Targeting Drug Resistance in EGFR with Covalent Inhibitors: A Structure-Based Design Approach

Article abstract of DOI:10.1021/acs.jmedchem.5b01082

Receptor tyrosine kinases represent one of the prime targets in cancer therapy, as the dysregulation of these elementary transducers of extracellular signals, like the epidermal growth factor receptor (EGFR), contributes to the onset of cancer, such as non-small cell lung cancer (NSCLC). Strong efforts were directed to the development of irreversible inhibitors and led to compound CO-1686, which takes advantage of increased residence time at EGFR by alkylating Cys797 and thereby preventing toxic effects. Here, we present a structure-based approach, rationalized by subsequent computational analysis of conformational ligand ensembles in solution, to design novel and irreversible EGFR inhibitors based on a screening hit that was identified in a phenotype screen of 80 NSCLC cell lines against approximately 1500 compounds. Using protein X-ray crystallography, we deciphered the binding mode in engineered cSrc (T338M/S345C), a validated model system for EGFR-T790M, which constituted the basis for further rational design approaches. Chemical synthesis led to further compound collections that revealed increased biochemical potency and, in part, selectivity toward mutated (L858R and L858R/T790M) vs nonmutated EGFR. Further cell-based and kinetic studies were performed to substantiate our initial findings. Utilizing proteolytic digestion and nano-LC-MS/MS analysis, we confirmed the alkylation of Cys797.

Full text of DOI:10.1021/acs.jmedchem.5b01082

Subscriber access provided by UNSW Library
Article
Targeting Drug Resistance in EGFR with Covalent
Inhibitors - a Structure-Based Design Approach
Julian Engel, Andre Richters, Matthäus Getlik, Stefano Tomassi, Marina Keul, Martin Termathe, Jonas
Lategahn, Christian Becker, Svenja Mayer-Wrangowski, Christian Grütter, Niklas Uhlenbrock, Jasmin
Krüll, Niklas Schaumann, Simone Eppmann, Patrick Kibies, Franziska Hoffgaard, Jochen Heil, sascha
Menninger, Sandra Ortiz-Cuaran, Johannes Heuckmann, Verena Tinnefeld, René Peiman Zahedi,
Martin L. Sos, Carsten Schultz-Fademrecht, Roman K. Thomas, Stefan M. Kast, and Daniel Rauh
J. Med. Chem., Just Accepted Manuscript • DOI: 10.1021/acs.jmedchem.5b01082 • Publication Date (Web): 14 Aug 2015
Downloaded from http://pubs.acs.org on August 16, 2015
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 69
Journal of Medicinal Chemistry
1
2
3
4
5
6
7
8
9
Thomas, Roman; University of Cologne, Department of Translational
Genomics
Kast, Stefan; Technische Universität Dortmund, Fakultät für Chemie und
Chemische Biologie
Rauh, Daniel; Technische Universität Dortmund, Chemie und Chemische
Biologie; Technische Universität Dortmund, Fakultät für Chemie und
Chemische Biologie
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
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 2 of 69
1
2
3
4
5
6
7
8
9
Targeting Drug Resistance in EGFR with Covalent
Inhibitors – a StructureꢀBased Design Approach
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
Julian Engel1, André Richters1, Matthäus Getlik2, Stefano Tomassi1, Marina Keul1, Martin
Termathe2, Jonas Lategahn1, Christian Becker1, Svenja MayerꢀWrangowski1, Christian Grütter1,
Niklas Uhlenbrock1, Jasmin Krüll1, Niklas Schaumann1, Simone Eppmann1, Patrick Kibies1,
Franziska Hoffgaard1, Jochen Heil1, Sascha Menninger3, Sandra OrtizꢀCuaran4, Johannes
Heuckmann4, Verena Tinnefeld7, René P. Zahedi7, Martin L. Sos4,6, Carsten Schultzꢀ
Fademrecht3, Roman K. Thomas4,5, Stefan M. Kast1,*, Daniel Rauh1,2,*
1 Faculty of Chemistry and Chemical Biology, TU Dortmund University, OttoꢀHahnꢀStraße 6, Dꢀ
44227 Dortmund, Germany
2 Chemical Genomics Centre of the MaxꢀPlanck Society, OttoꢀHahnꢀStraße 15, Dꢀ44227
Dortmund, Germany
3 Lead Discovery Center GmbH, OttoꢀHahnꢀStraße 15, Dꢀ44227 Dortmund, Germany
4 Department of Translational Genomics, University of Cologne, Weyertal 115b, Dꢀ50931,
Cologne, Germany
5 Department of Pathology, University of Cologne, JosephꢀStelzmann Straße 9, Dꢀ50931,
Cologne, Germany
6 Molecular Pathology, University Hospital of Cologne, Kerpenerstraße 62, Dꢀ50937, Cologne,
Germany
7 LeibnizꢀInstitut für Analytische Wissenschaften ꢀ ISAS ꢀ e.V., Dortmund, Germany
ACS Paragon Plus Environment
1

Page 3 of 69
Journal of Medicinal Chemistry
1
2
3
4
Abstract
5
6
7
8
9
Receptor tyrosine kinases represent one of the prime targets in cancer therapy as the
dysregulation of these elementary transducers of extracellular signals, like the epidermal growth
factor receptor (EGFR), contributes to the onset of cancer such as nonꢀsmall cell lung cancer
(NSCLC). Strong efforts were directed to the development of irreversible inhibitors and led to
compound COꢀ1686, which takes advantage of increased residence time to EGFR by alkylating
Cys797 and thereby preventing toxic effects. Here, we present a structureꢀbased approach,
rationalized by subsequent computational analysis of conformational ligand ensembles in
solution, to design novel and irreversible EGFR inhibitors based on a screening hit that was
identified in a phenotype screen of 80 NSCLC cell lines against about 1500 compounds. Using
protein Xꢀray crystallography, we deciphered the binding mode in engineered cSrc
(T338M/S345C), a validated model system for EGFRꢀT790M, which constituted the basis for
further rational design approaches. Chemical synthesis led to further compound collections
revealing increased biochemical potency and, in part, selectivity towards mutated (L858R and
L858R/T790M) vs. nonꢀmutated EGFR. Further cellꢀbased and kinetic studies were performed to
substantiate our initial findings. Utilizing proteolytic digestion and nanoꢀLCꢀMS/MS analysis,
we confirmed the alkylation of Cys797.
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
ACS Paragon Plus Environment
2

Journal of Medicinal Chemistry
Page 4 of 69
1
2
3
4
INTRODUCTION
5
6
7
8
The epidermal growth factor receptor (EGFR), a receptor tyrosine kinase (RTK), is a key
mediator in cellular signaling related to cell growth, proliferation, survival, and migration.1ꢀ3 Its
aberrant activity plays a pivotal role in the development and growth of tumor cells and is
associated with the onset and progression of nonꢀsmall cell lung cancer (NSCLC).4, 5 Mutations
in the catalytic domain of EGFR, accounting for increased kinase activity and ligandꢀ
independency, have been discovered as oncogenic drivers in NSCLC. The most prevalent
activating mutations are represented by the L858R, a single point substitution in exon 21, and the
exon 19 deletion (delE746ꢀA750). Patients harboring these specific activating mutations show a
significant clinical response (50ꢀ80 %) to firstꢀgeneration reversible EGFR TypeꢀI inhibitors
gefitinib6 and erlotinib.7ꢀ12 Therefore, EGFR activating mutations serve as predictive markers for
the treatment of NSCLC with EGFR tyrosine kinase inhibitors (TKIs), thus demonstrating the
potential of targeted cancer therapy. However, patients responding to these drugs develop
secondary drug resistance mutations and suffer from a dramatic relapse within months. In 50 %
of these cases, the acquired resistance results from a mutation of the gatekeeper residue
(T790M). This leads to increased affinity for ATP when compared to other oncogenic primary
mutant variants.13ꢀ16 Moreover, the sterically more demanding methionine provokes a clash with
4ꢀaminoquinazolineꢀbased drugs erlotinib and gefitinib and thereby induces a slightly different
binding geometry that results in a complete loss of inhibitory activity.17ꢀ19 Second generation
EGFR inhibitors, including neratinib20, dacomitinib21, and afatinib22 (Figure 1A) contain
electrophilic Michaelꢀacceptor systems to target a rare cysteine (Cys797) at the lip of the ATP
binding cleft of EGFR. These new inhibitors were thought to overcome T790M drug resistance
due to the covalent modification of Cys797 and the resulting increased target residence time.23ꢀ25
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
ACS Paragon Plus Environment
3

