Coupling mammalian cell surface display withsomatic hypermutation for the discovery andmaturation of human antibodiesPeter M. Bowers1, Robert A. Horlick, Tamlyn Y. Neben, Rachelle M. Toobian, Geoffrey L. Tomlinson,Jennifer L. Dalton, Heather A. Jones, Andy Chen, Laurence Altobell III, Xue Zhang, John L. Macomber, Irina P. Krapf,Betty F. Wu, Audrey McConnell, Betty Chau, Trevin Holland, Ashley D. Berkebile, Steven S. Neben,William J. Boyle, and David J. King

AnaptysBio, Inc., 10421 Pacific Center Court, Suite 200, San Diego, CA 92121

Edited by David V. Goeddel, The Column Group, San Francisco, CA, and approved October 12, 2011 (received for review August 31, 2011)

A novel approach has been developed for the isolation andmatura-tion of human antibodies that replicates key features of the adap-tive immune system by coupling in vitro somatic hypermutation(SHM) with mammalian cell display. SHM is dependent on theaction of the B cell specific enzyme, activation-induced cytidinedeaminase (AID), and can be replicated in non-B cells through ex-pression of recombinant AID. A library of human antibodies, basedon germline V-gene segments with recombined human D� �J re-gions was used to isolate low-affinity antibodies to human β nervegrowth factor (hβNGF). These antibodies, initially naïve to SHM,were subjected to AID-directed SHM in vitro and selected usingthe same mammalian cell display system, as illustrated by the ma-turation of one of the antibodies to low pM KD. This approachovercomes many of the previous limitations of mammalian celldisplay, enabling direct selection and maturation of antibodies asfull-length, glycosylated IgGs.

affinity maturation ∣ mammalian display

The immune system is exquisitely adapted to generate a highfrequency of functional antibodies starting from a small num-

ber of variable region genes. In the absence of antigen, pre-B cellsundergo VðDÞJ recombination and express IgM on their surface.Complementarity-determining region 3 (CDR3) sequences en-coded through such gene recombination are critical for antigenrecognition by unmutated B cell receptors, and may be largelyresponsible for the primary repertoire (1–3). Naïve B cells usehigh avidity with surface IgM to facilitate low-affinity binding,with subsequent activation-induced cytidine deaminase (AID)-dependent somatic hypermutation (SHM) and class switchingto generate high-affinity antibody responses (Fig. 1A). AID isessential for the initiation of SHM in B cells by the deaminationof cytidine residues directly in Ig genes (4, 5). To achieve this,AID is targeted to V-region DNA sequences, termed hotspots(e.g., WRCH) that result in mutations and amino acid substitu-tions, which are frequently in positions biased to modulate anti-gen binding (6). Expression of AID alone has been shown to besufficient to reproduce the salient features of SHM in both B cellsand other mammalian cells (4, 7, 8).

Recombinant antibodies represent the fastest growing class ofnew medicines, and generation of antibodies that meet specificcriteria is increasingly important for therapeutic applications. Atpresent, there are two predominant methodologies for therapeu-tic antibody generation: immunization-based and surface display–based approaches. Immunization of wild type or transgenic ani-mals (9, 10) is an effective method for generating antibodies tomany antigens, and has led to the majority of Food and Drug Ad-ministration–approved therapeutic antibodies (11). Nevertheless,immune tolerance may lead to difficulties generating neutralizingantibodies when antigens are well conserved or are toxic uponadministration to animals. Specific, immunodominant epitopes

may be preferentially selected, making it difficult to identify func-tional antibodies (12).

Display technologies such as phage, yeast and ribosome displayare based on the in vitro selection of antibody fragments fromlibraries (13) and overcome limitations of immune tolerance orepitope dominance in vivo. Selection from such libraries may notalways generate high-affinity antibodies without subsequent affi-nity maturation (14). Furthermore, antibody fragments isolatedfrom microbial display systems are not always easily reformattedto produce well-expressed IgGs, soluble enough to be formulatedfor subcutaneous delivery (15).

Mammalian cell expression systems offer a number of poten-tial advantages for therapeutic antibody generation, including theability to coselect for key manufacturing-related properties suchas high-level expression and stability, while displaying functionalglycosylated IgGs on the cell surface. However, mammalian celldisplay has been hampered by the smaller library sizes that can bescreened, making direct isolation of high-affinity binders from na-ïve libraries improbable. Although small libraries biased toward aparticular antigen have been used successfully (16–19), a moregeneralized approach to generate high-affinity human antibodiesfrom immunologically naïve libraries has not been reported.

By mimicking in vitro features such as germline recombinationof V-gene fragments, recovery of low-affinity initial binders, andsubsequent AID-mediated SHM, we have carried out de novodiscovery and maturation of an anti-human β nerve growth factor(hβNGF) antibody to high affinity.

