Respiratory virus genomic epidemiology during post-pandemic re-emergence of influenza in Australia

Abstract

Simultaneous genomic sequencing of multiple respiratory pathogens from clinical samples can provide real-time data on viral evolution, co-circulation, and co-infection during seasonal epidemics. Influenza is a major global respiratory pathogen, with a well-established genomic epidemiology framework, that has yet to be integrated with genomic surveillance of other co-circulating respiratory viruses. Leveraging existing well-integrated community influenza sampling during the first major Australian winter influenza season after the lifting of the Coronavirus Disease 2019 (COVID-19) pandemic restrictions in 2022, we examined the genomic epidemiology of influenza A alongside co-infections of common human respiratory viruses. Using a commercial respiratory viral sequencing method, Respiratory Virus Oligo Panel (Illumina), we recovered full-length human influenza A genomes from 75% (117/157) of samples with nucleic acid amplification test-confirmed influenza, as well as 19 genomes of co-infecting viruses from 17 samples, including respiratory syncytial virus (RSV-B), human bocavirus, Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2), human metapneumovirus, and coronaviruses 229E and OC43. The observed incidence of co-infecting viruses (RSV and SARS-CoV-2) was temporally consistent with national epidemic trends. In the majority of co-infections, the viral abundance was predominated by one of the infecting viruses, which may suggest either consecutive infections, or within-host dynamics favouring the dominance of one of the infecting viruses. The dominant influenza subtype was A/H3N2, with a minority of A/H1N1 infections persisting throughout the winter season, consistent with national surveillance. We contextualize our representative sample set within the global genomic diversity of A/H3N2 and the ongoing evolution of influenza A/H3N2 in the following Northern Hemisphere 2022/2023 winter. In addition to influenza A genomic epidemiology, multi-pathogen methodology enables simultaneous detection and characterization of co-infecting respiratory pathogens, providing insights into the role of viral dynamics during overlapping epidemics.

Introduction

During the Southern Hemisphere 2022 winter, the seasonal community influenza virus epidemic coincided with epidemic waves of multiple respiratory viral infections, including SARS-CoV-2, respiratory syncytial virus (RSV), and human metapneumovirus (hMPV) (NSW Health 2024). On a population scale, overlapping respiratory virus epidemics can rapidly escalate the demand on healthcare systems, both through a direct increase in the total number of affected individuals, and indirectly, via the potential for more severe infections in the presence of multiple co-infecting pathogens (Swets et al. 2022, Dagan et al. 2023). The dynamics and complexities of respiratory co-infections and overlapping epidemics are only beginning to be appreciated, with important consequences for predicting disease severity (Liu et al. 2023), understanding viral interference (Burstein et al. 2022, Dee et al. 2023), and uncovering novel pathogenesis due to pathogen–pathogen enhancement (Ho et al. 2023, Servellita et al. 2023).

The pivotal role of genomic epidemiology in real-time analysis of the COVID-19 pandemic has generated widespread expectations of similarly timely and high-resolution genomic monitoring of other circulating pathogens to support public health containment strategies, healthcare preparedness, and vaccine design (Hill et al. 2021, 2023, Ladner and Sahl 2023). Historically, influenza genomics capability was one of the earliest to be developed (Rambaut et al. 2008, Zhou et al. 2009) driven by the well-recognized pandemic potential of influenza A, which was responsible for four pandemics in the last century alone [H1N1 (1918), H2N2 (1957), H3N2 (1968), and H1N1 (2009)] (Rambaut et al. 2008, Wille and Holmes 2020), as well as the current epidemic of highly pathogenic H5N1 avian influenza first identified in 1997 (Xie et al. 2023). Even partial genomic information, such as sequences of the antigenically relevant haemagglutinin (HA) and neuraminidase (NA) genes, is vital for vaccine design and epidemic modelling. More recently, efforts have been made to develop genomic epidemiology for RSV (Zheng et al. 2021, Eden et al. 2022, Bardsley et al. 2023, Lin et al. 2024), which are likely to be expanded with the introduction of new RSV vaccines and therapeutics. However, genomic surveillance of other respiratory viruses remains minimal.

A limitation of most viral genomics methodologies is that laboratory protocols are designed around a single pathogen. Co-infections are seldom tracked, and no single public health workflow can accommodate multi-pathogen (syndromic) genomics. As broader sequencing strategies begin to generate syndromic genomic data (Hill et al. 2021, 2023), there is a strong incentive to design and evaluate workflows that can make use of these data to enhance epidemic management.

Epidemic dynamics of respiratory virus circulation are strongly influenced by pre-existing population immunity (Glass and Grenfell 2003). The global implementation of stringent public health measures due to the COVID-19 pandemic limited the circulation of many respiratory viruses, including Influenza A virus (Sullivan et al. 2020, Huang et al. 2021). As restrictions eased and international travel resumed, Australia experienced an early peak of seasonal influenza A in 2022 (Department of Health and Age Care, Australian Government 2022). Most local cases were diagnosed between May and June, with the highest case load occurring in the fortnight ending 26 June 2022 (Australian Centre for Disease Control 2022, WHO Collaborating Centre for Reference and Research on Influenza, Victorian Infectious Diseases Reference Laboratory 2022). Compared with the preceding 5-year average, the peak of infections occurred ~3 months earlier (May 2022) and resulted in a three-fold higher rate of notifications to public health authorities. In total, the 2022 season reported 225 332 of laboratory-confirmed influenza cases, with 82.7% determined to be influenza A virus (Australian Centre for Disease Control 2022). The rapid increase in cases coincided with limited vaccine protection, due to the early rise in influenza infections prior to the usual delivery of seasonal influenza vaccinations and low uptake of influenza vaccines in the community. Only 1.4% of children aged less than 5 years and 12.4% of adults greater than 65 years were vaccinated by mid-April 2022 (National Centre for Immunisation Research and Surveillance 2022). This low vaccine uptake could be a result of vaccine fatigue and hesitancy, and barriers to seeking healthcare during the COVID-19 pandemic (Khairat, Zou, and Adler-Milstein 2022, Su et al. 2022, Stamm et al. 2023). A similarly early spike in influenza A cases was also reported in Europe and North America during November and December 2022, the start of the Northern Hemisphere winter season (UK Government 2022, Centers for Disease Control and Prevention 2022).

