Bispecific antibody generated with sortase and clickchemistry has broad antiinfluenza virus activityKoen Wagnera,1, Mark J. Kwakkenbosa,1, Yvonne B. Claassena, Kelly Maijoora, Martino Böhnea,Koenraad F. van der Sluijsb, Martin D. Wittec,2, Diana J. van Zoelend, Lisette A. Cornelissend, Tim Beaumonta,Arjen Q. Bakkera, Hidde L. Ploeghc, and Hergen Spitsa,3

aAIMM Therapeutics, 1105 BA Amsterdam, The Netherlands; bLaboratory of Experimental Intensive Care and Anesthesiology, Academic Medical Center,University of Amsterdam, 1105 AZ Amsterdam, The Netherlands; cWhitehead Institute for Biomedical Research, Cambridge, MA 02142; and dDepartment ofVirology, Central Veterinary Institute, Wageningen University and Research Centre, 8200 AB Lelystad, The Netherlands

Edited by K. Christopher Garcia, Stanford University, Stanford, CA, and approved October 21, 2014 (received for review May 9, 2014)

Bispecific antibodies have therapeutic potential by expanding thefunctions of conventional antibodies. Many different formats ofbispecific antibodies have meanwhile been developed. Most aregenetic modifications of the antibody backbone to facilitateincorporation of two different variable domains into a singlemolecule. Here, we present a bispecific format where we havefused two full-sized IgG antibodies via their C termini using sor-tase transpeptidation and click chemistry to create a covalentlylinked IgG antibody heterodimer. By linking two potent anti-influenza A antibodies together, we have generated a full anti-body dimer with bispecific activity that retains the activity andstability of the two fusion partners.

antibody engineering | immunotherapy | influenza

With a steady increase of antibodies and antibody derivativessuch as antibody drug conjugates and bispecific antibodies

entering the clinic, monoclonal human antibodies are now anestablished source of new therapeutic agents (1, 2). The developmentof bispecific antibodies has generated particular interest, becauseit allows expansion of basic antibody functions (3, 4). Throughbinding two (or more) different targets, a bispecific antibody cansimultaneously engage two epitopes of a disease agent, block/acti-vate multiple ligands/receptors at once, or recruit immune effectorcells (i.e., T cells or B cells) to a specific (tumor) site (5). There isa growing interest in bispecific antibodies with anticancer proper-ties, which has led to an increase in bispecifics that have enteredpreclinical testing (5, 6).Bispecific antibodies with defined functions are generated by

means of genetic or biochemical engineering. Many differentmethods exist to engineer immunoglobulins, with more than 45bispecific antibody formats at last count (reviewed in ref. 5).These bispecific antibody formats fall into three broad subclasses(5): (i) single-chain double variable domain formats (50–100kDa) (7–9): Generally these bispecifics consist of multiple vari-able domains that are connected via peptide linkers. (ii) IgG withmultiple variable domains: In this type of bispecific antibody,a second variable domain is genetically linked to any desirableposition in the IgG molecule (i.e., the C or N terminus of eitherthe IgG heavy or light chain) (10–12). (iii) Asymmetric IgGmolecules: In an asymmetric IgG antibody, two different vari-able domains are incorporated into a single, asymmetric, anti-body molecule via heterodimerization of the constant domains.Heterodimerization may be achieved through engineering theCH3 domain (13–16) or the hinge region of the antibody (17,18). Depending on the engineering method, asymmetric IgGscan be made with a common light chain or with two differentlight chains (19).Each of these formats has its specific advantages and draw-

backs. Most of the limitations arise from the fact that their for-mats deviate significantly from the natural, highly stable, IgGstructure, which compromises stability and ease of manufacture.

Here, we present a bispecific antibody format, in which twoantibodies are fused at their C termini, using a combination ofsortase transpeptidation and click chemistry (20), to create an IgGheterodimer. This C-C fusion does not require mutations within theantibody constant domains that might interfere with Fc-receptorbinding or that would compromise antibody stability. Thus, thenative antibody structure is fully retained in our format.C-to-C fusion is a two-step process (Fig. 1), using a combina-

tion of sortase transpeptidation and click chemistry (20). Sortaseis a bacterial enzyme that functions to attach cell surface proteinsbearing an “LPXTG” motif to the cell wall of Gram-positive bac-teria via transacylation (21, 22). Sortase-catalyzed transpeptidationallows for efficient site-specific modifications under physiologicalconditions, with excellent specificity and near-quantitative yields(23–25). To facilitate site-specific linking of the C termini of twoantibodies, the fusion partners are labeled with either an azide ora cyclooctyne (DIBAC) functional group. The modified proteinsare then conjugated via a strain-promoted cycloaddition betweenthe azide and the cyclooctyne. This reaction is highly specific andreadily proceeds at room temperature in aqueous environments atneutral pH (26), allowing for efficient fusion under mild conditions.To test the robustness of this process and determine the features

of this bispecific antibody format, we fused two potent anti-influenza antibodies, each active against a different subgroup ofthe influenza A virus. Based on the hemagglutinin (HA) protein

Significance

Bispecific antibodies expand the function of conventional anti-bodies. However, therapeutic application of bispecifics is ham-pered by the reduced physiochemical stability of such molecules.We present a format for bispecific antibodies, fusing two full-sized antibodies via their C termini. This format does not requiremutations in the antibody constant domains beyond installationof a five-residue tag, ensuring that the native antibody structureis fully retained in the bispecific product. We have validated theapproach by linking two anti-influenza A antibodies, each activeagainst a different subgroup of the virus. The bispecific antibodydimer retains the activity and the stability of the two originalantibodies.

