1

Structure of the EGF Receptor Kinase Domain

Alone and in Complex with a

4-anilinoquinazoline InhibitorJennifer Stamosa, Mark X. Sliwkowskib and Charles Eigenbrota*

Departments of Protein Engineeringa and Molecular Oncologyb

Genentech, Inc

1 DNA Way

South San Francisco, California 94080 USA

running title: Structure of the EGF Receptor Kinase Domain

*Contact:

Charles EigenbrotGenentech, Inc.1 DNA WaySouth San Francisco, CA 94080

eigenbrot.c@gene.com650 225 2106650 225 3734 (fax)

Copyright 2002 by The American Society for Biochemistry and Molecular Biology, Inc.

JBC Papers in Press. Published on August 23, 2002 as Manuscript M207135200 by guest on M

arch 24, 2018http://w

ww

.jbc.org/D

ownloaded from

2

Summary

The crystal structure of the kinase domain from the epidermal growth factor receptor

(EGFRK), including forty amino acids from the carboxy-terminal tail, has been

determined to 2.6 Å resolution, both with and without an EGFRK-specific inhibitor

currently in Phase III clinical trials as an anti-cancer agent, erlotinib (OSI-774, CP-

358,774, Tarceva) . The EGFR family members are distinguished from all other known

receptor tyrosine kinases in possessing constitutive kinase activity without a

phosphorylation event within their kinase domains. Despite its lack of phosphorylation,

we find the EGFRK activation loop adopts a conformation similar to that of the

phosphorylated, active form of the kinase domain from the insulin receptor.

Surprisingly, key residues of a putative dimerization motif lying between the EGFR

kinase domain and carboxy-terminal substrate docking sites is found in close contact

with the kinase domain. Significant intermolecular contacts involving the carboxy-

terminal tail are discussed with respect to receptor oligomerization.

by guest on March 24, 2018

http://ww

w.jbc.org/

Dow

nloaded from

3

INTRODUCTION

Growth factor interactions with cell surface receptors influence proliferation,

survival, differentiation, and metabolism(1). The loss of control over these vital cellular

processes is a hallmark of oncogenesis(2). For instance, aberrant signaling from over-

expressed growth factor receptor ErbB2 is causal in about 30% of invasive breast

cancers(3). Growth factors bind to a cognate membrane-bound receptor system and

mediate changes in the intracellular portion of the receptor, often through the formation

of dimers or oligomers of receptors that initiate signal transduction cascades. The

epidermal growth factor receptor (EGFR, also ErbB1 or HER1) and its ligands epidermal

growth factor (EGF) and transforming growth factor-α (TGF−α) are among the earliest

characterized members of the growth factor/receptor tyrosine kinase (RTK) family. In

contrast to the widely applicable ligand-induced receptor dimerization paradigm, there

is evidence that EGFR family members exist as preformed dimers(4) and form higher

oligomer signaling complexes(5). Normal signaling in the EGFR system involves

ligand-induced, homo-oligomerization or hetero-oligomerization with the closely

related RTKs ErbB2 (HER2), ErbB3(HER3) and/or ErbB4 (HER4)(6).

Autophosphorylation of key tyrosine residues within the carboxy-terminal portion of

the receptor provides sites for direct interaction with SH2-containing proteins leading to

subsequent signal tranduction events.

The EGFR system, including receptor homologues and relevant ligands, is

complex. There are at least twelve different ligands that bind to the EGF receptor family

with partially redundant specificity for certain receptors. Several of the ligands,

including EGF, TGF–α, heparin-binding EGF, and betacellulin are reported to bind to

EGFR with nanomolar dissociation constants(7). Betacellulin also binds ErbB4 with high

by guest on March 24, 2018

http://ww

w.jbc.org/

Dow

nloaded from

4

affinity. Similarly, heregulin (HRG) binds to ErbB3 or ErbB4 with dissociation constants

in the nanomolar range. So far, a ligand that binds ErbB2 alone has not been identified,

although the affinity of an ErbB2/ErbB3 heterodimer for HRG is high, ~1011 M (8,9). In

addition, the kinase domain of ErbB3 has non-canonical amino acids at some key

positions which render it catalytically inactive(10). These factors, taken together, point

to a complicated interplay between cross-reacting ligands, functional diversity among

receptors, and differential expression in the EGFR signaling system.

In the non-signaling state, most RTKs possess low basal kinase activity that

increases substantially upon growth factor binding. This results from receptor

oligomerization and subsequent transphosphorylation of tyrosine residues within a

partner kinase domain. Specifically, initial phosphotyrosine (p-Tyr) modification of the

“activation loop” (A-loop) generates optimal catalytic activity and subsequent rapid

phosphorylation at substrate docking sites elsewhere on the receptor intracellular

domain (ICD). The EGFR, ErbB2, and ErbB4 receptors are the only known RTKs that do

not require this initial phosphorylation of kinase domain residues for full catalytic

competency. This unique feature may partially explain why EGFR family members are

frequently involved in cellular transformation. In the RTKs for which crystal structures

of both unphosphorylated and phosphorylated versions of the kinase domain are

available, phosphorylation in the A-loop causes it to undergo a large structural

reorganization that relieves steric and/or chemical restraints on the catalytic active

site(11).

Further distinguishing the EGFR family is a intracellular dimerization motif that

has been roughly assigned to reside between the kinase domain and the carboxy-

terminal phosphorylation sites. The greatest effects on receptor function seem to be

concentrated in the Leu955-Val956-Ile957 segment of EGFR (“LVI”) and other ErbB

by guest on March 24, 2018

http://ww

w.jbc.org/

Dow

nloaded from

5

receptors. This motif is necessary for ligand independent dimerization of EGFR

ICDs(12) and for transphosphorylation in ErbB2/ErbB3 heterodimers(13). Moreover,

alanine substitutions in this region override mutations in the transmembrane segment

of ErbB2 that would otherwise lead to constitutive signaling via non-ligand induced

dimerization(14). The molecular mechanism by which these residues influence receptor

activity is not well understood.

