There are two considerations for using primary human PBMCs for the direct amplification of naturally occurring full length antibody variable heavy-chain (VH) gene repertoires and selection of scaffolds for library construction: (1) to minimize the workload of humanization; and (2) to avoid introducing immunogenic epitopes. We and others have reported that several VH sub-families such as VH1-69, VH3-7, VH3-15, VH3-23, VH4-34 and VH6-1 showed high frequency in the memory compartment of the immune repertoire obtained from healthy donors (4, 19–22). We collected human PBMCs from 10 healthy donors. Full length natural occurring VH genes were amplified by conserved region primers to construct a small size (10^8) VH domain library. Primers were designed according to the alignments from international ImMunoGeneTics information system^® (IMGT, http://www.imgt.org), and are displayed in Table 1 (23, 24). We randomly sequenced 100 clones and no repeats were observed. The dominant sub-families in our VH library ( Figure 1A ) were consistent with the previous reports mentioned above.

We identified relatively stable VH scaffolds from a small VH phage-displayed library (10^8) by a heat denaturation and refolding strategy (4, 25). After 3 rounds of panning, output clones were enriched according to plate titration. We randomly sequenced 100 clones, and VH genes in VH3 family were largely enriched, composing up to 49% of sequenced genes ( Figure 1B ) (4, 18). We then sequenced more clones and evaluated the aggregation of the most enriched clones by size exclusion chromatography (SEC) and dynamic light scattering (DLS). The five most enriched clones (1 clone from VH 3-7, 2 clones from VH 3-30 and 2 clones from VH 4-34) were finally selected for construction of a large size VH domain library.

After selection of aggregation resistant scaffolds, we constructed a large size human VH domain library by grafting naturally occurring CDR2s and CDR3s from human PBMCs (from 50 healthy donors) ( Figure 1C ) (4, 18). Grafting of human CDR3 is an important step, since studies have verified that CDR3 of the VH domain is sufficient to distinguish between a variety of haptens and protein antigens with surprisingly high affinities (26). We optimized the primers for grafting naturally occurring human CDR3 ( Table 2 ), and observed that the optimized set of primers largely increased the diversity of the library constructed. The size of the final VH phage-displayed library constructed (after grafting) was estimated to be 1.3 × 10¹¹, with no repeated sequences found in 100 random sequenced clones, indicating a good diversity. 80% of the clones were productive.

This library was panned against human recombinant Transforming Growth Factor-β1 (TGFβ1), and a panel of binders was obtained with half maximal effective concentration (EC₅₀) ranging from 0.5 nM to 100 nM ( Figure 2A ). We performed 3 rounds of panning, which will be sufficient for enrichment of antigen-specific binders. The yield of both antibody VH domains ranged from 4 to 8 mg, from 100 ml 2x YT medium culture. To test aggregation, we incubated these binders at 3-5 mg/ml at 37°C for 24 hours and 7 days, and measured the aggregation by DLS ( Figure 2C ). 83% of binders showed aggregation resistance (aggregation percent < 5%) after 24 hours incubation while 67% of binders showed aggregation resistance after 7 days incubation. We kept incubating of aggregation resistant binders for another 7 days at 37°C, and all the binders showed durable aggregation resistance. We panned the library against another antigen, human recombinant mesothelin (MSLN) (by 3 rounds of panning), and created another panel of binders showing EC₅₀s ranging from 1 nM to 200 nM ( Figure 2B ), and measured the aggregation of binders by DLS after incubation at 37°C for 24 hours. 75% of binders (9 out of 12) showed aggregation less than 5%. The 9 aggregation resistant binders were incubated for another 13 days at 37°C, and 5 of the candidates showed persistent aggregation resistance, with aggregation accounting for less than 5% of the binder profile ( Figure 2D ).

