Simultaneous Surface Display and Secretion of Proteins from Mammalian Cells Facilitate Efficient in Vitro Selection and Maturation of Antibodies

Background: Simultaneous cell surface display and secretion is desirable for protein evolution and selection.

Results: Alternative splicing enables simultaneous cell surface display and secretion of the same protein or an alternate form to facilitate screening.

Conclusion: Analysis of secreted protein complements cell surface display for evolution of protein function.

Significance: This technology can enable rapid evolution and characterization of proteins with desired functional properties.

Abstract

A mammalian expression system has been developed that permits simultaneous cell surface display and secretion of the same protein through alternate splicing of pre-mRNA. This enables a flexible system for in vitro protein evolution in mammalian cells where the displayed protein phenotype remains linked to genotype, but with the advantage of soluble protein also being produced without the requirement for any further recloning to allow a wide range of assays, including biophysical and cell-based functional assays, to be used during the selection process. This system has been used for the simultaneous surface presentation and secretion of IgG during antibody discovery and maturation. Presentation and secretion of monomeric Fab can also be achieved to minimize avidity effects. Manipulation of the splice donor site sequence enables control of the relative amounts of cell surface and secreted antibody. Multi-domain proteins may be presented and secreted in different formats to enable flexibility in experimental design, and secreted proteins may be produced with epitope tags to facilitate high-throughput testing. This system is particularly useful in the context of in situ mutagenesis, as in the case of in vitro somatic hypermutation.

Introduction

Protein expression in mammalian cells is an established technology and is used to manufacture the majority of therapeutic proteins currently approved for human use. Mammalian cells are especially attractive for protein manufacturing because of their efficient machinery for protein folding, their ability to assemble multiple polypeptide chains, efficient secretion, and the authentic post-translational modifications that can be achieved when expressing proteins of mammalian origin. Expression of IgG molecules requires correct folding and assembly of multi-domain protein heavy and light chains, glycosylation, and in some cases other post-translational modifications such as sulfation (1). Although mammalian cell expression is the dominant technology for therapeutic antibody production, in vitro technology for isolation and engineering of antibodies has relied primarily on non-mammalian expression, using techniques such as phage display, requiring selections to be carried out with antigen-binding fragments of antibodies such as scFv and Fab (2). The resultant antibody fragments may be poorly expressed, can be time consuming to re-engineer, and do not always fold in the same manner when expressed in mammalian cells (3). To ensure the selection of high quality antibodies required for pharmaceutical development with the need for excellent biophysical properties and high level expression in mammalian cells, an in vitro system that we term “SHM-XEL” has been developed that allows direct selection and maturation of antibodies in the same mammalian cell types that can be used for efficient manufacturing of IgG (Fig. 1) (4).

FIGURE 1.

SHM-XEL enables the simultaneous display and secretion of multiple antibody formats. The schematic cell represented at the center of the figure is able to display full-length IgG on the surface (shown on the top half of the cell) and is also able to secrete Ab (bottom half), including versions with C-terminal epitope tags (represented by the orange stars). Of the methods listed (in blue), only SHM-XEL simultaneously permits display, secretion, and dynamic evolution without the need for any further recloning or other manipulation.

The system described here is particularly useful in the context of in situ diversity generation technologies, such as in vitro somatic hypermutation. Somatic hypermutation (SHM)² is well understood as the natural mechanism of antibody maturation and has been shown to be initiated through the action of the enzyme activation-induced cytidine deaminase (AID) (5). During SHM, AID is targeted to the DNA encoding the immunoglobulin variable domains and preferentially deaminates cytidine residues at hotspot motifs, resulting in a spectrum of amino acid substitutions that is non-random both with regard to identity and to location. This process has been replicated in a non-B cell milieu by transfection of non-B cells with AID (6) and recently has been used in the generation and maturation of both fully human and humanized antibodies (4, 7). For example, an anti-NGF antibody was isolated from a fully human germ line library expressed in HEK293 cells and, through SHM, initiated by the action of AID, was subsequently matured in the same cell background to low pm affinity with potent activity in in vitro assays (4).

Cell surface display of antibodies allows direct linkage of the antibody phenotype to genotype through facile cell cloning and gene recovery. This allows binding and selection experiments to be carried out by adding soluble antigen directly to cells, with detection by fluorescent labels or other techniques. To select antibodies with ideal properties for development, however, it would also be desirable to be able to test soluble antibody in a range of assays during the selection process. For example, binding to antigens presented on cell surfaces, measurement of biophysical properties such as stability and solubility, and functional activity. Functional testing is particularly important to ensure antibodies are selected for the intended properties rather than solely for high binding affinity to the antigen in question. The use of mammalian cells for antibody selection and maturation is particularly advantageous because soluble antibody can be in the same IgG format and with identical post-translational modifications as the antibody later manufactured in the same cell type.

