Structural basis for high-affinity HER2 receptorbinding by an engineered proteinCharles Eigenbrota,b,1, Mark Ultscha, Anatoly Dubnovitskyc, Lars Abrahmsénd,1, and Torleif Härdc,1

aDepartment of Structural Biology and bDepartment of Antibody Engineering, Genentech Inc., 1 DNA Way, South San Francisco, CA 94080; cDepartmentof Molecular Biology, Swedish University of Agricultural Sciences, Uppsala Biomedical Center, SE-751 24 Uppsala, Sweden; and dAffibody AB,Lindhagensgatan 133, SE-112 51 Stockholm, Sweden

Edited by Adriaan Bax, National Institutes of Health, Bethesda, MD, and approved July 6, 2010 (received for review April 14, 2010)

The human epidermal growth factor receptor 2 (HER2) is specif-ically overexpressed in tumors of several cancers, including anaggressive form of breast cancer. It is therefore a target for bothcancer diagnostics and therapy. The 58 amino acid residue ZHER2affibody molecule was previously engineered as a high-affinitybinder of HER2. Here we determined the structure of ZHER2 in solu-tion and the crystal structure of ZHER2 in complex with the HER2extracellular domain. ZHER2 binds to a conformational epitopeon HER2 that is distant from those recognized by the therapeuticantibodies trastuzumab and pertuzumab. Its small size and lack ofinterference may provide ZHER2 with advantages for diagnostic useor even for delivery of therapeutic agents to HER2-expressingtumors when trastuzumab or pertuzumab are already employed.Biophysical characterization shows that ZHER2 is thermodynami-cally stable in the folded state yet undergoing conformational in-terconversion on a submillisecond time scale. The data suggest thatit is the HER2-binding conformation that is formed transiently priorto binding. Still, binding is very strong with a dissociation constantKD ¼ 22 pM, and perfect conformational homogeneity is thereforenot necessarily required in engineered binding proteins. A compar-ison of the original Z domain scaffold to free and bound ZHER2structures reveals how high-affinity binding has evolved duringselection and affinity maturation and suggests how a compromisebetween binding surface optimization and stability and dynamicsof the unbound state has been reached.

protein engineering ∣ molecular recognition ∣ protein–protein interactions ∣protein conformational dynamics ∣ cancer therapy

The human epidermal growth factor receptor 2 (HER2, ErbB2)is a 185-kDa transmembrane glycoprotein receptor tyrosine

kinase involved in the signal transduction pathways leading to cellgrowth and differentiation (1). In contrast to the other threemembers of the epidermal growth factor receptor family, HER2is thought to be an orphan receptor, i.e. lacking a ligand (2). How-ever, HER2 forms heterodimers with any of the related receptors,resulting in receptor activation. Enhanced levels of HER2 havebeen shown to correlate with one form of aggressive breast can-cer, precipitating the development of the HER2-binding mono-clonal antibodies (mAb) trastuzumab (3) and pertuzumab (4).Trastuzumab is now established for treating breast cancers shownto be HER2 overexpressing. Pertuzumab binds to a separate siteon HER2 (4). It is more efficient at blocking HER2 heterodimer-ization (4, 5), and it is in clinical trials for its effects on breastcancer and other cancers (6). Trastuzumab conjugated to themicrotubule-depolymerizing agent maytansinoid (trastuzumab-DM1) was recently shown to be even more efficacious than nativetrastuzumab, leading to high response rates in patients who failedtrastuzumab therapy (7). These findings agree with previous dataindicating that cancer cells do not loose HER2 expression whenthey become refractory to trastuzumab, but that the biologicalcontext needed for efficacy of trastuzumab has been lost (8, 9).Thus, the role of the antibody moiety in trastuzumab-DM1 isprimarily to work as a carrier of a cytotoxin. This finding raisesinterest in evaluating other HER2-binding molecules displaying

alternative properties compared to antibodies, including biodis-tribution. One such molecule is the affibody molecule ZHER2:342(hereafter ZHER2), which binds HER2 with very high affinity[KD ¼ 22 pM (10)].

Affibody molecules were designed as an alternative to immu-noglobulin-based binding proteins. The parental “Z domain” is asmall protein of 58 amino acids organized into a three-helix bun-dle. It was developed as a stabilized variant of the “B domain” ofthe IgG-binding staphylococcal protein A (11). The original affi-body sequence diversity libraries were made by randomizing 13surface-exposed amino acid residues on the IgG-binding surfaceof helices 1 and 2 (12). Affibody molecules binding a variety ofdifferent target proteins have subsequently been selected fromsuch libraries (13).

The high affinity of ZHER2 for its HER2 target is a result ofso-called affinity maturation, which is a second round of combi-natorial protein engineering and phage display selection (10, 13).ZHER2 competes with neither trastuzumab (14) nor pertuzumab(15) for HER2 binding. ZHER2 also does not appear to mediatedetectable biological effects (16). These properties make it sui-table as a tracer when the aim is to study HER2 without affectingits function, even in the presence of therapeutic monoclonal anti-bodies. Hence, ZHER2 allows for molecular imaging of HER2(15, 17) without interfering with ongoing therapy using either ofthe two mAbs. Alternatively, it may be used in synergy with any ofthese, if used as a carrier to direct additional therapeutic agentsto their HER2 target.