ResultsWe sought to reproduce in vitro features of the adaptive immunesystem that are critical for generation of diverse high-affinityantibodies, including a discrete starting repertoire of human rear-ranged immunoglobulin variable region sequences, mammaliancell-surface presentation, high avidity for isolation of weak bin-ders, and AID-mediated SHM (Fig. 1B). Although this approachis potentially applicable to any mammalian cell line, the HEK293cell line was chosen because of its ease of transfection, handling,and rapid cell growth. A robust system for coexpressing IgG

Author contributions: P.M.B., R.A.H., W.J.B., and D.J.K. designed research; P.M.B., T.Y.N.,R.M.T., G.L.T., J.L.D., H.A.J., A.C., L.A., X.Z., J.L.M., I.P.K., B.F.W., A.M., B.C., T.H., andA.D.B. performed research; P.M.B., R.A.H., and A.M. contributed new reagents/analytictools; P.M.B. and S.S.N. analyzed data; and P.M.B. and D.J.K. wrote the paper.

Conflict of interest statement: All authors associated with this manuscript work forAnaptysbio Inc. and receive a salary, stock, and/or stock options as part of their employ-ment. The scientific work presented in this paper is one part of the Anaptysbio platformand technology.

This article is a PNAS Direct Submission.

Freely available online through the PNAS open access option.1To whom correspondence should be addressed. E-mail: pbowers@anaptysbio.com.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1114010108/-/DCSupplemental.

www.pnas.org/cgi/doi/10.1073/pnas.1114010108 PNAS ∣ December 20, 2011 ∣ vol. 108 ∣ no. 51 ∣ 20455–20460

APP

LIED

BIOLO

GICAL

SCIENCE

S

Dow

nloa

ded

by g

uest

on

Sep

tem

ber

17, 2

020

and AID in mammalian cells was created using episomal vectorsexpressing the heavy chain (HC), light chain (LC), and AID, eachalso encoding a separate antibiotic selectable marker (20, 21). Todisplay immunoglobulin on the surface of these cells, full-lengthhuman IgG1 HC was modified by addition of a C-terminal trans-membrane domain (22).

Construction of a Fully Human Library of IgGs. The IgG library wasbased on germline sequence V-gene segments joined to prere-combined Dð ÞJ regions isolated from a panel of human donors.A total of nine HC and nine LC (5κ and 4λ) human germlineV-gene segments were selected for the library based on the fre-quency of in vivo germline usage (23, 24) (Fig. 1C). V regionswere chemically synthesized and fused to Dð ÞJ region sequences(encoding CDR3 and FR4 diversity) isolated by PCR frompooled peripheral blood mononuclear cells (PBMCs) of normaldonors. Full-length V regions for HC and LC were assembledwith human HC and LC constant regions and transfected intoHEK293 cells. A sampling of the stably selected library by high-throughput sequencing (HTP) provided a lower estimate of6 × 107 total diversity of combinatorially expressed antibody se-quences (25). The library was designed to provide multiple initialcandidates with germline V-gene segments for further maturationby SHM, and is termed ABELmAb (AnaptysBio Evolving Libraryof monoclonal Antibodies).

Isolation of Novel Human Antibodies to hβNGF. A human cytokine,hβNGF, was selected as a target for antibody discovery becauseof its well-described role in modulating pain sensation followingtissue injury and inflammation (26). βNGF binds and activates itscognate receptor, tropomyosin-related kinase A receptor (TrkA),up-regulating the expression and activity of pathways that en-hance acute and chronic pain. Antagonism of the βNGF/TrkAsignaling pathway has been shown in animal and clinical studiesto be a potent means of attenuating pain sensation in a number ofclinical indications (27, 28).

The transfected library was expanded to 109 cells, and sub-jected to four rounds of negative selection against streptavidin(SA)-coupled magnetic beads, followed by a single round ofpositive selection against SA-coupled magnetic beads coated withbiotinylated βNGF (Fig. 1B). Positively selected cells were ex-panded, and two rounds of fluorescence-activated cell sorting(FACS) selection were performed under high avidity conditions.Single cell clones (SCCs) were isolated with the second round of

FACS selection, sequenced, and each characterized for binding toβNGFand the ability of soluble TrkA-Fc receptor to compete thisbinding (Fig. 2 A–C, Table S1, and SI Materials and Methods). Of37 isolated round 2 SCCs, six unique clones were chosen foraffinity maturation and stably transfected with AID. Threerounds of FACS selection were performed using progressivelylower concentrations of fluorescently labeled hβNGF antigen.Selected antibody-expressing cells exhibited improved hβNGFbinding by the third round of SHM, and HC and LC sequencingof each selected population revealed enriched mutations, con-tained primarily in HC CDR regions.

Affinity Maturation of a hβNGF-Specific Antibody. Expression in theHEK293 cells is stably maintained using an episomal system thatsupports a low copy number (3–5 per cell) of each vector (21).Unique HC and LCs from each SCC were cloned, combinato-rially paired, expressed in HEK293 cells, and assessed in Biacoreand flow cytometry-based antigen-binding assays. The HC/LCpair from each SCC providing the best binding to βNGF wereretransfected with AID for further maturation. The sole HC iso-lated from SCC C10A contained an enriched mutation, S31N, inCDR1. One of three distinct germline LC sequences recoveredfrom SCC C10A, in combination with this HC, was used forfurther maturation (APE391).

