Fully synthetic platform to rapidly generate tetravalent bispecific nanobody–based immunoglobulins

Significance Nanobodies are a promising class of biologics that can be used to prevent or treat viral infections. Here, we describe the production and validation of a discovery library that produces single-domain nanobodies using an engineered human antibody variable gene segment. As a test case, anti-SARS-CoV-2 nanobodies were isolated from this library and pairs of complementary nanobodies were incorporated into an antibody-like molecule that targets the receptor-binding domain using a biparatopic mode of engagement. This modular bispecific format enabled the rapid testing of nanobody pairs, and we show that incorporating pairs of nanobodies with different specificities can have synergistic effects on neutralization breadth and potency.


Supplementary Information Text
Library construction. The diversity in the CDRH3 was introduced with oligonucleotides synthesized with trimer phosphoramidite mixtures based on the frequencies of amino acids (AAs) found in antibody CDR3 sequences (originally synthesized for the production of a human antibody library, Table S13). All CDRH3 sequences begin in CAR and end in FDY, with 5-15 randomized AAs between the consensus residues. The oligonucleotides that were used, with [ Table S14. Double-stranded DNA was generated by combining the CDRH3 oligonucleotides with an invariant oligonucleotide encoding the nanobody framework 4 segment and 3' homology for the pYDSI2u_SiDir vector (CTCCTAGGAGTTCAGGTGCTGGTGATGGAGGTGACGTGTGAGTCTTGTCACCGGATCCAG ATGAAACAGTGACCTGCGTACCTTGTCCCCAGTAGTCG). This CDRH3/FW4 fragment was then combined with the invariant 5' fragment consisting of the hVHH323 (Fig. S1, codon optimized for yeast expression) and 5' homology for the pYDSI2u_SiDir vector (GGTTTGTCATCTACAAATACAACAATCGCATCCATAGCAGCTAAAGAGGAGGGTGTTCAGCT GGACAAGAGAGAAGCTAGTGAAGTTCAATTGCAAGAATCTGGTGGTGGTTTGGTTCAACCAG GTGGTTCTTTGCGTTTGTCTTGTGCTGCGTCTGGTTTTACTTTTTCTTCTTATGCTATGGGTT GGTATAGACAAGCTCCAGGTAAAGAAAGAGAATGGGTTTGTGCTATTTCCGGTTCTGGCGGT TCTACTTATTATGCTGATTCTGTTAAAGGTAGATTTACTTGTTCTAGAGATAATTCTAAAAACA CTTTGTATCTTCAAATGAATTCTTTGAAGCCAGAGGACACAGCTGTCTACTACTGCGCC) using isothermal assembly. Upon assembling the full nanobody library containing the necessary 5' and 3' homology for the target vector, the DNA was amplified and PCR cleaned for transformation. Each CDRH3 length was prepared independently and mixed at the desired distribution prior to transformation (Fig. S2). The nanobody library was then cloned into pYDSI2u_SiDir via homologous recombination in Saccharomyces cerevisiae YVH10 cells (ATCC, MYA4940) using protocol described in the reference (1). A total of 32 transformations were performed and pooled to achieve the desired library size. Dilutions of transformed yeast were then plated on dropout medium without uracil (SD-Ura Sunrise Science) as single colonies to obtain the estimate of library diversity of 3x10 9 unique clones. The plasmids described in the manuscript will be available by MTA.

Supplementary materials and methods
Magnetic-activated cell sorting (MACS) of naïve library. Initially, 400 µL of Super Mag Streptavidin Beads (50 nm diameter Ocean Nanotech) were pre-coupled with biotinylated SARS-CoV-2 RBD for 30 min at 4 °C in PBS, 2% (w/v) bovine serum albumin (PBSA 2%). 2 × 10 11 induced yeast cells from our naïve library (oversampling our library by a factor of ~100) were subsequently washed with PBSA 2% and incubated with the pre-coupled magnetic beads o/n at 4 °C. All three rounds of MACS were performed on an autoMACS Pro Separator using autoMACS columns (Miltenyi Biotec). In the first round of MACS (positive selection, Posseld2 program), 20 runs of 1 × 10 10 cells were sorted and the binders to the magnetic beads were selected and grown in SD-Ura o/n at 30 °C, and the following day induced o/n at 30 °C in in SGCAA + Trp induction medium (20 g galactose, 1 g glucose, 6.7 g yeast nitrogen base without amino acid, 5 g bacto casamino acids, 5.4 g Na2HPO4, 8.56 g NaH2PO4⋅H2O, 8.56 mg Trp in 1 L deionized water, pH 6.5, sterilized by filtration). To remove yeast-expressing nanobody that bound nonspecifically to magnetic beads, we performed a negative selection using only the magnetic beads. 2 x 10 10 induced cells were incubated with 200 µL Super Mag Streptavidin Beads in PBSA 2% for 30 min at 4 °C. The yeast cells that didn't bind to the magnetic beads were selected (negative selection, Possel program) and subsequently incubated with pre-coupled magnetic beads to RBD (as described above) o/n at 4 °C. For the third and final round of MACS (positive selection, Posseld2 program), binders to the magnetic beads were selected and grown in SD-Ura o/n at 30 °C.