Members of the EGFR family are frequently overactive in solid tumors(15). A

number of therapeutic approaches that interfere with aberrant EGFR family signaling

are being investigated(16). A relatively new therapeutic approach to kinase inhibition is

the use of ATP-competitive small molecules(17-20). Several groups have shown that

certain 4-anilinoquinazoline derivatives are both selective and effective inhibitors of the

EGFR kinase(21). Structural data exists for compounds of this general class bound to the

distantly related intracellular kinases CDK2 and P38 kinase(22). Many of these

inhibitors are being tested for the treatment of cancer, including erlotinib (OSI-774, CP-

358,774, Tarceva), which is currently undergoing Phase III clinical study.

Despite extensive study of EGFR, the only part of its molecular structure that has

been reported is a hexapeptide from the intracellular domain(23). The ECD has

relatively high homology with the ECD of the receptor for insulin-like growth factor-1

(IGF-1) for which a partial structure has been determined(24). X-ray crystal structures of

kinase domains of several RTKs have been reported, although, unlike EGFRK, all of

these kinases require phosphorylation for full activity. Previous computational studies

have suggested possible binding modes for the 4-anilinoquinazoline class of inhibitors

to the EGFR kinase domain(25,26), but no direct structural evidence has been generated

thus far. Here we present the crystallographic analysis of the EGFR kinase alone and in

complex with the inhibitor erlotinib.

by guest on March 24, 2018

http://ww

w.jbc.org/

Dow

nloaded from

6

EXPERIMENTAL PROCEDURES

Expression and Purification of EGFRK

DNA encoding residues 672 to 998 was amplified from full length EGFR cDNA

(27) by PCR, incorporating an N-terminal NheI restriction site and a C-terminal stop

codon and XhoI site. The product was digested with NheI and XhoI and ligated into

appropriately digested pET-28b (Novagen, Madison, WI). Further PCR was performed

on the EGFRK-pET-28b plasmid to acquire the histidine tag and thrombin site using an

N-terminal primer with a NotI site and a C-terminal primer with an XbaI site. The

product was digested with NotI and XbaI and ligated into similarly digested pVL1392

(Pharmingen, San Diego, CA).

Spodoptera frugiperda insect cells, SF9, were transfected with the EGFRK-pVL1392

plasmid using the Baculogold transfection system (Pharmingen) according to the

manufacturer’s protocol. One liter of High Five cells (Invitrogen, San Diego, CA and

Expression Systems, Woodland, CA) in suspension at 5 x 105 cells/mL was inoculated

with 8 mL of amplified EGFRK virus and incubated at 27 °C for 72 hours. Cells were

harvested by centrifugation at 4000 x g for 15 minutes. Cells were frozen on dry ice

and then thawed, two times. 150 mL of buffer (50 mM Tris pH 7.5, 200 mM NaCl, 1%

glycerol, 1 mM DTT, 0.1 mM benzamidine) was added to the cells. The cells were then

mechanically homogenized. The lysate was centrifuged at 25000 x g for 45 minutes to

remove insoluble material. The supernatant was passed over a 0.45 µm vacuum filter

with pre-filter. Filtrate was loaded onto a Ni-NTA agarose column (Qiagen, Emeryville,

CA). The column was then washed with buffer (50 mM Tris pH 8.0, 500 mM NaCl, 5

mM imidazole) for 10 column volumes. EGFRK protein was eluted from the column

by guest on March 24, 2018

http://ww

w.jbc.org/

Dow

nloaded from

7

with 4 x 1 column volume aliquots of elution buffer (50 mM Tris pH 8.0, 300 mM NaCl,

250 mM imidazole). Fractions containing EGFRK as assayed by SDS-PAGE were

pooled, combined with thrombin to remove the histidine tag, and then dialyzed against

50 mM Tris pH 8.0, 250 mM NaCl, 1 mM DTT. EGFRK was then concentrated to a

volume of 500 µL and loaded onto a Superdex 75 gel filtration column (Amersham

Biosciences, Piscataway, NJ) pre-equilibrated with 50 mM Tris pH 8.0, 500 mM NaCl, 1

mM DTT. Fractions containing EGFRK as assayed by SDS-PAGE were pooled and

dialyzed against 10 mM Tris 8.0, 1 mM DTT, 1 mM sodium azide, 0.1 mM benzamidine.

The EGFRK was then concentrated to approximately 8 mg/mL. Typical final yield for

each liter of High Five culture was 1-2 mg.

Structure determination

Small crystals of EGFRK formed over 1 day in hanging drops when protein was

mixed with the reservoir buffer (1.0M Na/K tartrate, 0.1 M MES pH 7.0) in a 1:1 ratio.

These crystals were used as macro seeds in a 10 µL sitting drop containing a 1:1

protein:reservoir ratio as above. Crystals grew to approximately 250 µm3 in 1 week.

Crystals of EGFRK complexed with erlotinib were obtained by soaking crystals of apo-

EGFRK in a solution containing 1.1M Na/K tartrate, 0.1 M MES pH 7.0, 3 µM erlotinib,

for 3 weeks. Crystals with and without the erlotinib treatment were immersed in a

reservoir solution with added glycerol (20%) before preservation with liquid nitrogen.

Diffraction data were collected at beamline 19-ID of the Structural Biology Center

(Advanced Photon Source, Argonne National Laboratory), and at beamline 5.0.1 of the

Berkeley Center for Structural Biology (Advanced Light Source, Lawrence Berkeley

by guest on March 24, 2018

http://ww

w.jbc.org/

Dow

nloaded from

8

National Laboratory) extending to 2.6 Å for both apo-EGFRK and EGFRK/erlotinib

crystals, respectively (Table 1). Data were reduced with HKL2000 and

Denzo/Scalepack(28). The high symmetry of the crystals affords very high redundancy

in a 90˚ sweep. We chose the high resolution limit of data used in refinement based on a

signal-to-noise criterion ( I/σ(I) ≥ 2) rather than on agreement factors.

Space group symmetry and cell parameters suggested one molecule per

asymmetric unit (Vm=3.5 Å3/Dalton) and 61% solvent. The apo-EGFRK structure was

solved (AMORE, CCP4(29))using a polyalanine version of FGFRK (pdb entry 1fgk) in

space group I23 using data to 4.0 Å from an in-house data set (rotating anode (RU-

200)/Mar345 scanner) reduced with Denzo/Scalepack. Before refinement, 4% of the

data were sequestered for calculation of Rfree, and the same set was used (and

extended to higher resolution) for the synchrotron data sets. Initial phases treated with

SIGMAA(30) and solvent flattened (DM) produced a map with indications of many

buried side chains, especially leucines and tryptophans. Model inspection and

adjustment were performed with XtalView(31) and refinement employed XPLOR98

(Accelrys, San Diego). The apo-EGFRK structure was used as a starting point for the

EGFRK/erlotinib work, and the final stages of refinement were performed similarly.