Affinity maturation of APE391 was carried out utilizing twoindependent but initially identical cell populations, strategiesS1 and S2. Rounds 1–3 of FACS selection for each strategy wereperformed using low nM concentrations of βNGF fused to wasabifluorescent protein (WFP), each round selecting approximately0.2–0.5% of the brightest cells. Subsequent rounds of FACS se-lection used low pM concentrations of fluorochrome-labeledβNGF in combination with unlabeled βNGF competition. Se-quences were monitored by the sequencing of approximately40 HC and LC after each round. HTP sequencing was also em-ployed to monitor enriching mutations during early rounds ofFACS selection of strategy S1 (Table S2 and Fig. S1). Affinitymaturation was observed starting with the third round of FACSselection, continuing through the final rounds of affinity matura-tion (Round 12 for S1, Round 9 for S2), as evidenced by improvedβNGF binding (Fig. 2D–I), sequence data (Table 1, Table S2, andSI Materials and Methods), and Biacore analysis of isolated clones(Figs. 3 and 4 and SI Materials and Methods).

Fig. 1. Design of the ABELmAb library for mamma-lian surface display (A) Pro-B cells initially recombineV, D and J regions to express a naïve repertoire ofavid IgM antibodies. Antigen binding to specificclones stimulates B-cell maturation, including classswitching (IgG), proliferation, and AID-mediatedSHM to produce antigen specific, high-affinity anti-bodies secreted by plasma cells and presented on cir-culating memory cells. (B) CDR3/FW4 IgG and IgMdiversity was isolated from pooled PBMCs andgrafted into selected V regions to produce a libraryof germline full-length Abs. Antigen-coated beadswere used to isolate low-affinity antibodies, withsubsequent FACS selection and AID-mediated ma-turation of high-affinity antibodies. (C) Design ofthe ABELmAb full-length library is shown highlight-ing placement of restriction sites that were used tograft CDR3/FW4 diversity (red, green, and orange re-gions), amplified from a pool of PBMCs from sevennormal donors, into selected IgH (IGHV1-2, 1-69, 3-7, 3-23, 3-30-3, 4-34, 4-59, 5-51, and 6-1), IgK(IGKV4-1, 3-20, 2D-30, 1D-39, and 1-33) and IgL(IGLV1-40, 2-11, 3-21, and 7-43) synthesized V re-gions. Constant domains were also synthesized,and a transmembrane (blue) and cytoplasmic domain(red) were added to the C-terminal end of the HC.

20456 ∣ www.pnas.org/cgi/doi/10.1073/pnas.1114010108 Bowers et al.

Dow

nloa

ded

by g

uest

on

Sep

tem

ber

17, 2

020

Mutational Trajectories Accompanying Affinity Maturation of C10A.Maturation of APE391 in strategy S1 continued with the obser-vation of the L45F mutation in round 3 (out of 32 sequences),with all recovered sequences containing this mutation by round4 (APE583, Fig. 4). Kinetic characterization of this mutation wasassociated with a significant improvement in binding affinity forhβNGF. Two HC FW1/CDR1 insertions, GDTFSNYA (position26) and TFSNYAI (position 28), were found to be enriched in thecontext of APE583 to a majority of recovered sequences fromrounds 6 and 7 (APE659 and APE646, respectively), each asso-ciated with an approximately 20-fold improvement in bindingaffinity. An HC CDR3 mutation, D100A, was observed in round12 in the context of the APE646 variant, improving kon 100-foldand leading to an affinity of 760 pM (APE803, Fig. 4).

Affinity maturation of APE391 in the S2 strategy commencedin the HC with the incorporation of a single amino acid insertion,A24 (APE608), coincident with incorporation of mutationsG44R and G55D in round 3, each associated with significantimprovements in affinity (Fig. 4). APE608 became the predomi-nate HC sequence by round 5, with additional CDR1, FW3 andCDR3 HC mutations, identified in subsequent rounds. L45F,identified early in the S1 evolution strategy, was not observedin strategy S2 until round 10 of affinity maturation. Other affinitymaturation events in the S1 strategy, such as D100A and the long-er CDR1 codon insertions, were not detected in the S2 strategy.

Sequence analysis identified a total of 18 enriched mutationswithin the HC that improved antigen-binding kinetics (Table 1and Fig. 4B). Seven of the 18 enriched mutations were observedin both S1 and S2 strategies; five of the remaining mutationsinvolved sequence changes at the same amino acid positions

(e.g., D100A vs. D100G) and were also found in both S1 and S2strategies. Eleven substitutions were cataloged early in the S1strategy (Fig. S2), nine of which were identical or related to thoseobserved in later rounds. Enriching mutations were also identifiedby HTP sequencing of the LC, one of which (A60T) was incorpo-rated into the affinity matured antibodies APE925 and APE928.

Four independent AID-mediated codon insertions were ob-served in the maturing HC sequence near the junction of FW1and CDR1 (Table 1), with insertions ranging from a single alanineto segments containing 7, 8, and 12 amino acids. All four inser-tions originate from local sequence duplications, three of the fourcontained secondary AID mutations in combination with the in-sertion, and each was associated with a significant improvementfor binding affinity for the antigen (Fig. 4), all hallmarks of inser-tions observed in in vivo derived antibodies (29–31).

