Selection of Single Domain Antibodies from Immune Libraries Displayed on the Surface of E. coli Cells with Two β-Domains of Opposite Topologies

Screening of antibody (Ab) libraries by direct display on the surface of E. coli cells is hampered by the presence of the outer membrane (OM). In this work we demonstrate that the native β-domains of EhaA autotransporter and intimin, two proteins from enterohemorrhagic E. coli O157:H7 (EHEC) with opposite topologies in the OM, are effective systems for the display of immune libraries of single domain Abs (sdAbs) from camelids (nanobodies or VHH) on the surface of E. coli K-12 cells and for the selection of high affinity sdAbs using magnetic cell sorting (MACS). We analyzed the capacity of EhaA and intimin β-domains to display individual sdAbs and sdAb libraries obtained after immunization with the extracellular domain of the translocated intimin receptor from EHEC (TirMEHEC). We demonstrated that both systems displayed functional sdAbs on the surface of E. coli cells with little proteolysis and cellular toxicity, although E. coli cells displaying sdAbs with the β-domain of intimin showed higher antigen-binding capacity. Both E. coli display libraries were screened for TirMEHEC binding clones by MACS. High affinity binders were selected by both display systems, although more efficiently with the intimin β-domain. The specificity of the selected clones against TirMEHEC was demonstrated by flow cytometry of E. coli cells, along with ELISA and surface plasmon resonance with purified sdAbs. Finally, we employed the E. coli cell display systems to provide an estimation of the affinity of the selected sdAb by flow cytometry analysis under equilibrium conditions.


Introduction
The expression of antibodies (Abs) in E. coli, both full-length immunoglobulin G (IgG) molecules and smaller antigen-binding fragments containing the variable (V) domains from heavy (H) and/or light (L) chains e.g. Fab, single-chain Fv (scFv), and single domain Ab (sdAb), provides a set of powerful technologies for the generation of Abs with novel specificities and improved properties [1,2]. Current selection of novel therapeutic Abs is based on hybridoma technologies using transgenic mice carrying human Ig genes [3] and screening of Ab gene libraries displayed on the surface of a biological entity [4].
The most common Ab display method is phage display, in which the V-genes are cloned in phagemids as fusions to the minor coat protein III (pIII) from filamentous bacteriophages of E. coli [5]. The Ab-pIII fusions contain a N-terminal signal peptide (SP) to translocate the Ab to the periplasm while the pIII moiety is anchored in the inner membrane (IM) [6]. Abs expressed in the periplasm of E. coli generally fold properly due to the presence of protein chaperones (e.g. Skp, FkpA) and disulfide bond forming and isomerization enzymes (e.g. DsbA, DsbC) [7]. Further, infection of E. coli cells expressing Ab-pIII fusions with a helper bacteriophage allows the production of phage particles displaying the Ab (Phabs), which can be incubated with the antigen of interest to recover antigen binding clones and amplified by infection of fresh E. coli cells (a process called biopanning).
An alternative technology for Ab display and selection in E. coli is the anchored periplasmic expression (APEx), in which the Ab fragments or full-length IgGs are expressed in the periplasm and are tethered to the IM by means of a short lipoprotein signal or an engineered lipoprotein binding the Fc region of IgGs [8,9]. In APEx, the outer membrane (OM) of E. coli is permeabilized and the generated spheroplasts are incubated with the antigen labeled with a fluorophore or biotin, and subsequently selected by fluorescence activated cell sorting (FACS).
Although phage display and APEx are robust technologies for Ab selection, alternative methods that enable the direct display of Abs or Ab libraries on the surface of E. coli cells, without the need for generation of Phabs or spheroplasts would be of great interest. In addition, E. coli cell display would facilitate selections by cell sorting methods using antigen in solution as well as the analysis of the selected clones by flow cytometry. Alternative successful cell display technologies developed for Ab selection utilize yeasts [10,11] and Grampositive bacteria [12]. In these cell display systems, the Ab fragments translocate across a single cell membrane and are anchored in the cell wall. Nevertheless, E. coli remains a more suitable microorganism for the generation, amplification and maintenance of large Ab repertoires owing to its high-efficiency of transformation and versatile expression systems. Despite these advantages, the presence of the OM has hindered the development of effective E. coli cell display methods for Ab selection, with the exception of the use of the chimeric lipoprotein Lpp-OmpA' for the display of scFvs and selection of variants with higher affinity after mutagenesis of the scFv (affinity maturation) [13][14][15]. Lpp-OmpA' consists of the Nterminal SP and first 9 residues of the mature Lpp fused to residues 46 to 159 of OmpA, which is a truncated fragment of its native 8-stranded β-barrel [16]. However, this chimeric construct lacks the characteristic stability of the β-barrel of native OM proteins (OMPs) [17] and its expression induces OM leakage as well as cellular toxicity [18,19], which may have limited its use to the affinity maturation of scFvs.
Other OMPs have also been used to display heterologous peptides and proteins on the surface of E. coli cells [20]. Among them, the autotransporters (AT) and Intimin/Invasin (Int/ Inv) proteins are very attractive display systems [21,22]. Interestingly, an AT protein (EspP) has been used in E. coli for the display and affinity maturation of an Anticalin protein scaffold binding human cytotoxic T-lymphocyte antigen 4 (CTLA-4) [23]. Protein members of the AT and Int/Inv families are large, secreted polypeptides that contain three functional regions: i) a N-terminal SP, that drives their Sec-dependent translocation across the IM; ii) a β-domain, that is anchored into the OM and comprises a 12-stranded β-barrel with a peptide linker running through the lumen of the β-barrel; and iii) a passenger region, that is secreted to the extracellular milieu [24]. Although their mechanism of secretion remains uncertain, AT and Int/Inv proteins are translocated into the periplasm and then use the β-barrel assembly machine (BAM) complex for insertion into the OM and translocation of the passenger region to the cell surface [24][25][26].
Despite their similarities, AT and Int/Inv proteins also have important differences. Firstly, they have opposite topological organization in the OM, being the passenger region located in the N-terminal portion of ATs whereas in the case of Int/Inv proteins is found in the C-terminus (Figure 1). The distinct topologies are also reflected in their β-domains. In the case of ATs, the β-barrel is preceded by α-helix linker that fills the lumen and connects its N-terminus with the passenger region [27,28]. In contrast, the β-barrel of Int/Inv proteins is followed by a peptide linker that runs through the lumen connecting its C-terminus to the passenger region [29]. In addition, the passengers of AT and Int/Inv proteins also have distinct structures, namely β-helical rods in most ATs and tandems of Ig-like domains in Int/Inv proteins [30,31].
In previous works, we have studied the secretion mechanism of ATs and reported their capacity to translocate a model sdAb to the surface of E. coli fused to the β-domain of ATs such as the IgA protease (IgAP) from Neisseria gonorrhoeae and EhaA from enterohemorrhagic E. coli (EHEC) O157:H7 [32,33]. A model sdAb fused to these β-domains was correctly folded in the periplasm with the canonical disulfide bond of Ig domains formed by the action of DsbA [32,33]. In a different study, we demonstrated that the Intimins from EHEC and enteropathogenic E. coli (EPEC) could be expressed in E. coli K-12 and displayed their native passenger domains on the cell surface with a disulfide bond formed by DsbA [25]. However, neither the display of heterologous Ig domains (e.g. sdAbs) fused to Intimin, nor the utility of the β-domains of ATs and Intimin for display of sdAb libraries and de novo selection of sdAbs against an antigen of interest was investigated.
In this work, we have demonstrated the capacity of the βdomains of EhaA and Intimin for the display of sdAb libraries on the surface of E. coli cells and for the selection of novel sdAbs against the extracellular domain of the translocated intimin receptor from EHEC (TirM EHEC ). E. coli display libraries were screened by magnetic cell sorting (MACS) and clones that specifically bound TirM EHEC were isolated, characterized by flow cytometry, as well as by enzyme linked immunosorbent assay (ELISA) and surface plasmon resonance (SPR) with the purified sdAb. In addition, E. coli display was used to provide an estimation of the affinity of the selected sdAb by flow cytometry analysis under equilibrium conditions before protein purification.

