Comparative epidemiology, phylogenetics, and transmission patterns of severe influenza A/H3N2 in Australia from 2003 to 2017

Abstract Background Over the last two decades, Australia has experienced four severe influenza seasons caused by a predominance of influenza A (A/H3N2): 2003, 2007, 2012, and 2017. Methods We compared the epidemiology, genetics, and transmission dynamics of severe A/H3N2 seasons in Australia from 2003 to 2017. Results Since 2003, the proportion of notifications in 0‐4 years old has decreased, while it has increased in the age group >80 years old (P < .001). The genetic diversity of circulating influenza A/H3N2 viruses has also increased over time with the number of single nucleotide polymorphisms significantly (P < .05) increasing. We also identified five residue positions within or near the receptor binding site of HA (144, 145, 159, 189, and 225) undergoing frequent mutations that are likely involved in significant antigenic drift and possibly severity. The Australian state of Victoria was identified as a frequent location for transmission either to or from other states and territories over the study years. The states of New South Wales and Queensland were also frequently implicated as locations of transmission to other states and territories but less so over the years. This indicates a stable but also changing dynamic of A/H3N2 circulation in Australia. Conclusion These results have important implications for future influenza surveillance and control policy in the country. Reasons for the change in age‐specific infection and increased genetic diversity of A/H3N2 viruses in recent years should be explored.


| INTRODUC TI ON
Influenza is a common, highly infectious virus that spreads from person-to-person by droplet and airborne routes. Influenza A/H3N2 is one of four subtypes that circulate seasonally in humans every year causing annual epidemics in both temperate and tropical regions; the others being influenza A H1N1pdm09 (A/H1N1pdm09) and influenza B Yamagata (B/Yam) and Victoria (B/Vic). 1 The predicted severity of seasonal influenza is not well understood but often depends on the predominate subtype in the circulation, vaccination coverage in the population, and the individual immune response of infected persons. 2 Studies have frequently observed higher rates of hospitalization and reduced vaccine effectiveness among A/H3N2 predominate seasons when compared to seasonal A/H1N1pdm09 (and the pre-pandemic seasonal A/H1N1) and influenza B viruses. [3][4][5] Increased disease burden and mortality in elderly people and children have also been associated with A/H3N2 infection compared with other seasonal strains, 6,7 while influenza B typically causes disproportionate morbidity and mortality in children. 8 Over the last 15 years, Australia has experienced four severe A/  9 An overall 33% vaccine effectiveness was estimated during the 2017 season in Australia for all strains. 9 Notably, the circulating Australian A/H3N2 virus from 2017 quickly spread to other countries and regions such as the United States and Europe during the following 2017/2018 northern hemisphere season causing similarly severe seasons. [9][10][11] In this study, we aimed to review the epidemiology and phylogenetics of the four most recent A/H3N2 predominant seasons in Australia as well as explore dynamics and temporal trends of A/ H3N2 transmission in Australia.

| Review of epidemiology
We searched the National Notifiable Diseases Surveillance System (NNDSS) for all laboratory-confirmed influenza notifications across all states and territories in Australia between 2001 and 2017 including data on subtype and age. 12 Age-stratified notification rates (per 100 000 population) were calculated using respective population estimates during the study years. Differences in notification rates between age groups and study years were assessed using a chi-square test as the proportion of the total notification rate per year. Crude age-stratified hospitalization and mortality rates due to influenza Database, [13][14][15] respectively. Differences of those two rates and notification rate of laboratory-confirmed influenza in each states/ territory between severe and mild A/H3N2 seasons were also assessed using a chi-square test. General practitioner surveillance reports on influenza-like illness (ILI), seasonal antigenic drift, and VE were obtained from annual National Influenza Surveillance reports. 16 Literature searches in Medline used a combination of keywords such as "vaccine effectiveness," "influenza season," "Australia," "H3N2," "vaccine," "2003 season," "2007 season," "2012 season," and "2017 season."

