Baseline gene signatures of reactogenicity to Ebola vaccination: a machine learning approach across multiple cohorts

Refined innate plasma signature after rVSVΔG-ZEBOV-GP immunization is shared among adult cohorts in Europe and North America

Background

During the last decade Ebola virus has caused several outbreaks in Africa. The recombinant vesicular stomatitis virus-vectored Zaire Ebola (rVSVΔG-ZEBOV-GP) vaccine has proved safe and immunogenic but is reactogenic. We previously identified the first innate plasma signature response after vaccination in Geneva as composed of five monocyte-related biomarkers peaking at day 1 post-immunization that correlates with adverse events, biological outcomes (haematological changes and viremia) and antibody titers. In this follow-up study, we sought to identify additional biomarkers in the same Geneva cohort and validate those identified markers in a US cohort.

Methods

Additional biomarkers were identified using multiplexed protein biomarker platform O-link and confirmed by Luminex. Principal component analysis (PCA) evaluated if these markers could explain a higher variability of the vaccine response (and thereby refined the initial signature). Multivariable and linear regression models evaluated the correlations of the main components with adverse events, biological outcomes, and antibody titers. External validation of the refined signature was conducted in a second cohort of US vaccinees (n=142).

Results

Eleven additional biomarkers peaked at day 1 post-immunization: MCP2, MCP3, MCP4, CXCL10, OSM, CX3CL1, MCSF, CXCL11, TRAIL, RANKL and IL15. PCA analysis retained three principal components (PC) that accounted for 79% of the vaccine response variability. PC1 and PC2 were very robust and had different biomarkers that contributed to their variability. PC1 better discriminated different doses, better defined the risk of fever and myalgia, while PC2 better defined the risk of headache. We also found new biomarkers that correlated with reactogenicity, including transient arthritis (MCP-2, CXCL10, CXCL11, CX3CL1, MCSF, IL-15, OSM). Several innate biomarkers are associated with antibody levels one and six months after vaccination. Refined PC1 correlated strongly in both data sets (Geneva: r = 0.97, P < 0.001; US: r = 0.99, P< 0.001).

Conclusion

Eleven additional biomarkers refined the previously found 5-biomarker Geneva signature. The refined signature better discriminated between different doses, was strongly associated with the risk of adverse events and with antibody responses and was validated in a separate cohort.