Analysis of circulating tumour DNA in plasma (ctDNA) is technically challenging, as it is usually dominated by cell-free DNA of non-cancerous origin 1,2. To increase the signal-to-noise ratio, current strategies focus on a subset of the genome 2-4. For untargeted sequencing, the only option is to increase depth of coverage, thus raising sequencing costs. Since it has been observed that tumour DNA in plasma tends to be shorter than non-tumour DNA 5,6, we hypothesised that filtering fragments by size could reduce the non-tumour fraction.
We established a catalogue of pan-cancer ctDNA fragmentation with shallow whole genome sequencing and characterised fragment lengths in 300 plasma samples from 183 individuals with 12 different cancer types and 48 healthy controls. We then quantified different facets of the fragment size distribution in each sample such as the periodicity, modality and the proportion of DNA in different size ranges. The ratio of fragment proportions between 20-150 bp and 180-220 bp was identified as a key feature that allowed the accurate classification of plasma samples as cancerous or healthy (AUC=0.855).
We studied the feasibility of ctDNA enrichment by in-vitro and in-silico size-selection in a subset of patients (N=51). An agarose gel-based filtering for DNA fragments between 90-150 bp, yielded enrichment of mutated DNA fractions up to 118-fold, in samples with starting ctDNA levels < 5%. Strikingly, in-silico selection of fragments from whole-exome sequencing alone allowed increased allele fractions and more mutation calls. Notably, 28% of mutations that were identified only in the plasma after in-silico size-selection were also present in the matched tumour tissue and included alterations in key genes such as BRAF, ARID1A, and NF1.
In conclusion, fragmentation features can be used to determine the presence of ctDNA with no prior knowledge of somatic alterations. Size-selection of fragments greatly improves the detection of tumour alterations in plasma, with potential to increase the sensitivity of liquid biopsies for early diagnosis and detection of minimal residual disease.