A small molecule-kinase interaction map for clinical kinase inhibitors

Abstract

Kinase inhibitors show great promise as a new class of therapeutics. Here we describe an efficient way to determine kinase inhibitor specificity by measuring binding of small molecules to the ATP site of kinases. We have profiled 20 kinase inhibitors, including 16 that are approved drugs or in clinical development, against a panel of 119 protein kinases. We find that specificity varies widely and is not strongly correlated with chemical structure or the identity of the intended target. Many novel interactions were identified, including tight binding of the p38 inhibitor BIRB-796 to an imatinib-resistant variant of the ABL kinase, and binding of imatinib to the SRC-family kinase LCK. We also show that mutations in the epidermal growth factor receptor (EGFR) found in gefitinib-responsive patients do not affect the binding affinity of gefitinib or erlotinib. Our results represent a systematic small molecule-protein interaction map for clinical compounds across a large number of related proteins.

Full text

A small molecule–kinase interaction map for clinical
kinase inhibitors
Miles A Fabian
1,3
, William H Biggs III
1,3
, Daniel K Treiber
1,3
, Corey E Atteridge
1
, Mihai D Azimioara
1,2
,
Michael G Benedetti
1,3
, Todd A Carter
1
, Pietro Ciceri
1
, Philip T Edeen
1
, Mark Floyd
1
, Julia M Ford
1
,
Margaret Galvin
1
, Jay L Gerlach
1
, Robert M Grotzfeld
1
, Sanna Herrgard
1
, Darren E Insko
1
, Michael A Insko
1
,
Andiliy G Lai
1
, Jean-Michel Le
´
lias
1
, Shamal A Mehta
1
, Zdravko V Milanov
1
, Anne Marie Velasco
1
,
Lisa M Wodicka
1
, Hitesh K Patel
1
, Patrick P Zarrinkar
1
& David J Lockhart
1
Kinase inhibitors show great promise as a new class of therapeutics. Here we describe an efficient way to determine kinase
inhibitor specificity by measuring binding of small molecules to the ATP site of kinases. We have profiled 20 kinase inhibitors,
including 16 that are approved drugs or in clinical development, against a panel of 119 protein kinases. We find that
specificity varies widely and is not strongly correlated with chemical structure or the identity of the intended target. Many novel
interactions were identified, including tight binding of the p38 inhibitor BIRB-796 to an imatinib-resistant variant of the ABL
kinase, and binding of imatinib to the SRC-family kinase LCK. We also show that mutations in the epidermal growth factor
receptor (EGFR) found in gefitinib-responsive patients do not affect the binding affinity of gefitinib or erlotinib. Our results
represent a systematic small molecule-protein interaction map for clinical compounds across a large number of related proteins.
Protein kinases are critical components of cellular signal transduction
cascades. They are directly involved in many diseases, including cancer
and inflammation, and have become one of the most important target
classes for drug development
1,2
. The approval of imatinib (Gleevec) for
chronic myeloid leukemia (CML), and gefitinib (Iressa) and erlotinib
(Tarceva) for non-small cell lung cancer (NSCLC) has provided proof-
of-principle that small molecule kinase inhibitors can be effective
drugs. Over 30 kinase inhibitors are currently in clinical development,
and many more are in preclinical studies. The vast majority of these
compounds target the kinase ATP site, and because all of the more
than 500 protein kinases identified in the human genome have an ATP
site
3
, there is great potential for cross-reactivity. Compounds must be
tested experimentally against many kinases to assess molecular speci-
ficity and to identify off-target interactions
4,5
. Binding specificity and
affinity are not readily predicted based on available sequence or
structural information, and conventional profiling methods based on
in vitro activity are limited by the difficulty of building and running
large numbers of kinase activity assays.
We describe an experimental approach to assessing the specificity of
kinase inhibitors that directly and quantitatively measures binding to
the ATP site. Importantly, the method does not require chemical
linking, labeling or immobilization of tested compounds (Supple-
mentary Notes online). The approach circumvents many of the
difficulties of conventional enzyme activity assays and allows rapid
development and efficient use of assays for large numbers of kinases.
We have applied the technology to develop assays for 113 distinct
protein kinases and six of the clinically observed imatinib-resistant
variants of the ABL kinase, and have determined quantitative binding
profiles for 20 kinase inhibitors, including staurosporine, imatinib,
gefitinib and 14 compounds which are now or have been in clinical
development. We have also assessed the effect of nine recently
identified gefitinib-sensitizing EGFR mutations
6,7
on the interaction
of EGFR with eight known EGFR inhibitors, including gefitinib and
erlotinib. The data constitute a small molecule–kinase interaction map
and represent a systematic exploration of binding behavior of clinical
compounds across a large protein class.
RESULTS
Binding assays for small molecule–kinase interactions
The approach uses ATP site–dependent competition binding assays
(Fig. 1a)
8
. The key assay components are human kinases expressed as
fusions to T7 bacteriophage and a small set of immobilized probe
ligands that bind to the ATP site of one or more kinases. The small set
of immobilized ligands is used to build the assays, but the ‘free’ test
compounds (e.g., the 20 molecules profiled here) are not linked,
labeled or immobilized. The kinases used in the assays can be viewed
as fusion proteins that are tagged to facilitate expression, purification
and detection. In this scheme the T7 phage particle is not unlike more
Published online 13 February 2005; doi:10.1038/nbt1068
1
Ambit Biosciences, 4215 Sorrento Valley Blvd., San Diego, California 92121, USA.
2
Present addresses: Genomics Institute of the Novartis Research Foundation, 10675
John Jay Hopkins Dr., San Diego, California 92121, USA (M.D.A.), Buck Institute, 8001 Redwood Blvd., Novato, California 94945, USA (M.G.B.), SeattleBiomedical
Research Institute, 307 Westlake Ave. N., Ste. 500, Seattle, Washington 98109, USA (J.L.G.) and Metabasis Therapeutics, 9390 Towne Centre Dr., San Diego, California
92121, USA (M.A.I.).
3
These authors contributed equally to this work. Correspondence should be addressed to D.J.L. (dlockhart@ambitbio.com) or P.P.Z.
(pzarrinkar@ambitbio.com).
NATURE BIOTECHNOLOGY VOLUME 23 NUMBER 3 MARCH 2005 329
ARTICLES
© 2005 Nature Publishing Group http://www.nature.com/naturebiotechnology

conventional protein tags such as glutathione S-transferase or green
fluorescent protein, except that it renders the attached protein
amplifiable and amenable to very sensitive and versatile detection.
