Single-stranded DNA library preparation uncovers the origin and diversity of ultrashort cell-free DNA in plasma

Abstract

Circulating cell-free DNA (cfDNA) is emerging as a powerful monitoring tool in cancer, pregnancy and organ transplantation. Nucleosomal DNA, the predominant form of plasma cfDNA, can be adapted for sequencing via ligation of double-stranded DNA (dsDNA) adapters. dsDNA library preparations, however, are insensitive to ultrashort, degraded cfDNA. Drawing inspiration from advances in paleogenomics, we have applied a single-stranded DNA (ssDNA) library preparation method to sequencing of cfDNA in the plasma of lung transplant recipients (40 samples, six patients). We found that ssDNA library preparation yields a greater portion of sub-100 bp nuclear genomic cfDNA (p 10(-5), Mann-Whitney U Test), and an increased relative abundance of mitochondrial (10.7x, p 10(-5)) and microbial cfDNA (71.3x, p 10(-5)). The higher yield of microbial sequences from this method increases the sensitivity of cfDNA-based monitoring for infections following transplantation. We detail the fragmentation pattern of mitochondrial, nuclear genomic and microbial cfDNA over a broad fragment length range. We report the observation of donor-specific mitochondrial cfDNA in the circulation of lung transplant recipients. A ssDNA library preparation method provides a more informative window into understudied forms of cfDNA, including mitochondrial and microbial derived cfDNA and short nuclear genomic cfDNA, while retaining information provided by standard dsDNA library preparation methods.

Figure 1. Schematic of sequencing library preparation methods.

Schematic illustration of key steps in the dsDNA and ssDNA library preparation protocols used in this work and their sensitivity to different types and forms of circulating cfDNA in plasma. cfDNA in plasma may be single-stranded (light blue), partially single-stranded or nicked (dark blue), short (sub-100 bp, red), or long (super-100 bp, green).

Figure 2. cfDNA fragment length distributions.

(A) Abundance of mitochondrial (green) and nuclear genomic cfDNA (black) measured by digital PCR assays with different amplicon lengths. Solid lines are model fits (see SI). (B,C) Fragment length histograms (frequency relative to total nuclear genomic) measured via sequencing for mitochondrial (B) and microbial (C) cfDNA following ssDNA (blue) and dsDNA (red) library preparation. (D) Density plot of the fragment sizes of nuclear genomic cfDNA measured after ssDNA (blue) and dsDNA (red) library preparation. The inset shows the sample-to-sample variability, as well as the difference in GC content for short (<100 bp) and long (>100 bp) fragments. (E) Density (scaled for clarity) of short length (segment lengths <100 bp) mitochondrial, microbial and nuclear genomic cfDNA measured by ssDNA library preparation. Vertical lines highlight most prevalent fragment lengths.

Figure 3. ssDNA library preparation yields greater fraction of non-human cfDNA.

(A) Comparison of the coverage of microbial genomes relative to the human genome for ssDNA and dsDNA library preparation. Data points are colored by domain of life. (B) Yield of bacterial sequences for ssDNA library preparation relative to dsDNA library preparation (74-fold mean increase). (C) Venn diagram representation of the number of species uniquely detected following ssDNA library preparation in blue (540/984, 54.9%), species uniquely detected following dsDNA library preparation in red (50/984, 5.1%), and species detected following both protocols (394/984, 40.0%).

Figure 4. Quantifying donor-specific nuclear genomic and mitochondrial cfDNA.

(A) Comparison of the fraction of donor-specific nuclear genomic DNA measured after dsDNA and ssDNA library preparation. In the inset, fraction of cfdDNA as function of time after transplant for a patient who suffered a severe acute rejection event at month 12 (cfdDNA measured by ssDNA (orange) and dsDNA (blue) library preparation). (B) Schematic representation of analysis workflow used to discriminate donor and recipient specific mt-cfDNA. Examples of an ambiguous assignment and a fragment assigned to the donor are shown. (C) Fraction of donor-specific mt-cfDNA as function of time post-transplant for five double lung transplant patients (25 samples, having excluded samples with fewer than 20 informative mitochondrial fragments); the inset compares the fraction of donor-specific mitochondrial and nuclear genomic DNA for the same samples (corr. = 0.463, Pearson, p = 0.0196). (D) Smoothed (distribution five nearest-neighbors, running mean) of donor mt-cfDNA is compared to the smoothed distribution of recipient mt-cfDNA. Inset: median fragment size for the donor mt-cfDNA compared to the fragment size of 10,000 subsets sampled from the recipient mt-cfDNA length set.