Exploring Pandora's Box: Potential and Pitfalls of Low Coverage Genome Surveys for Evolutionary Biology

High throughput sequencing technologies are revolutionizing genetic research. With this "rise of the machines", genomic sequences can be obtained even for unknown genomes within a short time and for reasonable costs. This has enabled evolutionary biologists studying genetically unexplored species to identify molecular markers or genomic regions of interest (e.g. micro- and minisatellites, mitochondrial and nuclear genes) by sequencing only a fraction of the genome. However, when using such datasets from non-model species, it is possible that DNA from non-target contaminant species such as bacteria, viruses, fungi, or other eukaryotic organisms may complicate the interpretation of the results. In this study we analysed 14 genomic pyrosequencing libraries of aquatic non-model taxa from four major evolutionary lineages. We quantified the amount of suitable micro- and minisatellites, mitochondrial genomes, known nuclear genes and transposable elements and searched for contamination from various sources using bioinformatic approaches. Our results show that in all sequence libraries with estimated coverage of about 0.02-25%, many appropriate micro- and minisatellites, mitochondrial gene sequences and nuclear genes from different KEGG (Kyoto Encyclopedia of Genes and Genomes) pathways could be identified and characterized. These can serve as markers for phylogenetic and population genetic analyses. A central finding of our study is that several genomic libraries suffered from different biases owing to non-target DNA or mobile elements. In particular, viruses, bacteria or eukaryote endosymbionts contributed significantly (up to 10%) to some of the libraries analysed. If not identified as such, genetic markers developed from high-throughput sequencing data for non-model organisms may bias evolutionary studies or fail completely in experimental tests. In conclusion, our study demonstrates the enormous potential of low-coverage genome survey sequences and suggests bioinformatic analysis workflows. The results also advise a more sophisticated filtering for problematic sequences and non-target genome sequences prior to developing markers.