Page 5 of 69
Journal of Medicinal Chemistry
1
2
3
4
5
6
7
8
9
Despite promising in vitro results, these inhibitors exhibit insufficient efficacy in patients at
clinically achievable concentrations. Thus, onꢀtarget toxicity represents the doseꢀlimiting factor.
In addition to the oncogenic mutant variants, the wild type form of EGFR is likewise inhibited
and severe side effects during therapy are observed (e.g., skin rash and diarrhea).26ꢀ28
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
For this reason, thirdꢀgeneration EGFR inhibitors including WZ4002,29 COꢀ1686
(Rociletinib),30, 31 and AZD929132, 33 (Figure 1A), have been developed and are based on an
amino pyrimidine scaffold. While WZ4002 has not progressed into human clinical trials, COꢀ
1686 (Rociletinib) and AZD9291 are currently at the leading edge of NSCLC treatment
exhibiting beneficial effects in terms of residence time, toxicity profile, and further ADME
parameters.34 Notably, these compounds do not interfere with larger gatekeeper residues such as
of the methionine in the T790M variant and, therefore, retain their inhibitory efficacy towards
these clinically relevant mutants. The work of Walter and colleagues showed that covalent
modification of EGFR by small molecule inhibitors is feasible for overcoming gatekeeper
mutations in cancer treatment30. However, only a few compounds possessing the necessary
structural features are under investigation in clinical trials, and further scaffolds are needed for
directed treatments.
Recently, we performed a phenotype screen of 80 NSCLC cell lines against 1500 small
molecule inhibitors and identified quinazoline 1a (Figure 1B) as having inhibitory impact on the
drugꢀresistant H1975 cell line (L858R/T790M).35 However, we also observed significant
nonspecific inhibitory effects, most likely caused by offꢀtarget activity. Nevertheless, the
identified candidate is comprised of a classical 4ꢀaminoquinazoline, but with a unique
substitution pattern. This kinase inhibitor is decorated with a sterically demanding phenyl moiety
at the 2ꢀposition that prevents the adoption of the typical quinazoline binding mode and which
ACS Paragon Plus Environment
4

Journal of Medicinal Chemistry
Page 6 of 69
1
2
3
4
5
6
7
8
9
utilizes the 4ꢀaminopyrazole moiety as an alternative hinge binding element (Figure 1B).
Moreover, a Michael acceptor system is incorporated in the 7ꢀposition of 1a, which may provide
the ability to target covalently Cys797 in EGFR. However, Xꢀray crystallography studies
utilizing engineered cSrc (T338M/S345C) as a validated model system for EGFR (T790M) did
not show evidence for the desired covalent modulation of Cys345 which is homologous to
Cys797 in EGFR (Figure 1C).18, 36 We used engineered cSrc (T338M/S345C) as initial model
system as it provides structural insights. However, crystal structures of EGFRꢀT790M in
complex with irreversible ligands have already been published and likewise guided our design
studies. Here, we present the followꢀup chemical optimization of potent and irreversible EGFRꢀ
inhibitors based on the screening hit from the phenotype screen of 80 NSCLC cell lines.35
Structureꢀbased design studies guided us to scaffolds including quinazolines and pyrimidines
both targeting the drug resistant T790M as well as the mutant variants of EGFR that harbor
activating mutations. Organic synthesis led to a focused inhibitor collection and subsequent
characterization using activityꢀbased assay systems elicited potent inhibition profiles. Using Xꢀ
ray crystallography, we gained valuable insights into the mode of action of the new series of
inhibitors and its structureꢀactivity relationships (SARs) that led to the design of further
derivatives with improved inhibitory activity and, accordingly, better mutant selectivity. Since
restricted conformational freedom of ligands in free solution is an important beneficial factor
governing binding affinity, we also characterized the population of relevant conformational
basins by computational means in order to rationalize the experimental observations. This work
demonstrates the strength of the interplay of medicinal chemistry methods, preparative organic
chemistry, biochemistry, structure biology, and computational methods in rational drug design
approaches.
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
ACS Paragon Plus Environment
5

Page 7 of 69
Journal of Medicinal Chemistry
1
2
3
4
RESULTS
5
6
7
8
The relevance of oncogenic EGFR for the treatment of NSCLC and the ongoing interest in the
role of acquired drug resistant mutations in targeted cancer therapy emphasize the need for
strategies to identify and improve new kinase inhibitors targeting gatekeeper mutant forms of
oncogenes, such as EGFRꢀT790M. Recently, these efforts were focused on the development of
irreversible inhibitors, which covalently modify Cys797 at the lip of the ATPꢀbinding cleft.20ꢀ22,
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
30, 32
Modulators that incorporate thiolꢀreactive groups are beneficial as they increase the target
residence time at the protein of interest upon covalent binding. Therefore, the offꢀrate of these
inhibitors is directly associated with the cellular turnꢀover rate of the respective protein.23, 37, 38
Rational Design of EGFR Inhibitors Based on Screening Hit Analysis.
Intrigued by the substitution pattern of 1a, we recently solved the crystal structure of 1a in
complex with mutated cSrc to gain deeper structural insights into the binding mode and to
constitute a structural basis for further design and optimization cycles (Figure 1C). Here, we used
the previously engineered and validated cSrcꢀkinase (T338M/S345C) as a model system for
drugꢀresistant EGFRꢀT790M that has been proven to be appropriate to provide primarily
structural information.18, 36
Notably, the coꢀcrystal structure of 1a and cSrc revealed a novel but reversible binding mode
in which the 4ꢀaminopyrazole moiety addresses the hinge region by three direct hydrogen bonds
to the peptide backbone residues of Glu339 and Met341. Furthermore, the hinge binding element
apparently does not affect the gatekeeper residue (Met338). Moreover, the Michaelꢀacceptor
system in the 7ꢀposition of the quinazoline is turned outwards from the ATP binding cleft,
whereas the 2ꢀphenyl moiety accesses the binding pocket preventing the formation of a covalent
ACS Paragon Plus Environment
6

