Concept and application of a computational vaccinology workflow

Abstract

Background: The last years have seen a renaissance of the vaccine area, driven by clinical needs in infectious diseases but also chronic diseases such as cancer and autoimmune disorders. Equally important are technological improvements involving nano-scale delivery platforms as well as third generation adjuvants. In parallel immunoinformatics routines have reached essential maturity for supporting central aspects in vaccinology going beyond prediction of antigenic determinants. On this basis computational vaccinology has emerged as a discipline aimed at ab-initio rational vaccine design.Here we present a computational workflow for implementing computational vaccinology covering aspects from vaccine target identification to functional characterization and epitope selection supported by a Systems Biology assessment of central aspects in host-pathogen interaction. We exemplify the procedures for Epstein Barr Virus (EBV), a clinically relevant pathogen causing chronic infection and suspected of triggering malignancies and autoimmune disorders.

Results: We introduce pBone/pView as a computational workflow supporting design and execution of immunoinformatics workflow modules, additionally involving aspects of results visualization, knowledge sharing and re-use. Specific elements of the workflow involve identification of vaccine targets in the realm of a Systems Biology assessment of host-pathogen interaction for identifying functionally relevant targets, as well as various methodologies for delineating B- and T-cell epitopes with particular emphasis on broad coverage of viral isolates as well as MHC alleles.Applying the workflow on EBV specifically proposes sequences from the viral proteins LMP2, EBNA2 and BALF4 as vaccine targets holding specific B- and T-cell epitopes promising broad strain and allele coverage.

Conclusion: Based on advancements in the experimental assessment of genomes, transcriptomes and proteomes for both, pathogen and (human) host, the fundaments for rational design of vaccines have been laid out. In parallel, immunoinformatics modules have been designed and successfully applied for supporting specific aspects in vaccine design. Joining these advancements, further complemented by novel vaccine formulation and delivery aspects, have paved the way for implementing computational vaccinology for rational vaccine design tackling presently unmet vaccine challenges.

Figure 1

Prototypic representation of a (computational) vaccine design workflow. The scheme spans the entire pre-clinical project life cycle from concept phase, determination of vaccine targets, further to detailed epitope analysis, formulation, and experimental validation. The entire process is optimally embedded in an integrated data and knowledge management framework.

Figure 2

Schematic overview of the relation between the Taverna Workbench, Taverna Remote Execution Server, pBone and pView: Workflow templates are generated in the Taverna Workbench and imported into the pBone system via pView. pView then acts as control element for project creation/extension, data import and processing in pBone. Individual components of pBone are set into relation including dependency on Taverna Remote Execution Server for processing of Taverna workflows.

Figure 3

A typical workflow situation utilizing pView. On top a multiple sequence alignment is provided. For the sequences various single sequence profiles are given in separate windows further including a 3D model of the protein of interest.

Figure 4

Presented is an overview of the epitope mapping process. The upper part depicts a subgraph comprising shortest paths between known epitopes and EBV gp110. The lower part of the figure shows the first 300 positions of a multiple secondary structure alignment of homologous envelope glycoproteins of EBV, HHV-5, HHV-1 and EHV-2. To improve readability secondary structures are color coded (helical areas in red, beta sheets in green, coils in blue, signal peptides in yellow and gaps in grey). The black strands above the multiple alignment mark possible mapping positions with respect to their position on the gp110 protein of EBV which are connected to their predecessor in the shortest path.

Figure 5

Representation of heterogeneous data in a network context. Provided is a subgraph encoding information available for EBV homology data enriched by IEDB object types, relationships and content. Red nodes represent EBV proteins from completely sequenced proteomes which are linked to IEDB data. Turquoise nodes represent proteins listed in the IEDB, orange triangles represent scientific publications, blue diamonds represent peptide epitopes, and green diamonds encode experimental assays.

Figure 6

Selection of vertices and edges from the EBV-human interaction graph centered around differentially regulated CD9. EBV proteins are shown in red, human proteins in green. Solid lines indicate physical interactions, dashed lines omicsNET connections. Red, blue and green edges indicate EBV-EBV, EBV-human and human-human interactions, respectively. Human genes significantly differentially regulated upon infection/reactivation are shown as hexagon.

Figure 7

Structural alignment of gp350 (PDB entry 2H6O, turquoise) and EBV type 1 gp350 N-terminal domain model (green). Mutations differentiating the two proteins are highlighted by red spheres. Glycosylations (which are part of the 2H6O structure) are drawn in blue and indicate which residues may not be directly accessible to antibodies. Mutations are located outside the CD21 interacting region which is non-glycosylated and has been implicated in neutralizing immunity. The arrow indicates the CD21/gp350 interface.

Figure 8

The optimized model for the EBV type 1 gp110 protein is given. Monomers are drawn in green, red and blue. The arrow indicates one of the large coils added by homology modeling. The lower part is close to the viral membrane while the stem and head extend into the solvent and are free for molecular interaction. Potential glycosylations were not further considered.

Figure 9

Representation of the gp110 putative trimer surface and secondary structure cartoons in lateral and top view (left to right). Monomers are drawn in green, violet and cyan. Regions covered by predicted, potentially neutralizing epitiopes are shown in blue, residues predicted to be glycosylated are given in brown. Areas coded in red were experimentally shown to be neutralizing in homologous proteins of other herpesviruses, while areas coded in orange were additionally predicted as epitopes. The orange spot at the stem of the molecule indicates the terminus of a neutralizing epitope close to the N-terminus of the protein (unfortunately only partially resolved in the structure model).

Figure 10

A snapshot for visualizing T-cell antigenicity of the N-terminal LMP2A. Data from numerous prediction methods were integrated and visualized in form of an HTML table using a Perl framework. Rows contain aligned EBV sequences; colors indicate degree of antigenicity for a particular allele. The snapshot was selected for three spots of potential HLA-B3501 antigenicity. Bars below the alignments indicate (in this order) allele, start position of the ligand, as well as minimum and maximum IC50 in nM of nanomer peptides. Red indicates high affinity ligands (IC50 in nM around 1), blue indicates low affinity ligands (IC50 in nM around 500).

Figure 11

Color coded display of variability of antigenicity comparing virus isolates in a region associated with an experimentally determined epitope. NetCTL (left block) and NetMHC (right block) predictions for supertype A24 and specifically HLA-A2301 are shown, respectively. The area represented is centered around known A2301/supertype A24 ligand peptide ‘PYLFWLAA’ starting at alignment position 137. Sequences in the alignments are in the same order as in Figure 10, where the first four and last two sequences are not of EBV origin.