Impacts of radiation exposure on the bacterial and fungal microbiome of small mammals in the Chernobyl Exclusion Zone

1. Environmental impacts of the 1986 Chernobyl Nuclear Power Plant accident are much debated, but the effects of radiation on host microbiomes have received little attention to date. 2. We present the first analysis of small mammal gut microbiomes from the Chernobyl Exclusion Zone in relation to total absorbed dose rate, including both caecum and faeces samples. 3. We provide novel evidence that host species determines fungal community compo -sition, and that associations between microbiome (both bacterial and fungal) com munities and radiation exposure vary between host species. Using ambient versus total weighted absorbed dose rates in analyses produced different results, with the latter more robust for interpreting microbiome changes at the individual level. We found considerable variation between results for faecal and gut samples

Over the last decade, there has been a growing interest in the effect of contaminants on the composition of the gut microbiome, with some studies reporting changes in the two most prevalent bacterial phyla within the gut, namely Firmicutes and Bacteroidetes (Jin et al., 2017;Weldon et al., 2015;Wu et al., 2016). Different chemical stressors have been found to affect Firmicute: Bacteroidete (F:B) ratios, with As (Lu et al., 2014), Cd , chlorpyrifos (Joly Condette et al., 2015), permethrin (Nasuti et al., 2016) and pentachlorophenol (Kan et al., 2015) leading to decreases in F:B, whereas Pb (Wu et al., 2016) and carbendazim  exposure increases F:B ratios. Environmental pollutants can impact the gut microbiota resulting in changes to metabolite production and the immune system (reviewed in Jin et al., 2017). For example, colonic inflammation in mice has been shown to be associated with reduced F:B in the gut microbiota following pesticide exposure (Jin et al., 2016). Similarly, in rats, silver nanoparticle exposure decreases Firmicute abundance in the gut, with associated disturbance of immunomodulatory gene expression in the ileum (Williams et al., 2015).
Consequently, pollutant exposure may increase the susceptibility of animals to some diseases.
The administration of bacterial probiotics (particularly Lactobacillus spp.) can compensate for this in both humans and model organisms, reducing radiation-induced diarrhoea (Demers et al., 2014;Goudarzi et al., 2016;Liu et al., 2017;A. Zhang & Steen, 2018). These responses of the gut microbiome to high acute radiation doses have led to the suggestion that the gut microbiota could be a potential biomarker of radiation exposure (Goudarzi et al., 2016;Zhang & Steen, 2018).
At lower radiation exposures in contaminated environments, such as the Chernobyl Exclusion Zone (CEZ), some studies report changes in particular taxa of the animal microbiome . For example,  report a reduction in F:B of faecal samples from bank voles Myodes glareolus at their most contaminated sites (mean ambient dose rate 30 μSv/hr), and radiation-induced changes in the bacterial communities of feathers have been suggested for birds in the Chernobyl region (Czirják et al., 2010;Ruiz-González et al., 2016). Conversely,  found no effect of radiation on the skin microbiome of bank voles. If consistent relationships between microbiome and chronic low-level (and environmentally relevant) radiation exposure can be identified, this would present a valuable biomarker for evaluating radiological exposure internationally.
The extent to which radiation exposure is affecting wildlife in Chernobyl is highly contested Mousseau & Moller, 2011). A fundamental problem with many of the studies undertaken to date is that they use ambient dose rates (often reported in units of absorbed radiation dose rate for humans, µSv/hr), rather than estimating the total absorbed dose rate of study organisms, accounting for both internal and external exposure (Beaugelin-Seiller et al., 2020). As such, it has not been possible to accurately determine dose-effect relationships, making interpretation of these studies difficult. Here we present the first study of gastrointestinal (GI) tract microbiome composition in CEZ small mammals for which individual total absorbed dose rates have been estimated. Previous studies in the CEZ have only considered the bacterial microbiome of one small mammal species (bank vole) using faecal samples; here we report on the faecal microbiome of four small mammal species using faecal samples, as well as the first direct analysis of the gut microbiome using caecum samples from bank voles. In addition, our microbiome analysis includes both bacteria and fungi, extending the limited general knowledge on the fungal component of animal microbiomes.

