A Novel Affinity Tag, ABTAG, and Its Application to the Affinity Screening of Single-Domain Antibodies Selected by Phage Display

ABTAG is a camelid single-domain antibody (sdAb) that binds to bovine serum albumin (BSA) with low picomolar affinity. In surface plasmon resonance (SPR) analyses using BSA surfaces, bound ABTAG can be completely dissociated from the BSA surfaces at low pH, over multiple cycles, without any reduction in the capacity of the BSA surfaces to bind ABTAG. A moderate throughput, SPR-based, antibody screening assay exploiting the unique features of ABTAG is described. Anti-carcinoembryonic antigen-related cell adhesion molecule 6 (CEACAM6) sdAbs were isolated from a phage-displayed sdAb library derived from the heavy chain antibody repertoire of a llama immunized with CEACAM6. Following one or two rounds of panning, enriched clones were expressed as ABTAG fusions in microtiter plate cultures. The sdAb-ABTAG fusions from culture supernatants were captured on BSA surfaces and CEACAM6 antigen was then bound to the captured molecules. The SPR screening method gives a read-out of relative expression levels of the fusion proteins and kinetic and affinity constants for CEACAM6 binding by the captured molecules. The library was also panned and screened by conventional methods and positive clones were subcloned and expressed for SPR analysis. Compared to conventional panning and screening, the SPR-based ABTAG method yielded a considerably higher diversity of binders, some with affinities that were three orders of magnitude higher affinity than those identified by conventional panning.

inTrODUcTiOn Over the past 25 years, in vitro display technologies, most notably phage and yeast display, have increasingly become the methods of choice for the isolation and ainity maturation of monoclonal antibody fragments. his is especially the case in instances where antibodies with particular properties are required (1,2), as in vitro display technologies allow for the selection process to be biased toward desired outcomes (3). To take full advantage of the power of these technologies it is preferable to screen relatively large numbers of clones ater one or two selection cycles to increase the odds of identifying clones with the desired properties. here is ongoing interest, therefore, in the A Novel Afinity Tag for sdAb Screening Frontiers in Immunology | www.frontiersin.org October 2017 | Volume 8 | Article 1406 development of more efective and rapid methods for screening antibodies for expression level and for target antigen speciicity and ainity. he classical protocol for isolating antibody fragments against a given target by phage display technology generally includes immunizing an animal, measuring the immune response, constructing a phage-displayed antibody fragment library and performing three to ive rounds of panning to enrich for clones that bind to the target. his generally gives a manageable number of clones for subsequent expression, puriication and characterization. he antibody fragments can be fragments antigen-binding (Fabs) (4), single-chain variable fragments (scFvs) (4), or single-domain antibodies (sdAbs). sdAbs can be the variable domains (VHHs) of camelid heavy chain antibodies (5,6), the variable domains (VNARs) of shark immunoglobulin new antigen receptors (7), the variable heavy domains (VHs) of conventional antibodies (8,9), or the variable light domains (VLs) of conventional antibodies (10). Ater panning, phage ELISA is performed on randomly picked clones to identify the leads. Leads are typically expressed and puriied for ainity measurements and functional characterization. his is a relatively slow, costly and laborious process that generally yields no more than a dozen binders, and oten fewer.
In this study, we compare diferent approaches to the isolation sdAbs against carcinoembryonic antigen-related cell adhesion molecule 6 (CEACAM6) from a phage-displayed immune llama VHH library. We have used a classical panning protocol followed by subcloning, expression and puriication of sdAbs for surface plasmon resonance (SPR) analyses. We also describe a screening method in which anti-CEACAM6 sdAbs are fused to an anti-bovine serum albumin (anti-BSA) sdAb characterized by extremely tight, but readily reversible, binding to BSA. he sdAb, initially termed BSA12, has a KD of 4 pM and a kd of 9 × 10 −6 s −1 , equivalent to an sdAb:BSA half-life of approximately 21 h, yet it can be completely dissociated from BSA surfaces with a 3 s pulse of 100 mM HCl (11,12). In the SPR-based screening method described here (Figure 1A), panning-enriched sdAb clones are expressed in fusion with the sdAb BSA12 and captured on BSA surfaces for CEACAM6 binding measurements. For applications such as this, we have given the anti-BSA sdAb BSA12 a new designation, ABTAG, with "AB" alluding to the fact that it is an antibody fragment or, alternatively, that it is an albumin-binding "TAG. " With this new methodology, we isolated several extremely high-ainity anti-CEACAM6 sdAbs that were missed by classical panning and screening of the same phage-displayed library. he ABTAG technology should have general utility in the isolation of antibody fragments from phage display libraries-i.e., it could be applied to VL, VH, VNAR, scFv, and Fab libraries. ABTAG technology has previously been used to screen sdAbs speciic to human Fc (fragment crystallizable) region (13).

