|Résumé :||Bacteriophage genomes show pervasive mosaicism, indicating that horizontal gene exchange plays a crucial role in their evolution. Phage genomes represent unique combinations of modules, each of them with a different phylogenetic history. Thus, a web-like, rather than a hierarchical scheme is needed for an appropriate representation of phage evolutionary relationships. Part of the virology community has long recognized this fact and calls for changing the traditional taxonomy that classifies tailed phages according to the type of genetic materials and phage tail and head/capsid morphologies. Moreover, based on morphological features, the current system depends on inspection of phage virions under the electron microscope and cannot directly classify prophages. With the genomic era, many phages have been sequenced that are not classified, calling for development of an automatic classification procedure that can cope with the sequencing pace. The ACLAME database provides a classification of phage proteins into families and assigns the families with at least 3 members to one or several functions.
In the first contribution of this work, the relative contribution of those different protein families to the similarities between the phages is assessed using pair-wise similarity matrices. The modular character of phage genomes is readily visualized using heatmaps, which differ depending on the function of the proteins used to measure the similarity.
Next, I propose a framework that allows for a reticulate classification of phages based on gene content (with statistical assessment of the significance of number of shared genes). Starting from gene/protein families, we built a weighted graph, where nodes represent phages and edges represent phage-phage similarities in terms of shared families. The topology of the network shows that most dsDNA phages form an interconnected group, confirming that dsDNA phages share a common gene pool, as proposed earlier. Differences are observed between temperate and virulent phages in the values of several centrality measures, which may correlate with different constraints to rampant recombination dictated by the phage lifestyle, and thus with a distinct evolutionary role in the phage population.
To this graph I applied a two-step clustering method to generate a fuzzy classification of phages. Using this methodology, each phage is associated with a membership vector, which quantitatively characterizes the membership of the phage to the clusters. Alternatively, genes were clustered based on their ‘phylogenetic profiles’ to define ‘evolutionary cohesive modules’. Phages can then be described as composite of a set of modules from the collection of modules of the whole phage population. The relationships between phages define a network based on module sharing. Unlike the first network built from statistical significant number of shared genes, this second network allows for a direct exploration of the nature of the functions shared between the connected phages. This functionality of the module-based network runs at the expense of missing links due to genes that are not part of modules, but which are encoded in the first network.
These approaches can easily focus on pre-defined modules for tracing one or several traits across the population. They provide an automatic and dynamic way to study relationships within the phage population. Moreover, they can be extended to the representation of populations of other mobile genetic elements or even to the entire mobilome.
Finally, to enrich the phage sequence space, which in turn allows for a better assessment of phage diversity and evolution, I devise a prophage prediction tool. With this methodology, approximately 800 prophages are predicted in 266 among 800 replicons screened. The comparison of a subset of these predictions with a manually annotated set shows a sensitivity of 79% and a positive predictive value of 91%, this later value suggesting that the procedure makes few false predictions. The preliminary analysis of the predicted prophages indicates that many may constitute novel phage types.
This work allows tracing guidelines for the classification and analysis of other mobile genetic elements. One can foresee that a pool of putative mobile genetic elements sequences can be extracted from the prokaryotic genomes and be further broken down in groups of related elements and evolutionary conserved modules. This would allow widening the picture of the evolutionary and functional relationships between these elements.