| Field sampling in the Red Forest (2017)
The study was undertaken in line with ethical approval obtained from the University of Salford. In August 2017, we sampled small mammals from the Red Forest, an area of c. 4-6 km 2 over which pine trees were killed by radiation in 1986; subsequently, there has been sparse regrowth of deciduous trees and some understorey vegetation. In 2016, approximately 80% of the Red Forest was damaged by fire (Beresford et al., 2021). Our 2017 sampling sites ( Figure 1) included a total of eight sites across three burn categories, namely 'burnt with regrowth' (n = 2), 'burnt with minimal regrowth' (n = 3) and 'unburnt' (n = 3). At each of these sampling sites, a 60 m × 60 m trapping grid was used, with traps positioned at 10-m intervals (each grid comprised a total of 49 traps). To maximise trapping success, the trapping grids were established 1 week prior to the beginning of the study and pre-baited with rolled oats and carrots/cucumber. Trapping occurred over 8 consecutive days; traps were baited and set each evening and visited early in the morning to retrieve captured small mammals. The small mammals were transferred to the Chernobyl field station where each animal was live-monitored to determine its (n = 27) and bank voles Myodes glareolus (n = 22; Table 1). Faecal samples were immediately placed into vials containing 100% ethanol and subsequently stored at −20°C. Samples were transported under licence to the University of Salford (UK); sample integrity was maintained during transit using dry ice, and the samples were then stored at −20°C prior to DNA extraction. We used fur clipping to mark each small mammal prior to release at the point of capture, ensuring each collection over subsequent days was from a new animal. Only faeces from new animal captures were included in this study.

| Field sampling across the CEZ (2018)
Small mammals were trapped in July/August 2018 over 10 consecutive days, with only bank voles included in this study. Twelve tran-  Table 1).
Captured animals were transferred to the Chernobyl field station, where each animal was live-monitored to quantify the wholebody activity concentrations of both 137 Cs and 90 Sr using a new field portable Radioanalysis of Small Samples (ROSS) detector developed at the University of Salford (Fawkes, 2018). ROSS comprises a holding chamber with a capacity of 170 × 60 × 50 mm. Two CsI gamma detectors (each measuring 70 × 40 × 25 mm) were mounted on opposite sides of the sample holding chamber and two plastic scintillator beta detectors were mounted, one above (100 × 50 × 0.5 mm) and one below (100 × 60 × 0.5 mm) the chamber. The entire assembly was enclosed within a lead shield (>10 mm thickness). ROSS was calibrated using 137 Cs and 90 Sr standards developed by Chornobyl Center; standards ranged from 4 to 20 g to represent small mammals. We included 137 Cs-only standards, 90 Sr-only standards and mixed ( 137 Cs and 90 Sr) standards. Counting of the 137 Cs standards on the beta detectors provided a correction for the influence of 137 Cs emissions on 90 Sr recordings. Multiple background counts were performed daily (at least nine per day), and the LOD was estimated using the method described by Currie (1968).
After live-monitoring of radiation, bank voles were killed by an overdose of anaesthetic (isofluorane) followed by exsanguination (in accordance with Schedule 1 of the Animals (Scientific Procedures) Act 1986). The sex and mass of each animal was recorded. Gastrointestinal tracts (n = 142) were dissected immediately under sterile conditions, with surfaces and instruments sterilised between each dissection. Gastrointestinal tracts were stored in laboratory vials containing 100% ethanol at −20°C. The frozen vials were transported to the University of Salford under licence and stored as described for faeces.

| Dosimetry: Ambient dose rate
All dose data are provided as Supporting Information. At every trapping location in 2017 and 2018, ambient dose rate (µSv/hr) was measured using an MKS-01R metre at 5 cm above the soil surface.

