Improving saliva shotgun metagenomics by chemical host DNA depletion

Abstract

Background: Shotgun sequencing of microbial communities provides in-depth knowledge of the microbiome by cataloging bacterial, fungal, and viral gene content within a sample, providing an advantage over amplicon sequencing approaches that assess taxonomy but not function and are taxonomically limited. However, mammalian DNA can dominate host-derived samples, obscuring changes in microbial populations because few DNA sequence reads are from the microbial component. We developed and optimized a novel method for enriching microbial DNA from human oral samples and compared its efficiency and potential taxonomic bias with commercially available kits.

Results: Three commercially available host depletion kits were directly compared with size filtration and a novel method involving osmotic lysis and treatment with propidium monoazide (lyPMA) in human saliva samples. We evaluated the percentage of shotgun metagenomic sequencing reads aligning to the human genome, and taxonomic biases of those not aligning, compared to untreated samples. lyPMA was the most efficient method of removing host-derived sequencing reads compared to untreated sample (8.53 ± 0.10% versus 89.29 ± 0.03%). Furthermore, lyPMA-treated samples exhibit the lowest taxonomic bias compared to untreated samples.

Conclusion: Osmotic lysis followed by PMA treatment is a cost-effective, rapid, and robust method for enriching microbial sequence data in shotgun metagenomics from fresh and frozen saliva samples and may be extensible to other host-derived sample types.

Keywords: Host depletion; Microbial enrichment; Microbiome; Propidium monoazide; Saliva; Shotgun sequencing.

Fig. 1

Percent of shotgun metagenome sequencing reads aligning to human genome varies by sample type. Data from the Human Microbiome Project (HMP; black) of healthy individuals demonstrates the percentage of human reads by sample type. Saliva data (blue) was collected from healthy individuals in this study. Stool n = 249, skin n = 29, vaginal n = 103, nasal cavity n = 112, inner cheek n = 175, tongue n = 208, gums n = 189, and saliva n = 24 (this study)

Fig. 2

Host DNA depletion in saliva reduces the percentage of sequencing reads aligning to the human genome. Saliva was collected from eight individuals and divided into triplicate aliquots for each of the processing methods. The fraction of quality filtered shotgun sequencing reads mapping to the human genome was assessed with Bowtie 2. One-way ANOVA with Tukey’s multiple comparison correction, significance p < 0.0001

Fig. 3

Differences in saliva microbiome driven by participant and not method of host depletion. Microbial reads cluster by participant (a) and not method of host depletion (b) in PCoA space using Bray-Curtis distance. c Pairwise Bray-Curtis dissimilarities: within participant, within method (WP-WM); within participant, between methods (WP-BM); and between participants, within methods (BP-WM). Each category is statistically significantly different from each other group (Kruskal-Wallis with Benjamini and Yekutieli FDR correction p < 0.0001)

Fig. 4

Bray-Curtis dissimilarity between host depleted and raw sample from the same participant. The pairwise Bray-Curtis dissimilarity value was calculated between each sample with every other sample in this study. The dissimilarity values between each sample and the matched participant raw sample are presented here. Statistical significance calculated with Kruskal-Wallis with Benjamini and Yekutieli FDR correction p < 0.05. raw-raw n = 22, raw-Fil n = 66, raw-NEB n = 63, raw-Mol n = 63, raw-QIA n = 69, raw-lyPMA n = 69

Fig. 5

Experimental overview. A graphical summary of the experimental design and results