PIPS: pathogenicity island prediction software

Abstract

The adaptability of pathogenic bacteria to hosts is influenced by the genomic plasticity of the bacteria, which can be increased by such mechanisms as horizontal gene transfer. Pathogenicity islands play a major role in this type of gene transfer because they are large, horizontally acquired regions that harbor clusters of virulence genes that mediate the adhesion, colonization, invasion, immune system evasion, and toxigenic properties of the acceptor organism. Currently, pathogenicity islands are mainly identified in silico based on various characteristic features: (1) deviations in codon usage, G+C content or dinucleotide frequency and (2) insertion sequences and/or tRNA genetic flanking regions together with transposase coding genes. Several computational techniques for identifying pathogenicity islands exist. However, most of these techniques are only directed at the detection of horizontally transferred genes and/or the absence of certain genomic regions of the pathogenic bacterium in closely related non-pathogenic species. Here, we present a novel software suite designed for the prediction of pathogenicity islands (pathogenicity island prediction software, or PIPS). In contrast to other existing tools, our approach is capable of utilizing multiple features for pathogenicity island detection in an integrative manner. We show that PIPS provides better accuracy than other available software packages. As an example, we used PIPS to study the veterinary pathogen Corynebacterium pseudotuberculosis, in which we identified seven putative pathogenicity islands.

Figure 1. Flowchart presenting each PAI analysis step performed by PIPS.

The procedure is divided into the following steps: (A) data treatment; (B) automatic analyses; and (C) manual analyses.

Figure 2. ROC curve showing the sensitivity and specificity of the Perl script for the identification of regions with GC content deviation.

Y-axis: sensitivity; X-axis: 100-specificity. The higher the accuracy is, the closer the curve is to the upper-left corner.

Figure 3. Frequencies of PAI features within the PICPs and in the full genomes of C. pseudotuberculosis strains 1002 and C231.

Y-axis: frequency in percentage; X-axis: PICPs and genomes of C. pseudotuberculosis strains 1002 and C231. The frequencies of the features in each PICP and in the whole genomes of the two strains are represented in the following colors: blue for codon usage deviation; red for GC content deviation; green for virulence factors; and purple for hypothetical proteins.

Figure 4. PICP3 and PICD3 (top and bottom, respectively) in the C. pseudotuberculosis and C. diphtheriae genomes.

Cp1002 and C. diphtheriae NCTC 13129 are shown at the top and bottom, respectively. Regions of similarity between the two genomes are marked in pink. Regions of similarity between two PAIs are marked in yellow, showing the presence of PICD3 in C. pseudotuberculosis with an insertion. Image generated by ACT (the Artemis Comparison Tool).

Figure 5. Replacement of the C. diphtheriae PICD8 (bottom) with C. pseudotuberculosis PICP5 (top).

Regions of similarity are represented by lines between the two genomes. The flanking regions of PICD8 and PICP5 are highlighted in yellow, showing the region of replacement. Image generated by ACT (the Artemis Comparison Tool).