Rational library design by functional CDR resampling

Abstract

Successful antibody discovery relies on diversified libraries, where two aspects are implied, namely the absolute number of unique clones and the percentage of functional clones. Instead of pursuing the absolute quantity thresholded by current display technology, we have sought to maximize the effective diversity by improving functional clone percentage. With the combined effort of bioinformatics, structural biology, molecular immunology and phage display technology, we devised a bioinformatic pipeline to construct and validate libraries via combinatorial assembly of sequences from a database of experimentally validated antibodies. Furthermore, we showed that the libraries constructed as such yielded a significantly increased success rate against different antigen types and generated over 20-fold more unique hits per targets compared with libraries based on traditional degenerate nucleotide methods. Our study indicated that predefined CDR sequences with optimized CDR-framework compatibility could be a productive direction of functional library construction for in vitro antibody development.

Figure 1. Schematic view of the PDC library strategy

In a library constructed from tens of unique CDR sequences of the L2, L3, H2 and H3 regions (four different ovals) of antibody V domains (gray box). The structural model was built based on the parental scFv sequence of PDC library using homological model building tool PIGS [26]. b,) CDR sets and the DNA arrangement of the small-set library and large set libraries. The scale of CDR sets for two different libraries are labelled above. CDRs L1, L2, and L3 are in red and CDRs H1, H2 and H3 in cyan. The Light chain framework is shown in white, and the heavy chain framework in dark blue.

Figure 2. Bioinformatic pipeline of DNA array design from the database of validated antibodies

Macro steps including database integration, sequencing filtering, antibody alignment as well as DNA sequence generation are shown in blue; routine steps that do not require extra input are highlighted in brown; sub-steps that can be customized to adapt to different purposes are in red. The entire bioinformatic pipeline was realized via Python/BioPython module.

Figure 3. Design, construction and validation of defined CDR libraries

Left panel: CDR sets and the DNA arrangement of small-set and large set libraries. The scale of CDR sets for two different libraries are labelled above. CDRs L1, L2, and L3 are in red and H1, H2 and H3 in cyan. Light chain framework is white, and heavy chain framework is dark blue. Right panel: the libraries are generated by Kunkel mutagenesis, where oligos are first annealed to ssDNA, and heteroduplex DNA is generated by filling in the gaps between different mutagenic oligos by T7 DNA polymerase and T4 ligase; the fully recombinant HD DNA will then be selectively amplified ~100-fold using Rolling Circle Amplification (RCA), linearization and recircularization.

Figure 4. Typical quality control data of PDC library by sequencing

Library quality control analysis for four CDRs (top row to bottom row: L2, L3, H2, H3); the left panel of each row shows whether the CDR sequence in the scFv was found in the predefined CDR database, where over 80% of CDRs can be identified as designed (y: CDR frequencies, x: all 96 scFv sequences by Sanger method). The middle panel is the experimental frequencies of any CDR observed in all 96 scFv sequences (y: CDR frequencies, x: all possible CDR sequences in the predefined database.), where an unbiased distribution of CDR frequencies found in 96 scFv sequences (represented by each circle points) should be comparable to the theoretically calculated distribution (represented by the bar Hight) shown in the right panel.

Figure 5. Typical dissection of CDRs for hits discovered from two PDC libraries, and the performance data of these libraries at different stages

Left panel: phage hits CDR dissection for the large library (p2146) and small library (p2148), both screens targeted on the same antigen (XIAP3B). Each sequence is compared with the predefined CDR database; an identifiable CDR ID is recorded as an integer; the same CDR IDs from the same screen but shared among different hits are highlighted in the same colour. Right panel, the percentage success rate across four key steps of antibody discovery are shown. Parallel comparison between small library (red bar) and large library (grey bar) are included. Zeros mean failure to match any designed sequences, probably due to sequencing error or mutations introduced in the library construction process.

Figure 6. Identification of CDRs for nonspecific binders

Based on the sequence data from five independent screens against different antigens, similar phage supernatant ELISAs were performed on 88 monoclonal hits per screen. Nonspecific hits (OD450nm >1 for both negative control plate and target coated plated) were pooled and sequenced. Raw DNA sequences were filtered and processed through the PDC library validation pipeline. A digital id per CDR sequence was assigned; one typical clone was unexpectedly enriched across different screens (L2:48, L3: 747, H2: 229 and H3: 591), indicating this combination of CDRs might yield nonspecific affinity as well as good growth advantage. Moreover, the H2:229 was further enriched in the majority of nonspecific clones combined with other CDRs, suggesting H2:229 itself conferring undesirable affinity the base material of the negative plate.

Figure 7. ELISA and western blot validation of discovery screen hits

Top panel, validating ELISA results for scFv: 20 typical scFv proteins were checked against their cognate target (dark grey bar), as well as three other non-target antigens (read, light grey and green); affinity is quantitated as OD absorption at 450nm. Bottom panel, scFv hits that were developed using full-length proteins were checked against human cell lysate (HEK293FS) spiked with its specific target protein (lane +) and lysate only (lane −) to validate their affinity as well as specificity in Western blot conditions. For ELISA, the specific targets are (peptides: XIAP1B, XIAP2B, GRAP2B, GRAP3B, LMNA3B, TDP_42_NLS, XIAP3B, LMNA2B, proteins: hIL-12p70, FGF basic protein, Dtk-Fc, BCMA-Fc, Nogo Fc, IL-13, myelin basic protein, hLeptin, IFN alpha, BDNF, IL-1 beta, IL3 beta) For western blot-targets, the specific targets are labelled respectively.