Structure-kinetic relationship reveals the mechanism of selectivity of FAK inhibitors over PYK2

Article abstract of DOI:10.1016/j.chembiol.2021.01.003

There is increasing evidence of a significant correlation between prolonged drug-target residence time and increased drug efficacy. Here, we report a structural rationale for kinetic selectivity between two closely related kinases: focal adhesion kinase (FAK) and proline-rich tyrosine kinase 2 (PYK2). We found that slowly dissociating FAK inhibitors induce helical structure at the DFG motif of FAK but not PYK2. Binding kinetic data, high-resolution structures and mutagenesis data support the role of hydrophobic interactions of inhibitors with the DFG-helical region, providing a structural rationale for slow dissociation rates from FAK and kinetic selectivity over PYK2. Our experimental data correlate well with computed relative residence times from molecular simulations, supporting a feasible strategy for rationally optimizing ligand residence times. We suggest that the interplay between the protein structural mobility and ligand-induced effects is a key regulator of the kinetic selectivity of inhibitors of FAK versus PYK2.

Full text of DOI:10.1016/j.chembiol.2021.01.003

Article
Structure-kinetic relationship reveals the
mechanism of selectivity of FAK inhibitors over PYK2
Graphical abstract
Authors
Benedict-Tilman Berger, Marta Amaral,
Daria B. Kokh, ..., Matthias Frech,
Rebecca C. Wade, Stefan Knapp
Correspondence
knapp@pharmchem.uni-frankfurt.de
In Brief
Berger et al. present a rationale for the
selectivity of PF-562271 on FAK over
PYK2. Investigation of an inhibitor series
by structural and biophysical,
computational, as well as cellular
techniques provided a structure-kinetic-
relationship and revealed a ligand-
induced helical DFG motif resulting in
kinetic selectivity of FAK inhibitors
over PYK2.
Highlights
d
d
d
A helical DFG motif causes slow ligand dissociation rates in
FAK but not PYK2
NanoBRET represents a versatile method measuring ligand
dissociation rates in cells
tRAMD simulation provides a tool for accurate ligand-binding
kinetic predictions
Berger et al., 2021, Cell Chemical Biology 28, 1–13
May 20, 2021 ª 2021 Elsevier Ltd.
https://doi.org/10.1016/j.chembiol.2021.01.003
ll

Please cite this article in press as: Berger et al., Structure-kinetic relationship reveals the mechanism of selectivity of FAK inhibitors over PYK2, Cell
Chemical Biology (2021), https://doi.org/10.1016/j.chembiol.2021.01.003
ll
Article
Structure-kinetic relationship
reveals the mechanism of selectivity
of FAK inhibitors over PYK2
Benedict-Tilman Berger,1,9,11 Marta Amaral,2,3,11,12 Daria B. Kokh,4,11 Ariane Nunes-Alves,4,7 Djordje Musil,2
2
Timo Heinrich, Martin Schroder,1,9 Rebecca Neil,4 Jing Wang,5 Iva Navratilova,6 Joerg Bomke,2 Jonathan M. Elkins,5
¨
Susanne Muller,1,9 Matthias Frech,2 Rebecca C. Wade,4,7,8 and Stefan Knapp1,9,10,13,
1Structural Genomics Consortium, Goethe University Frankfurt, Buchmann Institute for Molecular Life Sciences, Max-von-Laue-Straße 15,
60438 Frankfurt am Main, Germany
€
*
2Discovery Technologies, Merck KGaA, Frankfurter Straße 250, 64293 Darmstadt, Germany
3
´
´
˜
´
Instituto de Biologia Experimental e Tecnologica, Avenida da Republica, Estac¸ ao Agronomica Nacional, 2780-157 Oeiras, Portugal
4Molecular and Cellular Modeling Group, Heidelberg Institute for Theoretical Studies (HITS), Schloß-Wolfsbrunnenweg 35, 69118 Heidelberg,
Germany
5Structural Genomics Consortium, Nuffield Department of Medicine, University of Oxford, Old Road Campus Research Building, Roosevelt
Drive, Oxford OX3 7DQ, UK
6Division of Biological Chemistry and Drug Discovery, School of Life Sciences, University of Dundee, Dow Street, Dundee DD1 5EH, UK
7
€
€
Zentrum fur Molekulare Biologie der Universitat Heidelberg, DKFZ-ZMBH Alliance, Im Neuenheimer Feld 280, 69120 Heidelberg, Germany
8Interdisciplinary Center for Scientific Computing (IWR), Heidelberg University, Im Neuenheimer Feld 205, 69120 Heidelberg, Germany
9Institute of Pharmaceutical Chemistry, Goethe University Frankfurt, Buchmann Institute for Molecular Life Sciences, Max-von-Laue-Straße
9, 60438 Frankfurt am Main, Germany
10German Cancer network DKTK and Frankfurt Cancer Institute (FCI), Goethe University Frankfurt, Frankfurt am Main, Germany
11These authors contributed equally
12
Present address: Biologics Research, Sanofi-Aventis Deutschland GmbH, Industriepark Hochst, 65926 Frankfurt am Main, Germany
¨
13Lead contact
*Correspondence: knapp@pharmchem.uni-frankfurt.de
https://doi.org/10.1016/j.chembiol.2021.01.003
SUMMARY
There is increasing evidence of a significant correlation between prolonged drug-target residence time and
increased drug efficacy. Here, we report a structural rationale for kinetic selectivity between two closely
related kinases: focal adhesion kinase (FAK) and proline-rich tyrosine kinase 2 (PYK2). We found that slowly
dissociating FAK inhibitors induce helical structure at the DFG motif of FAK but not PYK2. Binding kinetic
data, high-resolution structures and mutagenesis data support the role of hydrophobic interactions of inhib-
itors with the DFG-helical region, providing a structural rationale for slow dissociation rates from FAK and
kinetic selectivity over PYK2. Our experimental data correlate well with computed relative residence times
from molecular simulations, supporting a feasible strategy for rationally optimizing ligand residence times.
We suggest that the interplay between the protein structural mobility and ligand-induced effects is a key
regulator of the kinetic selectivity of inhibitors of FAK versus PYK2.
INTRODUCTION
leading to kinase activation differ, with FAK being primarily acti-
vated by integrins, growth factor receptors, and cytokine
Focal adhesion kinase (FAK) and proline-rich tyrosine kinase 2 receptors, while PYK2 activation is dependent on intracellular
(PYK2) are two closely related non-receptor tyrosine kinases calcium mobilization (Avraham et al., 2000). FAKs promote tu-
that constitute the subfamily of adhesion kinases (Schlaepfer mor proliferation and metastasis and are overexpressed and
et al., 1999). Both kinases are involved in the regulation of upregulated in several types of cancers (Fan and Guan, 2011;
cell migration, adhesion, and survival in response to extracel- Tai et al., 2015; Wendt et al., 2013). PYK2 has been found to
lular signals, and they share a similar multidomain organization be upregulated when FAK activity is suppressed, suggesting
comprising an N-terminal FERM domain, a central catalytic ki- compensatory signaling pathways that may result in drug
nase domain (with 61% sequence identity), a proline-rich re- resistance. Efforts have therefore been made to develop dual
gion, and a C-terminal focal adhesion targeting domain (Lipinski FAK/PYK2 inhibitors that are currently in clinical testing
and Loftus, 2010). Despite their homology, signaling events (Lv et al., 2018).
Cell Chemical Biology 28, 1–13, May 20, 2021 ª 2021 Elsevier Ltd.
1

Please cite this article in press as: Berger et al., Structure-kinetic relationship reveals the mechanism of selectivity of FAK inhibitors over PYK2, Cell
Chemical Biology (2021), https://doi.org/10.1016/j.chembiol.2021.01.003
ll
Article
Figure 1. Comparison of the binding modes of PF-562271 (1) in FAK (PDB: 3bz3) and PYK2 (PDB: 5tob)
(A) Chemical structure of PF-562271 (1) and complexes of FAK (cyan) and PYK2 (gray) with PF-562271 (1, yellow). Kinase residues interacting with the inhibitor are
highlighted in blue (for detailed analysis see Table S1); hydrogen bonds are indicated by dashes. The DFG motif is depicted in orange for both kinases.
(B and C) Analysis of the interface between FAK and PF562271 (B, blue) and PYK2 and PF562271 (C, gray). Shown is the buried surface area (BSA) normalized by
the accessible surface area (ASA) per residue. Compound structure was drawn using Biovia Draw 2017, models were depicted using PyMol 1.7.6.2, and bar
charts were created using GraphPad Prism 8.
An interesting series of aminopyrimidines was reported by
Recently, several studies have suggested that the kinetic
Pfizer, resulting in the development of dual FAK/PYK2 inhibitors properties of a drug, specifically the residence time (t =
PF-562271 (1) (Roberts et al., 2008), PF-431396 (Buckbinder 1/koff), may provide a better indication of the duration of the
et al., 2007), and defactinib (PF-8554878 [Jones et al., 2011]), pharmacodynamic action in vivo than binding affinity alone
as well as selective FAK (PF-573228 [Slack-Davis et al., (Copeland, 2016; Schuetz et al., 2017). The observation of a he-
2007]) and PYK2 (PF-719 [Tse et al., 2012]) inhibitors, respec- lical conformation in FAK but not PYK2 suggests that it may
tively. Despite the high sequence similarity of the kinase do- provide a mechanism for the kinetic selectivity of 1. The struc-
mains of FAK and PYK2 (Figure S1), 1 adopts different back- turally very similar dual FAK/PYK2 inhibitor PF-431396 (10, with
pocket binding modes in the two proteins (Figure 1A). Strik- phenylsulfonamide instead of the aminopyridine in 1) shows a
ingly, the conserved tripeptide DFG-motif forms an a-helical similar selectivity in potency (FAK IC50 = 1.5 nM over PYK2
structure in FAK enabling multiple contacts between inhibitor IC50 = 11 nM) (Roberts et al., 2008). As 10 was also reported
and kinase, notably between the pyridine ring and L567 (Fig- to show fast binding kinetics on PYK2 (kon > 106 Mꢀ1
s
ꢀ1, koff
ure 1B). In PYK2, the same amino acid residues are positioned > 10ꢀ2 ꢀ1) (Han et al., 2009), we hypothesize that the about
s
in a linear conformation with only PYK2-D567 being able to 7-fold selectivity for FAK—and therefore also the about 9-fold
form contacts with the inhibitor (Figure 1C). This helical confor- selectivity (Roberts et al., 2008) of 1—may be related to very
mation of the DFG motif in FAK has been invoked as providing slow binding kinetics to FAK. As selectivity is often difficult to
a possible structural explanation for the approximately 9-fold achieve for related protein kinases, an understanding of the
greater potency of 1 for FAK (IC50 = 1.5 nM) over PYK2 mechanisms of kinetic selectivity of inhibitors of FAK and
(IC50 = 13 nM) (Roberts et al., 2008).
PYK2 would be beneficial for future medicinal chemistry efforts.
2
Cell Chemical Biology 28, 1–13, May 20, 2021