To date, there are several anti-human PD-L1 drugs approved by FDA (27). Most of them function in blocking the PD1/PD-L1 binding interface. We quickly (only 2 rounds of panning) isolated one binder (nominated 1-16) with specific binding to PD-L1. Epitope mapping demonstrated that binders 1-16, targeted the membrane proximal domain of PD-L1, which is far away from known PD1 binding epitopes ( Figure 3D ). We further modified the CDR3 of 1-16 by grafting and mutagenesis, and acquired one variant of 1-16, nominated 1-16-3, which showed improved binding and aggregation resistance. Surface Plasmon Resonance (SPR) was further performed to measure the equilibrium dissociation constant (K_D) of parental domain 1-16 and variant 1-16-3 to human recombinant PD-L1 ( Table 3 ). After conversion to VH-Fc format to further increase the avidity, 1-16-3 showed an EC₅₀ less than 10nM. PD-L1 positive CHO-K1 cells and PD-L1 negative CHO-K1 cells were used for confirming the cell specific binding for both VH 1-16 and VH 1-16-3 by flow cytometry ( Figure 3A ). The maturation of 1-16 to 1-16-3 did not block the cell specific binding nor shift the epitope since they competed to each other in a competition ELISA ( Figure 3C ). VH 1-16-3 has an improvement of aggregation resistance after incubation at 37°C for both 24 hours and 7 days by DLS ( Figure 3B ). In both VH and VH-Fc format, 1-16-3 shows an increase in binding to PD-L1 ( Figure 3C ). Both 1-16 and 1-16-3 cannot neutralize PD-L1 in a PD-1/PD-L1 Blockade Bioassay (data not shown), indicating the PD-L1 non-neutralization epitope is targeted by 1-16-3 ( Figure 3D ).

CD22, which is overexpressed in B cell malignancies, is a target of great interest. VH-Fc 1-16-3 was conjugated with MMAE for a DDC study exploring killing of CD22 positive leukemia cells. For control purposes, we identified two human VH domains (E1-2 and G10) that specifically bind human recombinant CD22. Cell specific binding of both domains were measured on two different B cell lines (Raji and Bjab) and one T cell line (Jurkat) by flow cytometry. The previously identified IgG1 antibody m971 served as positive control (28, 29). Both E1-2 and G10 can bind to Raji (CD22+) and Bjab (CD22+) cells, but cannot bind to Jurkat cells (CD22-) ( Figure 4A ). We measured the aggregation of both E1-2 and G10 by DLS. G10 aggregated after incubation 37°C for 24 hours ( Figure 4B ). In contrast to G10, VH E1-2 showed less than 5% aggregation after incubation at 37°C for 24 hours by DLS ( Figure 4B ), and an EC₅₀ below 10 nM by ELISA ( Figure 4C ). The equilibrium dissociation constant (KD) of VH domain E1-2 was measured by SPR ( Table 4 ). After conversion to VH-Fc format, the EC₅₀ of VH-Fc E1-2 was reduced, most likely due to the effect of the orientation of VH domains by human Fc protein ( Figure 4D ). We also conjugated VH-Fc E1-2 to MMAE for DDC study. VH-Fc E1-2 showed moderate killing of CD22 positive tumor cells, while VH-Fc 1-16-3 showed approximately 10-fold greater potency in comparison to VH-Fc E1-2. VH-Fc 1-16-3 DDC and VH-Fc E1-2 DDC worked efficiently in combination (with equivalent input) in CD22 positive tumor cells (Raji and Bjab) as well ( Figures 4E –G ). These data indicate that combination of PD-L1 therapy and CD22 therapy may achieve a superior effect to either alone. More extensive studies are required, but we present here an example of the potential applications of antibody VH domains as a therapeutic.

Fresh human normal peripheral blood mononuclear cells (PBMCs) (200 million cells/vial, CAT# SER-PBMC-200) were purchased from Zen-Bio, Inc. NC 27709. Consent and waiver was signed for purchasing purpose. Donor ages are ranging from 20 to 40 years old with unknown gender information. Fresh cells were directly used for mRNA extraction as previously described (4, 18).

