Systems Vaccinology: Enabling rational vaccine design with systems biological approaches

Introduction

Since Edward Jenner’s discovery in the late 1700’s that inoculation with the cowpox virus provided protection from smallpox infection, vaccines have emerged as one of the greatest public health tools in history. The last 60 years have established a golden age in the field of vaccinology, marked by events such as the eradication of smallpox by 1980 [1] and the development of polio vaccines in the 1950s, which have lead to near-eradication of the disease [2]. Despite the great success of these and other vaccines, there remain significant challenges for development of new vaccines against current global pandemics such as HIV and malaria. Among the many problems facing this field are: (i) most currently used vaccines were designed largely empirically. As a result there is little or no understanding of what the correlates and mechanisms of protection are for many vaccines. For example, although the two commercially available types of influenza vaccine, trivalent inactivated (TIV) and live attenuated (LAIV), provide similar levels of protection from infection [3], they generate significantly different immune responses. TIV induces higher levels of IgG antibody secreting cells (ASCs) in the blood as well as higher levels of serum antibodies than LAIV in adults. This is likely due to the different routes of administration, as LAIV, which is administered as an intranasal spray, is thought to produce a more local response in the nasal mucosa and upper respiratory tract, including IgA (mucosal) antibodies and cellular immune responses. As a result, the correlate of protection for TIV is generally considered to be serum antibodies, while the correlate of protection for LAIV is less clear [3]. (ii) The path to licensure of candidate vaccines involves very lengthy and expensive phase IIB and III clinical trials to assess their efficacy and safety. These trials involve thousands of subjects and can cost hundreds of millions of dollars to complete. As a result, very few vaccine concepts are tested in phase III trials. For example, during the past 30 years, only 4 HIV-1 vaccine concepts have been tested for clinical efficacy [4], and despite repeated failures, the correlates and mechanisms of protective immunity against HIV remain poorly understood.

The conventional immunological methods, such as ELISA, ELISPOT, flow cytometry, etc., used to study vaccines have played a valuable role in the field of vaccinology, and will remain essential in evaluating responses to vaccination in the future. However these approaches are generally only able to analyze a single or small number of components of the immune system at a given time, and are insufficient to analyze the full complexity of the structure and dynamics of the human immune system as a whole. This represents a critical obstacle towards understanding the molecular mechanisms by which vaccines generate protective immune responses and identifying meaningful correlates of protection.

To address this issue, vaccinologists have turned to systems biology. By examining how coordinated interactions at a molecular level give rise to immune responses, systems biology approaches enable a holistic view of the immune system and its many components. This developing field provides many promising tools to overcome the challenges facing current vaccine development. Enabling researchers to evaluate the immune responses of fewer subjects in a more in-depth and detailed fashion has the potential to dramatically improve our understanding of the mechanisms of protection of novel vaccines and decrease the length and costs of current clinical trials.

Systems Vaccinology

Within the past 20 years, advances in high-throughput technologies have granted researchers the ability to interrogate the properties and abundances of entire classes of molecular components within the cell. For example, development of lower cost next-generation sequencing technology has facilitated the growth of transcriptomics, which seeks to measure the expression of all RNA transcripts within a cell or population of cells. By sequencing and mapping mRNA transcripts, RNA-sequencing enables the accurate quantification of gene expression as well as simultaneous identification of RNA structure such as transcription start site and exon usage/splice junctions, the regulation of which has been shown to play an important role in many biological processes, including within the immune system [5, 6]. Simultaneously, in the growing domain of metabolomics, analytical chemistry techniques such as liquid chromatography-mass spectrometry (LC-MS) have been harnessed to identify and quantify the set of metabolites within cells or tissues [7]. Changes in metabolic activity are an important component of both innate and adaptive immune responses [8], such as the recognized role that lipid metabolism plays during inflammation [8–10].

Systems vaccinology is an emerging field that applies such ‘omics’ technologies, in combination with bioinformatics tools such as transcriptional network analysis and predictive modeling, to study immune responses to vaccination [11–13]. As a systems-based approach, it aims to use data generated through high-throughput measurements in the context of vaccination to characterize the interactions between individual components of the immune system in the interest of understanding and predicting behavior of the system as whole. This includes analysis of transcriptional, signaling, and metabolic pathways whose activity is perturbed in the various cells of the immune system in response to vaccination, as well as identification of molecular signatures that are predictive of various measurements of protection from infection. The knowledge obtained through these analyses can aid in the rational design of new vaccines that generate long-lasting protection and induce improved responses in populations with diminished immune function such as the elderly.

