Transcriptomic signatures induced by the Ebola virus vaccine rVSVΔG-ZEBOV-GP in adult cohorts in Europe, Africa, and North America: a molecular biomarker study

Abstract

Background: A recombinant vesicular stomatitis virus vector expressing the Zaire Ebola virus glycoprotein (rVSVΔG-ZEBOV-GP) vaccine has been reported as safe, immunogenic, and highly protective in a ring vaccination trial. We aimed to identify transcriptomic immune response biomarker signatures induced by vaccination and associated signatures with its immunogenicity and reactogenicity to better understand the potential mechanisms of action of the vaccine.

Methods: 354 healthy adult volunteers were vaccinated in randomised, double-blind, placebo-controlled trials in Europe (Geneva, Switzerland [November, 2014, to January, 2015]) and North America (USA [Dec 5, 2014, to June 23, 2015]), and dose-escalation trials in Africa (Lambaréné, Gabon [November, 2014, to January, 2015], and Kilifi, Kenya [December, 2014, to January, 2015]) using different doses of the recombinant vesicular stomatitis virus vector expressing the Zaire Ebola virus glycoprotein (rVSVΔG-ZEBOV-GP; 3 × 10⁵ to 1 × 10⁸ plaque-forming units [pfu]). Longitudinal transcriptomic responses (days 0, 1, 2, 3, 7, 14, and 28) were measured in whole blood using a targeted gene expression profiling platform (dual-colour reverse-transcriptase multiplex ligation-dependent probe amplification) focusing on 144 immune-related genes. The effect of time and dose on transcriptomic response was also assessed. Logistic regression with lasso regularisation was applied to identify host signatures with optimal discriminatory capability of vaccination at day 1 or day 7 versus baseline, whereas random-effects models and recursive feature elimination combined with regularised logistic regression were used to associate signatures with immunogenicity and reactogenicity.

Findings: Our results indicated that perturbation of gene expression peaked on day 1 and returned to baseline levels between day 7 and day 28. The magnitude of the response was dose-dependent, with vaccinees receiving a high dose (≥9 × 10⁶ pfu) of rVSVΔG-ZEBOV-GP exhibiting the largest amplitude. The most differentially expressed genes that were significantly upregulated following vaccination consisted of type I and II interferon-related genes and myeloid cell-associated markers, whereas T cell, natural killer cell, and cytotoxicity-associated genes were downregulated. A gene signature associated with immunogenicity (common to all four cohorts) was identified correlating gene expression profiles with ZEBOV-GP antibody titres and a gene signatures associated with reactogenicity (Geneva cohort) was identified correlating gene expression profiles with an adverse event (ie, arthritis).

Interpretation: Collectively, our results identify and cross-validate immune-related transcriptomic signatures induced by rVSVΔG-ZEBOV-GP vaccination in four cohorts of adult participants from different genetic and geographical backgrounds. These signatures will aid in the rational development, testing, and evaluation of novel vaccines and will allow evaluation of the effect of host factors such as age, co-infection, and comorbidity on responses to vaccines.

Funding: Innovative Medicines Initiative 2 Joint Undertaking.

Figure 1. Flow diagram of the four study cohorts

Overview of the number of participants in each study, of the timepoints at which peripheral whole blood samples were collected with day 0 as baseline before vaccination, and the number of participants who were given the various vaccine doses or placebo. pfu=plaque-forming units.

Figure 2. Effect of time and dose on gene expression profiles after rVSVΔG-ZEBOV-GP vaccination in the Geneva cohort

(A) Molecular degree of perturbation and principal component analysis performed on GAPDH-normalised, log₂-transformed gene expression data of the Geneva cohort to evaluate the effect of time by separating samples by timepoint (days 0, 1, 3, 7, 14, and 28). Day 0 samples of vaccinees were used as baseline controls. Timepoints were compared using Mann-Whitney U test. (B) Effect of dose (1 × 10⁷ and 5 × 10⁷ pfu vs 3 × 10⁵ pfu) evaluated by molecular degree of perturbation analysis at distinct timepoints. Day 0 samples of vaccinees were used as baseline controls. Timepoints were compared using Mann-Whitney U test. pfu=plaque-forming units. rVSVΔG-ZEBOV-GP=recombinant vesicular stomatitis virus vector expressing the Zaire Ebola virus glycoprotein.