The genomics of influenza A virus is crucial for providing data on the genetic diversity of circulating influenza viruses, when supported by phenotypic characterization, it enables an informed decision on candidate vaccine viruses for the following influenza season (Robertson et al. 2011). In this study, the extent of genome-wide viral diversity among circulating influenza A viruses in the 2022 season was assessed and identified epitope changes within the primary antigen of the dominant circulating influenza A subtype, A/H3N2, in addition to tracking corresponding rates of co-infections.

Here, we explored a commercially available method for multi-pathogen targeted sequencing to simultaneously detect and reconstruct influenza A genomes alongside other common respiratory viruses. Leveraging the existing well-integrated surveillance strategy for influenza in the community in Australia during the 2022 winter season, we used influenza-positive samples to analyse the genomic epidemiology of influenza A in the context of other respiratory viruses, including RSV and SARS-CoV-2.

Materials and methods

Ethics

Governance and human ethics approval for clinical metadata and use of respiratory specimens from cases positive for influenza A virus in New South Wales (NSW), Australia were obtained by Western Sydney Local Health District Human Research Ethics Committee (2022/ETH02426).

Samples for genomic surveillance

All clinical respiratory samples collected between 25 March and 14 July 2022 positive for influenza A virus by quantitative real-time polymerase chain reaction (RT-PCR) at the Institute of Clinical Pathology and Medical Research (ICPMR), Westmead and reference genomes provided by The WHO Collaborating Centre for Reference and Research on Influenza (WHO CCRRI) were included in the study. Details outlining the location of these Influenza cases can be found in Supplementary Fig. 2.

RNA extraction and RT-PCR

Total viral RNA was extracted using MagNA Pure 96 DNA and Viral NA Small Volume Kit (Roche Diagnostics Ltd., Rotkreuz, Switzerland), using 200 μL of primary respiratory sample and eluting in 100 μL in accordance with the manufacturer’s instructions. An in-house, diagnostic multiplex TaqMan® probe real-time PCR panel (Supplementary Table 3) detecting the influenza A virus matrix gene was used. Extracts which had a RT-PCR detection of influenza A virus were subject to a second confirmatory RT-PCR to confirm RNA integrity prior to sequencing. The reaction was carried out in 20 μL volumes with 4 μL of template DNA with AgPath-ID™ One-Step RT-PCR Kit (Applied Biosystems™, Thermo Fisher Scientific, Massachusetts, USA). The RT-PCR assays were conducted in a LightCycler® 480 System (Roche Diagnostics Ltd., Rotkreuz, Switzerland) using the following conditions: 45°C for 15 min, 95°C for 15 min, followed by 45 cycles of 95°C for 15 s and 60°C for 45 s, and 40°C for 30 s. Each run included a no template negative control (NTC).

Targeted respiratory virus genome sequencing

Genomic sequencing was attempted on all 180 clinical extracts (Supplementary Fig. 1). Viral enrichment was performed using the Illumina RNA Prep and Enrichment with the Respiratory Viral Oligo Panel version 2 (RVOP) (Illumina Inc., San Diego, California, USA). The RVOP comprehensive panel captures the genomes of 30 pathogenic human respiratory viruses, including SARS-CoV-2, and 12 common and recent influenza A virus subtypes (Supplementary Table 1). Total RNA extracts were used as input into the RNA Prep with Enrichment kit (Illumina Inc., California, USA). RNA denaturation, first and second strand cDNA synthesis, cDNA tagmentation, and library construction using unique dual indexes were performed according to manufacturer’s instructions. Individual libraries were purified then combined in equal volume 6-plex reactions for probe hybridization. Probe hybridization was performed overnight and held at 58°C, hybridized libraries were then captured and washed according to the manufacturer’s instructions and amplified as follows: initial denaturation 98°C for 30 s, followed by 14 cycles of 98°C for 10 s, 60°C for 30 s, 72°C for 30 s, and a final extension of 72°C for 5 min. Library pool quantities and fragment size were determined using Qubit™ 1× dsDNA HS Assay (Invitrogen™, Thermo Fisher Scientific, Massachusetts, USA) and Agilent HS D5000 ScreenTape (Agilent Technologies Inc., California, USA), respectively. Resulting library captures were combined into an equimolar pool and sequenced using paired end 148 base pair reads on the Illumina MiniSeq or iSeq (Illumina Inc., California, USA) with the aim of generating 1 × 10⁶ raw reads per specimen (Fig. 1).

Figure 1.

(a) Case notifications of influenza, RSV, and COVID-19 detected in Australia during the 2022 winter season. National Notifiable Disease Surveillance System (NNDSS) fortnightly data for influenza positive cases in NSW (orange triangles) and Australia (orange circles), COVID-19 (NSW, green triangles, Australia green circles), and RSV (Australia, purple circles) during 2022 are denoted by line graphs. (b) Local sequenced influenza A cases included in this study (bars) were typed as H1N1 (pink) and H3N2 (blue), or not typed (grey). Local and national influenza positive cases were separated on a fortnightly basis (excluding 17 April week ending). Each symbol represents a single case. Single cases with influenza A and RSV co-infection were denoted by a star symbol (★) when detected by RT-PCR detection only and a diamond symbol (♦) for sequencing confirmation of RSV RT-PCR results. Both symbols are displayed within the respective stacked bar corresponding to the influenza A typing results for that case. RSV became a nationally notifiable disease as of 1 July 2021. However, the notification numbers shown in this diagram do not represent a national picture, as these conditions are not yet notifiable in all states and territories.