Fluorescence-activated cell sorting (FACS)
. FACS is used after depleting the library of nonbinding clones by MACS to enrich the yeast cells in RBD-specific clones. In vitro engineering of antibodies or nanobodies can lead to constructs that are polyspecific (2), so we alternated between positive and negative selections to enhance the specificity of our synthetic constructs. The cells were alternatively selected as RBD binders (affinity sorts, AFF), or depleted against a biotinylated preparation of detergent solubilized biotinylated membrane proteins (polyspecific reagent or PSR) non-binders (negative sorts, PSR). In each round of selection, 1-5 x10 7 induced yeast cells were incubated for 60 min at 4 °C (rotating at 50 rpm) with biotinylated SARS-CoV-2 RBD for AFF sorts, or biotinylated HEK-cell soluble membrane protein extracts (3) for PSR sorts in 500 µl 1% PBSA (PBS containing 1% BSA) or PBS, respectively. Yeast cells were then washed twice with 1% PBSA (affinity sorts) or PBS (PSR sorts) and coupled to 2 fluorophores (1µg/mL) for 20 min: anti-V5-AF405-conjugated to check the yeast display, and anti-biotin-APC or streptavidin-PE to check RBD binding (we alternated the use of biotin-specific secondary antibody to prevent the enrichment in secondary antibody/fluorophore specific clones). Yeast cells were then washed once with PBSA 1% and resuspended in 1 mL PBSA 1% for sorting on a FACS Melody (BD Biosciences). Selected yeast cells were sorted into SD-Ura medium, grown and induced for consecutive rounds of selection. For AFF1, AFF2 and AFF3, 100, 20 and 4 nM of biotinylated SARS-CoV-2 RBD were used, respectively. For PSR1 and PSR2, 10 µg of biotinylated HEK-cell soluble membrane protein extracts were used. Sorted cells were either prepared for NGS sequencing, or serial dilutions of the last affinity sorts were plated on SD-Ura agar. After 3 days at 30°C, DNA of single colonies was amplified using Phire Plant PCR kit (Thermo Fisher) and sent for Sanger sequencing.
hVHH323 sequencing and analysis. Libraries were deep sequenced to determine the CDRH3 at each round of selection. The DNA from the sorted yeast cells was miniprepped (Qiagen) in the presence of zymolyase (Zymo Research) and amplified through two rounds of PCR as previously described (1). The first PCR reaction generates a ~ 200 bp amplicon containing flanking universal Nextera sequencing adapters using a set of six primers: hNb323_NGSSeq_Fa, hNb323_NGSSeq_Fb, hNb323_NGSSeq_Fc, hNb323_NGSSeq_Ra, hNb323_NGSSeq_Rb and hNb323_NGSSeq_Rc (Table S14). The second round of PCR adds a specific index primer pair (i5/i7) so the library could be pooled, cleaned, and sent for deep sequencing on an Illumina MiSeq (Illumina Incorporated, San Diego CA) with the paired-end MiSeq v2 500 bp kit.