5 year historic perspective

The first examples of the use of such approaches to study responses to vaccination were performed on the yellow fever vaccine [14, 15]. This vaccine contains a live-attenuated strain (YF-17D) of the yellow fever virus, which induces potent and long-lived CD8+ T cell and neutralizing antibody responses [16, 17]. By combining high-throughput measurements such as microarray gene expression profiling and multiparameter flow cytometry with computational modeling, we were able to detect a regulated network of interferon and innate antiviral genes that were induced post-vaccination in peripheral blood mononuclear cells (PBMCs) [14]. An independent YF-17D study by Gaucher et al. revealed induction of similar transcriptional responses to vaccination, including type I interferon and inflammatory pathways [15]. In addition to examining innate immune pathways activated by vaccination, we successfully identified unique gene signatures that were capable of accurately predicting the CD8+ T cell and neutralizing antibody responses, respectively [14]. The predictive CD8+ T cell signature contained complement protein C1qB and eukaryotic translation initiation factor 2 alpha kinase 4, which is an orchestrator of the integrated stress response. Meanwhile the B cell growth factor receptor TNFRSF17 was among the genes included in the antibody response signature. This work demonstrated for the first time that the immunogenicity of a vaccine could be successfully predicted using early transcriptional measurements within 1 week of vaccination.

Following these initial studies, systems biology approaches have been used to examine immune responses to vaccines against a wide range of pathogens, including influenza [18, 19], malaria [20], smallpox [21], and HIV [22]. In particular, as YF-17D is a live-attenuated vaccine that induces an acute viral infection, the study of influenza vaccination (TIV) enabled investigation into to whether or not similar methods could be used to identify molecular signatures predictive of response to an inactivated vaccine. We identified transcriptional signatures related to the expansion of plasmablasts and the unfolded protein response within B cells on day 7 post-vaccination that correlated with and were predictive of day 28 influenza-specific antibody responses [18]. Indeed, these findings were consistent with studies by Bucasas et al. [19] and Obermoser et al. [23]. Interestingly, TNFRSF17, which was predictive of antibody responses to YF-17D, also appeared in the signatures predictive of TIV response [18]. Recently, Tsang et al. [24], and Furman et al. [25] have extended this approach to search for baseline signatures capable of discriminating between high and low responders to vaccination. However, possibly due to limited sample sizes and weaker signal at baseline, neither study was able to successfully predict antibody response using baseline transcriptional measurements alone. Instead, Tsang et al. utilized cell subset frequencies, while Furman et al. generated a model based on transcriptional modules, serum cytokines, cell subset frequencies, and pre-existing antibody titers. Additionally, as these studies were conducted using cohorts from an individual flu season, the effect of changes in influenza strains included in the TIV vaccine on the performance of these models remains to be examined. To address this question, we are performing a comprehensive analysis of over 400 adults vaccinated with seasonal TIV during 5 consecutive influenza seasons (Nakaya et al., manuscript in preparation). This analysis is an important step towards generating robust and clinically relevant signatures that can be used to predict the efficacy of vaccines in clinical trials.

Among the vaccines under investigation, malaria is one for which a human challenge model exists, allowing for identification of subjects who are protected from or susceptible to infection [26]. Vahey et al. used this model, in which subjects vaccinated with the RTS,S malaria vaccine were challenged using mosquitos infected with an antimalarial-sensitive strain of Plasmodium, to identify transcriptional signatures in PBMCs capable of discriminating between protected and nonprotected vaccinees [20]. Protected subjects had increased expression 2 weeks post-vaccination (but before challenge) of genes involved in the immunoproteasome degradation pathway, involved in MHC peptide processing, compared to subjects who developed parasitemia [20]. These systems-based approaches can be used to identify early signatures of efficacy of other vaccines for which human challenge models have been developed, such as typhoid [27, 28] and Shigella [29], enabling rapid evaluation of novel candidate vaccines in future clinical trials.