The potential of the ZHER2 affibody molecule for diagnosticand therapeutic use motivates detailed characterization of itsstructural and biophysical properties. Previous structural studiesof affibody molecules (18–20) have revealed a large variability instructure and binding interface, and it is therefore also interestingto learn how the binding surface evolves during affinity matura-tion. We find that ZHER2 adopts the standard three-helix bundletopology, but that the extensive mutagenesis in this case results intwo or more similar conformations that interconvert on the sub-millisecond time scale. This dynamics is apparently not prohibi-tive of picomolar binding. We argue that it appears as a trade-offin affinity maturation when, for instance, intramolecular hydro-gen bonds are sacrificed in order to optimize the HER2 bindingsurface. The crystal structure of the complex between ZHER2 andHER2 allows comparison of the structures of free and bound

Author contributions: C.E., L.A., and T.H. designed research; C.E., M.U., A.D., and T.H.performed research; L.A. contributed new reagents/analytic tools; C.E., M.U., A.D., andT.H. analyzed data; and C.E., L.A., and T.H. wrote the paper.

Conflict of interest statement: L.A. is employed at Affibody AB.

This article is a PNAS Direct Submission.

Freely available online through the PNAS open access option.

Data deposition: The HER2/ZHER2, ZHER2, and ZHER2_ALT structures have been deposited inthe Protein Data Bank, www.pdb.org (PDB ID codes 3MZW, 2KZI, and 2KZJ, respectively).1To whom correspondence may be addressed. E-mail: eigenbrot.c@gene.com,lars.abrahmsen@affibody.com, or torleif.hard@molbio.slu.se.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1005025107/-/DCSupplemental.

www.pnas.org/cgi/doi/10.1073/pnas.1005025107 PNAS ∣ August 24, 2010 ∣ vol. 107 ∣ no. 34 ∣ 15039–15044

BIOCH

EMISTR

Y

Dow

nloa

ded

by g

uest

on

Sep

tem

ber

15, 2

020

ZHER2 and shows that the binding site on HER2 is distantfrom the epitopes previously determined for trastuzumab andpertuzumab.

ResultsSecondary Structure Content and Stability of Free ZHER2. The circulardichroism (CD) spectrum of ZHER2 (Fig. 1A) shows that α-heli-cal secondary structure is dominating. The contribution from thedisordered histidine tag to the CD at 222 nm is expected to beminimal (21). If the CD signal at 222 nm originates only fromthe 58-residue affibody domain, then the mean residue ellipticityMRE222 ¼ −26;100 deg cm2 dmol−1, which corresponds to a

helix content of 73% or ca. 42 residues (22). This value comparesto 77% (ca. 45 residues) in the parental Z domain (23).

The thermal stability of the present affibody molecule is alsogood with a melting temperature of 67 °C (Fig. 1B), which is closeto the 75 °C of the Z domain at pH 7.2. We measured foldingstability at room temperature by monitoring intrinsic proteinfluorescence as function of guanidinium chloride denaturant con-centration (Fig. 1C). The data can be fitted to a two-state modelin which the folded state is stabilized by a free energy ΔGunfold ¼3.1� 0.3 kcalmol−1. This value is lower than that of the Zdomain or the B domain [∼7 kcalmol−1; (24, 25)]. It may be ar-gued that a two-state model of unfolding is inappropriate when

190 200 210 220 230 240 250

Wavelength (nm)

Elli

ptic

ity (

mde

g)

-20

0

-20

40

A B

D

6789

115

120

125

n.a.

n.a.

Q55N6

N24 Q25N3Q0

N43N11N23Q26 minor

Q40N21

N52

minor

V59

K58

D60

N21

S39

G-10

S-9

D37

Q55S33

S41

R28F30

K27D52

N43L22

A17

L45I31

minor

N3E15

D53

A46Q40K50

E47

Q26 Q25

N23

Q0V1

L-1

D2

A42L44

L34A48

A29L51A54

A56L19

R10

I16

C

Elli

ptic

ity (

mde

g)

-10

-15

-20

-25

-30

20 30 40 50 60 70 80 90

Temperature (°C)

Tm = 66.8 ±0.2 °C

0 1 2 3 4 5 6 7

GdmCl concentration (M)

200

300

400

Flu

ores

cenc

e (a

.u.)

