Selection of antibodies for combinational use

In our previous study [21], we identified a broad-spectrum neutralizing antibody 2G1 that targets the tip of RBD through a small contact surface from our SARS-CoV-2 antibody repertoire, which makes it less possible in epitope clash with other antibodies and thus suitable for combinational use. We performed competition ELISA to map the epitope information of antibodies. As expected, all of the tested antibodies competed with themselves (Fig. 1a). Antibodies in the repertoire could be roughly divided into three groups according to their competition relation, of which antibody 2G1 was in the group 3.

Fig. 1: Selection of antibody cocktails from the SARS-CoV-2 neutralizing antibody repertoire.

a Epitope mapping of the antibody repertoire using a competition ELISA. A heatmap was used to show the competition percentages between two antibodies. The epitopes of antibodies were roughly classified into three groups according to their competitive relations. b Binding to SARS-CoV-1 S protein by antibodies in the repertoire. Trimeric SARS-CoV-1 S protein was coated on 96-well ELISA plates and incubated with 0.1 μg/mL of neutralizing antibodies. The binding was detected using a HRP-labeled goat anti-human IgG (Fc specific) antibody. c Concentration-dependent neutralization of SARS-CoV-1 pseudovirus by neutralizing antibodies. Serial diluted antibodies were incubated with the virus and used to infect 293T-ACE2 cells. The infection was quantified using a fluorescence quantification kit. Data from two replicates are shown as mean ± S.D. d Antibodies competitively blocked the ACE2 binding to SARS-CoV-2 S trimer as measured by ELISA. Recombinant human ACE2 protein and phosphate buffer solution (PBS) were used as controls.

Then, we examined whether these antibodies could cross bind to SARS-CoV-1 as the ability the binding to the Sarbecovirus could be a reflection of conserved activity by an anti-SARS-CoV-2 antibody. As shown in Fig. 1b, all antibodies in group 1 and 2, but not in group 3, efficiently bound to SARS-CoV-1 S protein at the concentration of 0.1 μg/mL. We then performed pseudo-typed SARS-CoV-1 virus neutralization (Fig. 1c). Antibodies 7G10, 8G4, 9E12, 9D11, 8C12 and 9A6 showed high neutralizing activity to SARS-CoV-1, while most of them were in epitope group 1. Subsequently, we explored the ACE2 competition profile of these antibodies and found that antibodies in group 1 and 3 showed excellent ACE2 competition while the group 2 antibodies had little or negligeable level of the competition (Fig. 1d).

These results revealed certain regularities. Antibodies in group 1 were able to bind to SARS-CoV-1 spike and neutralized the virus by blocking ACE2 interaction. The group 2 antibodies targeted non-ACE2 epitope in the spike of SARS-CoV-1 but were incapable of neutralizing the virus. The group 3 antibodies only showed SARS-CoV-2 binding and neutralizing ability, instead of SARS-CoV-1. To minimize the possibility of epitope clash, we intended to select antibodies from the group 1and 2 to make up combinations with 2G1. Finally, antibodies 9E12, 9D11, 8G4 and 11F7 from group 1 and antibodies 10D4 and 7B9 from group 2 were selected. Then, we used two formulating strategies, one was two monoclonal antibodies paired with 2G1, another was a bispecific antibody paired with 2G1. We designed bispecific antibodies 7B9-9D11, 10D4-8G4 and 10D4-11F7 using the BAPTS platform [24]. Eventually, four antibody cocktails were generated, including 9E12 + 10D4 + 2G1, 7B9-9D11 + 2G1, 10D4-8G4 + 2G1 and 10D4-11F7 + 2G1.

Testing the tolerance of antibody cocktails by long-term virus passaging

We then used a VSV-SARS-CoV-2 system that can generate random mutations during replication to investigate the antimutagenic potential of the four cocktails (Fig. 2a). The virus in this investigation experienced over 30 passages, and the entire 30 passaging cycles costed more than 60 days, which offered the virus an opportunity generating sufficient mutations in adequate time. The VSV-SARS-CoV-2 broke through the protection of monotherapy by 9E12, 10D4 and 2G1 at the 9th, 1st and 4th passages, respectively (Fig. 2b). The emergency use authorization (EUA) antibody REGN10933, which has a similar epitope with 2G1, was used as a control [21]. The virus broke through REGN10933 at the 3rd passage. As for the three bispecific antibodies, 10D4-8G4 lost activity at the 4th passage, 10D4-11F7 lost neutralization at the 1st passage. By contrast, 7B9-9D11 showed outstanding tolerance and was not broken through even at the designed endpoint of the 30th passage, despite the reduced neutralization.

