An NIH funded project was launched to rapidly develop and implement a high-throughput pipeline designed to generate high quality renewable recombinant antibodies to human transcription factors (U54 HG006436). TF antigens reported here were generated by Rutgers University, epigenetic antigens generated by the Structural Genomics Consortium (SCG), and in-house at the RAN facilities.

E. coli strain XL1-blue (Stratagene, Santa Clara, CA) or T1 phage resistant cells were used for phage propagation, whereas strains DH10B (Invitrogen, Grand Island, NY), or other T1 phage resistant cells were used for Fab expression C43 (DE3) Pro+ (16) or BL21 (DE3) were used. XL1-blue or T1 phage resistant cells were grown in the presence of 5 μg/ml tetracycline (Tet) to ensure expression of the F‘ pilus that allowed for phage infection and DH10B cells were grown with 50 μg/ml carbenicillin (Carb) for production of plasmid DNA. Fab expression cells, C43 (DE3) Pro+, were grown with 50 μg/ml Carb and 25 μg/ml kanamycin (Kan). Where biotinylation was required (Antigen or Fab) recombinant BirA co-expressed or added during purification (Avidity, LLC., Aurora, CO).

The genes for transcription factor domains were identified by bioinformatics (17–19) and synthesized (Life Technologies, Carlsbad, CA), and were provided as E. coli expression constructs. In order to generate constructs for antigen expression by in vitro transcription and translation (IVTT), and E. coli, gene fragments were PCR amplified from transfer plasmids and recombined into their specific expression vectors by SLICE or Ligase Independent Cloning techniques (20, 21). All expression plasmids were sequence verified prior to expression studies.

The antigen production group at Rutgers University or at the SGC Consortium provided purified human biotinylated human transcription factor domains or chromatin remodeling antigens as N-terminal tagged AviTag fusions. For Rutgers antigens, expression constructs were synthesized (GenScript Inc., Piscataway, NJ), cloned into an expression vector containing an N-terminal AviTag, HisTag, and TEV cleavage site, and expressed in E. coli Tuner (DE3) harboring a pLysSRARE2 helper plasmid (EMD Millipore, Billerica, MA). Cells were harvested, lysed by sonication, and clarified lysate was loaded onto HisTrap HP (GE Healthcare, Piscataway, NJ) columns for purification. If the protein of interest was predominantly found in inclusion bodies, pellets were lysed in solubilization buffer (binding buffer containing 6 m urea), bound to the HisTrap column, and subjected to a slow refolding gradient into binding buffer without urea prior to elution. Pooled elution fractions were in vitro biotinylated (Avidity LLC), concentrated, and applied to a HiLoad 16/600 Superdex 75pg (GE Healthcare, Piscataway, NJ) column to remove unreacted biotin and ATP. Fractions corresponding to the monomeric peak were pooled and concentrated before being snap frozen in liquid N₂ and stored at −80 °C until shipment (Campbell, E. Anderson, S. in preparation). SGC produced antigens were expressed in E. coli, purified by Immobilized Metal Affinity Chromatography, and tested for biotinylation prior to entry into the Fab selection pipeline (22, 23).

