2.1 Norovirus dataset curation and cleaning workflow

A total of 45,095 norovirus sequences exceeding the length of 100 nucleotides (nt) and available between the years 1995 and 2020 were downloaded from GenBank (https://www.ncbi.nlm.nih.gov/GenBank/) on 2 February 2021. The sequences were downloaded in FASTA format and were evaluated for GI or GII polymerase and capsid genotype identity and sequence coverage using the Norovirus Typing Tool (https://www.rivm.nl/mpf/typingtool/norovirus/).¹⁵ Genotype identity, sequence coverage, and other relevant metadata were retained and organised in RStudio. Virus sequences lacking sufficient sequence coverage to type either region of interest were discarded from the data set. The remaining virus sequence data (N = 43,999) were paired with relevant metadata (strain, organism, isolate, year of isolation, country, host, isolation source, publication title) available in the GenBank record. In instances where year and country of isolation metadata were not listed in one of the GenBank tags, such information was manually parsed from the strain or isolate tags where available. Viruses with ambiguous or missing year or location metadata were discarded from the dataset. Noroviruses collected during 2020 were omitted to only encompass 25 years of study (1995 and 2019).

To sort for human noroviruses, the data set was filtered for all instances of host metadata associated with Homo sapiens (‘Homo sapiens’, ‘Human’, etc). In instances where host metadata was not independently listed in GenBank tags, isolation source data was manually evaluated with the same key words. Viruses with missing host metadata were discarded from the dataset. Additionally, this dataset of human noroviruses was also screened for indications of immunocompromised hosts in the host, isolation source and title metadata (e.g., ‘Immunocompromised’, ‘HIV positive’), which were omitted upon detection. The final dataset totalled 30,810 human noroviruses between the years 1995–2019. A flowchart of the data cleaning process can be found in Figure 1a.

2.2 Data visualization

Statistical analysis and data visualization for the large-scale datasets were conducted in RStudio using tidyverse and associated packages, as well as Prism7 software. Geographic distribution of the human noroviruses was visualised in R using the maptools package. Administrative map shape file (1:10 m Cultural Vector, Admin 0 – Countries) was obtained from Natural Earth website (https://www.naturalearthdata.com/downloads/10m-cultural-vectors/10m-admin-0-countries/, accessed on 25 June 2021), and country boundaries were dissolved into eight geographic regions: North America, Latin America and Caribbean, North Africa and Middle East, Sub-Saharan Africa, Europe and Central Asia, East Asia and Pacific, South Asia, and Oceania. The number of sequences of each genotype in each region was counted and summarised in the pie charts using mapplots package. Heatmap order of polymerase types and capsid genotypes was derived from sequence alignments using MEGAX software from FASTA files containing up to two random representative sequences for each respective genogroup.^{14, 16, 17}

3.1 A quantitative and qualitative assessment of archival norovirus data

After stringent filters across multiple metadata parameters were applied to the original sequence database (N = 45,095), we obtained a final dataset comprised of 30,810 human norovirus sequences collected between the years 1995 and 2019 (Figure 1a). According to GenBank records, 16,972 (55.1%) of these sequences were reported in publications listed in PubMed. The length of these sequences spanned a range between 101 and 7778 nt, with a mean length of 961 nt and a median of 293 nt. On average, sequencing coverage begins around nucleotide position 4683 on the genome, allowing for typing of either the polymerase region, the capsid region, or both. The distribution of sequence coverage in this dataset with respect to the norovirus genome can be found in Figure 1b.

The spatiotemporal distribution of the dataset was also evaluated to better contextualise our investigation. A breakdown of archival norovirus distribution by country can be observed in Figure 1c, organised as the top 10 most represented countries with the remainder grouped in the Other category. While 90 countries were represented by this dataset, the distribution was highly skewed, with over 51% of all virus sequences contributed by the top four countries (China = 7690, Japan = 4551, Brazil = 2059, Thailand = 1948). Moreover, countries on the lower end of the distribution were largely underrepresented in terms of norovirus collection data, with 18 countries attributed to 10 or fewer norovirus sequences deposited into GenBank over a 25-year period. As a consequence, despite the necessity and utility of describing annual circulation trends at a global level, these results will be inherently biased from the overrepresentation of certain countries. Thus, it was additionally prudent to bin subsequent geographic distribution analyses into discrete regions.

The breadth and depth of norovirus sequences collected between the years 1995–2019 were also examined (Figure 1d). It was observed that only a minimal number of norovirus sequences were collected between the years 1995 through 1999. While modest increases in contribution to the norovirus record were observed in the years following 2000, it was not until 2006 that over 1000 sequences were recorded in GenBank in a single year. These annual trends have remained largely consistent, peaking in 2016 with 2615 reported norovirus sequences. Additionally, a steady increase in the length of sequence coverage has been observed over time, with the average length of submitted sequences consistently covering 1000 nt or more since 2011.