The modular structure of the antibody molecule is well suited to engineering with the ability to add and subtract independently folding protein domains without affecting antigen binding characteristics. Alternative splicing of pre-mRNA was first described through the observation that membrane bound and secreted antibodies are encoded through the same gene (8–10). Antigenic stimulation triggers the later stages of B cell differentiation via IgM on the surface of B cells, leading to secretion of IgM with the same antigen binding regions. We now demonstrate that a similar alternative splicing system may be used for antibody selection in non-B cells in vitro, with both cell surface displayed and secreted antibody generated from the same cell. This has been achieved with considerable flexibility in the experimental design to enable a number of novel assay formats. Simultaneous display and secretion of full-length IgG can be utilized, or alternatively either cell surface or secreted antibody may be utilized in a different format such as monovalent Fab. In addition, antibody fusion proteins may be secreted while maintaining IgG or Fab presentation at the cell surface. The ratio of cell surface displayed to secreted protein can be readily modulated by manipulation of the sequences flanking a cryptic splice donor. This allows a number of novel selection schemes to be devised which facilitate direct selection of antibodies with desirable properties. This methodology is readily applicable to in vitro evolution of a wide range of proteins.

EXPERIMENTAL PROCEDURES

Reagents

DNA was synthesized at DNA2.0 (Menlo Park, CA). Oligonucleotides were purchased from Valuegene (San Diego, CA). DNA sequencing was performed by Eton (San Diego, CA) or Genewiz (San Diego, CA). 293 c18 cells were purchased from ATCC (CRL-10852). The ABELMab library was assembled and transfected as described (4). Sequences for epitope tags were as follows: nt sequences corresponding to residues 410–419 of human c-Myc, (N-EQKLISEEDL-C, encoded by GAGCAGAAGCTGATCAGCGAGGAGGACCTGTGA), the synthetic FLAG tag (N-DYKDDDDK-C, encoded by GACTACAAAGATGACGATGAT AAAGGTTGA), or His-9 (CATCATCACCACCA TCACCATCACCATTGA) were added 8 nt distal to the SA site. HC and LC open reading frames were cloned into episomal vectors as described previously (4, 11).

Analysis of Spliced Forms of mRNA

Constructs were transfected into HEK293 c18 cells as described (4). mRNA was isolated from stably or transiently transfected cells and converted to dsDNA using SuperScript RT-PCR kits from Invitrogen as per the manufacturer's instructions. Reverse transcription step was performed using a combination of oligo(dT) and random hexamers as provided by the kit. cDNA was subsequently amplified using oligonucleotides situated at various points along the HC variable and constant regions, as shown schematically in Fig. 2. Amplicons were run on agarose gels and photographed on a FluorChem Imager (Alpha Innotech, Santa Clara, CA), and RT-PCR generated bands were quantified using AlphaEaseFC^TM analysis software. Several exposures for each gel were analyzed to ensure band intensities were within linear range of detection. A box of fixed size was drawn around each band, and the analysis software provided an estimate of band density as number of pixels. Background was determined for each lane by measuring an area, the same distance from the loading well for each lane, where there were no bands were visible. Background values were subtracted from the pixel value of each band for each lane.

FIGURE 2.

Constructs for simultaneous cell surface display and secretion. A, schematic representation of HC RNA with features as follows: IgHV, variable domain; IgHC, IgG1 constant domain; Lx1 and Lx2, 5′-proximal and -distal LoxP elements, respectively; designations SD1–SD3 indicate the relative order of major splice donors from the SA from 3′-proximal to distal; the juxtamembrane (jx), transmembrane (tm), and cytoplasmic (cy) domains of H-2K^k are located as indicated. Line width tying SD to SA indicates the approximate relative use of each as discerned in B. The relative positions of the stop codons for the secreted (S*) and transmembrane (Tm*) versions of the IgG are shown by diamond-shaped arrowheads. Templates in A are as follows: Ig1-1 schematically represents the original transmembrane-only IgG construct; Ig1-2 depicts the LoxP-flanked display-and-secreted version; and Ig1-3 represents a construct in which the 11-amino acid H2kk juxtamembrane moiety was deleted as described in the text. Arrows A through F indicate the relative locations of oligonucleotide primers used for PCR amplification. B, agarose gel showing results of RT-PCR. Black arrows labeled unspliced (U) and SD1 through SD3 identify the amplified DNA bands cloned and sequenced. Lane designations A through E correspond to which forward oligonucleotide primer was used for the PCR amplification step that gave rise to the bands visible in each lane.

