Archive Search Information Classified Subscribe

News & Views


Go to TOC
 
 
Previous Article AbstractFULL TEXTNext Article Abstract
Full Text PDF
 
 

References
FIGTAB

ISI related
 
doi:10.1038/nm0602-562
June 2002 Volume 8 Number 6 pp 562 - 563

 
 
Predicting evolutionary change in the influenza A virus
Neil M. Ferguson & Roy M. Anderson
 
Department of Infectious Disease Epidemiology, Faculty of Medicine, Imperial College of Science, Technology and Medicine, London University, London, UK e-mail: neil.ferguson@ic.ac.uk

Each year a new flu vaccine is produced, and judging which strains to target is a tricky business. A new study evaluating viral evolution suggests a more systematic approach to predicting next year's virus.

In any discussion of the public health impact of influenza, the catastrophic death toll of global pandemics such as that of 'Spanish flu' in 1918 comes first to mind1. But over the longer term the smaller flu epidemics that occur each winter between major pandemics are responsible for greater overall mortality. Underlying the damage caused by both forms of influenza is an unusual pattern of virus evolution. Pandemics are caused by rare events in which existing human strains recombine with bird or swine viruses, producing a novel (and potentially lethal) new virus to which the human population has had no prior exposure and therefore has no immune protection. The more common annual epidemics are the result of more gradual evolutionary processes termed 'antigenic drift'—the generation of new strains from existing ones through mutation. 'Drift' gives a somewhat misleading impression of the processes shaping influenza evolution, however. Although mutation is random, new variants are subject to the intense selective pressure2 of the immunity built up in the human population against their (antigenically similar) viral ancestors. Hence, although new viral lineages constantly emerge, this process of immunological selection generates fierce competition between strains, so on a timescale of a few years, only one survives and the others go extinct. The surviving dominant lineage is the one from which future variants arise. These dynamics give rise to the characteristic shape of the influenza A phylogenetic tree3—a dominant 'trunk' from which short branches emerge, describing a pattern of continuous genetic change through time, but approximately constant viral diversity at any instant (Fig 1a).


 
Top
References
FIGTAB

ISI related
 
Nature Medicine
 

Although it is of considerable theoretical interest, this pattern also has important implications for public health. The limited diversity of the virus at any point in time makes feasible the use of vaccines targeted against only a few viral strains, but viral evolution necessitates annual updating of that vaccine. Given that mass production of new vaccine takes several months, vaccine efficacy depends on accurately predicting the strain likely to dominate transmission in the next year. A major public health goal is therefore to refine the prediction techniques used to select vaccine strains. This challenging task motivates a recent study by Plotkin et al.4, which presents a novel approach to describing changes in the viral genome and, perhaps, to improving our ability to predict the unpredictable; namely, the direction of evolution.

Optimally, prediction methods should occur within a quantitative framework that melds the dynamics of both evolutionary and transmission processes. The growing volume of immunological and sequence data can help in this goal. The study by Plotkin et al.4 focuses on these data by interpreting the variability present in viral isolates recorded in the Los Alamos National Laboratory Influenza Sequence Database (http://www.flu.lanl.gov/). This database is the result of a global surveillance effort targeted at detecting new strains. Such global surveillance typically uses immunological screens to preselect isolates for antigenic novelty before sequencing—resulting in a set of sequences that are far from a random sample of circulating viral types5. To analyze these complex data, Plotkin et al. used statistical cluster analysis methods6 rather than the more usual phylogenetic models. There are some common elements to the two approaches (Fig 1bd), but also some key conceptual distinctions. Phylogenetic methods try to infer the ancestral sequences (sometimes long extinct) from which all the sequences examined emerged, whereas cluster analysis solely concentrates on ordering samples by some criterion of similarity to give insight into existing patterns of diversity.

Plotkin et al. classify viral isolates into non-overlapping clusters based on the amino-acid sequence of HA1, the immunogenic part of the surface hemagglutinin (HA) protein. They then propose that these clusters, or 'swarms', better characterize the scale at which influenza evolution occurs than individual strains—given that strains within the same swarm will be subject to very similar selective pressures from host immunity. Insight into viral evolution is then gained by analyzing how clusters change through time and examining the differences among sequences within and between clusters. This approach reveals a pattern of emergence of clusters of closely related strains, with a few clusters growing to dominate the viral population but then being replaced within two to five years. Perhaps most interestingly, by examining changes at key epitopic sites in HA1 alone, the authors suggest each transition from one dominant cluster to another is associated with substantial change at a single epitope, with the affected epitope varying from transition to transition.