Deep sequencing analysis. Paired-end fastqs were analyzed for sequence quality using the FastQC package (FastQC v0.11.9) (4). The forward and reverse reads were merged using BBMerge (version 38.87) from the BBTools suite using the default parameters (5). Merged reads were clustered using VSEARCH (v2.15.1) to quickly group reads with fully identical sequences (6). Clustering was done using the "cluster_fast" method and fasta files were written including the abundance of each unique sequence in the fasta header. This step substantially improved performance of downstream fasta parsing as each unique sequence was only analyzed once. A custom python (Python 3.7) script was written to parse the clustered fasta output, remove primer sequences, and translate the DNA sequences to amino acid sequences. The script then quantified the unique CDRH3 positions. (Table S2). The deep sequencing datasets were concatenated based on the CDRH3 sequence. To remove likely sequencing errors, a filter was applied to remove nanobodies that did not appear in either the C (compete) or the NC (noncompete) datasets of at least two of the antibody datasets. Additionally, nanobodies that had <10 counts were also removed. With these criteria, a total of 123 unique nanobody CDRH3s were obtained. Epitope bins were assigned by defining an overlapping epitope with a tested SARS-CoV-2 antibody as having a C/NC (compete/noncompete) ratio >10, and NC/C ratio >10 for a nonoverlapping epitope. All analysis was based on raw sequencing count data.

Affinity-maturation of LM18.
To select high-affinity nanobody variants, an affinity maturation library was prepared based on our previously reported SAMPLER strategy (1). The size of the theoretical starting library was 4.2 x 10 6 unique nanobodies (138 CDRH1 variants, 116 CDRH2 variants and 261 CDRH3 variants). We also included an M34L mutation that we identified as being potentially stabilizing. The nanobody library was cloned into pYDSI2u_SiDir using homologous recombination as described above. Four rounds of FACS-based selection (AFF1-AFF2-PSR1-AFF3) were performed to isolate populations of high-affinity clones. Serial dilutions of the AFF3 sort were plated on SD-Ura agar. After 3 days at 30°C, DNA of 96 single colonies was amplified using Phire Plant PCR kit (Thermo Fisher) and sent for Sanger sequencing.
Protein expression and purification. All recombinant soluble proteins from SARS-CoV, SARS-CoV-2 and their truncated protein versions (RBD) were expressed and purified as previously described (7). Nanobody and antibody expression and purification. Nanobodies-Fc and antibodies (HC and LC constructs) were transiently expressed with the Expi293 Expression System (Thermo Fisher). After five days, 24-deep well culture supernatants were harvested and purified using protein A magnetic beads (Thermo Fisher) and tested for binding and neutralization. Selected nanobodies and antibodies were re-expressed in small to medium scale cultures and IgG-purified on Protein A sepharose (GE Healthcare). Constructs were buffer exchanged in PBS and stored at 4°C. His10tagged Nbs used for SPR and crystallization were purified with the HisPur Ni-NTA Resin (Thermo Fisher). To eliminate nonspecific binding proteins, each column was washed with at least 3 bed volumes of wash buffer (25 mM Imidazole in TBS, pH 7.4). To elute the purified proteins from the column, we used five bed volumes of elution buffer (250 mM Imidazole in TBS, pH 7.4). Constructs were buffer exchanged in TBS and stored at room temperature.

Recombinant Protein ELISAs.
Anti Histag monoclonal antibody (Invitrogen, MA1-21315-1MG) was coated onto high-binding 96-well plates (Corning, 3690) at 2 μg/mL overnight at 4 °C. After washing, plates were blocked with PBSA 3% (3% BSA in PBS) for 1 h. Then His10-tagged recombinant RBD or spike protein were captured at 1 μg/mL in PBSA 1% and incubated for 1 h at room temperature. After washing, serially diluted nanobodies or antibodies were added into wells and incubated for 1 h at room temperature. Detection was measured with alkaline phosphataseconjugated goat anti-human IgG Fcγ (Jackson ImmunoResearch 109-005-008) at 1:1000 dilution for 1h. After the final wash, phosphatase substrate (Sigma-Aldrich, S0942-200TAB) was added into wells. Absorption was measured at 405 nm after less than 15 min. Positive and negative controls were systematically used. Non-linear regression curves were plotted using Prism 8 software.