Journal of Medicinal Chemistry
Page 8 of 69
1
2
3
4
5
6
7
8
9
bond to Cys345. Therefore, the observed inhibitory effect is apparently not caused by covalent
modification of EGFR, which is consistent with an analog inhibitory potency of the reversible
counterpart 1b as compared to 1a (Table 1). Both 1a and 1b did not show any significant kinase
inhibition of EGFRꢀWT and illustrated comparable inhibitory effects on the activating mutant of
EGFR (1.9 µM vs. 2.1 µM for 1a and 1b, respectively) and on the drugꢀresistant mutant variant
(2.2 µM and 2.2 µM), strengthening the conclusion that the Michaelꢀacceptor adopts a
nonreactive binding geometry, in which it does not impact the inhibitory efficacy of 1a as
compared to 1b. This remarkable observation, in turn, allows a structureꢀguided development of
inhibitors with respect to the improved inhibitory effects evoked by covalent modification of
Cys797. Based on this Xꢀray structure, we generated superimposed models to investigate further
the options for the covalent bond formation and proposed various scaffolds and substitution
patterns capable of covalent modification (Figure 2). The phenyl moiety in the 2ꢀposition of 1a
represented a convenient attachment point for incorporating a Michael acceptor (e.g.,
acrylamide) (Figure 2A). We also proposed a pyrimidine as an alternative scaffold. This variant
retains the aminopyrazole moiety as the hinge binding element and may increase flexibility due
to the reduced core structure (pyrimidine vs. quinazoline). We suggested two different linkages
of the Michael acceptor system to the pyrimidine core to investigate different conformations and
to evaluate the preferred geometry and orientation of the electrophile. Therefore, the electrophile
was attached directly (Figure 2B) and by a linked phenylether (Figure 2C, D) to the pyrimidine
core. Moreover, we paid particular attention to the optimization of the hinge binding element
with the aim to enhance the interdependency between the methionine gatekeeper and the ligand
by expanding the hydrophobic character of the 4ꢀsubstituent (e.g., 1ꢀmethylpyrazole in 5b,
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
ACS Paragon Plus Environment
7

Page 9 of 69
Journal of Medicinal Chemistry
1
2
3
4
indazole in 6a). Recently, this interaction was shown to have a dramatic effect on inhibitory
5
6
activity and mutantꢀselectivity in case of various EGFR inhibitors.39, 40
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
Synthesis of a Focused Small Molecule Library.
We set out to synthesize a focused library of pyrimidineꢀbased inhibitors with diverse
substitution patterns. Our efforts were directed towards the optimal orientation of the Michaelꢀ
acceptor system with respect to the alkylation of Cys797 in EGFR. Moreover, we focused on the
derivatization of the hinge binding element to enhance the hydrophobic interaction with the
gatekeeper residue while simultaneously retaining the crucial direct hinge interactions.
Therefore, seven different synthetic routes were established to produce pyrimidineꢀbased
compounds 3a-b, 4, 5l-m and 5a-k, 5n-o, 5p-q, 6a-e, 7. In case of 3a-b (Scheme 1), we started
with a nucleophilic aromatic substitution using 2ꢀphenyl substituted pyrimidine 8 and 5ꢀ
aminopyrazole, resulting in intermediate 9. Subsequent decoration of the pyrazole with the Bocꢀ
protection group led to 10, which was then converted into 3a and its reversible counterpart 3b by
amide formation using acryloyl or propionyl chloride and subsequent deprotection with TFA.
To retain the flexibility of the pyrimidine based inhibitors but to vary the type of substituent at
the 2ꢀ and 4ꢀpositions, we synthesized several similar derivatives by the synthetic route shown in
Scheme 2. Driven by our modeling studies (Figure 2), we proposed different attachment points
for incorporating a Michaelꢀacceptor moiety to scrutinize and optimize the geometry and
orientation towards Cys797. Therefore, we incorporated an electrophile utilizing a linker through
a carbonꢀcarbon bonding (4) and we likewise introduced a heteroatom linkage to gain increased
flexibility that allows for altered and potentially beneficial orientation towards Cys797 (5l). Also,
we proposed further molecules incorporating a proton at the pyrimidine 6ꢀposition to gain further
ACS Paragon Plus Environment
8

Journal of Medicinal Chemistry
Page 10 of 69
1
2
3
4
5
6
7
8
9
information about the significance of a solubility group on this position. 2,4ꢀDichloropyrimidine
was preferentially decorated with 3ꢀmethylꢀ1Hꢀpyrazolꢀ5ꢀamine at the 4ꢀposition to afford key
intermediate 12. Subsequently, the pyrazole moiety was protected using dihydropyran. Then, (3ꢀ
nitrophenyl)boronic acid was used to introduce the 3ꢀnitrophenyl moiety into the pyrimidine 2ꢀ
position under Suzukiꢀconditions resulting in intermediate 14a which was reduced to the
corresponding amine 14b. The acrylamide Michael acceptor system was introduced by amide
formation using acryloyl chloride. THPꢀdeprotection with TFA led to compound 4. Starting from
key intermediate 12, we likewise generated compounds with a bridging ether at the 2ꢀposition to
increase the conformational flexibility of the Michael acceptorꢀdecorated moiety. Nucleophilic
aromatic substitution of 12 utilizing 3ꢀnitrophenol gave intermediate 15a, and subsequent
reduction with Pd/C and ammonium formate led to the amino equivalent 15b. Reaction of amine
15b with various acid chlorides yielded compounds 5l-m.
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 a third approach (Scheme 3), we applied a generic route to build up 2,4,6ꢀsubstituted
pyrimidine compounds 5a-k, 5n-q, and 6a-d. Initially, 4,6ꢀdichloroꢀ2ꢀ(methylthio)pyrimidine
was oxidized to the corresponding sulfone 17 using mCPBA to generate an appropriate leaving
group at the pyrimidine 2ꢀposition. Consecutive nucleophilic aromatic substitution with the
orthoꢀ, metaꢀ or paraꢀsubstituted nitrophenolꢀderivatives led to intermediates 18a-c. Following
substitution of one aromatic chlorine with either aminoꢀpyrazoles or aminoꢀindazoles resulted in
intermediates 19a-c and 20-22, which were then derivatized with 1ꢀmethylpiperazine at the
pyrimidine 6ꢀposition, providing 23a-c and 24-26. The decoration of the pyrazoles or indazoles
with the Bocꢀprotection group led to 27b-c and 28-30. Subsequent reduction of the aromatic
nitroꢀgroup was performed by using Pd/C and ammonium formate and resulted in the
corresponding amino derivatives 31b-c and 32-34. Ultimately, these amines were decorated with
ACS Paragon Plus Environment
9

Page 11 of 69
Journal of Medicinal Chemistry
1
2
3
4
5
6
7
8
9
Michael acceptors using acid chlorides or carboxylic acids, under standard amide coupling
conditions (EDC/HOBt), respectively. The equivalent reversible counterparts were produced
accordingly utilizing propionyl chloride. Final Bocꢀdeprotection was conducted using TFA and
led to compounds 5a-k, 5n-q, and 6a-d (Table 1).
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
Synthetic schemes and preparations of quinazolineꢀbased compounds 1a-b and 2a-f have been
recently described elsewhere.35 Synthetic schemes of compounds 6e and 7 as well as detailed
synthetic procedures are shown in the supporting Information (Scheme S1 and S2, respectively).
Activity-Based Characterization of Quinazoline and Pyrimidine-Based Scaffolds on
EGFR inhibition.
We characterized the focused compound collection against various types of EGFR including
wild type and mutant forms L858R and L858R/T790M (Table 1) using an activityꢀbased assay
that quantifies phosphorylation of an artificial substrate by a particular kinase through
homogeneous timeꢀresolved FRET measurements (see Experimental Section). Since 1a was
decorated with an acrylamide warhead, we tested the reversible counterpart and observed equal
potency on EGFRꢀL858R and L858R/T790M (IC50 ~ 2.0 µM) while 1a displayed a reduced
effect on the wild type form of EGFR (IC50 >10 µM). However, due to these results and in
concordance with the observed reversible binding mode (Figure 1C), we generated further
quinazolineꢀbased compounds (2a-f) with the Michael acceptors attached in different positions
(ortho, meta, para) to increase the probability of alkylating Cys797. Compounds 2b and 2e
demonstrated significant inhibitory effects in the investigated EGFR variants. We observed IC50
values of 1.1 µM (wt), 0.5 µM (L858R) and 0.1 µM (L858R/T790M) for compound 2b, while
the reversible counterpart 2e was 4 times less potent (3.7 µM, 1.9 µM, and 0.4 µM, respectively)
ACS Paragon Plus Environment
10

