Sensitivity of noninvasive prenatal detection of fetal aneuploidy from maternal plasma using shotgun sequencing is limited only by counting statistics

Abstract

We recently demonstrated noninvasive detection of fetal aneuploidy by shotgun sequencing cell-free DNA in maternal plasma using next-generation high throughput sequencer. However, GC bias introduced by the sequencer placed a practical limit on the sensitivity of aneuploidy detection. In this study, we describe a method to computationally remove GC bias in short read sequencing data by applying weight to each sequenced read based on local genomic GC content. We show that sensitivity is limited only by counting statistics and that sensitivity can be increased to arbitrary precision in sample containing arbitrarily small fraction of fetal DNA simply by sequencing more DNA molecules. High throughput shotgun sequencing of maternal plasma DNA should therefore enable noninvasive diagnosis of any type of fetal aneuploidy.

Figure 1. Illustration of the procedures used to remove GC content dependent artifact in shotgun sequencing data using one of the patient samples as example.

(A) The number of sequence tag per 20 kb bin is plotted against GC content of the bin. (B) The average number of sequence tag per 20 kb is calculated for every 0.1% GC content. (C) A weight is calculated for a particular value of GC content, such that sequence tags falling within a 20 kb bin having such GC content would receive the same calculated weight.

Figure 2. Removal of GC bias reveals that sampling of sequence tags closely follows the Poisson distribution.

(A) Sequence tag distribution (number of sequence tags per 50 kb bin) across each chromosome before (blue) and after (red) applying GC dependent weighting to sequence tags, using one patient sample as example. (B) Plot of variance against mean of sequence tag density for each chromosome in every sample. (C) Comparing the cumulative distribution of the sequence tag density against that predicted by the Poisson distribution using one patient sample as example. The red diagonal line has a slope of 1 and an intercept of 0.

Figure 3. Comparing the distribution of sequence tag density of each chromosome among 19 patients after correcting for GC bias.

(A) Distributions of normalized sequence tag density for each chromosome (excluding chromosome Y) within each patient sample. Red: chromosome 21; blue: chromosome X; magenta: chromosome 18; green: chromosome 13; black: all other autosomes. For all but one (P19) male pregnancy sample, it is obvious that the distribution of chromosome X shifts towards the left, while the distributions of chromosomes 21, 18, and 13 for the respective cases of trisomy 21, 18, and 13 shift towards the right, relative to the distributions of all other chromosomes that are present in two copies. (B) The sequence tag distribution of each chromosome is compared to all other chromosomes (except chromosome Y) by calculating the z-statistic. If we require that the copy number of a chromosome to be significantly different from that of all other chromosomes at level α<0.001 to be flagged as abnormal, chromosome X is under-represented as compared to a normal female genome in all but one male pregnancy (P19), while chromosomes 21, 18, and 13 are over-represented in the respective cases of trisomy 21, 18, and 13. Plotted here is the minimum z-statistic for each chromosome when it is compared against 22 other chromosomes. The horizontal dashed line corresponds to the statistic associated with α<0.001.

Figure 4. Estimation of the requirement of sequencing depth for the detection of fetal aneuploidy in cell-free plasma as a function of fetal DNA fraction.

The estimates are based on level of confidence α<0.001 for chromosomes 13, 18, 21, and X, each having different length. As fetal DNA fraction decreases, the total number of shotgun sequences required increases. With a sequencing throughput of ∼10 million sequence reads per channel on the flowcell, trisomy 21 can be detected if >3.9% of the cell-free DNA is fetal (dashed lines). The total number of sequence tags and the estimated fetal DNA fraction from our set of 19 patient samples are also plotted. For one of the normal male samples (P19, indicated by the solid arrow), chromosome X was not detected as under-represented. This was probably due to insufficient sampling, as the total number of sequence obtained for this sample was close to the limit of detection given its fetal DNA fraction.