Quantification of Surface Fab and IgG

The number of surface IgG and Fab molecules per cell was quantified using a phycoerythrin fluorescence quantitation kit (Becton Dickenson). Only cells that had been transfected and stably selected for a minimum of 3 weeks were analyzed. Standard curves were generated from four populations of labeled spheres conjugated with calibrated numbers of binding sites per bead, as provided by the kit. Cells were treated identically to beads by staining with phycoerythrin-labeled goat anti-human antibodies as per BD Biosciences protocol. Samples were analyzed by flow cytometry on a BD Biosciences Influx instrument, and the number of sites per cell was calculated using BD FACS software (version 1.0.0.650).

Quantification of Fab and IgG Concentrations by ELISA

NeutrAvidin-coated 96 well React-bind plates (Pierce) were coated with 500 ng/ml of biotin-(spacer)-conjugated goat anti-human IgG-F(ab′)2 fragment (Jackson ImmunoResearch Laboratories). Following blocking with 3% BSA in PBS (200 μl/well) for 2 h at room temperature, 10 μl of test antibody was added with 90 μl of PBS-1% BSA. A standard curve was created using Chrome pure human IgG F(ab′)2 fragment (Jackson ImmunoResearch Laboratories). Antibody and standards were detected using 100 μl of goat anti-IgG-κ-HRP at 1:1500 dilution in PBS-1% BSA (Sigma) followed by addition of 100 μl of Sure Blue 3,3′,5,5′-tetramethylbenzidine peroxidase substrate (KPL). Reactions were stopped after 10 min by adding 100 μl of 3,3′,5,5′-tetramethylbenzidine Stop solution (KPL) and read at 450 nm.

Screening the ABELmAb Library Using Magnetic Beads

Invitrogen Dynabeads Biotin Binder no. 110.47 were washed with buffer and then coated with either 0.25 μg of biotinylated irrelevant antigen (for pre-clearing) or 0.5 μg of antigen (for positive selection) per 10⁷ beads according the manufacturer's protocol. Eight ABELmAb libraries, each composed of a distinct HC V-region paired with κ V-region library and displayed on the surface of HEK293-c18 cells, were combined into four pools of two sublibraries each by using 1 × 10⁸ cells from each sublibrary. 1 × 10⁸ beads coated with irrelevant antigen were then added and incubated at 4 °C with gentle tilting and rotation for 30 min. The tube was placed in a magnet for 3 min, and the supernatant containing the unbound cells was transferred to a fresh tube for next round of pre-clearing for a total of four rounds.

For the positive selection, 4 × 10⁷ beads coated with ActRIIB extracellular domain were added to each of the four pools of precleared cells (∼2 × 10⁸ cells in each pool) and incubated at 4 °C with gentle tilting and rotation for 30 min. The tube was then placed in a magnet for 3 min, and the supernatant (containing non-bound cells) was discarded. The bead-bound cells were gently washed four times by adding 8 ml buffer, placing the tube in the magnet for 3 min and discarding the supernatant by gently pouring off. Finally, the bead-bound cells from all four pools were resuspended in complete medium and combined into one T75 flask. The cells were expanded to ∼100 × 10⁶ cells and transiently transfected with AID. Six days following transfection, cells were stained with 100 nm Dyl650-ActRIIB-Fc fusion protein and FITC-labeled goat anti-human κ antibody for 40 min at 4 °C, and ∼0.5% of the top fluorescent cells were isolated using FACS. The cell populations were again expanded, transfected with AID, stained with 50 nm Dyl650-ActRIIB-Fc and FITC-labeled goat anti-human κ antibody, and followed by FACS isolation of individual cells into 96-well plates. Each well containing a cell population was grown to confluency, and the antibody secreted into the media was analyzed by surface plasmon resonance (SPR).

SPR Screening of Antibody Secreted into Cellular Medium

SPR experiments were conducted on a Biacore 4000 instrument (GE Healthcare) at 25 °C. Cell culture supernatants (in DMEM medium + 5% FBS) from discovery single cell clones were collected, spun down to remove cellular debris, and diluted 1:1 with SPR running buffer (1× HBS-EP+: 10 mm HEPES, 150 mm NaCl, 3 mm EDTA, 0.05% polysorbate 20, pH 7.6). Samples were captured for 10 min at 10 μl/min on a CM5 chip previously immobilized with 12.000 resonance units anti-human IgG (Fc-specific) capture antibody (GE Healthcare) yielding capture levels between 1000–2500 resonance units. A positive and an irrelevant control antibody were spiked at 1 μg/ml into 1:1 diluted spent DMEM medium + 5% FBS and captured analogously. After a 1-min wash step with 1 m NaCl, antigen and irrelevant control antigen were injected at concentrations of 500, 100, and 0 nm for 2 min at a flow rate of 10 μl/min and the association and dissociation sensorgrams recorded. 3 m MgCl₂ followed by 10 mm glycine, pH 1.7, were injected for 120 s at 30 μl/min for chip regeneration between cycles. SPR raw data were double-reference subtracted, and sensorgrams for the individual single cell clones are displayed side by side with the respective FACS plots.

