing a two-fold increase in response. (B) X-ray structures of the
Keap1 Kelch domain in complex with 1 (left, purple, PDB code 1X2R), 7 (middle, turquoise, PDB code 5CGJ), and
9
(right, violet, PDB code 5FNU). The sub-pockets P1–P5 have been indicated for the Kelch domain in complex
with 1. Figures were made in PyMOL.
Several non-covalent small-molecule Keap1 inhibitors have been reported, many of which exhibit
promising in vitro binding affinities to Keap1, are able to displace Neh2 peptides in competition assays, and/or
have demonstrated downstream cellular effects (Figure 1A).22, 24, 27-46 Additionally, a number of inhibitors (3, 4,
6
, 7, 9, 11, and 14) have been confirmed by X-ray crystallography to bind the Kelch domain (see examples in
Figure 1B).10 However, since the Kelch binding pocket is still rather large compared to classical enzyme or
receptor pockets and contains multiple arginine residues (Figure 1B), bulky ligands featuring carboxylic acids
2
6
have been enriched in medicinal chemistry programs. These features generally result in poor cell permeability
and might explain the discrepancy between binding affinity and cell activity observed for e.g. compound 5.22, 27
In relation to CNS drug discovery, large molecular weight (> 470 Da), large topological polar surface area (tPSA
>
90 Å), numerous hydrogen bond donors (HBD > 2), and the presence of carboxylic acid groups generally
impede blood-brain barrier (BBB) penetration.47, 48 The CNS permeability has been experimentally tested on a
few Keap1 inhibitors (3 and 13) but with discouraging results,30, 41 and based on their low CNS multiparameter
5
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 6 of 77
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
optimization (CNS MPO) scores,47, 48 the other reported inhibitors would not be expected to pass the BBB
either.10
One strategy for addressing the issues listed above is to optimize known Keap1 inhibitors for cell and CNS
activity. To facilitate such efforts, we found it crucial to verify the medicinal chemistry data of the reported
inhibitors. In general, concerns have been raised on the low reproducibility in chemical as well as biological
research,49-51 and specifically, we suspect some of the compounds identified in biochemical assays without
confirmatory biophysical binding data are artifacts. Indeed, some of the inhibitors seen in Figure 1A contain
pan-assay interference compounds (PAINS) substructures or other problematic structural elements that make
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
them unattractive in drug development.52, 53 Yet, uncritical use of PAINS substructure filters is fraught with risk
and might inappropriately exclude a viable starting point for optimization.54
In this comparative assessment study, we carefully scrutinize the scientific data on all known small-
molecule Keap1 inhibitor classes that we are aware of (3–21, Figure 1A). We describe the challenges and
reproducibility issues in synthesizing relevant inhibitors, and we establish whether the inhibitors bind to the
Keap1 Kelch domain using the three orthogonal assays, fluorescence polarization (FP), thermal shift assay
(TSA), and surface plasmon resonance (SPR). We also evaluate the compounds in an NQO1 induction cell assay.
This allows for a direct side-by-side comparison of the inhibitors, which have otherwise been tested in various
assays and different laboratories. Finally, we tested the compounds for common liabilities such as covalent
reactivity, redox activity, and aggregation. Hereby, we elucidate which of the published Keap1 inhibitors can
serve as reliable control compounds in future research, and we identify those most suitable as starting points
for further drug discovery efforts, including development of CNS active Keap1 inhibitors.
6
ACS Paragon Plus Environment

Page 7 of 77
Journal of Medicinal Chemistry
1
2
3
4
5
6
7
8
9
RESULTS
Selection and synthesis of known small-molecule Keap1-Nrf2 PPI inhibitors.
10
Based on a recent literature investigation, we commenced this study by including all the small-molecules we
know of and that have been reported to non-covalently inhibit the Keap1-Nrf2 PPI in competitive assays (FP or
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
2
D-FIDA) (3–12 and 18–21) or to directly bind the Kelch domain in biophysical assays (SPR) (13–15). For larger
compound series, we chose the most potent or well-characterized inhibitors as representatives. We also
included compounds 16 and 17, which show downstream Nrf2-dependent gene activation,44, 45 but for which
1
0
direct inhibition or binding data have not been provided. Compound 16 was interesting due to its high CNS
drug-likeness (CNS MPO scores ~5) and structural similarity with 12, while compound 17 was included due to
its high reported cellular activity and since non-covalent binding to the Kelch domain has been suggested by
45
molecular docking. Finally, the two well-characterized peptides 1 and 2 were included as controls, leading to
a total list of 21 compounds (Figure 1A).
The 21 compounds were either purchased (1, 2, 13, 15, and 18–21) or synthesized, generally following
previously reported procedures.22, 27, 29, 34, 36, 38-40, 42, 44, 45, 55-58 Compounds 9, 12, 14, and 16 were generated with
no or only small changes from literature protocols and in acceptable yields (Supporting Information Scheme
S1-4), while the syntheses of 3, 4–8, 10, 11, and 17 were more challenging, as described next.
The previously reported asymmetric protocol30 for synthesizing the active (SRS)-stereoisomer 3 was
attempted but found laborious. The synthesis is lengthy (11 steps) and includes synthetic steps, such as a
reduction of an aromatic ketone and an N-benzyl deprotection step, which were difficult in our hands. We
29
found the initially reported non-enantioselective procedure with an end-stage resolution by chiral HPLC to be
a more feasible route giving higher yields (7% vs. < 0.1% overall yield) in fewer steps (5 steps, Scheme 1).
Achieving the correct (SRS)-stereoisomer 3 was confirmed by testing the biological activities of all four isolated
stereoisomers, since 3 was reported to be at least a 100-fold more active than the other three isomers.29
7
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 8 of 77
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
Scheme 1. Synthesis of compound 3a
O
O
N
O
OH
N
O
OH
O
O
O
O
O
(S)
(S)
O
(R)
(R)
N
(R)
N
(
S)
O
N
N
O
OH
a, b
c
O
d, e
N
(cis)
3 (SRS)
25 (RSR)
O
O
OH
NH
N
O
O
O
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
N
O
OH
N
O
OH
22
23
24
O
O
O
(S)
(R)
O
(R)
(S)
N
N
(
S)
(R)
2
6 (RSS)
27 (SRR)
a
Conditions: (a) 2-phenylethan-1-amine, POCl , MeCN, RT, 85 °C, 46%; (b) NaBH CN, 4 M HCl in 1,4-dioxane,
3
3
MeOH, RT, 1 h, 76%; (c) (3aR,7aS)-hexahydroisobenzofuran-1,3-dione, p-Xylene, 50 °C, 1 h, RT, 24 h, yield ND;
(d) flash silica chromatography; (e) chiral separation on Chiralcel OD, 19% over step c–e.
