Comparable response of wild rodent gut microbiome to anthropogenic habitat contamination

Species identity is thought to dominate over environment in shaping wild rodent gut microbiota, but it remains unknown whether the responses of host gut microbiota to shared anthropogenic habitat impacts are species‐specific or if the general gut microbiota response is similar across host species. Here, we compare the influence of exposure to radionuclide contamination on the gut microbiota of four wild mouse species: Apodemus flavicollis, A. sylvaticus, A. speciosus and A. argenteus. Building on the evidence that radiation impacts bank vole (Myodes glareolus) gut microbiota, we hypothesized that radiation exposure has a general impact on rodent gut microbiota. Because we sampled (n = 288) two species pairs of Apodemus mice that occur in sympatry in habitats affected by the Chernobyl and Fukushima nuclear accidents, these comparisons provide an opportunity for a general assessment of the effects of exposure to environmental contamination (radionuclides) on gut microbiota across host phylogeny and geographical areas. In general agreement with our hypothesis, analyses of bacterial 16S rRNA gene sequences revealed that radiation exposure alters the gut microbiota composition and structure in three of the four species of Apodemus mice. The notable lack of an association between the gut microbiota and soil radionuclide contamination in one mouse species from Fukushima (A. argenteus) probably reflects host “radiation escape” through its unique tree‐dwelling lifestyle. The finding that host ecology can modulate effects of radiation exposure offers an interesting counterpoint for future analyses into effects of radiation or any other toxic exposure on host and its associated microbiota. Our data show that exposure to radionuclide contamination is linked to comparable gut microbiota responses across multiple species of rodents.