MaTerials anD MeThODs immunization and serology
A llama (Lama glama) was immunized ive times (days 1, 21, 35, 49, and 63) subcutaneously with approximately 100 µg of recombinant CEACAM6 [N-and A-domains, residues 35-232; (14)] per injection, kindly provided by Helix BioPharma (Aurora, ON, Canada), with Complete Freund's Adjuvant on day 1 and Incomplete Freund's Adjuvant on the remaining days. On days 1, 22, 36, 49, 64, and 71 blood (50 mL) was collected, from which sera and peripheral blood lymphocytes were isolated (15). his study was carried out in accordance with Animal Use Protocols approved by the National Research Council Canada Animal Care Committee.
Microtiter plates were coated with 10 µg/mL of CEACAM6 overnight at 4°C in 15 mM Na2CO3, pH 9.6, followed by blocking with 2% fat-free dry milk (BioRad Laboratories, Mississauga, ON, Canada) in PBS. Serially diluted serum was then added to the wells. Detection of llama IgGs was performed with goat anti-llama IgG (Bethyl Laboratories, Montgomery, TX, USA) and a horseradish peroxidase anti-goat IgG conjugate (Cedarlane Laboratories, Burlington, ON, Canada). Finally, peroxidase substrate (KPL, Gaithersburg, MD) was added followed by 1 M H3PO4 ater 15 min. he absorbance at 450 nm was then measured.

Phage-Displayed library construction
RNA was extracted from peripheral blood lymphocytes obtained from blood drawn on day 71 by using a QIAamp RNA Blood Mini Kit (Qiagen Inc., Mississauga, ON, Canada). cDNA was synthesized by using a first-strand cDNA synthesis kit (GE Healthcare, Mississauga, ON, Canada). Sense primers MJ1, MJ2, and MJ3 and antisense primers CH2 and CH2b3 (15) were used to amplify VHH and VH-CH1 encoding regions (600 and 900 bp, respectively). These two fragments were separated by agarose gel electrophoresis and the VHH band was purified from the gel. Nested PCR, using primers MJ7 and MJ8 (15), was performed to amplify all VHHs. The final PCR fragments were ligated into the phagemid vector pMED1 (16) using SfiI restriction sites. The ligated vector was used to transform electrocompetent Escherichia coli TG1 cells.
selection of sdabs by conventional Panning he VHH repertoire was expressed on phage surfaces ater rescuing with M13K07 helper phage. hree rounds of phage display panning were conducted ( Figure 1B) as described (17). Briely, speciic VHHs against CEACAM6 were enriched by in vitro selection on microtiter plates coated with antigen (10 µg/mL). Phage particles eluted with 100 mM triethylamine, pH 11, were immediately neutralized with 1 M Tris-HCl, pH 7.4, and used to infect exponentially growing TG1 cells. To assess the enrichment of phage particles carrying antigen-speciic VHHs, serial dilutions of the phage eluted from antigen coated vs non-coated control wells were used to infect exponentially growing TG1 cells. Individual colonies randomly picked ater all three rounds of panning were tested for binding to CEACAM6 by phage ELISA according to standard procedures (17). Unique clones identiied by DNA sequencing were subcloned in pSJF2H, expressed, and FigUre 1 | Summary of ABTAG technology and single-domain antibody (sdAb) isolation. Schematic diagram of the ABTAG/surface plasmon resonance (SPR)based screening method for the evaluation of enriched sdAb clones from phage display panning (a). While anti-CEACAM6 sdAb clones obtained after three rounds of panning were screened by phage ELISA, pools of phage clones enriched after either one or two rounds of phage display library panning were directly subcloned into pABTAG (3,575 bp, representative sdAb of average length), creating two sublibraries, for expression as sdAb-ABTAG fusions by capture on BSA surfaces (B,c). A representative sensorgram demonstrates sdAb-ABTAG fusion proteins are stably retained by BSA surfaces because of an extremely slow dissociation rate (D). BSA, bovine serum albumin; RU, resonance unit. Immobilizations were carried out at a protein concentration of 50 µg/mL in 10 mM acetate bufer, pH 4.5, using an amine coupling kit (GE Healthcare). Forty microliters of the E. coli culture supernatants of 48 randomly picked clones from the round 2 sublibrary, containing the anti-CEACAM6 sdAb-ABTAG fusions, were