Seasonal dynamics of the wild rodent faecal virome

Abstract Viral discovery studies in wild animals often rely on cross‐sectional surveys at a single time point. As a result, our understanding of the temporal stability of wild animal viromes remains poorly resolved. While studies of single host–virus systems indicate that host and environmental factors influence seasonal virus transmission dynamics, comparable insights for whole viral communities in multiple hosts are lacking. Utilizing noninvasive faecal samples from a long‐term wild rodent study, we characterized viral communities of three common European rodent species (Apodemus sylvaticus, A. flavicollis and Myodes glareolus) living in temperate woodland over a single year. Our findings indicate that a substantial fraction of the rodent virome is seasonally transient and associated with vertebrate or bacteria hosts. Further analyses of one of the most common virus families, Picornaviridae, show pronounced temporal changes in viral richness and evenness, which were associated with concurrent and up to ~3‐month lags in host density, ambient temperature, rainfall and humidity, suggesting complex feedbacks from the host and environmental factors on virus transmission and shedding in seasonal habitats. Overall, this study emphasizes the importance of understanding the seasonal dynamics of wild animal viromes in order to better predict and mitigate zoonotic risks.


| INTRODUC TI ON
Our knowledge of the global virosphere has rapidly expanded (Li, Shi, et al., 2015;Roux et al., 2016;Shi et al., 2018;Zhang et al., 2018), mainly due to decreasing costs and increasing efficiency of highthroughput sequencing. However, while it is now relaliutively straightforward to genetically characterize host viromes and discover new virus sequences, most studies provide only a glimpse of the circulating virus diversity due to infrequent, nonsystematic and spatially limited sampling of target species. As a result, it is unclear why some viruses are found in some species or populations at specific time points but not in others (Harvey & Holmes, 2022).
While viral discovery studies provide valuable data for understanding the evolutionary history and host range of viruses, they offer limited insights into what factors shape wild animal viromes.
In order to understand viral dynamics in wild populations, we need to move from descriptive host-virus associations to a mechanistic understanding of where and when viruses are transmitted and how entire viral communities (viromes) are shaped by the environment and local host communities (Bergner et al., 2019;Fearon & Tibbetts, 2021). For example, both decreases and increases in the number of parasites (i.e., richness) in wild animals have been associated with habitat loss and fragmentation (Mbora & McPeek, 2009;Morand et al., 2019), indicating that anthropogenically mediated changes in host species composition and population densities can directly impact parasite community compositions. Furthermore, the influence of anthropogenic land-use change on virus community compositions has also been observed for a broad range of taxa, suggesting it is a key determinant of host viromes (Campbell et al., 2020;Hermanns et al., 2021;Myer & Johnston, 2019;Susi & Laine, 2021).
These findings further highlight why studying community traits, such as parasite richness, is critical for understanding and forecasting zoonotic risk over time and space.
Current knowledge about what factors shape viral communities in animals comes from a small but growing number of studies.
A comparison of viromes from three parasitic wasp species reared in laboratory conditions suggests that host phylogeny influences viral community structure (Leigh et al., 2018). However, it is unclear whether viromes in wild animals are also commonly predicted by host evolutionary history. Indeed, a study of multiple wild waterbird species sharing habitats found discordance between the host phylogeny and virome composition (Wille et al., 2019). This finding suggests that interspecific interactions and transmission among waterbirds might break down the host phylogenetic structuring of viral communities in wild settings. However, as investigations into virus community dynamics in multihost systems are limited, it remains uncertain how viral communities vary across host and viral taxa or ecological contexts. Virome composition can differ within species due to demographic and environmental characteristics. For instance, a survey of 24 vampire bat colonies found that virus richness was positively associated with younger age structure, lower elevation and increasing anthropogenic influence (Bergner et al., 2019). Studies on waterbirds also similarly found higher viral richness in younger age groups (Hill et al., 2022;Wille et al., 2021), suggesting age structure is a critical determinant of virus diversity in wild animals.
Although evidence suggests that environmental and host factors influence viral communities in wild animals, these surveys have predominantly been cross-sectional, with any given population sampled at a single time point. As a result, we have a sparse understanding of temporal dynamics in wild animal viromes. Fundamental questions such as how viral diversity varies over time and what proportion of viruses are only detected intermittently or at certain times of year in seasonal environments remain unaddressed. These dynamic environments affect wild animals through seasonally varying birth and death rates and timing of specific behaviours, including mating and other social interactions. As a result, many factors affecting viral transmission vary seasonally (Altizer et al., 2006), such as recruitment of susceptible individuals, population density and contact rates. Furthermore, virus survival in the environment can fluctuate throughout the year and influence onward spread. For instance, the environmental persistence of avian influenza viruses is higher at lower temperatures (Brown et al., 2007).