Author contributions: K.W., M.J.K., T.B., H.L.P., and H.S. designed research; K.W., M.J.K.,Y.B.C., K.M., M.B., and A.Q.B. performed research; K.F.v.d.S., M.D.W., D.J.v.Z., and L.A.C.contributed new reagents/analytic tools; K.W., M.J.K., and H.S. analyzed data; and K.W.,M.J.K., H.L.P., and H.S. wrote the paper.

Conflict of interest statement: K.W., M.J.K., Y.B.C., K.M., M.B., T.B., A.Q.B., and H.S. areemployees of AIMM Therapeutics.

This article is a PNAS Direct Submission.1K.W. and M.J.K. contributed equally to this work.2Present address: Bio-Organic Chemistry, Stratingh Institute for Chemistry, University ofGroningen, 9747 AG Groningen, The Netherlands.

3To whom correspondence should be addressed. Email: hspits@aimmtherapeutics.com.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1408605111/-/DCSupplemental.

16820–16825 | PNAS | November 25, 2014 | vol. 111 | no. 47 www.pnas.org/cgi/doi/10.1073/pnas.1408605111

sequence, there are 18 different subtypes of influenza A, dividedinto two subgroups (27, 28). The HA protein is the common targetof almost all neutralizing antibodies, and several antibodies withbroadly neutralizing activity between influenza A subtypes in thesame group exist (29–33). Combining two such potent subgroup-specific antibodies may result in an IgG heterodimer with evenbroader anti-influenza A activity. This type of molecule wouldhave therapeutic relevance in a passive immunization setting,because influenza viruses continue to cause significant morbidityand mortality, despite efforts to contain them with seasonalvaccines (34). Because these vaccines are typically only effectiveagainst the specific seasonal viral strain, there is an urgent unmetmedical need for new treatments active against multiple subtypesof the influenza virus (35).

ResultsIsolation and Characterization of Potent Antiinfluenza Antibodies. Toobtain broadly neutralizing influenza A antibodies, we isolatedmemory B cells from influenza-vaccinated individuals. These Bcells were transduced with human Bcl6 and Bcl-xL as describedin Kwakkenbos et al. (36, 37) and screened for HA binding. Bcells that recognized HA molecules from multiple influenzasubtypes were cloned and the antibody derived from them pro-duced in 293T HEK cells. This approach resulted in the iden-tification of two broadly neutralizing antibodies: AT10-002and AT10-005.AT10-005 neutralizes two H1N1 and one H5N1 virus (Table

1) and binds four additional group 1 viruses (two H1N1, oneH5N1, and one H9N2) (SI Appendix, Table S1). In an ELISA,

this antibody binds full-length H5 HA protein and the HA2subunit. The subunits of HA are described in SI Appendix, TableS2. AT10-005 contains the IGHV1-69 gene segment and harborsthe hydrophobic signature commonly found in group 1-specificantibodies (SI Appendix, Table S3) (38, 39). Antibody competitionusing AT10-005 and the stem-binding antibody CR6261 (40) (SIAppendix, Fig. S1) for H1 binding on H1N1 (A/Hawaii/31/2007)-infected cells confirms that both antibodies bind similar regions.AT10-002 is specific for HA proteins of group 2 viruses (SI

Appendix, Table S1) and shows neutralizing activity against fourgroup 2 viruses (two H3N2, one H7N1, and one H7N7 virus)(Table 1). The antibody binds full-length H3 but not the separateHA1 portion (SI Appendix, Table S4). In addition, AT10-002competes with the group 2 HA stem-specific antibody CR8020(31) for binding to H3N2-infected cells (SI Appendix, Fig. S2). Tofurther analyze the binding site of AT10-002, we have isolateda third HA-specific antibody: AT10-003. AT10-003 was found tobind to three H3 viruses (SI Appendix, Table S5), and reactedwith both the full-length H3 and the HA1 portion of the mole-cule indicating that the HA head is sufficient for binding (SIAppendix, Table S4). Notably, AT10-003 was unable to blockbinding of AT10-002 to H3N2-infected cells (SI Appendix, Fig.S2). Therefore, the finding that AT10-002 binding is blocked byCR8020 and not by AT10-003 suggests that the stem region ofgroup 2 HA influenza has the largest contribution to the AT10-002 epitope. Linking the broadly reacting antibodies AT10-005and AT10-002 would potentially result in a molecule activeagainst a broad spectrum of group 1 and group 2 influenza Aviruses.