added to a 96-well microtiter plate and covered with selfadhesive foil (GE Healthcare). Sixty microliters of running bufer (10 mM HEPES, pH 7.4, containing 150 mM NaCl, 3 mM EDTA, and 0.01% surfactant P20) were added to each well to dilute the culture supernatants, followed by mixing. As the Biacore 3000 has four Fcs, three fusions were simultaneously analyzed to increase screening throughput-the fourth Fc served as an ABTAG reference surface. ABTAG and the fusions were captured on the BSA surfaces by sequentially injecting 40 µL of three diferent diluted culture supernatants over Fcs 2, 3, and 4 at a low rate of 5 µL/ min. For some clones over 4,000 RUs of the sdAb-ABTAG fusions were captured in this manner. For the reference surface, 20 µL of 80 nM ABTAG were injected over Fc 1, followed by injection of a 60 µL bufer blank. Next, 1 µM of CEACAM6 was injected over all four Fcs at a low rate of 20 µL/min and the dissociations were monitored for 3 min followed by surface regeneration with a 15 s injection of 10 mM glycine/HCl, pH 2.0. In all instances, analyses were carried out at 25°C in running bufer. he reference subtracted data were aligned and a corresponding bufer blank was subtracted from each sensorgram. he dissociation phase was normalized with the highest response in each sensorgram set at 100% and no response set at 0% for all data sets to rank binders by their dissociation rate. his qualitative analysis allowed for easy visual identiication of binders with fast, medium and slow dissociation rates, which generally correlate with low, medium, and high ainities. Of-rates were determined using the BIAevaluation v4.1.1 Sotware (GE Healthcare). DNA sequencing of CEACAM6 binding sdAb-ABTAG clones was then performed as described previously (15).

screening of ceacaM6 Binding to sdab-aBTag Fusions by Full Kinetic analysis
Full kinetic data for CEACAM6 binding to captured sdAb-ABTAG fusions were obtained by single-cycle kinetics (SCK) using a Biacore T200 (GE Healthcare). he same 48 clones (round 2 sublibrary) that were screened by of-rate analysis above were analyzed in this manner. Fresh microtiter plate cultures were grown as described above. Cultures were harvested and periplasmic extracts prepared as described (15). Approximately 9,500 RUs of BSA (Sigma-Aldrich Canada) were immobilized on all four Fcs of CM5 Series S sensor chips (GE Healthcare). Immobilizations were carried out at a protein concentration of 50 µg/mL in 10 mM acetate bufer, pH 4.5, using an amine coupling kit (GE Healthcare). Fity microliter of the periplasmic extracts of the 48 clones and 100 µL of running bufer (10 mM HEPES, pH 7.4, 150 mM NaCl, 3 mM EDTA, 0.05% surfactant P20) were added to a 96-well microtiter plate. ABTAG (80 nM) was captured on Fc 1 at a low rate of 5 µL/min for 4 min, as a reference, and the diluted periplasmic extracts of three clones were captured on Fcs 2, 3, and 4 at a low rate of 5 µL/min for 8 min per cycle. CEACAM6 was injected over all four Fcs at concentrations of 0.1, 1, 10, 100, and 1,000 nM at a low rate of 5 µL/min for 3 min at 25°C in running bufer. Dissociations were monitored for 5 min followed by surface regeneration with 10 mM glycine/HCl, pH 2.0, at a low rate of 10 µL/min for 60 s. Bufer-blank cycles preceded each antigen injection cycle. Data were analyzed using Biacore T200 Evaluation Sotware v3.0 (GE Healthcare). Moderate-throughput sdAb screening with ainity and kinetic determination was also carried out using the BioRad ProteOn SPR instrument (BioRad Laboratories Canada Inc.). Microtiter plate cultures of 576 randomly picked clones from the round 1 sublibrary were centrifuged and the supernatants analyzed for binding to CEACAM6 by ELISA. ELISA-positive clones were sequenced (15) and subjected to ProteOn screening. Approximately 1,200 RUs of BSA were immobilized, as described above, except for immobilizing the BSA in the horizontal direction to create horizontal referencing spots with a BSA surface. he screening procedure occurred in two steps with a ligand capture in the vertical direction followed by an analyte injection in the horizontal direction. One bufer injection for 30 s at low 100 µL/ min in the ligand