| INTRODUC TI ON
Gut microbiota provide many essential services to their animal hosts, including provision of diverse metabolites and nutrients, and protection against pathogens and toxic compounds (Parfrey et al., 2018;Suzuki, 2017). As persistence of these functions impact host health, variation of the gut microbiota composition holds importance for determining animals' adaptive potential in the face of environmental change (Alberdi et al., 2016). Despite such importance, responses of host-associated microbiota to environmental perturbations, particularly derived from human activity, remain largely unknown. At least partly, this is due to the lack of detailed characterization of microbial communities associated with animals in their natural habitat, where many anthropogenic stressors are now commonplace (Carthey et al., 2019;Rocca et al., 2019;Trevelline et al., 2019).
Wild animals living in sympatry can be expected to share exposures to common environmental conditions. These include, for example, persistent contaminants (e.g., xenobiotics, metals, radionuclides, microplastics) distributed due to their deliberate use or accidental release to the environment. Although a number of studies have shown that exposure to a variety of such contaminants can perturb gut microbiota diversity and composition (Chassaing et al., 2015;Goudarzi et al., 2016;Jin et al., 2019;Richardson et al., 2018), these studies were conducted with laboratory animals only (e.g., mice and rats) and lack broader ecological reality. In wild mammals, species identity is thought to dominate over environment in shaping gut microbiota (Knowles et al., 2019), and yet in many taxa (e.g., rodents, primates, other mammals) the strength of host species effects varies depending on features of host ecology including diet, lifestyle and space use strategy (Grond et al., 2019;Knowles et al., 2019;Moeller et al., 2013;Perofsky et al., 2019). Hence, it is unclear whether the responses of host gut microbiota to shared anthropogenic habitat impacts are species-specific (distinct responses) or if the general gut microbiota response is comparable (parallel responses) across host species.
Exposure to radionuclides presents diverse risks with overall negative impacts on animal health (Møller & Mousseau, 2006).
Numerous human activities worldwide have contributed to radionuclide contamination of the environment through nuclear weapons tests, uranium mining and nuclear accidents (Lourenço et al., 2016). The largest releases of radionuclides into the environment occurred due to the accidents at Chernobyl and Fukushima nuclear power plants (NPPs) on April 26, 1986 andMarch 11, 2011, respectively. In response to the accidents, humans were evacuated from a 4,760-km 2 area in Ukraine (and partly in Belarus) and from a 1,150-km 2 area in Japan; these abandoned areas are often respectively referred to as the Chernobyl Exclusion Zone (CEZ) and the Fukushima Evacuation/Exclusion Zone (FEZ). Due to environmental contamination by radionuclides with long half-lives (principally 137 Cs, 30 years; and 90 Sr, 29 years) access to these evacuation zones remains severely restricted for human habitation Harada et al., 2014), but not wildlife. Animals inhabiting the CEZ and FEZ thus provide the best-studied models of the biological impacts of exposure to radionuclide contamination.
A number of ecological studies have documented negative effects of exposure to radionuclides in wildlife from Chernobyl and Fukushima (reviewed by Lourenço et al., 2016;Møller & Mousseau, 2006;Strand et al., 2017). For example, radionuclide contamination at both accident sites has been associated with negative health impacts, evident as an apparent increase in genetic damage, oxidative stress (Bonisoli-Alquati et al., 2010;Boratyński et al., 2014;Einor et al., 2016;Hiyama et al., 2012;Lourenço et al., 2016) and various developmental abnormalities (Hayama et al., 2017;Møller et al., 2011), with fitness consequences and population-level effects . Only a few studies have attempted a direct hypothesis-driven comparison of the biological effects of radionuclide contamination at the two accident sites; a consistent decline in the abundance of birds with increasing levels of radionuclide contamination (Møller et al., 2012), and little notable effects of radiation on the abundance of large mammals (Lyons et al., 2020;Webster et al., 2016) were reported at both the CEZ and the FEZ.
Despite the efforts in surveying Chernobyl and Fukushima wildlife, only a few studies have quantified the effects of radionuclide contamination on host-associated microbiota; the available studies were all conducted in the CEZ only. For example, the total cultivable bacterial loads from feathers of birds nesting at Chernobyl were negatively correlated with radionuclide contamination levels (Czirják et al., 2010;Ruiz-González et al., 2016). Similarly, the diversity of the gut microbiota of earthworms collected from contaminated areas in the CEZ was reduced compared with uncontaminated control areas (but see effects of soil pH) (Newbold et al., 2019). In contrast, results from the only studied mammal, the bank vole Myodes glareolus, indicate that radiation exposure does not reduce estimates of gut or skin microbiota diversity . Instead, exposure to radionuclides was associated with a major change in bank vole gut microbiota composition, notable as a reduction in the proportion of Bacteroidetes and an increase in Firmicutes . As a result, bank voles inhabiting areas contaminated by radionuclides can be identified by their distinct gut microbiota. Importantly, it remains unknown whether such gut microbiota responses to radiation exposure are specific to bank voles or represent a more general pattern similar across other wild rodents from Chernobyl, or even other nuclear accident sites similarly contaminated with radionuclides, such as Fukushima.
To obtain a general assessment of the effects of exposure to radionuclide contamination upon rodent gut microbiota, we analysed faecal samples (n = 288) from two pairs of mouse species (Apodemus flavicollis, A. sylvaticus, A. speciosus, A. argenteus) that occur in sympatry in habitats surrounding the Chernobyl and Fukushima nuclear accident sites. We hypothesized that radiation exposure has general impact on rodent gut microbiota, and thus similar levels of exposure would (i) alter gut microbiota composition in all mouse species, with the overall patterns comparable to those reported in bank voles living under chronic radiation exposure . Specifically, we predicted (ii) that such altered community composition would be characterized by an increase in the ratio of the Firmicutes to Bacteroidetes phyla (F:B), which could possibly act as a "biomarker" of exposure to radiation; (iii) that exposure to radiation also would have little notable effect on the diversity of mouse gut microbiota; and (iv) that exposure to radiation would select for distinct gut microbiota structure in wild mice as found for bank voles.