Synthesis of the C-to-C Fused Bispecific Antibody Dimer. To enablethe C-to-C protein fusion, we modified the C termini of theheavy chains of both antiinfluenza antibodies with a small tag,consisting of a GGGGS (G4S) linker sequence, followed by thesortase recognition site LPETGG and a His6 tag. The His tag isused to monitor the sortase labeling reaction, because it will beremoved upon sortase-catalyzed transpeptidation. Triglycinepeptides containing either an azide or a DIBAC moiety weresynthesized as described (20). We chose to label AT10-002 withDIBAC and AT10-005 with azide. The extent of sortase labelingwas monitored by using a fluorescent azide-containing nucleo-phile (GGG-TAMRA-azide): We observed excellent labelingwith the azide-containing probe after 4 h at 41 °C in 25 mM Trisbuffer, pH 8.0 and 150 mM NaCl (Fig. 2A). For the TAMRA-DIBAC-modified peptide, sortase labeling is somewhat less ef-ficient (Fig. 2B). We attribute this difference to the nonspecificthiol-yne coupling reaction between DIBAC and free thiolgroups (41), such as the unpaired cysteine residues in the anti-body or in sortase (Fig. 2A). Because the active site of sortasecontains a free thiol (42), sortase itself would be particularlyvulnerable to the thiol-yne reaction (43). Nonetheless, weobserved efficient installation of DIBAC after 4-h incubationat 33 °C, in 25 mM Tris buffer, pH 6.8 and 150 mM NaCl,demonstrating that a lower incubation temperature and pHreduces thiol-yne coupling.Following sortase labeling, the antibodies were separated from

sortase, excess triglycine peptide, and other reaction productsby size exclusion chromatography. AT10-002–DIBAC was readilycoupled to AT10-005–azide at 20 °C. Antibody dimers were re-solved from any remaining antibody monomers by size exclu-sion chromatography (Fig. 2C). After prolonged incubation ofclick-labeled antibodies, a second peak was visible in the elutionprofile (Fig. 2C). This second peak consists of higher order an-tibody oligomers that may form when the two click-linked Ctermini of one antibody connect with two different antibodies (asshown in Fig. 2D). Formation of higher-order complexes alsooccurred when the click reactions were performed at highertemperatures or at a higher antibody concentration, suggesting

Fig. 1. Approach for synthesis of C-to-C–fused antibodies. (A) Antibodiesare labeled at the C terminus either with an azide (N3) or DIBAC with a clickpeptide by using sortase. (B) Click-labeled antibodies are fused via the clickreaction.

Table 1. Neutralizing activity of AT10-002 and AT10-005

Group Virus AT10-002 AT10-005

1 H1N1 (A/Hawaii/31/2007) — 1.29 (± 0.53)1 H1N1 (A/Neth./602/2009) — 16.8 (± 1.8)1 H5N1 (A/Turkey/Turkey/04) — 8.2 (± 6.1)2 H3N2 (A/Neth./177/2008) 0.76 (± 0.27) —

2 H3N2 (A/Swine/St. Oedenrode/1996) 1.79 (± 0.35) —

2 H7N1 (A/Chicken/Italy/1067/99) 22.3 (± 8.7) —

2 H7N7 (A/Chicken/Neth./621557/03) 0.78 (± 0.36) —

Shown are IC50 values in nanomolars. —, no neutralizing activitydetected. SI Appendix, Table S1 shows a complete overview of the reactivityof AT10-002 and AT10-005. Data represent mean ± SD of at least two in-dependent experiments.

Wagner et al. PNAS | November 25, 2014 | vol. 111 | no. 47 | 16821

IMMUNOLO

GYAND

INFLAMMATION

that this reaction requires careful control by choice of temper-ature and antibody concentration, to obtain the optimal yieldof the click-linked antibody dimer.We obtained pure and fully intact click-linked antibody dimers

as judged from analysis by SDS/PAGE (Fig. 2D). The presenceof a ∼100-kDa polypeptide (100 kDa equals twice the size of theantibody heavy chain) in the reduced dimer sample (Fig. 2D)confirms covalent fusion at the C termini of the heavy chains.The reduced dimer samples also showed some monomeric heavychains (Fig. 2D). The presence of these monomeric heavy chainsupon SDS/PAGE of the dimer sample demonstrates that not allantibody heavy chains have fused. This finding is not unexpected,because a single covalent HC-HC link suffices to generate anantibody dimer. This single HC-HC linkage is not due to in-complete sortase labeling; anti-His Western blotting demon-strates a near-complete loss of the His tag in the purified dimer(Fig. 2E). Apparently, after the first HC-HC linkage, a secondclick coupling between the two remaining modified HC C ter-mini is disfavored, presumably due to steric hindrance.

BiFlu Retains Structural Stability and Fc-Receptor Binding. To ensurethat the C-to-C fusion of two antibodies does not compromise

the stability of the resulting structure, we assessed the stabilityof our antibody dimer (BiFlu) by means of dynamic scanningfluorescence (DSF).The DSF curves (Fig. 3A) showed that the melting properties

of BiFlu are adequately described by summation of the curves ofthe two individual antibodies. Thermal stability of the two anti-bodies is therefore unchanged when they are linked via their Ctermini. In view of its small size (24 aa + the click product)relative to the mass of BiFlu (±320 kDa), the linker sequencethat connects the two antibodies contributes little if at all to theDSF curve of BiFlu.We further assessed the stability of the antibody dimer over

a more extended observation window. After 3 wk of incubation at37 °C in PBS (Fig. 3B) or in PBS plus (antibody-depleted) humanserum (Fig. 3C), the majority of BiFlu remains intact, suggestingthat the antibody dimer would also be stable at physiological con-ditions. Only a small amount of BiFlu (<< 10%) has decomposedinto the separate IgGmonomers. The covalent link between the twoC termini in the antibody dimer is stable under these conditions.In this bispecific full antibody dimer format, the two anti-

bodies are linked via the C termini of the heavy chain. Becausethe Fc portion contains the binding sites for the IgG Fc receptors