RESULTS

Simultaneous Surface Display and Secretion of IgG

Cell surface display of IgG on HEK293 cells was achieved through expression of a heavy chain fusion protein construct together with light chain. The human IgG1 heavy chain was fused to a short juxtamembrane region, the transmembrane domain and intracellular domain from the H-2K^k protein (12, 13) as shown schematically in Fig. 2A (template Ig1-1). To facilitate antibody screening, a number of alternative approaches were explored to enable the simultaneous presentation of cell surface antibody and secretion of soluble IgG. Experiments initially designed to excise a transmembrane region via DNA recombination, utilized constructs in which a LoxP site was placed on both sides of the H-2K^k domains (Fig. 2A, Ig1-2). The 5′ LoxP site is located six nucleotides downstream from the final Fc codon, preceded by a synthetic XhoI site. Cells transfected with this construct were found to secrete ∼1 μg/ml of IgG into the medium following 10 days in culture, as well as to display IgG on the cell surface, even in the absence of Cre recombinase, whereas IgG was not secreted from non-LoxP containing constructs (Ig1-1, Fig. 2) during an equivalent incubation period. To determine whether a transcriptional or post-transcriptional mechanism was responsible for this observation, RNA was isolated from transiently transfected cells and subjected to RT-PCR. Gel electrophoresis revealed two predominant, amplified bands in the construct that does not contain the LoxP sites (Ig1-1). Sequencing data showed that these bands represent unspliced RNA (Fig. 2B, band U), as well as a species in which the SV40 small t intron has been correctly removed (band SD1). Comparison of Ig1-2 (containing two LoxP sites) revealed two additional major bands (Fig. 2B.). Sequencing indicated that these represent two additional, differentially spliced products in which new splice donors (SDs) were used, with one (SD3 in Fig. 2) situated near the 3′ end of the human Fcγ1 coding sequence (CGGGTAAAC), and a second (SD2) with sequence CAGGTGGAA) in the H-2K^k cytoplasmic domain (Fig. 2). Thus, the introduction of LoxP sequences enabled alternative splicing to be initiated from two nearby GT sequences. When SD3 is used, the resulting spliced RNA is translated to yield secreted IgG, whereas translation of U, SD1, and SD2 RNA leads to retention of the transmembrane region resulting in cell surface retained IgG. A further construct, Ig1-3, was also prepared in which the H-2K^k juxtamembrane region of 11 amino acids was removed, such that the 11 amino acids encoded by LoxP would not alter the spacing between the IgHC and transmembrane domains as compared with the original Ig1-1 configuration. This construct resulted in the same RNA banding pattern as Ig1-2, and transfection of cells with this construct similarly resulted in simultaneous secretion of IgG and cell surface display. As a percent of all bands visualized, band SD3 (Fig. 2b) represents <10% of Ig1-1, 30% of Ig1-2, and ∼50% of Ig1-3, indicating some modulation of alternative splicing could be accomplished. These data suggested that alternative RNA splicing was responsible for the simultaneous secretion and cell surface presentation of the antibody, a theory that was explored in further experiments described below.

Titrating Cell Surface Antibody Expression

To test whether the same mechanism could be used for cell surface presentation and secretion of a monovalent antibody format, we tested the antibody as a Fab fragment. The human IgG1 constant region sequence was truncated to include the CH1 domain and the first seven amino acids of the hinge region that is required for assembly of the light chain and Fab heavy chain via a disulfide bond. Sequences downstream from the Fab heavy chain were kept identical to those used in the IgG construct Ig1-2, except that SD2 was changed from GT to GA to render it inactive. In the same context as Ig1-2, Fab1-2 (Table 1A) was well expressed on the cell surface, displaying nearly 400,000 sites per cell (Fig. 3), and Fab protein accumulated to 1.6 μg/ml in the culture medium by day 5 (Table 1). By contrast, secreted Fab was not detectable above background levels by ELISA (<0.01 μg/ml) from cells transfected with a version of Fab1-2 that does not contain LoxP sites and can produce only the transmembrane form. Sequence analysis of RT-PCR generated fragments from Fab1-2 indicated that the two most proximal GTs to the LoxP site were not used, but rather the vicinal LoxP sequence in this vector caused the sequence AAGGTGGAC to become an alternatively recognized SD in the Fab constructs versus CGGGTAAAC, which was used in the IgG constructs (see Table 1). In the case of the Fab, however, the use of this new SD results in the removal of six amino acids from the C-terminal end of CH1 plus all seven amino acids from the hinge region. Therefore, the cryptic SD sequence recognized in the IgG construct Ig1-2 was inserted between the 3′ end of the Fab coding region and first LoxP site, and the second, downstream LoxP sequence was removed to make vector Fab1-3. This strategy was successful in restoring the authentic Fab constant region sequence although expression of Fab on the cell surface was decreased by ∼2-fold (Table 1). Fab resulting from this construct accumulated to 1.7 μg/ml in culture medium by day 5. Thus, these results indicate that the alternative splicing mechanism allows the display and secretion of Fab as it had for IgG and, furthermore, that the presence of a single LoxP adjacent to the 3′ end of the IgHC domain sequence is sufficient to enable a functional system.

