Policy Document


Site requirements

1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | 12 | 13 | Figures | next


BURST was devised and developed by Ed Feil (University of Bath; e.feil@bath.ac.uk), while he was in the laboratory of Brian Spratt (now at Imperial College London; b.spratt@imperial.ac.uk), as a way of displaying the relationships between closely-related isolates of a bacterial species or population*. BURST, unlike cluster diagrams, trees or dendrograms, uses a simple but appropriate model of bacterial evolution in which an ancestral (or founding) genotype increases in frequency in the population, and while doing so, begins to diversify to produce a cluster of closely-related genotypes that are all descended from the founding genotype. This cluster of related genotypes is often referred to as a “clonal complex”.

The BURST algorithm first identifies mutually exclusive groups of related genotypes in the population (typically a multilocus sequence typing [MLST] database), and attempts to identify the founding genotype of each group (see here). The algorithm then predicts the descent from the predicted founding genotype to the other genotypes in the group, displaying the output as a radial diagram, centred on the predicted founding genotype. The procedure was developed for use with the data produced by MLST (sequence types or STs, and their allelic profiles), but using a suitable criterion for the definition of groups of related genotypes, it could be used with some other types of data, particularly data from multilocus enzyme electrophoresis (MLEE) or from molecular typing methods that produce strings of integers by using multiple repeat length polymorphisms.

The approach used in BURST greatly simplifies the problem of depicting the evolutionary relationships among closely related genotypes, which are poorly represented on a tree, as the focus for each genotype is its relationship to, and distance from, its predicted founding genotype, and the great majority of the pair-wise relationships between genotypes in the population are ignored. BURST does not make any inferences about the relationships between the more distantly related genotypes that belong to different groups. In many bacterial populations, the relationships between distantly-related genotypes may be difficult to discern, as the extent of homologous recombination may be sufficiently high that relationships may be poorly represented by a phylogenetic tree, and should be represented by a network.

The original version of BURST was implemented by Ed Feil and Man-Suen Chan (man-suen.chan@paediatrics.ox.ac.uk). A greatly enhanced version of BURST (eBURST v1) was developed as a Java™ applet by Bao Li (Bao@mac.com), and was integrated into the MLST websites by David Aanensen (d.aanensen@imperial.ac.uk), within the laboratory of Brian Spratt, with help and advice from Ed Feil, Jon Evans (jemevans@yahoo.com), Bill Hanage (w.hanage@imperial.ac.uk) and Christophe Fraser (c.fraser@imperial.ac.uk). The original version of BURST is still available at www.mlst.net and, if necessary, may be used in conjunction with eBURST, as the two algorithms differ slightly in the way they display the relationships between STs.

The enhanced version has been further developed by Bao Li, to produce eBURST v2, which introduces several new features:

  • Auto-editing – eBURST diagrams are produced that need minimal manual editing
  • Zooming, rotating, stretching and moving subgroups - New manual editing features
  • Average distances – the average distance of each ST to all other STs in an eBURST group
  • Subgroup bootstraps – bootstrap support for STs being subgroup founders
  • SLV and DLV links – all SLVs or DLVs of any ST (or multiple STs etc) can be displayed

* Feil, E.J., Li, B., Aanensen, D.M., Hanage, W.P. and Spratt, B.G. 2004. eBURST: Inferring patterns of evolutionary descent among clusters of related bacterial genotypes from multilocus sequence typing data. J. Bact. 186: 1518-1530.