Kinases that have been cloned into the phage vector can be produced
rapidly by simply growing phage in Escherichia coli.T7phage
replication leads to lysis of the bacterial host, and lysates containing
properly folded, tagged kinases are used directly in the assay with no
need for conventional protein purification. The small number of
immobilized small molecule ligands used to build the assays bind
kinases with high affinity (K
d
o 1 mM), and were amenable to
attachment of biotin through a flexible chemical linker. For the assay,
tagged kinases and immobilized ‘bait’ ligands were combined with the
‘free’ test compound (Fig. 1a). If the ‘free’ test compound binds the
kinase and directly or indirectly occludes the ATP site, fewer protein
molecules bind the immobilized ligand on the solid support. If the
‘free’ test compound does not bind the kinase, tagged proteins are able
to bind to the modified solid support. The results are read out by
quantifying the amount of fusion protein bound to the solid support,
which is accomplished with extraordinary sensitivity by either tradi-
tional phage plaque assays or by quantitative PCR (qPCR) using the
phage DNA as a template. Both quantitation methods enable near
single-molecule protein detection, allowing us to accurately detect and
‘count’ as few as 10–100 tagged protein molecules.
To test our new approach, we initially built an assay for p38 MAP
kinase. p38 is activated in response to extracellular signals and
regulates the production of pro-inflammatory cytokines
9
.Thiskinase
is considered an excellent target for inflammation, and a number of
p38 inhibitors are in clinical trials
9
. Building the assay required tagged
p38 protein and an immobilized ligand that binds the p38 ATP site. To
produce tagged p38, we cloned the coding region for p38a into the
phage genome in-frame with the gene encoding the major T7 capsid
protein. We chose as an immobilized ligand SB202190, a pyridinyl
imidazole that was one of the first p38 inhibitors described (Tab l e 1 ;
see Supplementary Table 1 online for compound structures)
10
.Biotin
with a flexible linker was chemically attached to SB202190 and the
biotinylated compound immobilized on streptavidin-coated magnetic
beads
11
. Tagged p38 was found to bind to beads on which SB202190
had been immobilized, but not to beads lacking the ligand (Fig. 1b).
Phage with no displayed protein did not bind to beads with or
without SB202190 (data not shown). Binding to the solid support is
therefore dependent on both the immobilized ligand and on the
displayed kinase.
Six different free (unlinked) compounds were tested for the ability
to compete with the interaction between p38 and immobilized
SB202190: SB202190 (without biotin modification); SB203580 (a
pyridinyl imidazole closely related to SB202190) (Ta b l e 1 )
10,12
;
SB202474 (a pyridinyl imidazole that does not bind p38)
10
;BIRB-
796 (Ta bl e 1 )
13
; VX-745 (Tab l e 1)
14
; and purvalanol A (a cyclin-
dependent kinase 2 inhibitor)
15
. Competition with unmodified
SB202190, SB203580, BIRB-796 and VX-745 decreased by 1,000-fold
or more the amount of tagged p38 bound to the solid support,
whereas neither SB202474 nor purvalanol A had an effect (Fig. 1b).
Because BIRB-796 binds predominantly in a position adjacent to the
ATP site and affects the conformation of the ATP site indirectly
13
,
whereas SB202190, SB203580 and VX-745 bind directly in the ATP
site, we can conclude that the assay can detect allosteric as well
as direct binding interactions (Supplementary Notes online). To
determine the affinity of the interactions, we quantified the
amount of tagged p38 bound to the solid support as a function
of free test compound concentration (Fig. 1c). The binding con-
stants measured in this manner agree well with published values
No test compound Free test compound
Competition No competition
_
Immobilized ligand
Free test compound
_
+
_
+
SB202190
+
SB203580
+
BIRB-796
+
VX-745
+
SB202474
+
Purvalanol
A
1,000
10,000
100,000
1,000,000
10,000,000
100,000,000
Pfu
1.0
0.8
0.6
0.4
0.2
0.0
0.01 0.1 1 10 100 1,000
BIRB-796
VX-745
SB202190
SB203580
[Test compound] nM
Pfu (norm.)
abc
Figure 1 Competition binding assay for measuring the interaction between unlinked, unmodified (‘free’) small molecules and kinases. (a) Schematic
overview of the assay. The phage-tagged kinase is shown in blue, ‘free’ test compound in green and immobilized ‘bait’ ligand in red. (b) Binding assay for
p38 MAP kinase. The immobilized ligand was biotinylated SB202190. The final concentration of test compounds during the binding reaction was 10 mM.
(c) Determination of quantitative binding constants. Binding of tagged p38 to immobilized SB202190 was measured as a function of unlinked test
compound concentration. Tagged p38 kinase was quantified by real-time quantitative PCR and the results normalized. Representative curves are shown,
and average binding constants from at least two independent experiments are listed in Supplementary Table 4 online.
Table 1 Kinase inhibitors for which specificity profiles were
determined
Inhibitor Primary targets
a
Status
a
Staurosporine Pan-inhibitor Research compound
SB202190 p38a Research compound
SB203580 p38a Research compound
VX-745 p38a Phase 2 (discont.)
BIRB-796 p38a Phase 3
SP600125 JNK Research compound
Imatinib ABL, KIT, PDGFR Approved
Gefitinib EGFR Approved
Erlotinib EGFR Approved
CI-1033 EGFR subfamily Phase 2
GW-2016 EGFR, ERBB2, ERBB4 Phase 3
EKB-569 EGFR, ERBB2 Phase 2
ZD-6474 VEGFR2, EGFR Phase 2
Vatalanib/PTK-787 VEGFR2 Phase 3
SU11248 VEGFR2, PDGFR, FLT3, KIT Phase 3
MLN-518 FLT3 Phase 1
LY-333531 PKCb Phase 3
BAY-43-9006 RAF1 Phase 3
Roscovitine/CYC202 CDK2 Phase 2
Flavopiridol CDK1, CDK2, CDK4 Phase 2 (discont.)
a
Source: Pharmaprojects database, V5 (PJB Publications, http://www.pjbpubs.com).
330 VOLUME 23 NUMBER 3 MARCH 2005 NATURE BIOTECHNOLOGY
ARTICLES
© 2005 Nature Publishing Group http://www.nature.com/naturebiotechnology

(Supplementary Table 2 online). These results demonstrate that
the new binding assay correctly discriminates between compounds
that bind to the kinase and those that do not, and yields accurate
binding constants.
Assays for 113 distinct protein kinases
To expand the approach, we produced tagged versions of additional
human kinases. We have so far cloned into the phage vector 440 of the
B518 protein kinases in the human genome
3
. To develop additional
assays, we next identified a small set of molecules that could be used as
immobilized ligands. To avoid the need to find a unique ligand for each
kinase, we sought molecules that bind with high affinity to the ATP site
of multiple kinases. One molecule with these characteristics is stauro-
sporine (Ta bl e 1 )
16
. Biotinylated staurosporine was synthesized, immo-
bilized and tested for binding to each of the 440 tagged kinases. To be
considered for inclusion in the assay panel, a kinase was required to
bind staurosporine-derivatized beads but not beads without ligand.