Protein expression system comparisons consisted of base-line TF expression in E. coli as intracellular GST fusion constructs (His-GST-Avi-TEV-Antigen) driven from the pTac promoter (supplemental Fig. S1C). E. coli based expression of soluble GST-TF fusion antigens was conducted by standard protein expression methods. Briefly E. coli cultures were grown in Terrific Broth with antibiotics to an OD₆₀₀ of 0.6–0.8 with 1 mm IPTG for 3–6 h at 37°C, then harvested by centrifugation. Cells were resuspended, lysed by sonication, and clarified prior to purification with GST Sepharose (24). As a second system we expressed the TFs as His-Avi-TEV N-terminal fusion driven from a T7 promoter using an E. coli in vitro transcription-translation system (IVTT) that was kindly provided by Sutro Biosciences (25) (supplemental Fig. S1D). The cell-free reaction mixture contained an ATP regeneration system supplemented with chaperones and redox enzymes. Briefly, 10 μg of IVTT plasmid constructs were combined with 1 ml cell free extract containing recombinant BirA protein and 50 μm biotin and allowed to shake at 650rpm for 2 h in an Eppendorf Thermomixer Expressed and biotinylated protein was purified over NiNTA agarose (Qiagen, Valencia, CA) with PBS based buffer system and eluted with PBS + 400 mm Imidazole. Protein samples were buffer exchanged with Amicon spin concentrators (EMD Millipore), concentrations determined by Bradford BSA assay (Thermo Fisher, Grand Island, NY), and analyzed for purity by SDS-PAGE prior to entering the selection pipeline (25, 26). In the third system, we displayed the antigen for selection on the surface of yeast (27, 28). To generate constructs for YAD, gene fragments were PCR amplified from gene synthesis transfer plasmids and then recombined into YAD plasmid. YAD constructs utilized a N-terminal Avi-6xHis-Aga2-TEV protease fusion partner with a C-terminal V5-His tag. The YAD strain, EBY100, was engineered to overexpress agglutinin AGA1 under the galactose inducible Gal1–10 promoter, which formed a disulfide bonds with the antigen-Aga2 protein in the extracellular space. Constructs were biotinylated during secretion by a cytoplasmic or ER localized BirA, which allowed for efficient capture to streptavidin magnetic beads (Promega, Madison, WI) (29–31). The TF domain was expressed with a C-terminal V5 epitope tag to monitor expression level (supplemental Fig. S1E and S4). The proteins generated from these three systems were then tested for the ability to generate antibodies by phage display.

All phage selections were done according to previously established protocols (24, 26) with several modifications as outlined below. Briefly, Fab-phage selections were automated allowing for multi-parallel processing of biotinylated target antigens either generated in-house or obtained from collaborators. Up to four rounds of soluble phage selections were conducted with biotinylated target antigens bound to streptavidin magnetic beads (Promega). The antigen concentration on the bead was systematically decreased with successive selection round as follows: 100 nm Round 1, 50 nm Round 2, 10 nm Rounds 3 and 4. To maximize throughput and reduce the amount of phage library used, up to 8 antigen-streptavidin beads complexes were combined into one selection well for round 1. Antigen-streptavidin beads were incubated with 1 × 10¹³ Fab-phage particles from either Library E or F for one hour with gentle mixing on a King Fisher Flex magnetic bead separator (Thermo Scientific) followed by three washes in phosphate buffered saline supplemented with 0.05% Tween-20 and 0.2% bovine serum albumin (PBST+BSA). To reduce the deleterious effects of nonspecific binding phage, we employed a “catch and release” strategy, where specific antigen binding Fab-phage were selectively eluted from the magnetic beads by the addition of 2 μg/well TEV Protease for 10 min. Once liberated from the streptavidin beads, Fab-phage were introduced to 5 ml (Selection Round 1) or 100 μl (Selection Rounds 2–4) of exponentially growing E. coli XL1-Blue or T1 phage resistant cells and propagated overnight at 37 °C with shaking.

Once propagated, Fab-phage were recovered from culture medium with Protein A magnetic beads (EMD Millipore) on the King Fisher Flex. Briefly, 20–50 μl of a Protein A magnetic bead slurry was incubated with up to 1 ml culture supernatant for 60 min then beads were collected and washed prior to elution of Fab-phage with 100 μm Acetic acid for 10 min. Tris-base pH 11 buffer was added to neutralize the purified Fab-phage prior to further processing. The initial Fab-phage library and amplified eluents can contain a small subpopulation of streptavidin binding Fab-phage, therefore prior to each selection round Fab-phage pools were incubated with 50 μl streptavidin-beads for 10 min then the beads were removed with the King Fisher Flex to deplete the library of any nonspecific binding Fab-phage. Plate based selections were done by immobilization of antigen on protein binding plates as described previously (24).