Consequently, it is not surprising that zoonotic virus surveillance studies in reservoir populations have consistently observed seasonal variation in virus prevalence in several host species, such as rodents, bats, birds and racoons (Amman et al., 2012;Fichet-Calvet et al., 2007;George et al., 2011;Hirsch et al., 2016;Páez et al., 2017).
However, except for a handful of studies focusing on specific virus families, for example, paramyxoviruses in bats and influenza viruses in mallard ducks (Latorre-Margalef et al., 2014;Wille et al., 2018), investigations into temporal variation in virus diversity in wild animals are rare, leaving a significant gap in our knowledge about viral community dynamics in changeable environments. For example, we may expect increases in viral richness during an animal's breeding season, driven by higher (primarily intraspecific) contact rates. Alternatively, viral community richness or composition may respond to seasonal changes in climate, for instance if ambient conditions affect viral persistence in the external environment (Brown et al., 2007;Sobsey et al., 1988), which could impact viral richness and abundance.
Rodents are a significant zoonotic reservoir globally, and Europe has been identified as a hotspot for rodent reservoir diversity (Han et al., 2015). Furthermore, viral metagenomic surveys confirm that wild rodents carry a high and diverse viral burden, which includes several viruses closely related to human pathogens (Drexler et al., 2012(Drexler et al., , 2013Firth et al., 2014;Kapoor et al., 2013;Phan et al., 2011;Williams et al., 2018;Wu et al., 2018), including coronaviruses (Wang et al., 2020). Therefore, understanding the composition and dynamics of rodent viromes is an important goal that can help shed light on when and where these host communities may pose the greatest risk of zoonotic spillover to humans. However, our understanding of virus diversity in rodents and what shapes variation in rodent viromes within and among sympatric species remains limited.
To address these questions, we utilized a long-term capturemark-recapture study of several sympatric rodent species in Wytham woods, Oxfordshire, UK. Specifically, we characterized the viromes of three common resident species, Apodemus sylvaticus (wood mouse), Apodemus flavicollis (yellow-necked mouse) and Myodes glareolus (bank vole). These three species are ubiquitous across Europe, particularly in woodland habitats. They have fastpaced life histories, with females capable of producing multiple litters in her lifespan, which is typically less than 1 year. To characterize seasonal variation in viral communities, we generated metaviromic data from pooled faecal samples collected longitudinally from each species over a single year. By combining local microclimate and demographic data from the same period, we explored key factors that predict seasonal variation in the wild rodent virome.  (Watts, 1968) and social interactions (Raulo et al., 2021). One night of trapping on a single ~2.4-ha trapping grid was carried out approximately fortnightly year-round. Small Sherman traps (baited with six peanuts, a slice of apple and sterile cotton wool for bedding material) were set at dusk and collected at dawn the following day.
Newly captured individuals were PIT-tagged for unique identification. Faecal samples were collected from the bedding material with sterilized tweezers and frozen at −80°C within 10 h of trap collection. Traps that showed any sign of animal contact (traps that held captured animals and trigger failures where an animal has interfered with the trap but not been captured) were washed thoroughly with bleach between trapping sessions to prevent cross-contamination.
All live-trapping work was conducted with institutional ethical approval and under Home Office licence PPL-I4C48848E.

| Sample selection and processing
We randomly selected 133 individual faecal samples (57 A. sylvaticus, 25 A. flavicollis, 51 M. glareolus). Five sampling intervals were defined, which took into account the breeding cycle of the three ro-  Table S1.
The samples were processed as follows to enrich for RNA within encapsulated viruses: (i) frozen archived faecal samples were first pooled, then suspended in DNA/RNA Shield Stabilization Buffer (Zymo), vortexed thoroughly, and the supernatant was filtered through a 0.45-nm pore filter; (ii) RNase treatment (RNase One) to remove nonencapsulated RNA from the sample; (iii) RNA extraction using Zymo Quick Viral RNA and RNA Clean and Concentrator 5 kits; (iv) DNA digestion following RNA extraction; (v) ribosomal depletion with an Illumina Ribo-Zero Plus kit, which allows for ribosomal RNA removal in human, mouse rat, and bacterial samples, during sequencing library preparation. The Oxford Genomics Centre carried out sequencing library preparation, which included cDNA synthesis and sequencing on an Illumina NovaSeq 6000 platform.

