One silver lining to the emergence of the COVID-19 pandemic has been the intense focus of numerous institutions and scientific actors to solving problems related to the pandemic. This has provided a unique global opportunity to directly compare different technologies applied to the same problem. Perhaps most striking are the vaccine development efforts. Presently, 86 vaccines against 18 different organisms (including strains) are approved for use in the US (https://www.fda.gov/vaccines-blood-biologics/vaccines/vaccines-licensed-use-united-states). However, for only a few (e.g., Herpes Zoster, Typhoid, Influenza) is more than one type of vaccine available (e.g. attenuated virus and recombinant protein); for most, there is just a single vaccine type, reflecting years of research. The advent of COVID-19 completely upended this slow development paradigm. In just over a year, 354 vaccines (194 preclinical, 135 clinical, and 25 approved), comprising eight different vaccine classes (protein subunit, replicating and nonviral vectors, DNA, RNA, live and attenuated virus, and virus-like particles) have been developed against this single organism (https://www.who.int/publications/m/item/draft-landscape-of-covid-19-candidate-vaccines). Although the outcome of this vaccine comparison is still pending, complicated by the challenge of analyzing the results of immunizing large cohorts in which infection rates and viral variants are a moving target, the RNA vaccines were the fastest to approval, and so far, appear to have the fewest side-effects. Another area of intense focus has been in the sphere of diagnostics, with over 500 different commercial tests available based on four main technologies (PCR, isothermal amplification and antigen or antibody detection) (https://www.360dx.com/coronavirus-test-tracker-launched-covid-19-tests), each of which has different advantages and applications. For example, PCR tests are the most sensitive, but may be too sensitive if positive in the absence of infectious virions.
The comparison of different methods to generate antibodies against the spike protein of SARS-CoV-2, and in particular, its RBD domain8,9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,25,26,27,28,29,30,31,48,49,50,51,52,53,54,55,56, may be more straightforward, in that in vitro antibody properties can be compared to one another using relatively few parameters: affinity and biological activity, which in the case of anti-spike antibodies can be represented by IC50, and the neutralization of variants of concern. Naïve antibody display libraries were first described in 1991, 30 years ago24,57, with the promise they would replace immunization as the means by which antibodies would be generated in the future. Although many different libraries have been described (e.g.58,59,60,61,62,63,64,65,66,67), and antibodies from some have entered the clinic and gained approval, phage antibodies have been described as having less favorable developability properties45 and lower affinities, and therefore routinely require subsequent maturation to develop utility68. However, it has been challenging to compare the properties of antibodies derived from different naïve or immune platforms, since they are frequently tested on different targets and are isolated by disparate approaches. The COVID-19 pandemic has provided the unique opportunity to carry out such a direct comparison, with multiple groups world-wide applying their platforms to generating antibodies against a single discrete target: the ~220 amino acids of the CoV-2 Spike protein receptor-binding domain (RBD). Most studies have used B-cells from immune sources, particularly convalescent COVID-19 patients9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,50,69,70,71,72,73,74,75,76,77,78,79, yielding many antibodies with potent neutralization properties, including some that have received Emergency Use Authorization. However, all RBD antibodies derived from immune sources are not equally potent, with best affinities spanning over four orders of magnitude, and only seven publications reporting best affinities <100 pM (Fig. 5)10,12,21,48,72,76,80, and a couple of exceptional antibodies with single-digit pM affinity7,76. In the case of naïve libraries, there is a three order of magnitude span of best reported affinities25,26,27,28,29,30,31,32,80, with only one publication80 reporting a single antibody with an affinity below 100 pM. IC50s for live virus neutralization show a similar trend, with the eleven best published immune antibodies having IC50s < 10 ng/ml13,15,17,18,21,48,50,71,72,78,81, and no published naïve best antibody having a live virus IC50 < 10 ng/ml.
a For each of the individual publications (see introduction) the IgG, Fab, or scFv-Fc with the highest affinity for the RBD is indicated (derived either from recombinant naïve antibody libraries “Library derived Abs” or from immunized animals or recovered Covid19 patients “Immune source Abs”), together with the affinities of 181 antibodies in the CoVIC Ab database recognzing the RBD (“CoVIC Abs”). For the antibodies selected here (“Selected Abs”), the ten IgGs with the highest affinities for the RBD are indicated. For three of the naïve library derived antibodies we showed the reported values and those obtained here (“Meas. Affinity”). b For each of the individual publications the IgGs with the lowest IC50’s obtained using pseudovirus are indicated, while for the CoVIC database the IC50’s of 306 reported antibodies are indicated For the antibodies selected here, the 10 IgGs with the lowest pseudovirus IC50s are indicated. c For each of the individual publications the IgGs with the lowest IC50’s obtained using live virus are indicated, while for the CoVIC database the IC50’s of 288 antibodies assessed by UTMB are indicated (76 antibodies had IC50’s > 25 µg/ml and are not included). For the antibodies selected here, the ten IgGs with the lowest livevirus IC50s are shown. For the publications as well as the antibodies selected here, the top antibody in each individual category is not necessarily the same antibody across all categories (e.g., a high-affinity antibody with poor IC50’s is reported in 5a, but not 5b or 5c). Furthermore, affinity, PSV, and live virus IC50s were not measured in all publications, so only the best measurements in each category that were reported are indicated. Source data are provided as a Source Data File.