TABLE 1.

Modulation of cell surface antibody density

Fab1 and Fab2 designations represent two different antibodies with unrelated variable regions. “Intact” in the LoxP column means the native LoxP sequence was used (ATAACTTCGTATAGCATACATTATACGAAGTTAT); Δ indicates that the site was deleted. LoxP variety v1* signifies that the sequence has been changed to ATGAAGCTTGGATATACTAAGCTTCATA, which alters the stem-loop structure found in the original LoxP arrangement from 13-8-13 to 10-9-10. Variety v2* retains the original 13-8-13 stem-loop-stem sequence lengths, but the stem sequence has been disrupted by mismatches (ATCACCTCATACAGTATACACTATACGAAGTTAT). The SD3 sequence used for each Fab is presented in the table. SD2 was either active (+) or inactivated (−), i.e. by mutating the canonical donor GT dinucleotide sequence to GA. NQ indicates not quantified; NT represents not tested; and NA signifies the element was not part of the construct. Unstained control values were subtracted from the total molecules of equivalent soluble fluorochrome (MESF) for each sample to yield sites per cell values. A and B represent the results of two independent experiments. C provides information for the remaining IgG and Fab entities discussed in this article.

Sample	Sites/cell	1st LoxP	2nd LoxP	SD3 seq	SD2	ELISA (μg/ml)
A
Fab1-3	190,000	Intact	Δ	CGGGTAAAC	−	1.7
Fab1-4	30,000	Intact	Δ	CAGGTAAAC	−	1.9
Fab1-5	170,000	Intact	Δ	CAGGTACCA	−	2.2
Fab1-6	5,500	Δ	Δ	CAGGTAAAT	−	2.0
Fab1-7	16,000	v2*	Δ	CAGGTAAAT	−	1.9
Fab1-8	11,000	Intact	Δ	CAGGTAAAT	−	1.7
Fab1-2	397,000	Intact	Intact	AAGGTGGAC	+	1.6
Fab1-s	1,900	None	None	NA	NA	>1
B
Fab1-2	770,000	Intact	Intact	AAGGTGGAC	+	1.1
Fab2-2	267,000	Intact	Intact	AAGGTGGAC	+	NT
Fab1-8	29,000	Intact	Δ	CAGGTAAAT	−	1.4
Fab2-8	47,000	Intact	Δ	CAGGTAAAT	−	NT
Fab1-9	15,000	Intact	Intact	CAGGTAAAT	−	1.5
Fab2-9	22,000	Intact	Intact	CAGGTAAAT	−	NT
Fab1-10	5,200	v1*	Δ	CAGGTAAAT	−	1.5
Fab2-10	4,300	v1*	Δ	CAGGTAAAT	−	NT
Fab1-11	2,900	v1*	Intact	CAGGTAAAT	−	1.2
Fab2-11	4,700	v1*	Intact	CAGGTAAAT	−	NT
Fab1-s	1900	NA	NA	NA	NA	>1
C
Ig1-1	NT	0	0	CGGGTAAAC	+	<0.01
Ig1-2	NT	Intact	Intact	CGGGTAAAC	+	∼1
Ig2-6a	NA	Intact	Intact	CAGGTAAAC	−	NT

^a Values describe samples Ig2-4 and Ig2-5.

FIGURE 3.

Modulation of cell surface display to control antibody avidity. The density of Fab presented on the surface of each cell population (A–G) was varied either by altering the nucleotide sequences adjacent to the alternatively recognized SD3 or by modifying the primary sequence within the nearby LoxP site as described in the text. Cells were analyzed by flow cytometry and the number of sites per cell for each population was evaluated using the BD Quantibrite system (BD BioSystems) (H).