direction was used to stabilize the baseline ater switching from the previous analyte injection. For each ligand capture, ive individual E. coli culture supernatants each containing an sdAb-ABTAG fusion were diluted 1:25 in running bufer containing 1 mg/mL carboxymethyldextran sodium salt (Sigma-Aldrich Canada) to reduce non-speciic protein interactions with the GLM chip surface and injected for 240 s at a low rate of 25 µL/min. his resulted in ive individual ligand samples on the GLM-BSA surface with up to approximately 200 RUs of fusion protein being captured for the highest expressing clones. A Novel Afinity Tag for sdAb Screening Frontiers in Immunology | www.frontiersin.org October 2017 | Volume 8 | Article 1406 he irst ligand channel was let empty for use as a blank control surface. his was immediately followed by two bufer injections to stabilize the baseline in the analyte direction, each for 30 s at a low rate of 100 µL/min, and then the CEACAM6 analyte injection. Five CEACAM6 concentrations (50, 16.7, 5.6, 1.85, and 0.62 nM) and bufer were simultaneously injected in individual analyte channels at 50 µL/min for 120 s with a 600 s dissociation, resulting in a set of binding sensorgrams with a bufer reference for each of the ive captured sdAb-ABTAG fusions. he BSA surfaces with bound sdAbs were regenerated by an 18 s pulse of 0.85% phosphoric acid at a low rate of 100 µL/min. Sensorgrams were aligned and double-referenced using the bufer blank injection and the resulting sensorgrams were analyzed using ProteOn Manager sotware v2.1.1. Kinetic parameters were determined by itting the referenced sensorgrams to a 1:1 interaction model using local Rmax, and ainity constants (KD, nM) were derived from the resulting rate constants [kd sPr analysis of ceacaM6 Binding to covalently immobilized sdabs and sdab-aBTag Fusions Puriied sdAbs identiied by conventional panning and the highest ainity binders identiied via ABTAG screening ater one and two rounds of panning were subjected to standard SPR analysis. Following puriication by immobilized-metal ainity chromatography (18), sdAbs and sdAb-ABTAG fusions were further puriied by size exclusion chromatography (SEC) using Superdex S75 and Superdex S200 Increase columns (GE Healthcare), respectively. CEACAM6 was subjected to Superdex S75 chromatography to remove possible aggregates. sdAbs from conventional panning and from ABTAG round 2 screening were immobilized on CM5 sensor chips (GE Healthcare) at 50 µg/mL in 10 mM acetate bufer, pH 4. Multiple cycle kinetics were performed on a Biacore 3000 (GE Healthcare) in 10 mM HEPES, pH 7.4, containing 150 mM NaCl, 3 mM EDTA, and 0.005% P20 by lowing appropriate concentrations of CEACAM6 at a low rate of 40 µL/min. Puriied sdAb-ABTAG fusions from ABTAG round 1 screening were immobilized on CM5 sensor chips (GE Healthcare) at concentrations of 12-25 µg/mL in 10 mM acetate bufer, pH 3.5. Single cycle kinetics (SCK) were performed on a Biacore 3000 (GE Healthcare) in 10 mM HEPES, pH 7.4, containing 150 mM NaCl, 3 mM EDTA, and 0.005% P20 by lowing appropriate concentrations of CEACAM6 at a low rate of 25 µL/min. Data were analyzed using BIAevaluation v4.1.1 Sotware (GE Healthcare).

next-generation Dna sequencing (ngs)
he phage-displayed immune llama VHH library was interrogated using an Illumina MiSeq instrument as previously described (1,2). Briely, replicative form DNA was puriied from phagemidbearing E. coli TG1 cells using a QIAprep ® spin miniprep kit (Qiagen Inc.). Amplicons for NGS were prepared using PelB-and FR4-speciic barcoded primers using ~25 ng of phagemid DNA as template and puriied by gel extraction followed by solid-phase reversible immobilization using Agencourt ® AMPure ® XP beads (Beckman-Coulter, Pasadena, CA, USA). he data were quality iltered using the FAST-X toolkit with a stringency of Q30 over 95% of the read.

resUlTs selection of ceacaM6-speciic Binders by conventional Panning
Immunization of a llama with recombinant CEACAM6 resulted in strong immune response against the immunogen as indicated by ELISA comparing the preimmune and postimmunization sera collected on day 71 from the animal ( Figure S1 in Supplementary Material).