Fig. 2: Long-term monitoring of the antibody cocktails protects cells from the infection of vesicular stomatitis viruses with S protein replacement (VSV-SARS-CoV-2).

a The passaging scheme of the long-term protection of antibody cocktails in a Vero E6 cell model. b Antibody cocktails were fivefold serially diluted from 100 μg/mL and mixed with VSV-SARS-CoV-2 before infecting Vero E6 cells. After 48 h, the cytopathic effect in each well was examined and medium from the highest concentration wells that the viruses showed positive infection was collected for the next round of passage. An authorized antibody REGN10933 was used as control. c Comparison of the replication rates of VSV-SARS-CoV-2 virus that broke through the protection of antibody cocktail. The replication rates were calculated according to the mean fluorescence intensity measured by flow cytometry. d Simultaneous binding of antibodies to WA1/2020 S trimer as measured by surface plasmon resonance. The simultaneous binding was confirmed by the accumulation of response values from different rounds of flow.

Combinational therapy by 9E12 + 10D4 significantly enhanced the anti-mutation ability and prevented the infection of the virus until the 29th passage (Fig. 2b). The triple combination by 9E12 + 10D4 + 2G1 showed more stronger resistance to viral mutation and had not been broken through even at the 33rd passage. The 7B9-9D11 + 2G1 cocktail showed not only outstanding tolerance, but also good neutralizing activity. The results indicated that the antibody cocktails we designed improved tolerance to virus from monotherapy substantially. However, not all combinations were promising in dealing with mutational escape. Combinations by 10D4-8G4 + 2G1 and 10D4-11F7 + 2G1 increased resistance to mutations merely, and were broken through at 5th and 7th, respectively.

The replication rate of a virus is in correlation with its transmissibility. In view of the good tolerance of combinations by 9E12 + 10D4 + 2G1 and 7B9-9D11 + 2G1, we collected the passaged viruses and measured their replication rates (Fig. 2c). The passaged viruses generally had lower growth rates than the WA1/2020 control. The slope of the replication curve of viruses from 9E12 + 10D4 + 2G1 and 7B9-9D11 + 2G1 were 0.120 (versus 0.220 in WA1/2020 control) and 0.148 (versus 0.193 in WA1/2020 control). Notably, although the virus broke through the protection of monotherapy by 2G1 quickly, the replication rate of the mutant virus was relatively slower (slope = 0.069 and 0.082). The slowed replication rates of these viruses might suggest less possibility of pandemic in the real world.

Subsequently, we verified whether antibodies in 9E12 + 10D4 + 2G1 and 7B9-9D11 + 2G1 cocktails could bind to S protein simultaneously using surface plasmon resonance (SPR). When 9E12, 10D4 and 2G1, or 7B9-9D11 and 2G1 flowed through the antigen-coated chip sequentially, the response signal increased accordingly, while that in the control channels remained unchanged, indicating that antibodies in both of the two cocktails could bind to S protein simultaneously (Fig. 2d).

ACE2 susceptibility of passaged VSV-SARS-CoV-2 viruses

The binding of S protein to ACE2 on the cell surface is a key step of virus invasion. Higher the affinity of S protein to ACE2 means the easier for the virus infecting cells. Meanwhile, the higher ACE2 affinity makes the virus easier to be neutralized by recombinant ACE2 protein in the medium. Thus, the susceptibility of viruses to recombinant ACE2 protein can be used to evaluate the difficulty of viruses invading cells. To further verify the reliability of the anti-mutational potential of the VSV-SARS-CoV-2 system in evaluating antibody cocktails, we examined the ACE2 susceptibility for 9E12 + 10D4 + 2G1 and 7B9-9D11 + 2G1. As shown in Fig. 3a and b, none of the mutant viruses had increased ACE2 susceptibility compared with the WA1/2020 control, indicating that mutations in these viruses did not increase the affinity to ACE2 and may not lead to an increased risk of transmission. Specifically, viruses from pressurized passaging by 9E12 + 10D4, 7B9-9D11 and 7B9-9D11 + 2G1 showed more than ninefold decreases in ACE2 sensitivity, and the 9E12 + 10D4 virus decreased by more than 27-fold. The reduced ACE2 susceptibility is correlated with the slowed cell invasion; this should be a contributive factor in the reduction of virus replication rates that observed in Fig. 2c. Reduced ACE2 susceptibility may result in degraded viral transmissibility without presenting an epidemic risk. Therefore, even though the mutant viruses escaped or weakened the protection of the two antibody cocktails, transmission of these viruses may not occur in nature due to the reduced transmissibility.

Fig. 3: Comparison of ACE2 susceptibility of VSV-SARS-CoV-2 viruses that broke through the protection of antibody cocktails.

a–b After the viruses broke through the protection of 9E12 + 10D4 + 2G1 (a) and 7B9-9D11 + 2G1 (b), or reached the designed endpoint, the multi-passaged viruses were mixed with gradient diluted recombinant ACE2 protein and were used to infect Vero E6 cells. The cytopathic effect was observed and the antigenic affinity shift of the viral S protein was assessed according to the neutralizing alteration to recombinant ACE2.

