Using complete genome comparisons to identify sequences whose presence accurately predicts clinically important phenotypes.

Abstract

In clinical settings it is often important to know not just the identity of a microorganism, but also the danger posed by that particular strain. For instance, Escherichia coli can range from being a harmless commensal to being a very dangerous enterohemorrhagic (EHEC) strain. Determining pathogenic phenotypes can be both time consuming and expensive. Here we propose a simple, rapid, and inexpensive method of predicting pathogenic phenotypes on the basis of the presence or absence of short homologous DNA segments in an isolate. Our method compares completely sequenced genomes without the necessity of genome alignments in order to identify the presence or absence of the segments to produce an automatic alignment of the binary string that describes each genome. Analysis of the segment alignment allows identification of those segments whose presence strongly predicts a phenotype. Clinical application of the method requires nothing more that PCR amplification of each of the set of predictive segments. Here we apply the method to identifying EHEC strains of E. coli and to distinguishing E. coli from Shigella. We show in silico that with as few as 8 predictive sequences, if even three of those predictive sequences are amplified the probability of being EHEC or Shigella is >0.99. The method is thus very robust to the occasional amplification failure for spurious reasons. Experimentally, we apply the method to screening a set of 98 isolates to distinguishing E. coli from Shigella, and EHEC from non-EHEC E. coli strains and show that all isolates are correctly identified.