The family of 1,4-diaminonaphthalene compounds 4–8 were synthesized divergently from a common 4-
nitronaphthalen-1-amine intermediate 29 following previously reported procedures (Scheme 2).22, 34, 36, 38, 56
A
key issue with these syntheses lies in the poor stability of the 1,4-diamino intermediates 30 and 33, which were
observed to rapidly degrade, presumably through oxidation; this might explain the reproducibility issues in
terms of yields, i.e. the previously reported yield for the synthesis of 4 (64%) could not be replicated here (13–
23%), which also applies for compound 7 (overall yield of 15% vs. literature yield of 35%). To minimize
degradation of the intermediates 30 and 33, workup of the reactions were carried out relatively quickly,
purifications or characterizations were not performed and the following reactions conducted immediately after
isolation of the compound.
8
ACS Paragon Plus Environment

Page 9 of 77
Journal of Medicinal Chemistry
1
2
3
4
5
6
7
8
9
Scheme 2. Synthesis of compounds 4–8a
O
O
O
O
O
O
S
S
S
NH
N
NH
O
O
O
O
O
l
m, n
O
OH
N
NH
N
O
N
O
S
S
H
S
O
O
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
O
O
O
34
35
8
O
k
O
O
O
S
S
R
R
NH2
NH2
NH2
NH
N
O
O
O
a
b
c
d, e / f
O
O
O
NO2
NO2
29
NH
N
S
S
28
30
O
O
O
4
O
R = -OH
R = -NH2
5
6
g
O
O
O
OH
OH
OH
(S)
(S)
(S)
Br
N
N
N
h
i
j
O
NO2
NO2
32
NH2
33
NH
S
31
O
7
a
Conditions: (a) NH OH·HCl, KOH, EtOH/MeOH, 60 °C, 2 h, 55%; (b) Pd/C, H (1 atm, balloon), EtOH, RT, 2 h,
2
2
yield ND; (c) 4-methoxybenzenesulfonyl chloride, pyridine, toluene, 100 °C, 2 h, 13–23% over 2 steps; (d) ethyl
bromoacetate, K CO , DMF, RT, 3 h, 90%; (e) NaOH, MeOH/H O, 65 °C, 3 h, 67%; (f) 2-bromoacetamide, K CO ,
2
3
2
2
3
DMF, RT, 5 days, 29%; (g) t-BuONO, CuBr , TBAB, camphorsulfonic acid, MeCN, 60 °C, 24 h, 83%; (h) (S)-
2
pyrrolidine-3-carboxylic acid, CuI, K CO , DMF, Ar, 150 °C, 24 h, 96%; (i) Pd/C, H (2 bar), MeOH, EtOAc, H-Cube
2
3
2
TM
Mini Plus , RT, 1.0 mL/min, yield ND; (j) 2,3,5,6-tetramethylbenzene-1-sulfonyl chloride, pyridine, DCM, RT,
16 h, 16% over two steps; (k) 4-ethoxybenzenesulfonyl chloride, pyridine, DCM, RT, 25 min, 48%; (l) ceric
ammonium nitrate, MeCN, RT, 18 min, 89%; (m) ethyl acetoacetate, Et N, toluene, RT, 2 h, 48%; (n) NH OH·HCl,
3
2
AcOH, H O, 80 °C, 2 h, 47%.
2
Compound 10 was synthesized according to the reported procedure39, 57 with some adjustments (Scheme
). The synthetic route relies on a central Suzuki-Miyaura cross-coupling between an enol triflate 38 and an aryl
3
9
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 10 of 77
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
boronate ester building block 41 as the crucial carbon skeleton-building step. Synthesis of the enol triflate using
the one-step NaHMDS-mediated enolization/PhNTf -induced trapping procedure reported in the patent
2
application39 was not efficient in our hands, giving low yield (43% vs. 88% reported in literature ) and
significant by-product formation. We found that using a freshly-made LiHMDS as an alternative base gave a
cleaner reaction and excellent yield (quantitative). The converging SM reaction step between 38 and 41 gave
several well-known by-products, including the boronic acid, aryl boronate homo-coupling product and
protodeboronation product, but could still afford the desired cross-coupling product 42 in good yield (69% vs.
39
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
3
3% reported in literature ). Catalytic hydrogenation to deprotect the carboxylic acid and reduce the alkene
3
9
double bond diastereoselectively furnished only the cis-cyclohexane 43 in accordance with literature; this was
revealed by nuclear Overhauser effect (NOE) NMR (Supporting Information Figure S1). This facial selectivity
can be explained by a steric directing effect of the carboxybenzyl group. The cyclohexane carboxylic acid of 43
was finally coupled with 2-butylpyrrolidine and the pyrazole carboxylic acid deprotected to give 10 as a mixture
of four stereoisomers. Attempted separation of the two diastereoisomers by preparative HPLC was
unsuccessful. In the patent application, purification by HPLC is reported to give two different fractions, each
containing all four stereoisomers in slightly different proportions, of which one was directly tested as a
3
9
mixture. Having this literature result as a reference point, we did not proceed with further purification of 10.
Scheme 3. Synthesis of compound 10a
O
O
O
a, b, c
O
OTf
d
BnO
BnO
O
O
O
O
O
HO
36
37
38
N
N
N
h
i
j, k
N
N
N
O
O
O
O
O
O
O
O
O
(cis)
(cis)
N
N
BnO
HO
N
O
e, f
N
g
N
42
43
10
O
Br
B
39
40
O
41
a
Conditions: (a) Cs CO , TMS-CN, H O/1,4-dioxane, 100 °C, 2 h, 82%; (b) 12 M aq. HCl, 80 °C, 45 min, 44%; (c)
2
3
2
BnBr, DBU, MeCN, RT, 1 h, 50%; (d) LiHMDS, PhNTf , THF, Ar, -78 °C–RT, 16 h, quantitative; (e) DMF-dimethyl
2
acetal, MW, 130 °C, 15 min (telescoping); (f) 3-bromophenylhydrazine·HCl, Et N, EtOH, RT, 18 h, quantitative;
3
1
0
ACS Paragon Plus Environment

Page 11 of 77
1
Journal of Medicinal Chemistry
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
(g) B pin , Pd(dppf)Cl , KOAc, 1,4-dioxane, reflux, 4 h, 76%; (h) Pd(PPh ) , Na CO , EtOH/H O/toluene, Ar, reflux,
2
2
2
3 4
2
3
2
3
h, 69%; (i) Pd/C, H (1 atm, balloon), EtOH, RT, 24 h, quantitative; (j) 2-butylpyrrolidine hydrochloride, HATU,
2
DIPEA, DCM, RT, 18 h, 44%; (k) NaOH, i-PrOH/H O, RT, 72 h, 41%.
2
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 synthesis of 11 has previously not been reported. The synthesis was carried out by utilizing 5-bromo-
1
,3-dimethyluracil 44 as the central building block (Scheme 4). A microwave-assisted nucleophilic aromatic
5
9
substitution (S Ar) procedure by Fang et al. was employed to obtain the sulfide intermediate 45. Following
N
oxidation to sulfone 46, the 1,3-dimethyluracil core was converted into thiouracil 47 upon treatment with
6
0
thiourea under basic conditions, employing a procedure by Hirota et al. Finally, an end-stage alkylation with
alkyl bromide 49 (obtained from treating 2-trifluoromethylaniline 48 with bromoacetyl bromide) gave target
molecule 11.
