Dengue virus has four serotypes (DV1, DV2, DV3 and DV4) that show substantial genetic diversity both within and between serotypes. Sequence comparison studies showed 30–40% difference in amino acid sequences between serotypes 12. The amino acid differences within each serotype are lower but the observed intra-serotype diversity is sufficiently large to warrant the definition of clusters of dengue virus variants 34. Studies of genetic diversity have focused on clade diversity and replacement 4, mutation spectra 5, conserved regions 6 and implications for clinical manifestations 3. Several studies have focused on the analysis of antigenic diversity (diversity of targets of immune responses in protein sequences) of dengue virus; these studies focussed on experimental mapping of antibody recognition sites 7891011 and T-cell epitopes 212131415 and subsequent analysis of their diversity. Recently, Simmons et al. 15 analyzed the T cell responses of individuals infected with DV2 by ELISpot assay and identified 34 peptides of several dengue proteins as potential novel T-cell epitopes.

Generally, there is a correspondence between genetic and antigenic evolution of viruses, but genetic changes may result in disproportionately large antigenic changes 1617. While genetic and antigenic diversity in dengue virus strains had become evident 18, large-scale and detailed systematic analyses that explore their relationship have not been reported. Understanding this relationship is important for the study of vaccine development, especially in rapidly mutating viruses. In this paper, we will focus on protein sequence diversity, and thus consider only genetic variations that affect the protein sequences.

Biological studies of antigenic diversity require great experimental effort, even for a single viral protein. Consequently, most research groups focus on studying small number of viral sequences. Rapid accumulation of sequence data from both classical and genomic/proteomic approaches makes the experimental studying of antigenic diversity difficult and time-consuming. A bioinformatics approach is necessary to support large-scale antigenic analysis of viral diversity, which can complement laboratory experiments.

In this study, we have developed a bioinformatics method to analyze antigenic diversity in the context of T-cell mediated immune responses. We studied antigenic diversity of more than 9000 dengue virus protein sequences reported in the NCBI Entrez protein database 19. The study aimed to identify a minimal set of sequences that encodes the complete antigenic diversity of short peptides from all known sequences in dengue virus serotypes. Short peptides, principally 9-mers were studied because they represent the predominant length of binding cores of T-cell epitopes. We analysed the relationship between short-peptide antigenic diversity and protein sequence diversity of dengue virus; the analysis was performed at two time points to help understand the effects of the accumulation of sequence data to the relationship. We have also analyzed the effects of sequence determinants on antigenic diversity of short peptides. This study provides a framework for large-scale, systematic analysis of antigenic diversity for the protein sequences of any virus. We did not analyze B-cell epitope antigenic diversity because of their complex conformational nature. Although linear B-cell epitopes exist and our method can be used to study them, very often, they also show conformational preferences and dependence on the context of a protein antigen 20. Further, only approximately 10% of B-cell epitopes from native proteins are linear 21.

In this study, we applied a systematic bioinformatics approach to collect, clean, organize and analyze the antigenic diversity of short peptides in reported protein sequence data of dengue virus. We have developed a computational method for the analysis of antigenic diversity in the context of T-cell mediated immune responses. The method was applied for the analysis of short-peptide antigenic diversity of dengue virus to determine a minimal sequence set that encodes the complete antigenic diversity of linear epitopes within each dengue virus serotype. We studied the relationship between short-peptide antigenic diversity and protein sequence diversity of DV and also explored the effects of sequence determinants on viral antigenic diversity. Our analysis showed that the minimal number of unique sequences required to represent complete antigenic diversity of linear epitopes in dengue virus is significantly smaller than that required to represent complete protein sequence diversity. Short-peptide antigenic diversity shows an asymptotic relationship to the number of unique sequences and linear relationship to the length of protein antigens.

The minimal sequence set that encodes the complete short-peptide antigenic diversity for each dengue virus serotype was derived through removal of identical sequences and antigenically redundant sequences (Table 5 and Figure 4). Both reductions occurred without any loss of information on antigenic diversity among the sequences. The largest reduction was accomplished through the removal of identical sequences, since only 36% (year 2004) or 25% (year 2005) of the sequences were unique. The identical sequences originated from dengue virus strains that were unique variants with respect to the whole polyprotein, but were identical to other dengue strains with respect to individual proteins, resulting in many duplicate protein sequences. The removal of antigenically redundant sequences also involved a significant proportion of the sequences, approximately one-third of all unique sequences (2004: 26%; 2005: 30%), reflecting the high antigenic redundancy among the dengue virus variants, which often differed by only a few amino acids. Despite significant reduction achieved by reducing the collected sequences to minimal sequences, a large number of protein sequences, 969 in 2004 and 1684 in 2005, were still required to represent the complete short-peptide antigenic diversity of dengue virus.