| Collection of sequence data
We searched GISAID for all human A/H3N2 hemagglutinin (HA) sequences collected during the study years with location and date of sampling metadata. 17 Table S1. While most sequences were isolated in each state or territories capital city, for simplicity and subsequent analysis, we refer to each state/territory as a single discrete location.
To reduce the impact of spatial sampling bias across seasons,

| Phylogenetics analysis
We aligned each seasons taxa against the corresponding vaccine strain using MUSCLEv3.8.4 19 and identified the proportion of unique mutations by subclade using Geneious v11.1.4. We measured the genetic diversity across each study season as the number of Single Nucleotide Polymorphisms (SNPs) per season using SeqMan via DNAstar Lasergene v7.1. 20 To control for differences in sample size by season, the average number of SNPs per season was calculated across three randomly subsampled subsets of 50 sequences each. We used an non-parametric Kruskal-Wallis H test in ibm ® spss Statistics 24 to determine significant differences in SNP counts of four seasons. 21 We estimated selection ratios (dN/dS) for each season using three separate methods: Single likelihood ancestor counting (SLAC) and fixed-effect likelihood (FEL) implemented via Datamonkey and default settings, 22 and Bayesian renaissance counting (BRC) implemented in beastv1.8.4. In BEAST, we specified 100 million Markov Chain Monte Carlo generations sampling every 10 000 steps. We considered selected sites as significant if they were supported by at least two of the three methods used.

| Phylogeography analysis
We generated time-scaled phylogenetic trees for each A/H3N2 season using beast v1.8.4 23-25 specifying a GTR + I + Γ4 nucleotide substitution model as determined using jModelTest. 26 We selected a relaxed (uncorrelated log-normal prior) molecular clock model over a strict clock model after additional model testing using path sampling (PS) and stepping-stone sampling (SSS) methods in beast.v1.8.4 (Table S2), and a non-parametric Bayesian skyline demographic tree prior. [27][28][29][30] We specified a symmetric discrete-trait model using the Bayesian stochastic search variable selection (BSSVS) framework in beast.v1. 8.4 to estimate transmission between each pair-wise location per season. [23][24][25] For each season, we again specified MCMC chains of 100 million generations sampling every 10 000 steps. We assessed for sufficient mixing and convergence using tracer.v1.6 after considering the first 10% of samples as burn-in. 31 We generated Maximum clade credibility (MCC) trees using treeannotator.v1.8. 4 32 and visualized each using figtree.v1.4.3. 33 We used spread3.v0.9.6 to visualize transmission and the calculate Bayes factor (BF) support for each route. 34 We considered statistical support for transmission as BF >3 which is convention in phylogeography studies. 35 (Table S3). There was no observable trend in increasing notification rates between states and territories when comparing severe and mild seasons (P > .05); however, the average notification rate of each state/territory was 1.75-5.53 times higher compared with mild seasons (Table S4).   (Table S5). There was also no significant difference of hospitalization rate and mortality rates between severe and mild seasons (P > .05).

| Influenza activity, ILI, and antigenic drift
Pediatric influenza outcomes in severe seasons seemed similar to those observed in mild seasons (Table S6). However, the average crude hospitalization rates (5.91 times) and crude mortality rates (3.11 times) of elderly people were much higher compared with mild seasons (Table S6).

| Phylogeographic analysis of A/H3N2
Between the eight states and territories of Australia across four A/  (Table S8).