Scheme 4. Synthesis of compound 11a
H
O
N
O
N
O
O
N
O
S
N
a
b
c
N
N
N
HN
Br
S
S
S
O
O
O
O
O
O
O
44
45
46
47
e
S
N
O
N
H
CF3
HN
S
O
O
d
O
11
Br
NH2
N
H
CF3
CF3
48
49
aConditions: (a) Sodium 2,4-dimethylbenzenethiolate, NMP, MW, 130 °C, 10 min, 62%; (b) mCPBA, DCM, 0 °C–
RT, 3 h, 71%; (c) (NH ) CS, NaOEt, EtOH, Ar, 100 °C (pressure vial), 16 h, 40%; (d) 2-bromoacetyl bromide, Et N,
2
2
3
1,4-dioxane, N , 0 °C–RT, 2 h, 89%; (e) Et N, DMF, 80 °C, 1.5 h, 55%.
2
3
The previously described one-step procedure by Ghorab et al. for synthesizing the quinazoline compound
4
5
1
7 by simply refluxing 50 and 51 in DMF for 24 h was futile in our hands with no observed cyclization of the
intermediate S Ar product 52. Instead, we conceived a two-step procedure (Scheme 5) consisting firstly of a
N
condensation of 50 and 51 by refluxing in neat acetic acid followed by an acetic anhydride-mediated ring-
1
1
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 12 of 77
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
closure to afford the target molecule in a good overall yield (75%) comparable to what was reported for the
one-step procedure (69%).45
Scheme 5. Synthesis of compound 17a
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
Cl
N
HN
N
N
CO2H
N
N
N
O
a
b
+
H2N
N
CO2H
51
50
52
17
a
Conditions: (a) AcOH, reflux 1 h, 91%; (b) Ac O, reflux, 1 h, 82%.
2
All compounds were thoroughly quality controlled by LC-MS and NMR, and the DMSO-d stocks were
6
1
concentration-determined by quantitative 1D H NMR (qHNMR) prior to assaying and checked for stability by
LC-MS during or after testing.
Inhibition activity in competitive FP assay.
6
1
FP is one of the most widespread fluorescence-based technique employed to identify modulators of PPIs and
26
is also the most commonly used assay for Keap1-Nrf2 PPI inhibitors; indeed, most of the compounds shown in
Figure 1A have been tested in an FP assay.22, 24, 27, 28, 30, 34-38, 40, 46, 62 Here, we established the FP assay using both
Cy5 and FAM as fluorophores attached N-terminally to a 9mer peptide-moiety of Nrf2 with the sequence
LDEETGEFL-NH . Although fluorescein has been used mostly to label Nrf2 peptides, readouts with red-shifted
2
fluorophores such as Cy5 are generally more advantageous in terms of reducing the occurrence of fluorescence
interferences.63, 64 The resulting peptide probes (here designated Cy5-Nrf2 and FAM-Nrf2) gave K values of
d
7
.5–13.6 nM when saturated by Keap1 (Supporting Information Figure S2A–C), which are similar to literature
values (e.g. 11.4 nM reported by Inoyama et al. for Cy5-LDEETGEFL-NH and 51.0 nM reported by Hancock et
2
al. for the comparable peptide probe FAM-LDEETGEFLP-OH).24, 28
1
2
ACS Paragon Plus Environment

Page 13 of 77
Journal of Medicinal Chemistry
1
2
3
4
5
Compounds 1–11 all showed robust inhibition (Figure 2A) with K values correlating with literature (Figure
i
1
vs. Table 1).22, 24, 27, 28, 30, 34, 36, 37, 39 Only compounds 8 and 11 showed some deviations from the literature (K =
i
6
0
7
.0021 µM vs. IC50 = 0.20 µM38 and K = 2.8 µM vs. IC = 118 µM , respectively). For compound 8, this
33
i
50
8
difference can be explained by a high protein-to-probe concentration ratio in the reported FP assay (250 nM
Keap1 to 10 nM probe) compared to our conditions (14 nM Keap1 to 3 nM probe); this is despite using the
same peptide probe.38 Thus, considering the affinity of the probe, such conditions would give rise to a
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
8, 65
significantly higher IC value than the K .
In fact, using the reported protein and probe concentrations and
5
0
i
assuming a K of 10 nM for the probe, a K ≈ 2 nM can be calculated from their measured IC value of 0.20 µM,
d
i
50
3
3
which is in line with our result. For compound 11, the IC value was derived from a different assay (2D-FIDA)
5
0
and might therefore not be comparable with our result. Another source of error could be ligand
concentrations, but noticeably, we here applied qHNMR to accurately quantify our stocks.
FP (Inhibition)
Cy5-Nrf2, 4% DMSO
FP (Inhibition)
Cy5-Nrf2, 8% DMSO
A
B
90
90
60
30
0
1
2
3
4
5
6
7
12
13
17
18
19
20
21
8
14
9
6
3
0
0
0
15
10
11
16
-9
-8
-7
-6
-5
-4
-3
-8
-7
-6
-5
-4
-3
-2
Log [Inhibitor/M]
Log [Inhibitor/M]
FP (Inhibition) - FP
FAM-Nrf2, 8% DMSO
FP (Inhibition) - FLINT
FAM-Nrf2, 8% DMSO
C
D
250
150000
12
12 (
18 (
19 (
20 (
21 (
1
8
200
19
100000
1
50
00
20
21
1
50000
50
0
0
-8
-7
-6
-5
-4
-3
-2
-8
-7
-6
-5
-4
-3
-2
Log [Inhibitor/M]
Log [Inhibitor/M]
Figure 2. Results from the competitive FP assay. Concentration-response curves for compounds 1–11 (A) and
12–21 (B) using the Cy5-Nrf2 probe. (C) Concentration-response curves for compounds 12 and 18–21 using the
1
3
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 14 of 77
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
FAM-Nrf2 probe and (D) the total fluorescence intensity (FLINT) measured in these experiments.
Representative curves of ≥ 3 individual measurements are here shown (as means of duplicates ± SEM).