Fig. 2. Preparation of BiFlu. (A) Determining optimal reaction conditions for sortase labeling. ST-tagged antibody AT10-005-ST (1.0 μM) is mixed with sortase(2.0 μM) and GGG-TAMRA-azide or GGG-TAMRA-DIBAC (125 μM), in labeling buffer (25 mM Tris·HCl, 150 mM NaCl, and 10 mM CaCl2) and incubated for 6 h atthe indicated pH and temperature. C, control reaction (incubation with GGG-TAMRA-DIBAC at 37 °C, pH 7.5). After incubation, the reaction mixture isanalyzed with reducing fluorescent SDS/PAGE (λex, 532 nm; λem, 580 nm). (B) Quantification of sortase-labeling results. Fluorescence is measured as relative tolabeling with GGG-TAMRA-azide at 37 °C, pH 7.5 (set at 1.00 RU). (C) Gel filtration chromatogram of click-reaction products. AT10-002-DIBAC (5.0 μM) ismixed with AT10-005-aizde (5.0 μM). After the indicated time, a sample is analyzed with gel filtration chromatography (column volume: 120 mL). (D)Coomassie-stained SDS/PAGE gel of purified BiFlu (1.5 μg). (E) Anti-His Western blot of purified BiFlu (0.5 μg). HC, antibody heavy chain; LC, antibody lightchain; M, Dual Color Protein Standard (Bio-Rad); S, sortase. Products of thiol-yne coupling are indicated with asterisks.

16822 | www.pnas.org/cgi/doi/10.1073/pnas.1408605111 Wagner et al.

(FcγRs) and the neonatal Fc receptor (FcRn) (44), we verifiedthat such a fusion did not impair interaction of the antibodydimer with Fc receptors. We measured binding of BiFlu to sol-uble Fc receptors by ELISA and found that BiFlu binds FcRnand all three Fcγ receptors with similar affinity as the parentalIgGs (SI Appendix, Fig. S3). We also tested binding of BiFlu toTHP-1 cells, which express Fc receptors on their cell surface(45), and observed equivalent binding of BiFlu and the parentalantibodies (SI Appendix, Fig. S3).BiFlu retains FcR-binding activity, implying that it should be

capable of engaging FcR effector functions. If in vivo neutralizationof virus by the single antibodies were to somehow involve engage-ment with Fc receptors, then the bispecific antibody retains func-tionality for this parameter as well.

BiFlu Retains Functional Activity and Neutralizes Influenza A. Havingdetermined stability and receptor binding of BiFlu, we measured

its functional activity. The binding of BiFlu to HA proteins wastested with capture surface plasmon resonance (SPR), using ei-ther heavy or light chain-specific anti-IgG antibodies, followedby subsequent injections with H1 or H3 HA protein. H1 HArepresents group 1 Influenza A viruses and should be detected byBiFlu component AT10-005; H3 is a representative of group 2Influenza A and should be recognized by BiFlu componentAT10-002 (Fig. 4A).When immobilized via a heavy chain-specific antibody, BiFlu

bound both H1 and H3 HA with similar, low picomolar (KD = ±15 pM for both antigens) affinity, as did the parental antibodies(Fig. 4A and Table 2). In the same setup, we determined bindingto H7 and H9 HA protein, finding that BiFlu binds these anti-gens with an affinity similar to that of the parental antibodies(Table 2).When using anti-light chain antibodies for immobilization,

BiFlu could be captured on both anti-kappa and anti-lambda,because AT10-002 and AT10-005 have different light chains(lambda and kappa, respectively). BiFlu captured via its lightchains binds both H1 and H3 HA (Fig. 4B), demonstrating thatBiFlu is a bivalent heterodimer. The fact that light chain-cap-tured BiFlu binds both antigens with similar kinetics as the singleantibodies (here: AT10-002 + H3) indicates that, in the BiFlusample, the two antibodies are linked in a 1:1 ratio and thathomodimers must be absent. We estimate that BiFlu contains >95% heterodimers.We then tested the in vitro neutralization activity of BiFlu

against H3N2 and H1N1. BiFlu neutralized both strains effi-ciently (Fig. 5) with an IC50 value of ∼1.0 nM, similar to the twosingle antibodies tested separately (Table 3). We then tested thein vivo activity of our antibodies in a murine H1N1 (A/PR/8/1934) challenge model. Prophylactic administration of 1 mg/kgAT10-005 protected the mice against lethal infection, all micepretreated with AT10-002 or rituximab lost more then 25% oftheir body weight and were killed by day 8 (SI Appendix, Fig. S7).We examined the in vivo activity of BiFlu in the same challengemodel, injecting mice with either BiFlu (2 mg/kg) or a mixtureof AT10-005 and AT10-002 (1 mg/kg each). All BiFlu-treated micewere protected against H1N1 challenge, similar to the AT10-002 +AT10-005–treated mice, whereas 50% of the mice treated withrituximab failed to recover from the infection (Fig. 6A) [thecontrol group survival is different from the first experiment (SIAppendix, Fig. S7); we attribute this result to variation in themodel]. Mice injected with BiFlu or AT10-002 + AT10-005 alsoshowed a significantly lower body weight loss compared with thecontrol group (Fig. 6B).