Full length human PD-L1 and CD22 sequences were synthesized by IDT (Coralville, Iowa) with sequence obtained from Uniprot (https://www.uniprot.org/uniprot/Q13421). The target sequences were cloned into an expression plasmid containing a CMV promotor, woodchuck posttranscriptional regulatory elements, and a His tag. Proteins were purified by Ni-NTA (GE Healthcare) chromatography. Target specific VH domains were cloned in pComb3x vector and purified from Escherichia coli HB2151 bacterial culture at 30°C in 2x YT medium (Y1003, Sigma-Aldrich) for 16 h with stimulation by 1 mM IPTG. Cells were lysed by Polymyxin B (Sigma-Aldrich). Lysate were spun down and supernatant was loaded over Ni-NTA (GE Healthcare). For conversion to Fc-fusion, the VH gene was re-amplified and re-cloned into pSectaq vector containing human Fc. VH-Fc proteins were expressed in the Expi293 expression system (Thermo Fisher Scientific) and purified with protein A resin (GenScript). Buffer replacement in protein purification used Column PD 10 desalting column (GE Healthcare). All protein purity was estimated as >95% by SDS-PAGE and protein concentration was measured spectrophotometrically (NanoVue, GE Healthcare). Further details can be found in our previous publication (4).

For ELISA assays, antigen protein was coated on a 96-well plate (Costar) at 50 ng/well in PBS overnight at 4°C. For the soluble VH binding assay, horseradish peroxidase (HRP)-conjugated mouse anti-FLAG tag antibody (A8592, Sigma-Aldrich) was used to detect VH binding. For detection of human Fc protein, HRP-goat anti-human IgG Fc secondary antibody (A18817, Thermo Fisher Scientific) was used. For the competition ELISA, 200 nM of VH-Fc 1-16 was incubated with serially diluted VH 1-16-3 proteins, and the mixtures were added to antigen-coated wells. After washing, competition was detected by HRP-goat anti-human IgG Fc secondary antibody (A18817, Thermo Fisher Scientific). All colors were developed by 3,3′,5,5′-tetramethylbenzidine (TMB, Sigma) and stopped by 1 M H₂SO₄ followed by recording absorbance at 450 nm by iMark™ Microplate Absorbance Reader (Bio-Rad Laboratories). Experiments were performed in duplicate and the error bars denote ± 1 SD.

The kinetics of the antibody fragments were determined using a Biacore X100 (GE Healthcare). Human PD-L1/CD22 was (10 mg/mL) was immobilized onto a CM5 sensor chip (GE Healthcare, BR100012) by amine coupling. The antibody fragments diluted in HBS-EP buffer (10 mM HEPES, 150 mM NaCl, 3 mM EDTA, and 0.005% surfactant P20, pH 7.4) were injected over an immobilized surface (200 - 400 RU) for 180 sec at a rate of 20 µL/min, followed by dissociation for 600 sec. After each sample injection, the surface was regenerated by injection of regeneration solution (10 mM Glycine/10% Glycerol pH 2.0). The kinetic values, ka, kd, and KD were calculated using the BiacoreX100 Evaluation Software (GE Healthcare).

A large phage-displayed human VH domain library was constructed by grafting of naturally occurring heavy-chain CDR2 s and CDR3 s (from PBMC pool of >30 healthy donors) to VH framework scaffolds. CDR2 s and CDR3 s were PCR amplified with the primers in Table 2 . After library construction, 100 dandom clones were sequenced and no repeats were observe, indicating a good diversity of the library. The library was preabsorbed on streptavidin-M280-Dynabeads in phosphate-buffered saline (PBS) for 1 h at room temperature (RT) and incubated with 50 nM biotinylated PD-L1 or CD22 for 2 h at RT with gentle agitation. Phage particles binding to biotinylated antigen were separated from the phage library using streptavidin-M280-Dynabeads and a magnetic separator (Dynal). After washing 20 times with 1 ml of PBS containing 0.1% Tween-20 and another 20 times with 1 ml of PBS, bound phage particles were eluted by 100 mM triethanolamine followed by neutralization with 1 M, pH7.5 Tris-HCl. For the second round of panning, 10 nM (2 nM for the third round) of biotinylated protein was used as antigen. After the third round of panning, 96 individual clones were screened for binding to target protein by phage ELISA. Panels of VHs were selected, sequenced and characterized.