Table 1. Activities of known small-molecule Keap1 inhibitors by FP (Cy5-Nrf2 probe), TSA, and SPR.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
Compound
FP (K /μM)
TSA (EC /μM)
TSA (∆T -max/◦C)
SPR (K /μM)
i
50
m
d
1
2
3
4
5
6
7
8
9
1
1
12
13
1
1
1
1
1
19
20
2
0.54 ± 0.009
0.051 ± 0.0002
0.56 ± 0.03
25 ± 1
13 ± 0.6
26 ± 1
5.8 ± 0.4
5.9 ± 0.6
12 ± 2
13 ± 2
7.2 ± 1
1.9 ± 0.2
9.4 ± 0.2
14 ± 0.1
12 ± 0.2
4.2 ± 0.3
22 ± 0.4
8.1 ± 0.1
18 ± 0.4
28 ± 0.3
25 ± 0.3
19 ± 0.3
15 ± 0.5
4.2 ± 0.5
0.74 ± 0.3
2.4 ± 1.0
0.37 ± 0.09
0.32 ± 0.07
0.25 ± 0.02
0.0061 ± 0.0002
0.053 ± 0.002
0.048 ± 0.002
0.0021 ± 0.0001
c
0.19± 0.03
0.47 ± 0.04
d,e
0.026
b
d,e
~0.001
0.0014
d,e
0
1
0.013 ± 0.0002
2.8 ± 0.040
NA (≤ 100)
NA (≤ 77)
NA (≤ 800)
NA (≤ 800)
NA (≤ 800)
NA (≤ 99)
10 ± 1
110 ± 8
0.019
51 ± 20
NA (≤ 354)
> 300
NA (≤ 582)
NA (≤ 452)
> 120
NA (≤ 50)
NA (≤ 14)
> 150
NA (≤ 100)
NA (≤ 77)
NA (≤ 800)
NA (≤ 800)
0.2-0.5 °C (200-800)
NA (≤ 99)
NA (≤ 49)
0.2-0.5 °C (18-72)
NA (≤ 50)
NA (≤ 25)
4
5
6
7
8
NA (≤ 49)
~280 (≤ 286)
~71 (≤ 50)
NA (≤ 200)
> 500
NA (≤ 286)
1
a
K (FP; Cy5-Nrf2 probe), EC (TSA), and K (SPR) from steady state affinity analysis of SPR sensorgrams shown
i
50
d
as mean ± SEM; based on at least three individual measurements unless otherwise stated. NA: No activity.
Maximum concentration tested is seen in parenthesis: Test concentrations of compounds 12, 13, 17, and 18
were limited by solubility-issues. For 16 and 19, EC could not be determined, but a ∆T of 0.2–0.5 °C were
50
m
found in the respective concentration ranges. Test concentrations of 19 were limited in FP by low stock
concentration due to low amount available from vendor, and further limited in TSA by a sudden drop in
fluorescence. Test concentrations of 20 and 21 were limited by increasing FP values at higher concentrations,
b
and by a sudden drop in fluorescence (20) or auto-fluorescence (21) in TSA. The affinity was too high to
c
calculate an exact K , but IC value was 0.008 μM (~2-fold lower than 8). Based on only two measurements.
i
50
d
e
Based on only one measurement. One concentration injection or OneStep gradient injection sensorgrams
were fitted to a kinetic 1:1 model.
1
4
ACS Paragon Plus Environment

Page 15 of 77
Journal of Medicinal Chemistry
1
2
3
4
To check that the inhibition curves were not due to fluorescence artifacts (e.g. auto-fluorescence or
63
chemical quenching of probe ) the total fluorescence intensity (FLINT) values were monitored and found
stable for compounds 1–11 (Supporting Information Figure S2D) as expected for genuine inhibitors that do not
disturb the fluorescence of the probe. Also, flat FP curves were seen when Keap1 was omitted from the assay
(Supporting Information Figure S2E), which further validates that the concentration-dependent reduction in FP
signal in the presence of Keap1 (Figure 2A) was due to inhibition of the Keap1-Nrf2 interaction.
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 12–18 and 21 did not show any inhibition in the FP assay, while 19 and 20 only gave a weak
signal (Figure 2B and Table 1). As these compounds were tested in higher concentrations than compounds 1–
11, a DMSO concentration of 8% was used instead of 4% as for 1–11. Under these conditions, compounds 12,
1
3, 17, and 18 still had solubility issues, and thus their test concentrations had to be reduced to less than
00 μM (Table 1). The maximum test concentration of 19 was limited to 286 μM due to low availability of the
8
compound, while compounds 20 and 21 showed increasing (artifactual) FP values at higher concentrations.
Still, all compounds were tested at concentrations relevant for assessing their potential inhibitory activity
(
Table 1). Within these concentrations, the FLINT was flat (Supporting Information Figure S2F) and hence did
not indicate any inner filter effects that would disturb the fluorescence-based (Cy5) FP measurements.
Compound 20 was found unstable in assay buffer, as purity was reduced to 80% compared to > 98% of the
original DMSO stock, with a main impurity corresponding in mass to a dehydration product.
Compounds 12 and 18–21 have all demonstrated activity in FP assays in literature with suggested affinity
values of 7–15 μM (Figure 1A).35, 40, 46, 62 However, in our hands only compounds 19 and 20 demonstrated some
inhibition with K values around 280 and 71 μM, respectively, which is 10–24-fold higher than published K
i
i
values of 15 and 3–7 μM, respectively,35, 62 while 12, 18, and 21 were completely inactive. To investigate this
discrepancy, a series of follow-up experiments were conducted. First, we tested if an apparent inhibition of 12
and 18–21 in the FP assay requires the use of a fluorescein-based probe, and thus we repeated the assay using
FAM-Nrf2. Again, compound 12 showed no inhibition, nor any sign of fluorescence interference as evidenced
by the flat FLINT (Figure 2C–D). Compound 18 was not measurable in this assay, as both FP and FLINT
1
5
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 16 of 77
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
dramatically increased with increasing concentrations. Compound 19 showed no activity when using the FAM-
Nrf2 probe (Figure 2C), in contrast to the apparently weak inhibition seen in the Cy5-Nrf2-based FP assay
(
Figure 2B and Table 1). Compound 20 gave an apparent IC value of ~100 μM (IC and the corresponding K
5
0
50
i
value were uncertain as inhibition did not go to baseline) and a stable FLINT throughout the concentration-
response test (Figure 2C–D). Finally, 21 was inactive up to 50 μM and could not be assessed with the FAM-
probe at higher concentrations due to increasing FP and FLINT values (Figure 2C–D). Similar data were obtained
for 12 and 18–21 when the reducing agent tris(2-carboxyethyl)phosphine (TCEP) was excluded from the assay
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
0
46
buffer (Supporting Information Figure S2G), and when mimicking the published assay conditions for 12, 18,
3
5
or 19–21 with respect to e.g. exact assay buffer, DMSO concentrations, incubation times, and fluorescent
probe (Supporting Information Figure S3).
Overall, 1–11 potently inhibited the Nrf2 peptide probe and Keap1 Kelch domain interaction, in
accordance with literature. For 12–21, no convincing activity was seen in FP, which is in opposition to reported
data.
Stabilization activity in TSA.
In the TSA, ligand binding is indicated by an increase in resistance to the temperature-dependent unfolding of
the protein, quantified as an increase in the melting temperature (T ) relative to that obtained in the absence
m
66
29, 32, 35, 62
of ligand. TSA has previously been employed to characterize the inhibitors 3 and 19–21.
Here, we
employed TSA to measure the T values induced by the compounds at varying concentrations (9 points) to
m
assess or confirm their ability to bind and stabilize the Keap1 Kelch domain (Figure 3 and Supporting
Information Figure S4).
