Bait-capture sequencing resolves respiratory virus genomes but often misses co-infections

Respiratory viruses remain among the most significant causes of illness and death worldwide, and when mutations in circulating strains undermine the primers and probes used in routine molecular tests, identifying infected individuals and tracing transmission chains becomes considerably harder. Whole genome sequencing offers a way around this vulnerability, because it does not depend on a single conserved target and can reveal viruses carrying novel mutations that conventional assays would miss. A new evaluation led by researchers at the UK Health Security Agency’s Virus Reference Department, working with the South London Specialist Virology Centre at King’s College Hospital, has now put three sequencing and amplification workflows head to head on the same clinical specimens, offering one of the most detailed comparisons to date of bait-enriched sequencing against commercial and laboratory-developed PCR for respiratory pathogens.

The study, published in Frontiers in Microbiology, compared the Roche Cobas ePlex respiratory pathogen panel, a rapid multiplex PCR assay that screens for many pathogens simultaneously, with an in-house semi-quantitative multiplex reverse transcription PCR and with whole genome sequencing using RNA bait capture. The material comprised 61 nasopharyngeal swabs collected in 2021 and stored in viral transport medium, each of which had tested negative for SARS-CoV-2 but positive for co-infections of other respiratory viruses on the ePlex. Because the swabs were selected precisely for their multiplex-detected co-infections, they posed a demanding test for any sequencing workflow, which must recover several divergent viral genomes from the same limited clinical extract.

Before touching the clinical samples, the team optimised the sequencing workflow using cultured isolates of six representative viruses: adenovirus type 14, influenza A H1N1, influenza A H3N2, respiratory syncytial virus A and B, and SARS-CoV-2, the last extracted directly from a clinical specimen rather than propagated in culture. Nucleic acids from these isolates were serially diluted to mimic the range of viral loads seen in real swabs, and three combinations of library preparation kit and bait panel were evaluated: the Takara SMARTer Stranded Total RNA-Seq Kit v2 paired with either the Illumina Respiratory Virus Oligo Panel v2 or the Twist Respiratory Virus Research Panel, and the Twist Library Preparation EF Kit 2.0 paired with the Twist panel. Enriched libraries were pooled and sequenced on Illumina instruments, and the resulting read counts were compared statistically using paired Wilcoxon signed-rank tests with Bonferroni correction.

The optimisation experiments revealed a clear trade-off. For undiluted, high-concentration extracts, the combinations built around the Twist Respiratory Virus Research Panel delivered substantially higher median read recovery, with significant fold changes of 7.4 and 9.4 favouring them over the Illumina panel pairing. Once the extracts were diluted, however, the picture reversed: the Takara kit combined with the Illumina oligo panel gave the most consistent read recovery across all dilution points, while the median number of mapped reads was always lower when the Twist library preparation was used on diluted material. The authors attribute these differences to four design features, including the absence of a fragmentation step in the Takara kit, adapter addition by PCR rather than ligation, an rRNA depletion step that enriches viral targets, and the much smaller bait pool in the Illumina panel, roughly 7,800 baits versus 41,047, which may increase the probability of bait-virus interaction at low target concentrations.

With the most sensitive protocol selected, the team turned to the 61 swabs and built an in-house bioinformatics pipeline for unbiased metagenomic analysis. Reads were stripped of human sequence by alignment to the GRCh38 reference, depth-normalised with khmer to remove redundancy, quality-filtered with Trimmomatic and assembled de novo with metaSPAdes. Assembled sequences longer than 150 bases were queried against the RefSeq viral database with BLASTN, and quality-filtered reads were then mapped to the best-matching reference to derive coverage statistics. A virus counted as detected only if its reads were absent from negative controls and aligned to the correct viral group, and genome recovery described the breadth of coverage toward the reference, with a near-complete genome defined as at least 95 percent recovered.

The headline result was that whole genome sequencing recovered respiratory virus genomes from most of the swabs, with an average genome recovery of 85.9 percent, and that its detections concurred with the ePlex result in the great majority of cases. RSV and metapneumovirus were the viruses recovered most often by sequencing, at 80 and 70 percent respectively, and the highest number of near-complete genomes came from RSV, with 14 of 16 RSV detections yielding genomes at least 95 percent complete. The authors note that the strong RSV performance may reflect both the relative lack of diversity in RSV genomes and the timing of sample collection, which coincided with a documented spike in RSV activity in the northern hemisphere after the easing of non-pharmaceutical interventions during the COVID-19 pandemic.