| Viral genome reconstruction
A total of 355,917,017 pair-end reads of 150 bp were obtained after sequencing. Illumina adaptors were removed, and reads were filtered for quality scores ≥30 and read length > 45 bp using cutadapt 1.18 (Martin, 2011). A total of 352,872,111 cleaned paired-end reads were de novo assembled into 435,021 contigs by megahit 1.2.8 with default parameters (Li, Liu, et al., 2015). Viral contigs were identified by comparing the assembled contigs against the NCBI RefSeq viral database using diamond 0.9.22 with an e-value cutoff of <10 −5 (Buchfink et al., 2014). To eliminate false positives, all contigs that matched virus sequences were used as queries to perform reciprocal searches on NCBI nonredundant protein sequence database with an e-value cutoff of <10 −5 (Altschul et al., 1990). We considered each viral contig as a viral operational taxonomic unit (vOTU). The abundance of each vOTU contig was assessed by iterative mapping reads against each contig using bowtie2 2.3.4.3 (Langmead, 2010) and bbmap 35.34 (Bushnell, 2014). For viral contigs corresponding to complete or nearly complete contigs, we examined open reading frames (ORFs) using orf finder (parameters: minimum ORF size of 300 bp, standard genetic code, and assuming there are start and stop codons outside sequences) in geneious prime 2019.1.1 (Kearse et al., 2012) to exclude misassembled genomes. Information on the number of raw, cleaned and viral sequence reads per pooled sample is outlined in Table S2. Output data (blast results, viral contigs, read abundance) from the bioinformatic analyses can be found on DRYAD: https://doi.org/10.5061/dryad.612jm 645s

| Virus abundance and diversity metrics
After the assignment of contigs to vOTUs, we normalized the abundance of contigs to the total reads and individuals used in a pool. To reduce the impact of contamination in our analyses, we excluded viral contigs with less than one read per 10 million. The abundance of viruses was then compared using normalized read abundance. Virus diversity was assessed using the number of virus genera (hereafter "richness") and the evenness of virus genera (hereafter "evenness"), which was measured by calculating the Shannon entropy of virus genera in the community using the Shannon diversity index function in the R library vegan. Consequently, viral evenness ranges from 0 and 1, and indicates the degree to which the virus community is dominated by a particular genus (i.e., evenness = 0) or whether different genera are equally abundant (i.e., evenness = 1). To identify unique and shared viruses across all time points, we visualized the distribution of viral contigs (200 bp or longer) with a minimum of 20 reads among host species with Venn diagrams (Yan, 2021). To determine if our methods were capturing the majority of virus genera in the system, we used rarefaction curves to assess the saturation of virus richness. We then calculated additive partitioning diversity to quantify how virus richness varied between species and time points (Oksanen et al., 2020). Finally, to assess how virus composition changes over time and which virus genera shift through time, we undertook a hierarchical PERMANOVA (Anderson, 2001), with sampling intervals and host species as covariates and constraining permutations to within species only. Together, these analyses inform how sufficient these sampling efforts are for understanding wild animal viromes.
To reconstruct the picornavirus phylogeny, we assembled a multiple protein sequence alignment of 93 whole picornavirus genome sequences from the NCBI RefSeq viral database and eight picornavirus genome sequences identified in this study. A maximumlikelihood phylogeny was inferred with iq-tree version 2.1.3 (Minh et al., 2020) using the best substitutional model identified by modelfinder (Kalyaanamoorthy et al., 2017).

