Enrichment of sequencing targets from the human genome by solution hybridization

Abstract

To exploit fully the potential of current sequencing technologies for population-based studies, one must enrich for loci from the human genome. Here we evaluate the hybridization-based approach by using oligonucleotide capture probes in solution to enrich for approximately 3.9 Mb of sequence target. We demonstrate that the tiling probe frequency is important for generating sequence data with high uniform coverage of targets. We obtained 93% sensitivity to detect SNPs, with a calling accuracy greater than 99%.

Figure 1

Experimental design. Genomic DNA fragment libraries were generated from two samples, Coriell (NA15510) and Wellderly (HE00069). Technical replicates of the target-enrichment steps for both samples NA15510 and HE00069 were performed (Capture 1 and Capture 2). The four target-enriched samples were loaded in separate lanes of a flow cell and sequenced by using the Illumina GAII.

Figure 2

Distribution of capture probes. (a) The 196-kb targeted 9p21 genomic interval (positions 21,938,000-22,134,000, top panel) and a magnified view (positions 21,955,500 to 22,002,000) within the interval (bottom panel). The locations of the 120-mer capture probes designed by eArray are shown (red bars). Capture probes were designed to all sequences in the interval except for repetitive elements marked as Repeat masked (black bars). Exons (grey rectangles) and introns (grey lines) for genes in the interval are shown. (b) A 9-kb interval encoding the 3' UTR of the targeted FOXO1gene. Capture probes were designed to FOXO1 exons (grey bars) and ECS (blue bars), such as the one on the right end of the panel. The 2× probe tiling-frequency parameter results in adjacent 120-bp probes overlapping by 60 bp.

Figure 3

Efficiency of target enrichment. The pie chart on the left illustrates the relative percentages of the targeted sequences, LINE elements, SINE elements, and all "other" sequences in the human genome reference sequence (3.08 Gb in total). The pie chart on the right shows the relative percentages of these sequences in the filtered sequence reads (293 Mb in total). Targeted sequences include those on or near (target ± 150 bp) target.

Figure 4

Uniformity of sequence coverage. Normalized coverage is the observed coverage of each base divided by the mean coverage of all the targeted bases to allow direct comparison among the four target-enriched samples. (a) Distribution of the normalized sequence coverage. The solid lines represent the cumulative fraction of bases (left axis) for each sample. The dashed lines (right axis) show a skewed normal distribution of the coverage for each sample. (b) A scatterplot of the normalized coverage of each capture probe versus the GC content of the probe. Normalized coverage was calculated by averaging across the four samples. (c) A box-whisker plot of the normalized coverage of capture probes versus tiling-probe frequency (see Methods). Each bin from 1× to 4× contains the data of 6,007, 15,513, 27,483, 1,132, and 3,184 probes, respectively. (d) A scatterplot of the normalized coverage versus the length of each targeted Exon/ECS; the x-axis is truncated because only a handful of exons are larger than 4 kb. The solid blue lines of (c, d) represent a polynomial regression of the scatterplot.

Figure 5

Reproducibility of target enrichment. (a) The normalized mean coverage of each capture probe is plotted for the technical replicates of NA15510 and HE00069 (Capture 1 versus Capture 2). Capture probes that lie outside the dashed lines on the plot have normalized coverage that differs by more than twofold in the technical replicates. (b) The normalized mean coverage of each capture probe is plotted for NA15510 (Capture 1) versus HE00069 (Capture 1). The values of coefficient of determination (r²) are shown.

Figure 6

Accuracy of sequence variant calls compared with microarray genotype calls. The detection rate (dashed lines) and concordance (solid lines) of variant calls versus the MAQ quality score [33] are shown for target sequences (a) and on or near (target ± 150 bp) targets (b). A filter requiring five or more reads was first applied, and then the detection and concordance rates at the various MAQ quality score thresholds was determined.