Fig. 3. Stability of the BiFlu antibody dimer. (A) DSF curves of BiFlu andmonomeric IgG (25 μg/mL) in PBS buffer. (B) Long-term stability test ofBiFlu and monomeric IgG in PBS buffer. BiFlu or monomeric IgG (AT10-002or AT10-005) is diluted into PBS buffer (final concentration: 250 μg/mL) andincubated at 37 °C. After the indicated number of days, 2.5 μg of antibodyis heated for 10 min at 55 °C (the lower temperature is used to minimizeantibody breakdown) and analyzed with SDS/PAGE. (C) Long-term stabilitytest of BiFlu and monomeric IgG in IgG-depleted human serum. BiFlu ormonomeric IgG (mixture of AT10-002 and AT10-005) is diluted into IgG-depleted human serum (final concentration: 250 μg/mL) and incubated at37 °C. After the indicated time, 0.25 μg of antibody (+ serum) is heated for10 min at 55 °C and analyzed with anti-IgG HC Western blotting. In B and C,BiFlu and monomeric IgG are not fully denatured; therefore, they run ata lower molecular mass than expected (IgG molecular mass: 160 kDa).

Table 2. Kinetic constants for HA binding

Antibody ka kd KD

H1 (A/New Caledonia/20/1999)AT10-002 — — —

AT10-005 19.5 (± 1.8) 0.29 (± 0.10) 15.0 (± 5.8)BiFlu 21.4 (± 2.8) 0.34 (± 0.17) 16.2 (± 8.4)

H3 (A/Wyoming/03/2003)AT10-002 19.2 (± 1.5) 0.23 (± 0.08) 11.8 (± 5.9)AT10-005 — — —

BiFlu 22.2 (± 2.5) 0.33 (± 0.16) 14.7 (± 9.2)H7 (A/Netherlands/219/03)AT10-002 0.59 (± 0.15) 0.25 (± 0.05) 441 (± 109)AT10-005 — — —

BiFlu 0.57 (± 0.12) 0.21 (± 0.01) 375 (± 88)H9 (A/Hong Kong/1073/1999)AT10-002 — — —

AT10-005 0.81 (± 0.06) 21.6 (± 0.1) 27,000 (± 2,000)BiFlu 0.94 (± 0.02) 27.1 (± 0.2) 29,000 (± 1,000)

ka in 104 sec−1·M−1, kd in 10−5 sec−1, KD in picomolars; —, no bindingdetected. Data represent mean ± SD of at least two independent experi-ments. SPR curves and fits are shown in SI Appendix, Figs. S4 (H1 + H3binding), S5 (H7), and S6 (H9).

Fig. 4. Capture SPR analysis of HA binding. (A) HA binding of BiFlu, AT10-002, and AT10-005, immobilized on an anti-HC–coated spot. (B) HA bindingof BiFlu, AT10-002, and AT10-005, immobilized on an anti-lambda light-chain coated spot. First, antibody is injected, followed by HA antigens (H3,then H1), and then anti-light chain antibody (either anti-kappa or anti-lambda). Because BiFlu is twice as large as IgG, BiFlu gives a greater captureresponse when binding to the capture antibody (anti-HC or anti-lambda).Kinetic constants are shown in Table 2.

Wagner et al. PNAS | November 25, 2014 | vol. 111 | no. 47 | 16823

IMMUNOLO

GYAND

INFLAMMATION

To test the integrity of the BiFlu molecule at the time of viralinfection, we performed an anti-human IgG HC Western blot(Fig. 6C), demonstrating that BiFlu remained intact as a dimer.The human IgG concentration in the mice, 24 h after injection,was determined with ELISA (Fig. 6D). For both the BiFlu andthe antibody mixture group, we found approximately 9.5 μg ofhuman IgG/mL, indicating that BiFlu remained in the circulationat similar levels as the single antibodies.BiFlu has the ability to bind H1, H3, H7, and H9 HAs, and

exhibits neutralization potency against both H1 and H3. The activityof BiFlu is consistent with the combined activities of the individualparental antibodies, which neutralize a wide range of influenzasubtypes. Thus, we have created a bispecific antibody dimer capableof broad HA binding and potentially broad neutralization potency.

