*Purpose: Rejection is a major post-transplant complication for organ transplant recipients. Biopsy remains the standard to diagnose rejection, but other non-invasive methods, such as transcript levels, cytokine biomarkers and detection of donor derived cell-free DNA (ddcfDNA) have been studied. An ideal method would be sensitive, specific, objective, non-invasive, cost-effective and be able to detect early rejection. Inferring the proportion of ddcfDNA in plasma is a powerful, non-invasive diagnostic method. A higher proportion of ddcfDNA correlates with a higher probability of organ rejection.

*Methods: Here we present a novel method for rapid ddcfDNA prediction using a bioanalytics pipeline which can predict the proportion of ddcfDNA, with inputs of low-depth, whole genome next generation sequencing (NGS) data of cell-free DNA (cfDNA) and the genotype of the recipient, within two hours. The cfDNA extracted from plasma collected in Streck BCT tubes is used to prepare libraries for analysis on the Illumina NextSeq. Recipient genotypes were determined using the Infinium Omni2.5-8 beadchip array, which interrogates approximately 2.5 million single-nucleotide polymorphism loci. The core prediction-pipeline is an extension of the SIGMA protocol (https://sigma.omicsbio.org), originally designed to calculate the proportion of microbial abundance at the strain level using metagenomic NGS reads. While the original method required the presence of the genomic sequences as a reference-set, our pipeline generates the required reference sequences (recipient and donor genomes) using cfDNA NGS data and the genotype of the recipient.

*Results: We tested 15 sets of simulated ad-mixtures of donor and recipient NGS reads, where donor fractions and read-pair counts ranged from 0.49% – 16.6% and 954,662 – 1,569,663, respectively. Computationally derived donor/recipient reference genomes, covering SNP loci homozygous to recipient, were comprised of ~1.3M bases. Given the relatively small sizes of reference genomes, the run time of the prediction module was ~10 minutes (15 CPUs/10GB RAM), and the whole simulation process for a single test set required ~30 minutes. Our predictions had strong correlation (r2=0.99) with the simulated proportions of the donor cell-free DNA reads in the mix.

*Conclusions: In summary, our conceptual model yields accurate results, significantly reducing the computational time required compared to published methods.

« Back to 2019 American Transplant Congress