Bioinformatic processing

Respiratory virus co-infections were detected using the castanet pipeline as previously described (Mayne et al. 2024). A positive detection was defined as >100 reads mapped to any reference target with >30% target coverage (minimum depth of 2 reads). For sequencing results to be included in the study NTC controls must contain <100 influenza virus reads and/or < 30% reference genome coverage to control for cross-contamination. Briefly, demultiplexed sequence read pairs were classified by Kraken2 (Wood and Salzberg 2014) using a custom database containing the human genome (GRCh38 build) and the full RefSeq set of bacterial and viral genomes. Sequences identified as either human or bacterial were removed using filter_keep_reads.py from the Castanet workflow. Remaining reads, composed of viral and unclassified reads, were trimmed to remove Illumina adapter sequences using Trimmomatic version 0.36 (60), with the ILLUMINACLIP options set to ‘2:10:7:1:true MINLEN:50’ to retain only high-quality reads. Trimmed reads were mapped to the influenza A virus genome of isolates A/Wisconsin/588/2019 [EPI_ISL_404460] for H1N1 and A/Darwin/9/2021 [EPI_ISL_16998754] for H3N2, using shiver version 1.5.7 (Wymant et al. 2018), with bowtie2 (Langdon 2015) as the mapper. Only properly paired reads with insert size <200-bp and with at least 70% sequence identity to the reference were retained. Consensus calls were derived for positions supported by a minimum of 10 mapped reads. Analysis of Influenza subtype co-infections was conducted by determining minor allele frequencies (MAFs), which were computed at every position using shiver restricted only to positions with minimum raw depth of 100 (tools/AnalysePileup.py), with the default settings of no BAQ and maximum pileup depth of 1 000 000. MAF were investigated to uncover subpopulation of virus with subtype defining mutations within H1N1 or H3N2 references.

Phylogenetic analysis

Genomes were included in phylogenetic analyses if the consensus sequence consisted of no more than 25% ambiguous characters. To place these data into the global phylogenetic context and to help resolve ancestry, a collection of consensus sequences for A/H1N1 and A/H3N2 viruses were downloaded from the GISAID database (Shu and McCauley 2017) on 15 December 2022, and included in the set of sequences to be aligned. Sequences included ancestral reference genomes, as well as all genomes spanning the period between the end of the Southern Hemisphere winter and the start of the Northern Hemisphere winter season (1 August 2022–31 December 2022). All available complete genomes for H1N1 (n = 425) and H3N2 (n = 1524) were downloaded and manually curated to retain only sequences listed as ‘original specimen’ to remove in vitro-generated viral diversity. Reference genomes from the 2022 global collection were downsampled using epidemiologically weighted random sampling, with each genome given a weight 1/x_i, where x_i is the number of sequences from location (country) i, for each of A/H3N2 and A/H1N1. This procedure reduced the impact of oversampled locations while retaining maximal geographic representation. The final subsampled data set contained 150 reference A/H3N2 and 50 A/H1N1 genomes collected between 2012 and 2022. The resulting reference genomes, together with sequences from this study, were aligned using MAFFT (Katoh et al. 2002) to produce one alignment each for H1N1 and H3N2. For segment specific analyses, the regions corresponding to each segment were selected from the whole-genome alignment.

Maximum-likelihood phylogenetic reconstruction was performed separately on the whole-genome alignments and all eight gene segment alignments, with automated model selection as implemented in IQ-TREE (Model: GTR + F + R6) (Nguyen et al. 2015). A total of 1000 bootstrap replicates were used for each IQ-TREE run, and zero-length branches were collapsed. For tree reconstruction, the first 7 codons (21 bases) of each segment were masked to prevent spurious phylogenetic signal arising from ambiguous reporting of segment ends due to different primer-based amplification protocols used in the production of global reference sequences.

Time resolved phylogenies and phylogeographic reconstruction were produced using TreeTime (ignoring outliers) (Sagulenko, Puller, and Neher 2018). The mugration model was corrected for sampling bias (correction factor of 5.0) to account for the over-representation of Australian sequences. To confirm robustness, we evaluated sampling-bias correction values of 0, 3, 5, and 10 and found that inferred ancestral locations were stable across this range. Predicted ancestral locations and their confidence values were exported and annotated on the final phylogeny.

RSV genotype classification

RSV genotypes A and B were identified by sequence similarity to representative consensus genomic sequences of RSV-A and RSV-B, generated as follows: complete RSV-A (n = 1295) and RSV-B (n = 834) genomes were downloaded from NCBI Virus (https://www.ncbi.nlm.nih.gov/labs/virus/vssi) and reduced to a set of representative sequences at 90% sequence identity using cd-hit-est (Fu et al. 2012) with cluster threshold 0.9, generating 4 clusters for RSV-A and 3 for RSV-B. Similarity to RSV-A and RSV-B consensus sequences was used for classification. In addition, consensus genomes generated for RSV-B and HMPV were genotyped using Nextclade (Aksamentov 2021).

Phylogenetic clustering of whole genomes of influenza A/H3N2

To visually represent the structure of the A/H3N2 whole genome phylogeny, we implemented a simplified procedure for phylogeny ‘painting’ conceptually similar to ChromoPainter and fineSTRUCTURE (Lawson et al. 2012) specifically, for each genomic segment, we painted the tree in four arbitrary colours corresponding to the most highly informative splits in the phylogeny, using informative sites within that segment to identify the splits. Phylogenetically informative sites (SNPs) were clustered using the k-means algorithm as implemented in the Python library SciPy (v12.0), testing values of k from 2 to 8 for the number of clusters. The optimal concordance with the deep structure of the whole genome phylogeny was obtained with k = 4, that is, the tree structure could be ‘painted’ using four colours. The predicted clusters for each segment were aligned to the corresponding node on the whole genome phylogeny using ggtree (Yu 2020).

Statistical analysis

Statistical analyses were conducted in RStudio (version 4.2.2; released 2022-10-31). Paired two-tailed t-tests were used to assess whether significant differences existed in age or hospitalization status between cases of each influenza subtype. Associations between viral co-infection and hospitalization were evaluated using Fisher’s exact test, given the categorical nature of these variables and the presence of small cell counts. Statistical significance was defined as P < .05.