Bacterial strains, bacteriophages, growth and induction conditions
were started from individual colonies (for single clones) or from a mixture of clones (in case of libraries), freshly grown and harvested from plates, diluted to an initial OD 600 of 0.5, and grown overnight (o/n) under static conditions. For induction, bacteria (corresponding to an OD 600 of 0.5) were harvested by centrifugation (4000 xg, 5 min), and grown in the same media with 0.05 mM isopropylthio-β-D-galactoside (IPTG), but without glucose for 3 h with agitation (160 rpm), unless indicated otherwise. For over-expression of soluble VHH in the periplasm, E. coli WK6 cells with the corresponding pCANTAB6-VHH plasmid (Ap R ) were induced with 0.3 mM IPTG for 3 h at 30 °C. Secretion of VHH into the culture media was performed using the hemolysin (Hly) secretion system of E. coli HB2151 cells carrying pVDL9.3 (HlyBD; Cm R ) and the corresponding pEHlyA4SD-VHH (Ap R ) plasmids, induced with 0.3 mM IPTG for 6 h [34,35]. Over-expression of soluble TirM EHEC with N-terminal His-tag was induced in E. coli BL21 (DE3) cells carrying the pET28a-TirM EHEC plasmid (Km R ) and grown at 37 °C in LB medium containing 1.0 mM IPTG for 2 h. Further details can be found in Materials and Methods S1.

Plasmids, DNA constructs and oligonucleotides
Plasmids used in this study are summarized in Table 1. DNA manipulation, ligation, transformation and plasmid preparation were performed following standard techniques. Oligonucleotides were synthesized by Sigma Genosys, except those used for VHH amplification, which were from Scandinavian Gene Synthesis (SGS). All DNA constructs were sequenced by Secugen SL. PCR reactions for cloning were performed with proof-reading Vent DNA polymerase (New England Biolabs) or Taq DNA polymerase (Roche). The plasmid pHEA (Cm R ), for in frame fusions to N-terminal PelB signal peptide and E-tagged C-EhaA (amino acids 989-1327 of EhaA from EHEC O157:H7 strain EDL933stx-), was reported previously [33]. The amino acid sequence of the E-tag is: GAPVPYPDLEPA. The plasmid pNeae2 (Cm R ) is a derivative of the pNeae vector [25], encoding Intimin residues 1-659 (from EHEC O157:H7 strain EDL933stx-) followed by the E-tag, the hexahistidine (His) epitope, and a C-terminal myc-tag (EQKLISEED). The details of plasmid constructions are described in Materials and Methods S1.

Immunization and generation of VHH sdAb libraries
Dromedary camel immunization (Camelus dromedarious) was performed with purified TirM EHEC (~1 mgr) mixed with veterinary vaccine adjuvant (GERBU). Amplification of V HH genes was done by RT-PCR of the mRNA isolated from ~2x10 7 lymphocytes from a peripheral blood sample of the immunized animal using standard methods and primers [36]. The camel immunization protocol followed the animal experimentation guidelines published by the Canary Islands Regional Government (Spain) and was approved by the Ethics Commission of the Department of Animal Medicine and Surgery, University of "Las Palmas de Gran Canaria" (Spain). The amplified VHH fragments were digested with SfiI and NotI restriction enzymes and ligated into the same sites of purified pHEA and pNeae2 backbone vectors and finally transformed in E. coli EcM1 cells by electroporation. The size of each library was ~2-3x10 6 clones, as determined by plating on LB-Cm agar plates with 2% w/v glucose incubated at 30 °C. Further details are provided in t Materials and Methods S1.