∆Gunf = 3.1 ±0.3 kcal/mol

0 200 400 600 800 10000

10

20

30

40

50

60

5 10 15 20 25 30 35 40 45 50 550

10

20

30

40

50

60

E

F

15N

R2

and

R2

+ R

ex (

s-1 )

15N

R2ef

f (s-

1 )

CPMG field strength (Hz)

Residue No

Ile16Ser33

Ala54Leu19

Tyr35Gln25

α1 α2 α3

15N resonance linewidth Black: < 28 Hz

Green: 28 - 40 Hz Red: > 40 HzBlue: doublet

1H Chemical shift (ppm)

15N

Che

mic

al s

hift

(ppm

)

Fig. 1. Biophysical characterization of the ZHER2 affibody molecule. (A) Circular dichroism spectrum at 20 °C. (B) Thermal denaturation monitored by CD at222 nm. The red line is a fit to a two-state folding model to determine the melting temperature Tm ¼ 66.8� 0.2 °C. (C) Chemical denaturation monitored byintrinsic fluorescence at 352 nm. The line is a fit with ΔGunfold ¼ 3.1� 0.3 kcalmol−1 and a value ofm ¼ −1.2� 0.1 kcalmol−1 M−1. (D) 15N HSQC NMR spectrumof ZHER2 at 900MHz and 30 °C. Resonances are colored according to apparent 15N resonance linewidths. These reflect four categories of backbone dynamics, asdiscussed in the text. n.a., two unassigned resonances; minor, three pairs of resonances from Asn or Gln side chains in unassigned alternate conformations.Many backbone amide resonances are broadened beyond detection by conformational exchange dynamics. (E) 15N R2 relaxation dispersion at 800 MHz(81.1 MHz 15N frequency). Effective R2 rates (R2

eff) of 38 resolved and assigned amide correlations were measured as a function of a constant-time CPMGspin lock field. Twenty-three resonances showed relaxation dispersion (99% confidence level) and were fitted (global R-square ¼ 0.981) to a two-state modelwith kex ¼ 3574� 206 s−1 to determine R2 relaxation rates and Rex. Others, such as Ala54 in the figure, were fitted to a straight line to determine R2.(F) Summary of global fit of all 15N relaxation dispersion profiles. Black bars, R2; red circles, R2 þ Rex with estimated standard deviations. α-helical secondarystructure is indicated at the top.

15040 ∣ www.pnas.org/cgi/doi/10.1073/pnas.1005025107 Eigenbrot et al.

Dow

nloa

ded

by g

uest

on

Sep

tem

ber

15, 2

020

there is more then one folded conformation, which we will show isthe case here. However, one must consider the folded state(s) as asingle thermodynamic state as long as interconverting conforma-tions cannot be distinguished by the experiments, and as long asthe two-state model fits the data as in Fig. 1 B and C.

The NMR Spectrum and Dynamics of Free ZHER2. The assigned 15NNMR heteronuclear sequential quantum correlation (HSQC)spectrum of ZHER2 at 30 °C is characterized by a large variationin NMR line shapes. These reveal at least four categories of dy-namics on different time scales, as indicated by the coloring ofresonances in Fig. 1D. Disordered residues at the N terminusare dynamic at the picosecond time scale with sharp NMR reso-nances. The residues in helix 3, most of helix 2, and in loopregions appear to be well ordered within the folded protein asjudged from linewidths and a large number of NOE connectivities(Fig. S1). Resonances from helix 1, and from mutated regions ofhelix 2, are broadened due to conformational exchange on themicrosecond to millisecond time scales. Some of these are tooweak to be assigned, and most are in fact broadened beyond de-tection. Several Asn and Gln side chain resonances are also broa-dened, and there are at least three sets of unassigned resonancesfrom Gln or Asn amino (15NH2−) groups in alternative slowlyexchanging conformations. The same pattern of dynamics is ap-parent in the 13C HSQC spectrum, where several resonances areundetectable and resonances from methyl groups at the hydro-phobic interface of helices 1 and 2 are observable, but broadened.

We used NMR relaxation dispersion experiments (26) to de-termine the rate of conformational exchange. As expected fromthe line broadening, many amide 15N resonances show a strongdependence of the R2 relaxation rate on an external CPMG(Carr–Purcell–Meiboom–Gill) field (Fig. 1E). Measured relaxa-tion dispersion profiles could be fit to a two-state model of fastexchange (27) using a single global value of the exchange ratekex ¼ 3574� 206 s−1, which is the sum of forward and reverseinterconversion rates. Best-fit R2 relaxation rates and dispersionamplitudes confirm that helix 1 and helix 2 are involved in thisinterconversion, whereas loop regions and helix 3 are less

affected (Fig. 1F). The populations of interconverting states can-not be resolved in the fast exchange limit unless there is addi-tional information on chemical shift differences. A global fit of800- and 600-MHz relaxation data to the full Carver–Richardsequation was still attempted to determine both populationsand shift differences, but it did not converge.

Resonances from C-terminal residues 59 and 60 appear asdoublets in the NMR spectrum (Fig. 1D), indicating that theycan adopt two slowly exchanging conformations. These are notpart of the parental Z domain and on the DNA level they corre-spond to a restriction site introduced to enable facile multimer-ization. Conformational heterogeneity at the C terminus isunlikely to be of relevance for ZHER2 stability or binding.

