Aga2p-fused variant libraries displayed on S. cerevisiae EBY100, titrated with fluorescent antigen and ranked by FACS to isolate high-affinity binders
Yeast surface display exploits the native yeast mating adhesion system: Aga1p, a GPI-anchored subunit permanently tethered to the cell wall, forms two disulfide bonds with Aga2p, a small secreted subunit. Fusing a protein of interest to the N- or C-terminus of Aga2p creates a covalently anchored display scaffold — roughly 30,000–100,000 copies per cell — with the crucial property that each cell displays only one genotype, physically linking phenotype (binding) to genotype (the encoding plasmid) for downstream recovery.
Library construction begins with PCR-amplified variant inserts (e.g., a randomized CDR-H3 loop, a combinatorial scFv repertoire, or an error-prone-PCR mutant pool of a lead binder) bearing 40–50 bp homology arms matching the linearized pCTCON2 vector on either side of the AGA2 fusion junction. Co-electroporation of insert and cut vector into EBY100 (a Trp1-, Ura3- auxotrophic strain carrying a genomic, galactose-inducible AGA1 copy) triggers homologous recombination in vivo — yeast gap-repair — reconstituting a circular, selectable plasmid only in cells that took up both fragments correctly assembled.
Electroporation parameters: 2.5 kV, 25 µF, ~4.5 ms time constant in a 0.2 cm cuvette; typical transformation efficiency is 10^7–10^8 transformants per µg of insert DNA, scaled across 10–40 parallel cuvettes to reach library sizes of 10^7–10^9 for naive antibody repertoires or focused affinity-maturation libraries. Transformants are recovered in YPD for 1 hour, then plated or grown in bulk on SD-CAA (glucose, casamino acids, Trp-dropout) selective medium, which represses AGA2 expression via the glucose-repressed GAL1 promoter — keeping the library in a non-displaying, purely genetic state until the experimenter chooses to induce.
Each clone's genotype (the variant DNA sequence) and phenotype (the folded, surface-displayed protein and its binding behavior) remain physically coupled within a single cell for the entire selection process — the central design principle that makes yeast display, like phage display, an in vitro directed evolution platform rather than a one-off binding assay. Library complexity is verified by serial dilution plating (colony counts) and, increasingly, by early-round Illumina sequencing of the naive population to confirm expected diversity and absence of severe bottlenecking during transformation and recovery.
Before any antigen is applied, the library must be shown to actually display protein — and to display it as a complete, correctly folded polypeptide rather than a truncated or degraded fragment. Switching the carbon source from glucose to galactose de-represses the bidirectional GAL1-10 promoter driving both AGA2-fusion and, in trans, AGA1 expression, and a two-color epitope-tag readout distinguishes full-length surface protein from truncations.
Induction is typically performed at 20°C rather than the yeast optimum of 30°C, and often extended to 16–24 hours in SG-CAA (galactose/raffinose, casamino acid, Trp-dropout) medium. Lower temperature slows the secretory pathway, reducing aggregation and unfolded protein response burden for disulfide-rich or otherwise fold-sensitive inserts (single-chain antibody variable fragments, in particular, are notoriously sensitive to expression temperature) and measurably improves the fraction of properly folded, functional surface protein without simply increasing total expression level.
The Aga2p fusion construct is flanked by two short epitope tags: an N-terminal HA tag and a C-terminal c-myc tag (in the common orientation; some vectors invert this). Because eukaryotic ribosomes occasionally stall or the mRNA/protein undergoes proteolysis, a cell can express and secrete a truncated fusion that carries the N-terminal tag but not the C-terminal one. Staining simultaneously with an anti-HA antibody (one fluorophore, e.g., Alexa Fluor 488) and an anti-c-myc antibody (a spectrally distinct fluorophore, e.g., Alexa Fluor 647) and analyzing by two-color flow cytometry resolves the population into double-negative (uninduced/non-displaying), single-positive (truncated), and double-positive (full-length, competent for antigen binding) subpopulations.
Only the double-positive gate is carried forward into antigen selection; typical healthy induced libraries show 40–75% double-positive cells, while poorly folding or toxic inserts can depress this to well under 20%, signaling a construct or induction-condition problem that should be resolved (lower temperature further, add an osmotic stabilizer such as 1M sorbitol, or reduce induction time) before committing reagent-intensive antigen titrations to a compromised population.
Yeast surface display converts a solution binding equilibrium into a single-cell fluorescence readout. Because thousands of identical fusion proteins decorate each cell, the mean fluorescence intensity (MFI) of bound, fluorescently labeled antigen scales with fractional receptor occupancy — allowing an entire equilibrium titration curve to be constructed from a single flow cytometry experiment across a dilution series, without ever purifying protein.
A dilution series of biotinylated antigen (commonly spanning 0.1–1000 nM, log-spaced across 8–10 points) is incubated with aliquots of the induced, double-positive-verified library or a single clonal population until equilibrium is reached — often 1 hour for micromolar-to-low-nanomolar affinities but extended to several hours or even overnight at 4°C for sub-nanomolar interactions, since equilibration time scales with 1/(kon·[Ag] + koff) and very tight, slow-off-rate binders equilibrate slowly at low antigen concentration. Bound antigen is revealed with fluorescent streptavidin (e.g., streptavidin-phycoerythrin), and c-myc co-staining again marks the display-competent population so that antigen signal can be normalized per unit of displayed receptor.
Mean antigen-channel fluorescence, gated on the display-positive population, is plotted against antigen concentration and fit to a one-site specific binding (Langmuir) isotherm: MFI = MFI_max · [Ag] / (Kd,app + [Ag]). The concentration producing half-maximal fluorescence is the apparent Kd. Because the antigen is added at a fixed concentration for the whole incubation (not depleted, given typical cell numbers of 10^5–10^7 versus antigen amounts in the µg range), this closely approximates a true equilibrium dissociation constant, unlike some solution assays where ligand depletion introduces systematic bias.
