Characterizing natural variation using next-generation sequencing technologies

Abstract

Progress in evolutionary genomics is tightly coupled with the development of new technologies to collect high-throughput data. The availability of next-generation sequencing technologies has the potential to revolutionize genomic research and enable us to focus on a large number of outstanding questions that previously could not be addressed effectively. Indeed, we are now able to study genetic variation on a genome-wide scale, characterize gene regulatory processes at unprecedented resolution, and soon, we expect that individual laboratories might be able to rapidly sequence new genomes. However, at present, the analysis of next-generation sequencing data is challenging, in particular because most sequencing platforms provide short reads, which are difficult to align and assemble. In addition, only little is known about sources of variation that are associated with next-generation sequencing study designs. A better understanding of the sources of error and bias in sequencing data is essential, especially in the context of studies of variation at dynamic quantitative traits.

Figure 1

A cumulative plot of the amount of DNA sequence (in bp) deposited in GenBank since 1995. The data were taken from http://www.ncbi.nlm.nih.gov/Genbank/genbankstats.html (data downloaded on August 10^th 2008; data for 2009 was not available yet).

Figure 2

RNAseq data from human and chimpanzee liver samples are plotted along the Vanin-family protein 3 (VNN3) gene region. As can be seen from the expanded section, the VNN3 gene is highly expressed in chimpanzee compared to human. However, using data collected with a multi-species microarray (4), we previously inferred that the VNN3 gene is highly expressed in humans compared with chimpanzee. The reason for the discrepancy between the two datasets is that the microarray probes were designed to complement a limited exonic region of the gene, which, in contrast to the entire gene, seems to be highly expressed in humans (the red lines represent the positions of the array probes for this gene).

Figure 3

A heuristic example of the inference that can be made from a combination of different sources of genomic data. In this example, we make the assumption that the resolution of all three approaches is sufficient to identify a specific causative nucleotide. In reality, at the moment at least, the resolution of these techniques is only sufficient to identify candidate small (few kb) genomic regions. A. Using RNAseq and genotyping data, an expression quantitative trait locus (eQTL) is identified. B. ChIPseq data indicates that the same eQTL also underlies differential binding of a transcription factor to the locus, thereby suggesting a regulatory mechanism that can explain the observed association between genotype and gene expression level. C. Bisulfite sequencing suggests that the mechanism that underlies difference in transcription factor binding across genotypes, and ultimately differences in gene expression levels, is DNA methylation.