Sensitivity of shotgun metagenomics to host DNA: abundance estimates depend on bioinformatic tools and contamination is the main issue.

Abstract

A recent study reported that increasing host DNA abundance and reducing read depth impairs the sensitivity of detection of low-abundance micro-organisms by shotgun metagenomics. The authors used DNA from a synthetic bacterial community with abundances varying across several orders of magnitude and added varying proportions of host DNA. However, the use of a marker-gene-based abundance estimation tool (MetaPhlAn2) requires considerable depth to detect marker genes from low-abundance organisms. Here, we reanalyse the deposited data, and place the study in the broader context of low microbial biomass metagenomics. We opted for a fast and sensitive read binning tool (Kraken 2) with abundance estimates from Bracken. With this approach all organisms are detected even when the sample comprises 99 % host DNA and similarly accurate abundance estimates are provided (mean squared error 0.45 vs. 0.3 in the original study). We show that off-target genera, whether contaminants or misidentified reads, come to represent over 10 % of reads when the sample is 99 % host DNA and exceed counts of many target genera. Therefore, we applied Decontam, a contaminant detection tool, which was able to remove 61 % of off-target species and 79 % of off-target reads. We conclude that read binning tools can remain sensitive to low-abundance organisms even with high host DNA content, but even low levels of contamination pose a significant problem due to low microbial biomass. Analytical mitigations are available, such as Decontam, although steps to reduce contamination are critical.