Purified phage from automated selections (rounds 3 or 4) were used to infect 100ul E. coli XL1-blue cells grown to log phase (OD600 0.6–0.8) for 20 minutes prior to plating on LB +50ug/ml Carb Omnitray (Nunc Thermo Fisher, Grand Island, NY) such that at least 200 individual colonies were grown per plate. Single colonies were then picked by either a K6–2 Colony Picker (KBiorsystems, San Diego, CA) or manually into 96-deep well blocks containing 0.5ml 2xYT +50ug/ml Carb and 1x10⁹ pfu/ml KO7 helper phage (New England Biolabs, Ipswich, MA). Liquid cultures were grown for 16–18 hours at 37°C with shaking at 600–900 RPM prior to further processing.

Two 96-well immunoassay plates (Corning, Corning, NY) coated with 50ng/well neutravidin (Thermo Fisher) were prepared for the ELISA validation of Fab-phage binders to one target antigen. A biotinylated target antigen was diluted to 20 nm in PBST + BSA. 50 μl/well were dispensed in two neutravidin plates, then incubated for 20 min. Direct Binding: Fab-phage containing culture supernatant was added to one antigen coated plate and incubated for 15 min. Competition ELISA: In a separate plate 20 nM antigen was mixed with phage containing culture supernatant and incubated for 15 min with shaking. After incubation, the phage-antigen complex was added to the Competition plate and incubated for 15 min. Both plates were then washed 3 times with PBST a BioTek EL406 or similar plate washer to remove unbound Fab-phage (or Fab-phage-antigen) followed by the addition of 50 μl anti-M13 phage-HRP antibody (GE Healthcare Piscataway, NJ; diluted 1:5000 in PBST + BSA) and incubated for 30 minutes. Plates were again washed 3 times with PBST or PBS then developed with 50 μl TMB substrate (KPL Labs, Gaithersburg, MD) and analyzed with a Tecan M1000 or similar plate reader monitoring OD650 for 5 min. Alternatively, signal development was stopped by the addition of 1M Phosphoric acid and OD450 signals were determined. ELISA data analysis was conducted by first plotting the OD650/min (or OD450). The Direct Binding signal was plotted on the Y-axis and the ratio of Direct/Competition signals that was plotted on the X-axis. In order to be considered a passing Fab-phage a competition ratio of <0.5 and a Direct Binding signal >0.005 units was required. Passing Fab-phage were then subjected to DNA sequence analysis to determine uniqueness of the Fab CDR sequences.

Analysis of the sequencing results was automated by the generation of several scripts where the sequences flanking CDR's were recognized and both DNA and amino acid translations were returned for the 4 CDR's of interest (LC3, HC1, HC2, and HC3). Duplicate Fab-phage sequences were removed and only unique Fab-phage clones passed into the cloning pipeline.

C43 (DE3) Pro+ E. coli containing expression plasmids were grown in Terrific Broth supplemented with 0.5% glycerol, Carb, Kan, Chlor, and 5 μm Biotin to an OD₆₀₀ of 0.6–0.8 at 37 °C then Fab expression was induced by the addition of 1 mm IPTG. At the time of induction incubation temperature was reduced to 30°C and allowed to shake for 16–18 h at which time cells were harvested by centrifugation. Recombinant Fabs were purified by Protein A chromatography and buffer exchanged into PBS buffer for subsequent storage and validation assays.

To assess the stability of the recombinant Fabs, we employed differential scanning fluorimetry (DSF) in which Sypro Orange (Invitrogen) binds to hydrophobic regions of partially or fully denatured proteins giving a high fluorescent signal (32). DSF was conducted either on Roche LC480 Lightcycler or similar qRT-PCR instrument in either 96- or 384-well formats. Briefly, purified recombinant Fab was diluted to 2 μm in DSF buffer containing Sypro Orange 4x and PBS then subjected to a temperature gradient (0.5°C/30 s) from 50 to 95°C. Data were continuously acquired at an ∼490 nm and ∼575 nm (excitation and emission wavelengths) then processed to generate first derivative curves where the curve peak corresponds to the melting temperature of the Fab.