Solution Structure of Free ZHER2.As many as 13 amide resonances inhelices 1 and 2 could not be detected and assigned (Fig. 1D), butother resonances in most of the corresponding side chains couldstill be sequentially assigned based on medium range dαβði;iþ 3ÞNOEs in the 3D 13C-NOESY spectrum, and yet others were as-signed based on residue type (for instance, Ala12). An ensembleof structures was calculated using simulated annealing (SA) basedon distance restraints derived from NOE connectivities, back-bone dihedrals derived from NMR chemical shifts, and backboneand side chain hydrogen bonds, which were added at the finalstage of refinement (SI Text and Table S1). The structure of mostof the molecule, including helix 3, most of helix 2, loop regions,and the hydrophobic core, is well defined (Fig. 2 A and B).

However, the backbone conformation of helix 1 and aroundmutated regions of helix 2 could not be calculated at high preci-sion. Hence, helix 1 appears as two helical fragments connectedby a loop at residues 13 and 14 in minimized SA structures, andthere are two conformations of residues 1 to 12 (Fig. 2 B and C).Still, chemical shifts and medium range NOE connectivities of allobservable resonances in helix 1 suggest that it indeed could becompletely helical. To examine this possibility, we calculated aset of alternative ZHER2 structures (ZHER2_ALT) in which anα-helical conformation of residues 6 to 14 was enforced, withoutremoving or changing any experimental restraints. These SA

D

90°

α1α2

α3

5 10 15 20 25 30 35 40 45 50 55

5 10 15 20 25 30 35 40 45 50 55

A

B

α1α2

α3

N

C

α1 α2 α3

α3

α2 HER2bindingsurface

EC

Fig. 2. Structure and dynamics of the ZHER2 affibody molecule. (A) Sequences of the Z domain and ZHER2. Gray, randomized positions in affibody libraries;yellow, selected side chains at HER2 interface; cyan, other residues at HER2 interface; black boxes, dynamic nonpolar side chains in core; red boxes, Asn and Glnside chains with alternating conformations. Insets above and below sequences show secondary structure and surface exposure, with darker blue shadingreflecting less surface exposure. Residues for which backbone NMR resonances in ZHER2 are broadened by dynamics are in gray and ordered regions in orange(matching coloring in C–E). (B) Backbone traces of the ZHER2 structure ensemble. The blue/gray colors of residues 1 to 12 at the N terminus show two conformersobtained with unrestrained SA. (C) ZHER2 conformational exchange dynamics: orange, no backbone dynamics; gray, backbone dynamics; white spheres, methylgroups at the interface between helices 1 and 2 that undergo conformational dynamics; sticks, dynamic Asn and Gln side chains. (D) The alternative ZHER2_ALTstructure with an intact helix 1. (E) Illustration of how dynamics (as in D) coincides with mutations at the HER2 binding surface (yellow sticks corresponding toyellow shading in A).

Eigenbrot et al. PNAS ∣ August 24, 2010 ∣ vol. 107 ∣ no. 34 ∣ 15041

BIOCH

EMISTR

Y

Dow

nloa

ded

by g

uest

on

Sep

tem

ber

15, 2

020

calculations also converged to high-quality structures withoutviolating any of the ca. 100 experimental restraints involvinghelix 1 (Table S1).

The reason why unbiased SA results in a distorted rather thanan intact helix 1 is that the former is slightly favored by the struc-ture calculation force field (conformational plus restraint ener-gies of −499� 20 versus −472� 24 kcalmol−1, respectively).However, an evaluation of Ramachandran statistics (Table S1)and packing of the hydrophobic core (Fig. 2D) suggest that anintact helix 1 is more favored. We conclude that both conforma-tions are possible. It is not unlikely that the observed dynamicsinvolves interconversion between these two conformations, asdiscussed below.

X-Ray Crystal Structure of HER2/ZHER2. The complex between theHER2 extracellular domain (HER2ecd) and ZHER2 was deter-mined at 2.9-Å resolution (Fig. 3 and Table S2). It shows ZHER2bound to an epitope that includes residues from both domain IIIand domain IVof HER2. This region is not part of the epitopes inthe previously characterized complexes of HER2ecd with Fabfragments of therapeutic antibodies trastuzumab or pertuzumab(Fig. 3A), although it is part of a crystal packing interaction in thepertuzumab-Fab/HER2 complex (4). The conformational ar-rangement of HER2ecd domains is essentially the same as in pre-vious studies (2, 4, 28). Upon ZHER2 binding about 1;250 Å2 ofsolvent accessible surface (1.4-Å probe radius), of which 58% isnonpolar, is sequestered on each side of the HER2ecd/ZHER2interface, and the shape complementarity statistic is Sc ¼ 0.65.There are 25 residues from HER2ecd and 14 residues fromZHER2 that are within 4 Å of the binding partner. These form11 direct interprotein hydrogen bonds with distances up to3.3 Å (Fig. 3B and Table S3). These metrics are all consistent withthe high-affinity binding.