Because avidity at the cell surface can affect binding kinetics, attempts to manipulate the ratio of cell surface to secreted Fab were undertaken. Huang and Gorman (14) reported that transfection of constructs containing the SV40 small t antigen intron into HEK293 and CV1 cells can result in aberrant splicing. This is due to the occasional unmasking of some upstream GTs as new, alternative SDs, with a consensus sequence of MAGGTRAGT (M = C or A; r = A or G). Therefore, the sequence of SD3 was altered to reflect either more closely or less closely the reported consensus sequence. This was tested with native as well as scrambled versions of the LoxP site in which the stem-loop structure and sequence were altered (stem shortened from 13 to 10 bp, loop lengthened from 8 to 9 nt) to make Fab1-10 and Fab1-11 (Table 1B, LoxP1 v1*), or in which the stem structure has been disrupted to generate Fab1-7 (Table 1A, LoxP v2*). In addition, a second IgG1 antibody that uses an unrelated variable region (the Fab2 series in Table 1B) was tested to determine whether there were V-region specific sequence effects. Upon expression, all constructs explored were able to generate the cell surface and secreted Fabs encoded by the alternatively spliced products. Secreted Fab accumulated by day 5 in culture to between 1 and 2 μg/ml as measured by ELISA for all LoxP-containing constructs. Estimates of cell surface retained Fab copy number, revealed that cell surface density could be varied from ∼2,900–770,000 copies per cell (Fig. 3), depending on how closely the SD3 sequence resembled the consensus sequence (14). Cells transfected with vectors that encoded more consensus-like SD3s displayed less surface Fab, whereas those whose sequences less related to the consensus splice site maintained higher surface densities. In the latter case, the presence of at least one nearby LoxP site was clearly required for alternative splicing because other similar vectors without LoxP sites did not produce detectable secreted antibody and demonstrated only very low quantities of the alternatively spliced RT-PCR fragments. Altering the sequence of the proximal LoxP site appeared to further lower the density of surface Fab by a factor of 5 to 10, although it did not abolish the recognition of the SD3 site (Table 1B, compare e.g. Fab1-8 with Fab1-10, and Fab2-8 with Fab2-10). RT-PCR analysis indicates that the ratio of secreted-to-surface retained Fab varies (from high to low) as follows: Fab1-8 > Fab1-6 ≈ Fab1-7 ≈ Fab1-4 > Fab1-5 > Fab1-3 (Fig. 4). The expression data support the hypothesis that SDs with sequence close to the consensus (matching 8 of 9 nt) tend to be used efficiently, whereas those with sequence more distant from the consensus (matching six or seven of nine) are used less effectively.

FIGURE 4.

Influence of sequence on efficiency of splice donor recognition. Fabs 1-2 through 1-8 differ from each other in SD3 and/or the presence or absence of native LoxP1 (Lx1) or variant (Lxv1) (further description provided in Table 1). A, lanes M1, 1-kb marker ladder; M2, 100-bp marker ladder. Gel lanes 1–7 display RT-PCR results. B, schematic representation of bands 1 through 5. Band 1, 1395-bp fragment derived from unspliced mRNA; band 2, 1329-bp fragment wherein the intron bounded by SD1 and SA has been spliced out; band 3, 931-bp fragment representing removal of intron bounded by SD3 (described in Fig. 2) and SA; band 4, 1066-bp band with SD2 to SA intron removed. Translation of RNA encoded by bands 1, 2, and 4 will generate transmembrane forms of the Fab, whereas translation of band 3 will result in secreted Fab. Band 5 represents a truncated product in which a GT within CDR3 was recognized as a cryptic SD in the Fab1-2 antibody sequence. The protein product from band 5 falls out of reading frame in CDR3 and presumably does not give rise to a functional antibody.

The experiments discussed above retained the 8 nt downstream from the SA to not disrupt its function or robust activity. Further experimentation found that these 8 nt did not significantly influence the efficiency of SA recognition and that native amino acid sequences of alternatively spliced proteins can be maintained by appropriately splitting a single codon across the facultative intron, thus eliminating the presence of unnecessary or exogenous amino acid residues.

For antibodies with weaker binding affinities, high surface densities may be useful to exploit avidity for the identification of target-specific antibodies. Thus, SD3 sequences that are less closely related to the consensus appear to provide the properties of relatively high surface density coupled with high levels of secretion. Indeed, by slightly modifying cell culture conditions, the production of IgG and Fab from vectors encoding these less conforming SD3 sequences routinely provides yields in the range of 5 to 50 μg/ml following 5 to 7 days in culture while still retaining the high levels of surface display desired for robust flow cytometry signals.

Secretion of Antibody Fusion Proteins

To further explore the versatility of the system, we investigated whether secreted antibody could be produced as a fusion protein. Toward this end, the coding sequence for each of three different epitope tags, His-9, Myc, or FLAG, were added in-frame, 8 nt downstream from the SA to generate three new, in-frame fusion constructs (Fig. 5A). After transfection, secreted antibodies were purified and analyzed by Western blot. Blots probed with antibodies specific for the epitope tags revealed that each was expressed and secreted as expected (Fig. 5B). The visualized, anti-epitope-reactive bands migrate at ∼53 kDa under reducing conditions, consistent with intact HC being linked to the epitope. All secreted products also reacted with anti-Fc as expected. RT-PCR analysis confirmed that mRNA for both secreted and transmembrane forms of the IgG are transcribed from these constructs (data not shown).

