This study illustrates how a systems biology approach can be applied in a clinical scenario to understand the complex molecular events that take place in vivo after trivalent influenza vaccination, and to develop better molecular biomarkers for vaccine responsiveness. The implications of the findings, therefore, extend from vaccine biology to vaccine development and clinical vaccinology.
The data support a model in which genes involved in interferon signaling and antigen presentation pathways are strongly upregulated in the initial 24 hours after vaccination, and the expression pattern of early-activation genes correlates strongly with the magnitude of the antibody response measured 14 and 28 days after vaccination. The simultaneous assessment of transcript abundance for over 25,000 genes and gene candidates offers an unbiased, genome-wide view of the transcriptional events that take place after vaccination. By performing this analysis before and at three time points after vaccination, we have gained a deeper understanding of these events and their timing than had been possible before. The three patterns of coexpression that became evident from our analysis are quite distinct and suggest a previously undescribed biphasic transcriptional response to vaccination. The observation of maximum expression changes in the initial 24 hours after vaccination and the correlation between early expression and the magnitude of the antibody response are important new contributions to the understanding of the vaccination response sequence. Content analysis suggests a central role in the early vaccine response for genes whose products are involved in viral sensing via TLR7 and TLR8, MHC Class I presentation, interferon signaling, IL-6, and the NFκβ and JAK/STAT signaling pathways. These genes are also known to be activated in innate cells (ie, macrophages and dendritic cells) during viral infections [26
]. Days 3 and 14 show similar patterns of expression, more suggestive of increased RNA processing and protein synthesis. For inactivated influenza vaccines, therefore, these findings suggest that the first 24 hours after vaccination are of great biological importance.
As the field of clinical vaccinology moves into an era of high-throughput experimental data generation and systems-level biology, it is imperative to give serious statistical consideration to the limitations of current methods of assessment of the magnitude of immune responses. With trivalent influenza vaccines, which in the case of influenza are of high clinical importance, three factors are known to complicate the quantitative analysis of antibody titer data in clinical studies. First, by young adulthood, most individuals have been exposed to influenza antigens through infection or prior vaccination. Indeed, despite our exclusion of individuals who had been vaccinated over the previous three years, most of the study participants had measurable prevaccination antibody titers. Second, individuals with higher prevaccination titers have smaller differences between pre- and postvaccination titers. Accordingly, we observed in our data an approximately linear inverse correlation between prevaccination titers and the rise in titer. For this reason, a simple calculation of the titer change (the titer delta) is inadequate as a way to classify the vaccine responsiveness of individuals. Finally, influenza vaccination currently involves the concomitant administration of three antigens and, while the immune response takes place simultaneously and separations are somewhat artificial, there is inter- and intraindividual variation in the response to each. These observations are depicted in Figure S2. We, therefore, developed the TRI (Titer Response Index) as a classification method for these data. The TRI accounts for an individual’s prevaccination antibody titers, their titers for the three antigens, and the magnitude of their response relative to other individuals in the population. The observed trimodal distribution clearly deviates from what would be expected by chance alone, permitting a statistically significant classification of trivalent influenza vaccine recipients on the basis of responsiveness.
Compared to previous studies, our study has a larger sample size, which provides sufficient power to detect significant correlations between transcript abundance and responsiveness. By assessing the relationship between the early patterns of gene expression and the magnitude of the antibody response, we have discovered 494 transcriptional biomarkers that strongly correlate with the humoral immune response to trivalent influenza vaccines. Our content and network analyses of this predictive gene expression signature again point clearly toward the interferon and antigen presentation pathways. Genes in these pathways, as well as other genes with levels of expression that are clearly different between the highest and lowest vaccine responders in our study (CD74, HLA-E, E2F2, and PTEN) are potential targets for functional studies linking the early molecular events that follow vaccination with the subsequent adaptive immune response. Genomic regions that control expression of these genes could be important for studies seeking to explain interindividual variation in vaccine responsiveness at the DNA level.
A full predictive analysis using the 92 individuals for which expression data was analyzed would be ideal for validating these correlated transcripts. However, our power to perform cross-validation on individuals in the intermediate response range is affected by the expression differences that result from interindividual variation. Fortunately, these subjects appear to develop protective immune responses after vaccination and are less crucial to the interpretation of the data. Since individuals who did not mount a response to the trivalent influenza vaccine are the most clinically interesting and, given that the vaccine response of the entire cohort follows a trimodal distribution, we sampled from the extreme ends of this distribution in order to maximize the power of the cross-validation analyses. Specific patterns of gene expression characterize individuals at the two extremes of the antibody response spectrum, and cross-validation illustrates their predictive value. The fact that a simple gradient of STAT1 to E2F2 expression can predict an individual’s response status at the extremes of the spectrum, and the fact that cross-validation could be performed using a small subset of representative genes, underscore the magnitude of the differences and suggest that simplified predictors based on early gene expression patterns are possible. Further validation of our gene expression-based prediction model in another and preferably larger group of volunteers is warranted. Ideally, this would include individuals of both sexes and different ethnicities.
Early molecular predictors of vaccine responsiveness can be useful in the development and comparison of new vaccines and adjuvants. They can also play a role in studies of vaccine response among different subgroups (children, the elderly, or immunocompromised patients, for example) and open the door for studies of individualized vaccine regimens. Past and recent influenza pandemics highlight the need for studies geared toward rapid implementation of clinical and translational research findings.