The conformation of ZHER2 in the crystal structure includes acontinuous helix 1 as in the alternate ZHER2_ALT NMR struc-ture, which provides close contacts to HER2 along its entirelength. The rms backbone differences between bound-stateZHER2 and the ZHER2 and ZHER2_ALT free-state structures are1.4 and 1.0 Å, respectively, for alpha carbons of residues 6 to 56,

and ∼3.0 and 0.9 Å for residues 6 to 13 (in helix 1). There is alsomostly good correspondence with side chain conformations inbound-state ZHER2 and ZHER2_ALT, whereas side chain corre-spondence to the (nonalternate) solution structure is relativelypoor for residues in helix 1. The mean temperature factor (B)for ZHER2 is the same as that for the entire complex structure,but B values are relatively high at the beginning of helix 1. ZHER2residues preceding helix 1 were not resolved in electron densitymaps and are absent from the final coordinates.

DiscussionConformational Dynamics in Free ZHER2. The present HER2-bindingaffibody molecule is thermodynamically stable and a very strongbinder, but it undergoes conformational interconversions on thesubmillisecond time scale in the free state. There is a clearcorrelation between sites of mutation and dynamics (Fig. 2 Aand C–E). ZHER2 has a melting temperature of 67 °C, but it issomewhat less stable to unfolding than the parental Z domainor the B domain of staphylococcal protein A. Still, the conforma-tional dynamics is not likely to be global unfolding, because onewould then not expect the localized effects of line broadening ob-served here. Studies of the B domain show that global proteinunfolding begins with helix 1 (29). It is therefore possible thatthe observed dynamics is due to local unfolding of helix 1.

However, the average helicity observed by CD implies that amajor fraction (>99%) of ZHER2 exists as a three-helix bundle at20 °C. An alternative explanation of the dynamics, which wefavor, is therefore that it reflects interconversions between twoor more folded conformations of helix 1. It is possible thattwo of these are the ones shown in Fig. 2 C and D, both of whichare consistent with experimental data. Interconversion betweenonly two conformations is supported by the 15N R2 relaxation dis-persion data, but the presence of additional conformational statescan on the other hand not be ruled out.

The fact that line broadening makes interconverting structuresnearly indistinguishable in structure calculations can be ex-plained: NMR resonances that would distinguish between confor-mations would be expected to experience the largest changes inchemical shift upon interconversion. These are therefore most

II

I

III

IV

pertuzumab Fab

trastuzumab Fab

HER2ecd

ZHER2

Asn6

Met9Arg10

Tyr13Trp14

Tyr35

Arg32

Pro38

Ala17

Asp36

Arg28

Asn24

Ile31

Leu18

A B

Fig. 3. The ZHER2 affibody epitope at the junction of domains III and IV on HER2. (A) Crystal structure of the complex between HER2ecd (surface colored bydomain) and ZHER2 (orange). For reference, Fab fragments of the therapeutic monoclonal antibodies trastuzumab (blue) and pertuzumab (cyan) are shownwhere they bind to HER2ecd. ZHER2 contacts HER2ecd on parts of domains III and IV far from both antibody epitopes. (B) Detailed view of ZHER2 binding. Thebackbone of residues 6 to 39 is colored in orange and side chains that interact with HER2ecd are shown as sticks (helix 3 omitted for clarity). Side chains thathave been randomized and selected by phage display are shownwith yellow color on carbon and side chains that are not varied, but still interact at the bindingsurface, are colored in cyan on carbons as in Fig. 2A. Three varied side chains (Asn11, Gln25, and Lys27; not displayed) do not interact with HER2. Some of theobserved intermolecular hydrogen bonds (Table S3) are shown as dashed lines.

15042 ∣ www.pnas.org/cgi/doi/10.1073/pnas.1005025107 Eigenbrot et al.

Dow

nloa

ded

by g

uest

on

Sep

tem

ber

15, 2

020

affected by exchange line broadening, and the critical experimen-tal discriminants (chemical shifts and NOEs) remain undetectedas they disappear with line broadening.

Previous structural studies of affibody molecules in complexwith target proteins reveal that the three-helix scaffold of theZ domain may remain intact in the complex (19, 20), but thatthe extensive mutagenesis also can lead to the selection of othervery different conformations (18). Affibody molecules may alsodiffer in their folding stability and even be largely disorderedwhile still showing strong binding (30). The present HER2-bind-ing affibody molecule represents yet another variation on thesethemes: It is thermodynamically stable and a very strong binder,but there is dynamics in mutated regions.

Conformational dynamics in engineered binding proteins willaffect binding affinity, though the effect does not have to be large.For instance, if there are two equally populated interconvertingstates of ZHER2 of which one binds the HER2 receptor, the bind-ing affinity is reduced by only 50%, in this case from a theoreticalKD ¼ 11 pM to the observed KD ¼ 22 pM. Still, the dynamicsleaves room for further optimization of binding affinity, perhapsby optimizing conformational stability as with the ZSPA-1 affibodymolecule (31).