Here, within the context of this international antibody comparison, we show that a semi-synthetic antibody library, relying on whole natural CDRs for diversity, can rapidly generate many antibodies with affinities and neutralization activities against CoV-2 that are superior to those previously reported from naïve libraries and comparable to the best from immune convalescent patients. Figure 5 compares the affinities and IC50s for the ten most potent antibodies selected here (in each category) of the 23 tested, against the single most potent antibody from every published report9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,25,26,27,28 in the same categories. As can be seen, the range of antibodies described here is generally comparable to, if not better than, the published record of the best antibodies generated by immunization from most publications, with the highest affinity antibody described here (34 pM) having a higher affinity than 79% of the best immune antibodies described, and all the antibodies from naïve libraries. Notably, of the 23 antibodies we tested, 19 possessed subnanomolar affinities, with two having binding affinities below 100 pM. One caveat to this comparison is that affinities in different publications were measured using different approaches in different laboratories (and we found some affinities to be lower when remeasured—see Fig. 5). Fortunately, a recent publication82 from the Coronavirus Immunotherapy Consortium (CoVIC) compared over 350 mAbs from 56 different labs for affinity and neutralization activity. The antibodies were derived from COVID-19 survivors, phage display, naïve libraries, in silico methods and other strategies, but were submitted blinded to CoVIC, making it impossible to define the origins (naïve or immune) of individual ones. Affinities were assessed using a single technique (surface plasmon resonance) in a single lab, on a single platform (Carterra LSA), which is the same as that used here. 181 of these antibodies recognized the RBD, and comparing their affinities to those described here, shows that the latter have higher affinities than the vast majority of those tested by CoVIC when measured in the same way on the same equipment. Eight of ten tested antibodies had IC50s < 10 ng/ml in live virus neutralization assays (and 13 of 23 in pseudoviral assays), including four with IC50s < 2 ng/ml and one with an IC50 of 1.3 ng/ml, values significantly better than all the best naïve antibodies, most of the best immune antibodies and all but a few of the CoVIC antibodies.
Although our selections were conducted using the entire S1 domain of SARS-CoV-2, all the high-affinity antibodies we describe here recognized targets in the RBD-A region. This may reflect either a higher immunogenicity or increased stability and folding of the RBD compared to the rest of S1, which may make in vitro antibody selection more effective.
The selected antibodies were compared with a series of previously characterized antibodies. Five were obtained from convalescent patients, two chosen for having high neutralization activities (CC6.29 and CC6.30), two because they bound a different epitopic region (CC12.17 and CC12.18)18, and one that was cross-reactive with SARS-CoV-1 and recognized an additional, epitope39. We also compared four antibodies derived from in vitro “naïve” phage library selections (naïve antibodies) with the highest reported affinities for which sequences were available39,40,41. The affinities of the antibodies from convalescent patients, as produced and measured here, were similar to those previously reported7,18. This was not the case for the antibodies from the naïve libraries, which had affinities significantly lower than those previously reported39,40,41, a finding we are unable to explain. For all but a few antibodies, values improved ~sixfold when RBD affinities were compared to a trimeric form of the Spike antigen (Fig. 2), reflecting the expected avidity effect due to multivalent binding. Interestingly, some antibodies (e.g., C7, G5, and A12) showed lower or similar avidities for the trimer compared to the monomeric RBD, which may be related to different trimer conformations83, allosteric effects (e.g., masking of recognized epitopes in the native trimer), or the fact that the soluble S trimers we used contained an artificial trimerization sequence, which may not conform with that of the natural membrane-bound trimer. It is intriguing that the affinity of the most potent neutralizing antibody (A12 – IC50 1.3 ng/ml) falls within this group, with affinities roughly equivalent for the RBD (52 pM) and trimer (83 pM).
The great majority of the selected antibodies binned together, targeting the RBD-A epitope, and overlapping with sites recognized by CC6.29/30 and most other neutralizing immune antibodies. Most of the tested antibodies were able to inhibit the B.1.17 and B.1.351 variants with IC50s similar to those of the wild type (Fig. 4d), indicating their potential utility in antibody cocktails designed to control variants and avoid escape mutants. Two of the selected antibodies (#41 and B3) belonged to a separate binning group, able to block binding to both RBD-A (CC6.29 and CC6.30) and one of the RBD-B antibodies (CC12.18)18, but these possessed reduced efficacy against the B.1.17 and B.1.351 variants (Fig. 4d).
In conclusion, the studies enabled by this global pandemic comparison show that carefully designed and constructed recombinant naïve antibody libraries are as efficient as immune sources for the generation of highly potent antibodies against a viral pathogen. This establishes that naïve antibody libraries have finally come of age, fulfilling their original promise of bypassing the need to identify immune sources with strong responses to generate powerful antibodies24. This library design has shown value not just for infectious disease, but also for therapeutic antibody discovery in general, having produced antibodies with comparable affinities and developability properties against other therapeutic targets33,34. The absence of developability concerns in the selected antibodies is manifested by the presence of few identifiable sequence liabilities (Supplementary Fig. 2), mean and median hydropathy and charge profiles for the merged CDRs better than the anti-Covid antibodies in the Cov-AbDab database (Supplementary Fig. 3), a lack of polyreactivity (Supplementary Fig. 4) and values for AC-SINS, Tm and accelerated stability considered developable within the clinical-stage antibody landscape45. These features are expected to accelerate scale up production and clinical approval of antibodies selected from this platform.