To assess Fab specificity, we utilized an immunoprecipitation method where target Fabs were first bound to streptavidin magnetic beads (Dynabeads M-280, LifeTechnologies, Grand Island, NY). Fab loaded beads were then incubated for 30 minutes with 50 nm biotinylated target antigen in the presence or absence of a HEK293 cellular extract (OD280 2.5 and OD260 3.68). The samples were washed and labeled with a Streptavidin-Fluorophore prior to analysis by flow cytometric methods described elsewhere (Koide et al. in preparation).

To assess the affinity or specificity of Fab's we utilized EC₅₀ or single point ELISA assays conducted in 96- or 384-well ELISA plates pre-coated with neutravidin. Briefly, 20 nm biotinylated antigen was bound to ELISA plate and incubated for 20 min prior to washing with PBST. Purified Fab protein was either diluted in a two- or threefold dilution series (EC₅₀) or at 20 nm (specificity assay) then transferred to prepared plates for 30 min prior to washing and development with anti-Flag-HRP antibody (1:5000 dilution; Sigma Aldrich, St Louis, MO) and TMB substrate.

Ninety-six-well cell carrier microtiter plates (Perkin Elmer, Waltham, MA) were seeded with 5 × 10³ A2780Cis, 3345A, A2780, HPAFII, HEPG2 or 293T cells in individual wells after which they were allowed to adhere by incubating overnight at 37 °C. After incubation, media was removed and cells washed 1× with ice-cold PBS pH 7.4, then fixed by incubating with ice-cold methanol at −20 °C. Fixed cells were then permeabilized with 0.25% Triton X-100 in PBS, washed 3X with 200 ml PBS pH 7.4 and blocked with 1% goat serum in PBS for 1h at RT, before the addition of a 10 μg/ml solution of primary Fab diluted in 1% goat serum in PBS. Cells were incubated with primary Fab for one hour at RT then washed 3x with PBS prior to the addition of and 30 min incubation with a 1:2000 dilution of FITC-labeled anti- human Fab secondary antibody protected from light.

Immuno-stained images were collected for Fab clones against 6 cell lines with an Opera High Content Screening System (Perkin Elmer) at identical incident light intensity and photomultiplier sensitivity and gain. Analysis was performed by visual inspection to determine localization and recorded using a limited number of commonly used GO terms to facilitate analysis and prioritization of downstream testing. Scanning multiple fields of view to determine nuclear or cytoplasmic staining localization was done in order to generate a nuclear/cytoplasmic ratio that could be visualized as a heat map.

The day before transfection, A2780cis cells (Sigma-Aldrich) were seeded at ∼5–7 × 10⁴ cells per well of a 96-well plate (Perkin Elmer) in 100 μl of RPMI with 10% FBS without antibiotics. For each transfection well, Silencer Select Pre-designed siRNA (Ambion Life Technologies, Grand Island, NY) was diluted into 25 μl Opti-MEM reduced serum medium for final concentration of 10 nm per well. In a separate tube, 0.5 μl of lipofectamine RNAiMAX transfection reagent was added per well and diluted into 25 μl Opti-MEM reduced serum medium and mix. After a 5-min incubation the tubes were combined then incubated at room temperature for 30 min to allow siRNA-liposome complexes to form. Once complexed, the mix was added to cells (growth media removed), and then diluted with growth media to 100 μl final volume. Cells are then incubated with siRNA for 18 h at 37 °C in a CO₂ incubator, at which time the mixture was removed and replaced with complete growth medium. Cells were screened for Fab binding after 48 h of siRNA addition.