1
6
ACS Paragon Plus Environment

Page 17 of 77
Journal of Medicinal Chemistry
1
2
3
4
5
6
7
8
9
A
B
TSA
TSA
3
2
0
0
2
1
2
3
4
5
6
7
12
13
14
15
16
17
18
19
20
21
8
1
0
1
2
9
10
11
2
5
50
75
100
125
-
-
[Ligand/M]
10
0
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
0
25
50
75
100
125
[
Ligand/M]
Figure 3. Concentration-dependent stabilization of the Keap1 Kelch domain by compounds 1–21 in the TSA. (A)
Concentration-∆T -max curves for compounds 1–11. (B) Concentration-∆T -max curves for compounds 12–21.
m
m
Average curves of ≥ 3 individual measurements are here shown (as means ± SEM).
Compounds 1–11 all show strong stabilization of the Kelch domain (∆T -max of 4–28 °C) in a
m
concentration-dependent manner (Figure 3 and Table 1). Comparing the results of the TSA with those
obtained from our FP assay, some correlation can be seen for the strongly binding compounds (low K values),
i
e.g. compounds 5, 8, and 9, exhibiting correspondingly low EC and high ∆T -max values in TSA (Table 1).
5
0
m
However, 1, 3, and 11 have fairly high ∆T -max values (9–15 °C) relative to their medium affinities (K values of
m
i
0.5–2.8 µM). This lack of correlation between FP and TSA might be because ligand-induced thermal shifts are
not only dependent on ligand affinity but potentially also specific contributions of enthalpy and entropy to
binding and on the extent to which the ligand stabilizes the denatured state of the protein.66-68
2
9,
Compound 3 has previously been reported with a ∆T -max of 1.3 °C, but was only titrated up to 38 µM,
m
3
2
while we test it up to 100 µM. Still, our data suggests a ∆T of 7.5 °C at 38 µM (Figure 3A); a deviation that
m
might be explained by differences in buffer and/or protein/dye concentrations. Unfortunately, the specific
conditions used in the literature TSA were not reported.
Compounds 12–15, 17, 18, 20, and 21 did not show any significant concentration-dependent stabilization
of the Kelch domain, while 16 and 19 showed only weak stabilization (Figure 3 and Supporting Information S4).
Testing of 12, 13, 17, and 18 was limited by solubility issues as also found in the FP assay, but no significant
1
7
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 18 of 77
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
thermal shifts were detected up to 100, 77, 99, and 49 µM, respectively (Table 1). Compounds 19–21 showed
sudden changes in fluorescence, i.e. the fluorescence promptly and significantly dropped at concentrations >
2
00 µM for compound 19 and at ≥ 100 µM for compound 20, indicating fluorescence quenching or protein
degradation, while the fluorescence increased for 21, suggesting auto-fluorescence (Supporting Information
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 S4). Compounds 19–21 have demonstrated thermal shifts in literature, but all showed negative ∆Tm
values (-2, -5, and -1 °C, respectively).35
To summarize, the TSA showed compounds 1–11 to potently stabilize Keap1 Kelch domain, while no
significant or comparable activity was seen for 12–21; this aligns with and confirms the FP results.
Binding activity in SPR assay.
To characterize direct binding between the Keap1 Kelch domain and compounds 1–21, SPR was applied. The
Kelch domain was immobilized covalently to a biosensor chip, and the compounds were injected in two-fold
dilution series. Binding curves for compounds 1–7 and 11 showed concentration dependent binding responses,
approaching or reaching saturation for the highest concentrations injected (Figure 4). Dissociation constants
(Kd) for these compounds could be derived by steady state affinity analysis, where the SPR responses at
equilibrium (R ), observed at the end of compound injection, were plotted against the injected concentrations
eq
and fitted to a Langmuir (1:1) binding isotherm model (Figure 4 and Table 1). Also, binding to the Kelch domain
was confirmed for compounds 8–10, which showed a very slow dissociation from immobilized protein,
indicative of very strong binding. Regeneration conditions to abrogate compound binding to protein between
analyte injections (so the response would return to baseline between analyte injections) could not be
established for 8–10. Thus, affinity analysis from dose-response experiments were not achievable for these.
Instead, compounds 8–10 were injected either by single concentration injection or by a gradient OneStep
injection (Figure 5),69, 70 and the SPR curves were globally fitted to a simple 1:1 kinetic interaction model from
where dissociation constants (K ) in the low nanomolar range were derived (Figure 5 and Table 1).
d
1
8
ACS Paragon Plus Environment

Page 19 of 77
Journal of Medicinal Chemistry
1
2
3
4
5
6
7
8
9
A 60
B 40
20
6
0
1
3
2
4
40
30
30
0
20
0
0
-6
-5
-6
-5
0
1
0
30
50
25
5.0x10
1.5x10
10
20
30
50
1.0x10
4.0x10
Time (s)
Conc (M)
Time (s)
Conc (M)
C 20
D 20
1
6
8
0
20
10
0
10
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
10
0
0
-6
-5
-6
-5
5
15
Time (s)
5.0x10
1.5x10
60
Time (s)
100
5.0x10
1.0x10
Conc (M)
Conc (M)
40
E 40
F 40
20
40
5
6
20
20
0
2
0
0
0
-
7
-7
0
-7
-7
4
0
100
Time (s)
160
2.0x10
8.0x10
40
100
Time (s)
160
2.0x10
8.0x10
Conc (M)
Conc (M)
G 40
40
H
120
60
0
120
60
0
7
11
20
20
0
0
-6
-5
-5
-4
20
60
Time (s)
100
10
30
Time (s)
50
4
.0x10
1.2x10
5.0x10
2.0x10
Conc (M)
Conc (M)
Figure 4. SPR sensorgrams of compounds 1–7 and 11 interacting with immobilized Keap1 Kelch domain. (A–H)
Compounds 1–7 and 11 were injected in two-fold serial dilutions (7–8 concentrations) over immobilized Kelch
domain (left panels). Sensorgrams are blank injection and reference surface subtracted. Right panels display
plots of equilibrium binding responses at the end of the analyte injections from sensorgrams in left panels
against analyte concentration. Steady state equilibrium dissociation constants (K ) from curves fitted to a 1:1
d
model are presented in Table 1.
Figure 5. SPR sensorgrams of compounds 8–10 interacting with immobilized Keap1 Kelch domain. (A)
Compound 8 was injected in one concentration (660 nM) over immobilized Kelch domain. Kinetic fit of the
sensorgram to a 1:1 Langmuir model was performed (red line), and the rate constants k and k were calculated
a
d
5
-1 -1
-3 -1
to 1.73 x 10 M s and 4.43 x 10 s , respectively. (B) Compound 9 was injected by OneStep gradient injection
1
9
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 20 of 77
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
up to 500 nM over immobilized Kelch domain. Kinetic fit of the sensorgram to a 1:1 Langmuir model was
6
-1 -1
-3 -1
performed (red line) and the rate constants k and k were calculated to 1.1 x 10 M s and 1.58 x 10 s ,
a
d
respectively. (C) Compound 10 was injected by OneStep gradient injection up to 1 µM over immobilized Kelch
domain. Kinetic fit of the sensorgram to a 1:1 Langmuir model was performed (red line), and the rate constants
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
-1 -1
-3 -1
k and k were found to be 1.16 x 10 M s and 2.26 x 10 s , respectively. Sensorgrams are blank injection
a
d
and reference surface subtracted. Dissociation constants (K ) from kinetic fits of sensorgrams are presented in
d
Table 1.