| Dosimetry: Estimation of small mammal total absorbed dose rate for the 2017 study
Soil samples (0-10 cm soil depth) were available from each of the trapping sites used in the 2017 study. These samples were analysed using laboratory detectors at Chornobyl Centre to determine 137 Cs and 90 Sr activity concentrations within the soil (see Beresford, F I G U R E 1 Location of the study sites in the CEZ where small mammals were trapped in 2017 and 2018; the approximate location of the Red Forest is indicated by the black rectangle. The underlying 137 Cs soil data shown (decay corrected to summer 2017) are from the study by Shestopalov (1996) Barnett, et al., 2020 for methodology). For each small mammal species, an external dose conversion coefficient was calculated using the ERICA Tool version 1.2 (Brown et al., 2016). To define the geometry for each species, the length, width and height were determined through literature review. Soil activity concentrations were inputted into the ERICA Tool, and external dose rates were estimated using the derived external dose conversion coefficients and appropriate occupancy factors (assuming 50% of time in soil and 50% on the soil surface for mice and 70% in soil and 30% on soil for bank voles).
The measured 137 Cs whole-body activity concentrations were used to determine the internal absorbed dose from 137 Cs. In 2017, the internal 90 Sr activity concentrations were not directly measured; these were estimated using the species-specific transfer parameters measured in the CEZ (Beresford, Barnett, et al., 2020) and the soil 90 Sr activity concentrations for the appropriate sampling site. For each small mammal species, an internal dose conversion coefficient was calculated using the ERICA Tool and the same assumed geometries as used for the external dose conversion coefficient derivation.
The ERICA Tool was then run using the default radiation weighting factors to calculate the total weighted absorbed dose rate. While other radionuclides (e.g. Pu-isotopes and 241 Am) are present in the CEZ, Beresford, Barnett, et al. (2020) demonstrated that the contribution of these isotopes to the total absorbed dose rate of small mammals within the Red Forest was low (<10%).
For each individual animal, the total weighted absorbed dose rate (hereafter referred to as the total absorbed dose rate) was calculated by summing the internal and external dose rates for that individual.

| Dosimetry: Estimation of small mammal total absorbed dose rate for the 2018 study
Soil activity concentrations were not available for all of the sites within the 2018 study. However, Beresford, Barnett, et al. (2020) and Beresford et al. (2008) demonstrated that, at worst, the estimated external dose from 137 Cs and the external ambient dose field at sites in the CEZ differ by a factor of 3. Based on our small mammal dose rate data from 2017, the mean ratio of external dose from 137 Cs to the external ambient dose field is 0.98. Therefore, the external gamma dose rates measured at each trapping location were used to estimate the external absorbed dose rate for each small mammal using the occupancy factors defined above. Note the ERICA Tool assumes a shielding effect from fur and skin for external beta exposure (Ulanovsky et al., 2008); this assumption was also adopted by the International Commission on Radiological Protection (ICRP, 2008).
The estimated contribution of 90 Sr (a beta emitter) to the external whole-body dose rate of small mammals is therefore negligible and could be ignored for the 2018 study.
The 137 Cs and 90 Sr whole-body activity concentrations measured using ROSS were input into the ERICA Tool, and the species-specific internal dose conversion coefficients were used to estimate the TA B L E 1 Sample sizes of host species used in the study, and associated sex, total absorbed dose rate, burn and site categories for these samples. Animals estimated to receive total absorbed dose rates of <4 µGy/hr were assigned 'low', those with estimated dose rates of 4-42 µGy/hr assigned 'medium', and those >42 µGy/hr assigned to the 'high' category. All samples from 2017 were collected within the Red Forest but from areas that had experienced different degrees of damage from forest fires. Samples from 2018 were either collected within or outside the Red Forest internal absorbed dose rate for each animal. At the lowest contamination sites in 2018, some of the whole-body activity concentrations for both 137 Cs and 90 Sr were below the LOD. Using these LOD values to determine total absorbed internal dose rate led to a maximum estimated dose of 0.6 µGy/hr, introducing some uncertainty in radiation exposure estimates at the lowest end of our total absorbed dose rate range.

