**3. Discussion**

We here report on the selection of human mAbs specific to the SARS-CoV-2 RBD (SARS-2 RBD) using a human synthetic Fab phage display library. Phage display is a powerful tool that has been used for both discovery and therapeutic applications against various malignancies, including infectious diseases [45–49]. In particular, phage display has been demonstrated to be highly effective for the selection of human antibodies against SARS-CoV-2 in both synthetic and immune phage display libraries built with immune repertoires of memory or plasma B cells from convalescent patients who recovered from the viral infection [27–30]. In our phage display panning, we employed two in-house human synthetic Fab phage display libraries: KFab-I and KFab-II. The KFab-I library was built on VH3 and Vk1 frameworks by randomizing their complementarity-determining regions (CDRs) and yielded ten human anti-SARS-2 RBD mAb clones, whereas no binder was yielded by panning with the KFab-II library, another in-house human synthetic Fab phage display library built on VH1 and Vk1 frameworks. Due to the same CDR randomization design being applied to the two libraries, we reasoned that the framework could have made a difference in the panning outcome. The human VH3 family has been shown to have the highest stability and soluble protein yield, and its germline usage out of 51 germline genes is about 43%, which is considerably higher than that of other families of human VH [43,50]. Indeed, for various antibody libraries, such as the Griffiths and the HuCAL libraries, it has been shown that a considerable number of antibodies selected from the libraries belonged to the human VH3 family (74% and 36% for the Griffiths and HuCAL libraries, respectively) [51,52], indicating that the human VH3 framework may be inevitably favored in the phage display selection due to its desirable properties. Moreover, our previous phage display panning against human YKL-40, which was also performed with the two Fab libraries, showed a similar outcome in that, unlike the KFab-I library, the KFab-II yielded no binders with desirable properties [43]. However, it is believed that a panning with a mixture of both VH3 and VH1 frameworks from KFab-I and KFab-II libraries might result in a different outcome from the previous panning performed separately with each Fab library.

Our study revealed that only one anti-SARS-2 RBD IgG clone (H1) cross-reacted with the S protein of SARS-CoV and the rest of the anti-SARS-2 RBD IgGs reacted specifically with the S protein of SARS-CoV-2, while no anti-SARS-2 RBD IgGs cross-reacted with the

S protein of MERS-CoV, as expected based on the low protein sequence identity between SARS-CoV-2 and MERS-CoV [5,53]. The difference in the cross-reactivity of the antibodies between SARS-CoV and SARS-CoV-2 could be related to whether the epitopes recognized by antibodies are located on regions that are conserved between SARS-CoV-2 and SARS-CoV. The amino acid sequence identity of the RBD (residues 387–516) between SARS-CoV-2 and SARS-CoV is quite high (86.3%), whereas the sequence identity of the receptor-binding motif (RBM; residues 438–505) is substantially lower (46.7%) [8,22,32]. This suggests that our antibodies likely recognize epitopes on the RBM of each virus, not on the conserved regions of the RBD. In addition, the RBM is known to have a loop structure and is thus likely subjected to the conformational variation, which may further reduce the structural homology between SARS-CoV-2 and SARS-CoV [8,22]. In addition, we also observed that all of the anti-SARS-2 RBD IgGs cross-reacted with the SARS-2 RBD variants tested, indicating that the antibody epitopes might not have overlapped with the regions of the RBD in which mutations occurred or that the anti-SARS-2 RBD IgGs might have been tolerable enough to bind to the RBD variants, although the antibody epitopes overlapped with regions where mutations occurred. Due to the limited numbers of the variants tested, it is hard to tell how tolerable our antibodies are to the genetic variations. More information on the antibody epitopes and antibody binding against increased numbers of the RBD variants will surely elucidate this.

Although in general human mAbs targeting the RBD of the S1 subunit have higher neutralizing potencies than those targeting other regions of the S protein, such as the S2 subunit and other regions of the S1 (e.g., the N-terminal domain (NTD)), it is still necessary to combine human mAbs that recognize different neutralizing epitopes due to the emergence of viruses carrying RBD mutations. Indeed, this was nicely demonstrated by the Regeneron antibodies (REGN10933 + REGN10987): an antibody cocktail consisting of these two antibodies, which recognize distinct, non-overlapping epitopes on the RBD, helped to avoid escape mutants after treatment thanks to the unlikely occurrence of simultaneous mutations on two distinct genetic sites [40]. By the same rationale, the neutralizing antibody (4A8 mAb) targeting the NTD of the S1 subunit could also be a good candidate for antibody cocktail therapy [35]. We are currently working to figure out whether our neutralizing antibodies (C2 and D12) can compete with other neutralizing antibodies, such as REGN-COV2, by charactering their antibody epitopes and by using a competitive enzyme-linked immunosorbent assay (ELISA) or bio-layer interferometry (BLI).

In order to identify potential neutralizing antibody candidates out of the ten Fab clones from the panning, we used in vitro competitive assays, namely an ELISA and an ACE2-overexpressed cell-based assay, to enable the Fab clones to compete with either a biotinylated ACE2 or the SARS-2 RBD, respectively. The two assays led to the identification of five Fabs as potential neutralizing antibodies, with the five candidate Fab clones behaving similarly in both assays. When the same assays were performed with the same five candidate antibodies that were reformatted from Fabs to IgGs, the two assays confirmed the five IgG antibodies as competitors, although the clone C12 (IgG) showed slightly less competition, albeit still significant, in both assays, especially in the ACE2-overexpressed cell-based assay. Although the two in vitro assays we adopted were not sensitive enough to discern their subtle differences and the antibodies could therefore not be ranked, it was strongly demonstrated that they could still be useful to handle many clones when screening potential candidates prior to a virus-mediated neutralization assay for either a pseudo-typed or authentic virus.