Results

Sample collection and viral genome sequencing

For this study, we retrieved remnant samples referred to the Institute of Clinical Pathology and Medical Research (ICPMR), NSW Health Pathology from collecting laboratories across the state of New South Wales (NSW) and the Australian Capital Territory (ACT). Together, NSW and ACT make up 34% of the Australian population, which includes Sydney, a city of over 6 million people and the most common point of entry to Australia for international arrivals during the study period. Samples were retrieved in near-to-real time throughout the 2022 influenza season and sequenced in batches every 2–4 weeks. From 25 March to 14 July 2022, 180 unique samples positive for influenza on routine RT-PCR had sufficient residual extract to be sequenced as part of this study (Supplementary Fig. 1). The majority (71%, 127/180) were from symptomatic individuals within the community, a quarter were hospital inpatients (26%, 46/180) and 3% were from high dependency or intensive care units. The cohort contained both children (<18 years, 44%, 80/180) and adults (55%, 100/180) (Supplementary Fig. 2).

It is notable that sequencing was done from remnant samples, which introduces an inevitable freeze thaw cycle, and this may consequently reduce RNA quality. On re-testing, 12.7% (23/180) of specimens were negative for influenza by RT-PCR. We included all 180 samples in our sequencing workflow but, as expected, influenza virus reads could not be recovered from the PCR-negative samples (Supplementary Fig. 1A).

Using the RVOP enrichment protocol with Illumina sequencing, 83% (130/157) of PCR-confirmed samples yielded sufficient influenza A sequence reads to infer subtype, and of these, 90% (117/130) produced a high-quality consensus influenza A virus genome (minimum read depth of ×10 and 75% reference genome coverage) (Supplementary Fig. 1A and B). Reconstructed genomes had a median depth of 682× and median genome coverage of 13 161-bp, with complete coverage of all 8 influenza A genomic segments (Supplementary Table 1). Successful genome reconstruction was highly concordant with the quantity of influenza RNA template, consistent with previous reports that targeted capture efficiency scales with viral load (Goh 2019, Bonsall et al. 2020, Lythgoe et al. 2021, Lin et al. 2024). Specimens from which a complete genome was obtained had mean cycle threshold (Ct) value of 22.5 (interquartile range 21–24), whereas specimens from which we could not reconstruct whole genomes had mean Ct of 30.2 (interquartile range 26–32), including the 23 specimens where influenza virus could not be detected despite repeat RT-PCR testing, consistent with RNA degradation (Supplementary Fig. 1A and B).

Spatiotemporal viral dynamics

Among the sequenced cohort, the temporal distribution of cases followed trends of influenza notifications at both state and national level, with the peak of cases in the fortnight ending 26 June 2022 (Fig. 1). The influenza A/H3N2 subtype predominated from the week commencing 29 May 2022, which also reflected state and national reports of A/H3N2 dominance (Australian Centre for Disease Control 2022, NSW Respiratory Surveillance Report—Week Ending 25 June 2022 2022). Concurrently, there was an ongoing wave of SARS-CoV-2 infection, dominated at the time by Omicron variant sublineage BA.2, and a rapidly developing RSV epidemic (Fig. 1). Over the peak influenza period, the week ending 25 June 2022, 55 510 COVID-19 cases, 13 409 laboratory-confirmed influenza cases, and 2700 RSV laboratory-confirmed cases were reported in NSW, with a combined incidence more than double the 5-year average of emergency department presentations for ‘influenza-like illness’ (Australian Centre for Disease Control 2022, NSW Respiratory Surveillance Report—Week Ending 25 June 2022 2022, NSW Health 2024).

The spatiotemporal patterns of influenza A cases among NSW local health districts during the study period (March to July 2022) were visualized using monthly maps of NSW (Supplementary Fig. 3). The first cases of influenza A/H3N2 (March) and A/H1N1 (April) from the study population were both diagnosed in Western Sydney. At least one case of each subtype was subsequently reported for each of the five local health districts represented in the study population.

Influenza A viral diversity in New South Wales and evidence of within-subtype reassortment

Within our dataset, influenza A/H3N2 predominated across all timepoints after 10 April 2022 (79%, 103/130 genomes), concordant with state and national reports (Fig. 1) (Australian Centre for Disease Control 2022, WHO Collaborating Centre for Reference and Research on Influenza, Victorian Infectious Diseases Reference Laboratory 2022, NSW Respiratory Surveillance Report—Week Ending 25 June 2022 2022). A minority (22%, 28/130 genomes) of the A/H1N1 population (WHO clade 6B.1A.5a.2) also persisted throughout the season, with limited genetic diversity (Fig. 2b). There was no difference in ages of cases infected with either subtype (P = .232, paired two tailed t-test), and neither subtype was associated with hospitalization at sampling (P = .342, paired two tailed t-test).

Figure 2.

Whole genome Influenza A virus phylogeny of circulating H3N2 (a) and H1N1 (b) clades during the Australian winter 2022. Age range of cases, hospitalization status, and the number of mutations of high interest detected (A only) of each node are indicated on the colour bars. The date of collection and co-infection (▴) of each case sequenced in the study is labelled parallel to each tip. The major, minor, and reassortant clades of A/H3N2 3C.2a1b.2a.2 are labelled in grey. (c) The mutations of high interest as defined by Flusurver detected for A/H3N2 cases. The study identifier is listed on the y-axis and the mutations detected listed on the x-axis, listing the segment the mutation is detected in first, followed by the amino acid change and position relative to A/Darwin/6. Detected mutations are indicated with a red box with a consistent set of mutations of high interest found in the minor and reassortant clades.

Phylogenetic reconstruction of the A/H3N2 genomes, contextualized with reference genomes from the national WHO CCRRI data, revealed the co-circulation of three distinct sub-clades of 3C.2a1b.2a.2, hereafter referred to as the ‘major’ (87%, 87/100 H3N2 genomes), ‘minor’ (10%, 10/100 genomes), and ‘reassortant’ clades (3/100 genomes from this study, and one genome from the WHO CCRRI reference dataset) (Fig. 2a).

Genomes belonging to the reassortant clade fell within the diversity of the major clade in segments PB2, PB1, NP, NA, and NS, with complete nucleotide identity to several other genomes within this sub-clade (Fig. 3, Supplementary Figs 4 and 5). However, in segments PA, HA, and MP, reassortant genomes clustered instead with the minor sub-clade (Fig. 3, Supplementary Figs 4 and 5). The four reassortant genomes were obtained from samples collected in two distinct geographic regions (Metropolitan Greater Sydney, 1/40 complete influenza A genomes, and rural Western NSW, 3/47 complete influenza A genomes), over a period from 23 April to 8 June 2022, indicating sustained transmission of a single genetic lineage. These results suggest that the reassortant sub-clade arose from a single reassortment event, which brought together distinct circulating clades of the two major immunodominant regions of HA and NA with different recent ancestries.