Figure 3. Identification of DEGs and key networks after rVSVΔG-ZEBOV-GP vaccination in the Geneva cohort

Differential expression analysis was performed on GAPDH-normalised log₂-transformed gene expression data of the Geneva cohort. (A) Volcano plots representing DEGs at different timepoints (days 1, 3, 7, 14, and 28) after rVSVΔG-ZEBOV-GP vaccination of all vaccinees (high dose 2 plus low dose) compared with their baseline gene expression levels. The y-axis scales of all plots are harmonised. p values are shown on a –log₁₀ scale for better visualisation. Genes with p<0·05 and log₂ fold change of less than –0·6 or more than 0·6 were labelled as DEGs. (B) Ingenuity pathway analysis interactive network analysis of DEGs identified between day 0 and day 1 following rVSVΔG-ZEBOV-GP vaccination of all vaccinees (high dose two and low dose) compared with their baseline gene expression levels. The shapes of the nodes represent the functional classes of the gene products. DEG=differentially expressed gene. IFITM_1/3=IFITM₁ or IFITM₃, or both. rVSVΔG-ZEBOV-GP=recombinant vesicular stomatitis virus vector expressing the Zaire Ebola virus glycoprotein.

Figure 4. Identification of signatures associated with rVSVΔG-ZEBOV-GP vaccination at the peak of the transcriptomic response (day 1) in a pooled dataset of the Geneva and USA cohorts

The pooled dataset of the Geneva and USA cohorts was used to train the model in which 70% of each dataset was used as training set and the remaining 30% of each dataset was used as test set. (A) Receiver operating characteristic curve and AUC show the classifying performance of the trained model. (B) Predicted probability plots showing the accuracy of the identified pooled biomarker signature across timepoints in box-and-whiskers plots (5–95 percentiles) either in the pooled cohort in which train-test split was performed or in the single validation cohorts. AUC=area under the receiver operating characteristic curve. rVSVΔG-ZEBOV-GP=recombinant vesicular stomatitis virus vector expressing the Zaire Ebola virus glycoprotein.

Figure 5. Identification of signatures associated with rVSVΔG-ZEBOV-GP vaccination at day 7 in a balanced pooled dataset of all four cohorts

The balanced pooled dataset of the four cohorts was used to train the model in which 70% of each dataset was used as train set and the remaining 30% of each set was used as test set. (A) Receiver operating characteristic curve and AUC show the classifying performance of the trained model. (B) Predicted probability plots showing the accuracy of the identified pooled biomarker signature across timepoints in box-and-whiskers plots (5–95 percentiles) either in the pooled cohort in which train-test split was performed or in the single validation cohorts. AUC=area under the receiver operating characteristic curve. rVSVΔG-ZEBOV-GP=recombinant vesicular stomatitis virus vector expressing the Zaire Ebola virus glycoprotein.

Figure 6. Correlation between gene expression profiles and ZEBOV-GP-specific antibody titres in response to rVSVΔG-ZEBOV-GP vaccination in cohorts from Geneva, USA, Lambaréné, and Kilifi and reactogenicity (arthritis) in the Geneva cohort

(A) Volcano plot showing genes that significantly correlate with ZEBOV-GP-specific antibody titres in all cohorts using random-effects meta-analyses (p<0·05). (B) Heatmap showing the Spearman’s rank correlation coefficient of the genes included in the five-gene signature (p<0·05) that correlates with the ZEBOV-GP-specific antibody titres. (C) The receiver operating characteristic curves show the predictive power of the identified five-gene signature to classify participants at day 1 into those that will and those that will not develop arthritis in training and test set using the regularised logistic regression model. The curves display the values extracted from the regularised logistic regression model after hyperparameters adjustments. AUC=area under the receiver operating characteristic curve. IFITM_1/3=IFITM₁ or IFITM₃, or both. rVSVΔG-ZEBOV-GP=recombinant vesicular stomatitis virus vector expressing the Zaire Ebola virus glycoprotein. ZEBOV-GP=Zaire Ebola virus glycoprotein.