A possible reason for destabilization and dynamics in ZHER2 isthe replacement of Gln9 at the N-terminal end with Met9 andGlu24/Glu25/Asn28 at the C-terminal end with Asn24/Gln25/Arg28. This is because as many as eight hydrogen bonds in theZ domain are lost by these mutations, and only a few alternativehydrogen bonds form (Fig. 4). These replacements can thereforeaccount for destabilization as well as dynamics of helix 1 and 2and alternative conformations of Asn and Gln side chains at theC-terminal end (Figs. 1D and 2 C and D). Homologous substitu-tions at positions 9, 24, and 25 could not be identified in initialselections from the full library (32). These were therefore againrandomized in the affinity maturation library resulting in selec-tion of Met9 and Asn24 that together with the native Asn6and the previously selected Arg28 participate in interactions withHER2 (Figs. 3B and 4). Thus, although reinstating original resi-dues might stabilize ZHER2, it will probably also affect bindingaffinity. We therefore suggest that affinity maturation hasinvolved a trade-off between binding surface optimization andfolding stability and/or structural homogeneity of unbound

molecules in the library. Or in other words, stability has beensacrificed during the selections in order that new side chainscan interact with HER2, and with free-state conformationallability and dynamics as by-products.

Bound-State ZHER2 Conformation and Interactions. The complexstructure determined using X-ray crystallography is of sufficientlyhigh resolution to permit confident placement of side chains.There is a good correspondence between the residues varied dur-ing initial selections and affinity maturation and those found inintimate association with HER2ecd. Of the 13 randomized aminoacid positions, 10 are within 4 Å and 6 interact with HER2 via 11hydrogen bonds (Fig. 3B and Fig. S2). Overall, the bound-statecoordinates of ZHER2 are clearly more similar to those of thealternate ZHER2_ALT than to the ZHER2 free-state structures,and most differences occur in regions that are dynamic in the freestate. This is consistent with a transient population of the bound-state conformation in free ZHER2. Furthermore, relatively hightemperature factors (B values) are observed for the first part ofhelix 1, where the two NMR structures differ the most. Hence,some of the flexibility observed with monomeric ZHER2 in solu-tion probably remains in complex with HER2. The similarity be-tween bound ZHER2 and ZHER2_ALT extends to almost all sidechains at the binding surface. Exceptions are Trp14, for which theconformation is the same as in only 4 of 23 structures of theZHER2_ALT ensemble, and Tyr35, which attains a ZHER2_ALT

side chain conformation different from that in the bound state(χ1 ¼ 180° versus 90°). Interactions with HER2 are clearly res-ponsible for the conformations of these side chains in the com-plex, indicating that some conformational adaptation occurs alsoat the binding surface.

The high-affinity ZHER2 binding is accounted for by the struc-ture. For instance, the binding surface is 50% larger than in any ofthe four other affibody molecules that have been structurallycharacterized (18–20). This is because 10 of 13 varied side chainsare involved, but also because interactions extend toward theN-terminal end of the molecule (Fig. 3B; e.g., Asn6 and Pro38).The Ile31 side chain, which is native to the Z domain, is centrallylocated at the ZHER2 binding surface as in all other bound-stateaffibody or Z domain structures. It was suggested that Ile31

Z domain free ZHER2 bound ZHER2

Gln9Asn6 Asn6 Asn6

Met9Met9

Glu24

Asn24Asn24

Gln26 Gln26 Gln26Glu25 Gln25

Gln25

Arg28Arg28Asn28

A

BHelix 1Helix 2

HER2 bindingsurface

HER2binding surface

Fig. 4. Trade-off between binding affinity and intramolecular hydrogen bonding during affinity maturation of ZHER2. (A) Hydrogen bonding (dashed lines) atthe C-terminal end of the Z domain (Left), free ZHER2 (Middle), and bound ZHER2 (Right). Replacements at positions 24, 25, and 28 result in HER2 interactions byAsn24 and Arg28 on the expense of wild-type Z domain hydrogen bonds. (B) Hydrogen bonds formed by Gln9 at the N-terminal end of the Z domain (Left) arelost upon replacement with Met9 in ZHER2 (Middle), but both Met9 and Ans6 can instead interact with HER2 (Right).

Eigenbrot et al. PNAS ∣ August 24, 2010 ∣ vol. 107 ∣ no. 34 ∣ 15043

BIOCH

EMISTR

Y

Dow

nloa

ded

by g

uest

on

Sep

tem

ber

15, 2

020

mediates some intrinsic property of Z domain binding that is in-herited in affibody molecules (19).

