New Tools and Strategy for TEs Annotation in Large Genomes

The recent successes of new sequencing technologies allow today to sequence increasingly large genomes at reduced costs. Transposable elements (TEs) constitute the most structurally dynamic components and the largest portion of nuclear sequences of these large genomes, e.g. 85% of the maize genome (Schnable et al. 2009), and 88% of the wheat genome (Choulet et al. 2010). Therefore, TEs annotation should be considered a major task in these genome projects.

However, this still remains a major challenge, since a good TE annotation relies critically on an expertly assembled reference sequence set, data that currently cannot be obtained in an automatic fashion. This crucial step is now a bottleneck for many genome analyses. To this end, we scale up a repeat detection and an annotation pipeline, both part of the REPET package (Flutre et al. 2011, now at its v2.1 release). The two pipelines called TEdenovo and TEannot respectively build  a TEs library and annotate TE copies in the genome. Recently, TEdenovo has been improved by integrating LTRHarvest (Ellinghaus D et al. 2008) in its framework for a fast and accurate LTR-TE detection. In addition, we propose now three new pipelines, one based on Tallymer (Kurtz S et al 2009), called TallymerPipe, as pre-processing tool for a fast repeated region detection, another, called PostAnalyseTElib,that gets information about a TE library, and finally, SegDup, a pipeline to detect segmental duplications.

Using these pipelines, we apply a new strategy, to cope with very large genomes such as the wheat, an allohexaploid with three homoeologous genomes. It is one of the largest plant genomes with ~17Gbp and 88% of TEs (Choulet, 2010). We started with the 3B chromosome, the first to be fully sequenced. This strategy is an iterative approach and can be summarized as following:

1) Detection of the most easy to found TEs, with stringent parameters, to build a first TE library. They often corresponds to young TEs and the less degenerate ones,

2) TE annotation and splicing of the corresponding sequences from the initial contigs. We then obtain a reduced genome sequence.

3) Detection of the other TEs with sensitive parameters on the reduced genome sequence to build a second TE library.

4) Annotation of the original contigs with the concatenation of the two TE libraries.

The rational here is that these large genomes are mostly made of few TE families easy to found because present in number of copies. They will be detected in the first step and this will allow reducing the genome size by an important factor. Using this approach we were able to reduce the genome from 986Mbp to ~230Mb, a reasonable size for a detection of TEs with sensitive parameters.