A High Through-put Platform for Recombinant Antibodies to Folded Proteins

Abstract

Antibodies are key reagents in biology and medicine, but commercial sources are rarely recombinant and thus do not provide a permanent and renewable resource. Here, we describe an industrialized platform to generate antigens and validated recombinant antibodies for 346 transcription factors (TFs) and 211 epigenetic antigens. We describe an optimized automated phage display and antigen expression pipeline that in aggregate produced about 3000 sequenced Fragment antigen-binding domain that had high affinity (typically EC50<20 nm), high stability (Tm∼80 °C), good expression in E. coli (∼5 mg/L), and ability to bind antigen in complex cell lysates. We evaluated a subset of Fabs generated to homologous SCAN domains for binding specificities. These Fragment antigen-binding domains were monospecific to their target SCAN antigen except in rare cases where they cross-reacted with a few highly related antigens. Remarkably, immunofluorescence experiments in six cell lines for 270 of the TF antigens, each having multiple antibodies, show that ∼70% stain predominantly in the cytosol and ∼20% stain in the nucleus which reinforces the dominant role that translocation plays in TF biology. These cloned antibody reagents are being made available to the academic community through our web site recombinant-antibodies.org to allow a more system-wide analysis of TF and chromatin biology. We believe these platforms, infrastructure, and automated approaches will facilitate the next generation of renewable antibody reagents to the human proteome in the coming decade.

Fig. 1.

Overview of the RAN recombinant Fab selection pipeline, selection process, and examples of primary validation. A, Target antigens expressed in E. coli or other formats (Left Panel) enter the Fab-phage pipeline as biotinylated or GST-fusion antigens. Three to four rounds of antibody-phage selection generate pools of Fab-phage that are subsequently validated in a single point competition Fabphage ELISA and sequenced to identify unique binding sequences (Middle Panel). Unique Fabs that pass primary validation tests proceed into various secondary validation assays including ELISA EC₅₀, Spiked-IP, and Immunofluorescence (Right Panel). B, UCSF robotic antibody production pipeline showing two liquid handling instruments for ELISA, KingFisher Flex magnetic bead separator for phage selections, and K6–2 colony picking robots. Similar robotics platforms are in-place at all three RAN locations. C, Solution based phage selection was conducted in a seven-step process that was repeated up to four cycles; 1. Biotinylated antigens were bound to streptavidin magnetic beads; 2. Bead-antigen complex was transferred to Fab-phage library; 3. Fab-phage-antigen-bead complex was transferred to wash buffer where any unbound or weakly associated Fab-phage was removed; 4. Specific Fab-phage bound to antigen was eluted from magnetic beads by TEV protease cleavage; 5. The processed magnetic beads and non-specifically bound Fab-phage were removed from well and discarded; 6. Antigen specific phage was propagated in E. coli; 7. Propagated Fab-phage were purified and the process repeated. D, Examples of Fab-phage that pass (green box) or fail the competition ELISA (red boxes) are shown. Direct binding ELISAs (y-axis) were conducted to measure the composite of expression and binding capability and single point competition ELISAs (x-axis) were measured to verify antigen specificity. For ELISAs 96 individual E. coli colonies were tested for each antigen with anti-M13-HRP phage secondary antibody with TMB chromagen development that is monitored kinetically at OD650/min. Each FAB-Phage expressing colony is represented as a single spot on the graph with the calculated Competition Ratio (x axis) and Total Fab-phage Binding (y axis). Two examples are shown for antigens selections that produced Fabs that pass and one that fail validation.

Fig. 2.

General parameters that affect successful Fab generation and characterization. Fabs expressed in E. coli were measured for antibody (A) expression, (B) stability, and (C) affinity. Fabs successfully passing primary validation were cloned into E. coli expression plasmids and expressed and purified for further testing. Fabs derived from Library E and F typically expressed between 1 and 10 mg/L of bacterial culture (n = 720). Fabs (n = 96) had melting points between 75C and 85C. EC₅₀ ELISAs were conducted on Fabs to determine binding affinity values with a subset shown here (n = 201) with ∼85% of Fabs binding with an EC₅₀ <20 nm.