The most striking discrepancy concerned co-infections. The ePlex detected up to four different viruses in a single swab, whereas the RT-qPCR and sequencing workflows never exceeded three, and each most frequently identified just one virus, accounting for 57.4 and 60.7 percent of samples respectively. Multiple viruses in the same swab were in the minority for RT-qPCR at 27.9 percent and for sequencing at just 13.1 percent. The authors suggest two possible explanations: the targeted nature of bait capture, and competition between viruses during bait hybridization or library preparation, in which a dominant pathogen outcompetes a weaker one for reagents and effectively masks its presence. They also caution that ePlex detections not confirmed by the other workflows, 49 viruses in total with cycle threshold values ranging from 23.4 to 39.9 and nearly two-thirds of them enteroviruses, raise uncertainty about whether such findings represent true clinical infection or merely viral carriage.

Agreement statistics told a nuanced story. Every virus found by the RT-qPCR or sequencing workflows was also found by the ePlex, with complete agreement on typing of seasonal coronaviruses, influenza, parainfluenza and RSV, giving a positive percent agreement of 100 percent for all viruses tested. Overall percent agreement, however, varied widely by virus: it was highest for RSV at 80 percent for sequencing versus ePlex and 85 percent for RT-qPCR versus ePlex, and for metapneumovirus at 70 and 60 percent respectively, but lowest for adenovirus and enterovirus, where sequencing reached only 12.5 and 11.76 percent against the ePlex. Enterovirus capture underperformed consistently, with a significant reduction in genome recovery relative to other viruses, and near-complete enterovirus genomes were rarely recovered even at cycle thresholds below 30. The authors link this to the considerable genetic diversity of enteroviruses, which may be incompletely represented in the bait panel, consistent with earlier reports of poor sequence-capture efficiency for this group.

Applying the study’s conservative operational criteria for genuine co-infection, which required either at least 75 percent genome recovery for each virus or cycle threshold values below 30, only two of the 61 swabs qualified: one containing RSV-A and metapneumovirus, and one with enterovirus, parainfluenza virus and RSV-B. The study also illustrates a subtle advantage of sequencing over probe-based assays: in one swab, an incomplete metapneumovirus genome was recovered by sequencing but missed by RT-qPCR, and one of the coverage gaps overlapped the binding site of the assay’s primer and probe, explaining the lost signal. Because capture-based sequencing is not constrained by the diversity assumptions built into diagnostic primer design, the authors argue it can both support surveillance and inform the design of better sequence-specific assays. The evaluation does carry limitations: the sample set was small and selected for co-infections, negative percent agreement could not be calculated for the ePlex against the other workflows, and freeze-thaw cycles between tests may have degraded some viral nucleic acid unevenly. Even so, the work demonstrates the power of bait capture to resolve respiratory pathogen genomes from limited clinical material, and suggests that as sequencing becomes more automated and accessible, capture-based metagenomic approaches are likely to play an expanding role in respiratory virus diagnostics and public health surveillance.

Subject of Research: Evaluation of whole genome sequencing protocols and target-capture panels for detecting respiratory viruses in clinical specimens

Article Title: Evaluation of whole genome sequencing protocols and target-capture panels for detection of respiratory viruses in clinical specimens

Article References: Kelly, D., Yogananthan, K., Subramaniam, P., Talts, T., Manso, C. F., Kele, B., Bibby, D. F., Zuckerman, M., & Mbisa, J. L. (2026). Evaluation of whole genome sequencing protocols and target-capture panels for detection of respiratory viruses in clinical specimens. Frontiers in Microbiology, 17, Article 1964559. https://doi.org/10.3389/fmicb.2026.1964559

Image Credits: AI Generated

DOI: 10.3389/fmicb.2026.1964559

Keywords: respiratory viruses, whole genome sequencing, bait capture, nasopharyngeal swabs, rapid multiplex PCR, RT-qPCR, co-infection, diagnostics, RSV, enterovirus, metagenomics, public health surveillance