| Predictors of picornavirus richness and evenness
We evaluated drivers of two outcome variables-picornavirus richness and evenness-using Gaussian distributed generalized linear models (GLMs). We modelled picornaviruses in wood mice and bank voles separately and only modelled these virus-host combinations as up to six picornaviruses were found, and these hosts were sam- Since predictor and outcome variables were calculated at different frequencies (daily to seasonally), we used locally estimated scatterplot smoothing (LOESS) and generalized additive models (GAMs; Wood, 2011) to model a continuous estimate of each variable over the study period (June 2016 to Jan 2018 for predictor variables; Jan 2017 to Jan 2018 for outcome variables). Bimonthly estimates for picornavirus richness, picornavirus evenness (see Section 2.4), and host population density were inferred with LOESS, while bimonthly estimates for microclimate data (temperature, humidity and rain) were inferred with GAMs.
Environmental and host density impacts may have delayed effects on observed picornavirus richness and evenness. Therefore, we first identified the appropriate time lags (if any) for each predictor variable. Significant relationships between picornavirus (i) richness and (ii) evenness and the four predictors were identified for each diversity metric and host species using cross-correlation analysis. Cross-correlation analysis (ccf function in R) compares two time series and identifies similarities between the variables. Values range from −1.0 to 1.0; the closer the absolute value is to 1.0, the more linked the two variables are across time. In addition to identifying contemporaneous correlation, cross-correlation can be used to evaluate if there are lagged correlations (i.e., delayed but significant similarities between time series). We evaluated lags from 0 to 14 weeks in 2-week increments and identified significant residual autocorrelation values for each increment. If multiple lags were identified as significant for a given predictor variable, we selected the lag with the highest significant residual autocorrelation value (see Table S3) to use in GLM construction below. The maximum lag was set at 14 weeks to reflect the average lifespan of the wild rodents in the study (approximately 3 months).
We considered four separate GLMs per host species and diversity metric (i.e., AS vs. viral evenness, MG vs. viral evenness, AS vs. viral richness and MG vs. viral richness) to evaluate drivers of picornavirus diversity. However, prior to undertaking a GLM analysis, correlations among the four variables (with or without lags as determined by the cross-correlation analysis; Table S9) for each metric and host species were visually assessed in each GLM in R using the library "corrplot" (Wei & Simko, 2021). If the correlation coefficient was ≥0.7 ( Figure S1), we reduced the sets of GLMs considered accordingly (Table S3). We used the library "AICmodvg" (Mazerolle, 2020) for model selection, which considers the Akaike Information Criterion (AIC) and the number of parameters to determine the best fit model. Lastly, the GLM results were plotted using the library "jtools" (Long, 2020). Statistical analyses and most plots were undertaken in R version 4.1.1 (The R Core Team, 2021). Adobe illustrator 2021 was also used to visualize the abundance of common vertebrate-associated and bacteriophage viruses over time.

| Virome dynamics in wild rodents
Over a 1-year period (January 2017-January 2018), we characterized viruses in faeces from a total of 133 individual rodents (57 Apodemus sylvaticus, 25 A. flavicollis and 51 Myodes glareolus). For each of the five 2-3-month sampling intervals, we randomly selected up to 13 individual samples per species to create species-and sampling interval-specific pools for metagenomic sequencing (see Methods, Table S1 for further details). This approach resulted in five pools for both A. sylvaticus (wood mouse) and M. glareolus (bank vole) and three pools for A. flavicollis (yellow-necked mouse) which are less abundant at the sampling site.
Of the total quality-filtered and trimmed reads, 3.20% (~22.7 million [M]/711.8 M) were taxonomically assigned to known viruses (see Methods). Figure 1 provides an overview of the viruses detected across all rodent species (hereafter, "Wytham rodents"). Clean virus abundance ranged from 1.06 M to 2.88 M reads per pooled sample (Table S2) single-stranded (ss) DNA genomes, although our protocols should also detect DNA viruses undergoing active replication or transcription. Alternatively, bacteriophages could be preferentially enriched in shotgun metagenomic data sets due to their large genome sizes (>100 kb) (Dion et al., 2020). While this might be a contributing factor, the most abundant bacteriophage virus families in the Wytham rodent virome were Leviviridae (+ssRNA) and Microviridae (ssDNA), which have genome sizes ranging from 4 to 6.5 kb (Table S4).   (Figure 2b), which corresponds to spring/ summer months when host population density is low ( Figure S3).
The relative abundance of viruses ( Figure 2) is affected by changes in both virus occurrence between individuals and abundance within individuals. Sample pooling does not allow us to disentangle these; therefore, it is likely that viruses at low prevalence across the population or low virus abundance within individuals are not detected. However, we can still determine how much variation in viral richness (genera level) was observed at different levels-within a (pooled) sample, between species and between sampling periods (Table 1)-with additive diversity partitioning (Crist et al., 2003).
When considering viromes of all three host species together, around a third of virus richness (28.5%) was observed within pooled samples, 18.5% was observed between pooled samples within a given sampling period (i.e., among species,  To understand how the viral community composition differs between host species and sampling intervals, we undertook a hierarchical PERMANOVA (Anderson, 2001). Overall, we found that the virus community composition shifts significantly over time (p < .007), with host species having a weaker effect (p < .043) (Table S5). However, sampling interval and host species were not significant when considering vertebrate-associated or bacteriophage viruses separately (Table S5). We also identified the main virus genera that shift between sampling intervals ( Figure S4), which included Eucampyvirinae