Protein extracts, SDS-PAGE and Western blots
Whole cell protein extracts were prepared by harvesting bacteria after induction (1 ml of OD 600 1.5), resuspended in 50 µl of 10 mM Tris HCl pH 8.0, mixed with the same volume of SDS-sample buffer (2X) or urea-SDS sample buffer (2X) and boiled for 10 min (pHEA constructs) or 30 min (pNeae constructs) respectively. The boiled samples were sonicated (5 sec; Labsonic B Braun), centrifuged (14,000 xg, 5 min) to pellet insoluble material, loaded onto 8% or 10% SDS-PAGE gels and run using a Miniprotean III electrophoresis system (Bio-Rad). For Western blot, the gels were transferred to a polyvinylidene difluoride membrane (PVDF, Immobilon-P, Millipore) using a semi-dry electrophoresis transfer apparatus (Bio-Rad), the membranes were blocked and incubated with anti-E-tag mAb (Phadia) or anti-c-myc-POD mAb (clone 9E10; Roche), as indicated. Bound anti-E-tag mAb was developed using anti-mouse IgG conjugated with peroxidase (POD) (Sigma). Streptavidin-POD conjugate (Roche) was employed to detect the biotinylated broad range SDS-PAGE protein markers (Bio-Rad). All mAbs and POD conjugates were used in a 1:5000 dilution. Membranes were developed by chemiluminescence and either exposed to an X-ray film (Curix, Agfa) or scanned in a Chemi-Doc XRS (Bio-Rad) and analyzed using the Quantity One software (Bio-Rad). To quantify the total number of V HH fusions expressed in E. coli, Western blots of whole cell protein extracts and dilutions of a purified Etagged V HH of known concentration [37] (hereafter referred to as "unknowns" and "standard", respectively) were visualized on a ChemiDoc XRS and analyzed using the Quantity One software (Bio-Rad). Composition of buffers and further details are described in Materials and Methods S1.

Purification of TirM EHEC
TirM EHEC with a N-terminal His-tag was purified from 500 ml cultures of E. coli BL21(DE3) cells carrying pET28a-TirM EHEC grown and induced as described above. Bacteria were lysed and the soluble protein extract loaded onto a chromatography column filled with a Cobalt-containing resin (Talon, Clontech). TirM EHEC was eluted with 150 mM imidazole, dialyzed against HEPES-buffer (20 mM HEPES pH 7.4, 200 mM NaCl, sterile filtered and degassed), and loaded onto a calibrated gel filtration column (HiLoad 16/600 Superdex 75 preparative grade, GE Healthcare). Fractions containing TirM EHEC were collected and checked for purity by SDS-PAGE. Protein concentration was estimated using the Bicinchoninic acid (BCA) Pierce protein assay kit (Thermo Scientific). Further details are described in Materials and Methods S1.

Protein biotinylation
Biotinamidocaproate N-hydroxysuccinimide ester (Biotin-NHS; Sigma) was re-constituted at 25 mg/ml in dimethylsulfoxide (DMSO, Fluka) and immediately used. Purified protein (0.1-1 mg) [TirM EHEC , GFP (Upstate, Merck Millipore), BSA (Sigma), anti-E mAb (Phadia)] was mixed with Biotin-NHS (20-fold molar excess) in 1 ml of PBS and incubated for 2 h at RT with slow agitation on a gyratory wheel. The reaction was stopped by addition of Tris-HCl pH 7.5 at final concentration of 50 mM and the samples were placed on ice for 1 h. The reaction mix was loaded onto a pre-packed column for gel filtration chromatography (Sephadex G25 PD-10; GE Healthcare) and the biotinylated protein was eluted in 500-µl fractions with PBS. Protein concentration was estimated using the Bicinchoninic acid (BCA) Pierce protein assay kit (Thermo Scientific).

Enzyme-linked immunosorbent assay (ELISA)
TirM EHEC or BSA (Sigma) proteins were adsorbed at 4°C o/n onto 96-well immunoplates (Maxisorb; Nunc) at a concentration of 5 µg/ml in PBS. Next, immunoplates were washed in PBS and blocked by incubation with 200 µl of 3% (w/v) Milk -PBS for 2 h at RT. The sdAbs (secreted or purified) were diluted in 3% (w/v) Milk-PBS, added at the indicated concentrations (0.1-100 nM) in duplicates and incubated for 1 h at RT. After incubation, the wells were washed three times with PBS (Immunowash 1575, Bio-Rad) and the bound sdAbs was detected by the addition of anti-c-myc-POD mAb (clone 9E10; Roche; 1:1000), or anti-E-tag mAb (Phadia; 1:1000) followed by anti-mouse-POD (Sigma; 1:1000) for E-tagged sdAb, and incubation of the plates for 1 h at RT. The plates were washed three times with PBS and developed with H 2 O 2 and o-phenylenediamine (OPD; Sigma) as previously described [38]. The plates were read at 490 nm using the iMark ELISA plate reader (Bio-Rad).

Magnetic Cell Sorting (MACS)
Induced E. coli cells (equivalent to a final OD 600 of 5.0) were harvested by centrifugation (4000 xg, 3 min), washed three times with 2 ml PBS (sterile filtered and degassed), and resuspended in a final volume of 1 ml of PBS. Biotinylated TirM EHEC (at concentrations of 50 nM or 250 nM, as indicated) was added to 100 µl of bacteria, the final volume was adjusted to 200 µl with PBS-BSA (PBS supplemented with 0.5% w/v BSA, sterile filtered and degassed), and incubation was carried out for 1 h at RT. After incubation, bacteria were washed three times with 1 ml of PBS-BSA, resuspended in 100 µl of the same buffer containing 20 µl of anti-biotin paramagnetic beads (Miltenyi Biotec) and incubated at 4 °C for 20 min. Next, bacteria were washed three times with 1 ml of PBS-BSA, resuspended in 500 µl of the same buffer, of which 10 µl was kept aside to calculate the input bacteria before the procedure, while the rest (490 µl) was applied onto a MACS MS column (Miltenyi Biotec), previously equilibrated with 500 µl of PBS-BSA and placed on the OctoMACS Separator (Miltenyi Biotec). The flow through of unbound cells was collected and the column was washed three times with 500 µl of PBS-BSA. The wash was combined with the flow-through as "Unbound fraction". Next, the column was removed from the OctoMACS Separator and placed onto a new collection tube, 2 ml of LB was added and the cells were eluted out. This fraction was labeled as the "Bound fraction". Serial dilutions of Unbound and Bound fractions were plated to determine CFU and to harvest Bound bacteria.

