Evaluating the Information Content of Shallow Shotgun Metagenomics

Abstract

Although microbial communities are associated with human, environmental, plant, and animal health, there exists no cost-effective method for precisely characterizing species and genes in such communities. While deep whole-metagenome shotgun (WMS) sequencing provides high taxonomic and functional resolution, it is often prohibitively expensive for large-scale studies. The prevailing alternative, 16S rRNA gene amplicon (16S) sequencing, often does not resolve taxonomy past the genus level and provides only moderately accurate predictions of the functional profile; thus, there is currently no widely accepted approach to affordable, high-resolution, taxonomic, and functional microbiome analysis. To address this technology gap, we evaluated the information content of shallow shotgun sequencing with as low as 0.5 million sequences per sample as an alternative to 16S sequencing for large human microbiome studies. We describe a library preparation protocol enabling shallow shotgun sequencing at approximately the same per-sample cost as 16S sequencing. We analyzed multiple real and simulated biological data sets, including two novel human stool samples with ultradeep sequencing of 2.5 billion sequences per sample, and found that shallow shotgun sequencing recovers more-accurate species-level taxonomic and functional profiles of the human microbiome than 16S sequencing. We discuss the inherent limitations of shallow shotgun sequencing and note that 16S sequencing remains a valuable and important method for taxonomic profiling of novel environments. Although deep WMS sequencing remains the gold standard for high-resolution microbiome analysis, we recommend that researchers consider shallow shotgun sequencing as a useful alternative to 16S sequencing for large-scale human microbiome research studies where WMS sequencing may be cost-prohibitive. IMPORTANCE A common refrain in recent microbiome-related academic meetings is that the field needs to move away from broad taxonomic surveys using 16S sequencing and toward more powerful longitudinal studies using shotgun sequencing. However, performing deep shotgun sequencing in large longitudinal studies remains prohibitively expensive for all but the most well-funded research labs and consortia, which leads many researchers to choose 16S sequencing for large studies, followed by deep shotgun sequencing on a subset of targeted samples. Here, we show that shallow- or moderate-depth shotgun sequencing may be used by researchers to obtain species-level taxonomic and functional data at approximately the same cost as amplicon sequencing. While shallow shotgun sequencing is not intended to replace deep shotgun sequencing for strain-level characterization, we recommend that microbiome scientists consider using shallow shotgun sequencing instead of 16S sequencing for large-scale human microbiome studies.

Keywords: human microbiome; metagenomics; microbiome; shotgun metagenomics.

Figures

FIG 1

Information content of deep and shallow shotgun sequencing. (A) Percentages of raw shotgun DNA sequences that are unique to one bacterial species across different human body habitats (7 distinct plaque samples, 30 distinct samples from other body sites). (B, C) Principal-coordinate analysis of Bray-Curtis beta diversity using deep (B) and shallow (C) sequencing (sample sizes were as described for panel A). (D, E) Shannon diversity estimates at varied sequencing depths for human stool (D) and subgingival plaque microbiomes (E) (sample sizes were as described for panel B). Boxplots show minimums, first quartiles, medians, second quartiles, and maximums, with outliers beyond 1.5 times the interquartile range plotted individually.

FIG 2

Comparison of species and function profiles with ultradeep sequencing data. (A, B) Correlation with ground truth species (A) and KEGG Orthology group (KO) (B) profile for known genes present in the reference database at different sequencing depths, showing that as few as 0.5 million sequences recover nearly the full species and function profiles (ground truth based on 2.5 billion reads per sample; 4,394 genes and 694 species were used at each subsampling level from the subject 1 ultradeep sequencing sample; comparable results from subject 2 are not shown). Gene and species profiles recovered from the ultradeep data include only direct matches to genes and genomes present in the database; de novo assembly of novel genes and contigs from deep data are expected to yield additional uncharacterized gene content and is not possible with shallow shotgun data. (C, D) Scatterplots of species (C) and KOs (D) at 0.5 million versus 2.5 billion reads per sample (we used the same sample size as used for panel A and B above).

FIG 3

Biomarker discovery using shallow shotgun sequencing. (A, B) Precision, recall for per-read species binning of different metagenomics analysis tools (“95” and “98” refer to the minimum alignment identity threshold used; 5 distinct replicates [rep] were performed per subsampling depth, and error bars show standard deviations). (C) Stacked bar plot of species abundances recovered from HMP mock community shotgun sequencing data. (D) Negative log10 false-discovery rate (FDR)-corrected P values using Mann-Whitney U tests for species associated with type 2 diabetes (17), compared between deep and shallow shotgun sequencing (43 healthy patients, 53 patients with type 2 diabetes).

FIG 4

Comparison of 16S sequencing and shallow shotgun recovery of species-level taxa. (A) Histogram of average Pearson correlation (R-squared) of species profiles between 16S sequencing and shallow shotgun sequencing from the same HMP sample (R-squared = 0.918), compared to the permutation-based null distribution of R-squared values for random pairings (P < 0.001). (B) Scatterplot of relative abundances of species in shallow shotgun sequencing versus 16S sequencing from the same HMP samples. Species found in only one data type are shown in a different color. (C) Fractions of all observed species, with relative abundance accounted for by species found by 16S sequencing only, shallow shotgun sequencing only, or both.