Journal of Medicinal Chemistry
Page 12 of 69
1
2
3
4
5
6
7
8
9
indicating that in contrast to 2b, no covalent bond was formed with 2e. Altering the attachment
point of the acrylamide/propionamide motif to the ortho and para sites, (2a, 2d and 2c, 2f) did
not result in notable inhibitory effects, except for 2a with moderate inhibitory effects on L858R
(2.8 µM) and L858R/T790M (1.1 µM).
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
When we decreased the spatial arrangement by utilizing a pyrimidine core instead of a
quinazoline 3a, we achieved similar IC50 values (0.9 µM, 0.8 µM, 0.3 µM), although the Michael
acceptor was not attached to the 2ꢀphenyl moiety as in the case of 2a-f but instead at the
pyrimidine 6ꢀposition. Both attachment points were likewise eligible to elicit potent inhibition on
EGFR. Furthermore, the reversible counterpart 3b did not significantly inhibit EGFR, supporting
our hypothesis of covalent bond formation between Cys797 and 3a.
Based on modeling studies (Figure 2), we attached the Michael acceptor to the 2ꢀphenyl
moiety (4) to increase inhibitory activity. We incorporated an additional methyl group at the
pyrazole 5ꢀposition, which we proposed would form hydrophobic interactions with the
methionine gatekeeper residue in EGFRꢀL858R/T790M, and increase the compound’s selectivity
towards this variant. Several EGFR inhibitors have been equipped with either halogen or methyl
substituents directed towards the methionine gatekeeper residue.29, 30 However, we did not
observe any beneficial effect for compound 4 with respect to the inhibitory potency towards the
investigated EGFR variants (IC50 >10 µM (wt), 2.5 µM (L858R) and 2.3 µM (L858R/T790M).
Further approaches were undertaken to position the Michael acceptor in proximity to Cys797 by
connecting the phenylacrylamide to a flexible ether linkage. We speculated that this would
translate into an improved orientation towards the reactive cysteine and favor subsequent
covalent bond formation. While the wild type form was not inhibited (IC50 >10 µM), we
observed moderate inhibition of the L858R (1.8 µM) as well as the L858R/T790M mutant
ACS Paragon Plus Environment
11

Page 13 of 69
Journal of Medicinal Chemistry
1
2
3
4
5
6
(0.9 µM) by 5l. The reversible counterpart 5m certainly did not elicit substantial inhibitory
activity towards any of the EGFR variants.
7
8
9
These results prompted us to introduce a solubilizing group (1ꢀmethylpiperazine) to the 6ꢀ
position of the pyrimidine to restrict the structural flexibility with respect to hinge binding and
optimally orient the 2ꢀphenyl moiety to preserve its conformational flexibility to reach Cys797
for covalent modification. Compounds 5a-c were generated to investigate the influence of orthoꢀ,
metaꢀ and paraꢀsubstituted acrylamides on the inhibition and also on the formation of a covalent
bond by the thiolꢀreactive warhead. For 5a (ortho), only moderate inhibition was observed for
wild type (IC50 = 0.9 µM), L858R (IC50 = 0.6 µM) and L858R/T790M (IC50 = 2.1 µM). Equal
behavior was demonstrated for the corresponding paraꢀsubstituted compound 5c (3.1 µM, wt;
1.8 µM, L858R; 2.4 µM, L858R/T790M). Notably, metaꢀsubstitution as in 5b led to remarkable
inhibition of EGFRꢀL858R (IC50 = 0.03 µM) whereas the double mutant form was less addressed
(IC50 = 0.5 µM). Interestingly, 5b did not show any significant inhibition of the wild type. With
respect to the reversible counterpart 5e, which is equipped with a propionyl moiety that cannot
undergo a Michaelꢀaddition to form a covalent bond to Cys797, we observed a dramatic loss in
inhibitory activity against both activating and drugꢀresistant mutant variants of EGFR (60ꢀfold
and about 20ꢀfold, respectively) as compared to its covalent counterpart 5b, thus indicating that
5b may interact with EGFR in a covalent manner. This observation further strengthened the
conclusion that covalently targeting EGFRꢀT790M is a crucial feature to achieve sufficient
efficacy. The corresponding 5n solely possesses a proton instead of the methyl substitution at the
pyrazole 5ꢀposition and therefore resulted in moderate inhibitory activity on the mutant variants
(2.0 µM and 1.9 µM for L858R and L858R/T790M, respectively) whereas no inhibitory activity
was observed against the wild type, supporting the concept that favorable hydrophobic
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
ACS Paragon Plus Environment
12

Journal of Medicinal Chemistry
Page 14 of 69
1
2
3
4
5
6
7
8
9
gatekeeper interactions may lead to improved proteinꢀligand interactions. To further strengthen
this concept, we introduced an indazole into the most promising compound 5b to increase the
spatial and hydrophobic character while retaining the crucial hinge binding motif as well as the
metaꢀsubstituted phenylether as a linker system for the electrophile, resulting in 6a. We observed
a strikingly increased effect on the drugꢀresistant mutant of EGFR (IC50 = 0.07 µM) for 6a. The
mutant variant harboring the activating mutation is 7ꢀfold less well inhibited (0.5 µM, L858R)
whereas only moderate activity could be observed against the wild type (IC50 = 1.7 µM),
indicating a favorable interaction of the methionine gatekeeper and the indazole moiety. With
respect to the reversible counterpart (6b), we did not observe an inhibitory effect on any of the
EGFR variants, thereby demonstrating the importance of a covalent bond. We intended to further
increase the gatekeeper interacting element by introducing a bromine in the 5ꢀposition of the
indazole and likewise to sterically constrain the inhibitor’s conformation (6d). We observed
similar behavior for 6d against the wild type (IC50 = 2.9 µM) as well as against the mutant form
EGFRꢀL858R (0.4 µM) while 6d displayed a slightly reduced effect against the drugꢀresistant
mutant variant (0.3 µM, L858R/T790M). In order to investigate the electrophilic characteristics
of the Michaelꢀacceptor system, we equipped the acrylamide with an additional dimethylamino
moiety (compound 6c) which is supposed to increase the reactivity of the thiol nucleophile for
subsequent Michaelꢀaddition (Cys797) by intramolecular base catalysis.41 This modification,
however, led to a significant loss of inhibitory activity against all investigated EGFR mutants
(>10 µM, wt; 3.6 µM, L858R; 0.7 µM, L858R/T790M). These observations indicate that the
acrylamide moiety represents a more favored electrophile for this scaffold in terms of size,
reactivity, and orientation.
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
ACS Paragon Plus Environment
13