Please cite this article in press as: Berger et al., Structure-kinetic relationship reveals the mechanism of selectivity of FAK inhibitors over PYK2, Cell
Chemical Biology (2021), https://doi.org/10.1016/j.chembiol.2021.01.003
ll
Article
Therefore, in this study we explored the structural features L567-ligand interactions increased potency and
that modulate the kinetic properties of a series of FAK and prolonged residence times in FAK
PYK2 inhibitors to support the development of kinetically Modifications in the selected inhibitor set significantly modu-
selective or balanced dual FAK/PYK inhibitors. To gain a lated binding affinity for FAK and PYK2. Interestingly, intro-
mechanistic understanding of the distinct kinetic profiles of ducing a phenylsulfonamide moiety at position R1 in 10
FAK and PYK2 binding, we performed surface plasmon reso- resulted in a 4-fold improvement in potency for FAK compared
nance (SPR) assays, binding experiments in cells, structural with the parent compound 1. We hypothesized that the ben-
and site-directed mutagenesis studies, and conventional mo- zene ring of 10 provided a more favorable hydrophobic inter-
lecular dynamics (MD) and t random acceleration molecular action with L567 compared with the pyridine ring in 1. Indeed,
dynamics (tRAMD) simulations. This combination of tech- if the polarity of the ring system was increased by incorpora-
niques allowed us to relate the structural and dynamic differ- tion of two nitrogen atoms as exemplified in 11, the affinity
ences between the two kinases to the kinetic selectivity of the decreased further. The corresponding change in off-rate was
inhibitors.
a result of a difference in binding free energy, as the on-rates
of 1, 10, and 11 were similar. For PYK2, we observed similar
affinities and kinetic parameters for 1, 10, and 11, in agree-
ment with the lack of the induced helical structure and interac-
tion with the conserved PYK2-L570.
RESULTS AND DISCUSSION
Design of a series of inhibitors to probe the structure-
activity relationship in FAK and PYK2
The introduction of a sulfonylphenyl moiety at R1 (5) resulted in
Based on the comparison of the crystal structures of 1 in com- a decreased affinity for FAK. This modification was accompa-
plex with FAK and PYK2 (Figure 1), we designed and synthe- nied by a 2-fold increase in the inhibitor off-rate. Compared
sized a series of inhibitors to probe the influence of interac- with 1, the removal of the sulfonamide in 8 led to a decrease in
tions with different parts of the ATP-binding site on binding binding affinity and interestingly, a substitution of the pyridine
kinetics in FAK and PYK2. The central aminopyrimidine hinge ring in 8 with benzene (9) or para-fluorobenzene (7) was not toler-
binding scaffold was modified at three positions, probing in- ated and resulted in an affinity loss of 350-fold and 1,300-fold,
teractions with the solvent-exposed pocket and interactions respectively. Introduction of an indoline-2-one ring system
with the non-conserved upper lobe residues (R1), the DFG he- was, however, tolerated (6), maintaining low nanomolar affinity.
lix induced in FAK after ligand binding (R2), and the back- Similar effects on inhibitor-binding affinity by R1 substitutions
pocket interface with the induced helix in FAK (R3) (Table 1). were observed for PYK2.
For a first assessment of the effect of the modifications on
inhibitor-binding kinetics and affinity, we performed SPR mining both potency and the binding kinetics profile. Changing
measurements.
the oxindole ring in 1 to a phenyl (3) resulted in a 93-fold decrease
Interestingly, the R2 substituent played a key role in deter-
To develop inhibitors that show different binding kinetics be- in potency and a 60-fold drop in the on-rate, most likely because
tween these two related kinases, we were particularly interested the phenyl ring does not allow the formation of a hydrogen bond
in the role of the hydrophobic interaction of the inhibitor with the with FAK-R426. Inhibitor 12 has been reported as a PYK2-selec-
induced helix located C-terminal to the FAK DFG motif (residues tive inhibitor (Tse et al., 2012), and we indeed measured tighter
D564–G566), which positioned FAK-L567 into the ATP-binding binding to PYK2 but observed only about 6-fold selectivity
site. The pyridyl-methanesulfonamide in R1 was therefore re- against FAK.
placed by diverse aromatic moieties (compounds 5–12), modu-
lating the electron density of the pendant aromatic ring system was intriguing that the kon-koff plot (Figure S3) clearly demon-
as well as substitution patterns in the ring.
strated that most inhibitors had very similar association rate con-
Comparing the selectivity profiles of both FAK and PYK2, it
Another interaction unique to FAK was a polar contact of stants kon. Remarkably, the variation was less than 10-fold over
the oxindole of 1 with FAK-R426. Despite conservation of all inhibitors tested. As a result, the different binding affinities for
this residue in PYK2 (R429), this interaction was not observed these two kinases were mainly due to differences in the dissoci-
in crystal structures of this kinase. To probe the contribution ation rate constant koff. Interestingly, despite similar IC50 values,
of this structural difference on binding kinetics, we introduced we observed that 10 showed a dissociation rate from FAK of
single aromatic ring systems instead of the oxindole moiety in 138 min, more than two orders of magnitude slower than that re-
1, including a phenyl moiety (3), a ring-expanded 3,4 dihydro- ported for PYK2 (Han et al., 2009).
quinoline-2-one (4 and 5), N-methylbenzamide (11, defacti-
nib), and N-methylbenzenesulfonamide (12). While we did Slow off-rate inhibitors induce a helical conformation of
not obtain data for a substitution of the oxindole in R1 with the activation loop in FAK
dihydroquinoline-2-one in 4, we observed a comparable po- To better understand the structural mechanisms giving rise to
tency with 1 by introducing a methylsulfonylbenzene in R1 the different binding kinetics of this inhibitor series, we solved
(5). In position R3, only a -CF3 group and a CF3-deleted variant six crystal structures of FAK in complex with different aminopyr-
(-H, 2) were tested. The removal of the trifluoromethyl group in imidine derivatives, including the most interesting substitutions
2 resulted in a significant reduction of affinity compared with in R1, R2, and R3 (Figure 2).
PF-562271 (1), associated with shorter residence time as ex-
The structure of the inhibitor with the slowest dissociation off-
pected for the key interaction of the trifluoromethyl group with rate—10—in complex with FAK confirmed the overall binding
the back pocket. The trifluoromethyl group was therefore kept mode previously described for 1 (Roberts et al., 2008). 10 formed
in all other inhibitors that we investigated.
hydrogen bonds between the inhibitor and the backbone of
Cell Chemical Biology 28, 1–13, May 20, 2021
3

Please cite this article in press as: Berger et al., Structure-kinetic relationship reveals the mechanism of selectivity of FAK inhibitors over PYK2, Cell
Chemical Biology (2021), https://doi.org/10.1016/j.chembiol.2021.01.003
ll
Article
Table 1. Design of a series of inhibitors probing the structure-activity relationship in FAK and PYK2
FAK (SPR)
PYK2 (SPR)
Compound no.
name
kon
ꢀ1 sꢀ1
koff
(10ꢀ2 sꢀ1
kon
ꢀ1 sꢀ1
koff
(10ꢀ2 sꢀ1
R1
R2
R3 (105
M
)
)
KD (nM)
(105
M
)
)
KD (nM)
1
-CF3
53 20.0
0.34 0.078
0.76 0.35
130 43
7.4 2.5
5.7 1.0
PF-562271
2
3
-H
8.8 2.4
29 8.8
320 27
70 22
9.0 3.3
75 32
820 96
-CF3 0.87 0.22
0.62 0.24
0.37 0.15 1.7 0.78 470 100
4
-CF3
-CF3
n.d.
n.d.
n.d.
n.d.
n.d.
n.d.
5
70.2 36
0.82 0.22
1.3 0,32
13 4.8
33 11
250 19
PF-573228
6
7
-CF3
-CF3
17 10
6.1 4.2
14 3.0
37 7.0
17 6.6
19 7.8
110 4.0
1.4 0.44
1,000 23
no binding
8
9
-CF3
-CF3
48 20
7.1 3.0
4.8 1.3
15 2.6
260 21
13 8.4
25 11
230 92
1.8 0.51
no binding
10
-CF3
30.8 14
0.054 0.027 0.18 0.05
58 15
4.0 1.2 6.9 0.54
PF-431396
11
-CF3
-CF3
24 5.8
0.44 0.061
1.9 0.50
15 3.8
3.6 1.1
24 3.1
PF-8554878,
defactinib
12
14 6.5
20 8.3
150 19
22 8.1
5.5 2.4
25 3.3
PF-719
Three substituent sites on the pyrimidine core structure of 1, R1 (solvent exposed), R2 (DFG adjacent), and R3 (back pocket) were considered to study
the interactions in the kinase binding sites. SPR, surface plasmon resonance; kon, association rate constant; koff, dissociation rate constant; KD, equi-
librium dissociation constant; n.d., not determined. Data of n = 5 individual experiments.
C502 located in the hinge region. The oxygen atom of the amino- structures with these slow off-rate inhibitors revealed an induced
oxindole (R2) was within hydrogen-bonding distance of the side helical structure in the activation loop region. The structure in
chain of R426. The trifluoromethyl group (R3) bound to the cavity complex with 4 induced a helical conformation in FAK. There-
created by the side chains of M499, D564, and L567. The phenyl- fore, we assumed similar kinetics compared with 1. Interestingly,
sulfonamide (R1) formed a hydrophobic interaction with L567. for 5, the FAK-L567 interaction was mainly due to a 180ꢁ rotated
Finally, the sulfonamide was stabilized by a hydrogen bond T-shaped p-stacking interaction of the aromatic ring system
with the backbone amide of FAK-D564. The conformation of based on the shorter distance provided by the phenylsulfodiox-
this helix and the binding mode of the inhibitor series discussed ide without the additional atoms present in 1 and additionally
here were similar to that reported in the structure of the complex formed a hydrophobic interaction with L553.
with the clinical FAK inhibitor TAE226 (Figure S2) (Lietha and
Eck, 2008).
We observed that the absence of the R1 methylsulfonamide
group in 8 prevented the formation of the DFG-motif helix, and
In addition, we also determined structures in complex with 3 this inhibitor was shown to have an increased off-rate compared
and 5, harboring a sulfonamide moiety in R1. Intriguingly, all with 1. The oxindole ring of 1 and 10 in position R2 interacted
4
Cell Chemical Biology 28, 1–13, May 20, 2021

Please cite this article in press as: Berger et al., Structure-kinetic relationship reveals the mechanism of selectivity of FAK inhibitors over PYK2, Cell
Chemical Biology (2021), https://doi.org/10.1016/j.chembiol.2021.01.003
ll
Article
Figure 2. L567-ligand interactions increased
potency and prolonged residence times in
FAK
Slow off-rate inhibitors induce a helical confor-
mation of the activation loop in FAK. The top of
each panel shows the chemical structure of
compounds (2 [A], 3 [B], 4 [C], 5 [D], 8 [E], 10 [F])
and their respective residence times t of the
inhibitor-FAK complexes. Shown below are the
determined inhibitor bound FAK structures (FAK
[cyan], inhibitors [yellow]) with highlighted DFG
segment in orange. Dashes indicate hydrogen
bonds. On the bottom of each panel, amino acid
residues contributing to the interface between
FAK and the inhibitor are shown. Displayed is
the buried surface area (BSA) normalized by the
accessible surface area (ASA) per residue (for
detailed analysis see Table S1). Accessions are
PDB: 6yt6 (2), 6yxv (3), 6yq1 (4), 6yoj (5), 6yvs
(8), and 6yr9 (10). Compound structure was
drawn using Biovia Draw 2017, models were
depicted using PyMol 1.7.6.2, and bar charts
were created using GraphPad Prism 8.
conformation by hydrophobic interac-
tions with FAK L567. Simulations sug-
gested that the very potent and slow off-
rate compound 11 had a binding mode
similar to that observed experimentally
for 10 and 1. All the inhibitors contained
a sulfoxide moiety coupled to an aromatic
ring system in R1. The overall results re-
vealed some general trends in inhibitor-
binding characteristics as a function of
the structural rearrangement of the DFG
motif. Compounds that induce a helical
conformation of the DFG motif via a hy-
drophobic interaction with FAK-L567
had significantly slower binding kinetics
withFAKR426viaahydrogenbond. ChangesintheR2 positiondid and higher potency when compared with compounds that
not lead to different interactions of the R2 groups. However, while assumed a type-I binding mode.
the oxindole moiety was withinhydrogen-bonding distanceof FAK
R426, this interaction appeared transient and was not present in all The kinetic selectivity profile of FAK and PYK2 is
chains in the asymmetric unit. This makes an involvement in the dictated by the formation of a helical conformation of
modulationoftheinhibitors’on-ratelesslikely. Weweresuccessful the DFG motif
insolvingthestructureofaninhibitorharboringdifferentmoietiesin We compared our structural data of FAK with published struc-
R3, which pointed toward the back pocket. Analysis of 2 in com- tures of PYK2 in complex with 1 (PDB: 5tob) (Figure 1), 10
plex with FAK revealed significant differences in the pyridyl-meth- (PDB: 3fzr, Figure 3A), and 12 (PDB: 3h3c, Figure 3B) to identify
anesulfonamide group orientation when compared with 10 and 1, crucial interaction residues that are relevant for the kinetic selec-
which inthecaseof2 did notallowa directinteraction withtheDFG tivity for either FAK or PYK2.
motif. Hence, its conformation was the canonical DFG-in arrange-
The superimposed crystal structures of FAK and PYK2 in com-
ment. The trifluoromethyl group was involvedin hydrophobic inter- plex with 10 showed a conformation of the amino-methyl-
actions with M499, D564, and L567 in the structures of inhibitors 1 phenyl-sulfonamide group similar to that reported for 1. In
and 3–11. The lack of these hydrophobic interactions in 2 ex- contrast to an interaction of 10 with the helical DFG motif in
plained the impact of this substituent on binding affinity and fast FAK, this interaction is missing for PYK2. The interaction of
binding dissociation kinetics.
10 with PYK2 is characterized by a DFG-out conformation and
As expected from the structure-activity relationship and struc- hydrophobic interactions with residues preceding the P loop
ture-kinetic relationship analysis, the slowest off-rate inhibitors (L431, G432), similar to FAK. We hypothesized that the differ-
of FAK, 5 and 10, formed interactions similar to those reported ences between the conformations of the DFG motif are likely to
for 1. In particular, the DFG motif was stabilized in a helical contribute to differences in potency and paralog selectivity of
Cell Chemical Biology 28, 1–13, May 20, 2021
5