A phage-displayed VHH library with a functional size of 2.9 × 10 7 clones was constructed from the lymphocytes of the immunized llama and used to select CEACAM6-binding sdAbs. Ater three rounds of panning, randomly picked clones were tested by phage ELISA to identify clones displaying CEACAM6speciic sdAbs. DNA sequencing revealed that the positive clones were comprised of ive diferent sequences. he unique clones were named 1B6, 1F6, 2A7, 2F8, and 2G9. Ater subcloning in an expression vector, all ive sdAbs were expressed in E. coli periplasms and puriied. ELISA showed that all ive sdAbs bound to CEACAM6 (data not shown). he actual ainities of these sdAbs were determined to be in the 2-10 nM range by SPR ( Table 1; Figure S2A in Supplementary Material).
sdab-aBTag Binding to immobilized Bsa he ainity of ABTAG for BSA, as a fusion with an anti-CEACAM6 sdAb, was determined to be 21 pM with a dissociation rate constant, kd, of 9.1 ± 0.7 × 10 −6 s −1 (Figure 1D). hese values are essentially identical to those previously reported for ABTAG (11). As the dissociation rate is extremely slow, it was determined over a long, 10,000 s, dissociation phase.

Dissociation rate and expression screening of sdab-aBTag Fusion Proteins
Eluted phage from the second round of panning were ampliied by PCR and cloned into the ABTAG fusion vector (Figures 1B,C). he size of this sublibrary was 1.4 × 10 7 independent transformants. Individual clones from this sublibrary were grown in microtiter plates and the culture supernatants were used to rank the dissociation rate constants by SPR analysis. A total of 17 cycles were performed to generate the binding data for 48 clones; two duplicates and a bufer injection were included (Table S1 in Supplementary Material). Surface regeneration between cycles was exceedingly eicient with identical amounts of ABTAG being captured on the reference Fc over the 17 cycles (Figure 2). Of the 48 clones, 19 were excluded from of-rate screening and sequence determination: (i) six because the sdAb-ABTAG capture levels were less than 100 RUs, (ii) 11 because there was insigniicant antigen binding to the captured fusion proteins and (iii) two because of low sdAb-ABTAG capture levels and poor 1:1 itting of antigen binding data. As shown in Figure 3 and Table  S1 in Supplementary Material the clones exhibited a wide range of of-rates (kds), namely, from 1.48 × 10 −4 to 1.67 × 10 −12 s −1 . As the clones were randomly picked, they were not all unique,  accounting for the nearly identical dissociation proiles of some of the clones. As the screening measurements were performed at a single CEACAM6 concentration, it was impossible to calculate accurate on-rates and, consequently, accurate KD values. Of 29 sequenced clones, one of the clones isolated by conventional panning, 2A7, was the most frequent, accounting for 9 of the 29 clones. Another clone isolated by conventional panning, 2G9, occurred twice. he remaining three sdAbs identiied by the conventional approach (1F6, 2F8, and 1B6) were not found by the of-rate screening method. hree clones with the slowest of-rates (2-03 = 6.87 × 10 −4 s −1 ; 2-15 = 1.48 × 10 −4 s −1 ; 2-35 = 8.56 × 10 −4 s −1 ) were subcloned into the pSJF2H expression vector and the sdAbs were puriied for full kinetic analysis over a range of antigen concentrations. hese three unique sdAbs exhibited ainities below 1 nM (0.5-0.8 nM), up to 10 times stronger than the ainities of the binders selected by conventional panning ( Table 1; Figure S2B in Supplementary Material). he ABTAG/sdAb-ABTAG capture step, prior to antigen low and dissociation rate ranking, gave expression level data for the 48 clones. he total RUs captured for the various clones were a good measure of relative expression levels (Table S1 and Figure S3 in Supplementary Material). It was possible to quantify the concentrations of sdAb-ABTAG fusions in the culture supernatants. A plot of fusion protein concentration vs initial fusion protein binding rate was generated from the linear, mass transportlimited, portions of the sensorgrams ( Figure S3 in Supplementary Material). his plot was used to derive fusion protein concentrations in culture supernatants. Approximately half of the clones

Full Kinetic and afinity screening of sdab-aBTag Fusion Proteins
A limitation of the dissociation rate screening with the Biacore 3000 instrument is that single antigen concentrations are utilized for screening and clones must be expressed as pure sdAbs to generate full kinetic information. In view of this, we explored two approaches for obtaining full kinetic data using microtiter plate cultures containing sdAb-ABTAG fusions.