Figure 3.

Influenza A/H3N2 whole genome phylogeny during the Northern and Southern hemisphere winters 2022. Influenza A/H3N2 WHO clade 3C.2a1b.2a.2 cases dominated the winter seasons in both the Northern and Southern Hemispheres during the winter 2022. Whole genome sequences recorded from A/H3N2 infections in Australia (cyan nodes, n = 103) were compared to international A/H3N2 genomes from Africa (orange nodes, n = 4), Asia (pink nodes, n = 19), Europe (purple nodes, n = 77), Middle East (green nodes, n = 5), northern (yellow nodes, n = 38) and southern (grey nodes, n = 7) America and historical reference genomes annotated by location and year collected. Ultrafast bootstrapping support for major nodes is annotated to indicate confidence in branching structure. The major, minor, and reassortant clades detected in Australia are annotated on the phylogeny. Nine clusters of putative reassortant clades, including the reassortant clade from the Australian, Europe, and an international reassortant can readily be identified and are labelled with grey bars. K-means clustering was performed on informative sites within each segment to visualize patterns of reassortment and support the identification of reassortants. The segment cluster for each genome is indicated by the colour bars with dashed white boxes highlighting segment clustering of minor and reassortant clades. Segments are labelled below each bar indicating the polymerase basic protein 2 (PB2), polymerase basic protein 1 (PB1), RNA polymerase subunit (PA), hemagglutinin (HA), nucleoprotein (NP), neuraminidase (NA), matrix protein (M1), and non-structural protein (NS1).

We used FluSurver (A*STAR Bioinformatics Institute 2009) to screen all genomes for presence of known phenotypically significant mutations that differed from the contemporaneous reference strain of the relevant subtype (H3N2 A/Darwin/6 and H1N1 A/Wisconsin/588). Mutations reported by FluSurver have been previously characterized to cause antigenic drift, change host specificity, antiviral susceptibility, and/or affect glycosylation (Fig. 2c). Genomes within the minor A/H3N2 clade contained four mutations of high interest within the HA (I208F and N112S), NA (S329N), and NS1 genes (K219E) (Fig. 2c). All genomes from the minor and reassortant A/H3N2 clades contained both the HA mutations, I208F and N112S, in concordance with the evolutionary history of the reassortant HA segment (Supplementary Fig. 5). Mutations HA I208F (HA1 192, classical H3N2 numbering) and HA N112S (HA1 96) are predicted to change host specificity and antibody recognition. Mutation I208F has been associated with adaptation in avian H5 influenza viruses by changing HA specificity from α2,6- to α2,3-linked sialic acid receptors (Yang et al. 2007). Mutation S329N (HA1 329) in the NA gene was detected in all genomes in the minor H3N2 clade. This mutation is predicted to generate an additional N-glycosylation site at position 329, which likely impacts the antigenic properties of the strain (Webster et al. 1987). All but one genome in the minor clade also contained mutation K219E (NS1 230) in the NS1 gene. NS1 blocks host cell antiviral responses through several mechanisms, and this mutation may contribute to increased virulence (Jackson et al. 2008). No evidence of NA or endonuclease inhibitor resistance was identified.

To understand the subsequent evolutionary fate of the reassortant and the major and minor circulating A/H3N2 sub-clades, and to better understand the history of the individual segments, we explored global A/H3N2 sequences from the following Northern Hemisphere autumn/winter season (December 2022–February 2023).

Influenza A/H3N2 genomic diversity between the Northern and Southern Hemisphere winters

We augmented our dataset with a similarly sized set of 150 publicly available A/H3N2 whole genome sequences from international A/H3N2 cases sequenced between 15 July and 9 December 2022 (n = 77 European, n = 38 Northern America, n = 19 Asia, n = 7 South American, n = 5 Middle East, n = 4 Africa). We found that the A/H3N2 clades that dominated the 2022/2023 Northern Hemisphere winter were phylogenetically distinct from strains that circulated during the preceding Southern Hemisphere winter, as represented in our collection (Fig. 3). Only 7 genomes (n = 5 from Europe and n = 2 from Asia) clustered with the major 3C.2a1b.2a.2 sub-clade circulating in Australia, and no genomes clustered with either the minor or the reassortant sub-clades.

Evidence of multiple instances of within-subtype reassortment could be readily inferred from phylogenetically incongruent clustering across the eight segments of the influenza A genome, just as was observed for the Australian reassortant clade (Fig. 3). To visually represent this complex history for the global phylogeny of H3N2 clade 3C.2a1b.2a.2, we clustered all phylogenetically informative mutations and assigned an arbitrary colour to each cluster, for all eight segments. Every genome could thus be labelled with a collection of eight colour-coded histories for its eight segments, and the tree was ‘painted’ in accordance with the evolutionary history of each genome (Fig. 3). Clades which predominated in Europe shared higher sequence similarity at either NA and/or the HA segment with the major Australian sub-clade (Fig. 3 and Supplementary Figs 4 and 5). Notable also were the nine clusters of putative reassortant clades, including the reassortant clade from the Australian season (Fig. 3). These reassortants were generally confined to a single international geographic region. Interestingly, the reassortant clade which contains genomes from diverse geographical regions (labelled international reassortant, Fig. 3), contained a highly divergent NA segment which emerged in 1993 (Supplementary Figs 4 and 5).

Mugration analysis supported Australia as the likely origin of the major, minor, and reassortant clades when analysing the whole-genome and NA phylogenies; however, analysis of the HA segment alone suggested an introduction of the HA segment from Spain into Australia (Supplementary Figs 6–8). This demonstrates that selective sequencing of HA and NA genes limits the ability to differentiate globally circulating strains and may mask the extent of diversification occurring through genome reassortment.

