Platform for high-throughput antibody selection using synthetically-designed antibody libraries

Abstract

Synthetic humanized antibody libraries are frequently generated by random incorporation of changes at multiple positions in the antibody hypervariable regions. Although these libraries have very large theoretical diversities (>10(20)), the practical diversity that can be achieved by transformation of Escherichia coli is limited to about 10(10). To constrain the practical diversity to sequences that more closely mimic the diversity of natural human antibodies, we generated a scFv phage library using entirely pre-defined complementarity determining regions (CDR). We have used this library to select for novel antibodies against four human protein targets and demonstrate that identification of enriched sequences at each of the six CDRs in early selection rounds can be used to reconstruct a consensus antibody with selectivity for the target.

Fig. 1. Library Strategy

(A) (top) A typical NNK-type synthetic library with 18 changes to the amino acid sequence of the 6 CDRs will have a maximum diversity of 20¹⁸ (=2.6 × 10²³) and nearly 1 stop codon per clone. Suppression of these stop codons is used to allow expression of full length in E. coli, although at greatly reduced levels. (Bottom) By replacing the NNK strategy with pre-defined CDRs we can achieve a similar maximum potential diversity and the same practical diversity (ie. the actual size of the library; dependent on the transformation efficiency of the host bacterium and the number of transformations used to create the library). (B) A schematic of the pre-defined CDR library built using 1000 oligos each for CDRs L1, L2, H1, and H2 and 2000 oligos each for CDRs L3 and H3. (C) For the pre-defined CDR approach all nucleotides within CDRs L1, L2, H1, and H2 were changed at a frequency based upon the somatic hyper-mutation rate found in vivo for that specific CDR. Any oligos containing stop codons, cysteines, or restriction sites were eliminated. The mutation rate multiplied by the length of the CDR results in the theoretical frequency of finding the wildtype germline amino acid sequence. CDRs L3 and H3 have a much higher mutation rate in vivo and thus were synthesized as described in the text.

Fig. 2. Library construction approach

Phase I – incorporation of diversity in L3 and H3. 2000 oligos each of L3 and H3 were synthesized and subsequently cleaved from a chip. Each pool of cleaved oligos is amplified using a reverse primer containing phosphorothioate linkages on its 5’ end. The resulting double-stranded DNA is treated with T7 exonuclease to selectively degrade the unmodified strand of the dsDNA molecule. The resulting single-stranded DNA, or ‘megaprimer’, is then annealed to the uracilated, circular, single-stranded phagemid DNA and used to prime in vitro synthesis by DNA polymerase. The ligated, heteroduplex product is then digested by SacII where the uracilated strand is cleaved by uracil N-glycosylase, favoring survival of the newly synthesized, recombinant strand containing the megaprimer and transformed into E. coli cells. Upon completion of a library that contains L3 and H3, the process is repeated in phase II with the incorporation of 1000 oligos each of L1, L2, H1, and H2 CDRs.

Fig. 3. Sequence analysis of selected clones

Specific CDR sequences were enriched during selection. The sequences identified with the highest frequency are labeled sequence A, B, and C. All other sequences were found at a frequency of less than 10%. (A) Oligonucleotide distribution of CDRs found against the EZH2 target after three rounds of biopanning shows a strong consensus in each CDR (represented as sequence A). (B) Oligonucleotide distribution after two rounds of biopanning against the EZH2 target. The same sequences A, B, and C found in round 3 were also identified in round 2. (C) The enriched sequences A, B, and C found after two and three rounds of panning against EZH2 protein fragment.

Fig. 4. Analysis of scFv specificity

The consensus sequence scFvs found after biopanning against four targets, ZMAT-3, EZH2, TDP-43, and ZNF622, were tested in an ELISA against the specific target antigen and the other three, non-relevant targets. For EZH2, two different scFvs with different L1 and H3 sequences were tested. All target proteins were biotinylated and attached to a neutravidin coated ELISA plate. Binding was detected using an HRP-conjugated anti-FLAG antibody that recognizes the tag on the scFv. All of the scFvs bound specifically to their cognate antigens.