FIGURE 5.

Secreted version of IgGs can be C-terminally labeled with a variety of epitope tags. A, schematic representation of the epitope-tagged deciduous HC, with His, FLAG, or Myc epitope indicated as the “tag” 3′ to the SA. Other landmarks are as described in the legend to Fig. 2. B, Western blot analysis of epitope-tagged IgGs. Nitrocellulose membrane strips were labeled using anti-Fc, anti-His, anti-FLAG, or anti-Myc antibodies as indicated. The His-tagged antibody was also used for probing with anti-Fc to demonstrate that the tags were directly linked to the heavy chain. The arrow indicates the expected size of the intact, epitope-tagged HC.

Analysis of Cell Surface and Secreted Ab Binding in ABELmAb Library Discovery

To further illustrate the utility of a system competent for the simultaneous cell surface display and secretion of antibodies, a library of full-length human antibodies containing germ line V-segments fused with rearranged (D)J segments isolated from human donors was assembled as described previously (4). The library, termed ABELmAb, was assembled in the vector Ig1-2 and subsequently screened for binding against a human antigen, activin receptor IIB (ActRIIB). The cell surface receptor ActRIIB binds a number of cytokines (e.g. myostatin, activin A) that negatively regulate muscle differentiation and growth. Inhibition of receptor signaling has been shown to reverse muscle wasting and increase muscle mass and strength in both animals and humans (15, 16). The soluble extracellular domain of ActRIIB, which binds the cytokines myostatin and activin A, was therefore selected as the antigen bait for the library screen.

The ABELmAb library was transfected into HEK293 cells, expanded to 10⁹ cells, and subjected to four rounds of negative selection against streptavidin (SA)-coupled magnetic beads, followed by a single round of positive selection against SA-coupled magnetic beads coated with biotinylated ActRIIB. Positively selected cells were expanded and transiently transfected with AID to initiate in vitro SHM as described (4). To isolate cells expressing antibodies with improving affinity for ActRIIB, two rounds of FACS selection were performed using 100 nm and subsequently 50 nm fluorescently tagged ActRIIB-Fc. Single cells were isolated into five 96-well plates following the second round of FACS selection and grown to confluency. The cell medium from each well of the five 96-well plates (containing secreted antibody) was characterized for binding to soluble human ActRIIB-Fc and to an equivalently tagged irrelevant antigen by SPR. In parallel, cells from each well were screened by FACS for binding to soluble, fluorescently tagged ActRIIB-Fc.

Fig. 6 shows the resulting sensorgrams and FACS scatterplots for a few representative examples of the identified clones. Cells in the depicted wells contained populations expressing a unique HC (human germ line IGHV3-23) and LC (human germ line IGKV4-1) pair (parental and muteins 50 and 10%) that bound soluble ActRIIB but did not recognize an equivalently tagged, irrelevant antigen (dotted line). FACS and Biacore analyses exhibited a wide variety of binding and kinetic properties, respectively, reflecting the incorporation and enrichment of AID-mediated SHM within each population. Ten to 50% of the sequences analyzed from the cell populations in well no. 2 (10% mutein) and no. 21 (50% mutein) contained an A50V mutation within the CDR2 of the HC, and 10–20% of the sequences taken from cells in well no. 21 also contained an L46V mutation in the LC. The phenotypic diversity of these populations can also be observed in the FACS dot plots, which reflect the mixture of antibodies with AID-mediated mutations exhibiting differential impact on binding kinetics (dotted circles in the mutein panels). In contrast, well no. 5 contained the unmutated germ line antibody HC and LC variable region sequences, with correspondingly more modest SPR and FACS profiles (parental Ab). The cells in well no. 3 (irrelevant Ab) expressed an antibody that did not bind the ActRIIB antigen. Templates encoding the HC A50V and LC L46V mutations were recloned into fresh expression vectors, and purified antibodies were further characterized by solution phase methods (KinExA) using monomeric ActRIIB under non-avid conditions. Results indicated that the affinity of the starting germ line (parental) antibody for ActRIIB was approximately in the single digit micromolar range, and the HC-A50V/LC-L46V double mutein had improved the affinity to a K_D of 160 nm. These data highlight the close correspondence between cell surface and soluble antibody binding characteristics, and the sensitivity of these assays for the detection of subtle differences in binding kinetics that can quickly link genotype with improved phenotype. The display and secretion methodology thus provides the ability to quickly identify and characterize cells that express antibodies with improving affinities.