Flow cytometry analysis
For standard flow cytometry, induced bacterial cells (equivalent to a final OD 600 of 1.0; ~10 9 CFU) were harvested by centrifugation (4,000 xg, 3 min), washed twice with 500 µl of PBS (filter-sterilized) and resuspended in a final volume of 400 µl of PBS. Next, 190 µl of this cell suspension (~3x10 8 CFU) was incubated with the primary antibody or antigen (as indicated) and PBS was added to adjust the total volume to 200 µl. The primary antibodies (for assay of expression levels) were anti-E-tag mAb (1:

Affinity determination by flow cytometry analysis
Induced E. coli cells (equivalent to final OD 600 of 1) were centrifuged (4000 xg, 3 min), washed twice with 1 ml of PBS (filter-sterilized) and resuspended in a final volume of 1 ml of PBS. Next, 50 µl of this cell suspension (~3x10 7 CFU) was incubated at room temperature for 90 min with a fixed amount of biotinylated TirM EHEC (2 pmols; ~30 ng) and increasing volumes of PBS (from 0.1 to 1.5 ml) to attain a final concentration range between 20 nM to 1 nM. After incubation, cells were centrifuged (4000 xg, 3 min), washed twice with 1 ml of PBS (filter-sterilized) and labeled with Streptavidin-PE as described for standard flow cytometry. After a final washing step with PBS, the mean fluorescence intensity (MFI) of Phycoerythrin (PE) was quantified in a cytometer (Gallios, Beckman Coulter). Data of MFI (relative values to maximum MFI) obtained from the cytometer were plotted against the concentration of TirM EHEC to obtain the dissociation constant (K D ). Curve was fitted according to non-linear least squares regression method and one site -specific binding saturation kinetics model using the data analysis tool in Prism software (GraphPad).

Secretion of sdAbs into E. coli culture media
Secretion of Ab fragmemts to E. coli culture media with the hemolysin secretion system has been reported previously [34,35]. Induced cultures of 10 ml E. coli HB2151 cells transformed with pVDL9.3 and pEHlyA4SD-V HH derivative were centrifuged (10000 xg, 10 min, 4°C) and the culture supernatants were adjusted to PBS 1X by adding 1/10th volume of PBS 10X. The amount of secreted V HH -HlyA fusion (~5 µg/ml) was estimated by densitometric analysis of silverstained SDS-polyacrylamide gels loaded with 10 µl samples of culture supernatants and dilutions of BSA standards of known concentration (Thermo Scientific). When needed, the V HH -HlyA fusions in culture supernatants were concentrated with 3-kDa or 10-kDa centrifugal filter units (Amicon Ultra-15).

Purification of sdAbs from the periplasm of E. coli
Soluble sdAbs with hexahistidine and myc tags in their Ctermini were induced in E. coli WK6 cells carrying pCANTAB6-VTIR1 or pCANTAB6-Vgfp. Cells were pelleted by centrifugation (4000 xg, 12 min, 4°C) from 1 L cultures, resuspended in 22.5 ml Periplasmic Extraction buffer [50 mM Sodium phosphate pH 7.4, 200 mM NaCl, 5 mM EDTA and 1 mg/ml polymyxin B sulphate (Sigma)] and stirred at 4°C for 2 h using a magnetic stirrer. The periplasmic extract was obtained by ultracentrifugation (40000 xg, 30 min, 4°C) and dialyzed o/n at 4°C against 5 L of PN2 buffer (50 mM sodium phosphate pH 7.4, 200 mM NaCl). Dialyzed extract was loaded onto a Cobaltcontaining affinity resin (Talon, Clontech), washed, and bound protein eluted in PN2 with 150 mM imidazole. Eluted sdAb was dialyzed, concentrated, and loaded onto a calibrated gel filtration column (HiLoad 16/600 Superdex 75 preparative grade, GE Healthcare) as described previously for TirM EHEC . The fractions corresponding to the monomeric sdAb were collected and concentrated in a 3-kDa centrifugal filter unit (Amicon Ultra-15). Protein concentration was estimated using the Bicinchoninic acid (BCA) Pierce protein assay kit (Thermo Scientific).

Surface Plasmon Resonance affinity determination
SPR measurements were performed using a Biacore 3000 instrument (GE Healthcare). All proteins solutions were dialyzed against HEPES-buffer [20 mM HEPES 200 mM NaCl (pH 7.4) sterile filtered and degassed] at 4°C o/n. Biotinylated TirM EHEC (0.1 µg/ml) was immobilized on a Streptavidin SA chip (GE Healthcare) at 150 response units (RU) at a flow rate of 10 µl/min in HEPES-buffer containing 0.005% (v/v) of the surfactant Polysorbate 20 (P20, GE Healthcare). For determination of binding kinetics, dilutions of purified sdAb (analyte) from 32 nM to 200 pM were flown at 30 µl/min in HEPES-buffer and sensograms were generated. The biotinylated TirM EHEC surface on the Streptavidin SA chip was regenerated after every cycle using three injections of 10 µl of 10 mM Glycine-HCl (pH 1.7). Sensograms with different concentrations of anaylte were overlaid, aligned and analyzed with BIAevaluation 4.1 software (GE Healthcare). All data were processed using a double-referencing method [39].