Influenza A co-infection with respiratory viruses

In addition to the genomic epidemiology of influenza A virus post-COVID-19 restrictions, we were interested in the role of other common respiratory viruses circulating within the same population during the study period. A key strength of the capture-based sequencing approach is its ability to simultaneously sequence a diverse range of pathogens within a single assay. In this study, the RVOP panel was used to enrich for 30 respiratory viruses (Supplementary Table 2); we used this sequence data to look for the presence of co-infecting viruses in our study and compared the results on RT-PCR based diagnostic assays for up to 8 common respiratory pathogens (Table 1).

Table 1.

Comparison of respiratory virus co-infections detected by RT-PCR and genomics. Comparison of Influenza A and respiratory viral co-infection detections using genomics and diagnostic RT-PCR assays. Respiratory diagnostic testing algorithms varied across referring laboratories during the study period. For example, some laboratories tested only for SARS-CoV-2 and influenza, whereas others included additional respiratory viruses such as RSV, HMPV, rhinovirus (HRHV), parainfluenza virus (PF), and enterovirus (EV). Diagnostic RT-PCR assays were not routinely available for some viruses detected in this study, including human bocavirus and seasonal human coronaviruses 229E and OC43. Consequently, RT-PCR results were unavailable for certain viruses and are indicated as ‘not tested’ in the table, either due to restricted diagnostic testing panels or the absence of a routinely performed RT-PCR assay. Influenza subtyping was derived from genomic sequencing data. Genome recovery for co-infecting pathogens was defined by reference genome coverage with a minimum median read depth of two reads. Where available, Nextclade was used to assign genotypes from consensus genomes of co-infecting pathogens, including RSV, SARS-CoV-2, and HMPV. Abbreviations: SARS-CoV-2, severe acute respiratory syndrome coronavirus 2; EV, enterovirus; HBoV, human bocavirus; HCoV-229E, human coronavirus 229E; HCoV-OC43, human coronavirus OC43; HAdV, human adenovirus, Bp, Bordetella pertussis, ND, not detected, NA, not available.

Influenza results				Co-infecting pathogen
Sample ID	Influenza A RT-PCR Ct value	Influenza A subtype	Genome coverage (read depth)	Co-infecting Pathogen	Co-infecting RT-PCR Ct value	RVOP Virus detections	Genome coverage	Genotype (Nextstrain)
8401	21	H1N1	100% (1948×)	RSV	32.4	ND	n/a	n/a
0201	28.5	H1N1	100% (98×)	ND	NA	SARS-CoV-2	16%	Insufficient coverage
10 901	22.9	H1N1	100% (286×)	Not tested	NA	HBoV	100%	n/a
14 601	22.2	H1N1	100% (472×)	Not tested	NA	HCoV-229E	100%	n/a
14 701	18.2	H1N1	100% (31 755×)	Not tested	NA	HCoV-OC43	86%	n/a
16 201	24.4	H3N2	100% (122×)	RSV	24.4	RSV-B	100%	B.D.E.4
18 501	19	H3N2	100% (160×)	RSV	28.1	RSV-B	40%	Insufficient coverage
10 501	23	H3N2	100% (355×)	RSV	32.4	ND	n/a	n/a
				HMPV	38.4	ND	n/a	n/a
18 101	25.5	H3N2	100% (4 311×)	RSV	34	RSV-B	90%	B.D.E.4
18 401	23.3	H3N2	100% (1 357×)	RSV	34.1	ND	n/a	n/a
8526	26.8	H3N2	48% (5×)	SARS-CoV-2	40	ND	n/a	n/a
16 901	24.5	H3N2	100% (122×)	HMPV	26.7	HMPV	40%,	Insufficient coverage
				hRHV	36.4	ND	n/a	n/a
17 101	26.8	H3N2	100% (130×)	HMPV	30.6	HMPV	82%	A2.2.2
8201	20.3	H3N2	74% (2×)	hRHV	35.4	ND	n/a	n/a
12 901	22.7	H3N2	100% (875×)	hRHV	36.5	ND	n/a	n/a
11 601	18.2	H3N2	100% (4 690×)	PF3	29.1	PF3	24%	n/a
11 801	24.3	H3N2	100% (1 040×)	Not detected	NA	RSV-B	30%	Insufficient coverage
15 101	22	H3N2	100% (286×)	Not tested	NA	HBoV	100%	n/a
17 401	27.2	H3N2	100% (7 099×)	Bp	n/a	n/a	n/a	n/a
2101	24.1	Untyped	<100 influenza reads	SARS-CoV-2	18.6	SARS-CoV-2	100%	BA.2.9
9701	19.1	H1N1	5% (1×)	EV	34.5	ND	n/a	n/a
1301	ND	Untyped	no Influenza reads	Not tested	n/a	HBoV	100%	n/a
1501	30.4	Untyped	no Influenza reads	Not tested	n/a	HMPV	100%	A2.2.2
						HAdV	97%	n/a
1801	27.4	Untyped	no Influenza reads	Not tested	n/a	HMPV	56%	Insufficient coverage
2301	33.9	Untyped	no Influenza reads	ND	n/a	RSV-A	11%	Insufficient coverage
4901	22.9	Untyped	no Influenza reads	ND	n/a	SARS-CoV-2	12%	Insufficient coverage
10 201	22.4	Untyped	<100 influenza reads	Not tested	n/a	HBoV	97%	n/a
15 501	ND	Untyped	<100 influenza reads	Not tested	n/a	HCoV-229E	100%	n/a

Among the 180 influenza A cases, 19 patients (10.6%) had respiratory viral co-infection identified by either qPCR (n = 16), genome sequencing (n = 20), or both (n = 7) (Fig. 2, Table 1, and Supplementary Fig. 2). Genomic sequencing uncovered 14 co-infecting viruses that had not been detected by routine RT-PCR respiratory virus assays. These include viruses which were not targeted by the RT-PCR assays used in our setting [human bocavirus HBoV (n = 4), human coronavirus 229E (HCoV-229E; n = 2), and HCoV OC43 (n = 1)], not included in the testing requested for individual patients, or false negatives on routine RT-PCR screening [SARS-CoV-2 (n = 3), HMPV (n = 4), RSV-A (n = 1), RSV-B (n = 4), human adenovirus hAdV (n = 1)]. We found no statistically significant association between viral co-infection and hospitalization (P = .797, Fisher’s exact test).

