The logistic regression model used for this immunobridging study was previously developed using data from NHP challenge studies24. Seven NHP challenge studies were conducted with a single 0.5 mL intramuscular administration of the well-characterized EBOV Kikwit virus recommended by the Filovirus Animal Non-Clinical Group (FANG)24,48 at a fully lethal target dose of 100 plaque-forming units. A penalized logistic regression model based on NHP data was developed using Firth’s method49, with survival outcome as a dependent variable and EBOV GP binding antibody concentrations (EU/mL, log10) measured 21 days post-Dose 2 as an independent variable. For the model described in Roozendaal et al.24, EBOV GP binding concentrations were analyzed by the Battelle Biomedical Research Center (Columbus, OH, USA). For the current immunobridging analysis, all NHP samples were reanalyzed by Q2 Solutions Vaccine Testing Laboratory (San Juan Capistrano, CA, USA) to enable a direct comparison to the human immunogenicity data. More information on the selection of the immune parameter and the model development can be found in Roozendaal et al.24.

The immunogenicity data of participants vaccinated with the two-dose, heterologous Ad26.ZEBOV, MVA-BN-Filo regimen in a 56-day interval were obtained from two phase 2 and three phase 3 randomized, observer-blind, and placebo-controlled studies in healthy adults, PLWH, and pediatric (aged 1-17 years) participants: EBL2001 (France, UK)27,28, EBL2002 (Burkina Faso, Côte d’Ivoire, Kenya, Uganda)29,30,31, EBL3001 (Sierra Leone)32,33,34, EBL3002 (USA)35,36, and EBL3003 (USA)36,37. All participants received an intramuscular injection (0.5 mL) with 5 × 1010 virus particles of Ad26.ZEBOV as Dose 1, followed 56 days later by 1 × 108 Infectious Units of MVA-BN-Filo as Dose 2 (heterologous two-dose Ad26.ZEBOV, MVA-BN-Filo vaccine regimen). The protocol-defined window around Dose 2 was ±3 days for all studies, except EBL2001 (±1 day) and EBL3001 (±7 days). In all five clinical studies, EBOV GP binding antibody concentrations were measured by Q2 Solutions Laboratory. Study details are available online27,29,32,35,37 and are previously published28,30,31,33,34,36.

For immunogenicity assessments in the previous NHP challenge studies and clinical studies, serum samples were obtained from all NHPs and clinical study participants immediately before the first vaccination and three weeks after the second vaccination. Immunoglobulin G responses against EBOV GP were analyzed in all NHP and clinical samples using the same EBOV GP (Kikwit) FANG ELISA28,50, validated for both human and NHP serum, in the same Q2 Solutions Laboratory. Responses were summarized as group GMCs with 95% CI. For human samples, all values below the lower limit of quantification (LLOQ [36.11 EU/mL]) were imputed with half of the LLOQ value (ie, LLOQ/2). Clinical study participants were considered responders if the post-vaccination concentration was >2.5-fold the LLOQ in baseline seronegative individuals or >2.5-fold the baseline value in baseline seropositive participants.

Two logistic regression models were constructed for the immunobridging analysis. The primary analysis model shown in Fig. 4 contained only data from NHPs vaccinated with Ad26.ZEBOV, MVA-BN-Filo in a 56-day interval (N = 66). The second model was based on all available data of NHPs (N = 108) vaccinated with either Ad26.ZEBOV or Ad26.Filo and MVA-BN-Filo in different vaccine sequences and intervals between doses. The second model was only used in a sensitivity analysis to evaluate the robustness of the primary immunobridging result. The data that were used for the calculation of the logistic regression models are shown in Supplementary Table 5 in the Supplementary Appendix, and the result is shown in Supplementary Table 6.

Figure depicts the logistic regression model and its 95% confidence band (bootstrap-derived using 10,000 bootstraps of the NHP data of the main regimen). EU enzyme-linked immunosorbent assay units, NHP non-human primate.

The fitted logistic regression model was used to predict a survival probability for each human ELISA value measured at 21 days post-Dose 2. The individual predicted human survival probabilities were averaged to calculate the mean predicted survival probability. First, a 95% CI was calculated using a non-parametric double bootstrap method. This method consisted of resampling the NHP and human datasets 10,000 times with replacement, repeating the fitting of a logistic model on the re-sampled NHP data, and calculating a mean predicted survival probability by inserting the re-sampled human ELISA data into the logistic model and averaging the predicted individual survival probabilities. As a result, 10,000 mean predicted survival probabilities were obtained. The 95% CI was derived as the 2.5th and 97.5th percentiles of the distribution of the mean predicted survival probabilities.