DiscussionWe have presented a bispecific antibody format, in which twofull-length IgG antibodies are joined at their C termini. Oneadvantage of this format is the stability of the C-C–linked IgGheterodimer, produced with minimal modification of the nativeIgG structure. The chemical structure of the C-C linkage includesthe sortase recognition site, plus a triazole moiety resulting from theclick reaction and eventual linker peptides. This product, like anyother nonnative protein modification, could be immunogenic,a notion that would require testing in a human host.The C-C–linked IgG heterodimer stands out from other IgG-

scFv formats; the latter bispecifics, in which the extra domainsare genetically fused to the antibody backbone, are often un-stable and aggregation-prone (46, 47). Several formats for asym-metric bispecific IgG antibody formats now exist. Antibodyasymmetry is facilitated through engineering of the CH3 domain(13–16) or the hinge region of the antibody (17, 18), promotingheterodimerization of the constant domains. Some of the constantdomain mutations required to enable IgG heterodimerizationcompromise stability and may affect binding to Fc receptors aswell (48). Solving these issues requires extensive antibody engi-neering (48). Also, an asymmetric IgG binds monovalently to itstarget because it contains only one copy of each variable domain,which may affect its activity. Production of these asymmetric IgGantibodies requires either a mild-reduction step to converthomodimers into heterodimers (17, 18) or coexpression of twodifferent antibodies (13–16), adding further complication.In contrast, the preparation of the C-C–fused IgG hetero-

dimers lends itself to large-scale manufacturing without loss ofproduct quality or the need for elaborate optimization. Theantibodies to be joined can be expressed separately as full-lengthantibodies, and the coupling reactions occur under physiologicalconditions. Likewise, sortase-catalyzed reactions enable large-scale preparation of modified biomolecules (49) and immuno-toxins (50). A panel of sortase-modified antibodies equippedwith click handles far more readily facilitates combinatorial

exploration of many different combinations of C-C–linked bis-pecific antibodies than genetic fusions would allow.Antibodies against the stem region of Influenza virus HA

antigens provide protection against virus in a prophylactic settingin animal models (30, 51) and based on this property, they arebeing tested in clinical trials. BiFlu combines the activities of twobroadly neutralizing antibodies into a single unit. From a de-velopmental and regulatory perspective, combining two anti-bodies into a single drug makes development less complex andmore cost-effective, because preclinical and clinical testing willbe reduced to a single molecule.

Materials and MethodsIsolation and Selection of Antiinfluenza Antibodies from Human B Cells. Im-mortalization of human memory B cells was performed as described (36, 37).Human memory B cells were isolated using FACS, out of peripheral bloodmononucleated cells (PBMCs) from an influenza vaccinated donor, and im-mortalized through retroviral transduction with a bicistronic construct cod-ing for Bcl6 and Bcl-xL. The use of human PBMCs was approved by theMedical Ethical Committee of the Academic Medical Center and was

Fig. 5. Virus neutralization assays. (A) Neutralization of influenza A H1N1virus (strain: A/Hawaii/31/2007). (B) Neutralization of influenza A H3N2 virus(strain: A/Netherlands/177/2008). Data represent mean ± SD of at least twoindependent experiments.

Table 3. IC50 values for influenza neutralization

Virus AT10-002 AT10-005 BiFlu

H1N1 (A/Hawaii/31/2007) — 1.29 (± 0.53) 0.36 (± 0.25)H3N2 (A/Neth./177/2008) 0.76 (± 0.27) — 1.37 (± 0.41)

IC50 values in nanomolars. —, no neutralizing activity detected. Data rep-resent mean ± SD of at least two independent experiments.

Fig. 6. In vivo protective activity of BiFlu. Kaplan–Meier survival curves (A)and body weight loss (mean ± SD) of C57BL/6J mice (B) that were i.v. injectedwith either BiFlu (2 mg/kg), AT10-002 + AT10-005 (1 mg/kg each), orrituximab (1 mg/kg). Twenty-four hours later, the mice were challengedintranasally with 50 μL of a 104.5 TCID50 H1N1 (A/PR/8/1934) preparation.Compared with mice treated with control antibody (rituximab), the survivaland body weight loss of mice treated with BiFlu or the AT10-002 + AT10-005antibody mixture was significantly improved (survival: P < 0.05; body weightBiFlu P < 0.01 from day 6, AT10-002 + AT10-005 P < 0.01 from day 3). Nosignificant difference was found between BiFlu- and AT10-002 + AT10-005–treated mice. (C) Integrity of BiFlu in mouse plasma, 24 h after injection.Samples containing 0.25 μg of antibody are heated for 10 min at 55 °C andanalyzed with anti-IgG HC Western blotting. (D) Antibody concentration inmouse plasma, 24 h after injection. IgG concentration of all mice in eachgroup (n = 6) is determined by ELISA.

16824 | www.pnas.org/cgi/doi/10.1073/pnas.1408605111 Wagner et al.

contingent on informed consent. B cells with reactivity to more than one HAtype were characterized for HA recognition by ELISA and binding to HA-expressing cells.

Preparation of SDS-PAGE and Western Blot Samples. Unless indicated other-wise, samples are prepared in 1x XT-sample buffer (BIORAD) and heated for10 min at 95 °C. If indicated, samples are reduced by adding 10 mM DTT.Mini-Protean TGX precast 4-20% gradient gels (BIORAD) were used forelectrophoresis.

SPR. SPR is performed on an IBIS Mx96 instrument (IBIS Technologies). Anti-human IgG heavy and anti-human light chain antibodies are immobilized onan amine-specific E2S gold-film SPR chip (Ssens Technologies) using a CFMmicrofluidics spotter device (Wasatch Microfluidics). Antibodies and full-

length HA-antigens are injected over the chip surface in cycles of concate-nated injections. Data is processed with SprintX software (IBIS Technologies).