A critical question is the degree to which transcriptional responses to vaccination were conserved across vaccines, and whether or not there are universal signatures capable of predicting antibody responses to vaccination. To this end, we used a systems-based approach to compare signatures induced by different types of vaccines [YF-17D, LAIV, TIV, the carbohydrate meningococcal vaccine (Menimmune), and the conjugate meningococcal vaccine (Menectra)] [30]. We observed that while recall antibody responses to inactivated vaccines (e.g. seasonal influenza vaccine, diphtheria toxoid component of the conjugate meningococcal vaccine) were associated with transcriptional modules related to plasmablast differentiation, the antibody responses of live-attenuated vaccines (e.g. yellow fever vaccine) were highly correlated with modules involving innate immunity and type I interferon responses. Meanwhile, antibody responses to polysaccharide components of the meningococcal vaccines were associated with increased proinflammatory cytokines as well as activation of antigen-presenting cells [30]. Thus, these preliminary results suggest that vaccine-induced signatures of immunity are significantly dependent on the class of vaccine in question.

Finally, systems approaches could also potentially be used to identify signatures of vaccine safety. Adverse reactions to vaccination pose major regulatory challenges to vaccine development. Relatively mild adverse reactions such as transient fever, local swelling at the injection site are fairly common and occur within a few hours or days after vaccination. The more serious adverse reactions such as the often fatal viscerotropic disease caused by yellow fever vaccination are very rare (1 in 250,000 cases [16]. The ability to identify signatures at baseline that would be predictive of such adverse reactions would thus represent an important advance, however a major obstacle to this is the rarity of such serious adverse events. In this context, detailed immunological characterization of a patient who developed viscerotropic disease after yellow fever vaccination, revealed a 200-fold expansion of CD14+CD16+ inflammatory monocytes during the period of disease, and a 20-fold elevation in the numbers of such cells even during the convalescent phase, long after the virus had been cleared [63]. This raises the possibility that the persistently elevated levels of the CD14+CD16+ inflammatory monocytes may have predisposed this subject towards viscerotropic disease after vaccination with the yellow fever vaccine. The further evaluation of this possibility is stymied by the rarity of these adverse reactions, and the difficulty in obtaining samples to perform detailed analysis. Future efforts should thus be aimed at creating a clinical infrastructure that would facilitate the acquisition of samples from vaccine adverse reactions at multiple time points, including the convalescent phase.

Challenges

Systems vaccinology analyses face challenges from both the biological and technological perspectives (Fig. 1). These problems must be tackled simultaneously in order to achieve a deeper understanding of how the immune system as a whole responds to vaccination. This is why it is important that systems vaccinology be carried out through collaboration between vaccinologists, immunologists and bioinformaticians. As biological knowledge and insight as well as rigorous and sound data analysis are both required for successful discovery, these two groups should work in close contact and coordinate throughout each phase of the scientific process, including hypothesis generation, experimental design, execution, and data analysis.

Drowning in a Sea of Big Data

The revolution in high-throughput technologies has created a dramatic change in the landscape of biological research: with such large amounts of information generated with each experiment, we are now ‘drowning in a sea of big data.’ A cornerstone of successful systems vaccinology is the ability to extract meaningful knowledge from this wealth of data, and then to harness this knowledge to generate improved biological understanding of the immune system’s response to vaccination. In the first transition, from data to knowledge, statistical approaches [76] are generally used to identify sets of genes, metabolites, etc. that undergo significant changes during system perturbation (e.g. before and after vaccination). However, a simple list of differentially expressed genes or metabolites is not highly informative, and may contain false positives. One approach to improve this type of analysis is to integrate expression measurements with a priori knowledge about the interactions and co-expression of genes within a given biological context. One such bioinformatics tool is gene-set enrichment analysis [77], which identifies enrichment of biologically relevant groups of genes (modules) within gene lists ranked by response to perturbation. This can be performed using manually curated sets of biological pathways, such as the Kyoto Encyclopedia of Genes and Genomes [78] and Reactome [79] databases, or through extraction of gene modules from existing data by searching for groups of co-expressed genes across experimental conditions or biological contexts [30, 80, 81]. MSigDB, a popular gene set database, contains collections of both types [82]. Examining changes in activity of pathways or modules rather than individual genes reduces noise and appearance of false positives while simultaneously providing improved biological meaning and functional context to the data analysis.