| Dosimetry: Incorporation of estimated dose rates with subsequent analyses
The ICRP has, over the last decade, developed an approach to ra- hr. As such, animals estimated to receive total absorbed dose rates of <4 µGy/hr (i.e. below the dose rate at which effects would be anticipated for mammals) were assigned 'low'. Those with estimated dose rates of 4-42 µGy/hr (i.e. within the band of anticipated effects) were assigned 'medium', and those >42 µGy/hr (i.e. above the dose rate at which effects would be anticipated for mammals) were assigned 'high'. The 'high' and 'low' total absorbed dose rates are ineffect also a comparison of inside and outside the Red Forest due to the high radionuclide deposition that occurred in the area now known as the "Red Forest" (i.e. the 'inside' and 'outside' Red Forest site categories; Table 1).
We correlated ambient and total estimated absorbed dose rates using a Spearman's rank correlation. To quantify whether correlation coefficients varied based on the radiation dose measure used, we also repeated the correlations for each total absorbed dose rate category separately and visualised these using a scatterplot in ggplot2 (Wickham, 2009).

| DNA extraction and molecular work
For faecal samples, we extracted DNA from the full sample (~0.1 g) of the four host species. For gut samples, we isolated ~25% of the distal end of the caecum of bank voles and homogenised the contents by hand in sterile Petri dishes, before weighing out ~0.1 g for DNA extraction. We conducted all DNA extractions using the PureLink™ Microbiome DNA Purification Kit (Invitrogen) according to the manufacturer's instructions.
To identify bacterial communities, we conducted 16S rRNA gene amplicon sequencing of the v4 region using F515 and R806 primers (~250 bp; Caporaso et al., 2010, according to Kozich et al., 2013Griffiths et al., 2018 We identified fungal communities by sequencing the ITS gene using ITS1F and ITS2 primers (~150-500 bp) using a modified protocol of Smith and Peay (2014) and Nguyen et al. (2015), as in Griffiths et al. (2019). We ran PCRs in duplicate using thermocycling conditions of 95°C for 10 min, followed by 28 cycles of 95°C for 30 s, 54°C for 45 s and 72°C for 60 s; with a final extension at 72°C for 10 min. We included extraction blanks and a mock community as negative and positive controls respectively. We quantified and normalised individual libraries as described above, before conducting full paired-end sequencing using Illumina v2 (2 × 250 bp) chemistry on an Illumina MiSeq at the University of Salford.

| Pre-processing of amplicon sequencing data
We conducted all data processing and analysis in RStudio (v1.2.1335; RStudio Team, 2016) for R (v3.6.0; R Core Team, 2017). A total of 13,371,018 raw sequence reads were generated during 16S rRNA gene amplicon sequencing, which we processed in DADA2 v1.5.0 (Callahan et al., 2016). Modal contig length was 253 bp once pairedend reads were merged. We removed sequence variants (SVs) with length >260 bp (26 SVs; 0.002% of total sequences) along with chimeras and five contaminant SVs identified using the decontam package (Davis et al., 2017). We assigned taxonomy using the SILVA v132 database (Quast et al., 2013;Yilmaz et al., 2014). DADA2 identified 20 unique SVs in the sequenced mock community sample comprising 20 bacterial isolates. We stripped out mitochondria from samples along with SVs with <0.0001% abundance across all samples.
We removed three samples from which poor sequence data were obtained (<1,000 reads), leaving a median of 28,563 reads (7, We obtained a total of 2,778,887 raw sequence reads during ITS gene sequencing. We trimmed the remaining adapters and primers using cutadapt (Martin, 2011) in RStudio. As with 16S rRNA sequence data, we pre-processed ITS amplicons in DADA2 v1.5.0 (Callahan et al., 2016). Modal contig length was 219 bp (167-457 bp) once paired-end reads were merged. We did not conduct additional trimming based on sequence length as the ITS region is highly variable (Schoch et al., 2012). We removed chimeras and one contaminant using the decontam package, and then assigned taxonomy using the UNITE v7.2 database (UNITE, 2017).
DADA2 identified 12 unique SVs in the sequenced mock community sample comprising 12 fungal isolates. We removed 54 samples from which poor sequence data were obtained (<500 reads),