Please cite this article in press as: Berger et al., Structure-kinetic relationship reveals the mechanism of selectivity of FAK inhibitors over PYK2, Cell
Chemical Biology (2021), https://doi.org/10.1016/j.chembiol.2021.01.003
ll
Article
Figure 3. The selectivity profile of FAK and PYK2 is dictated by the formation of a helical conformation of the DFG motif in FAK
(A) Overlay of crystal structures of the complexes of FAK-10 (blue and yellow, PDB: 6yr9) and PYK2-10 (gray and magenta, PDB: 3fzr [Han et al., 2009]).
(B) Overlay of the complexes of FAK-12 (blue and yellow, PDB: 6yvy) and PYK2-12 (gray and magenta, PDB: 3h3c [Walker et al., 2009]).
Models were depicted using PyMol 1.7.6.2.
the trifluoromethyl pyrimidine series, as previously reported the binding site (in particular the b1, aC, and DFG-motif regions)
(Heinrich et al., 2013; Roberts et al., 2008). The superimposed generally showed more mobility in PYK2 than in FAK (Figure 4A).
crystal structures of FAK and PYK2 also revealed a different The DFG-motif region was highly mobile for the complexes with
conformation of the P loop and a different position of aC. compounds 1 and 3 in PYK2 simulated starting from the crystal
Although this region did not directly interact with the inhibitor, structure with PDB: 5tob, where the position of the phenylalanine
the difference in one particular amino acid residue of the of the DFG fragment is different from that in other PYK2 crystal
P-loop region, F435 (PYK2) versus Q432 (FAK) (Figure S1), might structures and the side-chain position was found to be more var-
affect the stabilization of the helical conformation of the DFG iable in MD simulations (for details see Figure S4).
motif in FAK, as Q432 interacted with the backbone of G566.
To explore the possible reasons for the structural stability of
While FAK-R426 showed an interaction with 10, the correspond- FAK with respect to PYK2, we compared protein-ligand interac-
ing arginine R429 in PYK2 was not within hydrogen-bonding tion fingerprints in the MD trajectories. Although the pattern of
distance.
protein-ligand interactions in trajectories for PYK2 and FAK
In contrast to 10, which showed a 9-fold selectivity for FAK was generally similar (Figures 4B and 4C), there were several dif-
over PYK2, 12 shared the amino pyrimidine scaffold, but ferences. First, with exception of compound 2, an additional
harbored different moieties R1 and R2 and has been reported hydrogen bond with D564 appeared in the FAK complexes
as a PYK2-selective inhibitor (Tse et al., 2012). We solved the with compounds that have a methane sulfonamide R1 moiety ori-
crystal structure of 12 with FAK to provide structural insights ented toward the DFG motif (see Figure 4D).
into the mechanism of this selectivity. Exchanging R1 and R2 res-
The number and stability of protein-ligand hydrophobic inter-
idues, in contrast to 10, resulted in a shift in the position of the actions is larger in FAK. These interactions included not only
sulfonamide, which interacted with R426 (FAK) in b1 in both ki- the DFG motif (D564/D567, L569/570), having a helical confor-
nases. An interaction of the inhibitor with the conserved leucine mation in FAK, but also interactions with residues V484 located
residue L567 or the respective L570 in PYK2 was missing in both deep in the pocket and L553 in addition to interactions with
FAK and PYK2, stabilizing a DFG-in conformation. The FAK DFG conserved residues V436/V439 and A452/A455 (Figure 4D).
motif did not form the helical secondary structure in complex Additionally, the hinge residue L501 in FAK formed a temporary
with 12, as seen for 10, supporting the idea that differences in contact with most ligands. This suggests a tighter packing of the
the arrangement of the DFG motif were relevant for the inhibitor ligand in the binding pocket of FAK than in PYK2. Indeed, most of
selectivity of 12 for PYK2.
the compounds bound to PYK2 form a water-mediated
hydrogen bond with D567 through the R1 moiety, which indi-
cates that the pocket is relatively open and partially hydrophilic.
For FAK, this interaction is observed only for 2 and 6, which
Protein flexibility and the dynamics of the protein-ligand
interactions differ for FAK and PYK2
To investigate the origin of the differences in the binding rates bound to the helical DFG-loop conformation. In agreement
between FAK and PYK2 for the set of compounds considered, with our structural analysis, we did not observe stable hydrogen
we performed MD simulations. We carried out both conventional bonding to FAK R426 from the oxygen in the oxindole ring, and
MD simulations and tRAMD (Kokh et al., 2018) simulations for R426 was only transiently involved in a direct or water-mediated
the complexes of FAK with all 12 compounds listed in Table 1 hydrogen bond with ligands (Figure 4B).
and for PYK2 with 10 of these compounds (omitting compounds
Next, we employed the tRAMD procedure to simulate ligand
7 and 9 for which binding to PYK2 was not detected by SPR). unbinding and to evaluate the relative residence times of the
Conventional MD simulations of the protein-ligand complexes compounds studied. The computed tRAMD residence times
showed that the ligands and the parts of the protein that line overall correlate well with SPR data (R2 = 0.91 for the whole
6
Cell Chemical Biology 28, 1–13, May 20, 2021

Please cite this article in press as: Berger et al., Structure-kinetic relationship reveals the mechanism of selectivity of FAK inhibitors over PYK2, Cell
Chemical Biology (2021), https://doi.org/10.1016/j.chembiol.2021.01.003
ll
Article
Figure 4. Protein flexibility and the dynamics of the protein-ligand interactions differ for FAK and PYK2
Analysis of MD simulations of complexes of ligands with PYK2 and FAK is shown along with computed relative residence times.
(A) Root-mean-square fluctuation (RMSF) of the ligand and three regions of the binding site (b1, aC, and the DFG loop corresponding, respectively, to FAK
residues 421–434, 451–471, and 563–570) averaged over four MD trajectories (80 ns in total) and shown for 12 compounds for FAK and 10 compounds for PYK2.
(B and C) Protein-ligand interaction fingerprint maps for ligand-protein complexes obtained from the last 300 frames of four 20-ns conventional MD simulations for
FAK (B) and PYK2 (C). The occurrence of hydrogen bonds (HD/HA), aromatic (AR), hydrophobic (HY), and water-mediated (WB) protein-ligand interactions is
shown by the color (white, not observed; dark blue, present in all snapshots analyzed). The SPR residence times are shown on a logarithmic scale for all
compounds in the right-hand plots.
(D) View of compound 1 and residues in FAK that contribute to the protein-ligand interactions shown in B.
(E) Correlation between computed relative residence times (tRAMD) and 1/koff values measured by SPR for FAK and PYK2; R2 = 0.91 for the entire set of the data
(R2 = 0.95 and 0.63 for the FAK and PYK2 subsets, respectively). Compound 3 simulated starting from a complex with a helical conformation of the DFG motif in
PYK2 is denoted by ‘‘(3)’’ (see text and Figure S4A).
Figures were created using Python (library Matplotlib) and models were depicted using PyMol 1.7.6.2.
Cell Chemical Biology 28, 1–13, May 20, 2021
7

Please cite this article in press as: Berger et al., Structure-kinetic relationship reveals the mechanism of selectivity of FAK inhibitors over PYK2, Cell
Chemical Biology (2021), https://doi.org/10.1016/j.chembiol.2021.01.003
ll
Article
Figure 5. Binding potency and kinetics of FAK and PYK2 in living cells
(A) Target engagement assays for FAK and PYK2 for 1 and 10 as indicated on the figure. Representative data from at least n = 3 individual experiments.
(B) Correlation of the in cellulo potency with the potency determined in vitro by SPR.
(C) NanoBRET washout experiments for 1 and 10 in comparison with background data and tracer only (DMSO added instead of test compound). Representative
data from at least n = 3 individual experiments.
Figures were created using GraphPad Prism 8.
set of compounds, Figure 4E), including those modeled by sim- et al., 2015). For the target engagement assays, the apparent
ilarity or docked (such as 6, 7, and 11 for FAK, and most com- tracer affinity values of tracer K5 (Vasta et al., 2018) were deter-
pounds for PYK2, see STAR methods). The compounds that mined (46 nM and 78 nM for FAK and PYK2, respectively) and
dissociate more slowly from FAK were well separated from used as input parameters in the assay setup for FAK and PYK2
those with faster dissociation rates. However, the ranking of (Figure S5). Representative results of the target engagement
the fast dissociating compounds (koff > 0.03 sꢀ1) was less ac- assay for compounds 10 and 1 are shown in Figure 5A. The intra-
curate, which was potentially due to the uncertainty in the cellular apparent affinities (6, 7) were generally 10-fold weaker
structure of the modeled complex, as only a few crystal struc- than the binding affinities calculated from the in vitro SPR mea-
tures of PYK2 (1, 10, and 12) were available (see, for example, surements (Figure 5B). NanoBRET data were in very good agree-
the results of simulations with different docking positions for ment with previously reported IC50 values (Allen et al., 2009;
compounds 3 and 9 bound to PYK2 and FAK, respectively, Heinrich et al., 2013; Walker et al., 2008, 2009) for both FAK
Figure S4).
and PYK2. FAK in vitro selectivity for the present series of com-
All compounds with a sulfonamide moiety at R1 bound to the pounds was translated into the cellular context, where 5 was the
FAK complex with a helical DFG-motif conformation formed most selective compound for FAK with a 330-fold difference
hydrogen bonds with the DFG-motif backbone (residue D564, compared with PYK2.
Figure 4D) and showed long residence times (koff < 0.01 sꢀ1).
Next, we investigated the kinetic aspects in the cellular
This interaction, along with the hydrophobic contact with L567, context using NanoBRET technology in a washout format. The
seemed to be critical for residence time prolongation. Indeed, tracer used had the required rapid binding kinetics, making it
for two fast dissociating compounds, 7 and 9 (koff > 0.05 sꢀ1), suitable for the kinetic analysis of slower-binding inhibitors (Fig-
with aromatic R1 substitutions, we did not observe differences ure S5). After incubation of the individual inhibitors with their
in residence times when the simulated ligands were docked target at a concentration of 10-fold of their respective IC50 to
into the protein structure with a helical or extended conformation achieve 90% target occupancy, we removed the unbound com-
of the DFG motif (see, for example, simulation results for com- pound by washout and added the tracer whose association was
pound 9 for different conformations of the protein binding site monitored over time (Figure S5 and Table S3). Our residence
in Figure S4). Importantly, both compounds formed a hydropho- time analysis by NanoBRET revealed remarkably slow dissocia-
bic contact with L567 in the helical DFG-loop structure, although tion rates for 10 on FAK compared with PYK2 and distinguish-
it was much less stable than in the case of compounds with a sul- ably slower kinetics for 1 on FAK. For PYK2, in agreement with
fonamide moiety at R1. One more interesting feature was our SPR data, no inhibitors with slow off-rates were identified
observed in the series of simulations for compound 9: the ligand in cells. We confirmed that compounds 1 and 10, which had
docking positions for which a transient contact with R426 was the longest residence times among the entire series of pyrimi-
possible were found to dissociate several folds slower than dines in vitro, also had slow binding kinetics (Figure 5C) in cells.
those for which this hydrogen bond was not observed (see Fig- As expected from the SPR data, compound 10 showed fast
ure S4). Thus, both protein structural differences (tighter pocket binding kinetics for PYK2, providing an indication for kinetic
and more hydrophobic contacts in FAK) and the ligand’s ability selectivity. The observed rank order of intracellular residence
to form additional hydrogen bonds appeared important for times was in good agreement with the residence times deter-
slow dissociation rates.
mined in SPR experiments. It is noteworthy that in the Nano-
BRET assay, compound 4 showed a very similar potency to
compound 1 and an even longer half-life of dissociation.
Together with the relative residence time computations (Fig-
Binding potency and kinetics of FAK and PYK2 in
living cells
To validate the in vitro selectivity and slow binding kinetics of the ure 4E), these data indicate a similar kinetic behavior of 4
present series of compounds for FAK versus PYK2, we assessed compared with 1.
target engagement and binding kinetics in living cells under near
Interestingly, although 5 had a higher selectivity for FAK over
physiological conditions using NanoBRET technology (Robers PYK2, 10 showed longer occupancy due to its slow dissociation
8
Cell Chemical Biology 28, 1–13, May 20, 2021

Please cite this article in press as: Berger et al., Structure-kinetic relationship reveals the mechanism of selectivity of FAK inhibitors over PYK2, Cell
Chemical Biology (2021), https://doi.org/10.1016/j.chembiol.2021.01.003
ll
Article
Figure 6. Structural stability of the ligand-binding sites in FAK and PYK2 explored by computations and experimental site-directed muta-
genesis
(A–E) Computations. (A and B) Binding site in FAK (left, PDB: 3bz3) and PYK2 (right, PDB: 3et7) indicating residues that may affect protein flexibility (see dis-
cussion in the main text) in stick representation. (C and E) Results of simulations of unbound FAK and ‘‘chimeric’’ PYK2 structures with a helical conformation of
the DFG motif. (C) Variation of the structures of FAK (gray) and chimeric PYK2 (beige) in conventional MD simulations (four frames are shown for each protein);
L570 in PYK2 is highlighted in orange, and the ligand position in a FAK complex is shown for compound 1 (PDB ID 3bz3). (E) Fraction of MD frames where
hydrogen bonds between six residues of the DFG-motif helix and the rest of the protein were observed. (D) Protein-ligand interaction fingerprint map for
compound 3 bound to FAK and PYK2 (with helical or loop conformation of the DFG motif, see Figure 4B for key); residues are labeled by the name and number in
the FAK structure.
(F–H) Site-directed mutagenesis. (F) NanoBRET potency for mutant constructs displayed as a ratio of pIC50 mutant/wild type for mutants as indicated on the
figure. (G and H) Representative washout experiment for 10 on the mutants tested for FAK (G) and PYK2 (H) from at least n = 3 individual experiments. The signal
was normalized to the tracer control and fitted using a one-phase exponential decay.
Models in (A) to (C) were created using PyMol 1.7.6.2. Figures in (D) and (E) and (F) to (H) were created using Python (library Matplotlib) and GraphPad Prism 8,
respectively.
kinetics, providing evidence that the residence time can be a explanation for such structural differences is sequence varia-
valuable indicator of target occupancy.
tions between FAK and PYK2 (Figure S1), which may cause dif-
ferences in protein flexibility. The DFG motif and several of the
following residues are conserved in PYK2 and FAK, whereas in
other regions of the binding site, in particular in b1 and the hinge
Structural mechanisms affecting the protein binding
site flexibility in FAK and PYK2
Analysis of the crystal structures and the protein-ligand interac- region, some sequence variations might affect domain plasticity.
tion contacts in MD simulations suggests that the protein struc- The most notable specific interaction observed in the crystal
ture with a helical DFG-motif conformation formed a tighter structures of FAK was a hydrogen bond between the side-chain
binding pocket and enabled a larger number of hydrophobic amide nitrogen of Q432, located in the P loop, and the backbone
contacts between the protein and the ligand relative to that oxygen atom of L567, located next to the helical DFG motif
with a loop DFG-motif conformation. This analysis did not (Figure 6A). This interaction is not possible in PYK2 because
explain why PYK2 was unable to adopt the helical DFG-motif the hydrophobic F435 is at the Q432 position (Figure 6B),
conformation to form complexes with higher affinity. A possible although a hydrophobic residue at this position is unusual among
Cell Chemical Biology 28, 1–13, May 20, 2021
9