The C-terminus of our construct lacks structural similarity to any part of the template

FGFRK starting structure. The 13-residue section immediately following His964 is too

poorly ordered to be fit. Weak electron density was assigned starting with residue

Leu977, which leaves a 7 Å gap for the unassigned amino acids. Alternate connectivities

would bridge gaps of 22 Å or 29 Å. The final section of the C-terminus is well ordered

where it forms intermolecular contacts with two neighboring molecules within the

crystal. The activation loop is traced for its entire length. Individual isotropic

temperature factors were refined, and a bulk solvent term included. The average

by guest on March 24, 2018

http://ww

w.jbc.org/

Dow

nloaded from

9

temperature factors are high (~70 Å2), and the maximum permitted B was 120 Å2,

attained by a 1 – 2% of the atoms. Final (Fo-Fc) electron density maps lack interpretable

features. Coordinates of apo-EGFRK and EGFRK/erlotinib are available from the

Protein Data Bank, accession codes 1M14 and 1M17, respectively.

RESULTS

The EGFR kinase domain (EGFRK) adopts the bilobate fold characteristic of all

previously reported protein kinase domains (Fig 1). The N-terminal lobe (N-lobe) is

formed from mostly β-strands and one α-helix (αC), while the larger C-terminal lobe

(C-lobe) is mostly α-helical. The two lobes are separated by a cleft like those in which

ATP, ATP analogues and ATP-competitive inhibitors have been found to bind.

Important elements of the catalytic machinery bordering the cleft on the N-lobe include

the glycine-rich nucleotide phosphate binding loop (Gly695 to Gly700), while the C-lobe

contributes the DFG motif (Asp831 to Gly833) , the presumptive catalytic (general base)

Asp813, the catalytic loop (Arg812 to Asn818), and the A-loop (Asp831 to Val852).

N-lobe

The N-terminal lobe of EGFRK adopts a tertiary structure similar to previously

observed structures of RTKs (rmsd for superpositioning Cα atoms with the kinase

domain from the fibroblast growth factor receptor is ~1.2 Å), although a few features

distinguish the N-lobe of EGFRK from other kinase domains. The N-terminus of our

construct begins 25 amino acids before the first glycine of the nucleotide phosphate

binding loop and includes 5 additional amino acids prior to Ser671 that derive from the

expression construct. The first residue we identify in the electron density maps is

by guest on March 24, 2018

http://ww

w.jbc.org/

Dow

nloaded from

10

Gly672, with the succeeding 13 residues adopting an extended conformation. The N-

terminal 9 amino acids are influenced by several intermolecular contacts, including H-

bonds involving main chain atoms of residues Asn676 and Leu680, although

intramolecular H-bonds between Asn676 and both Tyr740 and Ser744 also contribute.

At Glu685, the polypeptide chain assumes a trace more similar to those of the

lymphocyte tyrosine kinase(32) (LCK, pdb entry 3lck), the insulin receptor kinase

phosphorylated form(33) (p-IRK , pdb entry 1ir3), and the unphosphorylated form of

the FGF receptor kinase(34) (FGFRK, pdb entry 1fgk). However, EGFRK lacks the

tryptophan-glutamate (“WE”) motif found in these related kinases, having instead

Arg681-Ile682. In the “WE” containing kinases, hydrophobic interactions of the

tryptophan and an H-bond between the glutamate and a threonine or serine and the

neighboring β-strand tie the N-terminal region to the N-lobe. In EGFRK, Arg681

projects into solvent but Ile682 contacts Leu782 and Ile756 on the neighboring β-strand

and thereby affords a similar effect.

Among the canonical features characterizing the N-lobes of active forms of

kinases is a salt bridge between two highly conserved side chains that, when ATP or a

close homologue is present, interact with the α- and β-phosphates. In both the apo- and

inhibitor-bound forms of EGFRK , we find such a salt bridge between Lys721 and

Glu738. Our observation of this salt bridge in apo-EGFRK indicates that EGFR does not

require large rearrangements within the N-lobe for catalytic competence.

C-lobe

The C-terminal domain of EGFRK contains the usual organization of α-helices

present in other kinase domain structures. Superpositioning of the C-lobes of kinase

by guest on March 24, 2018

http://ww

w.jbc.org/

Dow

nloaded from

11

domains from both LCK and p-IRK yield an rmsd of 1.1 Å. However, as with the N-

lobe, a few key features differ from previously elucidated RTK structures.

Activation Loop

In most protein kinases, the activation loop assumes its catalytically competent

conformation only if it first becomes phosphorylated on a Tyr or Thr. For these kinases,

the unphosphorylated activation loop is positioned many Ångstroms from the active

conformation, and may include a direct inhibitory element. For instance, the

unphosphorylated A-loop in FGFRK is incompatible with substrate binding, and the

unphosphorylated insulin receptor kinase A-loop blocks ATP binding as well as the

substrate tyrosine site.

The A-loop in apo-EGFRK (and EGFRK/erlotinib) differs significantly from

other apo-, unphosphorylated A-loop structures. Earlier work has shown that Tyr845 of

the EGFRK A-loop, at a position that is phosphorylated in other RTKs, can be replaced

by Phe without loss of function(35). Consistent with this, we see that the A-loop of

EGFRK adopts an “active” conformation, similar to the phosphorylated A-loop of p-

IRK (Fig. 2). Many energetically beneficial interactions stabilize this conformation, most

of which are also found in other active kinase A-loops. Tyr845 aligns well structurally

with p-Tyr1163 of p-IRK and makes van der Waals contact with the aliphatic part of

neighboring Lys836, a residue which occupies the space of Arg1155 in the p-IRK

structure. An H-bond between side chains of Tyr845 and Glu848 mimics that between

p-Tyr1163 and the main chain nitrogen of Gly1166 in p-IRK, but the electron density

supporting this Glu848 side chain conformation is weak (Fig. 3). This interaction may be

important for the loop conformation, but there is another more significant aspect of

Glu848. The relationship between Tyr845 and Arg812 (preceding the catalytic Asp813)

by guest on March 24, 2018

http://ww

w.jbc.org/

Dow

nloaded from

12

is the same as between the analogous residues in p-IRK and other tyrosine kinases. This

relationship is central to arranging the catalytic machinery and substrate for phospho-

transfer. In p-IRK, Tyr1163 is phosphorylated, and in EGFRK the Glu848 carboxylate

can assume a position closely analogous to that of the phosphate of p-Tyr1163 in p-IRK.