| Community analyses
For both bacterial and fungal community data, we normalised the clean count data using centred-log ratio (clr) transformations (Gloor et al., 2017) in phyloseq (McMurdie & Holmes, 2013), and visualised beta-diversity (based on species and sample types, i.e. gut or faeces) using PCA plots with Euclidean distances in gg-plot2 (Wickham, 2009). We used PERMANOVAs to test for differences in beta-diversity according to species, sample type, sex and total absorbed dose rate category using the adonis function in the vegan package, performing marginal tests for individual terms (Oksanen et al., 2018). We agglomerated the data to family level and visualised differences in clr-transformed data according to the five sampling groups (faecal samples from the three mice species, plus faecal and gut samples from bank voles) using jitter box plots in ggplot2 (Wickham, 2009 (Lahti & Shetty, 2017) to identify relationships between the two radiation dose measures (ambient and total) and clr-transformed 16S and ITS rRNA sequence data, agglomerated to genus level. These analyses were conducted separately for the gut and faecal samples, according to host species.
We then visualised the resultant correlation coefficients using heat maps in ggplot2 (Wickham, 2009).
We calculated F:B ratios in vole guts using both clr-transformed data and data converted to relative abundance (as done by . We also calculated F:B in faecal samples of all four mammal species using relative abundance data.
We visualised these ratios according to total absorbed dose rate category using jitter plots. We tested for differences between total absorbed dose rate categories within a set of data using Kruskal-Wallis nonparametric tests, with Dunn's pairwise tests and Hochbergadjusted p values where necessary.

| How do ambient dose rates compare to total absorbed dose rates?
Many previous studies of radiation effects in Chernobyl and other  Figure S1). Given that the estimated total absorbed dose rate provides a better estimation than ambient dose rate of each individual's radiation exposure and microbiome analysis is at the individual level, we used total absorbed dose rates for the majority of our analyses, but for a few comparisons we have also presented the relationship with ambient dose rate to illustrate the limitations of this commonly used approach.
In this study, we used the ERICA dosimetry approach, which assumes shielding by fur and skin of beta radiation and consequently the external dose rates from 90 Sr are estimated to be negligible. We acknowledge that estimates using different modelling approaches may lead to a higher estimated external dose rate from 90 Sr (e.g. Gaschak et al., 2011). We have estimated the external dose contributions using an alternative model (available from https://wiki.  p = 0.001) and total absorbed dose rate category (F 2,212 = 1.736, R 2 = 0.015, p = 0.001), but not sex (F 2,212 = 1.161, R 2 = 0.005, p = 0.148).

| How does microbiome beta-diversity vary according to host factors and total absorbed dose rate
Differences between host species were much more evident for bacterial community composition than for fungal community composition (Figure 2a,b), for which 22.4% and 6.6% of the variation was explained by host species respectively. There were a number of differences in the clr values of the most abundant bacterial and fungal families according to host species and sample type (Figure 2c,d); a full description with statistical testing can be found in Supporting Information.

| How do alpha-diversity and beta-diversity of faecal samples vary according to host species, burn category and total absorbed dose rate?
When using faecal samples (i.e. 2017 data) only, the PERMANOVA indicated that host species (F 3,128 = 11.944, R 2 = 0.217, p = 0.001; Figure S4), burn category (F 2,128 = 1.632, R 2 = 0.020, p = 0.004; Figure S4) and sampling site (F 5,128 = 1.562, R 2 = 0.047, p = 0.001) had a detectable effect on beta-diversity of faecal bacterial communities, but that total absorbed dose rate category (F 1,128 = 0.912, R 2 = 0.006, p = 0.616) and sex (F 1,128 = 1.103, R 2 = 0.007, p = 0.228) did not. There were only sufficient samples for yellow-necked mice to visualise differences in microbiome composition across all three burn categories (Table 1) There was a significant relationship between total absorbed radiation dose and fungal community alpha-diversity across all four host species (all p < 0.05; Table S1), whereby as radiation dose increased, fungal diversity decreased. Total absorbed radiation dose did not have a significant effect on bacterial community alpha-diversity (all p > 0.05; Table S1).