Polyspecificity reagent ELISA. According to the protocol described by Roger et al. (7), solubilized CHO-cell membrane proteins (SMP) and single strand (SS) DNA (Sigma-Aldrich, D8899) were used. SMP or SS were coated onto 96-well half-area high-binding ELISA plates (Corning, 3690) at 5 µg/mL in PBS overnight at 4˚C. After washing, plates were blocked with PBSA 3% for 1 h at RT. Antibody samples were diluted at 100 µg/mL in PBSA 1% with serial dilution and then added in plates and incubated for 1 h at RT. After washing, alkaline phosphatase-conjugated goat antihuman IgG Fcγ secondary antibody (Jackson ImmunoResearch, 109-055-008) was added in 1:1000 dilution and incubated for 1h at RT. After final wash, phosphatase substrate (Sigma-Aldrich, S0942-200TAB) was added into each well. Absorption was measured at 405 nm after 15 min.
Pseudovirus (PSV) Assay. PSV assays were performed according to the protocol described by Roger et al. (7). Assays were run with multiple batches of PSVs and at least in duplicate. As PSV titers can vary, values indicated in each graph or table were obtained with the same batch of PSV to enable accurate comparison between the tested constructs.
Surface Plasmon Resonance (SPR) Methods. SPR measurements were collected using a Biacore 8K instrument at 25°C. All experiments were carried out with a flow rate of 30 µL/min in a mobile phase of HBS-EP+ [0.01 M HEPES (pH 7.4), 0.15 M NaCl, 3 mM EDTA, 0.0005% (v/v) Surfactant P20]. Two chips were prepared in order to obtain data for the nanobodies (His-tagged) and nanobodies-Fc and bsNb4-Igs. One, anti-Human IgG (Fc) antibody (Cytiva) was immobilized to a density of ~2000-4000 RU via standard NHS/EDC coupling to a Series S CM-3 (Cytiva) sensor chip; a reference surface was generated through the same method. Two, recombinant CoV-2-RBD was immobilized to a density of ~250 RU via standard NHS/EDC coupling to flow cell 2 of a Series S CM-5 (Cytiva) sensor chip; a reference surface was generated through activation/deactivation of flow cell 1. For conventional kinetic/dose-response, bsNb4-Igs were captured to ~50-100 RU via Fc-capture on the active flow cell prior to analyte injection. A concentration series of CoV-2-RBD or CoV-2-Spike were injected across the antibody and control surface for 2 min, followed by a 20 min dissociation phase using a multi-cycle method. Regeneration of the surface in between injections of antigen was achieved by two 120 s injections of 3 M MgCl2. For conventional kinetics/doseresponse of the nanobodies, a CoV-2-RBD sensor chip was prepared as stated above. A concentration series of each nanobody was injected over CoV-2-RBD and a control surface for 3 min, followed by a 15 min dissociation phase using a multi-cycle method. Regeneration of the surface in between injections of nanobody was achieved with a single, 60 s injection of 10 mM glycine (pH 1.5), 200 mM NaCl. Kinetic analysis of each reference subtracted injection series was performed using the BIAEvaluation software (Cytiva). Sensorgrams were fit to either a 1:1 (Langmuir) binding or heterogeneous ligand model. Diffraction data were collected at cryogenic temperature (100 K) at the Stanford Synchrotron Radiation Lightsource (SSRL) on beamlines 12-1 and 12-2 for Nb-C4-225 and Nb-C4-240 complexes, and at beamline 23-ID-B of the Advanced Photon Source (APS) at Argonne National Laboratory for Nb-C4-255 complex. The X-ray data were processed with HKL2000 (8). The X-ray structures were solved by molecular replacement (MR) using PHASER (9) with MR models for the RBD and Nbs from 7KN5 (10) and for the Fab from 6XC3(11). Iterative model building and refinement were carried out in COOT (12) and PHENIX (12, 13), respectively.
