Phage Display Techniques

The mAb CR3014 was isolated from a semisynthetic single-chain variable antibody fragment (scFv) phage display library, expressed as human IgG1 molecules and purified as described previously [
22 (link),
27 (link)]. An immune scFv phage display library was constructed from lymphocytes of a convalescent SARS patient from Singapore essentially as described [
28 (link)]. From this library, CR3022 scFv was selected for binding to UV-inactivated SARS-CoV, essentially as described [
22 (link)]. SARS-CoV (Frankfurt 1 strain [FM1]) was prepared as described and UV-irradiated for 15 min (UVB radiation, 280–350 nm; λ
_max, 306 nm) at 4 °C. CR3022 scFv was converted into a human IgG1 format and expressed and purified as described. Anti-rabies mAb CRJA served as negative control.

Isolation of single-chain Fv fragments (scFv) clones specifically reacting with the human SARS-CoV-2 spike protein was performed according to our previous report with some modifications (23 (link)). Biotinylated human SARS-CoV-2 spike protein (#HAK-SPD _BIO-1, Hakarel Co., Ltd., Ibaraki, Japan) and biotinylated human ACE2 protein (#AC2-H82E6, AcroBiosystems, Inc., DE, U.S.A.) were used as antigens. First, the biotinylated ACE2 protein was mixed with the scFv phage display library constructed from naïve donors to remove scFv reacting with the ACE2 protein non-specifically (negative panning). Subsequently, the resultant library was mixed with SARS-CoV-2 protein to enrich for specific scFv (positive panning). After two-rounds of negative and positive panning, soluble scFv expression in Escherichia coli infected with the phage was induced. The resulting supernatant was immediately used for enzyme-linked immunosorbent assay (ELISA) screening. One scFv clone that reacted with SARS-CoV-2 but not with ACE2 was isolated. Sequences containing the IgH region, peptide linker, and IgK region are shown in Supplementary Data Sheet 2.

The PDZ domain dataset (SI Appendix, Table S1) was taken from ref. 20 (link) and consisted of 17 human PDZ domains with experimentally determined structures. Binder peptides for the 17 PDZ domains were downloaded from supplemental data for ref. 21 (link) (https://baderlab.org/Data/PDZ). Experimental amino acid frequency matrices (PWMs) were constructed from the PDZ binder peptides, with clone frequency weighting. For the AlphaFold simulations, 20,000 random peptide sequences of length equal to the peptide in the experimental structure were generated using NNK codon frequencies to match the amino acid bias in the phage-display libraries. The experimental structure listed in the template column of SI Appendix, Table S1 was used as the sole template, with the random peptide sequences aligned to the template peptide and single-sequence MSA information.
The SH3 domain dataset (SI Appendix, Table S2) consisted of 19 SH3 domains with experimentally determined structures extracted from the Database of Peptide Recognition Modules (http://prm-db.org/) (22 (link)). Experimental PWMs were downloaded from the PRM-DB. SH3 domains can bind peptides in two orientations, denoted Class I and Class II, which have opposite chain orientations: Class I peptides often match a +XXPXXP sequence motif, where “+” denotes a positively charged amino acid and X is any amino acid; Class II peptides often match a PXXPX+ motif. SH3-peptide PWMs from PRM-DB were annotated as Class I or Class II by choosing the class whose sequence motif had the highest PWM frequency (averaged over the three motif positions). Five of the SH3 domains had multiple PWMs in the PRM-DB, one of which was assigned as Class I and one as Class II; these domains were modeled twice, once in each orientation. For AlphaFold modeling, the native PDB structure listed in SI Appendix, Table S2 (“SH3 template” column) was used as the template for the SH3 domain. Four peptide-SH3 structures with peptides in the desired orientation (i.e., Class I or Class II) were chosen as “Peptide templates” based on SH3 domain sequence identity (SI Appendix, Table S2, Peptide templates column). The peptides in these structures were transformed into the reference frame of the SH3 domain template by structural superimposition to create hybrid template models for AlphaFold. Multiple structural alignment was used to identify the core motif positions (+XXPXXP or PXXPX+) in each template peptide. The peptide sequence modeled in the AlphaFold runs consisted of the core motif together with one residue on either side (nine residues for Class I and eight residues for Class II).
For comparison with experimental PWMs, predicted PWMs were constructed from the top-ranked peptide sequences. Peptides were ranked by protein–peptide inter-PAE: The sum of the residue–residue PAE scores for all (protein, peptide) and (peptide, protein) residue pairs, where PAE is AlphaFold’s “predicted aligned error” accuracy measure. The experimental PWMs were derived from phage-display experiments with random peptide libraries of size 10⁹ and greater, whereas the predicted PWMs were based on 20,000 modeled peptides. To account for this differential and better match the entropy of the amino acid frequency distributions, we squared the predicted amino acid frequencies and renormalized them to sum to 1. This had the effect of increasing the information content of the predicted PWMs without changing the order of amino acid preference. The exponent of 2 can be partly rationalized by the approximate twofold differential in log search-space size between predictions and experiments. Following ref. 20 (link), predicted and experimental PWMs for PDZ domains were compared over the last five C-terminal peptide positions. Predicted and experimental PWMs for SH3 domains were compared over the core 7 (for class I) or 6 (for class II) positions of the SH3 motif together with the immediately adjacent positions, if those positions were present in the experimental PWM. Two measures of PWM column divergence were used to assess predictions: average absolute difference (AAD) and the Frobenius metric. The AAD for a single PWM position equals the sum of the absolute frequency deviations for all amino acids, divided by 20; AAD ranges from 0.0 (perfect agreement) to 0.1 (maximal divergence). The Frobenius metric for a single PWM position equals the square root of the sum of the squared frequency deviations; it ranges from 0.0 (perfect agreement) to the square root of 2 (maximal divergence). The AAD and Frobenius values in Fig. 4 were averaged over all PWM columns.