Production of soluble intact protein is important for successful selection of antibodies to native proteins (7, 9). Unstable or mis-folded target antigens typically result in poor binding reagents from phage display. For example, in small test sets we observed that proteins that are significantly proteolyzed or expressed in low levels typically did not produce antibodies of high quality whereas intact and soluble proteins do generate high quality antibodies (supplemental Fig. S2). Therefore, to avoid pitfalls with potentially disordered (12) multi-domain (13) TF antigens, we focused on those regions of TFs (10 to 25 kDa) that had clear domain architecture based on structural and bioinformatic considerations (17–19) and thus a good chance of expressing a soluble protein (supplemental Table S1). Virtually all TF antigen domains were expressed in E. coli as fusion proteins to facilitate purification (His tag), antigen capture via a single biotinylated site (Avi tag), and tobacco etch viral (TEV) mediated protease cleavage to elute bound phage (supplemental Fig. S1A to S1C (9, 22, 23, 26). Purity and integrity of the TF to proteolysis was assessed by SDS-PAGE prior to entry into the antibody selection pipeline. Size exclusion chromatography (SEC) and thermostability measurements were occasionally conducted as a check of antigen quality; however these metrics were not usually integrated into the antigen expression pipeline because of the large number of antigens and the relatively low throughput nature of standard SEC.

We chose a Fab phage display format where the heavy chain constant domain-1 and variable domain is fused to the gene III and the light chain is co-expressed using a phagemid system (33, 34). The Fabs assemble in the periplasm of E. coli through disulfide bond formation as a fusion protein with gene III that are subsequently packaged and displayed in bivalent format on the phage coat (35). The Fab scaffold is derived from the 4D5 anti-Her2 antibody, a very stable IgG1 (T_m 80 °C) for which several approved antibody drugs are based (36–38). The Fab library is a synthetic and codon-restricted library that was designed to focus on the most important amino acids in four of the six complementary determining (CDR) loops (15). The CDRs were mutated in a combinatorial fashion based upon structural bioinformatic analysis of antibody–antigen structures in the PDB as previously described (15, 39, 40). This diverse library contains ∼3 × 10¹⁰ members and has been used successfully in manual selections for scores of target antigens with varying folds (41), altered conformations (26, 42, 43), and nucleic acids (44).

To industrialize the process we designed a pipeline approach augmented by custom assembled robotics to process antigens (Fig. 1A and 1B). This pipeline can accommodate various antigen formats through phage selection and primary validation of Fab affinity, sequence, and E. coli Fab expression. We typically used a “catch and release” proteolysis process for both soluble antigens and yeast cell-displayed antigen. To begin, biotinylated antigens are preferentially bound to streptavidin coated magnetic beads and excess antigen is washed away using a King Fisher Flex magnetic bead separator (Fig. 1B) (26, 39). The bead-based immobilization methods using streptavidin capture preserves the native fold of the antigen and reduces the possibility of localized antigen unfolding that sometimes occurs when antigens are adsorbed directly to plastic immunoassay plates (45). In some cases direct binding of antigen onto microtiter plates was conducted for selections, but this precludes the use of the King Fisher Flex device.

The Fab-phage library was mixed with the antigen-loaded magnetic beads and unbound phage was washed away (Fig. 1C, Steps 1–3). Beads were treated with TEV protease and eluted phage were recovered and propagated in E. coli for subsequent selection rounds (Fig. 1C, Steps 4–7). Unlike standard elution by low pH glycine elution from bare plastic plates (46), the proteolytic catch and release format ensures clean and efficient recovery of antigen bound Fab-phage regardless of antigen affinity. The catch and release approach, which is highly amenable to automation, reduces the recovery of nonspecific phage bound to the streptavidin bead or otherwise absorbed or aggregated onto the surface.

Several other improvements were made to the standard Fab-phage protocols. We found it useful to enrich phage particles that display functional Fab-phage with Protein A magnetic beads because only about 10% of phagemid have a Fab displayed. In addition, we exposed the phage to streptavidin magnetic beads to reduce the number of nonspecific or avidin binding Fab-phage. Lastly, the stringency of the selection was increased each round by systematically lowering the antigen concentration from 100 nm to 10 nm. From experience with hundreds of antigen selections, we found that ∼1nmol of antigen is needed for the selection process (∼25 μg for a 25 kDa antigen). This is considerably below what is typically used for animal immunizations. We monitored enrichment by positive binding Fab-phage versus the starting library and found that after three to four rounds of selection the Fab-phage pool was optimally enriched for antigen binding clones.