Antiparallel β-strand main chain/main chain H-bonds between Lys836 to Leu838

and Val810 to Arg808 are key anchors for the conformation adopted by the early part of

the EGFRK A-loop, for which analogous interactions appear in p-IRK. However, in an

additional interaction not seen in the p-IRK structure, the side chain of Arg808 H-bonds

to the main chain oxygen of Gly839. This interaction lends further stability to the

unphosphorylated EGFRK A-loop “active” conformation, and it is noteworthy that the

incidence of arginine at position 808 among kinases is low.

Underlying the central part of the A-loop, Tyr867 accepts an H-bond from

Arg812. While many kinases have a homologous Arg preceding the catalytic Asp813, its

interactions with a homologous Tyr (or sometimes Phe) vary in type. Some are purely

hydrophobic, while others involve Arg hydrogen atoms and either the hydroxyl oxygen

or π−electrons of the tyrosine. Tyr867 π-electrons interact with Arg813 in EGFRK, with

the details very similar to those found in p-IRK.

There are other protein kinases with known molecular structures that do not

require phosphorylation in their A-loop for optimal catalytic competence, among them

glycogen phosphorylase kinase(36) (pdb entry 1phk), casein kinase 1(37) (pdb entry

1csn), carboxy-terminal src kinase(38) (pdb entry 1k9a) and twitchin kinase(39)(pdb

entry 1koa). The most relevant of these is the phosphorylase kinase in which a Glu

residue (Glu182), at an A-loop position analogous to sites of phosphorylation, interacts

with the Arg preceding the catalytic Asp and thus reprises the p-Tyr or p-Thr role. The

by guest on March 24, 2018

http://ww

w.jbc.org/

Dow

nloaded from

13

EGFRK A-loop is rich in Glu residues, and the contribution of two of these (Glu842 and

Glu844) to EGFRK function was demonstrated using amino acid substitutions that

altered the in vitro kinetics of phospho-transfer(40). In EGFRK, the structural roles of

Tyr845 and Glu848 are a hybrid between the hydrophobic contacts of Tyr1163 in p-IRK

and the electrostatic effects of Glu182 in phosphorylase kinase.

Overall, the conformation adopted by the EGFRK A-loop appears to result more

from an energetic advantage for the “active” conformation rather than an energetic

disadvantage for an alternate “inactive” one. For instance, there is no apparent

impediment to the EGFRK A-loop adopting the conformation found for FGFRK, from

which it differs by up to 25 Å. Additionally, intermolecular crystal packing contacts do

not seem to play a role, since influential interactions of this type do not exist in our

structures, and deleterious ones would not arise if we were to imagine an FGFRK-like

A-loop in our crystal lattice.

LVI Motif

The C-lobe of EGFRK also contains the distinctive three amino acid sequence

(Leu955–Val956–Ile957) found previously to regulate, in a poorly understood manner,

the transphosphorylation of substrate tyrosines in oligomerized EGFR family

complexes. Of the three amino acids, substitutions for Leu955 are the most deleterious

for phospho-transfer activity. A model suggesting direct contact between this segment

and another protein, either another receptor or an unidentified adaptor protein, has

been proposed for the ErbB2/ErbB3 system(13). In our EGFRK structures, LVI is in

close contact with the C-lobe. The proximal polypeptide region is coupled to the C-lobe

by the completely buried Leu955 side chain (Fig. 4). This strongly suggests that the

Leu955 side chain does not contact another protein. It is more probable that

by guest on March 24, 2018

http://ww

w.jbc.org/

Dow

nloaded from

14

replacement of Leu955 with alanine uncouples it and nearby residues from the C-lobe.

It is possible that the tripeptide segment mediates subtle, allosteric effects in the

intracellular region that affect the viability of the C-terminal tail as a substrate for

transphosphorylation.

Following the LVI sequence the electron density continues strongly until

dropping off abruptly after residue Asp960, and 13 residues following His964 have

largely untraceable density. The disordered residues are part of an endocytotic signal

sequence which directs ligand-activated receptors to recycling or degradation via

clathrin-coated pits(41,42). The 19 residues starting at Leu977 become increasingly well-

ordered and are in close contact with neighboring molecules (crystal packing contacts)

beginning with Asp985. The last two amino acids in our construct are not observed.

There is no discernable secondary structure in the C-terminal extension except for a few

isolated (i, i+4) α-helical H-bonds. The crystal packing contacts experienced by the 19

well resolved C-terminal residues involve two neighboring molecules (Fig 5). The first

contact reduces solvent accessible surface area by about 1300 Å2 on each side, while the

other is characterized by an approximately 1000 Å2 reduction. The first contact, in

addition to being of greater size, is also slightly more complimentary with respect to

electrostatic charge. The C-terminal ordered region is highly acidic

(LMDEEDMDDVVDADEYLIPQ), and the opposing contact region of the neighboring

molecules is generally basic. The opposing region of the larger interface is at the back

side of the ATP-binding cleft. The smaller contact area for the C-terminal ordered

region is in the C-lobe of a different neighboring molecule.

by guest on March 24, 2018

http://ww

w.jbc.org/

Dow

nloaded from

15

Catalytic site

The well conserved sequence of the EGFRK catalytic loop (HRDLAARN), also

present in p-IRK and FGFRK, shares structural conservation with these RTKs as well,

with a rmsd of most atoms only 0.1 Å. The nearby conserved DFG sequence, important

for ATP coordination, adopts the p-IRK-like arrangement. The ATP binding site of apo-

EGFRK would require only limited rearrangement to accommodate the AMP-PNP

present in the p-IRK structure. The principle difference between the p-IRK and EGFRK

structures in this region is in the nucleotide phosphate binding loop. Although in both

apo-EGFRK and EGFRK/erlotinib the nucleotide phosphate binding loop is poorly

ordered, Phe699 is better discerned in the erlotinib complex as it is brought closer to the

C-lobe as a general consequence of inhibitor binding.