Comparison of EhaA and Intimin β-domains for display of sdAbs on E. coli
To test the potential of Intimin β-domain for the display of Ab fragments and compare it with the display capacity of EhaA βdomain, we employed sdAbs derived from heavy-chain antibodies (HCAbs) of camelids (e.g. dromedaries, llamas) also known as VHH (for VH of HCAbs) or nanobodies [40]. These sdAbs have a small size (ca. 12-14 kDa), high stability and solubility, as well as excellent antigen binding properties with high affinity and specificity. In addition, their sequence identity with human VH makes them attractive for therapy [41]. Initial assays were conducted using a model VHH clone binding GFP (Vgfp) [42] fused to the β-domains of EhaA and Intimin for comparison of the two systems before cloning of an immune library of sdAbs. Vgfp was cloned in pHEA vector (Table 1) in frame with the N-terminal SP of PelB [43] and the C-terminal fragment of EhaA (residues 989-1327; named as C-EhaA), bearing its native β-barrel with α-helix linker, and including the E-tag epitope between the VHH and C-EhaA ( Figure 1A) [33]. The Vgfp sequence was cloned in pNeae2 (Table 1) in frame with the N-terminal fragment of Intimin (residues 1-659; named as Neae), comprising its N-terminal SP, the periplasmic LysM domain (expected to bind the peptidoglycan), its native β-barrel with C-terminal linker, and the first Ig-like domain (D0) ( Figure  1B). Vector pNeae2 incorporated the E-tag and myc-tag epitopes flanking the VHH ( Figure 1B). Both E. coli display vectors contained unique SfiI and NotI restriction sites flanking the VHH in the same frame as those of conventional phagemids (e.g. pHEN6, pCANTAB6) [44]. The resulting fusion proteins were referred to as VHHA (fusions to C-EhaA) and NVHH (fusions to Neae).
The expression of VgfpA and NVgfp fusions in E. coli K-12 cells (strain UT5600; Table 1) was analyzed by Western blot after induction with 0.05 mM IPTG at 30 °C for 3 h (see Materials and Methods). Discrete protein bands corresponding to VgfpA and NVgfp were detected with anti-E or anti-myc monoclonal Abs (mAbs) in whole cell protein extracts from the induced cells ( Figure 1C and Figure 1D). Both fusion proteins showed a shift in their electrophoretic mobility, characteristic of native OMPs with correctly folded β-barrels, which are resistant to SDS denaturation at low temperatures and hence migrate faster than the unfolded polypeptides due to the compact structure of the β-barrel [17,45]. This heat-modifiable mobility was observed in both VgfpA and NVgfp, having the expected mobility according to their molecular weight (MW) after boiling (ca. 52 kDa and 84 kDa, respectively) and a faster mobility in non-boiled samples ( Figure 1C and 1D). Interestingly, NVgfp was resistant to 2% SDS and 4 M urea at low temperatures (i.e. 22 °C) and required boiling in this buffer to unfold, as previously reported for full-length Intimin and its β-domain [25]. The major protein bands detected with anti-E mAb in the boiled samples corresponded to full-length VgfpA and NVgfp fusions ( Figure 1C and 1D; labeled with arrows). Detection of NVgfp with anti-myc mAb confirmed the integrity of its C-terminal end ( Figure 1D). Minor bands of lower MW were also detected in Western blot with anti-E mAb, which likely represent proteolytic fragments of the full-length fusions ( Figure 1C and 1D; labeled with asterisks). The induced E. coli cultures expressing VgfpA or NVgfp fusions showed only a slight decrease in their growth rate compared to control cultures having the empty vector (pAK-Not), or expressing the β-domains C-EhaA (pHEA) or Neae (pNeae2), and reached final optical densities at 600 nm (OD 600 ) identical to controls ( Figure S1).
The accessibility of VgfpA and NVgfp fusions to the external milieu was initially compared by incubation of intact E. coli cells with externally added proteases. Trypsin digested full-length VgfpA leaving some weakly detectable proteolytic fragments with a size similar to C-EhaA and Vgfp domains ( Figure 1C, lane 3). The NVgfp fusion was resistant to Trypsin digestion ( Figure S2) but was sensitive to Proteinase K (ProtK) ( Figure  1D, lane 3). Nevertheless, ProtK digestion of NVgfp fusion left a resistant fragment comprising Neae. Hence, both the βdomains display the sdAb to the extracellular milieu, in a way that is accessible to externally added proteases but C-EhaA fusions are more sensitive to digestion than Neae fusions. Resistance to proteolysis was previously observed for fulllength Intimin [25].
Surface display of VgfpA and NVgfp was assessed by flow cytometry (Figure 2). Induced E. coli cells harboring pVgfpA, pNVgfp, or pAK-Not (control) were stained with anti-E or antimyc mAbs followed by anti-mouse IgG-Alexa488 (Figure 2, left panels). E. coli cells expressing VgfpA or NVgfp were positively bound by anti-E mAb, though cells expressing NVgfp were also positively bound with the anti-myc mAb. Control E. coli cells with pAK-Not were negative for both mAbs. Importantly, the presence of a single peak in the flow cytometry histograms indicated that most E. coli cells were expressing a homogenous level of the fusion proteins. The mean fluorescence intensity (MFI) of cells with anti-E-tag mAb suggested a higher expression and display level of NVgfp than VgfpA (~3-fold).
The antigen-binding activity of the surface displayed Vgfp sdAb using both display systems was compared by flow cytometry after incubation of E. coli cells expressing the fusions with 50 nM biotin-labeled GFP (positive antigen) or biotinlabeled BSA (negative antigen), followed by incubation with Streptavidin-Phycoerythrin (PE) conjugate (Streptavidin-PE) (Figure 2, right panels). This analysis showed the specific binding of the E. coli cells expressing VgfpA or NVgfp fusions to GFP, whereas control E. coli cells did not bind GFP. None of these cells bound BSA. Therefore, the sdAb is functional when displayed on the surface of E. coli cells with the β-domains of EhaA and Intimin. Nevetheless, the MFI of GFP binding was higher in E. coli cells expressing NVgfp than in those with VgfpA (ca. 8-fold) (Figure 2, right panels). The 3-fold difference in expression level of NVgfp does not appear to be sufficient to account for this difference in binding, suggesting that the sdAb may have a higher antigen-binding activity when fused to the βdomain of Intimin (see Discussion).

