Separating the signal from the noise in metagenomic cell-free DNA sequencing

Abstract

Cell-free DNA (cfDNA) in blood, urine and other biofluids provides a unique window into 24 human health. A proportion of cfDNA is derived from bacteria and viruses, creating opportunities for the 25 diagnosis of infection via metagenomic sequencing. The total biomass of microbial-derived cfDNA in 26 clinical isolates is low, which makes metagenomic cfDNA sequencing susceptible to contamination and 27 alignment noise. Here, we report Low Biomass Background Correction (LBBC), a bioinformatics noise 28 filtering tool informed by the uniformity of the coverage of microbial genomes and the batch variation in 29 the absolute abundance of microbial cfDNA. We demonstrate that LBBC leads to a dramatic reduction in 30 false positive rate while minimally affecting the true positive rate for a cfDNA test to screen for urinary 31 tract infection. We next performed high throughput sequencing of cfDNA in amniotic fluid collected from 32 term uncomplicated pregnancies or those complicated with clinical chorioamnionitis with and without intra-33 amniotic infection. The data provide unique insight into the properties of fetal and maternal cfDNA in 34 amniotic fluid, demonstrate the utility of cfDNA to screen for intra-amniotic infection, support the view 35 that the amniotic fluid is sterile during normal pregnancy, and reveal cases of intra-amniotic inflammation 36 without infection at term. 37 38 2 Introduction 1 2Metagenomic sequencing of cfDNA offers a highly sensitive approach to screen for pathogens in clinical 3 samples 1-4 . The sensitivity of metagenomic sequencing of cfDNA in plasma can be boosted by the 4 implementation of library preparations optimized to recover short, degraded microbial cfDNA 5 , or by 5 strategies that selectively enrich microbial DNA or deplete host DNA 6-8 . A major remaining challenge is 6the relatively poor specificity of the cfDNA metagenomic sequencing, which is limited by alignment noise, 7annotation errors in reference genomes and environmental contamination 9 . 8 9Here, we report low biomass background correction (LBBC), a tool to filter background contamination and 10 noise in cfDNA metagenomic sequencing datasets. We have applied LBBC to two independent datasets. 11We first re-analyzed a dataset from a previous study that investigated the utility of urinary cfDNA as an 12analyte to monitor urinary tract infection 2 (UTI). Next, we generated a new dataset of cfDNA in amniotic 13 fluid collected from uncomplicated pregnancies or those complicated with clinical chorioamnionitis at term, 14a common heterogeneous condition that can occur in the presence or absence of intra-amniotic infection 10 . 15We report a first, detailed study of the properties of cfDNA in amniotic fluid. For both datasets, detailed 16 microbiologic workups, including results from conventional bacterial culture and/or 16S rRNA sequencing, 17were available to benchmark the LBBC workflow. We demonstrate that LBBC greatly improves the 18 specificity of cfDNA metagenomic sequencing, while minimally affecting its sensitivity. ...

Interaction between microbiota and immunity in health and disease

“…252,253 In order to avoid fallacious conclusions, strategies to control contamination must be considered when working with low microbial biomass tissues, including experimental and computational measures. 250,[254][255][256] Although promising, these strategies largely still await proof that signals uncovered from low-biomass microbiomes reliably translate into verifiable mechanistic biological insights.…”

Peritoneal Effluent Cell-Free DNA Sequencing in Peritoneal Dialysis Patients With and Without Peritonitis

“…For 36 of the 68 samples, we detected a primer sequence utilized for experiments in the laboratory (Item S1), these adaptors were not overly abundant (of samples with primer, the median primer abundance was 0.004% of total reads), and were automatically removed in the low biomass background correction (LBBC) pipeline. 4…”

“…Filtered BLAST aligned reads were then merged (so read pairs were transformed to single reads) and a maximum likelihood estimator was used to estimate microbial abundance by taxa 11 . A low biomass background correction algorithm 4 was used to remove physical and digital contaminants using the following constraints: bacteria and eukaryotes (filter at tax_ID level; coefficient of variation filter = 2.5, batch variation filter = -7, negative control multiplier filter = 50), viruses (filter at tax_ID level; no coefficient of variation filter, batch variation filter = -8, negative control multiplier filter = 50) . All samples and replicates were included in the background correction pipeline.…”

“…We have recently demonstrated a 'digital culture' test, based on shotgun metagenomic DNA sequencing of cell-free DNA (cfDNA), that can detect a range of bacterial, viral, and fungal infections 2,3 . Importantly, we have applied this technique to body fluids that have been classically considered sterile such as urine and amniotic fluid [4][5][6] . Here, we investigate the utility of metagenomic sequencing of cfDNA in peritoneal effluent to monitor infections in PD patients.…”

Cell-free DNA Profiling Informs Major Complications of Hematopoietic Cell Transplantation

“…To identify microbial-derived cfDNA after WGBS, we first identified and removed host related sequences and we then aligned the remaining unmapped reads to a set of microbial reference genomes (0.9 ± 0.4% of total reads, Materials and Methods). We implemented a background correction algorithm to remove contributions due to alignment noise and environmental contamination 34 .…”