| Study species
Apodemus flavicollis and A. sylvaticus occur in sympatry throughout much of the Western Palearctic region, while A. speciosus and A. argenteus are also sympatric, but endemic to Japan (Michaux et al., 2005;Suzuki et al., 2008). These pairs of Apodemus species are also among the most common mammals in forested areas around the respective Chernobyl and Fukushima accident sites (Baker et al., 1996;Beresford et al., 2008;Kubota et al., 2015). These mice are typical granivorous rodents whose diets are largely dominated by mast seeds (e.g., acorns and other nuts) or weed seeds, but sometimes include vegetative plant material, invertebrates and/or fungi (Butet & Delettre, 2011). A. flavicollis and A. sylvaticus have similar ecology and diets (Knowles et al., 2019;Michaux et al., 2005;Ozaki et al., 2018), whereas despite overlapping habitats, soil-dwelling A. speciosus and tree-dwelling A. argenteus segregate vertically and differ in dietary preferences (specialist and generalist, respectively) (Oka, 1992;Sato et al., 2018).

| Study design
Trapping locations were designed so that animals had experienced low-dose rate radiation exposure (4-40 µGy hr -1 ) relevant for assessment of potential impacts of radiation on wildlife (Real & Garnier-Laplace, 2019). We assigned trapping locations as either (i) contaminated or (ii) uncontaminated (Figure 1; hereafter "treatment") based on a contrast in the ambient radiation dose rates (radionuclide contamination of the environment, measured at 1 cm above the ground using a hand-held Geiger counter ( Mice were sampled from mixed woodland areas with similar habitats across both locations of high and low radionuclide contamination. In the CEZ, we sampled replicate (spatially separated) locations within each treatment to deconfound effects of exposure to radiation from other potential environmental factors specific to a certain location. Treatment-level replicate sites in Ukraine were separated by 10-80 km (Figure 1), which exceeds seasonal dispersal capabilities of the study species (<1.5 km for adults, Stradiotto et al., 2009). In Japan, widespread agriculture around the FEZ prevented such replication. Hence, to minimize site-specific effects we set a large number (n = 32) of trapping locations across a ~25-km transect of the most contaminated "difficult-to-return" zone (Harada et al., 2014) and also sampled from diverse locations in uncontaminated areas (n = 19) ( Figure 1).

F I G U R E 1
Map of the study areas with mice trapping locations in Ukraine (left) and Japan (right) shown by points. Areas contaminated (CH, FH in orange) and uncontaminated (CL, KL, FL, green) with radionuclides within the Chernobyl and Fukushima Exclusion Zones (CEZ/ FEZ) are shown. Dashed line represents the border around the CEZ in Ukraine (area of ~2,050 km 2 ). The border around the FEZ in Japan is not shown; mice trapping locations in contaminated areas within the FEZ were located within the most contaminated "difficult-to-return" zone. The figure was created using the ggmap version 3.0.0 package in r

| Mouse trapping and sampling
Mice were caught by live trapping .  (Godsall et al., 2014;Oka, 1992;Stradiotto et al., 2009), in both study sites the minimum distance between trapping locations was at least 500 m.
Captured animals were transported to field laboratories for species' identification, and classification of sex. We also recorded body mass, head width and maturity (juvenile, subadult, adult) (