A phage display was performed to select a specific vNAR, starting with a naive vNAR library of H. francisci shark in the pCOMb3X plasmid previously generated^{37 (link)}. After reamplification, phages were obtained against rhTGF-β cytokine (Peprotech, 100-21) resuspended in 10 mM citric acid, pH 3.0, according to manufacturer instructions. Two wells of a 96-well plate coated with rhTGF-β (5 μg/mL) and incubated for 1 h at 37 °C. Wells were blocked with 150 μL of PBS-BSA 3% for 1 h at 37 °C. Then, 50 μL of phages were added and incubated at 37 °C for 1 h. Then, the washing steps are gradually increased to 7 for round 1, 14 for round 2, and 21 for round 3 to increase the stringency. These washes raise 150 μL of TBS-Tween 0.05% (TBST) per well five times and are allowed to stand 5 min between each wash. After the wash rounds, 50 μL of trypsin 10 μg/mL was added in 1% BSA, followed by 30 min at 37 °C incubation. The wells were washed by raising the solution volume vigorously ten times and using the eluted phages to infect a culture of 2 mL of E. coli strain ER2537 (OD_600nm = 1), followed by incubation of 15 min at room temperature. Finally, transferring the culture to a 50 mL tube containing 6 mL SB medium and 1.6 μL carbenicillin (100 mg/mL, Sigma, C1389) and incubated for 1 h at 37 °C at 250 pm. The output titration count was obtained with 2 μL of the initial 8 mL culture and diluted in 200 μL of SB medium, plating 10 μL and 100 μL in LB carbenicillin plates. To the input result, a culture of 2 mL of ER2537 cells was grown at an OD_600nm = 1. Then 50 μL was infected with 1 μL of a 1:10^–8 dilution of phages obtained after each panning round and incubated for 15 min at room temperature, finally plated onto LB agar plates with carbenicillin (100 μg/mL). The plates were incubated overnight (ON) at 37 °C. After incubation in standard conditions, the input and output titers were obtained by multiplying the number of colonies by the culture volume and dividing by the plating volume³⁸ .
After the 1 h incubation of the 8 mL culture, 2.4 μL of carbenicillin (100 mg/mL) was added, and the tube was incubated for another hour and transferred to a 500 mL flask. Next, 1 mL of helper phage VCSM13 phage VCSM13 (Stratagene, 200251), 91 mL of SB medium, and 46 μL of carbenicillin (100 mg/mL) were added to the flask and incubated for 2 h at 37 °C and 250 rpm. Then, 140 μL of kanamycin (Sigma, 60615) was added at 50 mg/mL and incubated for 12 to 16 h. This protocol was repeated in each round, except the next rounds used only one well with an immobilized cytokine and increased washed steps of 7, 14, and 21. Finally, a colony PCR screening selects clones with the vNAR sequence.