Usually 50–100 individual Fab-phage clones from the final selection pools were plated and grown for validation. These were tested in a high-throughput format (96- or 384-plate) for direct and competitive binding using a robotic ELISA assay (supplemental Fig. S3). We measured the OD₆₅₀/min for direct binding of the Fab-phage to biotinylated antigen bound to a neutravidin or streptavidin coated ELISA plate (or an antigen coated ELISA plate) that was then probed with an anti-M13 phage antibody (Fig. 1D, y-axis). This high throughput assay provided a composite estimate of functional Fab-phage affinity and expression. We estimated the relative affinity by a single point Fab-phage competition binding ELISA. Here the Fab-phage were mixed with 20 nm soluble antigen and then allowed to bind to the same antigen plate (Fig. 1D, x-axis) where OD₆₅₀/min was monitored and used to calculate a competition ratio. Competition ratio is defined as the OD₆₅₀/min of competitive binding divided by the OD₆₅₀/min of direct binding signal. Fab-phage clones were considered to have passed when they were shown to have expressed at OD₆₅₀ >0.005 (or OD₄₅₀ >0.1) and were competed by soluble antigen ratio <0.5 (Fig. 1D, upper left green quadrant). Representative graphs from ELISA validation show typical selection results for passing antigen (Fig. 1D; top right and lower left) and an unsuccessful antigen (Fig. 1D; bottom right). This simple ELISA triage step for 96 Fab-Phage generally requires less than 1nmol of antigen. The most promising clones (typically ∼48) were picked and prepared for DNA sequencing. Fab sequences are aligned and unique representative clones chosen for subsequent isolation, expression, and purification.

For expression of soluble monovalent Fab fragments, the Fab-phage clones were PCR amplified and subcloned into standardized E. coli expression plasmids, either Avi- or non-Avi-tagged (recombinant-antibodies.org/protocols), by SLICE or Ligase independent cloning methods (20, 21). Sequence verified plasmids were then transformed into C43 Pro+ E. coli (16) for expression studies. Expressed Fabs were purified by semi-automated Protein A chromatography that typically recovered 100–500 μg purified Fab proteins. Not all Fab frameworks bind to Protein A, but the Herceptin (4D5) framework chosen for our Fab-phage library binds efficiently to Protein A (47).

To better understand the global biophysical properties of the Fabs generated from this pipeline we analyzed a subset of primary validated Fabs for expression level, stability, and affinity. Expression levels in E. coli of over 700 monomeric Fabs showed that more than 90% express in the range of 1 to 10 mg/L, as assessed by Bradford assay or A₂₈₀ measurements (Fig. 2A). Given that commercial antibody reagents are generally provided in 100 μg aliquots, simple 50–100 mL-shake flasks culture provide access to purified Fabs for many applications. To test the stability of the Fab antibodies we measured the thermostability of 96 randomly chosen Fabs by differential scanning fluorimetry (DSF) facilitated by robotic cycling PCR instruments (48). The T_m values of the Fabs varied over a narrow range, 75–85 °C (Fig. 2B). Detailed affinity measurements for ∼200 Fabs showed EC₅₀ values that range from high pm to 50 nm with the average around 10 nm (Fig. 2C). The high expression level E. coli, the robust Fab stability, and high affinity encouraged us to scale up the process. To date, 537 TF antigens have been processed through the robotic Fab-phage selection pipeline and 435 antigens (81%) have produced about 3000 primary validated Fab antibodies that were moved into a secondary validation pipeline.