The SPR sensorgrams of compounds 12, 14, 15, 17, and 18 revealed no binding to the Kelch domain, while
some response was seen for 13, 16, and 19–21 at the tested concentrations (Supporting Information Figure
S5A–I and Table 1). However, 13 and 16 bound only very weakly to the domain; further, the shape of the
sensorgrams of 19–21 indicated non-specific binding, by either showing a disproportional increase in response
level with concentration, a response not reaching steady-state phase, or lack of binding saturation.
Compounds 3–6 and 13–15 have previously been described to show binding to Keap1 by SPR (Figure
1
A).29, 31, 34 For compound 3, the determined K of 2.4 μM (Table 1) was in similar range of the reported K
d
d
2
9
31
values of 1 μM and 0.8 μM . The SPR determined affinity (K ) of 4 was similar to the K value determined by
d
i
3
4
FP (Table 1), although lower than the K reported in literature. Compounds 5 and 6 showed K values of 20–
d
d
44 nM by SPR in earlier reports (Figure 1A),31, 34 which are similar to our FP K values, but several fold lower
i
than our SPR K values (Table 1); still, 5 and 6 are here confirmed by SPR to be potent binders. Compound 13
d
was previously found to bind to the Kelch domain with a K of 22.8 µM based on injections of three
d
4
1
concentrations in an SPR experiment not reaching binding saturation. In our hands, 13 only showed weak
binding responses when tested up to 307 µM. The reported SPR data for 14 showed very weak responses (≤ 5
RU) not approaching binding saturation level when tested up to 100 µM; thus, K in this previous study was
d
4
2
estimated to be higher than 100 µM. Here, no binding was observed for 14 to the Kelch domain, even when
tested at concentrations up to 582 µM. Compound 15 was reported to show a dose dependent SPR sensorgram
2
0
ACS Paragon Plus Environment

Page 21 of 77
1
Journal of Medicinal Chemistry
4
3
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
when tested up to 100 µM. However, the SPR sensorgram published, shows only very weak SPR responses for
43
5
0 and 100 µM injections (≤ 5 RU). We were not able to confirm any binding in our study when testing 15 up
to a concentration of 452 µM.
Compounds 7–11 have not previously been investigated by SPR. Compound 7 gave a K of 0.47 μM,
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
3
6
thereby confirming the nanomolar affinity seen in FP (Figure 1A and Table 1). Compounds 8–10 showed the
highest affinities among the tested compounds in our FP assay (K values of 1–13 nM). This was confirmed by
i
the single concentration SPR experiments (K values of 1.4–26 nM; Table 1) and is in line with the reported
d
affinity values of 1–200 nM based on FP or ITC (Figure 1A).27, 38, 39 For 11, we experienced an 18-fold
discrepancy in the FP K value and SPR K value (Table 1). This is likely explained by 11 binding 2:1 with the
i
d
33
Kelch domain, as observed by native LC-MS and X-ray crystallography, which would lead to a lower apparent
affinity in SPR measuring direct binding, but not in FP if the initial binding event leads to displacement of the
peptide probe. Noticeably, in SPR we see a ~2-fold higher Rmax for 11 compared to other active compounds
(Figure 4), which supports the suggested 2:1 stoichiometry of 11 binding to the Kelch domain.
In conclusion, compounds 1–11 showed direct binding to the Keap1 Kelch domain by SPR, thereby
confirming our own FP and TSA results as well as the literature reports. On the other hand, SPR did not reveal
any clear interaction with the Kelch domain for 12–21, as also seen by FP and TSA.
Cell activity in NQO1 induction assay.
To evaluate the downstream cellular activities of the compounds, we tested their ability to induce NQO1
enzyme activity at varying concentrations (1, 5, and 10 µM) in Hepa1c1c7 murine hepatoma cells using
7
1
sulforaphane (SFN) as a positive control. The peptide controls 1 and 2 were inactive, likely because of the
inherently limited physicochemical properties of peptides for permeating biological membranes, which also
2
5
correlates with the low cellular effect seen for other Neh2-derived peptides. The small-molecules confirmed
to be genuine Keap1-Nrf2 PPI inhibitors (3–11) in our binding assays modestly to strongly induced NQO1
specific enzyme activity in a dose-responsive manner, with exception of compounds 6 and 11 (Figure 6A).
2
1
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 22 of 77
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
Compound 6 did not show any statistically significant NQO1 induction, despite its strong binding to the Kelch
domain (e.g. 3 and 4 exhibit weaker affinities in FP and SPR, but did induce NQO1 activity). This could be due to
1
0
the lower cLogP and higher polar surface area featured by 6 relative to compounds 3 and 4, which might
decrease its passive diffusion across the cell membrane. Compound 5 also has a high polar surface area relative
to 4, but its high affinity towards the Kelch domain is likely to compensate for this, hence resulting in a cell-
active compound. The functional Nrf2 activation of 6 has not previously been reported, while 3–5 have shown
activity in ARE-inducing cell assays.22, 29, 33 For compound 11, the lack of cell activity may be explained by its
relatively weak binding activity as observed in FP, TSA, and SPR. In literature, 11 showed no activity in an ARE-
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
3
driven luciferase reporter cell-based assay. Compounds 7–10 were among the strongest binding ligands in our
FP, TSA, and SPR assays and were also potent NQO1 inducers, comparable to or exceeding SFN. These results
are in line with literature, where 7–10 have all demonstrated activity in NQO1-based or ARE-driven cell
assays.27, 36, 38, 39
2
2
ACS Paragon Plus Environment

Page 23 of 77
Journal of Medicinal Chemistry
1
2
3
4
5
6
7
8
9
A 2
NQO1 induction assay
1 M
5 M
.0
***
***
**
10 M
***
*
**
**
***
1.5
1.0
0.5
*
*
** ***
***
*
*
*
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
*
*
*
*
*
0.0
1
2
3
4
5
6
7
8
9
10
11
B
NQO1 induction assay
1
M
M
0 M
2.5
2.0
1.5
5
1
*
**
***
*
**
*
**
*
*
*
1.0
0.5
0.0
1
2
13
14
15
16
17
19
20
21
Figure 6. NQO1 fold induction at 1, 5, and 10 µM of compounds 1–11 (A) and 12–17, 19–21 (B) relative to the
positive control SFN (10 µM) shown as mean ± SEM; based on at least four individual measurements. Statistical
significance of differences relative to DMSO is indicated as follows: *p < 0.05; **p < 0.01; ***p < 0.001;
Student's two-tailed, unpaired t-test.