Cryo-electron microscopy. Trimeric SARS-CoV-2 6P-Mut7 S protein was incubated with a threefold molar excess of LM18/Nb-C2-136 bsNb4-Ig at room temperature for 100 minutes at a concentration of 0.85 mg/mL as determined by A280. n-dodecyl-β-D-maltopyranoside (DDM) was added to a final concentration of 0.06 mM and the sample deposited on plasma-cleaned Quantifoil 1.2/1.3 300 mesh grids. A Thermo Fisher Vitrobot Mark IV set to 4°C, 100% humidity, 3 s wait time, and a 3 s blot time was used to vitrify samples in liquid ethane. Data were collected using Leginon (14) on a Thermo Fisher Titan Krios operating at 300 keV and equipped with a Gatan K2 Summit direct electron detector. Movies were aligned and dose weighted using MotionCor2 (15). Aligned frames were imported into cryoSPARC v3.2 (15,16) and the contrast transfer function (CTF) was estimated using GCTF (15)(16)(17). Particle picking was done by automated picking using templates created from an initial round of 2D classification, then extracted and subjected to multiple rounds of 2D classification for cleaning. An ab initio model was generated and several rounds of non-uniform refinement (18), CTF refinement and 3D variability analysis were performed, resulting in a final global reconstruction (Fig. 5, Table S12). To further improve the resolution of the RBD and nanobody interactions, particles were exported to Relion 3.1 (19) and subjected to C3 symmetry expansion. A mask around a single RBD and nanobodies was created using University of California San Francisco Chimera (19,20) and used for focused 3D classifications without alignment. During focused 3D refinement, a mask of the trimeric core and a single RBD with nanobodies was applied, and angular sampling was restricted to prevent rotation of one protomer onto another. A summary of data collection and processing statistics can be found in Table S12. Initial models were generated by fitting coordinates from sAbPred (21) for the nanobodies and PDB 6VYB for the S protein into the focused refinement cryo-EM map. Several rounds of iterative manual and automated model building and relaxed refinement were performed using Coot 0.9.8 (22) and Phenix real_space_refine (23). Models were validated using EMRinger (24) and MolProbity (25). Kabat numbering was applied to the nanobody chains. Final refinement statistics and PDB/EMDB deposition codes can be found in Table S12.
Modeling LM18/Nb-C2-136 bsNb4-Ig. Based on the approximate placement of CH1/CL, we used RosettaRemodel (26) to build the bsNb4-Ig in three different steps. First, we modeled just the linkers (with the corresponding sequences) connecting CH1 and Fc domains, requiring the Ca and Cb atoms on the cysteine residues from the pairing heavy chain be 5.6 ± 1 Å and 4.0 ± 1 Å. Second, with "-bypass_fragments," "-remodel:match_rt_limit 2," and "-build_disulf" settings in RosettaRemodel, the geometries between the cysteine residues in models from step one were evaluated according to their closeness to known disulfide geometries in PDB. Third, for the structures with proper disulfide bond geometry, we modeled the linker region downstream from the disulfide and the Fc domains, requiring 5 sets of Ca distance pairs derived from a native Fc dimer interface be satisfied. We generated 926 samples from step 1, of which 406 passed step 2, and we built 474 representative Fc models in step 3. After filtering out models with severe clashes based on Rosetta scores, the calculations resulted in 191 final models (Fig. S15). We computed the center of mass for the Fc domains and showed them in Figure S15B to illustrate the range of motion for the Fc domain with respect to CH1/CL via flexible linkers.       S6. Evaluation of nanobodies and bsNb4-Igs for specific binding to Wuhan-1 SARS-CoV-2. Nanobodies-Fc (top two panels) and bsNb4-Igs were tested by ELISA for binding to Wuhan-1 SARS-CoV-2 RBD. CC6.30 was used as a positive control (green) and a nanobody-Fc from our library not selected for RBD binding was used as a negative control (red). We tested the bsNb4-Igs with LM18 either as a HC or LC and observed no notable difference for RBD binding, indicating that bsNb4-Ig building block can be linked indifferently to the CH1 or CL. Error bars indicate the standard deviation of the mean. Fig. S7. Evaluation of SARS-CoV-2 nanobodies-Fc and IgG-like bsNb4-Igs for polyreactivity. Nanobodies-C2 and nanobodies-C2-based bsNb4-Igs (A) and nanobodies-C4 and nanobodies-C4-based bsNb4-Igs (B) were tested by ELISA for binding to CHO-cell soluble membrane protein (SMP) extracts. Bococizumab (CAS: 1407495-02-6) was used as a control (ctrl) to determine nonspecific binding to SMP. We tested the bsNb4-Igs with LM18 either as a HC or LC and observed no notable difference for nonspecific binding.  