| Read data processing
Read data were demultiplexed, and adapters and spacers were removed by FIMM. Sixteen samples were not analysed due to a low TA B L E 1 Mean estimates of the internal, external and total radiation dose rates for each of the four sampled mouse species (n = 288) inhabiting areas contaminated with radionuclides (CH and FH) and uncontaminated areas (CL, KL, FL) in Ukraine and Japan A. speciosus n = 77, A. argenteus n = 53) were processed using dada2 (Callahan et al., 2016) in qiime2 version 2019.4 (Bolyen et al., 2019).
Briefly, reads were truncated at the 3′ end to remove low-quality base calls (<Q25, forward reads at 243 bp and reverse at 176 bp). Primer sequences were trimmed from the 5′ end of all reads (dada2 trim-left-f 19, trim-left-r 20) (Caporaso et al., 2011), after which data were denoised and dereplicated and putative chimeric sequences removed using default parameters in dada2 (Callahan et al., 2016). After quality control, the feature-table retained 11,930,717 sequences, with 4,510 amplicon sequence variants (ASVs) (Callahan et al., 2016). Taxonomy was assigned using a naïve Bayes classifier that has been pretrained on the Greengenes version 13_8 16S rRNA database (with sequences trimmed to the V4 region bounded by 515F/806R primers, and clustered at 99% identity) . The ASVs represented by fewer than 10 reads across all samples were removed in order to filter out potential errors and spurious artefacts (such a filtering strategy has little impact on the alpha-and beta-diversity estimates, Figure S1; see test details in Supporting Information). This step left 11,927,082 sequences (mean =43,849, range 10,553-263,628 sequences per sample) and 3,760 ASVs. These data were rarefied to 10,553 sequences per sample (3,731 ASVs) and this normalized (Weiss et al., 2017) feature-table was used for subsequent analyses.

| Radiation dose estimation
The total radiation exposure of each mouse was determined from the sum of the external (from surrounding environment) and internal (from ingested particles) radiation dose rates (Table 1; Supporting Information). External radiation exposure (absorbed dose) of mice was estimated from the ambient radiation dose rate levels (radionuclide contamination of the environment, typically measured using a hand-held Geiger counter) at trapping locations (Table S1).
This approach has been experimentally verified using implanted thermoluminescent dosimeters (TLDs) that directly measure absorbed external radiation dose rate in vivo (Beresford et al., 2008;Chesser et al., 2000;Lavrinienko et al., 2020). We also conducted a capture-mark-recapture pilot study and implanted TLDs (n = 10, CHP Dosimetry) in A. flavicollis mice from contaminated and uncontaminated areas within the CEZ (Supporting Information). While external dose readings for three recaptured individuals conform closely (Table S3) to those reported in other studies (Beresford et al., 2008), a low mice recapture rate (<25%) prevented such trials in other species. Internal radiation exposure of all mice (whole-body radiocaesium burden) was estimated using γ-spectrometry (SAM 940; Berkeley Nucleonics) ( Table 1; Supporting Information).
Total radiation exposure differed somewhat between species (Table 1), although consistently, most (~78% to >96%) of the total radiation exposure in all mice is derived from external sources; that is, animals are exposed to radiation simply by living in a contaminated area Kubota et al., 2015;Onuma et al., 2020).
The average total radiation dose rates for mice captured from contaminated areas were 0.51 mGy day -1 in CEZ and 0.21 mGy day -1 in FEZ, which is one to two orders of magnitude more than exposure received by mice from uncontaminated areas (Table 1). These radiation doses are similar to those reported in other studies of rodents from the Chernobyl and Fukushima accident sites Kubota et al., 2015;Onuma et al., 2020). For context, the dose rates in A. flavicollis and A. speciosus are equivalent to about two to four chest radiography scans (~0.10-0.15 mGy) each day (Baker et al., 2017;Brenner & Hall, 2007). The implication of the radiation dosimetry data is that mice inhabiting the contaminated (CH, FH), but not uncontaminated (CL, KL, FL), areas in Ukraine and in Japan experience significant chronic radiation exposure (Table 1).