Selection of antigen binding sdAbs from an immune library displayed on E. coli cells
To test the effectiveness of the E. coli display systems for the expression of a VHH library and selection of antigen-binding Selection of sdAbs by E. coli Display PLOS ONE | www.plosone.org clones, an immune library against the soluble extracellular fragment of the translocated intimin receptor (tir) from EHEC (named TirM EHEC , corresponding to residues 252-360 of fulllength Tir EHEC ) [46], was generated and cloned in vectors pHEA and pNeae2. TirM EHEC binds to the C-terminal Ig-like and lectinlike domains (D2-D3) of full-length Intimin EHEC [30,47] but not to the β-domain of Intimin (Neae) used for in this study. The VHH library against TirM EHEC was obtained by immunization of a dromedary with purified recombinant his-tagged TirM EHEC and subsequent amplification of the VHH gene segments from 2x10 7 lymphocytes isolated from a peripheral blood sample (Materials and Methods). The amplified VHH gene segments were cloned into the SfiI and NotI sites of pHEA and pNeae2 vectors, generating two E. coli display immune libraries of similar size (~2-3x10 6 clones). This relatively small library size is reported to be sufficient for having a good representation of the repertoire of V HH genes in the peripheral blood of camelids after immunization and to select the sdAbs of higher affinity raised against the antigen using conventional phage display [48]. The E. coli strain EcM1 (Table 1) was used as host for cell display. This strain is derived from the reference wild type K-12 strain (MG1655) with a deletion in the operon encoding type 1 fimbriae (ΔfimA-H) [49]. Sequencing of 40 clones picked randomly from each library confirmed the cloning of different VHH sequences in frame with the β-domains of EhaA and Intimin (data not shown).
The expression and display of the VHH libraries with the βdomains of EhaA (VHHA) and Intimin (NVHH) were analyzed by flow cytometry with anti-E and anti-myc mAbs, revealing a fairly homogeneous expression of both libraries in E. coli EcM1 ( Figure 3A). The MFI with anti-E mAb indicated a similar expression level of VHHA and NVHH libraries, in contrast to the significantly lower expression of VgfpA observed previously. Western blot analysis of whole-cell protein extracts from induced cultures revealed major protein bands with the expected size for full-length VHHA and NVHH fusions, upon boiling in SDS or SDS-urea buffer, respectively, and which have heat-modifiable electrophoretic mobility indicating the correct folding of their β-barrels ( Figure 3B). Quantification of the Western blot signals with anti-E mAb using ~1.5x10 8 bacteria (0.15 units of OD 600 ) expressing VHHA or NVHH fusions, was carried out using a standard curve generated with a purified E-tagged VHH of known concentration ( Figure S3), and allowed an estimation of ~6.5x10 3 molecules of VHHA and 7.8x10 3 molecules of NVHH per bacterium. The growth of E. coli cultures expressing VHHA or NVHH libraries was only slightly delayed compared to a control with pAK-Not and the cultures reached similar OD 600 after induction ( Figure S4). Plating of the induced cultures to determine the number of colony forming units (CFU) per OD 600 gave ~1.0x10 9 CFU/OD 600 in the control, ~ 0.9 x10 9 CFU/OD 600 in the VHHA library, and ~0.6x10 9 CFU/OD 600 in the NVHH library. Hence, although growth is not affected, expression of NVHH fusions appear to reduce the viability of E. coli cells compared to control cultures. Nevertheless, since the CFU/OD 600 after expression of NVHH fusions remains within the same order of magnitude as the control, the diversity of the sdAb library is not compromised. Cell toxicity for the expression of Intimin constructs has been reported previously in some E. coli K-12 strains [22].
E. coli EcM1 cells expressing VHHA and NVHH libraries were screened to isolate clones binding to TirM EHEC by magnetic cell sorting (MACS). In MACS, cells are incubated with a biotinylated target protein (e.g. antigen, mAb, etc.), washed to remove the unbound protein, and incubated with a suspension of paramagnetic microbeads coupled to Streptavidin or a mAb binding biotin (anti-biotin microbeads). The cell mixture is then passed through a small ferromagnetic column held in a magnetic holder, which retains cells coated with anti-biotin microbeads whereas unbound cells are washed out of the column ( Figure 4A). Elution of bound cells is carried out with buffer or fresh culture media (e.g. LB) upon removal of the column from the magnet. The CFU washed out of the column (Washed fraction) and eluted (Bound fraction) are determined by plating. The MACS conditions for capturing E. coli cells expressing VHHA or NVHH fusions were established using biotin-labeled anti-E mAb. An initial input of 0.1 units of OD 600 (~6-9x10 7 CFU of each culture) was incubated with 50-250 nM of biotinylated anti-E mAb allowing the recovery of a total of ~2-4x10 7 CFU in the Washed and Bound fractions. The Bound fractions contained ~95-99% of the CFU in the VHHA and NVHH libraries, and only ~0.2-0.5% in control E. coli cells with pAK-Not. Using these conditions, the VHHA and NVHH libraries were incubated with biotinylated TirM EHEC (250 nM) for the first selection step (MACS1) and ~0.3-0.6% of the total CFU from both libraries were collected in the Bound fractions ( Table 2). The colonies grown from the Bound fractions of each library were pooled independently and their plasmids purified and electroporated into fresh E. coli EcM1 cells to obtain VHHA and NVHH sublibraries (≥2x10 6 transformants). This step was added to avoid multiple rounds of induction of the same bacterium, which may favor the amplification of clones with reduced levels of expression of the fusions.
Next, the VHHA and NVHH sublibraries were subjected to a new round of selection with biotinylated TirM EHEC using conditions identical to those used in MACS1. Bacteria harvested from Bound fractions were pooled, their plasmids purified and transformed for the following rounds of MACS. Antigen concentration was reduced to 50 nM in the following MACS. The percentage of E. coli bacteria recovered in the Bound fractions showed a significant increase from the initial 0.3-0.6% to over 70% in MACS3 of NVHH and MACS4 of VHHA (Table 2), suggesting an enrichment of antigen binding clones in both libraries. Bacteria from the different rounds of selection were analyzed by flow cytometry to test their binding to biotinylated TirM EHEC (50 nM) ( Figure 4B), which demonstrated an enrichment of E. coli cells binding to TirM EHEC along the selection rounds, from ~0.2% positives in the original libraries to ~45% after MACS4 of the VHHA library and more than 75% positives after MACS2 and MACS3 of the NVHH library. No significant binding to biotinylated BSA was detected by flow cytometry in these pools (data not shown). The expression levels of the VHHA and NVHH fusions in the bacterial pools obtained after MACS was similar to those of the original libraries (data not shown).
Fifty colonies from the final round of selection of each library were randomly picked for plasmid isolation and DNA sequencing. A VHH sequence, named as VTIR1, was found in all NVHH clones and in 36 VHHA clones, the rest being different VHH sequences. Flow cytometry analysis confirmed the specific binding of biotinylated TirM EHEC (50 nM) by E. coli cells displaying VTIR1 fused to EhaA and Intimin β-domains whereas these cells did not bind to biotinylated BSA ( Figure 5A and 5B). Similar to the situation observed with Vgfp clone, the MFI of E. coli cells displaying VTIR1 was higher in the NVTIR1 fusion than with VTIR1A fusion, although both were expressed at similar levels ( Figure S5). We conducted a flow cytometry screening of the remaining 14 non-VTIR1 clones from MACS4 of VHHA library, identifying three other VHH sequences that bound to biotinylated TirM EHEC , referred to as VTIR2, VTIR3, VTIR4 ( Figure 5A, and data not shown). VTIR4 did not bind biotinylated BSA, but VTIR2 and VTIR3 clones exhibited non-specific binding to biotinylated BSA and were not further analyzed ( Figure 5A and 5B). We sought for additional VHH sequences binding TirM EHEC in the NVHH library by screening 96 colonies picked randomly after the first round of selection (MACS1), in which a higher diversity of binders with low and high affinities could be expected. PCR screening with a specific primer hybridizing the complementarity determining region 3 (CDR3) of VTIR1 enabled us to identify 17 VTIR1 clones out of these 96 colonies. This number is higher than the 1.3% positives found by flow cytometry after MACS1 of the NVHH library ( Figure 4B) but fits better with the percentage of clones recovered from this population after MACS2 (Table 2). This could indicate a lower sensitivity of flow cytometry to actually discriminate positive binders from the population. Flow cytometry screening of the remaining clones allowed the identification of two additional VHH sequences that specifically bound biotinylated TirM EHEC , VTIR4 (2 clones), which was found previously in the VHHA library, and VTIR5 (1 clone) ( Figure  5B). The MFI of these clones with 50 nM biotinylated TirM EHEC was low compared to VTIR1 ( Figure 5B) and increased at higher antigen concentrations (200 nM) (data not shown) suggesting that these clones had a lower affinity for TirM EHEC . The CDR3 amino acid sequences of the selected VHH are shown in Table 3. The sdAbs encoded by VTIR1, VTIR4, VTIR5 and one unrelated V HH binding α-amylase as a control, were secreted into E. coli culture media as soluble fragments with the hemolysin system [34,35], and used in ELISA plates coated with TirM EHEC and BSA (Materials and Methods). This experiment showed that soluble VTIR1, VTIR4, and VTIR5 sdAbs, lacking the β-domains of EhaA and Intimin, also bound specifically to TirM EHEC (Figure 6). VTIR1 was the clone with an apparent higher affinity, as could be inferred from its enrichment in both the E. coli display libraries.