Page 15 of 69
Journal of Medicinal Chemistry
1
2
3
4
5
6
7
8
9
Notably, initial ESIꢀMS analyses of EGFRꢀT790M treated with 3a, 5b or 6a resulted in mass
increases equivalent to the corresponding singleꢀlabeled EGFRꢀT790M (303 Da (3a), 436 Da
(5b), and 468 Da (6a), respectively), as compared to control EGFRꢀT790M treated with DMSO
indicating a covalent inhibitor adduct with EGFRꢀT790M (Figure S1). Furthermore, utilizing
proteolytic digestion of 5b and 6a and nanoꢀLCꢀMS/MS analysis we precisely confirmed
selective Cys797 modification for both compounds (Figure 3 and Supporting Information).
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
Activity-Based Characterization of Quinazoline and Pyrimidine-Based Scaffolds on the
EGFR surrogate system cSrc.
To account for the validity of our EGFR surrogate system, we likewise tested the focused set
of compounds on cSrc and selected mutant variants. In particular, we determined the IC50 values
with respect to cSrc wild type that lacks both, a large gatekeeper and a reactive Cys at the lip of
the ATPꢀpocket. Furthermore, we investigated the inhibitory activities on engineered cSrc
gatekeeper mutant variant T338M as well as on a mutant incorporating the gatekeeper mutation
T338M and the reactive cysteine (S345C) to serve as model system for EGFR T790M
(Table S1). For the parent compound 1a we observed moderate inhibitory activity towards cSrc
wild type (IC50 = 0.211 µM), while inhibition of cSrc T338M, as well as T338M/S345C, was
sixꢀ to nineꢀfold stronger, but did not improve with introducing S345C (IC50 = 0.065 µM on
T338M vs 0.043 µM on T338M/S345C) indicating that the mutant form is addressed more
efficiently, however, not covalently in case of T338M/S345C. The reversible counterpart 1b
likewise reflects improved inhibitory activity on the mutated variants as compared to wild type
cSrc substantiating our conclusion (IC50 = 0.336 µM vs. 0.060 µM vs. 0.048 µM). Improved
inhibitory effects towards the gatekeeper mutant forms of cSrc were observed and reflect a
ACS Paragon Plus Environment
14

Journal of Medicinal Chemistry
Page 16 of 69
1
2
3
4
5
6
7
8
9
general trend throughout the focused compound collection (Table S1). However, for 5b we
observed distinct differences among the mutated variants. In detail, an IC50 value of 0.170 µM
was determined for cSrc T338M whereas cSrc T338M/S345C was inhibited fiveꢀfold more
efficient (IC50 = 0.028 µM). On the contrary, we did not observe increasing potency for the
analogous reversible compound 5e (IC50 = 0.254 µM and 0.349 µM, respectively). Analogous
characteristics were found for 6a and 6b accounting for more efficient inhibition of cSrc
T338M/S345C in case of the Michaelꢀacceptor decorated 6a (IC50 = 0.017 µM) as compared to
the reversible analogue 6b (IC50 = 0.260 µM). These results validated further the use of
engineered cSrc as a surrogate system for EGFR and its mutants.
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
Kinetic Characterization of Covalent Bond Formation of 5b and 6a to EGFR Mutant
Variants.
Covalent binding of inhibitors to their respective target proteins requires distinct spatial
orientations that allow for efficient and irreversible modification of a unique cysteine. In
particular, Cys797 in EGFR serves as the anchor point for irreversible inhibitors equipped with
warheads. In order to investigate the rate and efficiency of covalent bond formation of our most
effective inhibitors 5b and 6a, we set out to determine the corresponding Ki and kinact parameters
with respect to the mutant variants L858R and T790M/L858R of EGFR utilizing an activityꢀ
based assay system (See Experimental Section). We monitored the respective IC50 values of 5b
and 6a after treatment of the respective proteins in a timeꢀdependent manner and correlated the
according values to the respective incubation times which then translated into the determination
of Ki and kinact as described in the literature (Table 2).42 Based on the kinetic characterization, we
demonstrated 5b as well as 6a to exhibit particularly high affinity towards EGFR L858R/T790M
ACS Paragon Plus Environment
15

Page 17 of 69
Journal of Medicinal Chemistry
1
2
3
4
5
6
7
8
9
(Ki = 0.64 nM and 0.32 nM, respectively) whereas they solely account for specific moderate
reactivity (kinact = 0.116 minꢀ1 and 0.137 minꢀ1, respectively) as compared to known reference
compounds.43 On the contrary, we likewise illustrated that both the affinity and specific
reactivity of 5b and 6a towards EGFR L858R is significantly impaired (Ki = 70.2 nM and
833 nM, respectively; kinact = 0.017 minꢀ1 and 0.055 minꢀ1, respectively) as compared to EGFR
L858R/T790M.
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
Complex Crystal Structures of 5b and 5c
In order to investigate the binding modes of our newly designed inhibitors and to substantiate
the observed SARs, we generated complex crystal structures of representative compounds of our
focused library. We coꢀcrystallized 5b and 5c (Figure 4) in complex with engineered cSrc
(T338M/S345C) as the model system validated for EGFRꢀT790M18, 36. We focused on 5b and 5c
as they exclusively differed in the substitution pattern of the attachment point of the Michaelꢀ
acceptor to the phenylether linker system, but exhibited a dramatic discrepancy in inhibitory
activity towards the investigated EGFR mutant variants.
In accordance to the Xꢀray structure of 1a, the complex crystal structures of both 5b and 5c
revealed a hinge contact addressed by three hydrogen bonds to the peptide backbone residues of
Glu339 and Met341. The pyrazole amine forms a hydrogen bond to the carbonyl group of
Glu339 while Met341 is addressed by N2 of the pyrazole that serves as the hydrogen bond
donor. The 3ꢀamino moiety of the pyrazole forms the third hydrogen bond to the carbonyl group
of Met341. The coꢀcrystal structure of cSrcꢀDMꢀ5b confirms the covalent bond between Cys345
and the inhibitor. The phenylether carrying an acrylamide in the metaꢀposition adopted a
perpendicular orientation towards the pyrimidine core, thereby facilitating a close proximity and
ACS Paragon Plus Environment
16

Journal of Medicinal Chemistry
Page 18 of 69
1
2
3
4
5
6
7
8
9
an optimal geometry of the electrophile towards Cys345 that was crucial for alkylation. The 6ꢀ
substituted methylpiperazine extended to the solvent exposed region of the ATP binding cleft
functioning as solubilizing group. As anticipated by modeling studies, the newly introduced
methyl group in the 5ꢀposition of the pyrazolo group accurately points towards the methionine
gatekeeper, most likely inducing a hydrophobic proteinꢀligand interactions. The coꢀcrystal
structure cSrcꢀDMꢀ5c, in contrast, revealed a reversible binding mode for 5c, in which the
phenylether (acrylamide in the paraꢀposition) adopted a stretched or slightly bent conformation
accessing a subpocket adjacent to the gatekeeper residue, thereby preventing a covalent bond
formation to Cys345.
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
To date, only covalent inhibitors have been sufficiently effective in overcoming T790M drug
resistance. Therefore, these Xꢀray structures demonstrate the exceptional importance of
optimized Michaelꢀacceptor systems for each inhibitor scaffold with respect to orientation,
basicity, and reactivity. These remarkable structural observations reveal a novel scaffold
possessing the ability to form a covalent bond and they likewise elucidate the opposing
inhibitory characteristics of 5b and 5c. Furthermore, this structureꢀguided approach provides
valuable insights into the SAR and may stimulate further approaches to develop covalent
inhibitors targeting the drug resistance in EGFR.
Evaluation of Selected Derivatives on Drug-Resistant NSCLC Cell Line
We investigated inhibitory effects of a selected subset of representative quinazolineꢀ and
pyrimidineꢀbased compounds in preliminary studies utilizing the drugꢀresistant NSCLC cell line
H1975 and we additionally included the HCC827 (exon 19 deletion) and metastatic lung cancer
cells harboring wild type EGFR (A431 and H661 Table 3). Each cell type was treated with
ACS Paragon Plus Environment
17