Several practical factors affect accuracy: avidity artifacts if antigen is allowed to dimerize or if streptavidin (tetravalent) cross-links surface antigen at high density, inflating apparent affinity; incomplete equilibration for slow binders, which underestimates Kd (i.e., seems weaker than it is); and photobleaching or antigen degradation during long overnight incubations. Well-controlled titrations on a first-round naive-library isolate typically report apparent Kd in the 100 nM–5 µM range — weak by therapeutic standards but a validated starting point for the affinity maturation performed via sequential FACS rounds.
Fluorescence-activated cell sorting is the selection engine of yeast display: a two-parameter gate on display level (c-myc) and normalized antigen binding isolates the small fraction of the library — often well under 1% — with the most favorable affinity-to-expression ratio, and iterative rounds under increasingly stringent conditions drive convergence toward the tightest binders present in the original diversity.
The core FACS gate plots c-myc (or HA) display intensity on one axis against antigen-binding fluorescence on the other. Because raw antigen signal is confounded by display level (a cell expressing more fusion protein will bind more antigen even at identical intrinsic affinity), sorting is typically done on a diagonal gate or on the ratio of antigen signal to display signal, selecting cells that bind more antigen per unit of displayed receptor than the population average — directly enriching for higher affinity rather than higher expression. Early rounds (round 1–2) commonly use a permissive top 1–5% gate to preserve diversity and rescue moderate-affinity improvements from a large but weak-affinity naive pool; later rounds tighten to the top 0.1–1% as the population affinity distribution shifts and the experimenter wants to resolve differences between already-improved clones.
A particularly powerful selection mode for late rounds is kinetic (off-rate) sorting: cells are first saturated with labeled antigen, then chased with a large excess of unlabeled antigen for a defined dissociation window (minutes to hours), so that only cells whose bound complex has the slowest dissociation rate (lowest koff) retain fluorescent signal at the time of sorting. Because affinity is Kd = koff/kon and kon is often close to diffusion-limited (~10^5–10^6 M^-1s^-1) across related variants, selecting for slow off-rate is frequently the most direct route to sub-nanomolar Kd, and avoids an artifact common to simple equilibrium titration at very low antigen concentration: avidity from residual antigen multimerization or from unintended cross-linking by the detection reagent, which can make a population of weak monomeric binders appear deceptively tight.
Sorted cells are collected directly into SD-CAA selective medium, expanded for 24–48 hours, then re-induced for the next round; between FACS rounds, an initial round of magnetic-activated cell sorting (MACS, using antigen-coated magnetic beads) is often used to cheaply and rapidly deplete 10^8–10^9 non-binding cells before the first, more precise and lower-throughput FACS round — a hybrid MACS-then-FACS funnel that is now standard practice for large naive libraries.
The final phase converts an enriched population back into individually interpretable hits. Recovered plasmid DNA is sequenced — historically by Sanger sequencing of a few dozen picked colonies, now routinely by next-generation sequencing of the entire sorted population across all rounds — to track which variants rose in frequency, and top clones are re-isolated, re-titrated, and confirmed by an orthogonal, label-free biophysical method.
Plasmid DNA is rescued from each round's sorted population using yeast miniprep protocols (zymolyase spheroplasting followed by standard alkaline lysis), then the variable region is PCR-amplified with primers that append round-specific sample barcodes and sequencing adapters. Illumina paired-end sequencing (typically 2×250 or 2×300 bp on a MiSeq, sufficient to span most scFv or nanobody CDR regions) of the naive library alongside each sort round's output allows direct calculation of an enrichment ratio for every observed variant: frequency in round N divided by frequency in the naive (round 0) library. Variants enriching monotonically and substantially (often >100-fold by round 3–4) across successive, increasingly stringent rounds are the strongest hit candidates — a trajectory that is far more informative than endpoint frequency alone, since it distinguishes genuine affinity-driven enrichment from clones that simply expressed or grew well.
Top-ranked unique sequences (commonly 24–96, arrayed into a 96-well format) are individually re-transformed into fresh EBY100, re-induced, and re-titrated exactly as in Stage 3 to obtain a clonal (not population-averaged) apparent Kd, removing any ambiguity from mixed populations. The most promising clones are then produced as soluble, secreted or bacterially expressed protein and characterized by a label-free method — surface plasmon resonance (SPR, e.g., Biacore) or bio-layer interferometry (BLI, e.g., Octet) — which reports true solution-phase kon, koff, and Kd without any avidity or surface-density confound, serving as the gold-standard confirmation of the yeast-display estimate.
Specificity counter-screens (binding to irrelevant antigens or to closely related paralogs), thermal stability (differential scanning fluorimetry, Tm), and, for therapeutic candidates, polyspecificity and aggregation propensity assays typically follow before a clone advances to expression scale-up.
A representative affinity-maturation campaign on a naive 10^9-member nanobody library against a cell-surface receptor started with a best hit of Kd ≈ 850 nM after round 1 equilibrium titration. Three additional rounds of kinetic (off-rate) FACS sorting — tightening the dissociation chase window from 2 hours to 24 hours — enriched a single CDR3-mutated variant 210-fold in NGS read frequency by round 4. SPR confirmed a clonal Kd of 3.4 nM (koff = 4.1×10^-4 s^-1, kon = 1.2×10^5 M^-1s^-1), a roughly 250-fold affinity improvement achieved entirely through yeast display selection without any structure-guided design.