Please cite this article in press as: Berger et al., Structure-kinetic relationship reveals the mechanism of selectivity of FAK inhibitors over PYK2, Cell
Chemical Biology (2021), https://doi.org/10.1016/j.chembiol.2021.01.003
ll
Article
the non-receptor kinases (Figure S6). To explore whether a res- 3 bound to the ‘‘chimeric’’ PYK2 structure in the same pose as in
idue corresponding to Q432 in FAK could stabilize the PYK2 pro- FAK (PDB: 3bz3; Figure S6, structures 3 and 4) and computed its
tein structure, we compared MD simulations of the apo wild-type residence time. The computed residence time of compound 3
PYK2 with F435Q and F435A mutants. Although the PYK2 muta- was doubled relative to that for the loop DFG-motif conformation
tion F435Q was slightly more stable (Figure S6), neither coiling of (see point denoted by ‘‘(3)’’ in Figure 4E) but was still ꢂ5-fold
the DFG-motif loop nor contact between Q435 and the DFG- shorter than for the same compound bound to FAK. The differ-
motif region were observed in the MD trajectories, most likely ence between the protein-ligand interaction fingerprints of FAK
because this would require a much longer simulation time. We and PYK2 is shown in Figure 6D. Although the hydrogen bond
also did not observe any stable aromatic stacking of F435 with between the sulfonamide moiety of the ligand and D569, as
the surrounding aromatic residues in PYK2, which might affect well as the hydrophobic interaction with L570, was still preserved
the position of the P loop in PYK2, although the Q and A residues in the complex with the ‘‘chimeric’’ PYK2 structure, some inter-
at position 435 were slightly less stable (Figure S6). Thus, the actions with b strands and the hinge loop (such as I428, V436,
simulations of single-point mutants of PYK2 did not provide clear A452, V484, and L501) lining the pocket entrance were less sta-
indicators of differences in the structural behavior of FAK ble than in FAK. This suggested that the protein-ligand interac-
and PYK2.
tions, particularly with b1 to b4, were less optimal in PYK2 rela-
We next investigated whether the helix-like conformation of tive to FAK even if the helical conformation of the DFG motif
the DFG-motif region would be unfavorable in apo-PYK2. For was present in both structures.
this purpose, we built a ‘‘chimeric’’ PYK2 structure with the heli-
We further explored the effect of the structural differences of
cal DFG-motif region mirroring the one observed in FAK com- the PYK2 and FAK binding sites by making 12 constructs with
plexes with slowly dissociating compounds (see STAR point mutations and subsequently performing NanoBRET target
methods). We then performed a series of MD simulations engagement assays and washout experiments (Figures 6F–6H
(ꢂ1 ms in total) for FAK and PYK2 (both with the helical DFG and S5; Table S3). First, we tested the interaction of the inhibitors
motif). The root-mean-square deviation (RMSD) of all structural with FAK-L567 by mutating this residue to an alanine and hy-
elements (the helical DFG motif, b1, and aC) of the binding site pothesized a change from slow to fast off-rate as the missing hy-
˚
was quite low in the unbound FAK (<2 A, Figure S6), which indi- drophobic interaction with the FAK protein. Indeed, the binding
cated that the DFG-helical form in FAK was relatively stable even affinity of the inhibitor series was lower for FAK L567A and, grat-
in the absence of a ligand. Surprisingly, the hydrogen bond be- ifyingly, the mutation changed the off-rates from slow to fast,
tween G566 and Q432 was not preserved and appeared on indicating a direct correlation of this residue and helix formation
average in only half of the frames (Figure 6E). Thus, the presence with dissociation kinetics (Figure 6G). To test the role of the polar
of Q432 in FAK instead of F435 in PYK2, while important, may not interaction of the oxindole and quinoline-2-ones with FAK R426,
be the only factor that determines the DFG-motif conformation. we mutated it to alanine. In agreement with computational and
In contrast to FAK, the binding site in the ‘‘chimeric’’ structure of crystallographic analysis, the potency did not change for the
PYK2 was flexible and underwent considerable reconstruction FAK R426A mutant, and we did not observe differences in the
˚
(backbone RMSD of about 3 A for the helical segment containing binding kinetics (Figure S5 and Table S3).
the DFG motif, Figure S6). Specifically, the aC moved away from
In the crystal structure, we observed a hydrogen bond between
the ATP-binding pocket toward a position similar to that in the FAK PYK2 E509 and the pyridine nitrogen of 1 (Figure 1), but this was
structure (Figure S6), and b1 slightly changed its conformation. not present for the equivalent FAK E506. To try to establish an
However, its first strand did not adopt the conformation observed interaction between the ligand aromatic ring system and FAK,
in FAK, i.e., distorted in the middle (PYK2 and FAK are compared we mutated FAK E506 to glutamine to test for an electron donor,
in Figure 6C). Instead, it preserved the well-defined b-sheet struc- and to isoleucine to test for a hydrophobic interaction at this po-
ture of the original PYK2 conformation. As mentioned above, the sition. Both FAK E506Q and FAK E506I showed decreased po-
difference in the b-sheet structure in PYK2 and FAK is likely tency for the series of inhibitors, indicating this interaction to be
caused by two residues: G425 and C427 in FAK are replaced by unfavorable in FAK (Figure S5 and Table S3). The P loop of
larger residues, N428 and I430 in PYK2, respectively. As a result, PYK2 is stabilized by a polar interaction of E432 side chain
in the ‘‘chimeric’’ PYK2 structure, b1 and the following P loop were (PYK2) with the backbone of G437. In FAK, the corresponding
shifted toward the helix of the DFG motif and the following A-loop residue E430 pointed toward the solvent and did not contribute
˚
region by about 4 A. This shift led to the repositioning of L570 to a stabilized P loop. Using FAK-E430S and FAK-E430Q mu-
as shown in Figure 6C, where FAK bound to compound 1 is tants, we investigated whether we can establish the same P-
compared with the ‘‘chimeric’’ PYK2 structure after about loop rigidity as in PYK2, preventing the DFG helix from forming.
200 ns of equilibration. Finally, hydrogen bonds between the heli- However, the mutant binding kinetics were similar to the wild-
cal DFG-motif fragment and the rest of the protein were different in type FAK binding kinetics (Figure S5 and Table S3). This result
‘‘chimeric’’ PYK2 relative to FAK. Apart from the missing interac- is also in accordance with the MD simulations showing that the
tion with G569 (hydrogen bond between G566 and R432 in FAK), hydrogen bond between G566 and R432 was not strictly pre-
the interactions of S571 and D572 were also less stable in PYK2 served. As we observed a striking difference in the sequence of
(Figure 6E). All these results clearly indicate that, even with the FAK and PYK2 in PYK2-F435 (versus FAK Q432), we tested the
same helical DFG motif, the apo structure and stability of the bind- PYK2 mutants F435Q and F435R. While the mutants did not
ing site in PYK2 differs from that in FAK. show a change in potency, we did observe a change to slower
To assess the effect of the helical conformation of the DFG- binding kinetics for PYK2 F435Q. As E474 is located in helix aC
motif loop on the ligand unbinding rate, we modeled compound in PYK2 and rigidifies the protein, we mutated E474 to glutamine,
10 Cell Chemical Biology 28, 1–13, May 20, 2021

Please cite this article in press as: Berger et al., Structure-kinetic relationship reveals the mechanism of selectivity of FAK inhibitors over PYK2, Cell
Chemical Biology (2021), https://doi.org/10.1016/j.chembiol.2021.01.003
ll
Article
B Lead contact
which resulted in decreased potency but similar binding kinetics
to wild-type PYK2 (Figure S5 and Table S3).
B Material availability
B Data and code availability
d EXPERIMENTAL MODEL AND SUBJECT DETAILS
B Microbe strains
Lastly, we tested a potential involvement of protein phosphor-
ylation in binding and kinetics by mutating the respective Y576
and Y577 in FAK and Y579 and Y580 in PYK2 to phenylalanine.
While in FAK we did not observe a change in potency, the differ-
ence compared with PYK2 wild type was a 2-fold increase in po-
tency. However, these mutations had no impact on binding
kinetics. In summary, for PYK2, none of the mutants showed a
change from fast to slow dissociation for this series of inhibitors.
As a representative for the inhibitor set, we chose 10, for which
only fast kinetics was observed for all PYK2 mutants (Figure 6H).
Based on both the mutagenesis study and our MD simulations,
we concluded that it is not one single amino acid residue change
that enables FAK to form a stable helical DFG motif in complexes.
B Cell lines
d METHOD DETAILS
B Compound chemistry
B Protein expression and purification
B Crystallization and structure determination
B Surface plasmon resonance
B NanoBRET cellular assays
B NanoBRET plasmid mutation
B Computations
d QUANTIFICATION AND STATISTICAL ANALYSIS
B Surface-plasmon-resonance data
B NanoBRET data
SIGNIFICANCE
SUPPLEMENTAL INFORMATION
In this study, we explored the structural mechanism of
aminopyrimidine inhibitor binding, confirming existing FAK
structure-activity relationships and providing a structure-
kinetic-relationship study for FAK. We observed slowly disso-
ciating inhibitors to induce a helical DFG-motif conformation
in FAK and showed that the gained selectivity of inhibitor
binding to FAK and PYK2 was based on a change in their un-
binding rates driven by several sequence dissimilarities in
their binding sites. This study demonstrates that small varia-
tions in the sequence of the protein binding site may trigger
different mechanisms of ligand-protein interactions and
result in ligand-induced stabilization of distinct protein con-
formations that affect ligand-binding kinetics. We demon-
strated that a combination of structural data and in silico
methods can be used to predict compound residence times
and to obtain insight into the molecular determinants
affecting kinetics of inhibitors. We have shown here that the
NanoBRET washout technique provides a unique opportunity
to explore target engagement and kinetic behavior in living
cells. The kinetic data translated into the in cellulo context
of a NanoBRET washout assay and validated the PF-431396
(10) surface plasmon resonance-based residence time of
138 min in living cells. Nevertheless, this cellular residence
time of PF-431396 (10) was still too short to significantly affect
the pharmacology of 10, indicating that future drug-discovery
efforts are necessary to optimize ligand kinetics in order to
affect clinical efficacy. Indeed, this aminopyrimidine series
was developed into a macrocyclic inhibitor that showed bet-
ter selectivity for PYK2 (Farand et al., 2016). The hydrophobic
interaction of inhibitors with the helical activation loop in FAK
should be included in future optimization processes for the
development of inhibitors showing prolonged residence
time for FAK.
Supplemental Information can be found online at https://doi.org/10.1016/j.
chembiol.2021.01.003.
ACKNOWLEDGMENTS
This work was supported by a Capes-Humboldt postdoctoral scholarship
to A.N.-A. (Capes process number 88881.162167/2017-01), the European
Union’s Horizon 2020 Framework Program for Research and Innovation un-
der the Specific Grant Agreement no. 785907 and no. 945539 (Human
Brain Project SGA2 and SGA3 to R.C.W.), EU/EFPIA Innovative Medicines
Initiative Joint Undertaking, K4DD (grant no. 115366 to R.C.W. and S.K.)
and the Klaus Tschira Foundation (D.B.K., A.N.-A., R.C.W.). B.-T.B.,
M.S., J.M.E., D.B.K., and S.K. are supported by the Innovative Medicines
Initiative 2 Joint Undertaking under grant agreement no. 875510 and the
SGC, a registered charity (number 1097737) that receives funds from Abb-
Vie, Bayer AG, Boehringer Ingelheim, Canada Foundation for Innovation,
Eshelman Institute for Innovation, Genentech, Genome Canada through
Ontario Genomics Institute [OGI-196], EU/EFPIA/OICR/McGill/KTH/Dia-
mond, Innovative Medicines Initiative 2 Joint Undertaking (EUbOPEN grant
875510), Janssen, Merck KGaA (a.k.a. EMD in Canada and USA), Merck
(a.k.a. MSD outside Canada and USA), Pfizer, Sa˜ o Paulo Research Foun-
dation-FAPESP, Takeda, and Wellcome. B.-T.B. and S.K. would like to
acknowledge funding by the DFG-funded CRC SFB 1399.
AUTHOR CONTRIBUTIONS
S.K., S.M., M.F., J.B., J.M.E., and R.C.W. supervised the research. T.H. syn-
thesized the inhibitors. B.-T.B, M.S., D.M., and M.A. performed structural
studies. B.-T. B., M.A., I.N., and J.W. performed biochemical and biophysical
binding studies. B.-T.B. and M.A. performed NanoBRET cell-based ligand
studies. D.B.K., A.N.-A., R.N., and R.C.W. performed in silico analysis. S.K.,
M.A., D.B.K., and B-T.B. drafted the manuscript, which has been edited and
approved by all authors.
DECLARATION OF INTERESTS
M.F., J.B., and T.H. are employees of Merck KGaA, Darmstadt, Germany. M.A.
was an employee of Merck KGaA, Darmstadt, Germany at the time of exper-
imental procedures and is now an employee of Sanofi-Aventis Deutschland
GmbH, Frankfurt, Germany. Other authors have no conflict of interest to
declare.
STAR+METHODS
Detailed methods are provided in the online version of this paper
and include the following:
Received: July 16, 2020
Revised: December 14, 2020
Accepted: January 4, 2021
Published: January 25, 2021
d KEY RESOURCES TABLE
d RESOURCE AVAILABILITY
Cell Chemical Biology 28, 1–13, May 20, 2021 11