Relationship Between N-lobe and C-lobe

The opening angle between the two lobes of kinase structures has been observed

to differ depending on the presence or absence of ATP or a close analogue. Those forms

with smaller opening angles (“closed”) bring the important catalytic elements into

proximity. Among available prior kinase structures, the relationship between EGFRK

lobes is most similar to those of LCK, p-IRK, and FGFRK, for which superposition of

about 250 Cα pairs yields rmsd values of about 1.2 Å. After superposition on the C-

lobes, the β-strands of both apo-EGFRK and EGFRK/erlotinib N-lobes align more

nearly with those of LCK and p-IRK than those of FGFRK, and the placement of the αC

helix is intermediate between those of LCK and p-IRK.

Comparing apo- and erlotinib complex structures, the rmsd for Cα atoms

excluding termini is 0.4 Å, and for the isolated lobes 0.25 Å (C-lobe) and 0.5 Å (N-lobe).

by guest on March 24, 2018

http://ww

w.jbc.org/

Dow

nloaded from

16

Most localized differences between the apo- and erlotinib complex structures arise

coincident with elevated B-factors and can be discounted. One exception is seen at

Arg752, where Cα atoms are 1.5 Å apart. The validity of this interpretation of the

respective electron density maps was gauged by inspecting an (Fo-Fo) difference map

calculated using phases from the erlotinib complex. This map supports the variation at

Arg752, and more generally indicates a shift of the N-lobe that reduces the inter-lobe

angle in EGFRK/erlotinib versus apo-EGFRK. This effect is small, and distances

between corresponding Cα atoms are about 0.6 Å on average, closing the inter-lobe

angle by ~3˚.

Inhibitor Interactions

Previous studies have indicated that 4-anilinoquinazolines such as erlotinib

cause inhibition through binding to the site occupied by ATP during phospho-transfer.

There is a vast literature describing ATP-competitive inhibitors directed against the

EGFRK(25,26,43,44). Computational methods have been applied to quinazolines

binding to EGFRK(25,26) , and where the results have been depicted, they differ in

detail from our result, though whether from inaccuracies in the protein homology

model, differences in quinazoline subsitituents, or other causes is not clear. In our

complex of erlotinib with EGFRK, we find the compound in an orientation very

reminiscent of those seen for closely related 4-anilino-quinazoline molecules complexed

with cyclin dependent kinase 2 (CDK2) and mitogen-activated protein kinase p38 (P38)

(pdb entries 1di8 and 1di9, respectively)(22). Erlotinib lies with the N1- and C8-

containing edge of the quinazoline directed towards the peptide segment connecting N-

and C-lobes, with the ether linkages projecting past the connecting segment into solvent

by guest on March 24, 2018

http://ww

w.jbc.org/

Dow

nloaded from

17

and the anilino substituent on the opposite end sequestered in a hydrophobic pocket

(Fig. 6). The N1 of the quinazoline accepts an H-bond from the Met769 amide nitrogen.

The other quinazoline nitrogen atom (N3) is not within H-bonding distance of the

Thr766 side chain (4.1Å), but a water molecule bridges this gap. Such a water was

observed by Shewchuk et al. in the P38/inhibitor complex(22) and was predicted by

Wissner et al.(26). The same water molecule contacts the side chain of Cys751, which

itself is disordered between two side chain conformers. The less robust nature of this

water-mediated H-bond between erlotinib and EGFRK parallels the relatively small

effect on inhibitor affinity seen for substitution with carbon for N3 among compounds

characterized by Rewcastle et al.(45).

The interplanar angle of aromatic ring systems in erlotinib is 42˚. This directs the

acetylene moiety into a pocket that many kinase domains share when the amino acid

side chain at position 766 is small (threonine in EGFRK). P38 also has threonine at this

position, but CDK2 has a phenylalanine, so that where the P38-bound inhibitor has a

dihedral angle similar to ours (39˚), in the CDK2-bound inhibitor the anilino ring is

coplanar with its quinazoline. We find Thr766, Lys721, and Leu764 are less than 4 Å

from the acetylene moiety on the anilino ring (Thr766 and Leu764 ~3.4Å). Both Met742

and Cys751 have been suggested to contact inhibitors very similar to erlotinib(25), but

contact distances seen here for both are greater than 4.5 Å. In the CDK2/inhibitor

complex, quinazoline carbon atoms C2 and C8 direct their attached hydrogen atoms

toward protein carbonyl oxygen atoms analogous to those of Gln767 and Met769,

respectively, with carbon-oxygen distances of about 3 Å. In the P38 complex, only the

first of these contacts is observed. In our complex, we observe both C2- and C8-carbonyl

oxygen contacts (3.1 Å and 3.2 Å). Thus the disposition of erlotinib with respect to the

by guest on March 24, 2018

http://ww

w.jbc.org/

Dow

nloaded from

18

interplanar angle and linker-region contacts is a hybrid of the CDK2 and P38

quinazoline complexes.

DISCUSSION

The inherent catalytic activity of the EGFR family kinases is unique among RTKs.

Regulation of the vital cellular processes influenced by EGFR signaling must be exerted

by control of the delivery of C-terminal substrate tyrosines to the active site. The

structure we have determined makes this emphatically clear, as we find all the catalytic

elements primed and ready for phospho-transfer, even though crystals were grown

without cofactor or substrate analogues. This fundamental difference from other RTKs

raises the possibility that the spatial relationships between the two or more kinase

domains in an EGFR-family complex may differ from those of even closely related

RTKs, for instance FGFR.

The reduction in the inter-lobe angle upon addition of erlotinib to apo-EGFRK

crystals suggests ATP binding would have a similar effect. Learning the true magnitude

of such effects will require crystals grown with the ATP-binding cleft already occupied,

however no essential element of the catalytic machinery requires much change from the

arrangements we see in this work. Crystals of the erlotinib complex were obtained by

crystallizing EGFRK in the absence of inhibitor, and then soaking in the inhibitor.