Furthermore, binding had to be suppressed at least 100-fold when
10 mM staurosporine (not biotinylated) was included as a free test
compound in the competition assay. With staurosporine as the
immobilized ligand, these criteria were met for 51 kinases. The panel
was expanded to 113 distinct protein kinases by testing SB202190 and
additional immobilized compounds against our entire collection of
tagged kinases and applying the same criteria as for staurosporine
(Fig. 2). Kinases from all the major subfamilies are represented in the
panel, and any known or potential kinase inhibitor can be tested for
binding against the entire set of assays in a single experiment without
the need for chemical modification. To validate the assays, we have
shown that there is a very high correlation between inhibitor binding
affinity measured in the competition assays and inhibition of enzy-
matic activity measured in traditional enzyme or cell-based assays, and
that this high correlation is observed for both known and novel
interactions and for known and novel inhibitors (Supplementary
Tab l es 2 and 3;seeSupplementary Notes online for details).
Specificity profiles for clinical kinase inhibitors
There are currently over 30 kinase inhibitors in clinical trials or
approved for use in humans. The particular set of kinases inhibited
by a compound may profoundly affect therapeutic usefulness, yet for
most molecules specificity has been determined against only relatively
small sets of kinases
13,17–27
. We systematically screened a set of well-
known kinase inhibitors against the entire panel of 113 different
kinases. Twenty compounds representing a diversity of chemical
scaffolds and targeting a variety of kinases (Ta b l e 1;seeSupplemen-
tary Table 1 online for compound structures) were profiled in two
steps. First, a primary screen against the entire kinase panel was
performed at a compound concentration of 10 mM. Second, quanti-
tative binding constants were determined for each hit in the primary
screen (Supplementary Table 4; in this table blank fields indicate
compound/kinase combinations for which no evidence of binding was
observed at 10 mM in the primary screen).
The results show that molecular specificity varies widely among
these known inhibitors (Fig. 3). Staurosporine is known to be a highly
promiscuous inhibitor of many different kinases, and indeed binds to
104 of the 113 kinases with affinities fairly evenly distributed from
20 pM to B7 mM(Fig. 4; Supplementary Table 4 online). Among the
compounds that have been in clinical trials, several, such as SU11248,
bind to many kinases, whereas others, such as Vatalanib (also known
as PTK-787) and GW-2016, bind very few kinases in addition to their
known, primary targets. For most of the compounds tested here, the
tightest interaction is with the kinase or kinases they were optimized
to inhibit, but the difference in affinity between the primary target or
targets and other kinases varies substantially (Fig. 4). For BIRB-796,
VX-745, erlotinib, GW-2016 and SU11248, there is at least a tenfold
difference in affinity between intended targets and off-targets, whereas
for SP600125, EKB-569 and ZD-6474 there is less than a twofold
difference (Supplementary Table 4 online). This suggests that efforts
to optimize kinase inhibitor potency against a particular target have
been generally successful, but that it may be possible and even
necessary to improve the ability of some compounds to discriminate
between intended targets and off-targets.
Specificity can vary substantially even among compounds that are
based on the same chemical scaffold or that target the same kinase.
For example, the five quinazoline-class and quinoline-class EGFR
inhibitors (Supplementary Table 1 online) range from highly specific
to quite promiscuous (Fig. 3; Supplementary Table 4 online).
GW-2016 binds only STK10 and SLK (in addition to EGFR and
ERBB2), both of them fairly weakly (K
d
4 1 mM), and is an example
of an inhibitor that targets one kinase subfamily with exquisite
specificity. EKB-569, in contrast, binds 56 of the 113 kinases, several
of them with affinities almost equal to that for EGFR. Specificity
also does not appear to be simply determined by the particular
targeted kinase. For example, there are specific as well as promiscuous
compounds among the EGFR inhibitors, as outlined above, and
among the vascular endothelial growth factor receptor (VEGFR)2
inhibitors. Vatalanib binds only four kinases in addition to VEGFR2,
whereas SU11248 and ZD-6474 bind 73 and 43 additional kinases,
respectively (Fig. 3; Supplementary Table 4 online). These results
illustrate that specificity is not dictated by the general chemical
scaffold of an inhibitor, or by the primary, intended kinase target.
RTK
TKL
CK
PKA
CAMK
CDK
CLK
TK
MAPK
Figure 2 Panel of binding assays for 113 different protein kinases. Each
kinase represented in the assay panel is marked with a red circle. Gene
symbols for kinases in the panel are shown in Figure 5,aswellasin
Supplementary Table 4 online. The kinase dendrogram was adapted
3
,andis
reproduced with permission from Science and Cell Signaling Technology,
Inc. (http://www.cellsignal.com). With sponsorship by Cell Signaling
Technology and Sugen, the figure was originally presented as a poster in
Science to accompany the first analysis of the complete human kinome. TK,
nonreceptor tyrosine kinases; RTK, receptor tyrosine kinases; TKL, tyrosine
kinase-like kinases; CK, casein kinase family; PKA, protein kinase A family;
CAMK, calcium/calmodulin dependent kinases; CDK, cyclin dependent
kinases; MAPK, mitogen-activated protein kinases; CLK, CDK-like kinases.
NATURE BIOTECHNOLOGY VOLUME 23 NUMBER 3 MARCH 2005 331
ARTICLES
© 2005 Nature Publishing Group http://www.nature.com/naturebiotechnology

There are many examples of off-targets
that are not closely related by sequence and
function to the primary, intended target.
Many compounds considered tyrosine kinase
inhibitors also bind to serine-threonine
kinases, and serine-threonine kinase inhibi-
tors frequently bind to tyrosine kinases
(Ta b l e 1 and Fig. 3). Examples include the
p38 inhibitors VX-745 and BIRB-796, which
are chemically unrelated to each other but
bind a number of tyrosine kinases, and the
receptor tyrosine kinase inhibitors SU11248
and EKB-569, both of which bind a number
of serine-threonine kinases. A meaningful
assessment of specificity, therefore, cannot
be achieved by testing only against kinases
within the same subfamily. Even screening
against a small number of representatives of
multiple kinase families can be misleading.
For example, CI-1033 was previously shown
to be highly active against three members of
the EGFR family, but to have no activity
against seven other enzymes representing
receptor tyrosine kinases, protein kinase C
and cyclin-dependent kinases
21
.Thisledto
the conclusion that CI-1033 is highly specific.
The profile against a much larger panel
shows, however, that CI-1033 binds at least
36 different kinases and is among the more
promiscuous compounds (Fig. 3; Supple-
mentary Table 4 online).