As a next level of antibody validation, we sought to determine how the Fabs would bind to their parent TF in a cellular lysate. Kelley and co-workers have shown that human antibodies that are rapidly cleared in serum exhibit high levels of nonspecific binding in assays containing cellular extract (49). Nonspecific binding was estimated by the reduction in the direct binding (ELISA EC₅₀) of antibody to antigen in the presence of non-cognate protein from baculovirus extract. We developed a similar assay termed Spiked Immunoprecipitation (Spiked-IP) designed to rapidly screen all Fabs to identify any sticky or nonspecific binding Fabs (50). This allowed us to assess the ability of the Fab to bind and pull-down exogenously added TF antigen in the presence of complex human cell lysates compared with buffer alone. Briefly, biotinylated Fab is bound to streptavidin magnetic beads and mixed with antigen for binding in the presence or absence of a HEK cell lysate as monitored by flow cytometry (Fig. 3A). The assay was also run in reverse mode where biotinylated antigens were bound to beads and allowed to bind Fab in the presence of the HEK cell lysate or in buffer alone.

In order to pass the Spiked-IP assay, lysate to buffer binding signals had to be within 2-fold of each other. Those Fabs having lysate signals more than twofold lower than the buffer condition were considered to have too much nonspecific or sticky binding to other proteins in the lysate as shown in an example for three antigens with three to four Fabs each (Fig. 3B). Although this assay does not directly address specificity of the Fab to its target antigen, we believe this assay robustly screens for Fabs with a high probability of being specific. About 56% of the antibodies that passed the primary validation steps (affinity, stability, and expression) passed the Spiked-IP secondary validation test (Table I). The endogenous levels of TFs are typically more than 100-fold lower than we estimate on the bead so it is unlikely we would discard passing antibodies from endogenous TF competition. Nonetheless, false negatives are possible, and we did not evaluate them further.

The ability to successfully generate at least one high affinity Fab that passed the primary and secondary validation depended on TF domain type (Table I). For example, antibody selections for SCAN domains succeeded 88% of the time, whereas Zinc finger domains were only successful 26% of the time (Fig. 3C). It is possible these differences reflect the stability of the domains or that antigens form complexes with a molecule in the lysate (e.g. DNA) in such a way that prevents Fab binding.

We also tested a subset of Fabs (n = 18) that were generated to SCAN domain-containing antigens in a single point pairwise ELISA analysis to address specificity of the Fabs to the different SCAN domains of varying homology. In almost all cases the Fabs generated here mono-specific (Fig. 4, left panel). In the few cases where the Fab was poly-specific, for example RAB-S181, S145, and S169, the target antigens were >93% identical and only differed in amino acid composition by up to 5 residues (Fig. 4, right panel). This degree of selectivity was remarkable given we did not impose a counter selection against the closely related antigen. We also tested a Fab that failed the Spiked-IP assay as a negative control and this indeed showed a higher degree of poly-specificity (Fig. 4, left panel).

We also noted that the number of unique Fabs obtained per antigen increased as antigen size increased over 14 kDa (Fig. 5A). This is not surprising, as the surface area for binding increases. We also find that the number of Fabs identified per antigen systematically decreases as the isoelectric point (pI) of the antigen increased from pI 4 to 10 (Fig. 5B). This could reflect the anionic nature of the phage coat to bind cationic proteins (51). These general parameters help to define antigen clones with high likelihood of success.

Overall, 56% of all TF antigens screened generated Fabs that passed primary and secondary validation criteria. We believed some of this failure could be from inefficient protein folding in E. coli. Therefore, we sought to develop additional high through-put antigen expression systems to supplement E. coli expression system. To systematically compare expression methods and selections from them, we studied the ability to produce Fabs from 32 different TF antigens expressed in three different formats: E. coli, in vitro transcription translation, and yeast antigen display.

We put these 32 antigens through the automated Fab selection pipeline and scored them based on the number of Fabs that passed primary validation criteria. Indeed there was considerable variation in the success of obtaining primary validated Fabs for the 32 TF antigens across the different expression formats (Fig. 5C and supplemental Table S2). Although the three formats in aggregate produced at least one primary validated Fab-phage per antigen, IVTT antigens were somewhat more successful, 66%, with E. coli constructs and yeast antigen display (YAD) averaging 41 and 47% success, respectively. Some antigens produced many passing antibodies (>45 Fabs per 96 tested) and others produced only a few. We conclude that having multiple antigen formats increases the likelihood of successful selection by providing back-up systems.

