Monoclonal antibody levels and protection from COVID-19

Here we demonstrate a relationship between the monoclonal antibody concentration and efficacy in preventing COVID-19. Further, we estimate the concentration of antibody required to have a high confidence of maintaining at least 50% protection. Our model fitting enabled us to quantify the uncertainty in this relationship and estimate that if a treated population can maintain a mean in vivo monoclonal antibody concentration of >96-fold (95% CI: 32–285) of the in vitro IC50 of the antibody to the circulating variant, they should maintain >50% efficacy against COVID-19. Analysis of the dose-response curve for monoclonal antibodies allows prediction of the level and duration of protection against different SARS-CoV-2 variants (Fig. 3). Our results suggest that casirivimab/imdevimab, cilgavimab/tixagevimab, and adintrevimab would provide >50% protection against the ancestral SARS-CoV-2 strain for 586, 581, and 548 days respectively. Unfortunately, this analysis also predicts that these mAb prophylaxis regimens will have lost protection against most of the more recent circulating variants because the change in IC50 to the variants is considerably greater than the thresholds determined here (Fig. 3d, Table S3). This supports the decisions of regulators who have now withdrawn recommendations for all mAbs for use as prophylaxis²⁷. We suggest that the analysis presented here will provide a tool for ongoing development of broadly neutralizing monoclonal antibodies, allowing in vitro data to be used to inform dosing choices of candidates in trials.

Counterintuitively, although antibodies with a longer half-life are expected in general to provide protection for longer, these are also expected to lose more ‘days of protection’ for a given fold increase in IC50 (to a new variant), compared to mAbs with shorter half-lives (Fig. 3). The higher susceptibility of therapeutics with longer half-lives to fold-shifts in the IC50 has been discussed previously for antimalarial products²⁸, and can be explained by considering that when antibodies lose 2-fold neutralization against a new variant it is equivalent to the mAb losing one half-life of time above a threshold. Therefore, for a 2-fold increase in IC50 a mAb with a 100-day half-life will lose 100 days above a specified threshold, whereas a mAb with only a 30-day half-life will lose 30 days above the same threshold.

The estimated in vivo concentration of antibody required for 50% protection from COVID-19 is much higher than the level of antibody required to neutralize virus in vitro (~100-fold), suggesting that in vivo neutralization may be much less efficient than the observed neutralization in vitro. This difference between in vitro IC50 and the in vivo 50% protective titer is not unexpected, given the major differences between infection in these environments. For example, in vitro neutralization assays usually involve pre-incubation of antibody and virus for an hour before exposure to cells. Similarly, the in vitro IC50 in plaque reduction assays estimates the antibody concentration required to neutralize 50% of virions. However, the dose required to completely neutralize large inocula may be considerably higher²⁹. In addition, in vivo antibody titers are assessed in the serum. However, antibody concentration at the mucosa is lower than the plasma level³⁰, and thus higher (serum) titers may be required to achieve neutralization on mucosal surfaces.

We and others have previously shown that neutralizing antibodies are a correlate of protection from COVID-19^{14,21,23,31,32}. A major question in understanding vaccine-mediated immunity is whether neutralizing antibodies are simply a surrogate marker of protection or are mechanistic in protecting individuals from symptomatic infection¹⁴. To-date it has only been possible to consider this question indirectly. For example, we have noted that the drop in neutralizing antibodies against new variants and over time both provide good predictions of the change in efficacy of vaccines over time and against new variants^1,20. Schmidt et al. recently showed that the monoclonal antibody adintrevimab lost neutralization to Omicron BA.1 and BA.1.1, and a corresponding loss of prophylactic efficacy compared with that predicted by the relationship between neutralization titer and vaccine effectiveness studies¹³. Here, we extracted data on antibody PK and temporal changes in efficacy from five monoclonal antibody studies (including from Schmidt et al.¹³), and compared this with an established immune correlate of protection after vaccination that has been validated across a number of settings^{1,14,20,33,34}. This allows a direct comparison of the protection provided by passively administered monoclonal antibodies versus vaccine-induced polyclonal antibodies. We have been able to show that the neutralization titer required for 50% protection by vaccination or monoclonal antibodies is comparable (Table S9), although a predicted trend towards higher efficacy of vaccines compared to mAb at high neutralizing antibody titers was observed (Fig. 4b and Figure S5). Importantly, the power to detect a difference in these comparisons is limited by the data available.

The difference in protection at a given neutralization titer between vaccination and monoclonal antibody therapy may be due to the additional benefit in vaccinees of a polyclonal antibody response, other non-neutralizing functions of antibodies, recall of immune memory, and/or other cellular immune responses. In particular, adintrevimab, tixagevimab, and cilgavimab all have modified Fc domains to increase the antibody half-life^13,35, and tixagevimab and cilgavimab have additional mutations to reduce Fc-related antibody functions. These functions may contribute to the estimated trend towards higher protection for vaccines at high neutralizing titers. While our analysis has shown that neutralizing antibodies alone are sufficient to provide a high level of protection from COVID-19 at the neutralization titers induced by vaccination, it is not possible to conclude from this analysis that neutralizing antibodies are necessary for protection. Also, since adintrevimab, casirivimab, and imdevimab have seemingly intact Fc receptor functions, differences in efficacy between these products may arise from other non-neutralizing functions that we have not directly considered here. We note that evidence in animal models supports the findings that neutralizing antibodies mediate protective immunity³⁶, with some showing an additional benefit of Fc receptor function³⁷.