To organize the results further we per-
formed a two-dimensional hierarchical clus-
ter analysis (Fig. 5a). One of the tightest
clusters in the compound dimension includes
BIRB-796, a p38 inhibitor, and BAY-43-9006,
a RAF1 inhibitor (Ta b l e 1 and Fig. 5). The
binding profile for BIRB-796 is largely a
subset of that for BAY-43-9006, with 25
targets shared between them. Ten additional
kinases bind BAY-43-9006, but not BIRB-796,
and only three kinases bind BIRB-796 but
not BAY-43-9006. The affinities of the two
compounds for the shared targets, however,
are very different (Supplementary Table 4
online). None of the four kinases with
binding constants below 100 nM for BAY-
43-9006 bind BIRB-796 with better than 1-
mM affinity. Conversely, only two of the eight
kinases with binding constants below 100 nM for BIRB-796 bind
BAY-43-9006 with better than 1-mM affinity, and none with better
than 100-nM affinity. Both molecules are substituted ureas with
similarly spaced aromatic ring systems (Supplementary Table 1
online), showing that compounds based on the same structural
scaffold can have closely related binding profiles even though they
have been optimized for inhibition of different primary targets.
Although the number of kinase inhibitors used here is not yet
sufficient for general patterns to emerge, clustering of kinases based
on small-molecule binding has the potential to reveal structural
relationships between active sites that may not be obvious from an
analysis of protein sequence
27
.
Assays for imatinib-resistant mutated versions of the ABL kinase
Imatinib has been a very successful drug for the treatment of
CML
28
. Unfortunately, a considerable fraction of patients treated
with the compound, including the majority of those with the
advanced, blast-crisis form of the disease, eventually develop resis-
tance
29–31
. Resistance in most cases is due to either amplification of
the BCR-ABL gene or to mutations in the ABL kinase that decrease
sensitivity to imatinib
32
. There is thus a great need for second
generation drugs that can inhibit the activity of imatinib-resistant
mutant ABL kinases
33
. To determine whether there are clinical kinase
inhibitors capable of inhibiting these therapeutically relevant mutated
kinases, we developed assays for six of the clinically observed mutant
TK TK
K
d
< 1 nM
1
_
10 nM
10
_
100 nM
100 nM
_
1 µM
1
_
10 µM
BIRB-796
RTK
TKL
CK
PKA
CAMK
CDK
CLK
TK
MAPK
BAY-43-9006
RTK
TKL
CK
PKA
CAMK
CDK
CLK
TK
MAPK
RTK
TKL
CK
PKA
CI-1033
CAMK
CDK
CLK
MAPK
RTK
TKL
CK
PKA
Erlotinib
CAMK
CDK
CLK
TK
MAPK
RTK
TKL
CK
PKA
Vatalanib/PTK-787
CAMK
CDK
CLK
TK
MAPK
RTK
TKL
CK
PKA
SU11248
CAMK
CDK
CLK
TK
MAPK
RTK
TKL
CK
PKA
Flavopiridol
CAMK
CDK
CLK
TK
MAPK
VX-745
RTK
TKL
CK
PKA
CAMK
CDK
CLK
TK
MAPK
RTK
TKL
CK
PKA
Gefitinib
CAMK
CDK
CLK
TK
MAPK
RTK
TKL
CK
PKA
ZD-6474
CAMK
CDK
CLK
TK
MAPK
RTK
TKL
CK
PKA
LY-333531
CAMK
CDK
CLK
TK
MAPK
RTK
TKL
CK
PKA
Roscovitine/CYC202
CAMK
CDK
CLK
TK
MAPK
RTK
TKL
CK
PKA
SB203580
CDK
CLK
TK
MAPK
RTK
TKL
CK
PKA
SP600125
CAMK
CDK
CLK
TK
MAPK
RTK
TKL
CK
PKA
SB202190
CAMK
CDK
CLK
TK
MAPK
RTK
TKL
CK
PKA
Imatinib
CAMK
CDK
CLK
TK
MAPK
GW-2016
RTK
TKL
CK
PKA
CAMK
CDK
CLK
TK
MAPK
RTK
TKL
CK
PKA
EKB-569
CAMK
CDK
CLK
MAPK
RTK
TKL
CK
PKA
MLN-518
CAMK
CDK
CLK
TK
MAPK
CAMK
RTK
TKL
CK
PKA
Staurosporine
CAMK
CLK
TK
MAPK
CDK
Figure 3 Specificity profiles of clinical kinase inhibitors. Kinase dendrograms were adapted
3
.TK,
nonreceptor tyrosine kinases; RTK, receptor tyrosine kinases; TKL, tyrosine kinase-like kinases; CK,
casein kinase family; PKA, protein kinase A family; CAMK, calcium/calmodulin dependent kinases;
CDK, cyclin dependent kinases; MAPK, mitogen-activated protein kinases; CLK, CDK-like kinases.
Circle size is proportional to binding affinity (on a log
10
scale). Binding constants were measured at
least in duplicate for each interaction identified in the primary screen. Complete quantitative results
are shown in Supplementary Table 4 online. The kinase dendrogram was adapted
3
and is reproduced
with permission from Science (http://www.sciencemag.org) and Cell Signaling Technology, Inc.
(http://www.cellsignal.com).
332 VOLUME 23 NUMBER 3 MARCH 2005 NATURE BIOTECHNOLOGY
ARTICLES
© 2005 Nature Publishing Group http://www.nature.com/naturebiotechnology

ABL kinases
29–31
and screened the set of 20 kinase inhibitors for
binding to these variants (Fig. 5b).
One of the most commonly observed mutant forms in patients,
T315I, is almost completely resistant to imatinib
30,31,34,35
, consistent
with the more than 1,000-fold difference in binding affinity between
wild type and the T315I mutant observed here (Fig. 5b; Supplemen-
tary Table 4 online). The E255K mutation confers greater than tenfold
resistance in cell-based assays, whereas the Q252H, Y253F, M351T and
H396P mutations have only moderate effects
30,31,36
. Consistent with
these observations, the E255K variant binds imatinib 50-fold more
weakly than the wild-type kinase in our assays, whereas the remaining
four mutations weaken binding only 6- to 30-fold. The relative
susceptibility of these variant kinases to imatinib is therefore faithfully
reflected in the binding constants measured here (Fig. 5b; Supple-
mentary Table 4 online), further illustrating that the ATP site–
dependent binding assays reflect biologically relevant behavior of
the kinases.
The 20 compounds tested fall into two classes based on their
binding behavior against wild-type and mutated ABL. Ten of the
compounds bound to all or most of the ABL variants, including wild
type, whereas the other ten bound neither wild type nor any of the
mutant forms (Fig. 5b; Supplementary Table 4 online). The most
striking finding is that the p38 inhibitor BIRB-796 binds ABL(T315I)
with a B40 nM binding constant (Supplementary Table 4 online).
Although several compounds have been described that can inhibit
some of the imatinib-resistant ABL forms, none of these compounds
effectively inhibit the important T315I mutant
33,37–39
.Theseresults
highlight the fact that mutations present as polymorphisms or that
emerge during tumor growth or in response to inhibitor treatment
may result in a mutant protein that can be inhibited with compounds
that do not bind well to the wild-type protein. This possibility can be
exploited by assessing compound binding not only to a panel of wild-
type kinases, but also to relevant variant forms.