| Extensive circulating virus diversity
Closer examination of the temporal patterns of the vertebrateassociated and bacteriophage viruses confirmed that considerable virus diversity was detected in the Wytham rodents, corresponding to different virus families, genera and genome architectures (i.e., single-or double-stranded, DNA or RNA genomes) displaying highly variable patterns of seasonal detection (Figure 3; Table S6).
In Figure 3,  Apart from picornaviruses, a more resolved taxonomic classification of the most common vertebrate-associated and bacteriophage virus families (e.g., Picobirnaviridae, Leviviridae and Microviridae) was not possible due to poor representation of these taxa in reference databases. As a result, it is challenging to ascertain more detailed information about these viruses. For example, which hosts do the bacteriophage infect and how many distinct virus species (i.e., virus genomes) are present? We aimed to partly address the latter by considering virus contigs similar in length to complete genomes (see Table S7), many of which are likely to represent new viruses. Based on this simple approach, there appear to be potentially 114 putative Picobirnavirus genomes (which are bisegmented), 21 putative Levivirus genomes and nine putative Microvirus genomes (Table S7).

| Seasonal cocirculation of picornaviruses
In Wytham rodents, picornaviruses were the most common and taxonomically well-characterized viruses. Furthermore, as they contain several important pathogens that affect human and animal health (e.g., Enterovirus and Apthovirus), we undertook a more detailed analysis to understand seasonal variation in picornavirus abundance and diversity. We assembled eight virus contigs for the most prevalent picornaviruses (see Table S8 for further details), representing partial and near-complete genomes. The eight genome sequences correspond to six distinct genera ( Figure 4a) and share between 48% and 95% amino acid sequence identity with their closest blast hits, which were primarily associated with mammalian hosts, such as bats and other rodent species. The normalized read abundance  When restricting the analysis to the eight picornavirus genomes, we observed broadly similar seasonally varying patterns in abundance, richness and evenness at the species level ( Figure S5).
However, there were notable differences compared with the analysis undertaken at the genera level. In wood mice, a comparable peak

| Drivers of picornavirus diversity
To explore the predictors of picornavirus diversity in Wytham woods, we focused on wood mice and bank voles, which were sampled in each of the five intervals. We evaluated three environmental variables (temperature, humidity and rain) using data collected from June 2016 to January 2017 from two weather stations located within the woodlands, together with approximately fortnightly estimates of host population density, calculated as the minimum number known alive (MNKA) per hectare from trapping data. Time series data on picornavirus diversity and the four variables were reconciled using interpolation techniques (see Methods). Specifically, we used a fortnightly interval to derive estimates of all variables at the resolution available for the host density data (Figures S3 and S6).
We undertook a cross-correlation analysis to select the single most informative time lag for each of the four variables (temperature, humidity, rain and host population density), as identified by the highest correlation coefficient (Table S9). The maximum time lag was set as 14 weeks to reflect the expected average lifespan of wood mice and bank voles (~3 months). We found a notable correlation between picornavirus diversity and the ecological conditions experienced by host species in the preceding weeks and months (r xy = 0.20-0.88; Table S9). For viral evenness, time lags in the four variables ranged from 10 to 14 weeks in wood mice and 6 to 14 weeks in bank voles, while for viral richness, time lags ranged from 2 to 14 weeks for both species (Table S9).
We constructed GLMs containing each time-lagged variable as predictors for each host species and diversity metric. Sets of GLMs were reduced accordingly to exclude highly correlated variables (i.e., >0.7; see Table S3, Figure S1). The results indicated that drivers of picornavirus diversity varied by species and diversity metric ( Figure 5). For both species, the temperature in the preceding 1-3 months was negatively correlated with viral evenness ( Figure 5; Table S10). In wood mice, viral evenness was additionally associated with a lower host population density 3 months previously, suggesting that the peak in viral evenness followed a period of low population density (in late winter) when the population mainly comprises overwintering individuals and when home ranges are largest and overlapping. In bank voles, viral evenness was negatively associated with rainfall in the previous 3 months and concomitant humidity ( Figure 5; Table S10). Viral richness was positively associated with concomitant host population density and temperature in both wood mice and bank voles ( Figure 5; Table S10).