Lastly, given the broad variety of sources collected for this study, it was prudent to verify whether any singular source or outbreak contributed a disproportionate number of submitted sequences to the dataset. An analysis of sequence contribution from 1124 unique titles showed that 98.0% of these sources contributed less than 1.0% to the dataset, with 77.6% of sources accounting for ≤0.01% (Figure S1A). Only nine sources were identified to individually exceed 1.0% sequence contribution, with a single outlier accounting for 3.93% of total sequences. However, a closer examination of per-year sequence contribution from these sources showed that they were the output of multi-year surveillance programs, spanning between two to 14 years in scope (Figure S1B). As such, it is unlikely that these identified sources introduce a concerning degree of skew to the total dataset.

3.2 Spatiotemporal distribution of archival norovirus polymerase and capsid genotype circulation

A total of 25,632 norovirus sequences from the sampled dataset presented sufficient coverage for successful genotyping of the capsid region. Thirty-seven unique capsid genotypes were detected from the curated dataset, with the top 10 most prevalent displayed in Figure 2a. Interestingly, the top 10 list of most prevalent annually circulating genotypes contained at least one representative genotype from nearly all norovirus immunotypes.¹⁸ GII.4 was the predominant circulating capsid genotype, accounting for 52.6% of all available sequences between the years of 1995–2019. The second most prevalent genotype was GII.3 at 11.5%, followed by GII.2, GII.17, GII.6, GI.3, GII.13, GI.4, GII.12, and GII.14. The remaining capsid genotypes accounted for only 8.4% of the total sequences. An analysis of the annual circulation of capsid genotypes shows a consistent presence of GII.4, accounting for close to 60% of all recorded sequences at certain years (Figure 2b). GII.3 was also present across all measured years at a consistent frequency (range of 3.9%–25.0%, with an average of 11.7%). In contrast, while circulation of GII.2 and GII.17 was observed at low levels in earlier years (respective ranges of 1.0%–7.5% and 0.0%–6.0%), there was a marked increase in prevalence of both genotypes in the last 6 years, with GII.2 becoming the predominant circulating genotype in 2017, accounting for 47.9% of all recorded sequences that year. To better visualise these granular changes, a second graph of annually circulating genotypes was constructed with GII.4 data omitted (Figure S2).

Given the enduring predominance of GII.4 and its pronounced epochal diversification,^{5, 19-23} we further analysed the temporal distribution of pandemic variants (Figure S3). An analysis of 13,483 GII.4 sequences between the years of 1995–2019 recapitulated the chronological emergence and replacement of variants previously reported in the literature,^{5, 22, 24} with the current variant Sydney_2012 maintaining predominance for a longer span of years than any of its previous counterparts. It is worth noting that a small contingent of GII.4 sequences were either unable to be assigned a variant or were assigned one outside the boundaries of commonly accepted circulation years. This was mostly observed for short (100–300 nt) sequences from highly conserved regions of the capsid protein that provided insufficient phylogenetic signal for variant identification.

A total of 11,919 norovirus sequences from the sampled dataset presented sufficient sequence coverage for polymerase typing. Quantification of the 56 unique polymerase types also saw GII.P4 as the predominant circulating type, albeit at only 28.5% of total recorded sequences (Figure 2c). GII.P16 and GII.P31 (formerly, GII.Pe) polymerases were the next most frequent at 16.3% and 12.1%, respectively. The rest of the top 10 most frequent types was comprised of GII.P17, GII.P12, GII.P21 (formerly, GII.Pb), GII.P7, GII.P2, GII.P33 (formerly, GII.Pg), and GI.P3, with all remaining types accounting for only 13.6% of available sequences. Interestingly, the temporal distribution of polymerase type circulation showed that while GII.P4 initially circulated at similar levels to its capsid counterpart, a steady decrease in prevalence beginning in 2007 gave way to a near-total replacement in 2013 by GII.P16, GII.P31, and GII.P17, which have alternated as the predominant polymerases between the years 2013–2019 (Figure 2d).

Following the observation that some countries have contributed a vastly disproportionate quantity of archival norovirus sequences compared to others, it was necessary to organise the global circulation of genotypes into eight distinct geographic regions: North America, Latin America and Caribbean, North Africa and Middle East, Sub-Saharan Africa, Europe and Central Asia, East Asia and Pacific, South Asia, and Oceania (Figure 3). While sample sizes varied greatly between regions, the cumulative analysis of circulating capsid genotypes largely reflected the previously described global trends. GII.4 was the predominant capsid genotype in all regions, accounting for half or more of all recorded sequences in each area with a minor exception in South Asia. Fluctuations in frequencies between the other top 10 capsid genotypes were also observed at a regional level, such as the higher prevalence of GII.2 over GII.3 in North America. Additionally, a larger presence of the GII.6 genotype was also seen in the Latin America and Caribbean region, as well as in Europe and Central Asia.