In the irst approach SCK experiments were performed in which increasing concentrations of CEACAM6 were lowed over captured sdAb-ABTAG fusions using a Biacore T200 instrument. Microtiter plate periplasmic extracts from the 48 clones from round 2 subjected to of-rate screening were screened by SCK to determine ka, kd, and KD (Table S2 in Supplementary Material). A wide range of CEACAM6 concentrations (0.1-1,000 nM) was lowed over the captured clones in order to obtain approximate KDs for clones with wide-ranging ainities as the of-rate screening of the 48 clones gave kds covering two orders of magnitude (Table S1 in Supplementary Material). As expected, the concentration range was not ideal for individual clones but in most instances two or three of the concentrations gave acceptable responses (Figure 4A). Fitting the sensorgrams to a 1:1 interaction model gave a surprisingly good approximation of the rate constants and KDs (Figure 4B; Table S2 in Supplementary Material), as indicated by the rate constants and ainities of some of the clones determined by more rigorous analyses ( Table 1). he median KD for these analyzed clones was 7.98 nM (Figure 4B).
In the second approach, ive diferent CEACAM6 concentrations were lowed over ive or six captured sdAb-ABTAG fusions in a single injection cycle using a BioRad ProteOn, which has a 6 by 6 grid of microluidic channels. To take full advantage of these higher throughput approaches, we made a second, more diverse, sublibrary by cloning phage-displayed sdAbs eluted from round 1 panning into the ABTAG-fusion protein expression vector. Of the 576 randomly picked clones from the round 1 sublibrary, 52 were positive for antigen binding by ELISA (data not shown). Microtiter plate culture supernatants containing sdAb-ABTAG fusions from the 52 positive ELISA clones were screened using a ProteOn to determine sdAb-ABTAG capture levels and, by itting the sensorgram data to a 1:1 interaction model, rate constants, KDs and observed Rmaxs (Table S3 in Supplementary Material). No antigen binding was observed for several of the clones-low capture levels and low ainities may account for this observation. hree of the ive clones identiied by conventional panning (2A7, 2G9, and 1F6) were found in this screen. Eleven of the 52 clones were 2G9, three or four were 2A7, and one was 1F6. Of the remaining 36 clones, 32 were unique; four clones were found twice. he sensorgram data were generally of high quality, itting well to a 1:1 interaction model (Figure 5A). he method gave excellent KD approximations (Figure 5B), as evidenced by the values obtained for clones that were subjected to a more stringent SPR analysis, i.e., the sdAbs that were analyzed by binding antigen to immobilized sdAbs ( Table 1). he median KD for these analyzed clones was 0.99 nM ( Figure 5B). Values are not given for clones with capture levels of less than 100 RUs. Several of the clones had higher ainities, attributable to slower of-rates, than those of clones from the round 2 library. While the very slow ABTAG kd results in stable capture of sdAb-ABTAG fusions, direct binding assays in which the ABTAG fusions are covalently linked to the surface and antigen is lowed are preferable in instances where the sdAb kd is very slow. While the of-rates determined by the capture method were derived using double referencing (reference cell and zero-concentration subtraction), accurate determination of very slow of-rates can be challenging, even in the best of circumstances. he four round 1 library clones showing the highest ainities by ProteOn screening were expressed and puriied as sdAb-ABTAG fusions for direct binding assays ( Table 1; Figure  S2C in Supplementary Material). Monomeric CEACAM6 peaks from SEC ( Figure S2D in Supplementary Material) were used for analysis. he ainities of these four clones, which ranged from 4 to 232 pM ( Table 1), were all higher than the ainities of the top round 2 clones (Figure 6; Table S2 in Supplementary Material) with clone 1-09 exhibiting an exceedingly slow of-rate ( Table 1). he 1-09 sdAb-ABTAG surface required stringent regeneration conditions ( Figure S2C in Supplementary Material).