PD-L1 positive aAPC/CHO-K1 (Cat # J1205) cells and PD-L1 negative aAPC/CHO-K1 (Cat # J1191) were purchased from Promega. Bjab cells were purchased from (ABC-TC514S, AcceGen, New Jersey, USA). Raji cells (Cat# CCL-86) and Jurkat cells (TIB-152) were purchased from American Type Culture Collection (ATCC). Both T cells and B cells were maintained in ATCC-formulated RPMI-1640 Medium (ATCC 30-2001) with 10% fetal bovine serum (ATCC 30-2020). To confirm the cell surface binding of isolated antibody binders, cells were treated with VH 1-16, VH 1-16-3, E1-2, G10 or m971 for 1 h at 4°C and then stained with anti-DYKDDDDK Antibody, FITC (130-127-934, Miltenyi Biotec), or anti-DYKDDDDK (flag tag) Antibody, PE(130-101-577, Miltenyi Biotec), or PE anti-human IgG Fc Antibody (410707, BioLegend) for 0.5 h at 4°C. After each incubation with antibodies, cells were washed with cold PBS. Data were acquired using the flow cytometry BD LSR II (San Jose, CA) and analyzed by FlowJo 10.7.1.

Monomethyl auristatin E (MMAE) was conjugated to VH-Fc via the cross-linker OSu-Glu-VC-PAB (SET0100, Levena Biopharma, USA) with a molar ratio mAb : OSu-Glu-VC-PAB-MMAE of 1:10. The conjugation was performed in buffer composed of 50 mM potassium phosphate, 50 mM sodium chloride, and 2 mM EDTA (pH 6.5) with the reaction run for 18-24 h at room temperature. The reaction was stopped by adding 50 mM sodium succinate (pH 5.0) followed by buffer replacement by using Column PD 10 desalting column (GE Healthcare). The protein was washed and concentrated using 30 kDa Amicon centrifugal filter unit (Millipore Sigma CAT#UFC8030).

Cell viability was measured using CellTiter-Glo or LDH-Glo (G7570, J2380, Promega). Briefly, CD22 positive or negative cells were plated into 96-wells, allowing attachment and growth for 24 hr, then triplicate wells were treated with DDCs, naked antibodies, free drugs, or DDCs plus competitor antibodies. Three to five days later, when untreated control wells were 70 to 90% confluent, reagent was added to the plates according to the supplier’s instructions. Wells treated identically but wells without cells were used to subtract background. Fluorescence (ex: 570 nm, Em: 585 nm) was measured using a CLARIOstar microplate reader (BMG Labtech) and data analyzed using GraphPad Prism 8.0.1 software. Significance was tested using one-way ANOVA, followed by the Tukey’s multiple post hoc test. ****P < 0.0001; ***P < 0.001; **P < 0.01; *P < 0.05 between the indicated groups. PD-L1 neutralization was done in a PD-1/PD-L1 Blockade Bioassay (J1250, Promega).

VH domains were buffer-changed to DPBS and filtered through a 0.22 μm filter. The concentration was adjusted to 5 mg/mL; 500 μL samples were incubated at 37°C. DLS measurement was performed on Zetasizer Nano ZS ZEN3600 (Malvern Instruments Limited, Westborough, MA) to determine the size distributions of protein particles.

Epitope mapping was performed by PEPperPRINT (Heidelberg, Germany). Briefly, a conformational PD-L1 peptide microarray was pre-stained with the secondary and control antibodies to investigate background interactions with the antigen-derived peptides that could interfere with the main assays. Subsequent incubation of conformational mesothelin peptide microarrays with human VH-Fc antibodies 1-16 at concentrations of 1 µg/ml, 10 µg/ml and 100 µg/ml in incubation buffer was followed by staining with the secondary and control antibodies. Read-out was performed with an Innopsys InnoScan 710-IR Microarray Scanner at scanning gains of 50/40 (red/green). The additional HA control peptides framing the peptide microarrays were simultaneously stained with the control antibody as internal quality control to confirm assay performance and peptide microarray integrity.

PD-L1 structure was extracted from the PD1/PD-L1 complex (PDB ID:5IUS). VH1-structure was modeled by Swiss-model (46) followed by energy minimization. Z-DOCK program (https://zdock.umassmed.edu/) is used for docking VH1-16 onto PD-L1 (47). The antibody binding and blocking regions on PD-L1 were selected based on the experimental epitope mapping results. Z-DOCK outputed top 10 optimal poses, which are visually scrutinized for compatibility at the interaction interface and are avoiding of side chain clashes. We choose the most favorable pose as the binding model for further analysis. The structural figures are prepared by PyMol (48).