Compound 12 potently and concentration-dependently induced NQO1 activity (Figure 6B), which is
consistent with the high cellular potency demonstrated previously in an analogous NQO1 induction assay (a 4-
40
fold induction at 10 µM was reported). However, this result is inconsistent with our binding assays, where we
2
3
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 24 of 77
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
see no binding to the Keap1 Kelch domain or inhibition of its interaction with Nrf2 for compound 12. A growing
body of literature has revealed a multitude of regulatory switches through which the activity of Nrf2 can be
modulated independently of Keap1.14, 72 We recognize that compound 12 has been shown to be able to disrupt
the low-affinity DLG-Keap1 interaction to form an open state complex in a FRET-based assay in live cells;40
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
however, this does not necessitate direct binding to the Keap1 Kelch domain, and the existence of an open
state conformation of the Keap1-Nrf2 complex has also been associated with multiple modes of regulation.9
Thus, it remains a possibility that the confirmed potent NQO1 induction exhibited by 12 is not dependent on
binding to the Keap1 Kelch domain.
Compound 13 induced NQO1 activity to a similar level as SFN, which is consistent with literature results
41
(
concentration to give a two-fold increase in response (CD) = 1.3 µM). As for 12, this contradicts our binding
assay results, where no binding to the Kelch domain was observed in relevant concentrations; thus, again Kelch
binding-independent Nrf2 regulation must explain the cellular activity as also discussed below.
Compounds 14 and 15 did not show any statistically significant increase in NQO1 activity, while 16 only
exhibited weak induction at 10 µM; this correlates with the inactivity or only very weak activity seen in our
binding assays. The cellular activity has not previously been demonstrated for 14, but 15 and 16 have both
shown modest activities in luciferase-based cell assays (CD = 1.36 µM and 2.5 µM, respectively).43, 44 We note
that different assay conditions, including different cell lines, were used in the literature for assessing the
cellular activities of 15 and 16 compared to what was used here.
Compound 17 was inactive in our binding assays and correspondingly showed no increase in NQO1
activity. However, this contradicts the literature, since the compound has been reported to potently induce
4
5
NQO1 in Hepa1c1c7 cells (CD = 0.07 µM). We cannot readily account for this discrepancy, and unfortunately
only scarce information has been reported about the details of the previously used assay, preventing cross-
examination for potential differences and identification of an underlying cause. We note that a positive control
4
5
(
e.g. SFN) appears not to have been employed in the literature assay. We also note that compound 17 had
very limited aqueous solubility and therefore had to be assayed with 1.0% DMSO.
2
4
ACS Paragon Plus Environment

Page 25 of 77
Journal of Medicinal Chemistry
1
2
3
4
Compound 18 was tested up to 10 µM without observing any NQO1 induction relative to DMSO (data not
shown). However, due to the low solubility of 18, a relatively high DMSO concentration (1.4%) was applied to
reach 10 µM of 18 in the assay, but under these conditions SFN was not active. Also, when testing a lower
concentration of 18 (0.9 µM) at a DMSO concentration of 0.13%, where SFN was active, no activity was
observed (data not shown). Compound 18 has previously been shown to activate Nrf2 in an ARE-driven
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
46
luciferase-based cell assay with a 10-fold induction at 200 µM and without cytotoxic activity. We speculate
however, whether the literature cellular activity could in fact be artifactual, since structurally, compound 18
bears much resemblance to known classes of urea- and hydrazine-based luciferase enzyme inhibitors, which
are able to confound readings from gene reporter assays.73
Compounds 19–21 did not show any significant NQO1 induction, correlating with the unobserved or
unconvincing activities seen in our binding assays. Cellular activities of 19 and 21 have not hitherto been
3
5
demonstrated, while compound 20 has previously been reported to elevate mRNA levels of NQO1; however,
the induction was weak and it was not shown whether it was statistically significant relative to the blank
control.
In conclusion, our NQO1 induction assay confirms the cellular potencies of the compounds also
established to bind Keap1 Kelch domain, i.e. compounds 3–5 and 7–10, while the weak or non-binding
compounds 14–21 correspondingly showed no or only weak increase in NQO1 activity. Outliers from this
correlation include the peptides 1 and 2 and compounds 6 and 11, where improper physicochemical properties
might be a possible reason for their cellular inactivity. Other aberrations include compounds 12 and 13 that
show modest to strong NQO1 induction despite not showing any activity in our binding assays, wherefore we
suggest these cellular effects are due to mechanisms independent from binding to the Keap1 Kelch domain.
Counter assays for covalent reactivity, redox cycling, and aggregations.
We next tested the 21 compounds for common liabilities, which can cause false-positive signals in biochemical
5
3
assays. First, we evaluated the compounds for covalent reactivity since several of these contain electrophilic
2
5
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 26 of 77
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
functional groups, which could potentially react with protein amino acid residues, especially cysteines. In
separate experiments, we incubated the compounds at 2–800 µM under our binding assay conditions with 2
µM
Keap1 Kelch domain and with 1 mM glutathione (GSH) as a cysteine residue surrogate, respectively.74, 75
Potential covalent binding was assessed by LC-MS or UPLC-MS. The known covalent modifier, SFN, was
employed as a positive control in both experiments, which as expected showed clear adduct formation with
both the Keap1 Kelch domain at ≥ 20 µM and with GSH at 500 µM (Supporting Information Figure S6 and S7A).