Characterization of VTIR1 sdAb and determination of its affinity by surface plasmon resonance and E. coli display
The VTIR1 clone was produced in the periplasm of E. coli WK6 cells as soluble sdAb with C-terminal His-and myc-tags and purified by metal-affinity chromatography followed by gelfiltration chromatography (Materials and Methods). As a control, Vgfp was also expressed and purified in the same manner. Both sdAbs behave as monomers with an apparent mass of ~15 kDa in gel filtration chromatography ( Figure S6A). The binding activity of the purified VTIR1 was confirmed in ELISA ( Figure S6B). In order to determine the apparent equilibrium dissociation constant (K D ) between VTIR1 and TirM EHEC their interaction was studied in surface plasmon resonance (SPR) experiments with a Streptavidin (SA) sensor chip coated with biotinylated TirM EHEC (Materials and Methods). The change in resonance units (RU) was recorded with time at different concentrations of purified VTIR1 from 0.2 to 32 nM showing a clear binding to TirM EHEC that reached the steady state equilibrium in ~220 s for the two highest concentrations used, but not for the lower concentrations ( Figure 7A). No binding was observed when Vgfp (40 nM) was flown over this Injection of buffer to evaluate the dissociation of VTIR1 (labeled with an arrow in Figure 7A) showed no loss of RU for > 200 s, indicating that VTIR1 remained stably bound to TirM EHEC over long periods of time. The absence of dissociation of VTIR1 from TirM EHEC prevented determination of its kinetic constants (k on and k off ) from the sensograms. Since reaching the steady state at the lower concentrations of VTIR1 would require very long injection times (>1 h), which are not practical due to the higher amounts of Ab needed and the possibility of conformational changes of the antigen bound to the sensor chip, in order to have an estimation of the apparent K D between VTIR1 and TirM EHEC we plotted the RU values obtained with the different concentrations of VTIR1 at 220 s, the time at which two concentrations had reached the steady state. This plot provided an apparent K D of ~2.2 x 10 -9 M ( Figure 7B). Nevertheless, the actual K D between VTIR1 and TirM EHEC should be below this estimated value (indicating an even higher affinity of VTIR1 for TirM EHEC ) since the steady state equilibrium has not been reached at the lower concentrations.
Flow cytometry analysis under equilibrium conditions has been used to estimate the apparent K D of Abs and Anticalins displayed on the surface of yeast and E. coli cells [14,23,50]. Thus, we tested whether the affinity of VTIR1 could also be estimated by flow cytometry analysis of E. coli cells with this sdAb on their surface and incubated with biotinylated TirM EHEC under conditions expected to be close to the equilibrium. We chose the Intimin display system given its superior MFI signals in flow cytometry with the antigen. E. coli EcM1 cells displaying NVTIR1 (~3x10 7 CFU) were incubated for 90 min with a fixed amount of biotinylated TirM EHEC (2 pmols) in two-fold increasing volumes of PBS (from 0.1 to 1.5 ml) to reach a final concentration range from 20 nM to 1 nM. After this incubation, cells were washed and labeled with Streptavidin-PE as previously described. The relative MFI of the cells was plotted against the antigen concentration used and the curve fitted by non-linear least squares regression, giving an estimated apparent K D of 1.7 x10 -9 M ( Figure 7C). This value is consistent with the estimated apparent K D determined by SPR analysis, indicating that these E. coli display systems could also be used to estimate the K D of selected sdAbs before purification.