Differential transcriptional pathways and networks can also be coupled with feature selection and machine learning algorithms (such as those mentioned in the previous section) to generate biologically informative predictive signatures of vaccine response (Fig. 2). These signatures can be used to assess vaccine efficacy and identify subjects at risk for infection. In addition, analysis of predictive pathways can yield insight into the mechanisms responsible for deficient immune responses. Once predictive signatures are generated using the data at hand, they should be validated in independent cohorts to ensure their robustness across diverse populations.

While the knowledge of biological pathways activated or suppressed in response to vaccination is useful in its own right, achieving deeper understanding and control over the mechanisms by which vaccines induce protective immune responses requires human knowledge and insight coupled with traditional methods of experimental validation. Ideally, observations made through high-throughput data analysis can help generate biological hypothesis that are then tested in animal models such as knockout mice. Recently our group has used this approach to identify a novel role for the integrated stress response, mediated by EIF2AK4 [eukaryotic initiation factor 2 α-kinase 4, also known as general control nonderepressible 2 kinase (GCN2)], in stimulating autophagy and antigen presentation within dendritic cells during response to YF-17D [83]. In our original systems biology analysis of the YF-17D vaccine, expression of GCN2 in PBMCs on day 7 post-vaccination was shown to be predictive of the later CD8+ T cell response [14]. As a regulator of the integrated stress response, accumulation of uncharged tRNA during amino acid starvation activates GCN2, which results in phosphorylation of eukaryotic initiation factor 2 α (eIF2α). This process leads to reduced activity of the eIF2 complex, resulting in reduced rates of protein synthesis and stress granule formation [84]. However, GCN2’s role in immune responses is not clear. The capability of GCN2 expression to predict CD8+ T cell responses to YF-17D led us to believe that this kinase may be important for priming of the adaptive immune response. Follow up work in mice revealed that YF-17D induced amino acid starvation in dendritic cells, inducing autophagy and antigen cross-presentation in a GCN2-dependent manner [83]. This reveals an unappreciated connection between the ancient nutrient sensing pathway and elicitation of adaptive immunity by dendritic cells.

Another mechanistic discovery enabled by a systems vaccinology is the role that the intestinal microbiota play in stimulating antibody responses to vaccination. This development was prompted by the finding in our systems-level analysis of human immune responses to TIV vaccination that expression of the toll-like receptor TLR5 on day 3 post-vaccination was significantly correlated with the day 28 antibody response [18]. As TLR5 is a receptor specific for flagellin, the protein subunit of bacterial flagellum, its association with antibody response to a viral vaccine was unexpected. The importance of TLR5 sensing of intestinal bacteria during TIV vaccination was confirmed in experiments with antibiotic treated, germ-free, and TLR5^−/− mice, all of which experienced significant reduction in antibody responses after TIV vaccination compared to controls [85]. Further work demonstrated that sensing of flagellin by TLR5 promoted plasma cell differentiation directly and by stimulating lymph node macrophages to produce plasma cell growth factors. Interestingly, the loss of intestinal microbiota did not result in reduction of humoral responses to adjuvanted vaccines or the live-attenuated YF-17D vaccine, indicating that the microbiota may be acting as a natural adjuvant in the absence of other sources of immune stimulation. As the microbiome is affected by diet and varies widely across the world [86], gut microbial composition adds another layer of diversity in the human population that should be considered in future studies of immune responses to vaccination. The experimental evaluation of the role of the microbiome in promoting immunity during vaccination in humans is ongoing.

Conclusions

Developments in high-throughput technologies are enabling vaccinologists to investigate the immune responses induced by vaccines at a greater depth than ever before. More importantly, these advances are facilitating the identification of robust molecular and cellular signatures of protective immunity, which can help to generate diagnostic tools that reduce the length and cost of current clinical trials. In this developing field, vaccinologists, immunologists, bioinformaticians, and systems biologists must work hand in hand to advance our understanding of the molecular mechanisms by which vaccines induce protective immunity and help drive development of the next generation of vaccines.