FIGURE 6.

Correlating cell surface binding with soluble antibody properties. The antigen binding characteristics of cell surface displayed and secreted antibody were determined by FACS binding analysis and by SPR, respectively, for clones isolated during antibody discovery from the ABELmAb library. SPR binding properties are shown for positive and negative control Abs, an antibody isolate that does not bind specifically to ActRIIB (irrelevant Ab; Irr. Ab), a germ line (parental) antibody from the ABELmAb library, and versions of the parental antibody in which one or more mutations have begun to accumulate (10 and 50% mutein). The parental and mutein isolates bind specifically to ActRIIB and not to an irrelevant, equivalently tagged antigen at the same concentration. Panels marked 50% mutein depict an isolate in which 50% of the HC sequences contained an A50V mutation in CDR2, and 10% of the LC templates encoded an L46V change in framework 2. Panels labeled 10% mutein contained 10% A50V mutation and no LC mutations. Secreted antibodies were screened directly from tissue culture medium without further purification by SPR using avid ActRIIB-Fc antigen at 500 nm (black line) and 100 nm (gray line) or 500 nm irrelevant Fc-tagged antigen negative control (dotted line). The corresponding FACS analysis for each population was performed by binding to 50 nm ActRIIB-Fc Dyl650 labeled antigen (x axis) relative to staining cell surface expressed IgG surface (y axis). Because the antigen is an Fc fusion protein and the displayed IgGs are bivalent, the interactions presented in this figure were measured under conditions of avidity. Differences observed between the binding properties of the parental versus mutein populations reflect the incorporation and enrichment of AID-mediated mutations that improve the binding kinetics of the germ line antibody sequence for ActRIIB. The dotted circles in the mutein panels delineate cell populations expressing antibodies with AID-derived mutations that improve binding affinities to antigen.

DISCUSSION

Antibody discovery technologies can be described in two major categories: first, harvesting of an in vivo immune response, for example by immunization of transgenic or wild-type animals paired with hybridoma generation and second, in vitro library screening technologies utilizing phage, bacterial, yeast or mammalian display of antibody repertoires. A significant advantage of hybridoma technology is the ability to immediately characterize secreted full-length IgG for function, specificity, and other properties without the need for reformatting, expression, and purification of antibodies that usually accompany library screening methodologies. We sought to create a display technology encompassing the advantages of in vitro library screening with the flexibility of a mammalian cell expression approach that supports titratable surface display and antibody secretion in functional form. This method supports hybridoma-like capabilities plus affinity maturation in the same cell using in vitro SHM directed by AID and also permits the sampling of soluble antibody during maturation to ensure desired properties are selected for and maintained.

This display and secretion system utilizes alternate splicing of pre-mRNA, a process first described during investigation of the immune system. Pre-B cells each display unique rearranged V(D)J sequences via cell surface presentation of IgM. Antigen binding and stimulation of individual pre-B cells can lead to differential splicing of pre-mRNA to remove the transmembrane region, leading to the secretion of the soluble, secreted form of IgM. The pentameric IgM molecule, capable of high avidity binding, facilitates antigen recognition by naïve rearranged antibodies during the early stages of an immune response. In addition, AID-mediated class switch recombination and SHM for affinity maturation are initiated, resulting in the generation of high affinity antibodies capable of a range of different effector functions. Although it has been estimated that ∼95% of multiexon human genes undergo some form of alternative splicing (17, 18), the complex rules for how efficiently and in which tissues the splicing occurs are only just now beginning to be understood (19).

By exploiting a methodology based on alternative splicing, we have developed an in vitro system that parallels many of the attributes associated with the presentation, maturation, and function of antibodies during adaptive immunity. Control of antibody avidity that can be displayed and secreted enables a flexible system in which cell surface levels may be titrated and novel fusion proteins secreted for ease of assay development. Furthermore, the ability to attach epitope tags to the secreted IgG scaffold offers the potential to devise novel selection schemes. It will be interesting to explore the feasibility of fusing additional unconventional payloads using this system such as toxins to generate antibody-drug conjugates or reporters such as a fluorescent protein or luciferase.

Simultaneous display and secretion has the potential to be used with a wide variety of proteins to facilitate directed evolution with selection for diverse properties. Proteins may be evolved using random mutagenesis techniques or via other approaches. For example, SHM has been used to evolve a number of different proteins with selectable phenotypes, including fluorescent proteins and anti-apoptotic proteins (20, 21), and yeast display coupled with synthetic libraries targeting the active site of a protein has been used to identify an enantioselective enzyme (22). The integration of in situ evolution with parallel mammalian surface display and secretion will enable the rapid evolution of a wider range of protein types through novel selectable properties and assay design.