A shot in the genome: how accurately do shotgun 454 sequences represent a genome?

BACKGROUND: Next generation sequencing (NGS) provides a valuable method to quickly obtain sequence information from non-model organisms at a genomic scale. In principle, if sequencing is not targeted for a genomic region or sequence type (e.g. coding region, microsatellites) NGS reads can be used as a genome snapshot and provide information on the different types of sequences in the genome. However, no study has ascertained if a typical 454 dataset of low coverage (1/4-1/8 of a PicoTiter plate leading to generally less than 0.1x of coverage) represents all parts of genomes equally. FINDINGS: Partial genome shotgun sequencing of total DNA (without enrichment) on a 454 NGS platform was used to obtain reads of Apis mellifera (454 reads hereafter). These 454 reads were compared to the assembled chromosomes of this species in three different aspects: (i) dimer and trimer compositions, (ii) the distribution of mapped 454 sequences along the chromosomes and (iii) the numbers of different classes of microsatellites. Highly significant chi-square tests for all three types of analyses indicated that the 454 data is not a perfect random sample of the genome. Only the number of 454 reads mapped to each of the 16 chromosomes and the number of microsatellites pooled by motif (repeat unit) length was not significantly different from the expected values. However, a very strong correlation (correlation coefficients greater than 0.97) was observed between most of the 454 variables (the number of different dimers and trimers, the number of 454 reads mapped to each chromosome fragments of one Mb, the number of 454 reads mapped to each chromosome, the number of microsatellites of each class) and their corresponding genomic variables. CONCLUSIONS: The results of chi square tests suggest that 454 shotgun reads cannot be regarded as a perfect representation of the genome especially if the comparison is done on a finer scale (e.g. chromosome fragments instead of whole chromosomes). However, the high correlation between 454 and genome variables tested indicate that a high proportion of the variability of 454 variables is explained by their genomic counterparts. Therefore, we conclude that using 454 data to obtain information on the genome is biologically meaningful.