Please cite this article in press as: Berger et al., Structure-kinetic relationship reveals the mechanism of selectivity of FAK inhibitors over PYK2, Cell
Chemical Biology (2021), https://doi.org/10.1016/j.chembiol.2021.01.003
ll
Article
REFERENCES
second-generation focal adhesion kinase (FAK) inhibitor, in patients with
advanced solid tumors. J. Clin. Oncol. 29, 3002.
Allen, J.G., Lee, M.R., Han, C.-Y.E., Scherrer, J., Flynn, S., Boucher, C., Zhao,
H., O’Connor, A.B., Roveto, P., Bauer, D., et al. (2009). Identification of small
molecule inhibitors of proline-rich tyrosine kinase 2 (Pyk2) with osteogenic ac-
tivity in osteoblast cells. Bioorg. Med. Chem. Lett. 19, 4924–4928.
Jorgensen, W., Chandrasekhar, J., and Madura, J.D. (1983). Comparison of
simple potential functions for simulating liquid water. J. Chem. Phys. 79, 926.
Kabsch, W. (2010). XDS. Acta Crystallogr. D Biol. Crystallogr. 66, 125–132.
Kath, J.C., and Luzzio, M.J. (2003). Pyrimidine derivates for the treatment of
Arcari, J.T., Bhattacharya, S.K., Brosius, A.D., Luzzio, M.J., Nelson, K.L., Pan,
G., Southers, J.A., Jr., Wishka, D.G., and Xiao, J. (2006). Pyrimidine derivatives
for the treatment of abnormal cell growth, WIPO patent application publication
WO2007072158A2, filed 11 December, 2006 and published 28 June, 2007.
abnormal
cell
growth,
WIPO
patent
application
publication
WO2004056786A2, filed Decemmber 17, 2003 and published August 7, 2004.
€
Kokh, D.B., Amaral, M., Bomke, J., Gradler, U., Musil, D., Buchstaller, H.-P.,
Dreyer, M.K., Frech, M., Lowinski, M., Vallee, F., et al. (2018). Estimation of
drug-target residence times by t-random acceleration molecular dynamics
simulations. J. Chem. Theor. Comput. 14, 3859–3869.
Avraham, H., Park, S.-Y., Schinkmann, K., and Avraham, S. (2000). RAFTK/
Pyk2-mediated cellular signalling. Cell Signal. 12, 123–133.
Bricogne, G., Blanc, E., Brandl, M., Flensburg, C., Keller, P., Paciorek, W.,
Roversi, P., Sharff, A., Smart, O., and Vonrhein, C. (2016). BUSTER Version
XYZ (Global Phasing Ltd.).
Kokh, D.B., Doser, B., Richter, S., Ormersbach, F., Cheng, X., and Wade, R.C.
(2020). A workflow for exploring ligand dissociation from a macromolecule:
efficient random acceleration molecular dynamics simulation and interaction
fingerprints analysis of ligand trajectories. J. Chem. Phys. 153, 125102.
Buckbinder, L., Crawford, D.T., Qi, H., Ke, H.Z., Olson, L.M., Long, K.R.,
Bonnette, P.C., Baumann, A.P., Hambor, J.E., and Grasser, W.A. (2007).
Proline-rich tyrosine kinase 2 regulates osteoprogenitor cells and bone forma-
tion, and offers an anabolic treatment approach for osteoporosis. Proc. Natl.
Acad. Sci. U S A 104, 10619–10624.
Krissinel, E., and Henrick, K. (2007). Inference of macromolecular assemblies
from crystalline state. J. Mol. Biol. 372, 774–797.
Lietha, D., and Eck, M.J. (2008). Crystal structures of the FAK kinase in com-
plex with TAE226 and related bis-anilino pyrimidine inhibitors reveal a helical
DFG conformation. PLoS One 3, e3800.
Case, D., Berryman, J., Betz, R., Cerutti, D., Cheatham, T., III, Darden, T.,
Duke, R., Giese, T., Gohlke, H., and Goetz, A. (2015). AMBER 2015 (UCSF).
Lipinski, C.A., and Loftus, J.C. (2010). Targeting Pyk2 for therapeutic interven-
Cheng, Y.-C., and Prusoff, W.H. (1973). Relationship between the inhibition
constant (K1) and the concentration of inhibitor which causes 50 per cent inhi-
bition (I50) of an enzymatic reaction. Biochem. Pharmacol. 22, 3099–3108.
tion. Expert Opin. Ther. Targets 14, 95–108.
Lo´ pez-Blanco, J.R., Canosa-Valls, A.J., Li, Y., and Chaco´ n, P. (2016). RCD+:
fast loop modeling server. Nucleic Acids Res. 44, W395–W400.
Copeland, R.A. (2016). The drug-target residence time model: a 10-year retro-
€
Ludemann, S.K., Lounnas, V., and Wade, R.C. (2000). How do substrates enter
spective. Nat. Rev. Drug Discov. 15, 87.
and products exit the buried active site of cytochrome P450cam? 2. Steered
molecular dynamics and adiabatic mapping of substrate pathways. J. Mol.
Biol. 303, 813–830.
Cornell, W.D., Cieplak, P., Bayly, C.I., and Kollman, P.A. (2002). Application of
RESP charges to calculate conformational energies, hydrogen bond energies,
and free energies of solvation. J. Am. Chem. Soc. 115, 9620–9631.
Luzzio, M.J., Autry, C.L., Bhattacharya, S.K., Freeman-Cook, K.D., Hayward,
M.M., Hulford, C.A., Nelson, K.L., Xiao, J., and Zhao, X. (2008). Sulfonyl amide
derivatives for the treatment of abnormal cell growth, WIPO patent application
publication WO2008129380A1, filed April 7, 2008 and published October
30, 2008.
Crooks, G.E., Hon, G., Chandonia, J.-M., and Brenner, S.E. (2004). WebLogo:
a sequence logo generator. Genome Res. 14, 1188–1190.
Dolinsky, T.J., Czodrowski, P., Li, H., Nielsen, J.E., Jensen, J.H., Klebe, G.,
and Baker, N.A. (2007). PDB2PQR: expanding and upgrading automated
preparation of biomolecular structures for molecular simulations. Nucleic
Acids Res. 35, W522–W525.
Lv, P.C., Jiang, A.Q., Zhang, W.M., and Zhu, H.L. (2018). FAK inhibitors in can-
cer, a patent review. Expert Opin. Ther. Pat. 28, 139–145.
Dolinsky, T.J., Nielsen, J.E., McCammon, J.A., and Baker, N.A. (2004).
PDB2PQR: an automated pipeline for the setup of Poisson-Boltzmann electro-
statics calculations. Nucleic Acids Res. 32, W665–W667.
Maier, J.A., Martinez, C., Kasavajhala, K., Wickstrom, L., Hauser, K.E., and
Simmerling, C. (2015). ff14SB: improving the accuracy of protein side chain
and backbone parameters from ff99SB. J. Chem. Theor. Comput. 11,
3696–3713.
Emsley, P. (2017). Tools for ligand validation in Coot. Acta Crystallogr. D
Struct. Biol. 73, 203–210.
Marks, C., Nowak, J., Klostermann, S., Georges, G., Dunbar, J., Shi, J., Kelm,
S., and Deane, C.M. (2017). Sphinx: merging knowledge-based and ab initio
approaches to improve protein loop prediction. Bioinformatics 33, 1346–1353.
Fan, H., and Guan, J.-L. (2011). Compensatory function of Pyk2 protein in the
promotion of focal adhesion kinase (FAK)-null mammary cancer stem cell
tumorigenicity and metastatic activity. J. Biol. Chem. 286, 18573–18582.
Matsuhisa, A., Kinoyama, I., Toyoshima, A., Nakahara, T., Takeuchi, M., and
Okada, M. (2001). Fused imidazolium derivatives, US patent application pub-
lication US20030114508A1, filed February 14, 2001 and published June
19, 2003.
Farand, J., Mai, N., Chandrasekhar, J., Newby, Z.E., Van Veldhuizen, J., Loyer-
Drew, J., Venkataramani, C., Guerrero, J., Kwok, A., Li, N., et al. (2016).
Selectivity switch between FAK and Pyk2: macrocyclization of FAK inhibitors
improves Pyk2 potency. Bioorg. Med. Chem. Lett. 26, 5926–5930.
Phillips, J.C., Braun, R., Wang, W., Gumbart, J., Tajkhorshid, E., Villa, E.,
Chipot, C., Skeel, R.D., Kale, L., and Schulten, K. (2005). Scalable molecular
dynamics with NAMD. J. Comput. Chem. 26, 1781–1802.
Gordon, M.S., and Schmidt, M.W. (2005). Advances in electronic structure the-
ory: GAMESS a decade later. In Theory and Applications of Computational
Chemistry: The First Forty Years, C.E. Dykstra, G. Frenking, K.S. Kim, and
G.E. Scuseria, eds. (Elsevier), pp. 1167–1189.
Robers, M.B., Dart, M.L., Woodroofe, C.C., Zimprich, C.A., Kirkland, T.A.,
Machleidt, T., Kupcho, K.R., Levin, S., Hartnett, J.R., Zimmerman, K., et al.
(2015). Target engagement and drug residence time can be observed in living
cells with BRET. Nat. Commun. 6, 10091.
Han, S., Mistry, A., Chang, J.S., Cunningham, D., Griffor, M., Bonnette, P.C.,
Wang, H., Chrunyk, B.A., Aspnes, G.E., Walker, D.P., et al. (2009). Structural
characterization of proline-rich tyrosine kinase 2 (PYK2) reveals a unique
(DFG-out) conformation and enables inhibitor design. J. Biol. Chem. 284,
13193–13201.
Roberts, W.G., Ung, E., Whalen, P., Cooper, B., Hulford, C., Autry, C., Richter,
D., Emerson, E., Lin, J., Kath, J., et al. (2008). Antitumor activity and pharma-
cology of a selective focal adhesion kinase inhibitor, PF-562,271. Cancer Res.
68, 1935–1944.
Heinrich, T., Seenisamy, J., Emmanuvel, L., Kulkarni, S.S., Bomke, J.,
Rohdich, F., Greiner, H., Esdar, C., Krier, M., Gradler, U., et al. (2013).
Fragment-based discovery of new highly substituted 1H-pyrrolo[2,3-b]- and
3H-imidazolo[4,5-b]-pyridines as focal adhesion kinase inhibitors. J. Med.
Chem. 56, 1160–1170.
¨ ¨
Rohm, S., Berger, B.-T., Schroder, M., Chaikuad, A., Winkel, R., Hekking, K.F.,
€
Benningshof, J.J., Muller, G., Tesch, R., and Kudolo, M. (2019). Fast iterative
synthetic approach toward identification of novel highly selective p38 MAP ki-
nase inhibitors. J. Med. Chem. 62, 10757–10782.
Jones, S., Shapiro, G., Bendell, J., Chen, E., Bedard, P., Cleary, J., Pandya, S.,
Pierce, K., Houk, B., and Hosea, N. (2011). Phase I study of PF-04554878, a
Schlaepfer, D.D., Hauck, C.R., and Sieg, D.J. (1999). Signaling through focal
adhesion kinase. Prog. Biophys. Mol. Biol. 71, 435–478.
12 Cell Chemical Biology 28, 1–13, May 20, 2021

Please cite this article in press as: Berger et al., Structure-kinetic relationship reveals the mechanism of selectivity of FAK inhibitors over PYK2, Cell
Chemical Biology (2021), https://doi.org/10.1016/j.chembiol.2021.01.003
ll
Article
Schuetz, D.A., Witte, W.E.A.d., Wong, Y.C., Knasmueller, B., Richter, L., Kokh,
D.B., Sadiq, S.K., Bosma, R., Nederpelt, I., and Heitman, L.H. (2017). Kinetics
for Drug Discovery: an industry-driven effort to target drug residence time.
Drug Discov. Today 22, 896–911.
Quantitative, wide-spectrum kinase profiling in live cells for assessing the ef-
fect of cellular ATP on target engagement. Cell Chem. Biol. 25, 206–214 e211.
Vriend, G. (1990). What IF: a molecular modeling and drug design program.
J. Mol. Graph. 8, 52–56.
Slack-Davis, J.K., Martin, K.H., Tilghman, R.W., Iwanicki, M., Ung, E.J., Autry,
C., Luzzio, M.J., Cooper, B., Kath, J.C., Roberts, W.G., et al. (2007). Cellular
characterization of a novel focal adhesion kinase inhibitor. J. Biol. Chem.
282, 14845–14852.
Walker, D.P., Bi, F.C., Kalgutkar, A.S., Bauman, J.N., Zhao, S.X., Soglia, J.R.,
Aspnes, G.E., Kung, D.W., Klug-McLeod, J., Zawistoski, M.P., et al. (2008).
Trifluoromethylpyrimidine-based inhibitors of proline-rich tyrosine kinase 2
(PYK2): structure-activity relationships and strategies for the elimination of
reactive metabolite formation. Bioorg. Med. Chem. Lett. 18, 6071–6077.
Tai, Y.-L., Chen, L.-C., and Shen, T.-L. (2015). Emerging roles of focal adhesion
Walker, D.P., Zawistoski, M.P., McGlynn, M.A., Li, J.-C., Kung, D.W.,
Bonnette, P.C., Baumann, A., Buckbinder, L., Houser, J.A., and Boer, J.
(2009). Sulfoximine-substituted trifluoromethylpyrimidine analogs as inhibitors
of proline-rich tyrosine kinase 2 (PYK2) show reduced hERG activity. Bioorg.
Med. Chem. Lett. 19, 3253–3258.
kinase in cancer. Biomed. Res. Int. 2015, 690690.
Trott, O., and Olson, A.J. (2010). AutoDock Vina: improving the speed and ac-
curacy of docking with a new scoring function, efficient optimization, and mul-
tithreading. J. Comput. Chem. 31, 455–461.
Tse, K.W.K., Lin, K.B.L., Dang-Lawson, M., Guzman-Perez, A., Aspnes, G.E.,
Buckbinder, L., and Gold, M.R. (2012). Small molecule inhibitors of the Pyk2
and FAK kinases modulate chemoattractant-induced migration, adhesion
and Akt activation in follicular and marginal zone B cells. Cell. Immunol.
275, 47–54.
Wang, J., Wolf, R.M., Caldwell, J.W., Kollman, P.A., and Case, D.A. (2004).
Development and testing of a general amber force field. J. Comput. Chem.
25, 1157–1174.
Wendt, M.K., Schiemann, B.J., Parvani, J.G., Lee, Y.-H., Kang, Y., and
Schiemann, W.P. (2013). TGF-b stimulates Pyk2 expression as part of an
epithelial-mesenchymal transition program required for metastatic outgrowth
of breast cancer. Oncogene 32, 2005–2015.
Vasta, J.D., Corona, C.R., Wilkinson, J., Zimprich, C.A., Hartnett, J.R., Ingold,
M.R., Zimmerman, K., Machleidt, T., Kirkland, T.A., Huwiler, K.G., et al. (2018).
Cell Chemical Biology 28, 1–13, May 20, 2021 13

