Unbiased Detection of Respiratory Viruses by Use of RNA Sequencing-Based Metagenomics: a Systematic Comparison to a Commercial PCR Panel

Abstract

Current infectious disease molecular tests are largely pathogen specific, requiring test selection based on the patient's symptoms. For many syndromes caused by a large number of viral, bacterial, or fungal pathogens, such as respiratory tract infections, this necessitates large panels of tests and has limited yield. In contrast, next-generation sequencing-based metagenomics can be used for unbiased detection of any expected or unexpected pathogen. However, barriers for its diagnostic implementation include incomplete understanding of analytical performance and complexity of sequence data analysis. We compared detection of known respiratory virus-positive (n= 42) and unselected (n= 67) pediatric nasopharyngeal swabs using an RNA sequencing (RNA-seq)-based metagenomics approach and Taxonomer, an ultrarapid, interactive, web-based metagenomics data analysis tool, with an FDA-cleared respiratory virus panel (RVP; GenMark eSensor). Untargeted metagenomics detected 86% of known respiratory virus infections, and additional PCR testing confirmed RVP results for only 2 (33%) of the discordant samples. In unselected samples, untargeted metagenomics had excellent agreement with the RVP (93%). In addition, untargeted metagenomics detected an additional 12 viruses that were either not targeted by the RVP or missed due to highly divergent genome sequences. Normalized viral read counts for untargeted metagenomics correlated with viral burden determined by quantitative PCR and showed high intrarun and interrun reproducibility. Partial or full-length viral genome sequences were generated in 86% of RNA-seq-positive samples, allowing assessment of antiviral resistance, strain-level typing, and phylogenetic relatedness. Overall, untargeted metagenomics had high agreement with a sensitive RVP, detected viruses not targeted by the RVP, and yielded epidemiologically and clinically valuable sequence information.

FIG 1

Respiratory virus detection by untargeted metagenomics and RVP. (A) Fractional abundance of human, bacterial, and viral sequences in untargeted metagenomics data from an influenza A virus-positive NP swab of a female infant determined by Taxonomer (; Flygare et al., submitted). Approximately 1.5% of reads were of viral and 2.7% of bacterial origin. (B) Of 1.74 × 10⁵ viral reads, 1.73 × 10⁵ (99.4%) could be classified to the species level (influenza A virus) and 7.9 × 10⁴ (45.3%) to the H1N1 subtype. (C) Untargeted metagenomics identified 36 of 42 (86%) respiratory viruses detected by the RVP. Four of the 6 viruses missed by untargeted metagenomics (blue bars) could not be detected by qPCR (hashed bars), resulting in detection of 36 of 38 viruses (95%) that were consistently detected by targeted methods. (D) Untargeted metagenomics detected more viral infections (n = 48, including 11 not targeted by the RVP [asterisk]) than the RVP (n = 37) in unselected NP swabs (n = 67) collected during a 12-month period. ADV, adenovirus; HMPV, human metapneumovirus; IAV, influenza A virus; PIV, parainfluenza virus; HRV, human rhinovirus; RSV, respiratory syncytial virus; HCoV, human coronavirus; CMV, cytomegalovirus; HBoV, human bocavirus; EV, enterovirus; MV, measles virus.

FIG 2

Overall taxonomic composition of RNA-seq reads and numbers of viral reads by respiratory virus. (A) Fractional abundance of reads binned as human, human mRNA (referred to as “mRNA”), any bacterial sequence (referred to as “bacterial”), bacterial 16S only (referred to as “16S”), viral, phage, any fungal sequence (referred to as “fungal”), fungal ITS only (referred to as “ITS”), ambiguous, and unknown is shown as median and interquartile range (box plots) and as violin plots. Only reads identified as viral (red, median ∼1:10⁻⁴ reads) were used for this analysis. (B) Viral read counts differed across 5 orders of magnitude. Viruses not targeted by the RVP are shown in red. ADV, adenovirus; HMPV, human metapneumovirus; IAV, influenza A virus; PIV, parainfluenza virus; HRV, human rhinovirus; RSV, respiratory syncytial virus; HCoV, human coronavirus; CMV, cytomegalovirus; HBoV, human bocavirus; EV, enterovirus; MV, measles virus; ITS, internal transcribed spacer.

FIG 3

Correlation of normalized read counts with viral burden and precision of viral read abundance within and between sequencing runs. (A) The correlation between viral copies per milliliter of viral transport medium determined by qPCR and normalized viral reads (viral reads per kilobase of viral genome size per million total reads [RPKM]) detected by untargeted metagenomics was assessed by a Spearman correlation test (rho = 0.7 P < 0.0001). (B) Reproducibility was evaluated by extracting and sequencing the same sample 5 (human rhinovirus [HRV] and human metapneumovirus [HMPV]) or 14 (human coronavirus [HCoV]) times. Replicate libraries were prepared independently and sequenced on the same lane (within run) or different lanes (between runs). Fractional abundance (viral reads per total reads) is shown for within-run replicates (same color) and between-run replicates (different colors). Precision is shown as percent coefficient of variation (CV).

FIG 4

High-resolution, sequence-based typing of 14 human rhinovirus strains based on RNA-seq directly from NP swabs. Most strains belonged to rhinovirus species C (n = 12; 86%), with 2 strains (14%) belonging to lineage 1, 4 strains (29%) belonging to lineage 2, and 6 strains (43%) belonging to lineage 3; 2 strains belonged to rhinovirus species A, and no rhinovirus species B strains were detected. Near full-length sequences of 14 human rhinovirus strains (strains A through N) were aligned (MUSCLE); a neighbor-joining consensus tree (1,000 replicates) is shown. Full-length reference sequences for rhinovirus A (HRV-A89), rhinovirus B (HRV-B14), and representative full-length genome sequences from each of the rhinovirus C lineages (GenBank accession numbers EF077280, GQ223227, and JN990702 [40]) were included for comparison. Poliovirus 1 was used as the outgroup. For strains sequenced as part of this study, month, year, and state of sample collection are indicated in parentheses. Colors represent species- and lineage-level clades.