The immunobridging analysis using data from the five clinical studies was originally pre-planned as an interim futility analysis, with no foreseen Type I error rate adjustment. In view of the 2018 to 2020 outbreak and the persisting public health need, this analysis was used as the basis for marketing authorization approval in Europe. To adjust the CI for alpha spending post-hoc, a 98.7% CI was calculated. The 98.7% CI was based on the O’Brien-Fleming alpha spending rules, as this approach is conservative and regularly used in interim analyses (approximately 65% of the pre-planned data available, resulting in an O’Brien-Fleming adjusted one-sided alpha of 0.0066 [obtained using Wang-Tsiatis bounds where Δ = 0]). The same non-parametric double bootstrap procedure was used to calculate the 98.7% CI but was based on 100,000 bootstraps to ensure sufficient resolution in the extreme regions of the distribution.

The primary analysis aimed to evaluate whether the lower limit of the CI was above a pre-defined success criterion of 20%, a cutoff agreed upon with the EMA. Immunogenicity data from healthy adult participants (aged 18-50 years) vaccinated with Ad26.ZEBOV, MVA-BN-Filo in the five contributing clinical studies were pooled. The FAS comprised all participants who were randomized (and non-randomized, open-label stage 1 of study EBL3001) and received ≥1 dose of study vaccine, regardless of the occurrence of protocol deviations. The PPI analysis set represented the primary analysis set and included all randomized (and non-randomized, open-label stage 1 of study EBL3001) and vaccinated participants who received Dose 1 and Dose 2 within the protocol-defined windows, had ≥1 evaluable post-vaccination immunogenicity sample, and had no major protocol deviations influencing the immune response. Only participants with an available 21 days post-Dose 2 ELISA result were included in the immunogenicity and immunobridging analyses. Immunogenicity data were analyzed descriptively, and immunobridging was performed on the PPI analysis set (primary analysis) and on the FAS.

Pre-specified immunobridging sensitivity analyses were conducted to evaluate the robustness of the primary analysis and assess potential influencing factors, such as baseline positivity in the ELISA, sex, age, race, and geographic region (Supplementary Tables 1, 7, and 8 in the Supplementary Appendix). Firstly, because Sierra Leone was the only country included in the five clinical studies that was previously affected by an EBOV outbreak, the analysis was repeated including only participants from study EBL3001 (which was conducted in Sierra Leone), stratified per baseline EBOV GP FANG ELISA levels (<LLOQ [36.11 EU/mL], LLOQ-100, >100-1000, >1000 EU/mL), to assess the impact of pre-existing EBOV GP binding antibody levels on the immunobridging analysis (Supplementary Fig. 1 in the Supplementary Appendix). Secondly, to further assess the effect of possible pre-exposure to EBOV, the analysis was repeated, excluding the participants of the Sierra Leone study (EBL3001; Supplementary Fig. 2)32. Thirdly, demographic subanalyses were conducted, stratified by age (18-30 and 31-50 years of age), sex (female and male), race (Asian, Black or African American, White, and other), and geographic region (East Africa [Kenya, Uganda], West Africa [Burkina Faso, Côte d’Ivoire, Sierra Leone], Europe [France, UK], and North America [USA]; Fig. 1 and Supplementary Fig. 2). Mean predicted survival probabilities with a 95% CI were also calculated based on the pooled data of healthy elderly participants (aged > 50 years), PLWH (aged 18-50 years), and children (1-17 years, and in three age categories: 1-3 years, 4-11 years, and 12-17 years) using the primary analysis model (Fig. 3). PLWH in this analysis were on a stable antiretroviral therapy regimen, had a CD4 + cell count >350 cells/μL, and were considered to be in otherwise reasonably good medical condition without an acquired immunodeficiency syndrome–defined diagnosis or a clinically significant disease. For all sensitivity analyses, the 95% CI were calculated based on the non-parametric double bootstrap method. Fourthly, post-hoc, a subgroup immunobridging analysis was conducted by country (Burkina Faso, Côte d’Ivoire, France, Kenya, Sierra Leone, Uganda, UK, and USA), with the 95% CI calculated based on the non-parametric double bootstrap method (Fig. 2). Finally, as a prespecified sensitivity analysis, the primary analysis was repeated using the second model based on all available ELISA data of NHPs (N = 108) vaccinated with either Ad26.ZEBOV or Ad26.Filo and MVA-BN-Filo in different vaccine sequences and intervals between doses (Supplementary Table 6)33. Results from the pre-specified sensitivity analyses are described in Section 1 (Supplementary Notes) in the Supplementary Appendix.

Further information on research design is available in the Nature Research Reporting Summary linked to this article.