Please cite this article in press as: Berger et al., Structure-kinetic relationship reveals the mechanism of selectivity of FAK inhibitors over PYK2, Cell
Chemical Biology (2021), https://doi.org/10.1016/j.chembiol.2021.01.003
ll
Article
STAR+METHODS
KEY RESOURCES TABLE
REAGENT or RESOURCE
Bacterial and virus strains
SOURCE
Novagen
IDENTIFIER
Cat#71400
Escherichia coli Rosettaꢀ 2(DE3)
Chemicals, peptides, and recombinant proteins
1 – PF-562271
This study
Roberts et al., 2008
Arcari et al., 2006
2
This study
Matsuhisa et al., 2001
Arcari et al., 2006
Arcari et al., 2006
Slack-Davis et al., 2007
Arcari et al., 2006
n/a
3
This study
4
This study
5 – PF-573228
This study
6
This study
7
This study
8
This study
Arcari et al., 2006
n/a
9
This study
10 – PF-431396
This study
Buckbinder et al., 2007
Jones et al., 2011
Tse et al., 2012
Cat# 11960044
Cat#10500064
Cat#15070063
Cat# 11058021
Cat# 600675
11 – PF8554878 (defactinib)
12 – PF-719
This study
This study
Gibcoꢀ DMEM
Invitrogen
Gibcoꢀ Heat inactivated FBS
Gibcoꢀ penicillin-streptavidin antibiotic
Gibcoꢀ OPTI-MEM
HercII polymerase kit
ATP
Invitrogen
Invitrogen
Invitrogen
Agilent
Sigma Aldrich
Molecular Dimension
Molecular Dimension
Instituto de Biologia
Cat# A9187-100MG
Cat# MD2-100-48
Cat# MD2-100-9
n/a
Magnesium formate dihydrate
PEG3350
Recombinant FAK (amino acid 410-689 with
P410G mutation)
´
Experimental e Tecnologica
Recombinant PYK2 (amino acid 420-691)
This study
n/a
Critical commercial assays
Surface Plasmon Resonance CM5 kit
NanoBRET Target engagement K5 kit
GE HealthCare
Promega
Cat# 29104988
Cat# N2500
Deposited data
FAK in complex with 1
PYK2 in complex with 1
PYK2 in complex with 10
PYK2 in complex with 12
FAK in complex with 2
FAK in complex with 3
FAK in complex with 4
FAK in complex with 5
FAK in complex with 8
FAK in complex with 10
Roberts et al., 2008
Farand et al., 2016
Han et al., 2009
Walker et al., 2009
This study
pdb id: 3bz3
pdb id: 5tob
pdb id: 3fzr
pdb id: 3h3c
pdb id: 6yt6
pdb id: 6yxv
pdb id: 6yq1
pdb id: 6yoj
pdb id: 6yvs
pdb id: 6yr9
This study
This study
This study
This study
This study
Experimental models: cell lines
HEK293T
ATCC
Cat# CRL3216ꢀ
(Continued on next page)
e1 Cell Chemical Biology 28, 1–13.e1–e7, May 20, 2021

Please cite this article in press as: Berger et al., Structure-kinetic relationship reveals the mechanism of selectivity of FAK inhibitors over PYK2, Cell
Chemical Biology (2021), https://doi.org/10.1016/j.chembiol.2021.01.003
ll
Article
Continued
REAGENT or RESOURCE
SOURCE
Eurofins
IDENTIFIER
Oligonucleotides
Mutagenesis primer pairs are listed in Table S4
https://www.eurofins.com/
Recombinant DNA
FAK
Merck KgaA
Merck KgaA
Promega
n/a
PYK2
n/a
NLuc-FAK
Cat# NV1921
PYK2-NLuc
Promega
Cat# NV1931
NLuc-FAK-R426S
NLuc-FAK-R426A
NLuc-FAK-L567A
NLuc-FAK-E506Q
NLuc-FAK-E506I
NLuc-FAK-E430S
NLuc-FAK-E430Q
NLuc-FAK-Y576F+Y577F
PYK2-F435R-NLuc
PYK2-E474Q-NLuc
PYK2-F435Q-NLuc
PYK2-Y579F+Y580F-NLuc
This study
This study
This study
This study
This study
This study
This study
This study
This study
This study
This study
This study
n/a
n/a
n/a
n/a
n/a
n/a
n/a
n/a
n/a
n/a
n/a
n/a
Software and algorithms
XDS
Kabsch, 2010
http://xds.mpimf-heidelberg.mpg.de/
PISA web server
Krissinel and Henrick 2007
https://www.ebi.ac.uk/msd-srv/prot_int/cgi-
bin/piserver
BUSTER
COOT
Bricogne et al., 2016
Emsley, 2017
https://www.globalphasing.com/buster/
https://www2.mrc-lmb.cam.ac.uk/personal/
pemsley/coot/
Prism
Graphpad
https://www.graphpad.com/scientific-software/
prism/
WHATIF web server
Vriend, 1990
https://swift.cmbi.umcn.nl/servers/html/index.html
http://server.poissonboltzmann.org/
http://vina.scripps.edu/
Pdb2pqr web server
Dolinsky et al., 2007
Trott and Olson, 2010
Marks et al., 2017
Gordon and Schmidt, 2005
Case et al., 2015
AutoDock Vina
Sphinx web server
https://pypi.org/project/sphinx-server/
https://www.msg.chem.iastate.edu/gamess/
https://ambermd.org/
GAMESS ab initio quantum chemistry package
AMBER15 MD simulation software
NAMD MD simulation software
RAMD tcl script
Phillips et al., 2005
https://www.ks.uiuc.edu/Research/namd/
https://www.h-its.org/downloads/ramd/
https://github.com/HITS-MCM/MD-IFP
http://rcd.chaconlab.org/
€
Ludemann et al., 2000
MD-IFP Python tools
Kokh et al., 2020
´
Lopez-Blanco et al., 2016
RCD+ web server
WebLogo web Server
Crooks et al., 2004
https://weblogo.berkeley.edu/
RESOURCE AVAILABILITY
Lead contact
Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Stefan
Knapp (knapp@pharmchem.uni-frankfurt.de).
Material availability
Requests for plasmids generated in this study should be directed to the lead contact. Inhibitors used in this study can be obtained
from commercial sources or for inhibitors generated in this study requests can be directed to Timo Heinrich.
Cell Chemical Biology 28, 1–13.e1–e7, May 20, 2021 e2

Please cite this article in press as: Berger et al., Structure-kinetic relationship reveals the mechanism of selectivity of FAK inhibitors over PYK2, Cell
Chemical Biology (2021), https://doi.org/10.1016/j.chembiol.2021.01.003
ll
Article
Data and code availability
The coordinates and structure factors of all complexes have been deposited to the protein data bank under accession codes: 6YT6,
6YXV, 6YQ1, 6YOJ, 6YVS, 6YR9, 6YVY.
EXPERIMENTAL MODEL AND SUBJECT DETAILS
Microbe strains
Escherichia coli, BL21 (DE3) strain was used for recombinant protein expression. Cells were cultured in terrific broth media at 37ꢁC
until they reached an OD600 of 2.6-3.0 prior to induction with 0.5 mM IPTG for protein expression at 18ꢁC overnight.
Cell lines
HEK293T (female) cells were obtained from ATCC (identifier CRL-3216). Cells were cultured at 37ꢁC/5% CO2 in DMEM (Gibco) sup-
plemented with 10% heat inactivated FBS (Invitrogen) and penicillin-streptavidin antibiotic (Gibco).
METHOD DETAILS
Compound chemistry
General synthesis of diaminopyrimidines 1 -6, 8, 10 - 12
The synthetic steps were performed as described in the patent literature (Arcari et al., 2006; Kath and Luzzio, 2003; Luzzio et al., 2008;
Matsuhisa et al., 2001) and are described in detail below.
Synthesis of N-Methyl-N-(3-{[2-(2-oxo-2,3-dihydro-1H-indol-5-ylamino)-5-trifluoromethyl-pyrimidin-4-ylamino]-methyl}-pyridin-2-
yl)-methanesulfonamide (PF- 562271) (1). The reaction of 5-trifluoromethyl-2,4-dichloropyrimidine and 5-amino-oxindole in a tert
butanol/dichloroethane = 1:1 and zinc chloride yielded the intermediate 5-{[4-chloro-5-(trifluoromethyl)pyrimidin-2-yl]amino}-2,3-di-
hydro-1H-indol-2-one with 59%. This compound reacted with N-[3-(aminomethyl)pyridin-2-yl]-N-methylmethanesulfonamide dihy-
drochloride in DMF and 2.5 eq of sodium carbonate at 85ꢁC over night to the final product in 36% yield. The yield for the final step was
optimized by reaction in toluene/isopropanol = 1:1 in the presence of 1.2 eq of diisopropylethylamine to 72%.
1H NMR (500 MHz, DMSO-d6) d 10.16 (s, 2H), 9.37 (s, 1H), 8.44 (dd, J = 4.7, 1.8 Hz, 3H), 8.21 (s, 3H), 7.67 (d, J = 8.0 Hz, 3H), 7.50 –
7.44 (m, 3H), 7.42 (dd, J = 7.8, 4.7 Hz, 3H), 7.33 (s, 2H), 7.19 (d, J = 8.3 Hz, 2H), 6.54 (d, J = 8.3 Hz, 2H), 4.77 (d, J = 5.8 Hz, 5H), 4.03 (q,
J = 7.1 Hz, 2H), 3.15 (s, 9H), 3.11 (s, 7H), 1.99 (s, 3H), 1.18 (t, J = 7.1 Hz, 3H).
Synthesis of N-Methyl-N-(3-{[2-(2-oxo-2,3-dihydro-1H-indol-5-ylamino)-pyrimidin-4-ylamino]-methyl}-pyridin-2-yl)-methanesulfo-
namide (2). The intermediate N-(3-{[(2-chloropyrimidin-4-yl)amino]methyl}pyridin-2-yl)-N-methylmethanesulfonamide was ob-
tained in 27% yield by reaction of 2,4-dichloropyrimidine and N-[3-(aminomethyl)pyridin-2-yl]-N-methylmethanesulfonamide in
the microwave at 120ꢁC for 20 minutes in the presence of 2 eq of triethyl amine. The final product was obtained after reaction of
the intermediate with 5-aminoindole under identical conditions in 30% yield as brown solid.
1H NMR (500 MHz, DMSO-d6) d 10.11 (s, 2H), 8.76 (s, 2H), 8.45 (dd, J = 4.7, 1.9 Hz, 2H), 8.15 (s, 3H), 7.85 – 7.79 (m, 4H), 7.56 (s, 2H),
7.52 (s, 2H), 7.44 (dd, J = 7.7, 4.7 Hz, 2H), 7.39 – 7.33 (m, 2H), 6.59 (d, J = 8.4 Hz, 2H), 5.99 – 5.95 (m, 1H), 4.65 (d, J = 5.9 Hz, 3H), 3.17
(d, J = 7.1 Hz, 13H).
Synthesis of N-Methyl-N-{3-[(2-phenylamino-5-trifluoromethyl-pyrimidin-4-ylamino)-methyl]-pyridin-2-yl}-methanesulfonamide
(3). The intermediate 4-chloro-N-phenyl-5-(trifluoromethyl)pyrimidin-2-amine was obtained in 58 % yield by reaction of 5-trifluor-
omethyl-2,4-dichloropyrimidine and aniline in dichloromethane/tert-butanol = 1:1 in the presence of stochiometric amounts of zing
chloride-diethylether complex and triethyl amine at RT over night. The final product could be isolated in 40% yield by reaction of the
intermediate with N-[3-(aminomethyl)pyridin-2-yl]-N-methylmethanesulfonamide in toluene/isopropanol = 1:1 at 80ꢁC over night and
column chromatography.
1H NMR (400 MHz, DMSO-d6) d 9.54 (s, 1H), 8.44 (dd, J = 4.7, 1.9 Hz, 1H), 8.26 (s, 1H), 7.66 (dd, J = 7.7, 1.9 Hz, 1H), 7.58 (t, J =
5.9 Hz, 1H), 7.42 (dt, J = 7.9, 4.5 Hz, 3H), 7.08 (t, J = 7.7 Hz, 2H), 6.88 (t, J = 7.3 Hz, 1H), 4.81 (d, J = 5.7 Hz, 2H), 3.15 (d, J = 11.9 Hz, 6H).
Synthesis of N-Methyl-N-(3-{[2-(2-oxo-1,2,3,4-tetrahydro-quinolin-6-ylamino)-5-trifluoromethyl-pyrimidin-4-ylamino]-methyl}-pyri-
din-2-yl)-methanesulfonamide (4). Reaction of 5-drifluoromethyl-2,4-dichloropyrimidine and 6-amino-3,4-dihydroquinolin-
2(1H)-one in tert butanol/dichloromethane in the presence of zinc chloride gave the intermediate 6-(4-Chloro)-5-(trifluoro-
methyl)pyrimidine-2-ylamino-3,4-dihydoquinoline-2(1H)-one in 64% yield. This compound reacted to the final product in 31%
yield with N-[3-(aminomethyl)pyridin-2-yl]-N-methylmethanesulfonamide.
1H NMR (500 MHz, DMSO-d6) d 9.89 (s, 1H), 9.45 – 9.33 (m, 1H), 8.43 (dd, J = 4.8, 1.8 Hz, 1H), 8.23 – 8.21 (m, 1H), 7.68 (d, J =
7.7 Hz, 1H), 7.49 (t, J = 5.8 Hz, 1H), 7.41 (dd, J = 7.8, 4.7 Hz, 1H), 7.32 – 7.26 (m, 1H), 7.19 (d, J = 8.5 Hz, 1H), 6.60 (d, J = 8.6 Hz,
1H), 4.79 (d, J = 5.7 Hz, 2H), 3.14 (s, 3H), 3.14 – 3.11 (m, 3H), 2.66 – 2.57 (m, 2H), 2.36 – 2.31 (m, 2H).
Synthesis of 6-[4-(3-Methanesulfonyl-benzylamino)-5-trifluoromethyl-pyrimidin-2-ylamino]-3,4-dihydro-1H-quinolin-2-one (PF-
573228) (5). The intermediate from the preparation of compound 4 reacted with (3-(methylsulfonyl)phenyl) methanamine HCl salt
in NMP in the presence of 3 eq of triethylamine to the final compound in 70% yield.
1H NMR (400 MHz, DMSO-d6) d 9.89 (s, 1H), 9.44 (br. s, 1H), 8.16 (s, 1H), 7.83 – 7.76 (m, 3H), 7.56 – 7.54 (m, 2H), 7.34 (s, 1H), 7.19 (d,
J = 7.3 Hz, 1H), 6.64 (d, J = 8.7 Hz, 1H), 4.70 (d, J = 5.5 Hz, 2H), 3.34 (s, 3H), 2.63 (t, J = 7.3 Hz, 2H), 2.34 (t, J = 11 Hz, 2H).
e3 Cell Chemical Biology 28, 1–13.e1–e7, May 20, 2021