Mutations observed in both strategies not utilized by APE803were recombined by overlap extension PCR in a combinatoriallibrary, each HC/LC pair expressed, and characterized on Biacoreutilizing direct capture of secreted antibodies (Fig. 4). Variantsidentified with dissociation constants in the low pM range werefurther characterized in Biacore experiments utilizing normalizedconcentrations of purified antibodies captured at low densities(Rmax < 30), with APE925 possessing a KD of 25 pM (Fig. 3C).

Antibodies spanning the S1 affinity maturation trajectory werecharacterized for their ability to antagonize binding of βNGF toTrkA, compete for binding to βNGF with known anti-βNGFantibodies, and inhibit βNGF dependent TrkA signaling in arat pheochromocytoma-derived cell line (PC12). C10A-derivedantibodies demonstrated a dose-dependent ability to inhibit bind-ing of an anti-βNGF antibody (Tanezumab) to βNGF in a homo-genous time-resolved fluorescence (HTRF) assay that correlatedwell with kinetic characterization of each clone (Fig. 3D; comparewith Fig. 4). Antibodies APE925 and APE928 demonstrated adose-dependent inhibition of TrkA-Fc binding to βNGF (Fig. 3E)and inhibited βNGF-induced TrkA signaling and activation ofextracellular regulated kinase (ERK)1/2 in PC12 cells (Fig. 3F).These results demonstrate that the fully human antibodies gen-erated from this in vitro mammalian cell affinity maturation strat-egy bind βNGF with low pM dissociation constants and exhibitpotent and functional inhibition of βNGF binding to its cognatereceptor.

DiscussionBy combining the critical features of adaptive immunity, we havedeveloped an in vitro system for generating potent, biologicallyactive antibodies in a mammalian cell context. This approachmakes possible both de novo discovery and maturation from a na-ïve antibody library and may also be used for maturation of pre-existing antibodies. In vitro affinity maturation by SHM has beendescribed in human B cell lymphoma lines (32), chickenDT40 cells(33, 34), CHO cells (4), and 18–81 cells (35). However, these celllines are difficult to transfect at efficiencies suitable for use withdiverse libraries. Nevertheless, the potential of utilizing SHM invitro has been apparent since Cumbers et al. (7) were able toevolve the endogenous IgM in Ramos cells to recognize SA withan apparent affinity of 11 nM after 19 rounds of FACS sorting.

Overlapping but distinct sets of mutations were identified intwo affinity maturation strategies (Table 1). Early or rare SHMevents (e.g., L45F in strategy S1 vs. G44R in strategy S2) maydirect the antibody to alternative evolutionary fates that areequally consistent with high-affinity binding and biological activ-ity, hinting at the nondeterministic nature of SHM in vitro andin vivo. HTP sequence analysis of early evolving populationsprovides a powerful method for the identification of large,unbiased sets of enriched mutations. Many of the beneficial muta-tions observed by HTP sequencing were not identified by Sangersequencing of small numbers of templates until later FACS rounds,or were never observed (e.g., V89L HC and A60T LC). Antago-

Fig. 2. Flow cytometry antigen-binding analyses of clone C10A, S1 and S2affinity maturation strategies scattergrams showing βNGF binding to isolatedABELmAb cell clone C10A (A–C) and subsequent affinity maturation in stra-tegies S1 (D–F) and S2 (G–I). Approximately 5 × 107 cells were sorted perround. (A) C10A cells did not bind 50 nM irrelevant myc-tagged antigen,(B) bound 50 nM myc-βNGF under identical SA-generated avidity conditions,and (C) 50 nM βNGF binding was competed in the presence of equimolar con-centration of 50 nM TrkA-Fc receptor. FACS scattergrams are shown for(D) round 1 (10 nM WFP-βNGF), (E) round 7 (50 pM Dyl-NGF), and (F) round12 (35 pM Dyl-NGF) from strategy S1, and (G) round 1 (10 nM WFP-βNGF),(H) round 5 (500 pM Dyl-NGF) and (I) round 8 (150 pM Dyl-NGF) for strategyS2. Emerging populations of βNGF binding cells expressing CDR1 insertionsGDTFSNYA and A were evident during intermediate rounds of selection instrategies S1 (E) and S2 (H), respectively (arrows).

Bowers et al. PNAS ∣ December 20, 2011 ∣ vol. 108 ∣ no. 51 ∣ 20457

APP

LIED

BIOLO

GICAL

SCIENCE

S

Dow

nloa

ded

by g

uest

on

Sep

tem

ber

17, 2

020

nistic pleiotropy has been shown to slow the compilation of advan-tageous mutations (36) within a protein by limiting the number ofevolutionary routes available for maturation. Early combination ofmutations identified by sequencing into arrayed libraries, pairedwith kinetic screening of secreted antibodies, is likely to furtherspeed the identification of optimal antigen-binding solutions.

