Cell-free expression of protein ligands
DNA constructs encoding the ligands, each fused to a C-terminal assay tag, were designed by reverse-translating the target protein sequences. The sequences were optimized for manufacturability and yield using codon optimization algorithms to maximize expression efficiency. The optimized DNA constructs of the variants and the 3′ fragment containing the linker and affinity tags were ordered as Gene Fragments from Twist Bioscience. Constructs were assembled using the NEBuilder HiFi DNA Assembly Kit (NEB) in 2 μL reactions. The assembled products were characterized by capillary electrophoresis (Agilent ZAG DNA Analyzer, ZAG-135-5000 Kit, FA/ZAG 96-Capillary Array Short, 33 cm), and their concentrations were measured using the Qubit DNA Quantification Kit (Invitrogen). Ligand expression was carried out in 8 μL reactions using an optimized prokaryotic in vitro translation system and 4 nM of assembled gene fragment. Reactions were incubated at 37 °C for 8 hours. Post-expression, protein concentration and yield were normalized using an affinity-based quantification assay.Affinity characterization by biolayer interferometry (BLI)
Kinetic binding measurements were performed on a BLI instrument (Gator Bio) using Strep-Tactin XT probes to capture Twin-Strep-tagged ligands.Sensor preparation and capture
Probes were equilibrated in running buffer before use. Ligands were captured with the following sequence:- Baseline 1: 120 s in running buffer
- Ligand loading: 120 s (target loading shift 0.5–1.0 nm)
- Baseline 2: 200 s in running buffer
Association and dissociation (multi-cycle kinetics)
Antigen solutions were prepared in running buffer as a half-log dilution series of 4 concentrations (1000 nM to 30 nM). Each cycle consisted of:- Association: 220 s in antigen solution
- Dissociation: 240 s in running buffer
Data processing and model fitting
Data were processed with Adaptyv fitting software. All sensorgrams underwent standardized preprocessing: trimming to the relevant phases (baseline, association, dissociation), correcting signal jumps at phase transitions, aligning the association and dissociation phases, and subtracting baseline and reference signals. Each design was fitted as a group: all of its association and dissociation curves across the concentration series were fitted simultaneously, with kon and koff shared across concentrations and Rmax fitted per curve. The default model is a 1:1 Langmuir interaction, and KD was calculated as koff / kon, with 95% confidence intervals for each parameter. Alternative binding models (bivalent analyte and conformational change) were also fitted. Where one described the curves better than 1:1, it was selected and its apparent KD reported; the portal shows this value, and the API returns it askd_app with the model name in binding_model.
Candidate fits were filtered on quality before selection. A fit was kept only if its residual error in both the association and dissociation phases, normalized to signal size, stayed below a fixed threshold, its Rmax was consistent across concentrations, and bulk refractive-index jumps were small relative to the binding signal. The remaining fits were ranked by residual error, then penalized when a rate constant sat at the edge of what the assay can resolve, then ranked by confidence-interval width.
For how to read the result, with examples of good and poor fits, see Judging fit quality and Common curve shapes in the binding characterization guide.
Ligands were classified as binders or non-binders based on the presence of quantifiable binding curves and calculated KD values. Where a ligand produced a significant signal shift during the association phase (≥300% over the negative control) but the signal could not be reliably fit, the binding label was assigned based on the magnitude of the observed shift.