Page 19 of 69
Journal of Medicinal Chemistry
1
2
3
4
5
6
7
8
9
quinazoline 1a, and 2,4,6ꢀ substituted pyrimidines 5b, 5e, 5j, 6a, 6b, 6c, 6d, 6e and 7 (0ꢀ30 µM)
respectively for 96 h. Although several covalent inhibitors of our focused library displayed
strong inhibitory effects in biochemical evaluations using isolated kinases, we observed poor
translation to cellular potency. While parent compound 1a equally addressed NSCLC lines
harboring wild type and gatekeeper mutated EGFR (GI50 = 0.7 µM, 1.9 µM, and 1.6 µM for
H661, HCC827, and H1975, respectively) none of the structurally altered pyrimidines displayed
improved effects in terms of inhibitory efficacy and mutantꢀselectivity. For instance, a distinct
biochemical selectivity profile was observed for 6a and its reversible counterpart 6b targeting the
mutant form of EGFR while sparing the wild type (Table 1). On the contrary, this trend was not
reflected in cellular evaluations since 6a solely exhibited moderate effects on the activating and
drug resistant cell lines HCC827 and H1975 (GI50 = 2.5 µM and 9.8 µM, respectively) and
likewise displayed comparable inhibitory activity on the wild type cell line A431
(GI50 = 8.6 µM). However, the reversible counterpart 6b did not display any significant effect on
the NSCLC cell lines, thereby indicating a certain beneficial cellular effect caused by the
inhibitor’s properties to form covalent adducts. We observed a slightly different tendency for 5b
and 5e since cellular evaluations led to equal GI50 values of 5.7 µM and 8.4 µM respectively on
EGFR wild type cell lines, but on the contrary 5e did not account for any significant inhibitory
effect in HCC827 and H1975 at concentrations up to 30 µM while the Michael acceptor
equipped derivative 5b illustrated moderate inhibition on both activating and drugꢀresistant
mutant cell lines (GI50 = 8.2 µM and 20 µM). Of note, nitroꢀdecorated compound 5h represents
one of the most active inhibitors in our cellular setting eliciting GI50 values of 3.5 µM, 8.9 µM,
and 6.5 µM towards H661, HCC827, and H1975 respectively.
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
ACS Paragon Plus Environment
18

Journal of Medicinal Chemistry
Page 20 of 69
1
2
3
4
5
6
7
8
9
Interestingly, we observed similar cellular effects for the majority of our compounds on the
cell line harboring the activating mutation (HCC827) as compared to the drugꢀresistant H1975
cell line without a dramatic loss in activity. However, the entity of investigated compounds did
not account for significant reduction of cell viability in cellular settings indicating the need for a
stronger inhibitory effect on the kinase activity and likewise strengthening the need for further
investigation with respect to pharmacokinetic properties such as solubility and cell permeability.
In order to assess these pharmacokinetic parameters and to pursue the poor translation of
biochemical activity into cellularꢀbased potency, we set out to investigate the solubility and the
cell permeability for the most promising compounds 5b and 6a using in vitro assays (see
Experimental Section). The kinetic solubility for 5b and 6a was measured at pH 7.4 and revealed
a high aqueous solubility of 432 µM and 277 µM for 5b and 6a, respectively (Table S3).
Furthermore, we gained valuable insights into the cellular absorption using an artificial
membrane permeation assay (PAMPA). 5b exhibited a rather low cellular penetration at pH 7.4
(PAMPA, flux 7%), whereas 6a, equipped with an indazole instead of a pyrazolo moiety,
showed a 7ꢀfold higher permeability (PAMPA, flux 47%) representing a sufficient cellular
penetration in an artificial membrane system according to the assay’s classification.44 However,
we further investigated the cell permeability of 5b and 6a using a Cacoꢀ2 cell assay, in which the
ratio of influx/efflux across Cacoꢀ2 cells is determined, and, therefore, is more appropriate to
reflect the setting in a cellular environment. Strikingly, we observed a high efflux rate for both
compounds (efflux over cell penetration: 32:1 and 35:1 for 5b and 6a) indicating a poor overall
absorption that leads to a low receptor occupancy in the cells, thus providing a reasonable
explanation for the compound’s moderate cellular efficacy as compared to their inhibitory
activity with respect to isolated kinases.
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
ACS Paragon Plus Environment
19

Page 21 of 69
Journal of Medicinal Chemistry
1
2
3
4
5
6
Computational Studies of Ligands’ Conformation in Solution
7
8
9
The drug COꢀ1686 as one of the current frontꢀrunners in EGFR related cancer therapy is
supposed to be conformationally constrained and therefore most likely to adopt a preconfigured
conformation which is very similar to the observed binding mode in the WZ4002ꢀEGFR
complex (PDB code 3ika). Preconfigured compounds take advantage of the fact that the free
energy penalty incurred upon binding a ligand in an optimum conformation to the protein of
interest is markedly lower than for those compounds that present considerably different
conformational states in aqueous solution and therefore must undergo unfavorable
conformational changes upon ligand binding. For this reason, our current efforts, especially by
introducing an indazole moiety, were directed to the design of compounds that take advantage of
a certain preconfiguration to minimize disfavored conformational changes upon intrinsic
compound rearrangements. In order to test these propositions, we performed theoretical
calculations of the conformational ensembles for selected compounds in order to identify most
favored conformations that could be adopted in aqueous solution and their similarity to the
crystallographic binding mode.
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
Conceptually, (near) exhaustive search of the conformational space of typical EGFR ligands in
solution is problematic due to the large number of accessible states which requires sampling
schemes followed by geometrical clustering approaches in order to identify representative
structures.45ꢀ47 Moreover, empirical force fieldꢀbased methods, for instance based on the
“General Amber Force Field” (GAFF48, 49) are not accurate enough to discriminate possibly
subtle population shifts arising from free energy differences below 1 kcal molꢀ1. This represents
the typical soꢀcalled “chemical accuracy” which poses a systematic limit to force field
ACS Paragon Plus Environment
20

Journal of Medicinal Chemistry
Page 22 of 69
1
2
calculations and simulations for instance of small ligands in solution50 or of proteinꢀligand
interactions51. To overcome this uncertainty, we developed a strategy to estimate the
configurational free energy of conformational “basins” of free ligands in solution based on
quantumꢀchemical calculations taking into account an aqueous solution environment. To address
the complexity of the conformational space, we started by generating a, for all practical
purposes, uniformly distributed ensemble of structures by sampling from a highꢀtemperature
simulation trajectory of the ligand with fixed charges immersed in a PoissonꢀBoltzmannꢀtype
continuum model of the solvent. After applying a sufficiently large energetic cutoff of
3 kcal molꢀ1 (corresponding to a Boltzmann weight of skipped conformations of less than 0.007
at ambient temperature) in order to remove highꢀenergy structures, the resulting snapshots were
geometrically clustered. The minimum energy geometries of each cluster (still from a fixed
charged force field) were subjected to further quantumꢀchemical treatments. These consisted for
each cluster optimum in a preoptimization step under vacuum conditions followed by final
optimization taking into account a solvent by the polarizable continuum model (PCM52), yielding
the configuration’s free energy in solution GPCM. Finally, the free energies of the resulting
structures in solution were rescored by the embeddedꢀcluster reference interaction site model
(ECꢀRISM53, giving values GECꢀRISM) integral equation theory, using settings adopted earlier for
tautomer54 and chemical shift55 predictions in water. The number of cluster members was taken
as degeneracy factor gi,j for cluster i in the conformational “basin” j, where the latter has been
defined by grouping clusters that have visually sufficient distance between each other (quantified
in the caption to Figures 5 and 6) as measured by their root mean square deviation (RMSD) with
respect to the crystallographic binding mode. Such a basin corresponds to a certain distribution
of dihedral angles (data and atom indices are discussed below) which define a specific
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
ACS Paragon Plus Environment
21