| Statistical analyses
Statistical analyses were performed using r version 3.6.1 (R Core Team, 2019). We analysed alpha-diversity by calculating ASV richness, Faith's phylogenetic diversity and Shannon index for each sample. Significant differences between groups (i.e., host species, treatment and sex) were determined using either Wilcoxon ranksum or Kruskal-Wallis tests (when more than two groups were compared), followed by a post hoc Dunn test with a Benjamini-Hochberg false discovery rate (FDR) correction for multiple comparisons in the r package dunn.test (version 1.3.5) (Dinno, 2017). We examined potential associations between continuous metadata variables (e.g., total radiation dose rate, body condition) and mouse gut microbiota alpha-diversity estimates using Spearman's rank correlations.
To examine differences in beta-diversity between treatments we calculated pairwise sample dissimilarities using four beta-diversity metrics (Bray-Curtis dissimilarity, Jaccard similarity index, unweighted and weighted UniFrac distances) in the r package phyloseq (version  (Oksanen et al., 2019). Dispersion tests using permdisp in qiime2 were used to assess whether significant treatment effects could be influenced by differences in group dispersion rather than centroids (Anderson, 2001;Bolyen et al., 2019).
To identify which ASVs significantly differed in abundance between treatments, we used the permuted mean difference tests with 10,000 permutations and discrete FDR correction at alpha 0.1, using ds-fdr (version 0.0.2) (Jiang et al., 2017). In addition, the random forest (RF) machine-learning algorithm (Breiman, 2001) was used to determine the accuracy with which samples could be assigned to one of the study sites (Ukraine vs. Japan), and host species, based on gut microbiota community composition. Models were built using the q2-sample-classifier , with 10,000 trees and a five-fold cross-validation scheme. Input samples were randomly split into training and test sets at a 4:1 ratio, and automatic feature (ASV) selection and parameter optimization steps were enabled to tune the model. ASVs that maximize predictive accuracy were identified and assigned an importance score using a cross-validated recursive feature elimination procedure .
Additionally, we re-analysed the data set (e.g., EBI accession no. ERP104266) from the study that examined the gut microbiota of bank voles (Myodes glareolus) exposed to radionuclide contamination in Chernobyl . We directly compared the generality of radiation-associated changes in the bank vole gut microbiota (e.g., differences in alpha-and beta-diversity between treatments) to the patterns observed in A. flavicollis and A. sylvaticus; thus, we examined three species that occur in sympatry in habitats affected by the Chernobyl nuclear accident. All details about data processing and results provided in the Supporting Information.

| Wild mouse gut microbiota composition
We identified 3,731 ASVs from 15 bacterial phyla in the gut microbiota of four species of mice (n = 272) inhabiting areas surrounding the Chernobyl and Fukushima nuclear accident sites (Table S4; Figure 1).

| Interspecific variation in wild mouse gut microbiota
When compared across species, irrespective of radionuclide contamination, Apodemus mice harbour distinct gut microbiota, both between Europe and Japan and in sympatry.  samples (Figure 2b; Table S5). Consistent with this observation, RF supervised learning models could correctly classify all samples from Ukraine and Japan (baseline accuracy 0.53, accuracy ratio 1.9) ( Figure S2a). We also trained the RF classifier to identify host species; RF models assigned samples to respective mouse species with 90% accuracy (baseline 0.42, ratio 2.13) (see Table S6 for the top 20 most important ASVs). Notably, the predictive accuracy of this RF classifier decreased due to misclassifications only between samples from A. flavicollis and A. sylvaticus hosts ( Figure S2b,c).  (Figure 2b; see Table S5 for PERMANOVA results testing for species differences in each country). any of the mouse species except A. speciosus (Figure 3). That the gut microbiota diversity did not differ significantly between treatments was also consistent in bank voles (χ 2 = 2.66, df = 2, p > .264, Kruskal-Wallis test for ASV richness; see Figure S3a). In A. speciosus, animals from contaminated areas (FH) had significantly (W = 382, p < .004,

| The effects of radiation on mouse gut microbial community diversity
Wilcoxon rank-sum test for ASV richness) higher alpha diversity compared with animals from uncontaminated (FL) areas (Figure 3; Figure S4). The significant positive trend seen in A. speciosus, but not the other host species, was also retained when total radiation dose rate for individual mice was used as a continuous explanatory variable (rho =0.26, p < .020, Spearman's correlation for ASV richness; Figure S5, Table S7). These patterns in gut microbiota diversity were consistent across all three alpha diversity measures (e.g., ASV  Table S7). This is because, in A. argenteus, host body condition exhibited a weak positive association (rho =0.28, p < .042, Spearman's correlation for ASV richness; but nonsignificant for Shannon index, see Table S7) with gut microbiota alpha diversity ( Figure S7); this latter association was not related to radiation exposure as body condition was not correlated with total radiation dose in any of the sampled mice ( Figure S8).