| How do alpha-diversity and betadiversity of bank vole gut samples vary according to site category and total absorbed dose rate?
The PERMANOVA showed total absorbed dose rate category The partial Mantel test for association between microbial community beta-diversity and total absorbed dose rate distance There were no significant effects of total absorbed radiation dose rate on alpha-diversity of microbial communities (all p > 0.05; Table S2).

| How do different microbial taxa correlate with the two radiation dose measures?
Faecal and gut samples of bank voles showed considerably different fungal and bacterial association patterns (Figures 3 and 4; note that faecal samples were collected in 2017 from the Red Forest and guts collected from different animals and sites over a wider area of the CEZ in 2018). Fungal and bacterial association patterns of faecal samples from the four small mammal species were also markedly different to one another (Figures 3b and 4b).
The association analysis identified one bacterial family in bank vole gut samples that positively correlated with total absorbed dose rate (Lachnospiraceae, p < 0.001; Figure 3) and one that negatively correlated with total absorbed dose rate (Muribaculaceae, p < 0.001; Figure 3). Lachnospiraceae also significantly correlated with ambient dose rate (p < 0.001; Figure 3). The association analysis also identified one bacterial family from bank vole faeces that correlated with total absorbed dose rate (Saccharimondaceae, p = 0.004; Figure 3).
Two fungal families in bank vole gut samples were correlated with total absorbed dose rate, with Steccherinaceae negatively correlated (p = 0.006) and Strophariaceae positively correlated (p = 0.006; Figure 4). Steccherinaceae also correlated with ambient dose rate (p = 0.002; Figure 4). There were no fungal families from faecal samples that were correlated with total or ambient dose rate (Figure 4).

| How do Firmicute: Bacteroidete ratios vary according to total absorbed dose rate category?
When using clr-transformed data, F:B was <0 in vole guts ( Figure S12a). When using relative abundance data, voles in the 'high' total absorbed dose rate category had slightly higher F:B than those in the 'low' and 'medium' categories ( Figure S12b). The Kruskal-Wallis model indicated no effect of total absorbed dose rate category on F:B in bank vole guts (X 2 = 5.556, df = 2, p = 0.062). For the faecal sample data, only striped field mice and yellow-necked mice had data for animals in more than one absorbed dose rate category, that is, medium and high for both. The Kruskal-Wallis analysis was not significant for either striped field mice (X 2 = 0.012, df = 1, meaning there were no differences in F:B between the 'medium' and 'high' categories for these two species ( Figure S12c).

| D ISCUSS I ON
In this study we present the first analyses of small mammal faecal and gut microbial communities from the CEZ for which individual total absorbed dose rates have been estimated. This study also presents the first data from Chernobyl on the fungal component of the gut microbiome, and considers a wider range of small mammal species than previously studied, which have been limited to bank voles . Previous papers used faecal samples to characterise the small mammal gut microbiome , whereas our study also provides the first data on the true gut microbiome of Chernobyl bank voles using samples from the distal section of the caecum.