Therefore, any artifact from the crystal lattice is more likely to have limited the

observed inter lobe shift than to have augmented it. “Closed” forms of protein kinase

domains are found when both A-loop phosphorylation and ATP (or analogue) are

present. The constitutive catalytic competence of EGFRK and our finding of the apo-

by guest on March 24, 2018

http://ww

w.jbc.org/

Dow

nloaded from

19

form’s A-loop in an “active” conformation are consistent. Both the crystal lattice and the

significant differences between erlotinib and ATP are likely to have limited the

magnitude of the inter-lobe shift we observe upon inhibitor binding, and suggest that

our erlotinib complex may not represent the fully closed form, but instead an

intermediate of the transition.

Our finding that Leu955 is concealed in the C-terminal lobe argues against a

direct binding role with either partnered receptors or accessory proteins. A more

plausible explanation for the impacts on transphosphorylation caused by substitution

for Leu955 is an allosteric effect caused by loss of the Leu955–C-lobe connection. It

seems somewhat more plausible that such allosteric effects would arise in the C-

terminal tail following Leu955. The intermolecular contact regions present in our crystal

lattice are larger than is typical in protein structures. They are clearly required to form

these crystals. The relevance of these contacts involving the C-terminal ordered region

is difficult to gauge. The lack of traceable density for the preceding 13 residues presents

the possibility that our arbitrary assignment of “self” and “neighbor” proteins is

incorrect. The larger of the two contacts of the C-terminal ordered region is at a site on

the “neighbor” protein that includes parts from both N- and C-lobes and the connecting

segment between them, and thus large changes in the inter-lobe relationship may

disrupt this interaction. Such a scenario is consistent with previous results

demonstrating in vitro dimer disruption upon ATP binding(46).

The present data do not permit us to conclude that these contacts also arise in a cellular

context, but the high local concentration of membrane-bound EGFR and evidence for

non-ligand induced dimers(4) suggest close association between some parts of the ICDs

is likely.

by guest on March 24, 2018

http://ww

w.jbc.org/

Dow

nloaded from

20

Acknowledgement

We thank M. Ultsch, S. Hymowitz, and M. Franklin for data collection help, I.

Massova for help with analysis, as well as F. Rotella and the staff at SBC, and T. Earnest,

K. Henderson and staff at ALS. Use of the Argonne National Laboratory Structural

Biology Center beamlines at the Advanced Photon Source, was supported by the U. S.

Department of Energy, Office of Biological and Environmental Research, under

Contract No. W-31-109-ENG-38. The Advanced Light Source is supported by the

Director, Office of Science, Office of Basic Energy Sciences, Materials Sciences Division,

of the U.S. Department of Energy under Contract No. DE-AC03-76SF00098 at Lawrence

Berkeley National Laboratory.

by guest on March 24, 2018

http://ww

w.jbc.org/

Dow

nloaded from

21

REFERENCES

1. Schlessinger, J., and Ullrich, A. (1992) Neuron 9, 383-391

2. Hunter, T. (2000) Cell 100, 113-127

3. Ross, J. S., and Fletcher, J. A. (1998) Stem Cells 16, 413-428

4. Moriki, T., Maruyama, H., and Maruyama, I. N. (2001) J. Mol. Biol. 311, 1011-1026

5. Schlessinger, J. (2000) Cell 103, 211-225

6. Yarden, Y., and Sliwkowski, M. X. (2001) Nature Rev. Mol. Cell Biol. 2, 127-137

7. Jones, J. T., Akita, R. W., and Sliwkowski, M. X. (1999) FEBS Lett. 447, 227-231

8. Karunagaran, D., Tzahar, E., Beerli, R. R., Chen, X., Graus-Porta, D., Ratzkin, B. J.,

Seger, R., Hynes, N. E., and Yarden, Y. (1996) EMBO J. 15, 254-264

9. Sliwkowski, M. X., Schaefer, G., Akita, R. W., Lofgren, J. A., Fitzpatrick, V. D.,

Nuijens, A., Fendly, B. M., Cerione, R. A., Vandlen, R. L., and Carraway, K. L., III.

(1994) J. Biol. Chem. 269, 14661-14665

10. Guy, P. M., Platko, J. V., Cantley, L. C., Cerione, R. A., and Carraway, K. L., III.

(1994) Proc. Natl. Acad. Sci. USA 91, 8132-8136

11. Hubbard, S. R., and Till, J. H. (2000) Ann. Rev. Biochem. 69, 373-398

12. Chantry, A. (1995) J. Biol. Chem. 270, 3068-3073

13. Schaefer, G., Akita, R. W., and Sliwkowski, M. X. (1999) J. Biol. Chem. 274, 859-866

14. Penuel, E., Akita, R. W., and Sliwkowski, M. X. (2002) J. Biol. Chem. in press

15. Khazaie, K., Schirrmacher, V., and Lichtner, R. B. (1993) Cancer Metast. Rev. 12, 255-

274

16. Shawver, L. K., Slamon, D., and Ullrich, A. (2002) Cancer Cell 1, 117-123

17. Druker, B. J., Sawyers, C. L., Kantarjian, H., Resta, D. J., Reese, S. F., Ford, J. M.,

Capdeville, R., and Talpaz, M. (2001) New Eng. J. Med. 344, 1038-1042

18. Schindler, T., Bornmann, W., Pellicena, P., Miller, W. T., Clarkson, B., and Kuriyan, J.

(2000) Science 289, 1938-1942

19. Arteaga, C. L. (2001) J. Clin. Oncol. 19, 32S-40S

20. Kelloff, G. J., Fay, J. R., Steele, V. E., Lubet, R. A., Boone, C. W., Crowell, J. A., and

Sigman, C. C. (1996) Cancer Epidem. Biomark. Prevent. 5, 657-666

21. Woodburn, J. R. (1999) Pharma. Therapeut. 82, 241-250

22. Shewchuk, L., Hassell, A., Wisely, B., Rocque, W., Holmes, W., Veal, J., and Kuyper,

L. F. (2000) J. Med. Chem. 43, 133-138

by guest on March 24, 2018

http://ww

w.jbc.org/

Dow

nloaded from

22

23. Owen, D. J., and Evans, P. R. (1998) Science 282, 1327-1332

24. Garrett, T. P., McKern, N. M., Lou, M., Frenkel, M. J., Bentley, J. D., Lovrecz, G. O.,

Elleman, T. C., Cosgrove, L. J., and Ward, C. W. (1998) Nature 394, 395-399

25. Palmer, B. D., Trumppkallmeyer, S., Fry, D. W., Nelson, J. M., Showalter, H. D. H.,

and Denny, W. A. (1997) J. Med. Chem. 40, 1519-1529

26. Wissner, A., Berger, D. M., Boschelli, D. H., Floyd, M. B., Greenberger, L. M., Gruber,

B. C., Johnson, B. D., Mamuya, N., Nilakantan, R., Reich, M. F., Shen, R., Tsou, H. R.,

Upeslacis, E., Wang, Y. F., Wu, B. Q., Ye, F., and Zhang, N. (2000) J. Med. Chem. 43,