| Radiation-associated differences in mouse gut microbiota composition
Radiation exposure was associated with altered gut microbiota composition in all species of mouse except A. argenteus (p < .05, permutation test with discrete FDR (dsFDR) correction; Figure 4a). We found that among the 305 bacterial ASVs (from taxa with >0.5% mean relative abundance) that were differentially abundant be- It is important to note that members of the S24-7, Ruminococcacae, Lachnospiraceae and other Clostridiales were both negatively and positively affected (most taxa in ~1:1 ratio, but see 46 vs. 84 S24-7 ASVs, respectively) by radiation exposure (Figure 4a;  Figure S10). This implies that, contrary to our prediction, the F:B ratio is not a general biomarker of radiation exposure, because changes in the gut microbiota of mice inhabiting radioactively contaminated areas operate on the level of distinct ASVs. Hence, effects of radiation on mouse gut microbiota composition appear to be taxonomically widespread across many bacterial families, with members of each family both positively and negatively associated with radiation exposure.

| Radiation-associated differences in mouse gut microbiota structure
Variation in the mouse gut microbiota structure reinforces the patterns of community composition described above. The gut microbiota samples from contaminated with radionuclides and uncontaminated areas were clearly separated in the ordination space (Figure 4b, Bray-Curtis dissimilarity; see also ; this pattern was also consistent in bank voles (see Figure   S3b and Supporting Information for more details). Accordingly, gut microbiota significantly differed among radionuclide contamination treatments (p < .01, PERMANOVA based on Bray-Curtis dissimilarity), but not between sexes or due to variation in body mass, head width (but see A. speciosus) or maturity (Table S5, and Supporting Information for details on bank voles). While exposure to radionuclide contamination predicted gut microbiota structure in mice, within each treatment the total radiation dose rates explained little (1%-4%) additional variation in the gut microbiota profiles among sampled mice. These patterns were largely consistent across all four beta diversity metrics used (see Table S5). A no- contributions to the total variation in the gut microbiota profiles ( Figure 4b; Table S5). We found no significant (p > .05, permdisp) difference in dispersion between treatment groups, and thus the distinct gut microbiota profiles in A. flavicollis, A. sylvaticus and A. speciosus mice between treatments resulted from differences in group centroids. Hence, exposure to radionuclide contamination alters the gut microbiota structure in three out of the four species of Apodemus mice.