Please cite this article in press as: Berger et al., Structure-kinetic relationship reveals the mechanism of selectivity of FAK inhibitors over PYK2, Cell
Chemical Biology (2021), https://doi.org/10.1016/j.chembiol.2021.01.003
ll
Article
Synthesis of 5-({4-[(2-oxo-2,3-dihydro-1H-indol-5-yl)amino]-5-(trifluoromethyl)pyrimidin-2-yl}amino)-2,3-dihydro-1H-indol-2-one
(6). The intermediate 5-{[4-chloro-5-(trifluoromethyl)pyrimidin-2-yl]amino}-2,3-dihydro-1H-indol-2-one obtained for the synthesis
of 8 was repeated with an access of 5-amino-oxindole and the final product was isolated in 77% after chromatography.
1H NMR (400 MHz, DMSO-d6) d 10.44 (s, 2H), 10.16 (s, 2H), 9.47 (s, 1H), 8.56 (s, 1H), 8.26 (s, 2H), 7.43 (s, 2H), 7.24 – 7.16 (m, 6H),
6.84 (d, J = 8.1 Hz, 2H), 6.55 (d, J = 8.4 Hz, 2H), 3.47 (s, 4H).
Synthesis of 5-{4-[(Pyridin-3-ylmethyl)-amino]-5-trifluoromethyl-pyrimidin-2-ylamino}-1,3-dihydro-indol-2-one (8). To solution of
5-trifluoromethyl-2,4-dichloropyrimidine in 1:1 DCE/tBuOH was added Zinc chloride 1 M solution in ether. After 0.5 hour,
5-amino-oxindole was added followed by triethylamine keeping temperature at 25ꢁC. The reaction was allowed to stir at room tem-
perature overnight, then was concentrated and the product triturated (repeated several times, slightly soluble in MeOH) from meth-
anol as a yellow solid (5-{[4-chloro-5-(trifluoromethyl)pyrimidin-2-yl]amino}-2,3-dihydro-1H-indol-2-one, 65% yield). This intermedi-
ate was dissolved in DMF and 1-(pyridin-3-yl)methanamine and sodium carbonate were added. The mixture reacted for 3 days at RT
and the final product could be obtained after chromatography in 34% yield.
1H NMR (400 MHz, DMSO-d6) d 10.21 (s, 1H), 9.41 (s, 1H), 8.53 (s, 1H), 8.42 (dd, J = 4.9, 1.7 Hz, 1H), 8.17 (d, J = 5.3 Hz, 2H), 7.73 (t,
J = 6.0 Hz, 1H), 7.66 (d, J = 7.6 Hz, 1H), 7.38 (s, 1H), 7.33 (dd, J = 7.8, 4.7 Hz, 1H), 7.24 (d, J = 8.3 Hz, 1H), 6.65 (d, J = 8.4 Hz, 1H), 4.64
(d, J = 5.9 Hz, 2H).
Synthesis of N-Methyl-N-(2-{[2-(2-oxo-2,3-dihydro-1H-indol-5-ylamino)-5-trifluoromethyl-pyrimidin-4-ylamino]-methyl}-phenyl)-
methanesulfonamide (PF-431396) (10). The intermediate 5-{[4-chloro-5-(trifluoromethyl)pyrimidin-2-yl]amino}-2,3-dihydro-1H-in-
dol-2-one prepared for the synthesis of compound 8 was reacted with N-[2-(aminomethyl)phenyl]-N-methylmethanesulfonamide
under identical conditions and yielded the title compound in 45% yield.
1H NMR (400 MHz, DMSO) d 10.24 (s, 1H), 9.75 (br. s, 1H), 8.26 (s, 1H), 7.76 (br. s, 1H), 7.56 – 7.53 (m, 1H), 7.41 – 7.32 (m, 3H), 7.27 –
7.22 (m, 2H), 6.62 – 6.60 (m, 1H), 4.81 – 4.79 (m, 2H), 3.32 – 3.27 (m, 2H), 3.12 – 3.09 (m, 6H).
Synthesis of 4-(4-{[3-(Methanesulfonyl-methyl-amino)-pyrazin-2-ylmethyl]-amino}-5-trifluoromethyl-pyrimidin-2-ylamino)-N-methyl-
benzamide (PF-04554878) (11). The reaction of N-(3-Aminomethyl-pyrazin-2-yl)-N-methyl-methanesulfonamide and 2,4-Di-
chloro-5-trifluoromethyl-pyrimidine in dichloromethane in the presence of diisopropylethylamine from -20ꢁC to 0ꢁC yielded the
intermediate N-{3-[(2-Chloro-5-trifluoromethyl-pyrimidin-4-ylamino)-methyl]-pyrazin-2-yl}-N-methyl-methanesulfonamide in 32%
yield. The final product was obtained by reaction of this intermediate with 4-amino-N-methyl-benzamide in butanol and catalytic
amounts of acetic acid from RT to 100ꢁC in 14% yield.
1H NMR (400 MHz, DMSO) d 9.84 (s, 1H), 8.68-8.67 (d, J = 4 Hz, 1H), 8.57-8.56 (d, J = 4Hz, 1H), 8.30 (s, 1H), 8.20-8.19 (s, J= 4 Hz,
1H), 7.65-7.58 (m, 3H), 7.43 (s, 1H), 4.99-4.98 (d, 2H), 3.21-3.18 (d, 6H), 2.74-2.73 (d, 3H)
General procedure for the preparation of aminopyrimidines 7 and 9
Intermediate 5-{[4-chloro-5-(trifluoromethyl)pyrimidin-2-yl]amino}-2,3-dihydro-1H-indol-2-one from the synthesis of compound 8
was coupled to the corresponding aryl acetylene in the presence of catalytic amounts of bistriphenylphosphine palladiumdichloride,
copper(I)iodide and two equivalents of triethylamine in DMF at 100ꢁC under N2 atmosphere to the corresponding diaryl-acetylene.
The triple bond was reduced on Pd/C contact at room temperature under an atmosphere of H2 for 1h in methanol.
Synthesis of 5-{4-[2-(4-Fluoro-phenyl)-ethyl]-5-trifluoromethyl-pyrimidin-2-ylamino}-1,3-dihydro-indol-2-one (7). The intermediate
5-({4-[2-(4-fluorophenyl)ethynyl]-5-(trifluoromethyl)pyrimidin-2-yl}amino)-2,3-dihydro-1H-indol-2-one was obtained in 92% and the
final product in 75% yield.
1H NMR (400 MHz, DMSO) d 10.31 (s, 1H), 10.07 (s, 1H), 8.61 (s, 1H), 7.58 (s, 1H), 7.48 (d, J = 8.6 Hz, 1H), 7.24 7.24 (dd, J = 8.2 Hz,
5.8 Hz, 2H), 7.10 (t, J = 8.8 Hz, 2H), 6.76 (d, J = 8.4 Hz, 1H), 3.48 (s, 2H), 3.13 – 2.93 (m, 4H).
Synthesis of 5-(4-Phenethyl-5-trifluoromethyl-pyrimidin-2-ylamino)-1,3-dihydro-indol-2-one (9). The intermediate 5-{[4-(2-phenyl-
ethynyl)-5-(trifluoromethyl)pyrimidin-2-yl]amino}-2,3-dihydro-1H-indol-2-one was obtained in 90% and the final product in
80% yield.
1H NMR (400 MHz, DMSO) d 10.31 (s, 1H), 10.07 (s, 1H), 8.61 (s, 1H), 7.58 (s, 1H), 7.48 (d, J = 8.6 Hz, 1H), 7.29 (t, J = 7.4 Hz, 2H), 7.20
(dd, J = 15.2 Hz, 7.1 Hz, 3H), 6.76 (d, J = 8.4 Hz, 1H), 3.48 (s, 2H), 3.03 (s, 4H).
Protein expression and purification
Human FAK kinase domain was expressed and purified by Instituto de Biologia Experimental e Tecnologica. The FAK catalytic
´
domain used for crystallization experiments and SPR assays was expressed with an NH2-terminal 6XHis-tag and comprises residues
410 to 689 after thrombin cleavage (FAK sequence with residue 410 changed from a Pro to a Gly). FAK was expressed in Hi5 insect
cells using the Bac Magic kit (Invitrogen Corp.) Cells were harvested by centrifugation and resuspended in lysis buffer (20 mM Na-P,
500 mM NaCl, 0.1% NP 40, 5 mM MgCl2, pH 7.5, 1 mM DTT, benzonase and protease inhibitor cocktail III). Cells were lysed by high
pressure homogenization and cleared by centrifugation at 31.000x g for 40 min at 4ꢁC. The supernatant was loaded onto a Ni-NTA
affinity column and the target protein was eluted in step gradient with buffer 20 mM Na-P, 500 mM NaCl, 500 mM Imidazole, 1 mM
DTT, pH 7.5. Peak fractions were desalted and treated with hu-alpha Thrombin for histidine tag removal. The cleaved proteins were
separated by passing through Ni-Sepharose resin and further purified using size-exclusion chromatography (Superdex 75 26/60).
The resultant pure recombinant FAK was concentrated to 5.9 mg/mL and stored at ꢀ80ꢁC in a buffer containing 10 mM HEPES,
200 mM ammonium sulfate, and 0.1 mM TCEP, pH 7.5.
The human PYK2 kinase domain used for SPR measurements was expressed with an NH2-terminal 6XHis-tag and comprised res-
idues 420-691. PYK2 was cloned into pET28a including an N-terminal HIS-TEV-tag and expressed in Escherichia coli Rosetta 2 (DE3)
Cell Chemical Biology 28, 1–13.e1–e7, May 20, 2021 e4