3244-3256

27. Ullrich, A., Coussens, L., Hayflick, J. S., Dull, T. J., Gray, A., Tam, A. W., Lee, J.,

Yarden, Y., Libermann, T. A., and Schlessinger, J. (1984) Nature 309, 418-425

28. Otwinowski, Z., and Minor, W. (1996) Methods in Enzymol. 276, 307-326

29. CCP4. (1994) Acta Crystallogr. D50, 760-763

30. Read, R. J. (1986) Acta Crystallogr. A42, 140-149

31. McRee, D. E. (1999) Practical Protein Crystallography, 2 Ed., Academic Press, San

Diego

32. Yamaguchi, H., and Hendrickson, W. A. (1996) Nature 384, 484-489

33. Hubbard, S. R. (1997) EMBO J. 16, 5572-5581

34. Mohammadi, M., Schlessinger, J., and Hubbard, S. R. (1996) Cell 86, 577-587

35. Gotoh, N., Tojo, A., Hino, M., Yazaki, Y., and Shibuya, M. (1992) Biochem. Biophys.

Res. Comm. 186, 768-774

36. Owen, D. J., Noble, M. E., Garman, E. F., Papageorgiou, A. C., and Johnson, L. N.

(1995) Structure 3, 467-482

37. Xu, R. M., Carmel, G., Sweet, R. M., Kuret, J., and Cheng, X. (1995) EMBO J. 14, 1015-

1023

38. Ogawa, A., Takayama, Y., Sakai, H., Chong, K. T., Takeuchi, S., Nakagawa, A.,

Nada, S., Okada, M., and Tsukihara, T. (2002) J. Biol. Chem. 277, 14351-14354

39. Kobe, B., Heierhorst, J., Feil, S. C., Parker, M. W., Benian, G. M., Weiss, K. R., and

Kemp, B. E. (1996) EMBO J. 15, 6810-6821

40. Timms, J. F., Noble, M. E., and Gregoriou, M. (1995) Biochem. J. 308, 219-229

41. Chen, W. S., Lazar, C. S., Lund, K. A., Welsh, J. B., Chang, C. P., Walton, G. M., Der,

C. J., Wiley, H. S., Gill, G. N., and Rosenfeld, M. G. (1989) Cell 59, 33-43

by guest on March 24, 2018

http://ww

w.jbc.org/

Dow

nloaded from

23

42. Wells, A., Welsh, J. B., Lazar, C. S., Wiley, H. S., Gill, G. N., and Rosenfeld, M. G.

(1990) Science 247, 962-964

43. Bridges, A. J. (2001) Chem. Rev. 101, 2541-2571

44. Garcia-Echeverria, C., Traxler, P., and Evans, D. B. (2000) Med. Res. Rev. 20, 28-57

45. Rewcastle, G. W., Denny, W. A., Bridges, A. J., Zhou, H., Cody, D. R., McMichael, A.,

and Fry, D. W. (1995) J. Med. Chem. 38, 3482-3487

46. Basu, M., Sen-Majumdar, A., Basu, A., Murthy, U., and Das, M. (1986) J. Biol. Chem.

261, 12879-12882

by guest on March 24, 2018

http://ww

w.jbc.org/

Dow

nloaded from

24

Figure Legends

1. The EGFRK structure with key features indicated. Erlotinib is found in the cleft

between amino-terminal and carboxy-terminal lobes. The C-terminal ordered region

including residues Leu977 to Pro995 is not included here.

2. Activation loops. The close structural correspondence between the EGFRK A-loop

(blue) and the A-loop from the phosphorylated form of the insulin receptor kinase(33)

(gold). The hydrophobic interaction between Lys836 and Tyr845 almost exactly reprises

that between Arg1155 and Tyr1163 of p-IRK (underlined). The presence of four

glutamate residues in this part of EGFRK has been suggested as a cause for its intrinsic

catalytic activity.

3. Representative electron density from the EGFRK/erlotinib structure. Map (2Fo-Fc,

contoured at 1.0 rmsd) in part of the A-loop, with additional residues Arg812

(immediately precedes the catalytic Asp813) and Arg808 (H-bonds to main chain of

Gly839). The placement of the Glu848 side chain is not supported by electron density,

but is among possible low energy conformers. Extra electron density at the Tyr845

hydroxyl suggests that it interacts with a solvent molecule, Glu848, or both. Glu848 in

the conformer shown mimics the phosphate of p-Tyr1163 of the activated insulin

receptor kinase (Fig 2).

4. The LVI tripeptide segment of EGFRK is found in close association with the C-lobe. A

solvent accessible surface from EGFRK with LVI removed is depicted. Residue Leu955,

by guest on March 24, 2018

http://ww

w.jbc.org/

Dow

nloaded from

25

the most important as gauged by mutagenesis studies, is found within what, in its

absence, would be a hydrophobic pit.

5. Stereo view of the intermolecular contacts of the extreme C-terminus of the

crystallized construct, showing solvent-accessible surfaces and worm representations of

the protein backbones. The reference molecule (blue) is shown from Leu977 to Pro995.

There are 13 residues preceding Leu977, and two residues following Pro995, that were

too poorly ordered to be fit. The yellow molecule shares a contact area of about 1300 Å2

(each side) with the blue one, with specific contacts between Asp985/Lys715 ‘(main

chain/main chain), Val987/Lys715’ (main chain/main chain), Asp988/Gln767 ‘(side

chain/side chain), Asp990/(Lys822’+Lys828’)(side chain/side chain) and

Tyr992/Glu712’ (side chain/side chain). The green molecule shares about 1000 Å2 (each

side) with the blue one, and contacts include Asp985/Lys946’’ (side chain/side chain),

Glu991/Lys799’’ (side chain/side chain), and Leu993/Arg938’ ‘(main chain/side chain).