Complete or near complete genomes (>70% coverage) were obtained from 60% (12/20) of co-infecting pathogens, including HBoV (n = 4/4), HCoV-229E (n = 2/2), RSV-B (n = 2/4, both genotype B.D.E.4), RSV-A (n = 0/1), HMPV (n = 2/4, both genotype A2.2.2), HCoV-OC43 (n = 1/1), SARS-CoV-2 (n = 1/3), and hAdV (n = 0/1). Phylogenetic analysis of complete genomes obtained from RSV-B and HMPV demonstrate clustering with other Australian genomes (Supplementary Figs 9 and 10).

Among our cohort, the most common respiratory viral co-infection with influenza A was RSV-B and HBoV (Table 1). Most patients with co-infections were within the 34–48-year-old age group (Supplementary Fig. 2). Only one patient was confirmed positive for influenza A and two additional respiratory viruses (HMPV/RHV); this patient was within the 0–2-year-old age group (Fig. 2a and Supplementary Fig. 2). In summary, we detected numerous and diverse co-infections with influenza A infections (Fig. 2a and b). Where multiple pathogens were detected and Ct values were available for both (n = 14), most samples had large differences between the viral load of the co-detected pathogens, with median Ct difference (∆Ct) of 9.2 cycles (range 0–15.4, Table 1, Supplementary Fig. 11). Across all co-infections, the virus with the lowest Ct values (corresponding to the highest viral load) consistently yielded the greatest number of deduplicated reads, reflecting a correlation between viral load and read recovery using the RVOP sequencing method (Supplementary Fig. 11). These results also suggest consecutive rather than simultaneous infections, or with virus–virus interactions affecting the relative infection dynamics.

Discussion

During the Southern Hemisphere winter of 2022 (June–August), numerous pathogens that cause symptomatic respiratory infections co-circulated in Australia after international travel resumed. International air travel for fully vaccinated residents resumed on 1 November 2021, travel between Australian states eased by March 2022 and fully vaccinated tourists were permitted to enter from 21 February 2022 (Australian Government 2021, 2022, Government of Western Australia 2022). The rapid and early rise of influenza A cases in the autumn of 2022 raised concerns about low influenza vaccine uptake and effectiveness, importation and establishment of new influenza virus strains (by displacing current circulating influenza virus strains), increased disease severity due to coinfections, and the impact of co-circulating respiratory pathogens on healthcare resource utilization. The season was dominated by circulation of A/H3N2 (3C.2a1b.2a.2) with persistent low prevalence of A/H1N1 infections (WHO Collaborating Centre for Reference and Research on Influenza, Victorian Infectious Diseases Reference Laboratory 2022). Indeed, the major A/H3N2 Australian clade reported in this study cause a spike in A/H3N2 in Victoria in autumn, this virus likely persisted and spread to other Australian states (Pendrey et al. 2023). A/H3N2 also dominated in the following Northern Hemisphere winter (December 2022–February 2023) but from genetically distinct sub-clades of A/H3N2 3C.2a1b.2a.2 (Fig. 3).

Our findings demonstrated the added value of targeted enrichment such as RVOP, which allows the whole influenza viral genome to be sequenced in a single assay, in contrast to the traditional PCR-based methods for selective sequencing of only HA and NA genes. Our analyses of complete influenza A genomes highlight evidence of gene segment reassortment within a single season and a high prevalence of reassortment between Northern and Southern Hemisphere A/H3N2 genomes. A/H3N2 whole genome phylogenies showed a deep split between Australian and subsequently sampled global A/H3N2 viruses (Fig. 3), suggesting that phylogenetically dissimilar clades caused A/H3N2 cases in the Northern Hemisphere. While acknowledging the global disparities in genomic surveillance capacity, mugration analyses provided further support of transition rates of individual segments between international locations (Brito et al. 2022, Hill et al. 2023, de Jong et al. 2026) (Supplementary Figs 6–9). Despite their highly divergent backbones, a large proportion of the European and North American A/H3N2 genomes share similar NA and HA segments with the major clade circulating in Australia during its earlier 2022 winter season, as revealed by the phylogenetic structure of each segment (indicated by colour bars in Fig. 3). In addition, up to nine reassortant A/H3N2 clades emerged, the vast majority of which were detected only within geographically distinct regions. This demonstrates the frequency and rapid diversification of influenza A viruses via within-subtype reassortment, which can be missed when only the HA and NA segments are sequenced.

We found no direct evidence of co-infection by distinct influenza viruses within a single sample, as expected given the limited sample size and previous reports indicating very low rates of dual influenza virus detection (Gregianini et al. 2019). However, the presence of multiple reassortments throughout the local and global A/H3N2 phylogeny strongly argues for relatively frequent and ongoing within-subtype co-infections, where multiple distinct lineages with each subtype co-circulate in the same population (Holmes et al. 2005). Theses within subtype co-infections would likely be masked by PCR-based screen methods and maybe more apparent as genomic surveillance matures. The minimal phylogenetic divergence of the resulting reassortant lineages additionally supports the view that reassortment events do not require host shift or significant adaptation and regularly produce evolutionarily fit viruses capable of sustained transmission within susceptible populations (Westgeest et al. 2014, Matsuzaki et al. 2024).

Examining influenza A, we assessed the extent to which the presence of other viruses, consistent with co-infection, was contributing to the patient load in our healthcare setting (Asner et al. 2014, Mao and Wu 2024). We found that evidence of recent or concurrent infection by multiple pathogens, defined here as the relative proportion of sequence-detectable virus in symptomatic patients between 9% and 11% (16/180 and 19/180) had evidence of possible co-infection. This is comparable to previous studies that report coinfection rates of 14%–15% (Mantelli et al. 2024), where co-infections are more common in children less than 5 years old. Of the sequence-confirmed possible influenza A co-infections, 26% (5/19) were RSV (4/5 RSV B), consistent with the epidemic trend in Australia showing RSV to be co-circulating widely at the same time as the influenza A wave was peaking (Australian Centre for Disease Control 2022). Notably, no single virus dominated the co-infection profile with influenza A: we detected various viruses, including RSV, HMPV, and HBoV, but also a variety of genotypes, with both RSV-A and RSV-B co-circulating. The predominant influenza A genotype was H3N2 (78% of samples), and correspondingly, only 5 of 19 co-infections (26%) involved H1N1. The detection of RSV co-infection with A/H1N1 demonstrates that this is not a genotype-specific phenomenon, but rather reflects the epidemiological dynamics.