SHM events observed in our heterologous system mirror thoseobserved during in vivo affinity maturation. Nucleotide trans-versions and transitions, double mutation events (Table 1 andTable S2), and the nonsynonymous mutations that result, sug-gested that AID activity alone is sufficient to generate functionalamino acid substitutions (37). Mutation patterns and amino aciddiversification produced by AID on HCs and LCs in this in vitrosystem have been examined in the absence of functional selec-tion, and closely resemble that seen in antibody sequences thathave undergone in vivo affinity maturation (Fig. S1).

Codon insertions and deletions, particularly within CDRregions, are a rare but important feature of in vivo SHM that sig-nificantly expands the sequence space that may be sampled dur-ing affinity maturation (30, 38, 39). It has been estimated thatapproximately 6.5% of native human antibodies contain inser-tions or deletions within the variable domains as a consequenceof SHM (40), and a number of potential mechanisms for theirgeneration via AID-mediated mutagenesis have been proposed(39, 41). The identification of multiple, localized codon insertionswithin CDR regions in this and other affinity maturation pro-grams substantiates this as a robust feature of AID-directedSHM, and another aspect in which the in vivo process can bereproduced in vitro.

The use of AID-mediated mutagenesis and flow cytometry toevolve antibodies against fluorescent-labeled antigen(s) allowscomplex selection methodologies to be applied to select antibo-

dies evolved toward multiple properties of interest. We utilizedFACS selection in two dimensions to ensure that only well-expressed human IgGs with high binding affinity were selected.FACS-based selection may be carried out using multiple para-meters with control of stringency used to exert selection pressureto drive maturation to a desired endpoint. The simultaneous sur-face display and secretion of full-length antibodies allows for thedirect kinetic characterization of maturing cell clones or popula-tions, enabling parallel screening for desired functional or phy-siochemical properties.

Combining key elements of adaptive immunity in a single invitro system allows selection and maturation of antibodies tohigh-affinity, while avoiding the intrinsic limitations of the in vivoimmune response. A library of rearranged immunoglobulins withgermline V-gene segments was displayed on the surface ofHEK293 cells as a means to mimic the initial recombined reper-toire presented on the surface of B cells prior to affinity matura-tion. The combination of the use of high binding avidity to isolateinitial binders, and AID-induced SHM enables selection and af-finity maturation of high-affinity antibodies utilizing a minimumof mutations from the germline sequence and offers the flexibilityto select for antibody characteristics, such as multiple antigen orepitope binding, high expression level in mammalian cells andstability not achievable using other antibody discovery systems.

Materials and MethodsTransfection, Stable Expression and Selection of HEK293 c18 Cells. Stable epi-somal HEK-293 c18 cell lines expressing IgG HCs modified with a C-terminaltransmembrane domain (28) or LCs, together with AID were generated byseeding a T75 culture flask with 3 × 106 HEK293 c18 cells in 10 mL Dulbecco’smodified Eagle’s medium (DMEM) containing 10% FBS (Life Technologies).Plasmids were transfected using 500 μL OptiMEM (Life Technologies) and20 μL HD-Fugene (Roche Diagnostics). Three days posttransfection, cell

Table 1. AID-mediated mutations and insertions observed during affinity maturation

Category Corridor Starting sequence Ending sequence Start nucleotide End nucleotide Mutation

Mutation S2 AGGCT AGGGT C G A24GMutation S1,S2 AGGCT AGCCT G C A24PMutation S1 TGCTA TGTTA C T A33VMutation S1 TGATT TGCTT A C D100AMutation S2 TGATT TGGTT A G D100GMutation S2 AGGGC AAGAC, AAGGC G A G44RMutation S2 TGGTA TGATA G A G55DMutation S1 GGGCA GGACA G A G65DMutation S1,S2 GGCTT GGTTT C T L45FMutation S1 CAAAC CACAC A C N58HMutation S1,S2 ATCGT ATAGT C A R100gSMutation S1 CAGCA CATCA G T S30IMutation S1,S2 CAGCA CAACA G A S30NMutation S1,S2 CAGCT CAACT G A S31NMutation S1,S2 CAGCT CAACT G A S35NMutation S1,S2 GAGCA GAACA G A S76NMutation S2 GGGTG GGCTG G C V37LMutation S1 ATGCT ATATT G,C A,T GDTFSNYIInsertion S2 AGGCTT AGGCGGCTT A (24)Insertion S1 CTGGAGGCACCTTC

AGCAACTATGCTATCCTGGAGGCACCTTCAGCAACTATGCTAT

CACCTTCAGCAACTATGCTATC

TFSNYAI (28)

Insertion S1 CTGGAGGCACCTTCAGCAACTATGCTATC

CTGGAGACACCTTCAGCAACTATGCTACTGGAGACACCTT

CAGCAACTATGCTATC

GDTFSNYA (26)

Insertion S2 CAGGTGCAGCTGGTTCAGTCTGGGGCTGAGGTGAAGAAGCCTGGGT

CAGGTGCAGCTGGTTCAGTCTGGGGCTGAGGTGAACTGGGTGCAGCTGGTTCAGTCTGGGGCTGAGGTGAAGAA

NWVQLVASGQEV(12)