Virus Neutralization Assay. MDCK-SIAT cells are incubated with virus andantibody. Cells are fixed 24 h after infection. The amount of infected cells isdetected with FITC-labeled antiinfluenza nuclear protein (NP) antibody; totalcell count is measured with DAPI staining.

Additional detailed information is described in SI Appendix, SI Materialsand Methods.

ACKNOWLEDGMENTS. We thank Carla Guimaraes and Juan-Jose Cragnolinifor their assistance with the purification of sortase and setting up the sortaselabeling assays. Chris Theile is acknowledged for synthesis of GGG-DIBAC.This work is supported by the FLUNIVAC programme, European CommissionFP7, Project no. 602604.

1. Nelson AL, Dhimolea E, Reichert JM (2010) Development trends for human mono-clonal antibody therapeutics. Nat Rev Drug Discov 9(10):767–774.

2. Reichert JM (2013) Which are the antibodies to watch in 2013? MAbs 5(1):1–4.3. Reichert JM (2011) Bispecific antibodies and ADCs: Once and future kings? MAbs 3(4):

329–330.4. Vincent KJ, Zurini M (2012) Current strategies in antibody engineering: Fc engineer-

ing and pH-dependent antigen binding, bispecific antibodies and antibody drugconjugates. Biotechnol J 7(12):1444–1450.

5. Kontermann RE (2012) Dual targeting strategies with bispecific antibodies.MAbs 4(2):182–197.

6. May C, Sapra P, Gerber HP (2012) Advances in bispecific biotherapeutics for thetreatment of cancer. Biochem Pharmacol 84(9):1105–1112.

7. Holliger P, Prospero T, Winter G (1993) “Diabodies”: Small bivalent and bispecificantibody fragments. Proc Natl Acad Sci USA 90(14):6444–6448.

8. Schoonjans R, et al. (2001) A new model for intermediate molecular weight re-combinant bispecific and trispecific antibodies by efficient heterodimerization ofsingle chain variable domains through fusion to a Fab-chain. Biomol Eng 17(6):193–202.

9. Kellner C, et al. (2008) A novel CD19-directed recombinant bispecific antibody de-rivative with enhanced immune effector functions for human leukemic cells.J Immunother 31(9):871–884.

10. Coloma MJ, Morrison SL (1997) Design and production of novel tetravalent bispecificantibodies. Nat Biotechnol 15(2):159–163.

11. Wu C, et al. (2007) Simultaneous targeting of multiple disease mediators by a dual-variable-domain immunoglobulin. Nat Biotechnol 25(11):1290–1297.

12. Dong J, et al. (2011) Stable IgG-like bispecific antibodies directed toward the type Iinsulin-like growth factor receptor demonstrate enhanced ligand blockade and anti-tumor activity. J Biol Chem 286(6):4703–4717.

13. Ridgway JBB, Presta LG, Carter P (1996) ’Knobs-into-holes’ engineering of antibodyCH3 domains for heavy chain heterodimerization. Prot. Eng 9(7):617–621.

14. Davis JH, et al. (2010) SEEDbodies: Fusion proteins based on strand-exchange en-gineered domain (SEED) CH3 heterodimers in an Fc analogue platform for asymmetricbinders or immunofusions and bispecific antibodies. Protein Eng Des Sel 23(4):195–202.

15. Gunasekaran K, et al. (2010) Enhancing antibody Fc heterodimer formation throughelectrostatic steering effects: Applications to bispecific molecules and monovalentIgG. J Biol Chem 285(25):19637–19646.

16. Schaefer W, et al. (2011) Immunoglobulin domain crossover as a generic approach forthe production of bispecific IgG antibodies. Proc Natl Acad Sci USA 108(27):11187–11192.

17. Strop P, et al. (2012) Generating bispecific human IgG1 and IgG2 antibodies from anyantibody pair. J Mol Biol 420(3):204–219.

18. Labrijn AF, et al. (2013) Efficient generation of stable bispecific IgG1 by controlledFab-arm exchange. Proc Natl Acad Sci USA 110(13):5145–5150.

19. Spiess C, et al. (2013) Bispecific antibodies with natural architecture produced by co-culture of bacteria expressing two distinct half-antibodies. Nat Biotechnol 31(8):753–758.

20. Witte MD, et al. (2012) Preparation of unnatural N-to-N and C-to-C protein fusions.Proc Natl Acad Sci USA 109(30):11993–11998.

21. Navarre WW, Schneewind O (1999) Surface proteins of gram-positive bacteria andmechanisms of their targeting to the cell wall envelope. Microbiol Mol Biol Rev 63(1):174–229.

22. Popp MW, Ploegh HL (2011) Making and breaking peptide bonds: Protein engi-neering using sortase. Angew Chem Int Ed Engl 50(22):5024–5032.

23. Antos JM, et al. (2009) Site-specific N- and C-terminal labeling of a single polypeptideusing sortases of different specificity. J Am Chem Soc 131(31):10800–10801.

24. Möhlmann S, Mahlert C, Greven S, Scholz P, Harrenga A (2011) In vitro sortagging ofan antibody fab fragment: Overcoming unproductive reactions of sortase with waterand lysine side chains. ChemBioChem 12(11):1774–1780.

