Comparison of three next-generation sequencing platforms for metagenomic sequencing and identification of pathogens in blood

Abstract

Background: The introduction of benchtop sequencers has made adoption of whole genome sequencing possible for a broader community of researchers than ever before. Concurrently, metagenomic sequencing (MGS) is rapidly emerging as a tool for interrogating complex samples that defy conventional analyses. In addition, next-generation sequencers are increasingly being used in clinical or related settings, for instance to track outbreaks. However, information regarding the analytical sensitivity or limit of detection (LoD) of benchtop sequencers is currently lacking. Furthermore, the specificity of sequence information at or near the LoD is unknown.

Results: In the present study, we assess the ability of three next-generation sequencing platforms to identify a pathogen (viral or bacterial) present in low titers in a clinically relevant sample (blood). Our results indicate that the Roche-454 Titanium platform is capable of detecting Dengue virus at titers as low as 1X102.5 pfu/mL, corresponding to an estimated 5.4X104 genome copies/ml maximum. The increased throughput of the benchtop sequencers, the Ion Torrent PGM and Illumina MiSeq platforms, enabled detection of viral genomes at concentrations as low as 1X104 genome copies/mL. Platform-specific biases were evident in sequence read distributions as well as viral genome coverage. For bacterial samples, only the MiSeq platform was able to provide sequencing reads that could be unambiguously classified as originating from Bacillus anthracis.

Conclusion: The analytical sensitivity of all three platforms approaches that of standard qPCR assays. Although all platforms were able to detect pathogens at the levels tested, there were several noteworthy differences. The Roche-454 Titanium platform produced consistently longer reads, even when compared with the latest chemistry updates for the PGM platform. The MiSeq platform produced consistently greater depth and breadth of coverage, while the Ion Torrent was unequaled for speed of sequencing. None of the platforms were able to verify a single nucleotide polymorphism responsible for antiviral resistance in an Influenza A strain isolated from the 2009 H1N1 pandemic. Overall, the benchtop platforms perform well for identification of pathogens from a representative clinical sample. However, unlike identification, characterization of pathogens is likely to require higher titers, multiple libraries and/or multiple sequencing runs.

Figure 1

Read mapping against DENV-1 genome at 1X10^2.5 pfu/ml. Reads resulting from 1 454 Titanium sequencing run of a cDNA library made from blood spiked with DENV-1 at titer of 1X10^2.5 pfu/ml were aligned to the reference genome NC_001477.1 using CLC Genomics Workbench version 6.0.4 at default parameters.

Figure 2

Read mapping against representative segments of Influenza genome. In A), reads resulting from a representative run of 454 Titanium sequencing, Ion Torrent PGM sequencing, and Illumina MiSeq sequencing were mapped to the reference Influenza A H1NI segment 5 [NC_002019.1], using CLC Genomics Workbench version 6.0.4 at default parameters, whereas in B), they were mapped against segment 8 [NC_002020.1]. In B), no Roche-454 reads mapped to the reference. In both A) and B), coordinates of reference genome segment are displayed along the top and G/C content is graphed below reference in pink.

Figure 3

Read mapping against segment 5 of Influenza A of replicate sequencing runs. Reads resulting from a replicate run of Ion Torrent PGM sequencing (top) and Illumina MiSeq sequencing (bottom) were mapped to the reference Influenza a H1NI segment 5, [NCBI accession: NC_002019 ] using CLC Genomics Workbench version 6.0 at default parameters. Coordinates of reference genome segment are displayed along the top and G/C content is graphed below reference in pink.

Figure 4

Mathematical modeling of the likelihood of detecting a genetic modification in B. anthracis. The expected number of hits to an inserted gene of size 1 kb, at 5 copies, was simulated as a function of the number of organism-specific reads collected from the metagenomic sample. The relative size of each rectangle indicates the proportion of samplings for which a specific number of hits is expected.