| D ISCUSS I ON
In this study, we have reviewed the epidemiology of severe and mild A/H3N2 predominant seasons in Australia since 2003 and investigated the genetics and transmission patterns of four severe A/ H3N2 seasons. We have observed a significant difference (P < .001) between age-stratified notification rates across the study years, with increasing rates among those 80 years and over and decreasing rates among 0-4 years (Figure 1). This suggests the elderly are an increasingly important risk group for A/H3N2, perhaps even more so than currently expected, as Australia's population continues to age: The proportion of over 80 population was 3.31%, 3.61%, 3.82%, and 3.91% in 2003,2007,2012, and 2017, respectively. [42][43][44][45] Recent increases in influenza vaccine coverage for children (children >6 months of age), from 10.1% in 2014 to approximately 30% in 2017 season may also explain this trend. 46,47 Influenza A/H3N2 is typically characterized by the limited circulating diversity of HA such that entire lineages are replaced within 2-8 years. 48 In our study, however, we observed that the genetic diversity of A/H3N2 per season has increased significantly (measured as the average number of SNPs per season) even when controlling for increased sampling effects ( Figure 2B). This may be due to changing population dynamics rather than inherent viral changes and hypothesize that increasing global population size may provide the potential for increased diversity of Our study has also identified patterns and possible transmission trends in circulating A/H3N2 between states and territories in Australia. In each of the study years, we found strong evidence consistently linking transmission between Victoria and other states and territories across the continent. As cold temperatures are known to affect the transmissibility of influenza, 52 this could be related to the cool-temperate weather the state experiences during winter, along with a large growing population. 26 New South Wales, Australia's most populous state, was also frequently implicated in transmission to or from other states and territories, and experiences similarly but less severe cool-temperate winters. International tourism and Australian residents returning from overseas may also explain the observed results, with 63% of international overseas arrivals intending to stay in New South Wales and Victoria, 53 and 62% of all short-term Australian departures and arrivals coming from the same states 54 increasing the likelihood of importation and ignition of seasonal outbreaks from these states, especially in the cities of Sydney (New South Wales) and Melbourne (Victoria). Interestingly, Queensland, F I G U R E 4 Phylogeographic projection of supported routes of influenza A/H3N2 transmission between states and territories in based on HA across four predominant H3 seasons a state which experiences rather mild winters and subtropical climates, was also frequently implicated in transmission across the study years. Potential reasons for this observation warrant further investigations but suggest demographic factors are more impactful on influenza transmission between Australian states and territories than climactic ones. Recent evidence of synchronization of seasonal onset across the continent also supports this conclusion. 55 There was also an apparent trend towards increasing circulation on the east coast of Australia from 2003 to 2017. As Australia's population has continued to grow, particularly concentrated in major cities along the east coast, 26 along with increasing domestic travel 28 this is not unexpected but does highlight opportunities for epidemic control. For example, the emergence of a severe novel strain in Victoria or New South Wales might be expected to reach most major cities along the east coast in a relatively short period, perhaps weeks, 55 and epidemic/pandemic plans should therefore account for this potential scenario.
There are a few of limitations to our study. We can only include data on laboratory-confirmed influenza notifications to indicate the impact of A/H3N2 viruses. These notifications are likely to be a significant underestimate of the true incidence of influenza infection in the population. 56 Furthermore, large numbers of un-subtyped notifications also likely conceal the true proportion of subtype A/H3N2 among typed influenza A. We therefore make the assumption that the typed sample is representative of the true composition of circulating influenza A. The relative lack of sequence data in the earlier seasons studied here also means our analysis is likely to be affected by sampling bias, even when controlling for the known effects of oversampled locations in phylogeography studies. 57 More frequent subtyping of influenza-positive cases as well ensuring sequence data is routinely collected and uploaded onto publicly accessible databases would help to improve the accuracy of influenza surveillance in Australia and understandings of A/H3N2 severity.

| CON CLUS ION
We identify increasing rates of influenza notifications among those aged over 80 years in Australia as proportion of the total notification rate, while rates for those under 5 years have decreased as a proportion. This indicates an increasing influenza-associated health burden for elderly Australians and may correlate with Australia's aging population. The genetic diversity of circulating A/H3N2 viruses has also increased since 2003 with more diversified strains observed in recent severe seasons. We hypothesize that increasing global population size and travel to Australia provides the potential for increased A/H3N2 circulating diversity, importation, and reduced VE, which has implications for domestic and international influenza vaccination policy. This, however, warrants further investigation. Lastly, we identified the states of Victoria and New South Wales as important locations for the dissemination of A/H3N2 viruses to other states and territories, possibly due to climate and/or population effects.
Improving testing procedures to include routine subtyping and sequencing should be a priority for improving the accuracy and detailed data from influenza surveillance in Australia and enable a better understanding of the true burden of disease due to A/H3N2 influenza viruses.

ACK N OWLED G EM ENT
We gratefully acknowledge the authors and originating and submitting laboratories of the sequences from GISAID's EpiFlu™ Database on which this research is based. The list is detailed below.