Deep sequencing identifying key viral escape sites

To understand how these two cocktails protect cells from mutant virus infection, we performed next generation sequencing to identify mutation sites in the S protein of the VSV-SARS-CoV-2 variants. Sole pressurized passaging by 9E12 generated three high proportion mutations, i.e., G485R, V503E and V615M (Fig. 4a). While both antibodies 10D4 and 2G1 generated only one mutation site, K529N and F486V, respectively. The more mutations the virus generates may mean that the antibody is more difficult to be broken through, which could explain why the virus required more passages to break through the protection of 9E12 than 10D4 or 2G1 (Fig. 2b). In the antibody cocktail by 9E12 + 10D4, the virus generated six mutation sites in S protein and two of them were in RBD region. Correspondingly, the virus did not break through 9E12 + 10D4 until the 29th passage (Fig. 2b), suggesting that 9E12 + 10D4 might be a relative tolerant combination that is difficult to be broken through. The virus in the 2 μg/mL wells of 9E12 + 10D4 + 2G1 developed three mutations including H69R, G485R, and R683G. Although the three mutations did not completely break through the protection of 9E12 + 10D4 + 2G1, it weakened the neutralizing efficacy moderately. In terms of the 7B9-9D11 + 2G1 cocktail, four and five mutation sites came about from the 7B9-9D11 and 7B9-9D11 + 2G1 viruses respectively to weaken the antibody activity (Fig. 4b).

Fig. 4: Deep sequencing of VSV-SARS-CoV-2 identifies key escape mutational sites for viruses from pressurized culture in antibody cocktails.

a–b Virus-containing medium from pressurized culture in 9E12+10D4+2G1 (a) and 7B9-9D11+2G1 (b) was collected and the next generation sequencing was performed to identify nucleotide mutations in the spike gene. The single nucleotide polymorphism (SNP) was analyzed with sufficient sequencing depth coverage to ensure the reliability. The RBD region is colored in light green and mutation sites are colored in red.

It is worth noting that the mutation sites of the viruses resistant to cocktails were not the addition of mutations from that of single antibodies. This may imply that more than one scheme for the virus to escape and be involved in the overall conformational change of the S protein. Besides, most of these mutations are not found in VOC strains. This suggested that even if an antibody neutralizes most of the current variants, there might be still unknown mutations that can escape. Therefore, the anti-mutational potential of antibody combinations designed based on the current variants still needs to be explored reliably.

All these antibodies in our repertoire recognize the RBD region; however, there occurred non-RBD mutations. Studies suggested that mutations in the non-epitope region likely represent tissue culture adaptations [27, 28]. We also infer that these mutations may together change the S protein in conformation and thus reduce antibody binding.

Verifying the breadth of cocktails picked from VSV-SARS-CoV-2 passaging

Using a replication incompetent pseudovirus neutralizing system, we tested the activity of both 9E12 + 10D4 + 2G1 and 7B9-9D11 + 2G1 cocktails against all current VOC and VOI strains, as well as the SARS-CoV-1. As for monoclonal antibodies, 9E12 showed good neutralization against variants except for Omicron (BA.1); 10D4 inhibited the infection of most variants, especially the Omicron (half maximal inhibitory concentration IC50 = 1.4180 μg/mL), but its neutralization efficiency was relatively low, the IC50 values were generally over 0.1 μg/mL; 2G1 ultra-potently neutralized most variants, but had minimized activity against Omicron and had no activity against SARS-CoV-1 (Fig. 5a). By contrast, the combinations of 9E12 + 10D4 and 10D4 + 2G1 not only neutralized Omicron, but also improved the efficiency against other variants. The 9E12 + 2G1 presented both high efficiency and the ability to neutralize SARS-CoV-1. Furthermore, the combination by 9E12 + 10D4 + 2G1 showed good neutralization to all variants, which was not achieved by monoclonal antibodies. Similar results were also observed on the 7B9-9D11 + 2G1 cocktail (Fig. 5b). Antibody 2G1 increased the anti-viral efficiency and bispecific antibody 7B9-9D11 improved the anti-viral breadth.

Fig. 5: Evaluation of broad neutralizing ability of antibody cocktails in pseudo-typed SARS-CoV-2 and SARS-CoV-1 virus models.

a–b Antibody cocktails 9E12+10D4+2G1 (a) and 7B9-9D11+2G1 (b) were tenfold serially diluted from 101 μg/mL to 10−5 μg/mL and incubated with pseudo-typed viruses for 30 min, and then the mixture was added into a plate seeded with 293T-ACE2 cells. The infection of viruses to cells was quantified 48 h later.