However, SFN was not active in the FP assay (Supporting Information Figure S8B), demonstrating that this
assay is generally not sensitive to covalent modifiers. Interestingly, incubation of the Keap1 Kelch domain with
compounds 11, 13, 19, and 20 all resulted in significant changes of the mass spectrum of the protein
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
(Supporting Information Figure S6A). For 11, 13, and 20 distinct mass adducts were formed at 800, 200 and 20
µM, respectively, and these could be attributed to the nucleophilic displacement of labile thiolate moieties in
their structures by protein cysteines (Figure 7A). For compound 19, strong perturbation of the mass spectrum
was observed at 800 µM with the apparent disappearance of mass signals, indicating protein denaturation or
degradation. This phenomenon was also observed at 200 µM of 13 (Supporting Information Figure S6A) and
2
0 (not shown). For compound 20, covalent reactivity was further observed in the incubation experiment with
GSH with detection of a clear adduct formation (Supporting Information Figure S7B), while this was not seen
for 11, 13, and 19. This discrepancy might be explained by the known effect of the protein microenvironment,
including solvent exposure and local pH, on the reactivity of cysteine thiols.76, 77
2
6
ACS Paragon Plus Environment

Page 27 of 77
Journal of Medicinal Chemistry
1
2
A
Blank Keap1
Keap1 + 11
Keap1
SH
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
Keap1
Keap1
O
S
N
O
S
N
O
N
 H
Keap1
O
H
CF3
HN
HN
+
SH
S
S
N
Keap1-11
Keap1-13
H
O
O
O
O
CF3
SH
11
Keap1-11
~33130 Da)
(
Keap1
SH
Keap1
S
NO2
S
NO2
+
N

H
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
N
N
Keap1 + 13
Keap1 + 20
N
O
N N
N
O
O
N
Keap1
Keap1
13
Keap1-13
~33043 Da)
(
O
S
O
O
S
S
HN
HN
SH
Keap1-20
 H
+
N
N
Keap1
S
NH
HS
OH
N
OH
N
Keap1
NH
20
Keap1-20
(~33210 Da)
Deconvoluted mass (Da)
4_wo Triton/Tween20
4_w 0.005% Tween20
B
Redox activity
C
FP (Inhibition)
Cy5-Nrf2, 4% DMSO
8% DMSO
1
1
0
0
.5
60
40
20
0
4
4
4
_w 0.01% Triton
_w 0.02% Triton
_w 0.05% Triton
3-Methyltoxoflavin
.0
.5
.0
H2O2
13
6
_wo Triton/Tween20
6_w 0.005% Tween20
6
_w 0.01% Triton
_w 0.02% Triton
6
-7
-6
-5
-4
-3
-2
-9
-8
-7
-6
-5
-4
6_w 0.05% Triton
Log [Inhibitor/M]
Log [Inhibitor/M]
Figure 7. (A) Covalent reactivity of compounds 11, 13, and 20 towards the Keap1 Kelch domain as assessed by
LC-MS. All compounds were incubated with 2 µM protein at various concentrations/stoichiometries (1:1, 10:1,
1
00:1, and 400:1). Shown are deconvoluted mass spectra for the Kelch domain incubated with 11 at 800 uM
(no adduct formation seen at lower concentrations), with 13 at 200 uM, and with 20 at 20 uM (no adduct
formation seen at lower concentrations, protein signal disappearance seen at higher concentrations). The raw
charge envelope spectra can be found in the Supporting Information Figure S6A. The proposed reactions
occurring are shown to the right. (B) Redox activity of 13 as measured by H O generation in presence of 1 mM
2
2
TCEP and in comparison with known redox-active 3-methyltoxoflavin and externally added H O .
2
2
Representative curves of two individual measurements are shown as means of duplicates ± SEM. (C) FP
concentration-response curves for 4 and 6 without (wo) detergent or with (w) various concentrations of
Tween20/Triton X-100. Representative curves of two individual measurements (wo Triton/Tween20, w 0.005%
2
7
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 28 of 77
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
Tween20, and w 0.01% Triton X-100) or one measurement (w 0.02 and 0.05% % Triton X-100) are shown as
means of duplicates ± SEM.
Redox cycling compounds (RCCs) generate H O in the presence of strong reducing agents like TCEP and
2
2
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
DTT, which can affect proteins and lead to false readouts.53, 78-81 As we have TCEP in our standard assay buffer,
we tested if 1–21 produced H O using an assay based on oxidation of phenol red by H O catalyzed by
2
2
2
2
8
1
horseradish peroxidase. Interestingly, compound 13 gave a clear TCEP-dependent response starting at 77 µM,
with absorbance values about 2–3-fold lower than corresponding concentrations of pure H O added as a
2
2
positive control. Compared to the control compound 3-methyltoxoflavin, 13 was ~20-fold less potent (Figure
B). The remaining compounds did not give any signal (Supporting Information Figure S8A). 3-
Methyltoxoflavin and H O showed no activity in the FP assay (Supporting Information Figure S8B); thus this
7
2
2
assay is not sensitive to RCCs.
Aggregate-forming compounds are typically much more potent without presence of detergent and can
thereby be identified by performing the biochemical assay with and without 0.01% Triton X-100.82, 83 Here, we
observed no difference for 1–3, 5, 7–11, and 19–20 in the FP assay with 0.01% or without Triton X-100
(
Supporting Information Figure S8C) or when compared to our usual assay conditions using 0.005% Tween20
Fig. 2). Compound 12-18 and 21 were inactive without Triton X-100 and were therefore not tested further.
(
Benzyl benzoate and clotrimazole were inactive in our FP assay both with and without detergent (Supporting
84
Information Figure S8B), indicating that our FP assay is robust to typical promiscuous aggregators. Compound
4
and 6, however, were both 2–4-fold more potent without detergent than with (0.01/0.02/0.05% Triton X-100
and 0.005% Tween20 gave similar results) (Figure 7C).
DISCUSSION AND CONCLUSIONS
Nrf2 activation through disruption of the Keap1-Nrf2 PPI has been recognized as a promising strategy to
combat oxidative stress in peripheral as well as CNS diseases, such as stroke and neurodegenerative disorders.
2
8
ACS Paragon Plus Environment

Page 29 of 77
1
Journal of Medicinal Chemistry
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
Although our artillery of direct small-molecule Keap1-Nrf2 PPI inhibitors has grown significantly throughout the
years, many of these compounds feature physicochemical properties inappropriate for passing the BBB.
However, further optimization might address this problem. Other reported inhibitors contain potentially
problematic substructural motifs or have only been scarcely characterized. In order to rid the literature of
problematic compounds, investigate the reproducibility of their synthesis and reported activities, and allow
selection of the most promising starting points for lead optimization we have here performed an extensive
comparative assessment study of all reported non-covalent small-molecule Keap1-Nrf2 PPI inhibitor classes
that we are aware of (Figure 1A).
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
Initially, compounds 3–12, 14, 16, and 17 were synthesized, generally following reported procedures.
Adjustments and attention to procedural details were necessary to obtain 3, 4–8, and 10 in reasonable yields,
and new procedures were developed for 11 and 17.
To evaluate the binding efficiency and cellular potency of the obtained compounds 3–21, we tested them
together with two positive peptide controls 1 and 2 in our triad of orthogonal binding assays, FP, TSA, and SPR,
and further in a cellular NQO1 induction assay. To our surprise, we were not able to demonstrate convincing
binding activity of half of the compounds, i.e. 12–21. By careful use of control experiments, we revealed
compounds 18, 19, and 21 to exert fluorescence interfering activities, which for 19 and 21 correlates with the
inclusion of PAINS moieties in their structure.10, 52 Although we did detect weak inhibition for compound 19 and
20 in our FP assay, the binding efficiency was unsteady across different assay conditions. Importantly, we
observed stability issues of 20 during QC by LC-MS subsequent to assaying; furthermore, the compound
formed a covalent adduct with GSH and, as 19, severely distorted the mass spectrum of Keap1 indicating
various forms of reactivity or deteriorations of the protein. We recognize that SAR has been conducted around
6
2
the scaffold of 20 to produce modestly potent compounds. However, our observations, and the fact that
degradation under assay conditions, protein-adduct formations, and even protein denaturation have been
reported in literature for similar 4-hydroxyarylsulfonamides,75 raise serious concerns for the utility of this
particular chemotype in drug discovery.
2
9
ACS Paragon Plus Environment