Fig. 3.

Overview of Spiked-IP with examples of data analysis and relative successes of two protein domains. A, Diagram of Spiked IP showing Fabs immobilized on magnetic beads binding to biotinylated antigens (Left) and after transfer to fresh well (Middle) the antigen-Fab-bead complex was bound to fluorescently labeled streptavidin then binding was quantified by Flow Cytometry. Samples were analyzed for binding characteristics in both buffer and a complex Hek293T lysate that showed specificity of the Fab to its cognate antigen. This assay was designed to run in either forward or reverse mode where antigen could be immobilized on beads with Fabs in solution. B, An example data set that shows 3 antigens with 3–4 Fabs each with both Passing and Failing Spiked-IP tests. Negative control (Blue) consists of beads without Fab in the presence of antigen then (Green) bars represent the assay run in the presence of buffer and (Red) lysate. To pass the assay, the differences in median fluorescence intensity between buffer and lysate tests must not be >2x. Fabs that passed the Spiked-IP test were denoted with (*). C, Representative success of Fabs generated from SCAN (n = 39) and Zinc Finger (n = 42) domains through secondary Spiked-IP test.

Fig. 4.

Assessment of the specificity of anti-SCAN domain Fabs. Left panel shows specificity heatmap of 18 Fabs tested against 17 closely related SCAN domains using the direct ELISA method. Fabs used in experiment were sorted by homology according to target protein homology as shown in the phylogenetic tree (left) and colors on the heatmap represent strength of the ELISA signal. anti-GFP Fab and eGFP protein was used as a positive control. An antibody raised again ZNF496 that failed Spiked-IP validation was used as a negative control to show nonspecific binding. Anticipated signals on diagonal represent interactions where Fabs recognize their intended antigens. Left panel shows pairwise identity heatmap of SCAN domain antigens tested. Multiple sequence alignment and phylogenetic tree of SCAN domains, as defined in SMART database, was constructed in MAFFT using l-INS-i and NJ methods respectively (63, 64). All pairwise identities were calculated in Jalview (65). Phylogenetic tree and heatmap annotations were visualized using EvolView (66).

Fig. 5.

Analysis of Fab selection success rates. Success rates were analyzed as a function of antigen (A) molecular weight or (B) isoelectric point. Greater Fab selection success was appreciated for antigens with a MW ≥14kDA and pI <8 (n = 447 antigens; n = 3343 Fabs). C, Multiple antigen expression pipelines can increase likelihood of obtaining high quality Fabs. Test antigens (n = 32) were expressed in three different expression formats: E. coli, IVTT or YAD. Fab-phage selections were conducted for each antigen using the robotic Fab-phage selection pipeline. The number of primary validated Fabs is shown for each antigen.

Fig. 6.

Photomicrographs and heat map showing Immunofluorescence staining patterns (cytosol in blue to nuclear in red) for 1017 Fabs directed to 270 TFs. A, Expressed and purified Fabs were screened against six commonly used human cell lines and images were screened by the Columbus imaging software to generate signal intensities across multiple fields. Intensities were used to generate a heatmap that shows the ratio of intensities of nuclear to cytoplasmic localization of Fabs. B, Representative photomicrographs for immunofluorescence staining of Fabs (n = 1017) against TFs (n = 270) in six different fixed and permeabilized mammalian cell lines known to express the TF. Scoring was represented as Cyto, Cyto/Nuc, Nuc, or Misc. C, Similar analysis was conducted by visually inspecting images and assessing staining patterns based on 4 GO terms; Nuclear (Nuc), Nuclear/Cytoplasmic (Nuc/Cyto), Cytoplasmic (Cyto), and Miscalenaous staining (Misc). Mean localization results were calculated and showed that ∼67% of all Fabs labeled a protein in the cytoplasm compared with ∼19% in the nucleus.