S8. Epitope binning of Fc-tagged nanobodies using an Octet RED384 platform. His10-tagged RBD was captured using a Ni-NTA biosensor, and indicated monoclonal antibodies (CR3022, CC6.30, CC12.1) or nanobody (LM18), or ACE2 at a concentration of 100 μg/ml were first incubated (*), followed by an incubation with 25 μg/ml of competing nanobody (**). A) LM18; B) Class 2 nanobodies and C) Class 4 nanobodies. The tested nanobody # is indicated on the left of each graph.         Figure 5, a complete bsNb4-Ig can be modeled to assess the plausibility of 1:1, bsNb4-Ig to spike, binding. Fc is highly labile in the sample and has no discrete density in the EM reconstruction, the center-ofmass for ~200 alternative Fc locations are represented as yellow spheres here.  Table S1. IC50 values and potency for neutralization of Wuhan-1 SARS-CoV-2 PSV by four nanobodies-Fc identified by Sanger sequencing. Assays were run in triplicate with a starting nanobody concentration of 100 µg/mL. Table S2. NGS analysis and CDRH3 sequences from competitive sorts with CC12.1, CR3022, and CC6.30. In yellow are indicated the NGS counts for the 6 competition sorts, in grey the 4 sequences that correspond to nanobodies competing with the 3 epitopes that were not taken into account. Table S3. IC50 values and potency for neutralization of Wuhan-1 SARS-CoV-2 PSV by bsNb4-Igs (Class 2 nanobody as HC/LM18 as LC, top and Class 4 nanobody as HC/LM18 as LC, bottom). Assays were run in duplicate to select for the best constructs to be tested with other PSVs. Table S4. IC50 values and potency for neutralization of SARS-CoV-2 PSVs by a selection of Nb-C2/LM18 and Nb-C4/LM18 bsNb4-Igs . No notable difference for neutralization between bsNb4-Igs with LM18 either as a HC or LC could be observed. Assays were run in duplicate. wt =Wuhan-1, NN = non-neutralizing, nd = not determined, n.a = not applicable. Table S5. IC50 values and potency for neutralization of SARS-CoV-2 PSVs by class 2 and 4 nanobody-based bsNb4-Igs. Assays were run in duplicate. wt = Wuhan-1, NN = non neutralizing, n.a = not applicable Table S6. IC50 values and potency for neutralization of SARS-CoV-2 omicron PSV by LM18-and Nb-C2-136based bsNb4-Igs. Monoclonal antibodies CC12.1 and CC6.30, as well as the clinical antibody candidates LY-CoV555, REGN-10933 and REGN-10987 were tested for reference. Assays were run in duplicate. NN: not neutralizing, n.a: not applicable Table S7. IC50 values and potency for neutralization of SARS-CoV-2 PSVs by Nb4-Igs (tetravalent monospecific constructs). Assays were run in duplicate. NN: not neutralizing, n.a: not applicable, nt: not tested. wt = Wuhan-1 Table S8. Summarized results of Wuhan-1 SARS-CoV-2-RBD binding to nanobodies, Fabs and antibodies. Association and dissociation rate constants calculated through a 1:1 Langmuir binding model when possible or heterologous ligand binding model using the BIAevaluation software. Table S9. Summarized results of Wuhan-1 SARS-CoV-2-Spike binding to Fabs or antibodies. Association and dissociation rate constants calculated through a 1:1 Langmuir binding model using the BIAevaluation software. Table S10. Neutralization of SARS-CoV-2 PSVs by matured LM18-based bsNb4-Igs. Assays were run in duplicate. wt = Wuhan-1 Table S11. X-ray crystallography data collection and processing statistics (26). a Numbers in parentheses refer to the highest resolution shell. b Rsym = Σhkl Σi | Ihkl,i -<Ihkl> | / Σhkl Σi Ihkl,i and Rpim = Σhkl (1/(n-1)) 1/2 Σi | Ihkl,i -<Ihkl> | / Σhkl Σi Ihkl,i, where Ihkl,i is the scaled intensity of the i th measurement of reflection h, k, l, <Ihkl> is the average intensity for that reflection, and n is the redundancy. c CC1/2 = Pearson correlation coefficient between two random half datasets. d Rcryst = Σhkl | Fo -Fc | / Σhkl | Fo | x 100, where Fo and Fc are the observed and calculated structure factors, respectively. e Rfree was calculated as for Rcryst, but on a test set comprising 5% of the data excluded from refinement. f From MolProbity. Table S12. Cryo-EM data collection, processing, model refinement and validation statistics. Table S13. Amino acid frequencies used to generate the trimer phosphoramidite mixtures for the construction of the naïve library. Table S14. Primer sequences used in this study.