In the characterization of the antibodies in terms of affinity and neutralization, we also noticed that the affinity of the antibodies seemed to correlate with the neutralization potency. That is, the order of the affinity, C2 > D12 > H1 > F7 > C12, strongly correlated with the order of the neutralization potency, C2 > D12 > H1 > F7 > C12. This observed correlation is strongly supported by previous studies: (1) in a study of the mAb IIB4 recognizing influenza A virus haemagglutinin (HA), a strong positive correlation between its affinity and viral neutralization was found [54]; (2) in a study with potential SARS-

CoV-2 neutralizing antibodies from convalescent human patients, RBD binding and viral neutralization were well correlated [55].This therefore suggests that further maturation of the affinity of the mAbs may somehow enhance their neutralization potency accordingly; studies are underway to explore this. Moreover, the neutralization potencies determined by an in vitro neutralization assay for pseudo-typed and authentic SARS-CoV-2 also correlated with each other: the order of neutralization potency from the pseudo-typed virus, C2 > D12 > H1 > F7 > C12, nicely correlated with the order of potency from the authentic virus, C2 > D12 > F7 > H1 > C12, indicating that a neutralization assay for a pseudo-typed virus can be reliably applied to assess the neutralization potency of clones prior to the authentic virus-based assay, which, unlike the pseudo-typed viral assay, must be done under Biosafety Level 3 (BSL3) conditions. Consistent with the antibody binding against the D614G S1 variant, the order of neutralization potency for the pseudo-typed virus (carrying the D614G S1 variant) remained the same as the order for the pseudo-typed virus (carrying the D614 wildtype S1), with C2 (IgG) and C12 (IgG) showing the highest and the lowest neutralizations, respectively, thus confirming that the antibodies were tolerable to the D614G variation on the S1. This result highlights that current vaccinations relying on the neutralization of antibodies targeting the wildtype S protein of SARS-CoV-2 in vivo may somehow also be effective in coping with the new SARS-CoV-2 variants, including the D614G variant [56].

In conclusion, we selected human anti-SARS-2 RBD mAbs from a human synthetic Fab phage display library. We characterized the resulting Fabs and IgGs in order to observe their desirable biophysical properties, such as their affinity, non-aggregation, and thermal stability. We conducted in vitro assays to assess their neutralizing activities against pseudo-typed and authentic SARS-CoV-2 and identified two clones, C2 and D12, which demonstrated an exceptional ability to block the viral entry into cells. Further refinement of the mAbs should allow for the development of promising anti-SARS-CoV-2 therapeutics, as well as reagents for diagnosis.

#### **4. Materials and Methods**

#### *4.1. A Phage Library Display Panning*

Human synthetic Fab phage display libraries produced in-house (KFab-I and KFab-II, respectively built on human VH3/Vk1 and human VH1/Vk1 germline-based scaffolds, with randomized complementarity-determining regions) were used for the selection of specific binders against a SARS-CoV-2 spike protein (SARS-2 RBD) (Sino Biological, Cat. 40592-V08H, Beijing, China). The SARS-2 RBD was coupled to beads following the protocol for dynabeads (Thermofisher Scientific, Cat. 14301, Waltham, MA, USA). After removing the supernatant on the beads, the coated beads were blocked with 5% skimmed milk (BD, Cat. 232100, Franklin Lakes, NJ, USA) in PBS for 1 h at room temperature. At the same time, the phage library was incubated in 2% skimmed milk in PBS for 1 h at room temperature. The blocked phages were transferred to the beads coated with SARS-2 RBD and incubated for 2 h at room temperature. After separating the beads from the supernatant, the beads bound with phages were washed three times with PBST (PBS containing 0.05% Tween 20) and bound phages were eluted from the beads with 100 mM triethylamine (Sigma-Aldrich, Cat. 90335, St. Louis, MO, USA) for 10 min at room temperature, followed by neutralization with 1 M Tris-HCl (pH 7.4) (Biosesang, Cat. T2016-7.5, Seongnam, Korea). The eluted phages were used to infect *E. coli* TG1 cells at OD600 0.6~0.8. Phage particles were prepared for subsequent rounds of panning by amplification and rescue using VCSM13 helper phages (provided by Dr. Hong from Kangwon National University, Chuncheon, Gangwondo, Korea) according to standard procedures. The amplified phage was used for the next round of panning, and so forth.

#### *4.2. Polyclonal Phage ELISA*

A polyclonal phage ELISA was performed using pools of purified phage from each library stock. A 96-Well Half-Area Microplate (Corning, Cat. 3690, New York, NY, USA) was coated overnight at 4 ◦C, with 30 μL per well of 1 μg/mL SARS-2 RBD (Sino Biological, Cat. 40592-V08H, Beijing, China), and each well was blocked with 5% skimmed milk in PBS (MPBS) for 1 h at room temperature. Phage pools (~10<sup>12</sup> phage particles) were also blocked in MPBS for 1 h at room temperature and then blocked phage pools were added to the SARS-2 RBD-coated plate and incubated for 1 h at 37 ◦C. After washing four times with PBST, the horseradish peroxidase (HRP)-conjugated anti-M13 antibody (1:5000, Sino Biological, Cat. 11973-MM05, Beijing, China) was incubated for 1 h at 37 ◦C. After washing four times with PBST, a TMB substrate solution (Sigma-Aldrich, Cat. T0440, St. Louis, MO, USA) was added for 8 min, and the reaction was stopped with 1 N sulfuric acid (Merck, Cat. 100731, Darmstadt, Germany). The absorbance was measured at 450 nm using a SpectraMax 190 Microplate Reader (Molecular Devices, Sunnydale, CA, USA).