25. Guimaraes CP, et al. (2013) Site-specific C-terminal and internal loop labeling ofproteins using sortase-mediated reactions. Nat Protoc 8(9):1787–1799.

26. Becer CR, Hoogenboom R, Schubert US (2009) Click chemistry beyond metal-catalyzedcycloaddition. Angew Chem Int Ed Engl 48(27):4900–4908.

27. Nobusawa E, et al. (1991) Comparison of complete amino acid sequences andreceptor-binding properties among 13 serotypes of hemagglutinins of influenzaA viruses. Virology 182(2):475–485.

28. Webster RG, Bean WJ, Gorman OT, Chambers TM, Kawaoka Y (1992) Evolution andecology of influenza A viruses. Microbiol Rev 56(1):152–179.

29. Sui J, et al. (2009) Structural and functional bases for broad-spectrum neutralizationof avian and human influenza A viruses. Nat Struct Mol Biol 16(3):265–273.

30. Ekiert DC, et al. (2009) Antibody recognition of a highly conserved influenza virusepitope. Science 324(5924):246–251.

31. Ekiert DC, et al. (2011) A highly conserved neutralizing epitope on group 2 influenzaA viruses. Science 333(6044):843–850.

32. De Marco D, et al. (2012) A non-VH1-69 heterosubtypic neutralizing human mono-clonal antibody protects mice against H1N1 and H5N1 viruses. PLoS ONE 7(4):e34415.

33. Friesen RHE, et al. (2014) A common solution to group 2 influenza virus neutraliza-tion. Proc Natl Acad Sci USA 111(1):445–450.

34. Lambert LC, Fauci AS (2010) Influenza vaccines for the future. N Engl J Med 363(21):2036–2044.

35. Wang TT, Palese P (2009) Universal epitopes of influenza virus hemagglutinins? NatStruct Mol Biol 16(3):233–234.

36. Kwakkenbos MJ, et al. (2010) Generation of stable monoclonal antibody-producing Bcell receptor-positive human memory B cells by genetic programming. Nat Med 16(1):123–128.

37. Kwakkenbos MJ, et al. (2014) Genetic manipulation of B cells for the isolation of raretherapeutic antibodies from the human repertoire. Methods 65(1):38–43.

38. Lingwood D, et al. (2012) Structural and genetic basis for development of broadlyneutralizing influenza antibodies. Nature 489(7417):566–570.

39. Avnir Y, et al. (2014) Molecular signatures of hemagglutinin stem-directed hetero-subtypic human neutralizing antibodies against influenza A viruses. PLoS Pathog10(5):e1004103.

40. Throsby M, et al. (2008) Heterosubtypic neutralizing monoclonal antibodies cross-protective against H5N1 and H1N1 recovered from human IgM+ memory B cells. PLoSONE 3(12):e3942.

41. van Geel R, Pruijn GJM, van Delft FL, Boelens WC (2012) Preventing thiol-yne additionimproves the specificity of strain-promoted azide-alkyne cycloaddition. BioconjugChem 23(3):392–398.

42. Ilangovan U, Ton-That H, Iwahara J, Schneewind O, Clubb RT (2001) Structure ofsortase, the transpeptidase that anchors proteins to the cell wall of Staphylococcusaureus. Proc Natl Acad Sci USA 98(11):6056–6061.

43. Ekkebus R, et al. (2013) On terminal alkynes that can react with active-site cysteinenucleophiles in proteases. J Am Chem Soc 135(8):2867–2870.

44. Shields RL, et al. (2001) High resolution mapping of the binding site on human IgG1for Fc γ RI, Fc γ RII, Fc γ RIII, and FcRn and design of IgG1 variants with improvedbinding to the Fc γ R. J Biol Chem 276(9):6591–6604.

45. Tsuchiya S, et al. (1980) Establishment and characterization of a human acutemonocytic leukemia cell line (THP-1). Int J Cancer 26(2):171–176.

46. Demarest SJ, Glaser SM (2008) Antibody therapeutics, antibody engineering, and themerits of protein stability. Curr Opin Drug Discov Devel 11(5):675–687.

47. Miller BR, et al. (2010) Stability engineering of scFvs for the development of bispecificand multivalent antibodies. Protein Eng Des Sel 23(7):549–557.

48. Von Kreudenstein TS, et al. (2013) Improving biophysical properties of a bispecificantibody scaffold to aid developability: quality by molecular design. MAbs 5(5):646–654.

49. Steinhagen M, Zunker K, Nordsieck K, Beck-Sickinger AG (2013) Large scale modifi-cation of biomolecules using immobilized sortase A from Staphylococcus aureus. Bio-org Med Chem 21(12):3504–3510.

50. Kornberger P, Skerra A (2014) Sortase-catalyzed in vitro functionalization of a HER2-specific recombinant Fab for tumor targeting of the plant cytotoxin gelonin. MAbs6(2):354–366.

51. Koudstaal W, et al. (2009) Pre- and postexposure use of human monoclonal antibodyagainst H5N1 and H1N1 influenza virus in mice: Viable alternative to oseltamivir.J Infect Dis 200(12):1870–1873.

Wagner et al. PNAS | November 25, 2014 | vol. 111 | no. 47 | 16825

IMMUNOLO

GYAND

INFLAMMATION