A new target for imatinib
Although imatinib is not a promiscuous compound, our results reveal
that in addition to the known targets ABL, platelet-derived growth
factor receptor and KIT, it also binds tightly to the SRC-family
tyrosine kinase LCK (K
d
¼ 62 nM) (Fig. 3; Supplementary Table 4
online). To further support this finding, we tested imatinib in a LCK
enzyme activity assay, which showed the IC
50
to be 160 nM at 10 mM
ATP (Upstate Biotechnology; data not shown). Imatinib does not bind
SRC itself, consistent with the lack of SRC inhibition reported
previously
40
, but does bind weakly to two additional SRC-family
kinases, FRK and FYN (K
d
of 3.5 and 5.5 mM, respectively; Supple-
mentary Table 4 online). As an additional control, imatinib was tested
in a FYN enzyme activity assay, which showed the IC
50
to be 3.7 mMat
10 mM ATP (Invitrogen; data not shown), once again consistent with
our binding measurements. SRC, FRK, FYN and LCK are closely
related, and the fact that imatinib can discriminate between them
presents an interesting subject for molecular recognition studies
41
.
LCK is a key regulator of T-cell maturation and activation and a
potential target for immunosuppression
42
. It has been reported that
treatment with imatinib suppresses T-cell proliferation
43,44
and affects
the LCK pathway
44
, leading to the suggestion that imatinib may be
clinically useful as an immunosuppressant
44
. Our results may provide
a molecular explanation for these observations.
Clinical EGFR mutations do not affect inhibitor binding
It was recently shown that the presence of specific somatic mutations
in the EGFR tyrosine kinase in the vicinity of the ATP site correlates
with sensitivity of NSCLC to the EGFR inhibitor gefitinib
6,7
.This
discovery at least partially explains the clinical observation that only a
subset of patients respond to treatment with gefitinib and provides a
means to identify those patients most likely to benefit from the drug.
Two important questions are whether the mutations have a direct
effect on the interaction between gefitinib and the EGFR, and whether
the mutations are likely to sensitize tumors to treatment with other
EGFR inhibitors currently in development.
To determine whether binding of gefitinib to the ATP site of EGFR
is affected by the mutations, we developed assays for nine of the
described NSCLC EGFR mutants and measured their binding affi-
nities for gefitinib (Supplementary Table 5 online). Gefitinib bound
with similar affinity to wild type and mutant forms of EGFR, with the
measured binding constants for all nine mutants within threefold of
that for the wild type (Fig. 6; Supplementary Table 5 online). To
further determine whether other EGFR inhibitors, including several
currently in clinical development, bind the EGFR mutant forms, we
measured binding constants for seven additional compounds known
to bind the ATP site of wild-type EGFR. We found very little difference
between binding affinities for the wild type and any of the nine
mutants for the inhibitors tested (Fig. 6; Supplementary Table 5
online). For each compound, the affinities for all nine mutants were
within approximately threefold of that for the wild type. The only
exception was binding of SU-11464 to the G719C mutant, which was
almost sevenfold weaker than binding to wild-type EGFR (Fig. 6;
Supplementary Table 5 online).
The mutations in the EGFR tyrosine kinase that confer sensitivity
to gefitinib therefore do not fundamentally affect the intrinsic
interaction between the ATP site of EGFR and small-molecule EGFR
inhibitors. This contrasts with other activating catalytic domain
mutations that have a profound effect on the interaction with imatinib
and other inhibitors, such as mutations in KIT associated with
mastocytosis
45
. Compounds that discriminate between wild-type
and mutant EGFRs may be able to achieve a therapeutic benefit
while avoiding side effects such as the skin rash that has been observed
with several EGFR inhibitors
46
. That is to say, there is no privileged
interaction between gefitinib and gefitinib-sensitizing EGFR mutants,
and our results predict that tumors with gefitinib-sensitizing muta-
tions should also respond to treatment with other small-molecule
EGFR inhibitors. After our study was completed, it was reported that
mutations similar to those found in gefitinib-sensitive patients are also
Flavo
p
iridol
Roscovitine
LY-333531
MLN-518
SU11248
Vatalanib
ZD-6474
EKB-569
GW-2016
Cl-1033
Erlotinib
Gefitinib
Imatinib
SP600125
VX-745
BAY-43-9006
BIRB-796
SB203580
SB202190
Staurosporine
5
6
7
8
9
10
11
pK
d
Figure 4 Distribution of binding constants. For each compound the pK
d
(log K
d
) was plotted for all targets identified. Primary targets, as shown in
Table 1, are in blue, and off-targets in red. Staurosporine does not have a
particular primary target or targets, and the primary targets for BAY-43-9006
(RAF1) and LY-333531 (PKCb) were not part of the assay panel.
NATURE BIOTECHNOLOGY VOLUME 23 NUMBER 3 MARCH 2005 333
ARTICLES
© 2005 Nature Publishing Group http://www.nature.com/naturebiotechnology

present in tumors from patients who have responded to treatment
with erlotinib
47
.
DISCUSSION
The binding assays described here have several significant advantages
relative to traditional in vitro enzyme activity assays. First, the entire
panel can be run in parallel in a single experiment. All of the assays are
performed under the same conditions using similar reagents. Second,
the entire set of kinases can be prepared in parallel in about 2 h
immediately before use. Third, because of the low concentration of
kinase used, the assays have a wide dynamic range and can measure
binding affinities as low as 1–10 pM. Fourth, binding affinity provides
a ‘common denominator’ that allows direct comparisons between all
kinases in the panel. The assays are performed without added ATP or
substrate and measure binding, rather than activity. The results,
therefore, are not dependent on ATP concentration (or the K
m
for
ATP) or on the specific choice of substrate. Fifth, many protein kinases
that do not fold efficiently when overexpressed in E. coli are amenable
to our approach because of the extraordinary detection sensitivity of
the assays; only a small fraction of the total expressed protein, less than
1%, needs to be properly folded to yield sufficient signal.
We have described a systematic small molecule–kinase interaction
map for clinical kinase inhibitors. Integration of the information
provided here with results from cell-based or animal studies, and
ultimately with clinical observations, should enable a more complete
understanding of the biological consequences of inhibiting particular
combinations of kinases. Binding profiles for larger numbers of
chemically diverse compounds, combined with the phenotypes elicited
by these compounds in biological systems, will help identify kinases
whose inhibition leads to adverse effects, kinases that are ‘safe’ to
inhibit and combinations of kinases whose inhibition can have a
synergistic beneficial effect in particular disease states. This knowledge
should enable the development of inhibitors with ‘appropriate’
specificity that target multiple kinases involved in the disease process
while avoiding kinases implicated in side effects. The ability to rapidly
screen compounds against multiple kinases in parallel and the
incorporation of specificity profiling during initial lead discovery
and optimization should greatly facilitate and accelerate the drug
development process.