All HC mutations and insertion events observed during the S1 and S2 evolution corridors are shown. Column one indicates the category of SHM event, andthe second column indicates the strategy in which it was identified (S1 or S2). Columns three and four show the starting and ending nucleotide sequencesurrounding the site of mutation (codon in bold, mutation underlined) or insertion (original sequence in bold, duplicated sequence underlined). Columns fivethrough seven highlight the starting and ending nucleotide and resulting mutation, referenced by Kabat. Seven of the 19 mutations involved nucleotidetransversions, while the remaining 12 involved nucleotide transitions. Sixteen of the 19 mutations were initiated at the nucleotides G or C, three at nucleotideA, and thirteen of 19 occurred at known AID hotspots (WRCH) located primarily within CDRs 1, 2, or 3

20458 ∣ www.pnas.org/cgi/doi/10.1073/pnas.1114010108 Bowers et al.

Dow

nloa

ded

by g

uest

on

Sep

tem

ber

17, 2

020

growth medium was exchanged with 10 mL DMEM containing 10% FBS,50 μg∕mL Geneticin, 10 μl∕mL Antibiotic-Antimycotic Solution, 1.5 μg∕mLpuromycin, 15 μg∕mL blasticidin, and/or 350 μg∕mL hygromycin (all from LifeTechnologies), and the cells were incubated for approximately 4 weeks withperiodic reseeding and exchange of the cell culture medium.

ABELmAb Library Assembly and Transfection. CDR3 diversity was amplifiedfrom PMBCs from a pool of normal donors and inserted into nine HC andnine LC synthesized germline V regions (DNA2.0) selected based on their re-presentative use in vivo. Constant domains for the IgHC-γ1, IgKC, and IgLC3chains were also synthesized. A type I transmembrane and cytoplasmic regionwere added to the C-terminal end of the HC to enable presentation on thecell surface of HEK293 cells. The synthesized HC and LC constant region genefragments were assembled with the respective V regions in episomal vectors(ABELmAb library). The ABELmAb library DNA was transfected as describedabove as separate IGHV sublibraries, for each of the nine germline IGHV andIGKV templates. A random sampling of the stably selected library by HTP pro-vided a lower estimate of complexity of at least 50,000 unique HC sequencesand at least 1,200 unique LC sequences, for a lower estimate of 6 × 107 totaldiversity (25) of combinatorially expressed antibody sequences.

Target Cell Isolation Using Dynal Magnetic Biotin Binder Beads. Nine HC sub-libraries were transfected in combination with the pooled LC library intoHEK293 c-18 cells and used to isolate de novo βNGF binders. Cells (1 × 108

from each sublibrary) were subjected to four rounds of negative selectionby incubating with 2 × 108 biotin binder beads (Dynabeads® Biotin Binder;Life Technologies) for 0.5 h and retaining unbound cells. Antigen-coatedbeads were prepared by washing 2 × 107 beads in PBS, 0.1% BSA (Sigma-Aldrich) and incubating with 600 ng biotinylated βNGF (50 μL PBS, 0.1%BSA) for 1 h at 4 °C with rotation. Beads were washed twice, negatively se-lected HEK293 library cells were added and incubated for 2 h at 4 °C withrotation. Positively selected HEK293 library cells were resuspended in growthmedium and expanded for AID transfections and FACS selection.

Antigen and Antibody Expression and Purification. Antibody variants weregenerated as full-length secreted IgG molecules lacking the transmembraneand cytoplasmic domains, and expressed transiently in HEK293 c-18 cells.Transfected cell supernatants were loaded on a protein A/G agarose resin(Thermo Scientific), washed with 6 column-volumes of PBS, pH 7.4, andeluted with 100 mM glycine, pH 3.0. Human βNGF-His used in the ABELmAblibrary discovery and affinity maturation was expressed transiently in HEK293

Fig. 4. Sequence and kinetic analysis of affinity matura-tion trajectories. (A) The kinetic properties of affinitymatured anti-βNGF antibodies, secreted from SCCs in stra-tegies S1 and S2 and monitored by SPR, are shown in a kon

vs. koff plot. Unsequenced cell clones recovered during theS1 and S2 strategies are shown in gray. (B) HC chain se-quence variants corresponding to the affinity maturingclones shown in A are displayed, starting at Kabat position22, with mutations highlighted in orange and codon inser-tions shown in blue.