 
Top
References
FIGTAB

ISI related
 
Nature Medicine
 

For vaccine selection (Fig 1e), it is the shorter-term evolutionary trends that are most relevant. The analysis of Plotkin et al. predicts that new strains are more likely to arise from the cluster of largest size (that is, the one containing the largest number of isolates collected over the previous year). Thus, focusing on that dominant cluster can improve predictions of the next season's influenza strain.

These are interesting results, but clearly more work is required before their techniques might replace current methods used in the annual process of vaccine strain designation by the World Health Organization. A number of key methodological issues also suggest caution at present. First, should the optimal measure of evolutionary distance between isolates be based on nucleotides, amino acids, stereochemical properties or immunogenic properties? Plotkin et al. use the number of HA1 amino-acid differences between two strains as a more phenotypically correlated description of isolate similarity than nucleotides (which give the greatest evolutionary information for phylogenetic analyses), but recognize the need for more research to develop predictive metrics capable of inferring antigenic properties from genotype data.

Second, although Plotkin et al. highlight the difficulties caused for phylogenetic analyses by the biased sampling scheme used to collect influenza sequence data5, the methods they adopt are also vulnerable to such biases. If viral evolution had been more intensively observed (for example, using 5,000 sequence samples, rather than 500), sampling all single amino-acid changes, then single-linkage cluster analysis would have grouped all sequences into the same cluster—dramatically reducing insight into evolutionary processes. A consequent concern is that the classification of strains into clusters is at least in part a result of phenotypically biased sampling or random temporal variation in sequence collection, as well as any underlying discrete directional changes in viral evolution. Non-random sampling also undermines assumptions that cluster size reflects the frequency of member strains in the overall viral population. These issues need to be addressed through use of cluster analysis methods that are more robust to sampling biases6, 7, and by placing greater emphasis on representative sampling.

Understanding how cluster analysis relates to phylogenetic studies (Fig 1) is also clearly important to a deeper insight into the dynamics of influenza evolution. The power of phylogenetic approaches is typified by past work identifying the HA1 gene to be under intense positive selection2. These studies demonstrated an association between mutation at 18 key codons and lineage survival that might form the basis of a vaccine-strain prediction algorithm3. Phylogenetic techniques (which are also a type of cluster analysis) have the advantage of being based on statistically robust models of the random mutational process governing nucleotide substitutions. Testing evolutionary hypotheses, sensitivity analysis and sample-size power calculations can all be undertaken more easily for such models5 than for the deterministic algorithms underlying simple cluster-analysis methods.

The novelty of the work of Plotkin et al. therefore lies less in the specifics of their analysis method, but more in the conceptual focus on influenza evolution as a dynamic process. The explicit representation of the development of 'swarms' of closely related viral strains over time gives new insight into processes of strain and quasispecies emergence and extinction. Despite the statistical concerns, by focusing less on inferring the microscopic structure of evolution, cluster methods do offer some potential advantages over conventional phylogenetic analyses—such as their ability to integrate both diversity and frequency data into approaches for predicting short-term viral evolution. Such ever more sophisticated descriptive analyses bring us closer to uncovering what is still missing from our understanding of influenza epidemiology and evolution: a robust theoretical framework capable of explaining the key population processes that shape the complex spatiotemporal patterns observed in epidemiological and sequence data.


 
Top
References
FIGTAB

ISI related
 
Nature Medicine
 
REFERENCES
  1. Glezen, W.P. Emerging infections: pandemic influenza. Epidemiol Rev. 18, 64-76 (1996). | PubMed  | ISI | ChemPort |
  2. Fitch, W.M., Leiter, J.M., Li, X.Q. & Palese, P., Positive Darwinian evolution in human influenza A viruses. Proc. Natl. Acad. Sci. USA 88, 4270-4270 (1991). | PubMed  | ChemPort |
  3. Bush, R.M., Bender, C.A., Subbarao, K., Cox, N.J. & Fitch, W.M. Predicting the evolution of human influenza A. Science 286, 1921-1925 (1999). | Article | PubMed  | ISI | ChemPort |
  4. Plotkin, J.B., Dushoff, J. & Levin, S.A. Hemagglutinin sequence clusters and the antigenic evolution of influenza A virus. Proc. Natl. Acad. Sci. USA 99, 6263-6268 (2002). | Article | PubMed  | ChemPort |
  5. Bush, R.M., Smith, C.B., Cox, N.J. & Fitch, W.M. Effects of passage history and sampling bias on phylogenetic reconstruction of human influenza A evolution. Proc. Natl. Acad. Sci. USA 97, 6974-6980 (2000). | Article | PubMed  | ChemPort |
  6. Gordon, A.D. Classification. (Chapman & Hall/CRC, London, 1999).
  7. Everitt, B.S., Landau, S. & Leese, M. Cluster Analysis. (Edward Arnold, London, 2001).

 
top

Home
Copyright 2002 Nature Publishing Group  Site
GuideFeedback