next-generation Dna sequencing
To better understand why some of the highest-ainity sdAbs against CEACAM6 were not recovered through conventional panning of the phage-displayed VHH library, we interrogated the library to moderate depth (approximately 2.5 × 10 5 qualityiltered sequences) using amplicon sequencing on an Illumina MiSeq instrument ( Table 2). Both the medium-ainity sdAbs recovered via conventional panning as well as the extremely high-ainity ones recovered only via ABTAG screening were observed in the library at roughly similar frequencies. As panning progressed, the frequencies of all CEACAM6-speciic sdAbs increased, indicating that all of the sdAb-phages bound plateadsorbed CEACAM6 and were eluted from it under high pH. However, several medium-ainity sdAbs (in particular 2A7 and    2G9) rapidly outcompeted the higher-ainity sdAbs, becoming enriched 50-100-fold in the irst round of panning and making up the majority of CEACAM6-speciic sdAbs by round 2. hus, utilizing a high-throughput screening technology was necessary in this case to recover the most desirable antibody speciicities.

DiscUssiOn
he present study has shown that ABTAG can be a useful addition to the toolbox for antibody discovery by in vitro display methods. Compared to conventional phage display approaches to antibody isolation, ABTAG technology yielded a much higher number of unique sequences while simultaneously giving detailed antigen binding information. With the rapid rise in antibody drugs in recent years, the number of diferent applications for this class of biologics is ever increasing. Along with each application is also a desirable set of kinetic parameters and functional properties (19). For instance, naked inhibitory mAbs targeting a surface receptor would preferentially have tight-binding with slow of-rates. Conversely, tumor-imaging molecules, or those carrying a toxic payload, require shorter half-lives and faster of-rates. Selection of antibodies based on kinetics using a full kinetics screening method, such as the ProteOn methods described here, is an obvious advantage, and afords a moderate throughput of up to 200 clones per day, extrapolating from the time required to screen the 52 round 1 clones described here, with generation of high quality kinetic and ainity data. Maximizing the number of clones available for functional screening is also clearly a key aspect of the search for antibodies with desired functionalities such as crossspecies reactivity and binding to speciic epitopes, for example, epitopes that deliver activation, inhibitory, internalization or no biological activity. he ABTAG technology supports the growing trend away from largely blind selection of limited numbers of binders as evidenced, for example, by the use of next generation sequencing to more efectively capture the full diversity of the immune response and bypass B-cell or antibody library screening (1,20). Other ainity tags are available for use in puriication and in screening technologies such as SPR; however, none of these tags are ideal. he use of streptavidin requires chemical or enzymatic biotinylation of the proteins being screened; SPR screening of biotinylated proteins is not always straightforward and rapid. His-tagged proteins can be captured on Ni 2+ -surfaces but the capture is not always stable enough for the analysis of high ainity interactions; assays require steps for Ni 2+ activation and regeneration which slows screening throughput and Ni 2+chelation surfaces are relatively expensive. His-tagged proteins can be captured by anti-His-tag antibodies but commercial sources of these antibodies can be costly and the antibodies can have variable binding ainity depending on the presentation of the His-tag in the context of the protein. Monoclonal antibodies can be captured on anti-Fc surfaces but the use of Fcs as fusion tags generates homodimeric fusion proteins which are undesirable for some assays. he ABTAG capture system described here is in many ways ideal: (i) the capture molecule (BSA) is readily available, inexpensive and stable on SPR surfaces, (ii) there is no indication that ABTAG negatively impacts the expression or activity of the molecules being screened, (iii) the BSA-ABTAG interaction is of very high ainity due to a very slow of-rate, and (iv) the BSA-ABTAG interaction can be completely reversed without any damage to the BSA surface.