The analysis presented here has a number of limitations. Firstly, our dose-response analysis requires comparison of the in vivo measured antibody concentrations and the estimated in vitro IC50s. This relied on a meta-regression of estimates of the IC50 of each mAb against different variants, and we observed here a large between-study variation (Figure S1). This means that the results of any particular study or assay may vary quite considerably from the central estimate of the IC50 for analysis of all studies. Despite the limitations of the IC50 meta-analysis, we found that the fitted model for the relationship between antibody concentration and efficacy was robust to the uncertainties in the IC50s (Table S7 and Supplementary Methods). Further, when restricting the analysis to only studies conducted when non-VOC predominated, unadjusted antibody concentrations were similarly predictive of efficacy.

An additional limitation is that we did not have access to raw data from the clinical studies. Data were requested from all corresponding authors of the original studies (31 March 2023) but were not provided by the time of revision (30 June 2023), thus we relied on extraction of data from the published reports. This involved manual extraction of data from figures in some cases, which carries implicit risks of error. Data were extracted independently by two authors, and the results were compared to resolve discrepancies³⁸. Further, we are reliant upon population level rather than individual-level data on antibody concentration, half-lives, and clinical outcomes broken down by time intervals as reported in, or extracted from, the published studies. Thus, we could not account for between-subject variability, or subjects lost to follow-up (although fortunately, these numbers are relatively small). Additionally, the analysis is strongly influenced by the results from the Herman et al.¹⁵ and Schmidt et al.¹³ studies, given their contribution of data at lower effective antibody concentrations (Fig. 2). It is also evident that each study data point on efficacy contained considerable uncertainty (with wide confidence intervals), this contributed somewhat to wide confidence intervals in the overall fitted model (Fig. 2). To gain more precise estimates of the dose-response curve for the use of monoclonal antibodies for prophylaxis, combining the results from more studies, preferably with individual-level data available, and with longer follow-up times (where this is ethical) would be helpful.

Other differences between the design of the included studies may have impacted the analysis, which we could not account for directly. For example, seropositivity was zero at baseline in all studies except Isa et al.⁵, and vaccination of participants after treatment was allowed in Herman et al.¹⁵ (Table S2). Fortunately, in our sensitivity analysis, we found that the correlation between antibody correlation and efficacy was robust to removing either of these studies (Table S4). Also, uptake of vaccination in Herman et al. was similar between treated and control arms (~35%). This likely lowered the power of the study to detect efficacy at later time points, but otherwise was not expected to have a large impact on the efficacy estimates in this study (Table S2). In addition, studies used different modes of delivery of the monoclonal antibodies, i.e., casirivimab/imdevimab was administered subcutaneously in the Isa⁵, O’Brien⁶, and Herman¹⁵ studies, whereas cilgavimab/tixagevimab and adintrevimab were administered intramuscularly^4,13. Plasma antibody concentrations appear to increase more slowly following intramuscular administration compared with subcutaneous administration of casirivimab/imdevimab (Fig. 1 and Table S8), and thus it is possible that there is a delay until protective antibody concentrations are achieved. To avoid this difference and also to account for the risk of infection around the time of antibody administration, we omitted the earliest time interval from our analysis (which encompassed the first 7, 10, 27, 28, 30, or 90 days across different studies, Table S1).

Finally, when comparing mAb prophylaxis and vaccination, there are some additional limitations. In particular, this comparison involves converting the antibody concentration data to a fold of convalescent scale. However, we have not shown that such normalization can account for all the differences between neutralization titers achieved by vaccination and mAb administration. For example, it is possible that the normalization employed here oversimplifies the comparison since it does not appreciate potential differences between neutralization readouts for polyclonal sera and monoclonal antibodies. Further, only a subset of studies (n = 19) in our meta-regression reported the geometric mean neutralization titer of a suitable panel of convalescent sera. In addition, the definition of convalescent sera was specified differently in each study, which introduced some potential confounders to these aggregated estimates. Even so, this approach has revealed surprisingly similar prophylactic and vaccine efficacy for a given neutralization titer on the fold of the convalescent scale (within the statistical power of the data).

Our conclusion that monoclonal antibody therapy and vaccination provide similar protection at equivalent neutralizing antibody titers is consistent with a recent publication by Follmann et al.³⁹. In that study the authors performed an analysis using simulated individual neutralizing antibody titers from one trial of casirivimab and imdevimab^6,15 (from the study in Fig. 1b here) and one mRNA-1273 vaccine trial^23,26. The authors found that monoclonal antibodies could explain most of the observed vaccine protection at high antibody levels, although they note that such a comparison was not possible at low neutralizing titers due to limited power. Our analysis suggests that vaccination and monoclonal antibody administration also provide similar levels of protection at low antibody levels.

Vaccination has provided a high level of population immunity to COVID-19. However, there remain a number of subgroups in which vaccination is either not possible or ineffective (largely due to immunodeficiency). The use of monoclonal antibodies for prophylaxis in these cohorts has the potential to provide long-term protection from both symptomatic and severe COVID-19 for these vulnerable groups. However, the frequent observation of novel SARS-CoV-2 variants that escape antibody recognition has raised significant challenges in predicting monoclonal antibody protection against new variants. Further work is required to obtain more data on protection at low antibody levels, as well as to validate predictions of prophylactic efficacy against SARS-CoV-2 variants. Within this context, the work presented here provides a quantitative and evidence-based framework for predicting monoclonal antibody efficacy that can be used in the assessment of novel therapeutics or in designing optimal regimes for new SARS-CoV-2 variants.