Fig. 3. Biacore and in vitro analysis of affinity maturing S1 strategy anti-βNGF antibodies. Selected sensorgrams and in vitro assays show the pro-gression of antigen-binding kinetics to antibodies produced by HEK293clones selected over the course of the S1 affinity maturation strategy (SIMaterials and Methods). (A) Incorporation of mutation S31N (APE391) inthe HC CDR1 results in detectable binding on Biacore. RU, resonance units.(B) Addition of mutation L45F in CDR2 of HC (APE583) improves affinity tolow micromolar affinity (kon ¼ 9.5 × 104 M−1 s−1, koff ¼ 0.08 s−1). Incorpora-tion of insertion CDR1 GDTFSNYA and a CDR3 mutation D100A (APE803)resulted in an affinity of 760 pM (kon ¼ 7.8 × 106 M−1 s−1, koff ¼ 6.0 ×10−3 s−1). (C) Introduction of the remaining HC mutations into a combina-torial library resulted in the identification of an antibody (APE925) with aKD ¼ 25 pM (kon ¼ 1.4 × 107 M−1 s−1, koff ¼ 3.5 × 10−4 s−1). (D) Antibodiesoriginating from the S1 strategy and an antibody to an irrelevant antigen(APE273) were characterized in an HTRF assay for their ability to inhibitbinding of tanezumab (APE081), an anti-hβNGF antibody. (E) Affinity ma-tured antibodies APE803, 925, and 928 were able to inhibit binding ofhβNGF to its high-affinity receptor TrkA-Fc, (F) and demonstrated potentinhibition of βNGF-dependent ERK1/2 phosphorylation in neuronal PC12cells (42).

Bowers et al. PNAS ∣ December 20, 2011 ∣ vol. 108 ∣ no. 51 ∣ 20459

APP

LIED

BIOLO

GICAL

SCIENCE

S

Dow

nloa

ded

by g

uest

on

Sep

tem

ber

17, 2

020

c-18 cells and purified using standard his-tag affinity purification methodol-ogies. FITC-labeled antigens utilized for FACS were prepared using standardamine coupling chemistry.

FACS Selection with Antigen Avidity. Binding analysis: Antibody transfectedHEK293 cells (5 × 105 in 0.5 mL PBS, 0.1% BSA) were incubated with variousconcentrations of βNGF-WFP-Myc for 0.5 h at 4 °C. Unlabeled Goat anti-Myc(AbCam) was added at a 2∶1 (antigen:anti-Myc antibody) molar ratio, andcells were incubated for 0.5 hr at 4 °C. FITC-AffiniPure Fab Fragment Goatanti-Human IgG (Hþ L) (Jackson ImmunoResearch) was added (1∶2;000)for 0.5 h at 4 °C. Cells were then pelleted and resuspended in 0.3 mL

0.2 μg∕mL DAPI in PBS, 0.1% BSA (Sigma-Aldrich) and analyzed for fluores-cence on a BD Influx cell sorter (BD Biosciences). FACS: Antibody transfectedHEK293 cells (5 × 107 in 20 mL PBS, 0.1% BSA) were incubated with 25 nMβNGF-WFP for 0.5 h at 4 °C. Goat anti-Myc andFITC-Goat anti-Human IgGwere added as above to the cells. Stained cells were resuspended in1.0 mL 0.2 μg∕mL DAPI in PBS, 0.1% BSA and sorted for the strongest anti-gen-binding cells on a BD Influx cell sorter.

ACKNOWLEDGMENTS. The authors thank Michael Neuberger, Phil Patten,Marilyn Kehry, and Matthew Scharff for their insightful comments.

1. Ignatovich O, Tomlinson IM, Jones PT, Winter G (1997) The creation of diversity in thehuman immunoglobulin V(lambda) repertoire. J Mol Biol 268:69–77.

2. Davis MM (2004) The evolutionary and structural ‘logic’ of antigen receptor diversity.Semin Immunol 16:239–243.

3. Williams SC, et al. (1996) Sequence and evolution of the human germline V lambdarepertoire. J Mol Biol 264:220–232.

4. Martin A, Scharff MD (2002) Somatic hypermutation of the AID transgene in B andnon-B cells. Proc Natl Acad Sci USA 99:12304–12308.

5. Maul RW, Gearhart PJ (2010) AID and somatic hypermutation. Adv Immunol105:159–191.

6. Ohm-Laursen L, Barington T (2007) Analysis of 6912 unselected somatic hypermuta-tions in human VDJ rearrangements reveals lack of strand specificity and correlationbetween phase II substitution rates and distance to the nearest 3′ activation-inducedcytidine deaminase target. J Immunol 178:4322–4334.

7. Cumbers SJ, et al. (2002) Generation and iterative affinity maturation of antibodies invitro using hypermutating B-cell lines. Nat Biotechnol 20:1129–1134.

8. GoodmanMF, Scharff MD, Rornesberg FE (2007) AID-initiated purposeful mutations inimmunoglobulin genes. Adv Immunol 94:127–155.

9. Lonberg N (2005) Human antibodies from transgenic animals. Nat Biotechnol23:1117–1125.

10. Almagro JC, Fransson J (2008) Humanization of antibodies. Front Biosci 13:1619–1633.11. Carter PJ (2006) Potent antibody therapeutics by design. Nat Rev Immun 6:343–357.12. Wilson IA, et al. (1984) The structure of an antigenic determinant in a protein. Cell

37:767–778.13. Hoogenboom HR (2005) Selecting and screening recombinant antibody libraries. Nat

Biotechnol 23:1105–1116.14. Hoet RM, et al. (2005) Generation of high-affinity human antibodies by combining

donor-derived and synthetic complementarity-determining-region diversity. Nat Bio-technol 23:344–348.