It is clear that antigenic diversity in the reported dengue sequences is large. With many asymptomatic human and animal carriers of dengue viruses representing a huge reservoir for emergence of new strains 62428, the diversity is expected to increase, although at a progressively slower pace. This is because antigenic redundancy increases when the number of sequences increases; we observed that when the dataset for a particular protein reaches approximately 200 sequences, the effect of addition of new sequences to increasing antigenic diversity is marginal.

Our study of factors that affect antigenic diversity provided insight into dealing with the increasing T-cell epitope antigenic diversity in the context of vaccine development. Length of sequences had the largest effect on short-peptide antigenic diversity. The asymptotic behaviour of antigenic diversity increase was observed for the increase in the number of sequence variants. For practical purposes of vaccine formulation, antigenic diversity cannot be represented by whole protein sequences because it is not feasible to use these sequences for systematic experimental analysis: they are long and their number is increasing rapidly. The implication is that conventional vaccination strategies, which utilize whole attenuated pathogen with little knowledge of the specificity of immune responses they elicit, may not be suitable for providing protection from multiple variants of viruses. Furthermore, it may be difficult to optimize such vaccine according to the human leukocyte antigen (HLA) profile of the population receiving the vaccine 2930, as neither the identities of the HLA molecules that bind T-cell epitopes, nor the epitopes themselves are known.

The more effective vaccine strategy that we propose is to focus on short segments of proteins (~<100 aa) that are known to be specific targets of immune responses (such as T-cell epitopes specific to particular HLA alleles), particularly those that have high concentration of T-cell epitopes 31. By combining selected sets of short antigen fragments that represent T-cell epitope antigenic diversity, complete sets of viral targets can be covered in a "divide-and-conquer" approach. This may provide a promising basis for multivalent peptide-based vaccines against dengue virus. However, this strategy does not address the dengue virus-specific problem of protection versus immunopathology during secondary infections with a different serotype 2.

Several caveats need to be considered in a study such as this. First, it is well-known that not all HLA-restricted epitopes are 9-mers 32. This may impact the interpretation of our results, which were based only on 9-mers, and hence may not give a true representation of dengue T-cell epitope antigenic diversity. We selected 9-mers because they represent the typical size of HLA class I T-cell epitopes, as well as the binding core of HLA class II T-cell epitopes 32. We performed the same analysis with peptides of 8-mers and 10-mers. The results showed no significant difference as compared to the analysis of 9-mers (data not shown).

The second caveat is the sampling bias in dengue virus sequences reported to the public databases. Only dengue sequences that have been studied are reported, and viruses collected in accessible locations, associated with notable disease outbreaks or of known immunological properties are preferentially studied. Consequently, certain dengue proteins have been studied intensively, while the others remained largely unstudied. For example, sequences of the envelope protein, known to be important for immunological activity and viral entry into host 2633, were the most abundant in our dataset (3183 sequences for all four serotypes), while that of NS4a, which is relatively unknown for immunological activity, was under-represented. In addition, for majority of the proteins, a large portion of the reported sequences were incomplete in length. For example, 95% of DV2 NS5 collected sequences were incomplete in length (data not shown). However, the data used in this study was the most representative available and the large sample size for majority of the proteins helps to decrease the margin of error due to sampling bias. In addition, the reported sequences represent highly pathogenic strains isolated during dengue outbreaks.

There has been no significant increase in the number of unique sequences for dengue virus since the last analysis (December 2005). The September 2006 data set contained a total of 2661 (793 DV1, 784 DV2, 759 DV3 and 325 DV4) dengue unique sequences. This was an increase of 242 unique sequences from the 2005 data set. The increase, approximately 10%, was not expected to significantly affect the results observed for 2005 data set. Therefore, we did not perform the analysis of antigenic diversity on the 2006 data set because of the small increase in the number of unique sequence.

This study has provided evidence that there are limited numbers of antigenic combinations in variant protein sequences of a viral species and that short regions of the viral proteins are sufficient to capture antigenic diversity of T-cell epitopes. The approach described herein has direct application to the analysis of other viruses, in particular those that show high diversity and/or rapid evolution, such as influenza A virus and human immunodeficiency virus (HIV).

DV- Dengue Virus; DV1- Dengue Virus Serotype 1; DV2- Dengue Virus Serotype 2; DV3- Dengue Virus Serotype 3; DV4- Dengue Virus Serotype 4; aa- amino acids; NCBI- National Center for Biotechnology Information; HIV- Human Immunodeficiency Virus; HLA- Human Leukocyte Antigen.

The author(s) declare that they have no competing interests.

AMK performed the in silico experiments and drafter the manuscript. ATH, KXL, KNS, TWT and JTA participated in the design of the study. VB conceived the study, participated in its design and coordination and helped to draft the manuscript. JTA and TWT critically reviewed the manuscript. All authors read and approved the final manuscript.