The HER2/ZHER2 interface also reveals how the evolution ofZHER2 into a high-affinity binder has occurred. First, residuesArg10 (or a Lys10), Trp13, Tyr14, Arg28, Arg32, and Tyr35 wereall present in two initial binders obtained from a full library with13 randomized positions (32). These, together with Ile31, Asp36,and Pro38 that were not varied, already constitute the core bind-ing surface (Fig. 3B) that supports dissociation constants ofKD≈50 nM and KD≈140 nM, respectively. The maturation li-brary then allowed for optimization around the core HER2 inter-face. It contained six completely randomized positions and twopositions with a more restricted variation. The side chains ofMet9, Ala17, Leu18, and Asn24 at four of the six completely ran-domized positions were selected because they extend the inter-face and Arg10 was this time present in all selected binders.Presumably also the nonvaried Asn6 became involved at thispoint. Side chains at three remaining randomized positions thatdid not participate in binding then “returned” to those present inthe original Z domain (Asn11) or as homologous side chains(Gln25 at the position of Glu25, and Lys27 replacing Arg27)to complete the high-affinity binder (10).

ZHER2 Recognizes a Previously Unexploited Epitope on the HER2 Recep-tor. The ZHER2 binding site on HER2 is distant from those ofboth trastuzumab and pertuzumab. This is consistent with thelack of interference between either trastuzumab or pertuzumaband ZHER2. Any antibodies that may have been raised againstthis epitope would probably not be well studied due to an absenceof any effect on HER2 function. Nonetheless, the present struc-ture demonstrates that this and other nominally occult epitopesare available for exploitation for imaging or for targeted deliveryof linked chemotherapeutic agents. The relatively small size ofZHER2 and the associated reduced potential for steric overlapwith therapeutic antibodies make it attractive for use as a mole-cular imaging tracer during antibody treatment. Indeed, a

molecular imaging agent based on ZHER2 has been investigatedclinically and has been shown to yield images of HER2 over-expressing metastases within hours after injection in patientsalso undergoing trastuzumab therapy (17).

Materials and MethodsZHER2 with an N-terminal His6 tag (10) used for biophysical and NMR studieswas produced in Escherichia coli and purified by immobilized metal ionchromatography followed by size exclusion chromatography (SEC). NMRwas performed at 30 °C on a ca. 1 mM uniformly 13C;15N-labeled ZHER2 sam-ple in 160 mM NaCl and 16 mM potassium phosphate at pH 6.0 with 0.1%NaN3 and 5% D2O. Structures were calculated using simulated annealingwith distance restraints derived from NOEs, backbone dihedral angle re-straints derived from chemical shifts, and hydrogen bond restraints derivedfrom hydrogen bonds observed in initially calculated structures.

The 58 amino acid ZHER2 peptide used for X-ray crystallography was pre-pared by solid phase synthesis. HER2ecd was expressed in Chinese hamsterovary cells and purified by affinity chromatography using HERCEPTIN®,diethylaminoethane anion exchange, and SEC. Crystals of the HER2ecd:ZHER2complex formed in sitting drops at 10 mg∕mL in 0.1 M NaCl, 5 mM MOPS(pH 7.3) and reservoir containing 15%wt∕vol PEG 3350, 0.1M sodium acetate(pH 5.0), 0.2 M ammonium acetate. Diffraction data extending to 2.9-Åresolution was collected at ESRF beamline ID23-2. The structure was solvedby molecular replacement.

Complete descriptions of protein production, biophysical characteriza-tion, NMR spectroscopy, and crystallography are in SI Text.

ACKNOWLEDGMENTS. We thank Jeremy Murray and MXpress for collectingdiffraction data, Clifford Quan for peptide synthesis, and Prof. VladislavOrekhov at the Swedish NMR Centre at the University of Gothenburg forassistance with NMR experiments. Portions of this research were carriedout at the Stanford Synchrotron Radiation Lightsource (SSRL), a national userfacility operated by Stanford University on behalf of the U.S. Department ofEnergy, Office of Basic Energy Sciences. The SSRL Structural Molecular BiologyProgram is supported by the Department of Energy, Office of Biological andEnvironmental Research, and by the National Institutes of Health,National Center for Research Resources, Biomedical Technology Program,and the National Institute of General Medical Sciences. The research wasin part supported by a grant from the Swedish Research Council (to T.H.).

1. Olayioye MA, Neve RM, Lane HA, Hynes NE (2000) The ErbB signaling network:Receptor heterodimerization in development and cancer. EMBO J 19:3159–3167.

2. Garrett TP, et al. (2003) The crystal structure of a truncated ErbB2 ectodomain revealsan active conformation, poised to interact with other ErbB receptors. Mol Cell11:495–505.

3. Cho H-S, et al. (2003) Structure of the extracellular region of HER2 alone and incomplex with the Herceptin Fab. Nature 421:756–760.

4. Franklin MC, et al. (2004) Insights into ErbB signaling from the structure of the ErbB2-pertuzumab complex. Cancer Cell 5:17–28.

5. Scheuer W, et al. (2009) Strongly enhanced antitumor activity of trastuzumab andpertuzumab combination treatment on HER2-positive human xenograft tumormodels. Cancer Res 69:9330–9336.

6. Makhija S, et al. (2010) Clinical activity of gemcitabine plus pertuzumab in platinum-resistant ovarian cancer, fallopian tube cancer, or primary peritoneal cancer. J ClinOncol 28:1215–1223.