The hybrid capture WGS-based approach is a powerful method and a useful intermediate tool between amplicon-based sequencing, which must be conducted through a separate laboratory workflow for each virus, and metagenomics, which is expensive and impractical for high-throughput genomic epidemiology (Buddle et al. 2024). Capture-based enrichment has the added advantage of multiple viral targets, remaining effective within regions of high genetic diversity, and permitting the detection of rapidly evolving, and potentially novel viruses (Briese et al. 2015, Williams et al. 2018, Pogka et al. 2022).

The method deployed in this study, the commercially available Illumina RVOP panel, was able to detect both RSV genotypes (A and B), with total RSV detection comparable with RT-PCR (Table 1). However, this method may have been less sensitive to diverse enteroviral genomes such as rhinoviruses as it failed to recover human rhinovirus (HRV) from two samples where HRV was originally detected by PCR. This finding is consistent with similar reports of low sensitivity to enterovirus genomes by the competing Twist Comprehensive Viral Panel (Kapel et al. 2023). In contrast, probe-based methods that use a denser set of probes to cover diverse viruses have substantially higher analytical sensitivity relative to the RVOP panel, and may be better suited to situations where monitoring for enteroviruses and other genetically diverse viral families is needed (Goh 2019, Lin et al. 2020).

Our study has limitations: this was a proof-of-concept deployment study in a public health setting, using a relatively limited number of remnant samples from routine microbiology testing, and thus does not explore the full capability of the methods. Research studies that used freshly collected samples and a custom probe panel designed to optimize sensitivity for genetically diverse pathogens (Lin et al. 2024) demonstrate that improved performance is achievable in practice, with dedicated processes. Our sample numbers and subsampling of contextual international genomes were also necessarily limited by available resources and timing, reducing power to detect reassortment, international A/H3N2 diversity, co-infections or pathogens not circulating at the time of the study. Our limits of detection were likewise conservative, and we considered only potential co-infections with sufficient viral load to return a positive sequencing result. However, we sampled representatively relative to the epidemic wave in our setting (Fig. 1), and the close concordance between our results and national trends lend support to our conclusions. Unfortunately, influenza B which was at low levels of circulation during this study and was not detected in the cohort and, therefore, could not be assessed. In addition, antiviral resistance can be readily inferred from genomic data, importantly as we were able to generate complete influenza A genomes, we had the capacity to infer resistance to NA and endonuclease inhibitors used to treat influenza infections, but neither were identified in our study.

Our study demonstrated that multi-pathogen genome sequencing can be beneficially deployed in a public health setting during epidemic waves of co-circulating respiratory viruses. This enables more efficient genomic epidemiological analysis and streamlines laboratory workflows, which currently are conducted independently for each individual virus species. We deployed a commercially available probe panel (RVOP, Illumina) to detect and sequence all commonly occurring respiratory viruses from clinical samples, using remnant material from routine microbiology testing. We demonstrated that in this setting, the sensitivity of probe panels is not inferior to routine RT-PCR in the diversity and multiplicity of respiratory viruses detected. Further refinement and customisation of such panels to increase probe density for genetically diverse pathogens could further enhance the sensitivity of this approach (Goh 2019). As the integration of pathogen genomics into public health surveillance matures, the sensitivity of capture-based sequencing methods is expected to improve through the adoption of standardized sample processing protocols that minimize RNA degradation, which was observed in this study. Expansion of sequences targeted by hybridization panels have enabled identification, quantification of carriage load and typing of bacterial pathogens which may increase disease severity, such as Streptococcus pneumoniae and Haemophilus influenzae (Lin et al. 2024). Although increasing the number of probes adds to sequencing costs of capture-based sequencing methodologies, the ability to sequence a variety of pathogens, including coinfections, with a single assay could have significant cost advantages to public health laboratories. Costs can also be mitigated through workflow efficiencies and the development of new innovations to decrease costs per genome.

Further work is warranted to improve and expand the application of multi-pathogen genomics combined with syndromic surveillance, particularly in complex epidemic scenarios such as seasonal respiratory infections with enhanced surveillance of influenza subtypes to monitor for H5 and H7 infection in humans. Ongoing ripple effects from the COVID-19 pandemic are likely to continue to impact the dynamics of co-circulating pathogens and argue in favour of a shift towards multi-pathogen sequencing in communicable disease control.

Author contributions

Study concept, administration and supervision by RR, TG. Methodology, data curation and investigation by CL, JA, WF, RR, SC. Formal analysis by RR, WF, TG, JA, SC, AA, CS. The first manuscript was written by JA, WF, RR, TG with additional editing from VS, JK, SC, DD, JA, SC, WF, SA, IB. The final manuscript was approved by all authors.

Conflicts of interest

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. S.G.S. reports consulting for CSL Seqirus.

Funding

This study was supported by the Prevention Research Support Program funded by the New South Wales Ministry of Health. R.R. and T.G. are supported by Investigator Grants (R.R. GNT2018222; T.G. GNT2025445) from the National Health and Medical Research Council, Australia (NHMRC). The funders of this study had no role in the study design, data collection, data analysis and interpretation, or writing of the report. The corresponding author had full access to study data and final responsibility for the decision to submit for publication.

Data availability

The Influenza A virus genome sequences and associated metadata can be found in the NCBI Sequence Read Archive online repository which can be accessed through (https://www.ncbi.nlm.nih.gov/sra/). Data pertaining to this study can be found in the BioProject PRJNA911331. SRA accessions, bioinformatic and study metadata (Supplementary Data Table 1) has been deposited on Zenodo 10.5281/zenodo.20197772.