Effective Optimization of Antibody Affinity by Phage Display Integrated with High-Throughput DNA Synthesis and Sequencing Technologies

Abstract

Phage display technology has been widely used for antibody affinity maturation for decades. The limited library sequence diversity together with excessive redundancy and labour-consuming procedure for candidate identification are two major obstacles to widespread adoption of this technology. We hereby describe a novel library generation and screening approach to address the problems. The approach started with the targeted diversification of multiple complementarity determining regions (CDRs) of a humanized anti-ErbB2 antibody, HuA21, with a small perturbation mutagenesis strategy. A combination of three degenerate codons, NWG, NWC, and NSG, were chosen for amino acid saturation mutagenesis without introducing cysteine and stop residues. In total, 7,749 degenerate oligonucleotides were synthesized on two microchips and released to construct five single-chain antibody fragment (scFv) gene libraries with 4 x 10(6) DNA sequences. Deep sequencing of the unselected and selected phage libraries using the Illumina platform allowed for an in-depth evaluation of the enrichment landscapes in CDR sequences and amino acid substitutions. Potent candidates were identified according to their high frequencies using NGS analysis, by-passing the need for the primary screening of target-binding clones. Furthermore, a subsequent library by recombination of the 10 most abundant variants from four CDRs was constructed and screened, and a mutant with 158-fold increased affinity (Kd = 25.5 pM) was obtained. These results suggest the potential application of the developed methodology for optimizing the binding properties of other antibodies and biomolecules.

Fig 1. Overall scheme of the HuA21 affinity maturation process by phage display.

(A) Candidate residues from the L1, L3, H1, H2 and H3 CDRs were targeted for diversification. (B) NWG, NWC and NSG degenerate codons were used for amino acid saturation mutagenesis. (C) Degenerate oligonucleotides containing CDR mutations were synthesized on microchips. (D) Mutant CDR libraries were amplified from the mixture of released oligonucleotides. (E) Mutant scFv gene libraries were assembled by OE-PCR and transformed into TG1. (F) Recombinant phages were amplified with helper phages, captured on antigen-coated magnetic beads, and recovered by re-infecting TG1. (G) Phage ELISA was used to isolate antigen-binding clones. (H) Positive clones were submitted for Sanger sequencing. (I) Pool of mutant CDR libraries was sequenced on the Illumina HiSeq 2000 platform. (J) NGS data was analyzed after quality filtration. (K) ScFv-Fc mutants were expressed in Expi293 cells to determine binding EC50s.

Fig 2. Characterization of library CDR sequences.

Data analysis was performed for the five individual CDRs from 13.9 million qualified sequences in the unselected libraries. (A) Sequence abundance analysis. Each point represents the total number of variants with a specific copy number. Only variants with ≤10 copies are shown. (B) Length distribution analysis. Each point represents the total number of variants of a specific length. Only variants with ≤5 nucleotide deletions or insertions are shown.

Fig 3. Amino acid composition at different CDR positions.

(A) Map of the theoretical amino acid composition using NWG, NWC and NSG degenerate codons according to the SPM strategy. (B) Map of the observed amino acid composition from 10.3 million functional sequences in the unselected libraries. Forty diversified positions and four non-diversified positions are shown. Z represents a stop codon.

Fig 4. Affinity gains by target-binding screening and NGS approaches.

The scFv-Fc mutants were expressed in Expi293 cells to determine their apparent binding affinities to ErbB2 by screening ELISA. The fold affinity enhancement was calculated as the EC50 ratio of wild-type HuA21/mutant. The numbers of mutants from different CDR libraries are indicated.

Fig 5. Frequency distribution of CDR sequences.

Data analysis was performed for five individual CDRs from 13.9 and 4.5 million qualified sequences in the unselected (BS) and selected (AS) libraries, respectively. Total counts of all variants representing that particular frequency group were calculated.

Fig 6. Heat-map of amino acid enrichment values.

The EVs of all single amino acid substitutions were calculated from 10.3 and 4.5 functional sequences in the unselected and selected libraries, respectively. Rows represent forty diversified positions from the five CDRs. Columns represent the amino acids listed in single-letter code, where Z represents a stop codon. The wild-type CDR residues in HuA21 are labeled.