Page 23 of 69
Journal of Medicinal Chemistry
1
2
3
4
5
6
7
8
9
conformation. Different basins are therefore characteristic dihedral angle groups, allowing for
limited fluctuations represented by the other members within a basin. Neglecting contributions
from vibrational entropy associated with each cluster well, the cluster population is given by 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
Boltzmann weight at ambient target temperature T,
PCM|EC−RISM
i, j
gi, j exp(−G
/ RT)
PCM|EC−RISM
i, j
p
=
(1)
PCM|EC−RISM
i, j
gi, j exp(−G
/ RT)
∑i ∑j
where R is the gas constant, while the population of each conformational basin is defined by
PCM|EC−RISM
pPj CM|EC−RISM
=
p
i, j
i
(2)
∑
This methodology has been applied initially to WZ4002 in order to test the performance with
respect to the proposition that a substitution of chlorine by hydrogen could lead to more
conformational freedom and, correspondingly, less inhibitory activity.40 Results are summarized
in Tables 4 and 5 as well as Figure 5. Most importantly, only four dihedral angels are relevant for
defining the accessible conformational space, as shown in Figure 5. Here, WZ4002 is the only
compound which exhibits a close match between reference crystal structure and basin 1 (smallest
RMSD) in terms of three out of four dihedrals, while the population of this basin is nonzero. In
this sense, WZ4002 is indeed preconfigured since only a single torsional rearrangement of
members of basin 1 is required in order to bind. In contrast, the structurally completely differing
basin 4 (largest RMSD) has to rotate all four dihedrals to facilitate binding. Although we cannot
estimate the binding pathway (in the sense of making assumptions about the possibility of
internal rotations at certain points on the binding reaction coordinate), it is clear that the visually
elongated form is not likely to bind directly. However, its population is basically zero in free
solution, so all essentially contributing conformations of the compound are to a certain degree
preconfigured, as measured by the preconfiguration ratio (PR) defined in Table 5 being 100:0.
ACS Paragon Plus Environment
22

Journal of Medicinal Chemistry
Page 24 of 69
1
2
3
4
5
6
7
8
9
In contrast, dihedral and population data for WZ4002ꢀH show that the latter derivative exhibits
a form closest to the binding mode where at least two dihedrals have to rotate. This is already an
indication of less likely binding. Moreover, the PR in this case shows a shift toward elongated
conformations (basin 3) whereas the other two conformations are Uꢀshaped as defined by
dihedral angles spanned by atoms (2,3,4,5) and (5,6,7,8) being close to 0 (see Table 4). Note that
the diffuse RMSD boundary between basins 2 and 3 is nevertheless clearly characterized by a
change of angle (5,6,7,8) by almost 180°. In summary, the reduced WZ4002ꢀH activity is indeed
correlated with a loss of conformational preconfiguration propensity.
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 order to understand the conformational space of the present set of compounds, we examined
analogously 5b, 5n, and 6a. Numbers for dihedral angles, populations, and graphical
representations of main conformers are also shown in Tables 4 and 5 as well as in Figure 6.
Again, we find that it is possible to reduce the conformational space to four essential dihedral
angles. Remarkably, at least 5n possesses an accessible state where three out of four dihedrals
match the crystal binding mode of reference compound 5b fully, but this state is practically
unpopulated for all 5b, 5n, and 6a – in contrast to the best matching WZ4002 conformer. All
three compounds have to undergo a rotation of at least one dihedral angle by 180° whereas
WZ4002’s best matching conformation only has to rotate a single dihedral (6,7,8,9) by 57° as
mentioned in Table 4. Also, all three compounds show some tendency to form mirrorꢀlike image
relations, defined by the inverted angles (1,2,3,4) and (2,3,4,5) between basins 2 and 3, which are
similarly populated. Some degree of preconfiguration exists for this class of compounds
measured by summing populations of basins 1 (which contributes only marginally) and 2 that
come closest to the binding. It is important to note that 6a, the most active compound against the
L858R/T790M double mutant of EGFR, shows indeed the some increased propensity toward the
ACS Paragon Plus Environment
23

Page 25 of 69
Journal of Medicinal Chemistry
1
2
3
4
preconfigured conformation as compared to 5b and 5n, corroborating the proposition that
5
6
conformational restriction is a valid design goal that can be predicted by computational means.
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
DISCUSSION AND CONCLUSIONS
Drug resistance mutations within the catalytic pocket, which occur as the consequence of
secondary mutations after firstꢀline treatment, represent a major challenge in modern cancer
therapy, as the target proteins have to be considered as entirely different enzymes and, therefore,
require newly developed inhibitors.31, 33, 56, 57 However, in case of the T790M mutation in EGFR,
recently designed thirdꢀgeneration inhibitors bearing a warhead for covalent modification of a
unique and reactive cysteine at the lip of the ATPꢀbinding cleft in EGFR provide evidence for a
successful mode of action without evolving severe side effects.31, 33, 58 Covalent drugs equipped
with Michael acceptors were not appreciated in medicinal chemistry for a long time. Such drugs
were mainly seen as acting nonspecifically because of their reactive groups and thereby driving
toxicity such as idiosyncratic drugꢀrelated toxicity by eliciting adverse immune response.59ꢀ61
Recently, it has been found that the specific fineꢀtuning of these covalent drugs leads to distinct
therapeutic and selective effects, which are not associated with toxicity in vivo. It has been
turned out that an acrylamide warhead, which is perfectly attached to a drug in terms of
geometry and orientation towards the reactive cysteine, appears to represent an excellent
compromise regarding reactivity and toxicity as it is feasible to specifically alkylate the target
protein in vivo while omitting unwanted irreversible interactions with arbitrary proteins.31, 33, 62, 63
Current design approaches take advantage of cysteine mapping along the kinome and therefore,
the concept of covalently modifying target protein to increase the target residence time and elicit
temporal target selectivity has been revived in the last decade. Current efforts in medicinal
ACS Paragon Plus Environment
24