7. Phillips GDL, et al. (2008) Targeting HER2-positive breast cancer with trastuzumab-DM1, an antibody-cytotoxic drug conjugate. Cancer Res 68:9280–9290.

8. Berns K, et al. (2007) A functional genetic approach indentifies the PI3K pathway as amajor determinant of trastuzumab resistance in breast cancer. Cancer Cell 12:395–402.

9. Junttila TT, et al. (2009) Ligand-independent HER2/HER3/PI3K complex is disrupted bytrastuzumab and is effectively inhibited by the PI3K inhibitor GDC-0941. Cancer Cell15:429–440.

10. Orlova A, et al. (2006) Tumor imaging using a picomolar affinity HER2 binding affibodymolecule. Cancer Res 66:4339–4348.

11. Nilsson B, et al. (1987) A synthetic IgG-binding domain based on staphylococcalprotein A. Protein Eng 1:107–113.

12. Nord K, et al. (1997) Binding proteins selected from combinatorial libraries of anα-helical bacterial receptor domain. Nat Biotechnol 15:772–777.

13. Grönwall C, Ståhl S (2009) Engineered affinity proteins—generation and applications.J Biotechnol 140:254–269.

14. Lee SB, et al. (2008) Affibody molecules for in vivo characterization of HER2-positivetumors by near-infrared imaging. Clin Cancer Res 14:3840–3849.

15. Tolmachev V, et al. (2007) Radionuclide therapy of HER2-positive micro-xenograftsusing a 177Lu-labeled HER2-specific Affibody molecule. Cancer Res 67:2773–2782.

16. Ekerljung L, et al. (2008) Dimeric HER2-specific affibodymolecules inhibit proliferationof the SKBR-3 breast cancer cell line. Biochem Biophys Res Comm 377:489–494.

17. Baum RP, et al. (2010) Molecular imaging of HER2-expressing malignant tumors inbreast cancer patients using 111 In- 68Ga-labeled Affibody molecules. J Nucl Med51:892–897.

18. Hoyer W, Grönwall C, Jonsson A, Ståhl S, Härd T (2008) Stabilization of a β-hairpin inmonomeric Alzheimer’s amyloid-β peptide inhibits amyloid formation. Proc Natl AcadSci USA 105:5099–5104.

19. Lendel C, Dogan J, Härd T (2006) Structural basis for molecular recognition in anaffibody:affibody complex. J Mol Biol 359:1293–1304.

20. Wahlberg E, et al. (2003) An affibody in complex with a target protein: Structure andcoupled folding. Proc Natl Acad Sci USA 100:3185–3190.

21. Kelly SM, Price NC (1997) The application of circular dichroism to studies of proteinfolding and unfolding. Biochim Biophys Acta 1338:161–185.

22. Scholtz JM, Quian H, York EJ, Stewart JM, Baldwin RL (1991) Parameters of helix-coiltransition theory for alanine-based peptides of varying chain lengths in water.Biopolymers 31:1463–1470.

23. Dincbas-Renqvist V, Lendel C, Dogan J, Wahlberg E, Härd T (2004) Thermodynamics offolding, stabilization, and binding in an engineered protein–protein complex. J AmChem Soc 126:11220–11230.

24. Bai Y, Dyson HJ, Wright PE (1997) Absence of a stable intermediate on the foldingpathway of protein A. Protein Sci 6:1449–1457.

25. Cedergren L, Andersson R, Jansson B, Uhlén M, Nilsson B (1993) Mutational analysis ofthe interaction between staphylococcal protein A and human IgG1. Protein Eng6:441–448.

26. Mulder FA, Mittermaier A, Hon B, Dahlquist FW, Kay LE (2001) Studying excited statesof proteins by NMR spectroscopy. Nat Struct Biol 8:932–935.

27. Palmer AG, Massi F (2006) Characterization of the dynamics of biomolecules usingrotating-frame spin relaxation NMR spectroscopy. Chem Rev 106:1700–1719.

28. Bostrom J, et al. (2009) Variants of the antibody herceptin that interact with HER2 andVEGF at the antigen binding site. Science 323:1610–1614.

29. Bottomley SP, et al. (1994) The stability and unfolding of an IgG binding protein basedupon the B domain of protein A from Staphylococcus aureus probed by tryptophansubstitution and fluorescence spectroscopy. Protein Eng 7:1463–1470.

30. Hoyer W, Härd T (2008) Interaction of Alzheimer’s Aβ peptide with an engineeredbinding protein—thermodynamics and kinetics of coupled folding-binding. J Mol Biol378:398–411.

31. Wahlberg E, Härd T (2006) Conformational stabilization of an engineered bindingprotein. J Am Chem Soc 128:7651–7660.

32. Wikman M, et al. (2004) Selection and characterization of HER2/neu-binding affibodyligands. Prot Eng Des Sel 17:455–462.

15044 ∣ www.pnas.org/cgi/doi/10.1073/pnas.1005025107 Eigenbrot et al.

Dow

nloa

ded

by g

uest

on

Sep

tem

ber

15, 2

020