Asymmetrical barcode adapter-assisted recovery of duplicate reads and error correction strategy to detect rare mutations in circulating tumor DNA

Abstract

Deep sequencing is required for the highly sensitive detection of rare variants in circulating tumor DNA (ctDNA). However, there remains a challenge for improved sensitivity and specificity. Maximum-depth sequencing is crucial to detect minority mutations that contribute to cancer progression. The associated costs become prohibitive as the numbers of targets and samples increase. We describe the targeted sequencing of KRAS in plasma samples using an efficient barcoding approach to recover discarded reads marked as duplicates. Combined with an error-removal strategy, we anticipate that our method could improve the accuracy of genotype calling, especially to detect rare mutations in the monitoring of ctDNA.

Figure 1. A schematic of the asymmetrical barcoding method.

(a) Design of the asymmetrical random index sequencing adapter oligonucleotide. Self-annealing sites 1 and 2 are referred to as R1 and R2, respectively. (b) Sequencing library-preparation scheme using the asymmetrical random index sequencing adapter. The self-annealed barcodes in a were used in the adapter ligation step and subsequently amplified using the P7 flanking primer and the P5 index primer. (c) Comparison of the removal of duplicate candidates with a conventional duplication removal strategy. Molecules with the same barcodes and the same start/end positions were considered true duplicate pairs.

Figure 2. Analysis of the performance of the barcoding strategy in five clinical samples.

The observed allele frequency distributions after error correction for each sample are plotted. The central peaks (amino acid numbers 12 and 13) of chromosome 12 in the graph are the primary mutations detected in the sample. The major-mutation peaks of samples (a) ctDNA1, (b) ctDNA2, (c) ctDNA3, (d) ctDNA4 and (e) ctDNA5 are indicated by the arrows. The allele frequency was calculated within a specific 200-bp region of chromosome 12.

Figure 3. Simulation of sequencing reads in the small target region (KRAS gene, five exonic regions) and the impact of the barcoding strategy.

(a) Size distribution of the sequencing data from the ctDNA and tissue DNA. Two different distributions were used to model each sample type. (b) Simulation of the estimated duplication fraction according to the sample type and number of sequencing reads. The duplication rate of the ctDNA was higher than that of the tissue DNA. (c) Simulation of the duplication rate according to the target size (assuming 0.1% sequencing errors).