The kinase binding profiles also provide valuable information to
guide structural studies. In many cases kinases that tightly bind the
same compound have no obvious sequence similarity (for example,
p38 and ABL(T315I) binding to BIRB-796). In other cases, com-
pounds can discriminate between kinases closely related by sequence,
such as imatinib binding to LCK but not SRC. ABL and the imatinib-
resistant ABL mutants are of particular structural interest because
some compounds bind with good affinity to all forms (e.g., ZD-6474),
whereas BIRB-796 has a strong preference for a particular mutant. Key
insights should result from an analysis of selected co-crystal structures
of kinase-compound combinations identified through profiling
studies, and the large, uniform data set presented here should serve
as a valuable training set for computation-based inhibitor design.
Finally, the use of phage-tagged proteins in quantitative biochemical
assays circumvents the need for conventional protein production and
ABL1
(
T315I
)
ABL1 (Y253F)
ABL1 (Q252H)
ABL1 (M351T)
ABL1
ABL1 (H396P)
ABL1 (E255K)
Vatalanib
LY-333531
Flavopiridol
MLN-518
Roscovitine
GW-2016
BAY-43-9006
BIRB-796
EKB-569
SU11248
Cl-1033
ZD-6474
Staurosporine
Erlotinib
SB202190
Gefitinib
SP600125
SB203580
Imatinib
VX-745
NEK6
p38-β
p38-α
p38-γ
JNK3
JNK3
JNK1
PKMYT1
STK36
RIPK2
PTK6
GAK
CSNK1E
ERBB2
EGFR
MKNK2
PIM1
TNIK
TTK
STK17A
STK18
FGFR1
SYK
PAK1
PAK7/PAK5
PAK6
PAK4
PAK3
MAP3K5
NEK9
ULK3_m
STK17B
PHKG2
PHKG1
STK38L
MYLK2
NEK2
PTK2
FER
JAK2
ACK1
FYN
RPS6KA2
RPS6KA5
PIM2
CAMKK2
STK4
STK3_m
DAPK3
CAMK1D
AAK1
MARK2
DAPK2
Aurora3
Aurora2
BIKE
FGFR2
STK16
CAMK1G
CAMK1
CLK4
PRKAA1
FGFR3
RPS6KA3
INSR
PRKACA
CAMK2G
CAMK2D
CAMKK1
CAMK2A
CAMK2B
PCTK1
CLK2
CLK1
CLK3
VEGFR2
FLT4
EPHA7
PDGFRB
KIT
NTRK1
FLT3
LIMK1
SLK
STK10
EPHA6
MAP4K5
SRC
YES
MAP3K4
LYN
HCK
FGR
CSK
BTK
TEK
EPHA3
FRK
EPHA8
EPHA4
EPHA2
EPHB1
EPHA5
LCK
ABL2
ABL1
EPHB4
BMX
JAK1
CSNK1G2
CDK2
CDK5
CSNK1G1
MLN-518
Vatalanib
Flavopiridol
Roscovitine
LY-333531
SU11248
Staurosporine
SP600125
EKB-569
GW-2016
Cl-1033
Erlotinib
Gefitinib
BAY-43-9006
BIRB-796
ZD-6474
VX-745
Imatinib
SB203580
SB202190
a
b
Figure 5 Hierarchical cluster analysis of specificity profiles. Lighter colors
correspond to tighter interactions. (a) Twenty kinase inhibitors profiled
against a panel of 113 different kinases. (b) The same compounds
screened against a panel of six imatinib-resistant variants as well as
the wild-type ABL kinase.
334 VOLUME 23 NUMBER 3 MARCH 2005 NATURE BIOTECHNOLOGY
ARTICLES
© 2005 Nature Publishing Group http://www.nature.com/naturebiotechnology

purification, and should help reduce one of the major bottlenecks in
modern proteomics and drug discovery research.
METHODS
Kinase cloning. Kinases were cloned in a modified version of the commercially
available T7 Select 10–3 strain (Novagen). The head portion of each phage
particle includes 415 copies of the major capsid protein, and in this system
approximately one to ten of these are kinase fusion proteins. The fusion
proteins are randomly distributed across the phage head surface. The N
terminus of the kinase is fused to the C terminus of the capsid protein through
a flexible peptide linker; the kinases are linked to the T7 phage particle but are
not incorporated into the phage head. The fusion proteins are randomly
incorporated, and therefore distributed across the phage head surface. In
general, for single-domain kinases the entire coding region was cloned, whereas
for large, multidomain proteins, such as receptor tyrosine kinases, the catalytic
domain was cloned (Supplementary Table 6 online). Clones for each kinase
were sequenced, compared to an appropriate reference sequence, changed by
site-directed mutagenesis where necessary to exactly match the reference
sequence throughout the kinase domain, and then transferred into the phage
vector. Reference sequences for most of the kinases were obtained from the
RefSeq database. Whenever available we chose a curated sequence as the
reference (‘NM’ entries in RefSeq). For some of the kinases, there was evidence
from genomic and/or EST sequences that RefSeq entries were not the most
appropriate reference. In these cases alternative reference sequences consistent
with genomic and EST sequences were obtained from GenBank or from
Manning et al.
3
. Each kinase was fully resequenced after archiving to produce
a clone collection that is highly curated and matched to human genome
reference sequences.
Kinase assays. T7 kinase-tagged phage strains were grown in parallel in 24- or
96-well blocks in an E. coli host derived from the BL21 strain. E. coli were
grown to log phase and infected with T7 phage from a frozen stock (multi-
plicity of infection B0.1) and incubated with shaking at 32 1C until lysis
(B90 min). The lysates were centrifuged (6,000g) and filtered (0.2 mm) to
remove cell debris. Streptavidin-coated magnetic beads were treated with
biotinylated small molecule ligands for 30 min at 25 1Ctogenerateaffinity
resins for kinase assays. The liganded beads were blocked with excess biotin and
washed with blocking buffer (SeaBlock (Pierce), 1% BSA, 0.05% Tween 20,
1 mM DTT) to remove unbound ligand and to reduce nonspecific phage
binding. Binding reactions were assembled by combining phage lysates,
liganded affinity beads and test compounds in 1 binding buffer (20%
SeaBlock, 0.17 PBS, 0.05% Tween 20, 6 mM DTT). Test compounds were
prepared as 1,000 stocks in DMSO and rapidly diluted into the aqueous
environment (0.1% DMSO final). DMSO (0.1%) was added to control assays
lacking a test compound. All reactions were carried out in polystyrene 96-well
plates that had been pretreated with blocking buffer in a final volume of 0.1 ml.
The assay plates were incubated at 25 1C with shaking for 1 h, long enough for
binding reactions to reach equilibrium (data not shown), and the affinity beads
were washed four times with wash buffer (1 PBS, 0.05% Tween 20, 1 mM
DTT) to remove unbound phage. After the final wash, the beads were
resuspended in elution buffer (1 PBS, 0.05% Tween 20, 2 mM nonbiotinylated
affinity ligand) and incubated at 25 1C with shaking for 30 min. The phage titer
in the eluates was measured by standard plaque assays or by quantitative PCR.