Discussion
In this work, we have demonstrated that the β-domains of EhaA and Intimin from EHEC O157:H7 [51] are effective platforms for the display sdAb libraries on the surface of E. coli K-12 cells, and allow the selection of high affinity sdAbs from immune libraries using biotinylated antigen and MACS. Despite their opposite topologies, both systems express stable fusion proteins with the native β-barrel correctly folded in the OM and display a functional sdAb with antigen-binding capacity on the surface of E. coli cells. Most sdAbs in the immune library were  displayed on E. coli at good expression levels with both C-EhaA and Neae, with an average between 6000-8000 molecules/bacterium. From the immune libraries constructed in our study, we obtained a total of five independent camelid V HH sequences binding the antigen used in the immunization (TirM EHEC ), being the sdAb of higher affinity and specificity (VTIR1) the more frequent clone found in the selections with both β-domains. These results demonstrate that both E. coli display systems can be used to retrieve high-affinity binders from immune libraries of camelid V HH sequences. Fewer different binders were found compared to other reported studies using an immune library of camelid V HH displayed on the surface of phages and Staphyloccous carnosus cells [12]. However, we think that the relatively small number of binders obtained in our study reflects a limitation of the anti-TirM EHEC library employed, and not of the E. coli-display systems described. Firstly, we have performed selections of the anti-TirM EHEC V HH library on phage that failed to retrieve other specific binders than those selected by E. coli-display (data not shown). Secondly, ongoing work in our laboratory with other immune V HH libraries also seem to indicate that essentially the same pool of binders can be isolated from phage and E. colidisplay systems. Thus, it appears that rather than the display system used, the actual diversity of binders retrieved from an immune V HH library is more dependent on factors such as immunogenicity of the antigen, number of animals used in the immunization, and library size.
The small molecular weight, high protein solubility and stability of camelid V HH domains are likely to help their effective translocation across the OM fused to the β-domains of EhaA and Intimin. Similar properties are also found in other natural sdAbs, like VNARs from sharks [52], and have been engineered in synthetic libraries based on human V H s and V L s [53][54][55]. Thus, most types of sdAbs could be efficiently displayed on E. coli cells with EhaA and Intimin β-domains. Larger Ab fragments based on a single polypeptide, such as scFvs, also have the potential to be displayed with these βdomains. However, the tendency of some scFv clones to oligomerize and aggregate may hinder their translocation across the OM [32]. Nonetheless, this could be advantageous for the selection of highly stable and soluble scFvs from scFv libraries. Lastly, Ab molecules with separate H and L polypeptide chains (e.g. Fabs, IgGs) cannot be displayed with the β-domains reported in this work (at least in their current configuration), and phage display and APEx should be used for their selection in E. coli [8,9,56].
Despite the functionality of the β-domains of EhaA and Intimin, we observed some important differences between these E. coli display systems. Firstly, NVHH fusions were found to be more stable than VHHA fusions in vivo (to E. coli proteases) and in vitro (to externally added proteases). This difference could be due to the natural resistance of Intimin to proteolysis and denaturation [25] and/or to the susceptibility of certain ATs to bacterial proteases as part of their secretion mechanism [24,26]. Secondly, expression of VHHA clones appeared to be more variable than NVHH fusions, with some clones showing significantly lower expression levels (e.g. VgfpA). This suggests that the N-terminal fragment of Neae could have a positive effect on the expression of sdAbs, similar to other N-terminal fusion partners (i.e. MBP, Trx1, GST) used for production of recombinant proteins and Ab fragments [57]. Thirdly, we found that the antigen-binding activity of NVHH fusions was at least 3-fold higher than that of VHHA fusions, as indicated by the flow cytometry signals of E. coli cells displaying these fusions when incubated with their cognate antigens (GFP or TirM EHEC ). The lower antigen binding signals of E. coli cells displaying VHHA fusions was not explained by a different expression level and indicated the existence of additional factors. Although partial misfolding of VHHA fusions cannot be excluded, this possibility seems unlikely because both EhaA and Intimin β-domains use a common secretion pathway exposing the sdAb to periplasmic chaperones and DsbA before their translocation across the OM [25,33]. Alternatively, the longer linker region in NVHH fusions (with the D0 domain) could make the sdAb more accessible for the extracellular antigen by increasing its distance from the OM. The improved stability, expression and antigen-binding activity of NVHH fusions could explain why selection of TirM EHEC binders was more efficient in the NVHH library than in the VHHA library, reaching a higher percentage of positive antigen binding clones in fewer selection rounds. From our data, the only limitation of the Neae display could be the ~40% reduction in viability of E. coli cultures expressing NVHH fusions (estimated as CFU/OD 600 ). This reduction in viability does not have a significant effect on the representation of immune libraries with diversity ~10 7 clones, since an excess of input bacteria over the Ab library size is used during MACS.
We chose MACS to select and recover E. coli cells bound to the antigen since this technology does not require the use of expensive cell-sorting equipment as in the case of FACS, and multiple samples along with controls can be processed in parallel. We employed a manual MACS system that can hold up to eight mini-MACS columns simultaneously, each with a capacity for ~10 8 bacteria. In addition, MACS can be scaled up using multiple columns of higher capacity (each with a capacity of 10 9 -10 10 bacteria) and it can be automated, which would allow the screening of large Ab libraries faster and more efficiently than FACS. Given the maximum density of E. coli cells that can be manipulated at ease (~2-3x10 10 CFU/ml), naive and synthetic E. coli display libraries should have a maximum diversity between 10 9 -10 10 clones in order to ensure a representation of all clones during selection. In fact, these numbers are very similar to the size of most naive and synthetic libraries of Ab fragments constructed in phage display vectors. Although bacteriophages can be produced and handled in higher densities than E. coli bacteria (i.e. ~10 13 phages/ml) the diversity of phage display libraries is strongly limited by E. coli factors (i.e. transformation efficiency and the ability to infect cultures with sufficient number of bacteria). Hence, E. coli display is suitable for immune libraries with 10 6 -10 7 clones, as shown in this work, but it could also be applied to naive and synthetic libraries of ~10 9 clones [54,55,58]. For screening of Ab libraries with higher diversities, phage display and, especially, cell-free display systems (e.g. ribosome display) would be more appropriate [4,59].
In conclusion, the E. coli cell display systems reported in this work represent a good alternative to phage display and APEx display systems for selection of positive binders from sdAb libraries. The major benefit of E. coli display over phage display is the use of flow cytometry for the direct determination of the expression levels and antigen binding specificity and affinity (K D ) of the selected sdAb clones. Secondly, E. coli display makes possible the use of FACS, alone or in combination with MACS, for selections with antigen in solution. In addition, the multivalent Ab display and the less sticky properties of E. coli cells compared to filamentous bacteriophages, may reduce the background binding when complex antigenic surfaces (e.g. mammalian cells, tissues and organs) are used for selections. Although E. coli cells are more sensitive than bacteriophages to extremes of pH, temperature and other strong denaturants, E. coli cells having the OM can be washed with most common buffers and tolerate significant concentrations of detergents (e.g. 0.1-0.4% w/v) such as TX-100, SDS or deoxycholate [60]. This represents an improvement over the washing conditions tolerated by spheroplasts in APEx. Future work would aim to exploit the above-mentioned advantages of E. coli display for selections of novel Ab fragments of biomedical and biotechnological interest.