Journal of Medicinal Chemistry
Page 26 of 69
1
2
3
4
5
6
7
8
9
chemistry, for instance, are directed towards the design and development of selective and
irreversibly bound compounds to overcome secondary drug resistance mutations, as in the case
of EGFRꢀT790M.20ꢀ22, 64 Irreversible inhibitors such as COꢀ1686 and AZD9291 are currently at
the leading edge of clinically relevant therapeutics for the treatment of advanced NSCLC and
may stimulate further rational design approaches adapted from covalent modification of target
proteins.30, 32 Here, we have presented a structureꢀguided development of irreversible inhibitors
to selectively target the L858R/T790M mutant form of EGFR. Initial screening of 1500
compounds against 80 NSCLC cell lines led us to the identification of 1a, revealing an attractive
chemotype and showing effective inhibitory impact on H1975, an EGFRꢀL858R/T790M mutant
lung cancer cell line. However, the pharmacological effect was not limited to H1975 but to the
majority of investigated cell lines which indicates rather broad specificity. We set out to utilize
the general quinazoline scaffold of 1a as the structural basis for further rational design
approaches that would result in compounds with increased specificity while retaining cellular
potency. Therefore, we utilized Xꢀray crystallography and discovered the complex crystal
structure of 1a in complex with genetically engineered cSrc as the initial model system for the
therapeutically relevant kinase EGFR. By methods of synthetic chemistry, we produced focused
compound libraries and particularly paid attention to the variation of the quinazoline core
structure resulting in further quinazolines (2a-f) as well as various substituted pyrimidines (3a-b,
4, 5a-q, 6a-e, 7). The latter were extensively elaborated with respect to different substitution
patterns (e.g., Michael acceptor position, 2,4ꢀ vs. 2,4,6,ꢀsubstitution) and revealed distinct
inhibitory effects in biochemical characterizations using wild type as well as L858R and
accordingly L858R/T790M mutated EGFR. While quinazolineꢀbased inhibitor 2b illustrated
increasing inhibitory impact in biochemical characterizations (IC50 = 1.1 µM (wt) > 0.5 µM
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
ACS Paragon Plus Environment
25

Page 27 of 69
Journal of Medicinal Chemistry
1
2
3
4
5
6
7
8
9
L858R > 0.1 µM L858R/T790M) and thereby displaying mutant selectivity, the related pendants
2a and 2c, differing only by the position of the attachment of the Michaelꢀacceptor, basically did
not evoke any effect. Shortened pyrimidine scaffolds 3a and 5l, however, illustrated the same
mutant selectivity. Notably, for the 2,4,6ꢀsubstituted scaffold 5b, the mutant selective inhibition
profile was somehow inverted since the L858R mutant form of EGFR was predominantly
addressed (IC50 = 0.4 µM (wt) > 0.03 µM (L858R) < 0.5 µM (L858R/T790M)) whereas 6a
exhibited an excellent as well as a mutantꢀselective effect on EGFR mutant variants
(IC50 = 2.2 µM (wt) > 0.5 µM L858R > 0.07 µM L858R/T790M). Constitutive protein Xꢀray
experiments using a validated model system of cSrc (T338M/S345C) representing EGFRꢀ
T790M and promising inhibitors from our focused library revealed binding modes that are in
accordance with covalent modification of the target protein as in the case of 5b. Moreover, MSꢀ
based analyses further supported our result of covalent modification of EGFR by demonstrating
increased molecular weight of the proteinꢀligand complex matching the respective monolabeled
species as in the case of EGFRꢀ3a, EGFRꢀ5b, and EGFRꢀ6a. In addition, we confirmed the
alkylation of Cys797 for both inhibitors by conducting proteolytic digestion of 5b and 6a and
subsequent nanoꢀLCꢀMS/MS analysis. As a proof of concept study, we likewise tested our
focused compound collection with respect to their inhibitory activities on cSrc as our initial
EGFR surrogate system and found the majority our covalent inhibitors to more efficiently inhibit
cSrc T338M/S345C as compared to cSrc T338M and cSrc wildtype.
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 a further experimental setup, we set out to investigate the kinetic properties of 5b and 6a
with respect to timeꢀdependent affinity and specific reactivity towards EGFR mutant variants
(L858R and L858R/T790M) by determination of the respective Ki and kinact parameters. Based
on this analysis we found that the inhibitors address the gatekeeper mutant form of EGFR
ACS Paragon Plus Environment
26

Journal of Medicinal Chemistry
Page 28 of 69
1
2
3
4
5
6
7
8
9
(L858R/T790M) with high affinity reflected by Ki values in the subnanomolar range whereas the
rate of inactivation remained fairly moderate (kinact). On the contrary, the activating mutant
(L858R) was addressed with significantly less affinity revealing Ki values of 70 nM (5b) and
833 nM (6b) and beyond that the rate of inactivation was marginal (kinact ≤ 0.055 minꢀ1).
Covalent inhibitors achieve their potency through a twoꢀstepꢀmechanism. In the first step, a
covalent inhibitor reversibly interacts with the target kinase to build a crucial nonꢀcovalent drugꢀ
target complex. In a second step, a covalent bond is exclusively formed in vivo in case the
inhibitor incorporates an optimized electrophile located in close proximity towards the reactive
cysteine that allows a rapid alkylation. Therefore, a potent covalent inhibitor should comprise
two essential feature: (i) a strong reversible binding interaction and (ii) an optimized electrophile
that facilitates rapid covalent bond formation.
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
We observed discriminative inhibitory activities of 5b and its reversible counterpart 5e that
clearly demonstrates the importance of covalently targeting the drugꢀresistant mutant of EGFR,
as 5e exhibited a 60ꢀfold and >20ꢀfold loss in inhibition against EGFRꢀL858R and EGFRꢀ
L858R/T790M. As a Michaelꢀacceptor represents the only structural difference between these
two inhibitors, the shift in activity might directly correlate with beneficial effects due to the
covalency of the compound and its associated prolonged drugꢀtarget residence time. In
conjunction with the MSꢀbased analyzes and the Xꢀray crystal structures our most promising
pyrimidine based inhibitors 5b and 6a seem to form a covalent adduct with EGFR in vitro. These
results explicitly revealed the potential of covalent inhibitors and highlighted the benefits of
covalent design strategies for further drug development.
We also conducted theoretical calculations investigating the potentially preconfigured
conformation of our 2,4,6,ꢀsubstituted pyrimidine scaffolds (5b, 5n and 6a) and the
ACS Paragon Plus Environment
27

Page 29 of 69
Journal of Medicinal Chemistry
1
2
3
4
5
6
7
8
9
WZ4002/COꢀ1686 shape. Comparative analysis of both structure types revealed that the Uꢀshape
of WZ4002, the significantly populated geometry (out of four basins) that resembles the binding
mode most closely, to a large extent, already exists in solution prior to interacting with the target
protein. This favorable preconfiguration triggers the superior inhibitory effect of this compound
because the free energy penalty that must be paid in order to bind to the target protein is much
lower than that incurred by nonꢀpreconfigured species. In the case of our compounds, we
similarly found three discrete conformational basins of which only two are significantly
populated with mirrorꢀlike conformations that are adopted in solution. For the nonzero populated
basin closest to the binding mode, at least two dihedrals have to rotate which may result in an
increased penalty and, therefore, reduced the inhibitory effect. Preconfiguration exists for this
class, but not to the same extent and similar relevance as for WZ4002. The compound that is
most active against the double mutant exhibits the largest average preconfiguration ratio,
although the effect is too small to be the only reason for the observed activity trend. The results
presented in this paper indicate that conformation control during ligand design is important and
furthermore demonstrate the feasibility of computational tools to address this challenge.
A further investigation of pharmacokinetic parameters such as solubility, cell permeability and
drugꢀabsorption for 5b and 6a revealed an excellent aqueous solubility for both compounds and a
sufficient cellular permeability for 6a in an artificial membrane system. However, we observed a
dramatic efflux rate for 5b and 6a illustrating a moderate drugꢀabsorption that leads to low
cellular receptor occupancy, and thereby providing a reasonable explanation for their moderate
cellular efficacy in the NSCLC cell lines.
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
ACS Paragon Plus Environment
28