15. Persic L, et al. (1997) Targeting vectors for intracellular immunization. Gene 187:1–8.16. Lanzavecchia A, Corti D, Sallusto F (2007) Human monoclonal antibodies by immor-

talization of memory B cells. Curr Opin Biotechnol 18:523–528.17. Beerli RR, et al. (2008) Isolation of human monoclonal antibodies by mammalian cell

display. Proc Natl Acad Sci USA 105:14336–14341.18. Kwakkenbos MJ, et al. (2010) Generation of stable monoclonal antibody-producing B

cell receptor positive human memory B cells by genetic programming. Nat Med16:123–8.

19. Zhou C, Jacobsen FW, Cai L, Chen Q, Shen WD (2010) Development of a novel mam-malian cell surface antibody display platform. MAbs 2:508–518.

20. al-Shawi R, Kinnaird J, Burke J, Bishop JO (1990) Expression of a foreign gene in a lineof transgenic mice is modulated by a chromosomal position effect. Mol Cell Biol10:1192–1198.

21. Horlick RA, et al. (2000) Combinatorial gene expression using multiple episomal vec-tors. Gene 243:187–194.

22. Rogers J, et al. (1981) Gene segments encoding transmembrane carboxyl termini ofimmunoglobulin gamma chains. Cell 26:19–27.

23. Glanville J, et al. (2009) Precise determination of the diversity of a combinatorialantibody library gives insight into the human immunoglobulin repertoire. Proc NatlAcad Sci USA 106:20216–20221.

24. Knappik A, et al. (2000) Fully synthetic human combinatorial antibody libraries(HuCAL) based on modular consensus frameworks and CDRs randomized with trinu-cleotides. J Mol Biol 296:57–86.

25. Weinstein JA, Jiang N,White RA, Fisher DS, Quake SR (2009) High-throughput sequen-cing of the zebrafish antibody repertoire. Science 324:807–810.

26. Mantyh PW, Koltsenburg M, Mendell LM, Tive L, Shelton DL (2011) Antagonism ofnerve growth factor-TrkA signaling and the relief of pain. Anesthesiology115:189–204.

27. Shelton DL, Zeller J, Ho WH, Pons J, Rosenthal A (2005) Nerve growth factor mediateshyperalgesia and cachexia in auto-immune arthritis. Pain 116:8–16.

28. Lane NE, et al. (2010) Tanezumab for the treatment of pain from osteoarthritis of theknee. N Engl J Med 363:1521–1531.

29. Krause JC, et al. (2011) An insertion mutation that distorts antibody binding sitearchitecture enhances function of a human antibody. MBio 2:e00345-10.

30. Miura Y, et al. (2003) Diversification of the Ig variable region gene repertoire ofsynovial B lymphocytes by nucleotide insertion and deletion. Mol Med 9:166–174.

31. Wu X, et al. (2011) Focused evolution of HIV-1 neutralizing antibodies revealed bystructures and deep sequencing. Science 333:1593–1602.

32. Delker RK, Fugmann SD, Papavasiliou FN (2009) A coming-of-age story: Activation-induced cytidine deaminase turns 10. Nat Immunol 10:1147–1153.

33. Seo H, et al. (2006) An ex vivo method for rapid generation of monoclonal antibodies(ADLib system). Nat Protoc 1:1502–1506.

34. Seo H, et al. (2005) Rapid generation of specific antibodies by enhanced homologousrecombination. Nat Biotechnol 23:731–735.

35. Wang CL, Yang DC, Wabl M (2004) Directed molecular evolution by somatic hypermu-tation. Protein Eng Des Sel 17:659–664.

36. Weinrich DM, Delaney NF, DePristo MA, Hartl DL (2006) Darwinian evolution can fol-low only very few mutational paths to fitter proteins. Science 312:111–114.

37. Zheng NY, Wilson K, Jared M, Wilson PC (2005) Intricate targeting of immunoglobulinsomatic hypermutation maximizes the efficiency of affinity maturation. J Exp Med201:1467–1478.

38. Reason DC, Zhou J (2006) Codon insertion and deletion functions as a somatic diver-sification mechanism in human antibody repertoires. Biol Direct 1:24.

39. Wilson PC, et al. (1998) Somatic hypermutation introduces insertions and deletionsinto immunoglobulin V genes. J Exp Med 187:59–70.

40. DeWildt RM, van Venrooij WJ, Winter G, Hoet RM, Tomlinson IM (1999) Somatic in-sertions and deletions shape the human antibody repertoire. J Mol Biol 294:701–710.

41. Sale JE, Neuberger MS (1998) TdT-accessible breaks are scattered over the immuno-globulin V domain in a constitutively hypermutating B cell line. Immunity 9:859–869.

42. Greene LA, Tischler AS (1976) Establishment of a noradrenergic clonal line of rat adre-nal pheochromocytoma cells which respond to nerve growth factor. Proc Natl Acad SciUSA 73:2424–2428.

20460 ∣ www.pnas.org/cgi/doi/10.1073/pnas.1114010108 Bowers et al.

Dow

nloa

ded

by g

uest

on

Sep

tem

ber

17, 2

020