Staged heavy-chain preselection reallocates Fab library capacity from heavy-chain breadth to light-chain sampling depth: a single-target, same-repertoire comparison with direct Fab screening

Combinatorial fragment antigen-binding (Fab) libraries encode a heavy–light chain pairing space that can exceed 10 14 combinations, far beyond the 10 7 –10 9 diversity eukaryotic display systems can construct and screen. Direct Fab screening commits both chain dimensions at once, so how finite library capacity is distributed between them is a design choice. We implemented a sequential heavy-chain (HC) preselection–remating workflow in yeast surface display: antigen-compatible HC variants are enriched as haploids, then remated with a light-chain (LC) repertoire. Using 18,023 sorted cells as an upper bound on staged HC variants, nominal LC pairing opportunities per HC were approximately 16 for conventional screening and at least 7.8 × 10 3 for staging, an estimated 500-fold difference. HC-first filtering acts only on HC variants that remain displayable and antigen-competent without a partner LC; in a five-antibody reference panel, four display-competent HCs showed no detectable cognate-antigen binding in the HC-only format at 100 nM cognate antigen (CD40, 500 nM). In a SARS-CoV-2 spike-targeted campaign, observed HC clonotypes among sampled clones declined from 15 to 4 across the staged workflow, the dominant CLR201 clonotype accounting for 60.9% with 12 distinct LC complementarity-determining region patterns among its 14 clones; direct Fab screening from the same repertoires returned 8 clonotypes with a dominant frequency of 30% (23 and 10 clones sampled, respectively). CLR201 bound the receptor-binding domain (RBD) with an apparent on-yeast half-maximal effective concentration (EC 50 ) of 2.75 nM, versus 3.70–9.73 nM for directly screened clones. As full-length IgG1, CLR201 bound wild-type and five variant RBDs with sub-nanomolar apparent EC 50 values and neutralized D614G spike-pseudotyped virus. Staged HC preselection reallocated finite diploid library capacity from HC breadth toward LC sampling depth, an estimated 500-fold increase per retained HC, without expanding the number of HC–LC pairings constructed and screened. That reallocation follows from library architecture and is not antigen-specific. Whether deeper LC sampling improves the binders recovered is antigen-dependent, and is addressed here as a single-target proof of principle in which the two campaigns were not stringency-matched: direct screening retained broader observed HC representation, whereas staging yielded the clone with the lowest apparent on-yeast EC 50 tested.