F I G U R E 4 Correlations between the two radiation dose measures (total absorbed and ambient dose rates) and (a) clr values of fungal genera in vole guts and (b) clr values of fungal genera in faecal samples from four small mammal species
We provide novel evidence that radiation has limited effects on the microbial communities associated with the gut of a wild mammal species. Across all four host species, partial Mantel tests demonstrated that animals experiencing similar radiation exposure also had similar microbiome composition; this relationship was independent of sampling location. Although we identified detectable effects of radiation exposure category on both bacterial and fungal gut microbiome composition of bank voles, the effect sizes were limited, and the results were not robust once geographic distances were included in the analysis. Given that microbes are actually highly resistant to death when exposed to radiation (e.g. a high acute (>10 kGy) radiation dose is needed to eliminate fungi and bacteria from soils; (McNamara et al., 2003;Whicker & Schultz, 1982), environmental radiation exposure at sites such as Chernobyl is unlikely to affect the gut microbiome directly (but see discussion below about indirect drivers).
For bank voles, we observed differences in microbial communities associated with the gut and faeces, in agreement with previous studies of various host species (Ingala et al., 2018;Leite et al., 2019).
We also observed significant differences in the relationships be- cluding from a number of sites that had been recently burnt), which may also be influencing the observed differences between the gut and faecal samples. In addition, the host microbiome, including the fungal component, is influenced by the immune system (Enaud et al., 2018), which in turn is affected by radiation exposure (Jin et al., 2017), and so there is a clear route for radiation exposure to indirectly influence the host microbiome. However, given the vital role of the microbiome in host functioning, perhaps more important than compositional changes in the microbiome are the functional changes that radiation exposure confers. Further work with shotgun metagenomics or metatranscriptomics is required to determine the effect of radiation exposure on microbiome function, and the implications of this for host fitness.
We found that sampling site is also a significant predictor of bacterial beta-diversity. Geographical location is known to affect bacterial community composition Griffiths et al., 2018), and here we also provide novel evidence that geography affects fungal community composition.  (Beresford, Scott, et al., 2020;Garnier-Laplace et al., 2013;Smith, 2020) and/or an active increase in the consumption of plant-based foods. Indeed, F:B in faeces has previously been used as a marker of changes in diet (Carmody et al., 2015). However, as    (Heisel et al., 2017;Wegley et al., 2007), degradation of dietary carbohydrates (Yang et al., 2018), resistance to infectious disease (Kearns et al., 2017), modulating host immune responses (Enaud et al., 2018;Yeung et al., 2020), and even behavioural traits such as host dispersal (Lu et al., 2010). We also show host species predicted bacterial community composition, which supports the results of previous studies on a range of host species (Davenport et al., 2017;Mazel et al., 2018;Youngblut et al., 2019), including small mammals (Knowles et al., 2019). We found no effect of sex on bacterial or fungal communities of the gut or faecal samples from any host species; previous studies have found mixed effects of sex on microbiome composition of small mammals (Knowles et al., 2019;Weldon et al., 2015).

| CON CLUS IONS
Here we show the gut communities of small mammal species are not directly affected by total radiation dose in the CEZ, but they are susceptible to differences in habitat type or quality. We found evidence that faecal communities are associated with radiation exposure independent of geographic location, but that these communities were not representative of true gut communities. In this study, we have identified two bacterial (Lachnospiraceae and Muribaculaceae) and two fungal (Steccherinaceae and Strophariaceae) families in the guts of bank voles, which may serve as biomarkers of exposure to radiation. However, our findings would need verification in further studies considering a range of host species before these families could be recommended as robust biomarkers of small mammal radiation exposure.
In contrast to the findings of a published study of small mammal microbiomes in the CEZ , we did not see any effect of estimated radiation exposure on the F:B ratio (the earlier papers considered ambient dose rate only). Recognising the high variability in the individual-level radiation exposure measurements at some of our sites, our results provide evidence that total absorbed radiation dose should be used for radiation effects studies rather than sitelevel ambient dose rate measurements. Ambient dose rate was also not a reliable predictor of comparative total dose rates.
Given the importance of the microbiome to host health and the limited studies on microbiome (especially fungal microbiome and gut microbiome) in wild animals, further studies are required to understand whether different host species respond differently to radiation exposure in the CEZ, and the mechanisms of host physiology that regulate these. For this, it is important to establish directionality, that is, whether the host microbiome alters host physiology in response to radiation exposure or vice versa. Furthermore, changes in diet resulting from the impacts of radiation on the ecosystem (e.g. in an area such as the Red Forest), rather than the host, may also be expected to affect the gut microbiome. More work is required to understand the mechanisms that are driving changes in host microbiomes of wildlife in general, and the implications of this for host function and fitness.

ACK N OWLED G EM ENTS
The work described in this paper was conducted within the TREE

DATA AVA I L A B I L I T Y S TAT E M E N T
Sequence data are available from the NCBI SRA database under project numbers PRJNA594002 and PRJNA592322.