Binding constant measurements. The equilibrium binding equations yield the
following expression for the binding constant for the interaction between the
free test compound and the kinase (K
d(test)
), assuming that the phage
concentration is below K
d(test)
: K
d(test)
¼ (K
d(probe)
/(K
d(probe)
+[Probe]))
[test]
1/2
. K
d(probe)
is the binding constant for the interaction between the kinase
and the immobilized ligand, [Probe] is the concentration of the immobilized
ligand and [test]
1/2
is the concentration of the free test compound at the
midpoint of the transition. If [Probe] is below K
d(probe)
the expression
simplifies to K
d(test)
¼ [test]
1/2
. Under these conditions the binding constants
measured for the interaction between kinases and test compounds (K
d(test)
)are
therefore independent of the affinity of the immobilized ligand for the kinase
(K
d(probe)
)(seeSupplementary Notes online). T7 phage grow to a titer of 10
8
–
10
10
plaque forming units (PFU)/ml, and the concentration of phage-tagged
kinase in the binding reaction is therefore in the low picomolar range. The
concentration of the immobilized ligand is kept in the low nanomolar range,
below its binding constant for the kinase. Binding data were fit to the equation
PFU ¼ L + ((H – L)  (K
d(test)
/(K
d(test)
+ [test]))), where L is the lower
baseline, H is the upper baseline, K
d(test)
is the binding constant for the
interaction between the test compound and the kinase, and [test] is the free test
compound concentration. Binding constants measured in duplicate on the
same day as part of the same experiment generally were within twofold.
Duplicate measurements performed on separate days generally varied by no
more than fourfold. Clustering and visualization was performed with Cluster
3.0 (M. Eisen, Stanford University) and Mapletree software (M. Eisen, Stanford
University; L. Simirenko, Lawrence Berkeley National Lab). For kinase/com-
pound combinations where no interaction was observed, the binding constant
was arbitrarily set to 1 M. K
d
values were converted to pK
d
(–log K
d
), and
clustering was based on the Pearson correlation.
Note: Supplementary information is available on the Nature Biotechnology website.
ACKNOWLEDGMENTS
We thank Tony Hunter, Nicholas Lydon and Webster Cavenee for a critical
reading of the manuscript and helpful discussions, Dan Lockhart for writing
software tools to facilitate data analysis, David Austin for helpful suggestions
regarding compound synthesis and Nicholas Olney and Victor Perez for expert
technical assistance.
COMPETING INTERESTS STATEMENT
The authors declare competing financial interests (see the Nature Biotechnology
website for details).
Received 16 September; accepted 20 December 2004
Published online at http://www.nature.com/naturebiotechnology/
1. Dancey, J. & Sausville, E.A. Issues and progress with protein kinase inhibitors for
cancer treatment. Nat. Rev. Drug Discov. 2, 296–313 (2003).
2. Cohen, P. Protein kinases–the major drug targets of the twenty-first century? Nat. Rev.
Drug Discov. 1, 309–315 (2002).
3. Manning, G., Whyte, D.B., Martinez, R., Hunter, T. & Sudarsanam, S. The
protein kinase complement of the human genome. Science 298, 1912–1934
(2002).
4. Davies, S.P., Reddy, H., Caivano, M. & Cohen, P. Specificity and mechanism of action
of some commonly used protein kinase inhibitors. Biochem. J. 351,95–105
(2000).
5. Bain, J., McLauchlan, H., Elliott, M. & Cohen, P. The specificities of protein kinase
inhibitors: an update. Biochem. J. 371, 199–204 (2003).
6. Lynch, T.J. et al. Activating mutations in the epidermal growth factor receptor under-
lying responsiveness of non-small-cell lung cancer to gefitinib. N. Engl. J. Med. 350,
2129–2139 (2004).
7. Paez, J.G. et al. EGFR mutations in lung cancer: correlation with clinical response to
gefitinib therapy. Science 304, 1497–1500 (2004).
8. Sche, P.P., McKenzie, K.M., White, J.D. & Austin, D.J. Display cloning: functional
identification of natural product receptors using cDNA-phage display. Chem. Biol. 6,
707–716 (1999).
10
1
0.1
K
d
(mut.)/K
d
(w.t.)
ZD-6474
SU-11464
EKB-569
GW-2016
Cl-1033
PKI-166
Erlotinib
Gefitinib
Figure 6 Relative binding affinities of EGFR inhibitors for wild type and
mutant forms of EGFR. For each compound, the ratio of the K
d
for each
mutant relative to wild type was calculated. Dashed lines mark a threefold
difference between binding constants for wild-type and mutant EGFRs
(K
d
(mut.)/K
d
(w.t.) o0.33 or 43).
NATURE BIOTECHNOLOGY VOLUME 23 NUMBER 3 MARCH 2005 335
ARTICLES
© 2005 Nature Publishing Group http://www.nature.com/naturebiotechnology

9. Kumar, S., Boehm, J. & Lee, J.C. p38 MAP kinases: key signalling molecules as
therapeutic targets for inflammatory diseases. Nat. Rev. Drug Discov. 2, 717–726
(2003).
10. Lee, J.C. et al. A protein kinase involved in the regulation of inflammatory cytokine
biosynthesis. Nature 372, 739–746 (1994).
11. Godl, K. et al. An efficient proteomics method to identify the cellular targets of protein
kinase inhibitors. Proc. Natl. Acad. Sci. USA 100, 15434–15439 (2003).
12. Tong, L. et al. A highly specific inhibitor of human p38 MAP kinase binds in the ATP
pocket. Nat. Struct. Biol. 4, 311–316 (1997).
13. Pargellis, C. et al. Inhibition of p38 MAP kinase by utilizing a novel allosteric binding
site. Nat. Struct. Biol. 9, 268–272 (2002).
14. Fitzgerald, C.E. et al. Structural basis for p38alpha MAP kinase quinazolinone and
pyridol-pyrimidine inhibitor specificity. Nat. Struct. Biol. 10, 764–769 (2003).
15. Gray, N.S. et al. Exploiting chemical libraries, structure, and genomics in the search for
kinase inhibitors. Science 281, 533–538 (1998).
16. Meggio, F. et al. Different susceptibility of protein kinases to staurosporine inhibition.
Kinetic studies and molecular bases for the resistance of protein kinase CK2. Eur. J.
Biochem. 234, 317–322 (1995).
17. Bennett, B.L. et al. SP600125, an anthrapyrazolone inhibitor of Jun N-terminal
kinase. Proc. Natl. Acad. Sci. USA 98, 13681–13686 (2001).
18. Buchdunger, E., Matter, A. & Druker, B.J. Bcr-Abl inhibition as a modality of CML
therapeutics. Biochim. Biophys. Acta 1551, M11–M18 (2001).