Single-domain antibody-ABTAG screening identiied 53 anti-CEACAM6 sdAbs that were not found by a conventional phage display panning approach. he ive sdAbs identiied by A Novel Afinity Tag for sdAb Screening Frontiers in Immunology | www.frontiersin.org October 2017 | Volume 8 | Article 1406 conventional panning grouped into three families-the 2A7/2G9 family, the 1F6/2F8 family, and 1B6. Of the 32 round 2 sublibrary sdAbs, 23 belonged to the 2A7/2G9 family (12 were 2A7 or 2G9) and two belonged to the 1F6/2F8 family. he remaining seven were characterized by longer CDR3s, up to 24 residues; one of these clones was 2-15, which was the round 2 clone with the highest ainity and slowest of-rate ( Table 1). Of the 52 round 1 sublibrary sdAbs, 20 belonged to the 2A7/2G9 family (11 were 2G9 and three or four were 2A7-there was some sequence ambiguity) and one belonged to the 1F6/2F8 family. he remaining 31 were characterized by CDR3 lengths of 6-25 residues. he CDR3 lengths of clones 1-04, 1-07, 1-09, and 1-19 were 12, 11, 16, and 12 residues, respectively. Clones 1-04, 1-07, 1-09, and 1-19 had of-rates in the 6.24 × 10 −4 to 7.76 × 10 −6 s −1 range (Table 1; Figure 2) and required stringent surface regeneration conditions ( Figure S2C in Supplementary Material). It is unclear why the group of high-ainity, long-CDR3 sdAbs was less eiciently enriched by panning than the 2A7/2G9 family of medium-ainity, short-CDR3 sdAbs given that no obvious bias in their starting frequencies was observed in the library. We speculate that this observation may relate to the very slow of-rates of the former group of sdAbs which may translate into less eicient elution of these sdAb-phages by high pH; however, a stochastic element to the panning process cannot be ruled out. Clearly, as has been the experience of many groups, phage display selection is inefective in discriminating antibodies based on ainity and probably depends instead on other factors (e.g., starting frequency; fast of-rates; pH-sensitive binding; growth advantages in E. coli). he capture levels (Response; RUs) of the 2A7/2G9 moderateainity family and the rarer set of high-ainity clones (1-04, 1-07, 1-09, and 1-19) were very similar (Table S3 in Supplementary  Material), suggesting, at least at the protein level, there were no signiicant diferences in the expression of these sdAbs in E. coli supernatants. Capture levels (Response; RUs) were clearly shown to correlate with sdAb-ABTAG concentrations ( Figure S3 and Table S1 in Supplementary Material). However, it is unknown whether the display eiciency of these sdAbs on the surface of phage is comparable to their soluble expression. Another beneit of the screening method described here is that both E. coli culture supernatants and periplasmic extracts of microtiter plate cultures can be used for ABTAG SPR-based screening. Culture supernatants were used directly for of-rate screening and periplasmic extracts were used for SCK screening of the 48 round 2 clones. Similar capture levels were observed for individual clones with both sdAb-ABTAG sources.
he screening of up to 10 4 clones in a week by the approach described here is realistic, provided that antigen cost is not prohibitive. his represents very good sampling of the phage clones eluted ater a single round of panning, typically 10 5 -10 6 . Of the methods described here, the ProteOn method has the highest throughput, followed by the of-rate screening method and the SCK method, although the throughput potential of the latter two methods is not that diferent. Analysis of about 30 clones a day is possible with the SCK method. he ProteOn instrument used here has recently been discontinued and will be supported only until 2020. However, as newer higher throughput instruments become available, ABTAG screening throughput equaling or exceeding that achieved here with the ProteOn should not be an issue (21).
Beyond its usefulness in screening sdAb libraries, ABTAG has been applied to other antibody and protein engineering projects (JZ, unpublished data). Both scFv and Fab libraries, including humanization and ainity maturation libraries, can be screened in a manner similar to that described here for sdAbs. In addition to screening for ainity, ABTAG technology can be applied to screening for protein stability, an unmet need in the ield of protein engineering. When pure to very pure proteins are needed ABTAG can be used as a puriication tag for BSA-ainity matrix puriication. When the eicacy of a protein needs to be evaluated in vivo, the protein-ABTAG can be mixed with BSA to overcome the limitation of short serum half-life of the protein. Finally, screening libraries constructed from other proteins, receptors, ligands or enzymes, for which binding partners are available, could also be exploited with ABTAG. In summary, ABTAG technology provides a powerful and versatile tool for antibody discovery and protein engineering applications.

eThics sTaTeMenT
his study was carried out in accordance with Animal Use Protocols approved by the National Research Council Canada Animal Care Committee.
aUThOr cOnTriBUTiOns GH and CRM conceived experiments and planned and wrote the manuscript. GH supervised the Biacore analyses and created the igures. TB constructed the libraries, conducted the panning and coordinated the of-rate and ProteOn screening of the round 1 and round 2 library clones. JB devised and conducted the ProteOn analyses. HvF conducted the of-rate screening and SPR characterization of the puriied sdAbs and sdAb-ABTAG fusions. SR conducted the SCK screening of the round 2 clones. KH performed and analyzed NGS experiments. JZ devised the application of ABTAG to the screening of antibody libraries.

acKnOWleDgMenTs
We thank Yanal Murad and Shenghua Li for help with the round 1 and round 2 library screening. We thank Shannon Ryan for sdAb-ABTAG expression and puriication.