Please cite this article in press as: Berger et al., Structure-kinetic relationship reveals the mechanism of selectivity of FAK inhibitors over PYK2, Cell
Chemical Biology (2021), https://doi.org/10.1016/j.chembiol.2021.01.003
ll
Article
cells. Cells were harvested by centrifugation and resuspended in lysis buffer (20 mM Na-P, 500 mM NaCl, 1 mM DTT, 20 mM Imid-
azole, 5 mM MgCl2, pH 7.5, protease inhibitor cocktail III and Benzonase). Cells were lysed by high pressure homogenization and
cleared by centrifugation at 20.400x g for 60 min at 4ꢁC. Subsequent steps were similar to previously described for FAK. The resultant
pure recombinant PYK2 was concentrated to 10.7 mg/mL and stored at ꢀ80ꢁC in a buffer containing 30 mM HEPES, pH 7.5, 150 mM
NaCl, 0.1 mM EGTA, 1 mM DTT.
Crystallization and structure determination
FAK was incubated for 2 hours with 2 mM adenosine 50-triphosphate magnesium salt. Crystals were obtained using the hanging drop
vapor diffusion method and equilibrating against 1 mL of the reservoir solution (0.2 M Magnesium formate dihydrate, 10-20% PEG
3350) at 20ꢁC. The protein was mixed 1:1 with the reservoir solution. For complex formation with the inhibitor, the crystals were trans-
ferred to a stabilizing solution (0.2 M Magnesium formate dihydrate, 30% PEG 3350) containing 5 mM 1 and 5% DMSO and were
soaked for 24 h. All data sets were collected at 100 K on beamline SLS X106 and processed with the XDS software package (Kabsch,
2010). The structures were solved by molecular replacement, using BUSTER (Bricogne et al., 2016). Model building was performed in
Coot(Emsley, 2017), with compounds and waters fitted into the initial |Fo|–|Fc| map, and the structures were refined using BUSTER.
The coordinates of the holo-structures of FAK have been deposited in the RCSB Protein Data Bank. The refinement statistics and
PDB accession codes are given in Table S2.
Protein -ligand interfaces were calculated using the PISA webserver (Krissinel and Henrick, 2007). Buried Surface Area (BSA, in A )
2
˚
2
and Accessible Surface Area (ASA, in A ) were calculated per residue and normalized by dividing the BSA by ASA. Figures contain the
˚
interface profile of one chain of the asymmetric unit.
Surface plasmon resonance
SPR measurements were performed on a Biacore 4000 instrument from GE Healthcare. Recombinant FAK [huFAK (410-689)] and
PYK2 [huFAK (420-691)] were immobilized on a Biacore CM5 chip at 25ꢁC at a flow rate of 10 mL/min using amine coupling at pH
4.50 and pH 5.0 respectively, according to Biacore’s standard protocol. HBS-N (10 mM Hepes pH 7.40, 0.15 M NaCl, 0,05 %
Tween-20) served as running buffer during immobilization. FAK and PYK2 were applied at a concentration of 5 mg/mL in a buffer con-
taining a 60-fold excess of PF-431396. An unmodified carboxydextran matrix served as a reference surface. FAK and PYK2 inhibitors
stored as 10 mM stock solutions in 100% dimethyl sulfoxide (DMSO) were dissolved in running buffer (20 mM HEPES pH 7.50,
150 mM NaCl, 0.05% Tween 20, 1 mM DTT, 0.1 mM EDTA, 2% DMSO) and analyzed using two-fold dilution series. Kinetic titration
experiments were performed at 25ꢁC with a flow rate of 30 mL/min, a sample contact time of 120 s and a dissociation time between
300 and 600 s. Data sets were processed and analyzed using the Biacore 4000 Evaluation software, version 1.1. Solvent corrected
and double-referenced association and dissociation phase data were fitted to a simple 1:1 interaction model with mass transport
limitations.
NanoBRET cellular assays
The NanoBRET cellular target engagement assay is a tracer displacement assay. A BRET can here be observed between the NLuc
(tagged to the full-length protein kinase) and the tracer molecule used (a commercial tracer that uses a BODIPY fluorophore) in living
cells that were transiently transfected 20 h prior to assay performance. A compound of interest that binds to the protein kinase dis-
places the tracer and thereby reduces the BRET signal in a dose-dependent manner. The FAK and PYK2 NanoBRET assays that were
¨
performed here were conducted as described before (Rohm et al., 2019; Vasta et al., 2018). In brief: full-length FAK and PYK2 ORF
(Promega) cloned in frame with a N-terminal NanoLuc-fusion and C-terminal NanoLuc-fusion, respectively, were transfected into
HEK293T cells and proteins were allowed to express for 20h. For the target engagement assay serially diluted inhibitor and Nano-
BRET Kinase Tracer K5 (Promega) at 1 mM for FAK and at 2 mM for PYK2, respectively, were pipetted into white 384-well plates
(Greiner 781 207). The corresponding FAK or PYK2-transfected cells were added and reseeded at a density of 2 x 105 cells/mL after
trypsinization and resuspending in Opti-MEM without phenol red (Life Technologies). The system was allowed to equilibrate for 2
hours at 37ꢁC/5% CO2 prior to BRET measurements. To measure BRET, NanoBRET NanoGlo Substrate + Extracellular NanoLuc
Inhibitor (Promega) was added as per the manufacturer’s protocol, and filtered luminescence was measured on a PHERAstar plate
reader (BMG Labtech) equipped with 450 nm BP filter (donor) and 610 nm LP filter (acceptor). Competitive displacement data were
then graphed using GraphPad Prism software using a 4-parameter curve fit with the following equation: Y=Bottom + (Top-Bottom) /
(1+10^((LogIC50-X)*HillSlope))
For the kinetic wash-out assay, the inhibitor at a concentration of 10 times its previously determined IC50 was pipetted into
white 384-well plates (Greiner 781 207) and the corresponding FAK or PYK2-transfected cells were added and reseeded at a density
of 2 x 105 cells/mL after trypsinization and resuspending in Opti-MEM without phenol red (Life Technologies). The system was al-
lowed to equilibrate for 2 hours at 37ꢁC/5% CO2 to reach approximately 90% target occupancy prior to the wash-out via
removal of the medium to extract any unbound inhibitor and addition of the removed volume of Opti-MEM medium. The tracer
was added prior to BRET measurements at a final concentration as described for the target engagement assays. To measure
BRET, NanoBRET NanoGlo Substrate + Extracellular NanoLuc Inhibitor (Promega) was added as per the manufacturer’s protocol,
and filtered luminescence was measured on a PHERAstar plate reader (BMG Labtech) equipped with 450 nm BP filter (donor) and
610 nm LP filter (acceptor) for 2h. The preincubated and saturated inhibitor-target complex dissociates during the wash-out and the
tracer added associates which results in an increasing BRET signal. The Background of the kinetic data were corrected and then
e5 Cell Chemical Biology 28, 1–13.e1–e7, May 20, 2021

Please cite this article in press as: Berger et al., Structure-kinetic relationship reveals the mechanism of selectivity of FAK inhibitors over PYK2, Cell
Chemical Biology (2021), https://doi.org/10.1016/j.chembiol.2021.01.003
ll
Article
graphed using GraphPad Prism software using a one-phase association fit with the following equation: Y=Y0 + (Plateau-Y0)*(1-
exp(-K*x))
NanoBRET plasmid mutation
Plasmids obtained by PROMEGA were point-mutated using Quickchange Mutagenesis with HercII polymerase (Agilent). PCR cycles
were prolonged to 35 cycles to obtain sufficient amounts of plasmid. PCR products were then transformed into E.coli MACH1 and
potential clones sequenced for positive DNA sequence. DNA obtained was then amplified and purified using Midi-Prep Kits (QIAGEN)
and used in the NanoBRET experiments as described above.
Computations
Computational preparation of structures
For FAK, we employed crystal structures for 8 complexes (compounds 1, 2, 3, 4, 5, 8, 10 and 12, PDB IDs 3bz3, 6yt6, 6yxv, 6yq1, 6yoj,
6yvs, 6yr9 and 6yvy, respectively). The structure of the complex with compound 11 was built by similarity with 1 (PDB 3bz3). The FAK-
bound structures of compounds 7 and 9 were modelled using the complex with 8 as a template (PDB 6yvs). For compound 9, we
employed both DFG loop-type and helix-type conformations of FAK to build complexes and performed simulations with both struc-
tures (see Figure S4).
For PYK2, we employed crystal structures for 3 complexes (compounds 1, 10 and 12, PDB IDs 5TOB (Farand et al., 2016),
3FZR(Han et al., 2009) and 3H3C (Walker et al., 2008) , respectively). In the structures from PDB files 3FZR and 3H3C, there was
a missing region (residues 573-584 and 571-584, respectively). This region was modeled using the Sphinx web server to model loops
(Marks et al., 2017). The structure of PDB file 3FZR was used as input. The missing region in PDB file 3H3C was modeled using PDB
file 3FZR as a reference. In the structure from PDB file 5TOB, there were four regions missing (residues 446, 461-465, 495-496 and
´
570-587). Residues 570–587 were modeled using the RCD + web server to model loops (Lopez-Blanco et al., 2016). The remaining
regions were modeled using the PDB file 3FZR as a reference. Missing side chains were added using the WHATIF web server (Vriend,
1990). The protonation state of the histidine residues and charged residues at pH 7.4 was checked using pdb2pqr (Dolinsky et al.,
2004, 2007). To keep consistency among the different structures, their residues were modeled with the same protonation states. Res-
idues were assigned unusual protonation states only when such a state appeared in two or more structures. GLU 617 was protonated
because it had a pKa value higher than 7.4 in two different structures.
Docking of compounds to PYK2
Complexes for compounds without a crystal structure available were modeled using the protein coordinates and the ligand pose in
the PDB file 3FZR as a reference. Compounds 4 and 5 had bulkier substitutions compared to 10 in 3FZR, so their positions were
modeled by docking. Compound 2 had a substitution to a less bulky group, and its position was also modeled by docking. Docking
˚
of compounds 2, 4, and 5 to PYK2 (PDB 3FZR) was done with AutoDock Vina (Trott and Olson, 2010). Grids with 0.375 A spacing and
70 points along each side were centered in the known binding site. Rotation about bond torsions was allowed in ligands and the pro-
tein structure was kept frozen. The most favorable binding pose making two hydrogen bonds with Tyr 505 (known to be important
interactions from crystal structures) was used for simulations.
Modeling of PYK2 with a FAK-like DFG-motif helix
A model of PYK2 with an FAK-like DFG-motif helix was built to understand why PYK2 is not able to form this structure. The structures
of PYK2 (PDB 3FZR) and FAK (PDB 3BZ3) were aligned using PyMol and the coordinates of the atoms of the DFG-motif helix of FAK
and residues nearby (residues 562–569) were used to replace the original coordinates in PYK2 (residues 565–572). Moreover, a mu-
tation to Cys at position 462 was introduced in PYK2 to create the same disulfide bond present in FAK (residues 456 and 459 in FAK,
residues 459 and 462 in PYK2). New rotamers for Glu 474 and Lys 457 in PYK2 were selected using PyMol to avoid steric clashes with
the inserted DFG-motif helix. Missing residues 573–584 were modeled using the Sphinx web server to model loops (Marks et al.,
2017). The structure of PYK2 containing the DFG-motif helix was used as input. Missing side chains were added using the WHATIF
web server (Vriend, 1990). Crystallographic waters from PDB 3FZR were kept in the structure of PYK2 with the DFG-motif helix. The
protonation states of the residues were kept the same as those defined for WT PYK2 (PDB 3FZR). To evaluate the effect on protein
stability of the disulphide bridge that is located in the loop preceding the a-helix C in FAK (C456-C459) but is not present in PYK2, we
ran simulations twice, with and without the corresponding disulphide bridge. The simulations, however, showed only minor differ-
ences in the protein dynamics for the structures with and without disulphide bridges.
MD and tRAMD simulations
The tRAMD method for simulation of dissociation trajectories of protein ligand complexes and computation of relative residence
times has been reported elsewhere (Kokh et al., 2018), where the method was assessed on 70 N-HSP90 ATP-pocket binding com-
pounds, and we used the same method and parameterization. Here, we will give a brief description of the procedure (a complete
workflow is illustrated in Figure S7) (Panel A was prepared using weblogo web server (Crooks et al., 2004)).
We employed the Gaff force field (Wang et al., 2004) for ligands and the Amber ff14 force field (Maier et al., 2015) for the protein.
RESP (Cornell et al., 2002) partial atomic charges for ligands were obtained using molecular electrostatic potentials from ab initio
quantum mechanical calculations performed at the HF level with a 6-31G*(1d) basis set using the GAMESS software (Gordon and
Schmidt, 2005).
First, the system was solvated in a periodic box of TIP3P water molecules (Jorgensen et al., 1983) with NaCl ion concentration of
150mM. Then it was energy minimized, gradually heated and equilibrated stepwise with decreasing harmonic restraints on all heavy
Cell Chemical Biology 28, 1–13.e1–e7, May 20, 2021 e6

Please cite this article in press as: Berger et al., Structure-kinetic relationship reveals the mechanism of selectivity of FAK inhibitors over PYK2, Cell
Chemical Biology (2021), https://doi.org/10.1016/j.chembiol.2021.01.003
ll
Article
atoms of the protein and ligand using the AMBER15 software (Case et al., 2015). Then the system was run for 10 ns without restraints
and the final snapshot was used as an input for equilibration simulations carried out with the NAMD (Phillips et al., 2005) simulation
´
software. The equilibration was performed in the NPT ensemble using the Langevin and Nose–Hoover methods for the temperature
(300K) and pressure (1 atm) control, respectively. At least four replicas of equilibration simulations (20ns each) were prepared for each
ligand.
€
The last snapshot of each replica was used to simulate ligand dissociation. For this, the RAMD method (Ludemann et al., 2000) was
˚
applied, where an additional force with a magnitude of 13 kcal/molA and a random direction was applied to the center of mass of the
ligand in MD simulations. Every 100 fs, the force direction was changed randomly if the ligand center of mass did not move further
˚
˚
than a threshold of 0.025A and was retained otherwise. Simulations were stopped if the ligand center of mass moved further than 30 A
from the original position. At least 10 RAMD dissociation trajectories were generated starting with each equilibration replica. The
times required for ligand dissociation in each trajectory were stored and used for computation of the relative residence times as
described in (Kokh et al., 2018) and illustrated in Figure S7.
˚
Additionally, we repeated computations with a random force magnitude of 8 kcal/molA for the fast dissociating compounds only
(koff > 1.eꢀ3 sꢀ1, compounds 2, 6, 7, 8, 9 and 12 with FAK and all compounds with PYK2) in order to sample the egress process more
slowly. However, we did not obtain a significant improvement in the correlation with SPR data.
Conventional MD simulations for the apo-protein structures were performed using the same procedure and parameters as
described above for the equilibration step of the tRAMD workflow. For each protein, three 500 ns trajectories were simulated.
Analysis of MD trajectories
To analyze the protein-ligand interactions in the complexes, we computed protein-ligand interaction fingerprints, IFP, for the last 500
frames of each compound and equilibration trajectory using the MD-IFP tools (Kokh et al., 2020). In this approach, the IFP include
hydrophobic interactions, salt-links, hydrogen bonds, halogen bonds, aromatic interactions and water-mediated protein-ligand
˚
bridges, as well as all unspecific contacts within a threshold of 5A (aromatic interactions and salt bridges make very minor contribu-
tions and, therefore, are not presented in plots). From these data, the occurrence of each type of interaction was obtained for each
compound and the binding site protein residues. For the computation of BSA, the occurrence of unspecific protein-ligand contacts
was employed.
QUANTIFICATION AND STATISTICAL ANALYSIS
Surface-plasmon-resonance data
Data from at least n=5 individual experiments are reported as kon, koff, KD average SD. For analysis, the Biacore 4000 Evaluation
software, version 1.1 was used for solvent correction and double-referenced association and dissociation phase data fitting to a sim-
ple 1:1 interaction model with mass transport limitations.
NanoBRET data
Data from at least n=3 individual experiments are reported as KI, kobs average SEM. Analysis for IC50 determination was performed
using GraphPad Prism 8 software using a 4-parameter curve fit with the following equation: Y=Bottom + (Top-Bottom) /
(1+10^((LogIC50-X)*HillSlope)). IC50s were then converted to KI, obs using the Cheng-Prusoff-Relation(Cheng and Prusoff, 1973) tak-
ing the tracer parameters and tracer concentration into account. Analysis for kobs determination was performed using GraphPad
Prism 8 software using a one-phase association fit with the following equation: Y=Y0 + (Plateau-Y0)*(1-exp(-K*x)).
e7 Cell Chemical Biology 28, 1–13.e1–e7, May 20, 2021