19. Wakeling, A.E. et al. ZD1839 (gefitinib): an orally active inhibitor of epidermal
growth factor signaling with potential for cancer therapy. Cancer Res. 62, 5749–5754
(2002).
20. Moyer, J.D. et al. Induction of apoptosis and cell cycle arrest by CP-358,774,
an inhibitor of epidermal growth factor receptor tyrosine kinase. Cancer Res. 57,
4838–4848 (1997).
21. Allen, L.F., Lenehan, P.F., Eiseman, I.A., Elliott, W.L. & Fry, D.W. Potential benefits
of the irreversible pan-erbB inhibitor, CI-1033, in the treatment of breast cancer.
Semin. Oncol. 29, 11–21 (2002).
22. Rusnak, D.W. et al. The effects of the novel, reversible epidermal growth factor
receptor/ErbB-2 tyrosine kinase inhibitor, GW2016, on the growth of human normal
and tumor-derived cell lines in vitro and in vivo. Mol. Cancer Ther. 1, 85–94 (2001).
23. Torrance, C.J. et al. Combinatorial chemoprevention of intestinal neoplasia. Nat. Med.
6, 1024–1028 (2000).
24. Wedge, S.R. et al. ZD6474 inhibits vascular endothelial growth factor signaling,
angiogenesis, and tumor growth following oral administration. Cancer Res. 62,
4645–4655 (2002).
25. Wood, J.M. et al. PTK787/ZK 222584, a novel and potent inhibitor of vascular
endothelial growth factor receptor tyrosine kinases, impairs vascular endothelial growth
factor-induced responses and tumor growth after oral administration. Cancer Res. 60,
2178–2189 (2000).
26. Kelly, L.M. et al. CT53518, a novel selective FLT3 antagonist for the treatment of acute
myelogenous leukemia (AML). Cancer Cell 1, 421–432 (2002).
27. Vieth, M. et al. Kinomics-structural biology and chemogenomics of kinase inhibitors
and targets. Biochim. Biophys. Acta 1697, 243–257 (2004).
28. Druker, B.J. Perspectives on the development of a molecularly targeted agent. Cancer
Cell 1, 31–36 (2002).
29. Gorre, M.E. et al. Clinical resistance to STI-571 cancer therapy caused by BCR-ABL
gene mutation or amplification. Science 293, 876–880 (2001).
30. von Bubnoff, N., Schneller, F., Peschel, C. & Duyster, J. BCR-ABL gene mutations in
relation to clinical resistance of Philadelphia-chromosome-positive leukaemia to
STI571: a prospective study. Lancet 359, 487–491 (2002).
31. Shah, N.P. et al. Multiple BCR-ABL kinase domain mutations confer polyclonal
resistance to the tyrosine kinase inhibitor imatinib (STI571) in chronic phase and
blast crisis chronic myeloid leukemia. Cancer Cell 2, 117–125 (2002).
32. Gambacorti-Passerini, C.B. et al. Molecular mechanisms of resistance to imatinib in
Philadelphia-chromosome-positive leukaemias. Lancet Oncol. 4, 75–85 (2003).
33. Shah, N.P. et al. Overriding imatinib resistance with a novel ABL kinase inhibitor.
Science 305, 399–401 (2004).
34. Branford, S. et al. High frequency of point mutations clustered within the adenosine
triphosphate-binding region of BCR/ABL in patients with chronic myeloid leukemia or
Ph-positive acute lymphoblastic leukemia who develop imatinib (STI571) resistance.
Blood 99, 3472–3475 (2002).
35. Azam, M., Latek, R.R. & Daley, G.Q. Mechanisms of autoinhibition and STI-571/
imatinib resistance revealed by mutagenesis of BCR-ABL. Cell 112,831–843
(2003).
36. Roumiantsev, S. et al. Clinical resistance to the kinase inhibitor STI-571 in chronic
myeloid leukemia by mutation of Tyr-253 in the Abl kinase domain P-loop. Proc. Natl.
Acad. Sci. USA 99, 10700–10705 (2002).
37. Warmuth, M. et al. Dual-specific Src and Abl kinase inhibitors, PP1 and CGP76030,
inhibit growth and survival of cells expressing imatinib mesylate-resistant Bcr-Abl
kinases. Blood 101, 664–672 (2003).
38. La Rosee, P., Corbin, A.S., Stoffregen, E.P., Deininger, M.W. & Druker, B.J. Activity of
the Bcr-Abl kinase inhibitor PD180970 against clinically relevant Bcr-Abl isoforms that
causeresistancetoimatinibmesylate(imatinib,STI571).Cancer Res. 62,7149–
7153 (2002).
39. Huron, D.R. et al. A novel pyridopyrimidine inhibitor of abl kinase is a picomolar
inhibitor of Bcr-abl-driven K562 cells and is effective against STI571-resistant Bcr-abl
mutants. Clin. Cancer Res. 9, 1267–1273 (2003).
40. Druker, B.J. et al. Effects of a selective inhibitor of the Abl tyrosine kinase on the growth
of Bcr-Abl positive cells. Nat. Med. 2, 561–566 (1996).
41. Nagar, B. et al. Crystal structures of the kinase domain of c-Abl in complex with the
small molecule inhibitors PD173955 and imatinib (STI-571). Cancer Res. 62,4236–
4243 (2002).
42. Goldberg, D.R. et al. Optimization of 2-phenylaminoimidazo[4,5-h]isoquinolin-9-ones:
orally active inhibitors of lck kinase. J. Med. Chem. 46, 1337–1349 (2003).
43. Mattiuzzi, G.N. et al. Development of Varicella-Zoster virus infection in patients with
chronic myelogenous leukemia treated with imatinib mesylate. Clin. Cancer Res. 9,
976–980 (2003).
44. Dietz, A.B. et al. imatinib mesylate inhibits T-cell proliferation in vitro and delayed-type
hypersensitivity in vivo. Blood 104, 1094–1099 (2004).
45. Ma, Y. et al. The c-KIT mutation causing human mastocytosis is resistant to STI571
and other KIT kinase inhibitors; kinases with enzymatic site mutations show different
inhibitor sensitivity profiles than wild-type kinases and those with regulatory-type
mutations. Blood 99, 1741–1744 (2002).
46. Lynch, T.J. et al. Novel agents in the treatment of lung cancer: conference summary
statement. Clin. Cancer Res. 10, 4199s–4204s (2004).
47. Pao, W. et al. EGF receptor gene mutations are common in lung cancers from ‘‘never
smokers’’ and are associated with sensitivity of tumors to gefitinib and erlotinib.
Proc. Natl. Acad. Sci. USA 101, 13306–13311 (2004).
336 VOLUME 23 NUMBER 3 MARCH 2005 NATURE BIOTECHNOLOGY
ARTICLES
© 2005 Nature Publishing Group http://www.nature.com/naturebiotechnology