F I G U R E 4 Picornavirus diversity and abundance. (a) Evolutionary relationship of picornavirus assembled genomes identified in Wytham rodent faecal viromes (coloured by their predominant host association) and a subset of known mammalian picornaviruses (in grey). (b)
Overall normalized read abundance of six picornavirus genera over time. Read counts per pooled sample below one per 10 million read threshold were excluded. Colours indicate association with host species (green, orange and blue correspond to AS, MG and AF). (c) Diversity of picornaviruses, measured as virus evenness and richness, over time. AS = wood mouse, AF = yellow-neck mouse, MG = bank vole. Although we found similar predictors associated with viral evenness and richness in both host species, as the same set of predictors was not evaluated in each GLM, interspecific differences should be interpreted with caution. Specifically, the absence of a predictor in our analyses does not necessarily mean it is not associated with viral diversity. Therefore, to better understand the extent of interspecific variation in shaping virus diversity, additional field data will be required to characterize viral communities on finer temporal scales (e.g., fortnightly or monthly). Our study corroborates previous findings that rodents harbour a substantial and diverse virus burden in the gastrointestinal tract (Firth et al., 2014;Phan et al., 2011;Williams et al., 2018;Wu et al., 2018) with individuals probably encountering a shifting array of seasonally abundant viruses over their lifetimes (Abolins et al., 2018;Firth et al., 2014), contributing to their highly activated immune state (Abolins et al., 2017). Most vertebrate-associated virus genera identified in the Wytham rodents have been detected previously in wild rodents in the USA and China (Firth et al., 2014;Phan et al., 2011;Williams et al., 2018;Wu et al., 2018). Cardiovirus and

| DISCUSS ION
Picobirnavirus have previously been reported in four major untargeted viral metagenomic surveys of wild rodents (Firth et al., 2014;Phan et al., 2011;Williams et al., 2018;Wu et al., 2018) (Horzinek et al., 1987;Jamieson et al., 1998).  (Begon et al., 2009;Carslake et al., 2006;Knowles et al., 2012;Telfer et al., 2002Telfer et al., , 2007. While these studies have focused on specific DNA viruses that are endemic in these species, they also observed heterogeneity in rodent virus epidemiology, including between years, host species, individuals and across different viruses (Begon et al., 2009;Carslake et al., 2006;Knowles et al., 2012;Telfer et al., 2002Telfer et al., , 2007 (Sobsey et al., 1988). In contrast, other picornaviruses, such as foot and mouth disease virus (genus Aphthovirus), appear to be less stable in the environment, with longer survival times observed at higher humidity and moderate temperatures (Abad et al., 1994;Mbithi et al., 1991;Mielke & Garabed, 2020). Although we observed clear seasonality in picornavirus detection and abundance, given the substantial temporal turnover in viral diversity, it is reasonable to assume that other viruses in Wytham rodents also circulated seasonally, especially those detected transiently in the population (e.g., Coronavirus, Paramyxovirus). In the future, we plan to develop mechanistic transmission models in these systems using field studies with a higher temporal resolution. Mechanistic models could be adapted to other rodent systems to forecast peaks and troughs in epizootics and test potential interventions in settings where zoonotic viruses are a risk to human populations.
Understanding viral community dynamics is key to predicting and mitigating human risk from known and unknown rodent zoonoses.
Improvements in sequencing technology that enable the identification and monitoring of RNA viruses longitudinally in wildlife are crucial to establishing the spatial, temporal and environmental factors that determine zoonotic risk. Previous work has shown that specific rodent-borne zoonotic viruses exhibit strong seasonal dynamics in the reservoir population (Fichet-Calvet et al., 2007;Luis et al., 2015;Tian et al., 2017). Nevertheless, by quantifying the virome dynamics, Simmonds, Sarah C. L. Knowles and Oliver G. Pybus. Accepted Manuscript version arising from this submission.

CO N FLI C T S O F I NTE R E S T
The authors declare there is no conflict of interest.

O PEN R E S E A RCH BA D G E S
This article has earned an Open Data Badge for making publicly available the digitally-shareable data necessary to reproduce the reported results. The data is available at Dryad, doi:10.5061/ dryad.612jm 645s, while associated code is available via Github: https://github.com/jnara g/Wytha m-roden t-virome.

DATA AVA I L A B I L I T Y S TAT E M E N T
The raw sequencing data generated in this study have been depos-

B EN EFIT-S H A R I N G S TATEM ENT
Benefits generated: Benefits from this research accrue from the sharing of our data and results on public databases as described above.