A quantitative comparison of single-cell whole genome amplification methods

Abstract

Single-cell sequencing is emerging as an important tool for studies of genomic heterogeneity. Whole genome amplification (WGA) is a key step in single-cell sequencing workflows and a multitude of methods have been introduced. Here, we compare three state-of-the-art methods on both bulk and single-cell samples of E. coli DNA: Multiple Displacement Amplification (MDA), Multiple Annealing and Looping Based Amplification Cycles (MALBAC), and the PicoPLEX single-cell WGA kit (NEB-WGA). We considered the effects of reaction gain on coverage uniformity, error rates and the level of background contamination. We compared the suitability of the different WGA methods for the detection of copy-number variations, for the detection of single-nucleotide polymorphisms and for de-novo genome assembly. No single method performed best across all criteria and significant differences in characteristics were observed; the choice of which amplifier to use will depend strongly on the details of the type of question being asked in any given experiment.

Figure 1. Design of experiments.

Overview of experiments, where n denotes the number of experiments of a given type. In the “E. coli single cells” column, the box that straddles both the “Microfluidic” and the “Tube” fields corresponds to the method of carrying out a first round of amplification in a microfluidic chamber and then a second round of amplification in a test tube. This method will be denoted by “microfluidic+tube” in subsequent figure captions.

Figure 2. Sequence read classification.

(A) Breakdown of read pairs in each experiment according to type of mapping achieved. (B) Breakdown of unmapped reads by organism of origin, expressed as a fraction of the total number of reads.

Figure 3. Amplification bias and uniformity.

(A) Local mapping density from properly mapped reads (at fixed 5x sampling depth), normalized to average 1, as a function of position along the reference sequence, for single-cell MDA (microfluidic in red, microfluidic+tube in orange). (B) Same as panel A, but for single-cell MALBAC and bulk NEB-WGA. (C) Fractional genome coverage from properly mapped read pairs, plotted as a function of gain. Here, each set of properly mapped read pairs was randomly down-sampled to 5x depth. Experiments that did not generate this many properly mapped reads were not included in the figure. (D) Fraction of the genome covered by mapped read pairs when the set of raw read pairs was down-sampled to a fixed depth of 20x, plotted as a function of gain. Filled symbols signify bulk experiments, open symbols single-cell experiments.

Figure 4. CNV resolution.

Size of resolvable duplications (W, minimum width of a sliding window average filter that gives rise to a relative genome mapping density smaller than 2 across all positions in the genome) versus gain. Filled symbols signify bulk experiments, open symbols single-cell experiments.

Figure 5. Combined single-nucleotide error rates.

Main panel: Experimental error rates D versus gain G, for low gains. Here D is the fraction of bases differing from the reference in the mapped reads. Linear fits for D as a function of are also shown: their slope approximately indicates the per-base per-cycle replication error rate. Inset: D versus G over the entire gain range. Filled symbols signify bulk experiments, open symbols single-cell experiments.

Figure 6. De-novo assemblies.

LG50, the minimal number of assembled contigs (≥500 bp) needed to cover 50% of the E. coli reference genome, versus reaction gain (at fixed raw sequencing depth 30x). Assemblies that failed to cover 50% of the reference sequence were symbolically assigned the maximum value that LG50 can take in this scenario ((50%•genome length/500) = 4686). Filled symbols signify bulk experiments, open symbols single-cell experiments.