Transcription factors can reside in almost any compartment in the cell from the cell membrane, to cytosol to nucleus (52). Having an unprecedented collection of renewable antibodies to TFs allows us to begin to probe the cellular distribution of these TFs by immunofluorescence (IF). We chose six human cell lines that are known to express detectable levels of mRNA based on RNA sequence data (not shown) for each of 270 TFs which we had multiple passing Fabs. We used a high-throughput IF assay to screen 1017 Fabs in permeabilized cells (supplemental Table S3). Columbus software associated with an Opera imager (Perkin Elmer) was utilized to quantify the fluorescence intensities across multiple fields for each antibody to generate a heat map showing a ratio of nuclear intensity to cytoplasmic intensity (Fig. 6A). Staining patterns were also scored by visual inspection using four standard GO-based descriptors typically applied to categorize cellular localization: cytoplasmic, nuclear, mixed cytoplasmic/nuclear, and miscellaneous. Interestingly, TF proteins appear to have different localization patterns depending on cell type and growth conditions. These results suggest that the Fabs produced here were specific for their target TF and could be used to monitor TF trafficking. Examples of each of these categories are shown in Fig. 6B. When analyzing the binding patterns of all Fabs tested by IF, we found that on average ∼70% of the Fabs tested localized predominantly to the cytosol and ∼20% to the nucleus (Fig. 6C). There was remarkable consistency between these two scoring methods and across all cell lines in terms of the proportion in the nucleus versus cytosol (20% versus 70%). Moreover, some TFs were found in different locations depending on the cell type likely reflecting their activation state differences. We also compared our IF data with those in common from the Protein Atlas, which generated affinity purified polyclonal antibodies against segments of TFs. Remarkably, we found that of the 55 antigens where an overlapping antibody has been developed, 50 (90.1%) show identical localization patterns (supplemental Table S5).

An example siRNA knockdown (K_D) validation was done using the BATF Fab/antigen combination and showed significant reduction in staining intensity when cells were transfected with antigen specific siRNA and compared with the anti-flag secondary antibody only control, that showed no nonspecific signal because of secondary antibody binding (supplemental Fig. S5). Quantitation was done by ImageJ analysis and showed a significant decrease in staining with BATF (p < 0.0001) antibodies in siRNA-transfected cell in contract to control siRNA-transfected cells. Further studies will be useful to study what external stimuli alter the trafficking of these TFs. Fabs shown in supplemental Table S4 are currently available to the scientific community for further study.

In a parallel project, we collaborated with the Structural Genomics Consortium (SGC; http://www.thesgc.org/) to generate Fabs to an extensive group of chromatin remodeling proteins (supplemental Tables S1 and S4). Similar to the TFs, these chromatin-remodeling proteins are large multi-domain molecules that are difficult to express and are prone to aggregation in their full-length forms. Thus, as was the case for the TFs, the design of the antigens was based on identifying domain types that could be expressed as independent and stable entities. In many cases we had direct structural information to guide the domain expression experiments. From the phage display selections against 211 antigens we obtained 334 Fabs that passed primary affinity, and secondary Spiked-IP or IP-MS validation. Marcon and co-workers (53), developed an IP-MS method and tested 1154 Fabs to 154 of the epigenetic antigens for binding their endogenous target in HEK293 cells. They found that 452 of these antibodies pulled down the endogenous target for 98 antigens where the antigen (and in some cases known binding partners) were among the top three proteins identified. The other 54 targets were presumed not expressed at high enough endogenous levels to detect. This success rate from primary to secondary validation was roughly the same as the Spiked-IP method. Although the IP-MS assay is more labor intensive than Spiked-IP, it provides direct mass data for antigen capture in the presence of complex cellular lysate like the Spiked-IP assay. The overall success rate for antigens producing primary and secondary passing Fabs was 68% (123 out of 200 tested). The generally higher success rate for the epigenetic targets was probably because of the fact that the expressed domains benefited from knowledge of the actual structures of many of the domains, which had been determined previously in the SGC.