Additional regional differences were detected when we visualised the temporal distribution of the respective changing frequencies of circulating capsid and polymerase types (Figure S4 A, B). For example, the increased presence of GII.2 between the years 2015–2019 was mostly from North America, Europe and Central Asia, and East Asia and Pacific regional circulation trends. Additionally, some circulation patterns were entirely unique to particular regions, such as the overwhelming majority of GII.13 circulation occurring in South Asia. Regarding polymerase type circulation, all regions presented the shift in predominance from GII.P4 to that of GII.P16, GII.P31, and GII.P17 that began in 2010–2014. However, the respective proportions of these genotypes varied greatly between regions. GII.P16 circulation was predominant in the North America, Latin America and Caribbean, Europe and Central Asia and South Asia regions, while GII.P31 eventually became the predominant polymerase genotype circulating in the North Africa and Middle East and Sub-Saharan Africa regions. In contrast, GII.P17 circulated at high levels in the East Asia and Pacific region as well as Europe and Central Asia and was the predominant genotype of Oceania during the years 2015–2019. Some of these data trends could be attributed to the extremely limited sequence collection of certain regions, particularly during the years prior to 2007, which complicates the interpretation of these results.

3.3 Incidence and frequency of genotype recombination events

Recent studies have indicated that recombination events between polymerase and capsid regions may be a significant driver in the emergence and predominance of norovirus genotypes and variants.^25-27 Thus, we analysed the dataset for the incidence of archival recombinants to provide a quantitative description of the role of recombinant strains on human norovirus diversity. Only 6471 (21.9%) sequences were able to be genotyped for both the polymerase and capsid regions; 4067 of these (62.8%) sequences were determined to be recombinants. The temporal distribution of this dataset showed many years with under 100 double-typed genomes until 2007, complicating analysis (Figure 4a). However, the proportion of recombinant sequences also rose dramatically alongside these increased sampling trends, with recombinant strains accounting for the majority of double-typed sequences from 2012 onward.

While 97 unique recombinant strains were uncovered from the analysis, three account for half (∼50.2%) of the 4067 total sequences: GII.2[P16], GII.4[P31] and GII.4[P16] (Figure 4b). Notably, these recombinants correlate with the recent emergence and rapid proliferation of capsid genotype GII.2 and polymerase types GII.P16 and GII.P31. Indeed, recombination events with GII.P16 account for four of the top 10 most represented recombinant genomes. In contrast, the remaining 87 recombinant strains only represented 20.7% of all analysed sequences, with many of these only accounting for a single recorded instance of recombination. Beyond the predominant recombinant strains, another point of interest was elucidating how recombination events might aid in perpetuating the long-term circulation of less prevalent polymerase types. For example, the circulation of polymerase type GII.P12 is characterised by a large spike in prevalence between the years 2004–2005, followed by moderate levels of circulation to present day (Figure S5A). To the degree that double-typed sequences are available for this time, these trends are associated with recombination events with GII.4 and GII.3, respectively (Figure S5C). In contrast, GII.P21 circulation is characterised by multiple boom-and-bust cycles between the years 2000–2019, which also seems to be associated with the relative levels of the GII.3[P12] recombinant strain (Figure S5B, D). Taken together, these data suggest that recombination events may not only be responsible for the sudden appearance of major shifts in genotype predominance but may also play a role in the prolonged persistence of minor genotypes circulating in the background.

Potential limitations to recombination types have been recently described using nearly-full norovirus genome sequences.¹⁴ Accordingly, we assembled a heatmap for all double-typed sequences in our dataset to better assess the full scope of all norovirus recombination data available on GenBank (Figure 5). Broad analysis showed a far higher incidence of recombination events within GII than GI. Despite a single report from a study in India,²⁸ there was no single incidence of recombination events between GI and GII genogroups. Unexpectedly, two instances of recombination events between GII and GVIII or GIX viruses were also identified. However, the parental viruses for GVIII and GIX have not been fully described, so the proper nature of these discrepancies on the phylogenetic clustering remain uncertain. Many of the polymerases identified in the breakdown of the most prevalent recombinant strains were shown to associate with five to 10 different capsids. In particular, the GII.P16 polymerase associated with 10 different capsids (including its nonrecombinant GII.16 counterpart), though recombination events with GII.2, GII.4 and GII.13 made up the majority of recorded double-type sequences. In contrast, polymerases such as GII.P31 had recorded associations with five capsid genotypes, but the recombinants with GII.4 accounted for the overwhelming majority of all double-typed sequences. Shifting to capsid genotypes with broad polymerase associations, our analysis found that eight capsids had documented associations with six or more polymerase types. Of these, GII.2 and GII.3 were associated with 10 and eight different polymerase types (including non-recombinants), with the most prevalent recombinant pairings being GII.2[P16] and GII.3[P21], respectively. A breakdown of the top 10 capsid and polymerase types with the highest number of associations with their counterpart regions can be found in Figure 5b, c.