The charge complimentarity for these interactions is good, with a net negative charge

on the blue molecule and net positive charges for contact regions from both neighboring

molecules. Because of the disordered 13 amino acids preceding Leu977, the assignment

of “reference” and neighbor is arbitrary, with the result that the blue depicted here may

also be considered to belong to either the yellow or the green molecules.

6. Stereo view of the inhibitor binding site and nearby residues from EGFRK/erlotinib.

Dashed line indicates an H-bond from the Met769 amide nitrogen to erlotinib. The light

blue sphere is a water molecule. The electron density for the side chains of Cys751 and

Asp831 was modeled using two conformers, but only one is depicted for clarity.

by guest on March 24, 2018

http://ww

w.jbc.org/

Dow

nloaded from

26

Table I. Structure Statistics for EGFRK/erlotinib and apo-EGFRK

Data Collection and ReductionResolution(Å) Nmeas1 Nref2 Complete3 I/σ Rmerge4 Rwork5 Rfree6

EGFRK/erlotinib30.0–5.59 19226 1744 100 35 0.054 0.268 0.3225.59–4.44 18910 1663 100 38 0.058 0.179 0.2364.44–3.88 18897 1664 100 37 0.067 0.192 0.2133.88–3.53 18717 1674 100 31 0.088 0.218 0.2913.53–3.28 18369 1656 100 22 0.130 0.224 0.3093.28–3.08 18139 1663 100 14 0.234 0.267 0.2783.08–2.93 17534 1626 100 8.8 0.334 0.366 0.3712.93–2.80 17629 1648 100 5.6 0.519 0.340 0.4392.80–2.69 17751 1655 100 3.6 0.792 0.516 0.4982.69–2.60 17173 1635 100 2.8 0.585 0.57730.0–2.60 182345 16628 100 27 0.093 0.251 0.294apo-EGFRK50.0–5.60 28108 1689 100 64 0.043 0.261 0.3235.60–4.45 29446 1642 100 80 0.055 0.186 0.2584.45–3.88 29452 1626 100 66 0.094 0.201 0.2223.88–3.53 29430 1617 100 43 0.149 0.221 0.2393.53–3.28 29555 1605 100 25 0.270 0.230 0.3093.28–3.08 30123 1625 100 14 0.245 0.243 0.3363.08–2.93 23682 1582 100 7.7 0.346 0.259 0.2652.93–2.80 17036 1617 100 4.4 0.436 0.266 0.3092.80–2.69 16220 1612 100 2.4 0.732 0.407 0.5012.69–2.60 15093 1568 100 1.8 0.964 0.454 0.55350.0–2.60 248145 16183 100 34 0.108 0.238 0.286

Refinement Statistics contents of model rms deviations

residues atoms7 waters bonds angles B-factors(bonded atoms)

EGFRK/erlotinib 314 2560(37) 20 0.011Å 1.5˚ 7.7 Å2

apo-EGFRK 307 2469(30) 17 0.010Å 1.4˚ 6.2 Å2

1 Nmeas is the total number of observations measured.2 Nref is the number of unique reflections measured at least once.3 Complete is the percentage of possible reflections actually measured at least once.4 Rmerge = Σ||I| - |<I>||/Σ|<I>|, where I is the intensity of a single observation and <I> the averageintensity for symmetry equivalent observations. The value for the highest resolution shell ofEGFRK/erlotinib is not reported by SCALEPACK.5 Rwork = Σ|Fo-Fc|/Σ|Fo|, where Fo and Fc are observed and calculated structure factor amplitudes,respectively.6 Rfree = Rwork for 679 (apo-EGFRK) or 689 (EGFRK/erlotinib) reflections sequestered from refinement.7 Number in parenthesis is number of atoms assigned zero occupancy.

by guest on March 24, 2018

http://ww

w.jbc.org/

Dow

nloaded from

Glycine rich loopDFG

catalytic loop

activation loop

Leu955

His964

N

Stamos, Sliwkowski & EigenbrotFigure 1

by guest on March 24, 2018

http://ww

w.jbc.org/

Dow

nloaded from

Glu848

Tyr845

Glu844Lys843

Glu842

Glu841

Arg808

Lys836

pTyr1163

Arg1155

Stamos, Sliwkowski & EigenbrotFigure 2

by guest on March 24, 2018

http://ww

w.jbc.org/

Dow

nloaded from

Arg808

Arg812

Glu848

Tyr845

Lys836

Arg808

Arg812

Glu848

Tyr845

Lys836

Stamos, Sliwkowski & EigenbrotFigure 3

by guest on March 24, 2018

http://ww

w.jbc.org/

Dow

nloaded from

Leu955

Val956

Ile957

Stamos, Sliwkowski & EigenbrotFigure 4

by guest on March 24, 2018

http://ww

w.jbc.org/

Dow

nloaded from

Pro995

Asp982

Glu710’

Lys713’

Glu939"

Lys946"

Pro995

Asp982

Glu710’

Lys713’

Glu939"

Lys946"

Stamos, Sliwkowski & EigenbrotFigure 5

by guest on March 24, 2018

http://ww

w.jbc.org/

Dow

nloaded from

Met769

Cys751

Thr766

Leu820

Thr830

Glu738

Met742

Asp831

Leu764

Lys721Val702

Leu694

Cys773

Ala719

Leu768

Met769

Cys751

Thr766

Leu820

Thr830

Glu738

Met742

Asp831

Leu764

Lys721Val702

Leu694

Cys773

Ala719

Leu768

Stamos, Sliwkowski & EigenbrotFigure 6

by guest on March 24, 2018

http://ww

w.jbc.org/

Dow

nloaded from

Jennifer Stamos, Mark X. Sliwkowski and Charles Eigenbrota4-anilinoquinazoline inhibitor

Structure of the EGF receptor kinase domain alone and in complex with

published online August 23, 2002J. Biol. Chem. 

  10.1074/jbc.M207135200Access the most updated version of this article at doi:

 Alerts:

  When a correction for this article is posted• 

When this article is cited• 

to choose from all of JBC's e-mail alertsClick here

by guest on March 24, 2018

http://ww

w.jbc.org/

Dow

nloaded from