| DISCUSS ION
Species identity is thought to dominate over environment in shaping wild rodent gut microbiota (Knowles et al., 2019), but it remains unknown whether host gut microbiota responses to similar anthropogenic habitat characteristics are species-specific or comparable across host species. To obtain a general assessment of the effects of exposure to environmental contamination on gut microbiota, we F I G U R E 4 Radiation-associated differences in mouse gut microbiota community composition and structure. (a) The 305 bacterial ASVs that were differentially abundant among radionuclide contamination "treatments" in mice inhabiting areas surrounding either the Chernobyl (Apodemus flavicollis, A. sylvaticus) or Fukushima (A. speciosus) nuclear accident sites. Data for three mouse species are shown as none of the ASVs were differentially abundant between treatments in A. argenteus. ASVs below the dashed line (t-statistic from the permutation test with dsFDR correction) are more abundant in samples from uncontaminated (KL, CL, FL) areas, while those above are more abundant in samples from contaminated (CH, FH) areas. ASVs are grouped by family, coloured by phyla and sized by mean taxon relative abundance.
(b) Differences in the mouse gut microbiota beta diversity associated with exposure to radionuclide contamination. PCoA plot based on the Bray-Curtis dissimilarity in mouse gut microbiota profiles among areas contaminated with radionuclides (CH, FH) or uncontaminated (KL, CL, FL) surrounding either the Chernobyl (A. flavicollis, A. sylvaticus) or Fukushima (A. speciosus, A. argenteus) nuclear accident sites. Each point represents a single sample, while shape indicates host species, coloured according to radionuclide contamination "treatments" sampled two pairs of mouse species (Apodemus flavicollis, A. sylvaticus, A. speciosus, A. argenteus) that occur in sympatry in habitats affected by radionuclides derived from the Chernobyl and Fukushima nuclear accidents. We compared the patterns of microbiota response to radionuclide contamination among these four mouse species, with the general prediction that patterns observed in the bank vole  should emerge also in mice exposed to similar environmental contamination. In agreement with our hypotheses, we found that three mouse species, like the bank vole, were characterized by (i) similar alpha diversity, irrespective of host radiation exposure levels, (ii) altered community composition in contaminated areas, and (iii) distinct gut microbiota structure when exposed to radionuclide contamination. Contrary to the pattern observed in the bank vole, however, (iv) selection for a more distinct gut microbiota in these mice does not manifest in a systematic increase in Firmicutes to Bacteroidetes (F:B) ratio. The notable lack of gut microbiota response to radiation in one mouse species from Fukushima (i.e., A. argenteus) may be due to host escape from most radiation exposure through its unique tree-dwelling lifestyle. Taken together, our data show both general and species-specific impacts on gut microbiota of murine, and also arvicoline (see Figure S3; and also , rodents living under chronic radiation exposure. Human activity typically has negative impacts on the biodiversity of macro-and microorganisms (Carthey et al., 2019;Foley et al., 2005), but the effect of environmental radionuclide contamination on microbial diversity is unclear. A negative association between radiation dose rate and alpha diversity has been observed in soil, bird feather and earthworm microbiota (Czirják et al., 2010;Newbold et al., 2019;Romanovskaia et al., 1998;Ruiz-González et al., 2016), but not in some communities of free-living microbes isolated from areas surrounding the Chernobyl and Fukushima accident sites (Hoyos-Hernandez et al., 2019;Ragon et al., 2011;Theodorakopoulos et al., 2017). We found no negative impact of radionuclide contamination on the diversity of mouse gut microbiota ( This result suggests that low-dose (~2 mGy day -1 ) radiation exposure experienced by wildlife within the CEZ and FEZ Kubota et al., 2015) does not reduce host-associated microbial diversity. This is perhaps not surprising as many bacteria can withstand chronic radiation exposure (Ragon et al., 2011;Shuryak, 2019). Moreover, a "rewilding" of the Chernobyl and Fukushima landscapes that followed human abandonment of these sites (Deryabina et al., 2015;Lyons et al., 2020;Perino et al., 2019;Webster et al., 2016) might have widened the niche space available to microbes (Gellie et al., 2017). This in turn could even have concomitant positive effects on the diversity of host-associated microbiota (Hanski et al., 2012). As such, a similar diversity of mice gut microbiota irrespective of radiation exposure levels reinforces patterns found in the bank vole gut microbiota . The implication is that exposure to environmental radionuclides does not reduce gut microbiota diversity in wild rodents.
Host diet, habitat niche segregation features and/or host phylogeny can potentially explain differences and similarities in the gut microbiota of two pairs of mouse species (Figure 2; Figure S2; see also Figure S3 for differences between mice and voles) (Perofsky et al., 2019;Youngblut et al., 2019). Despite these effects of host species, our data show a general impact of exposure to radionuclides on structuring the gut microbiota of three out of four species of Apodemus mice. The sympatric species pairs share the same environments, and thus should experience comparable radiation exposure.
This assumption may well hold true except when there is an apparent "radiation escape" due to some features of host ecology that modify radiation exposure intensity Shuryak, 2020).
That the A. argenteus gut microbiota was not structured by radionuclide contamination is striking and suggests the microbiota species escape radiation. The most plausible explanation here is that the arboreal lifestyle of A. argenteus (Oka, 1992) lessens the average external radiation exposure for this species compared with the sympatric, soil-dwelling A. speciosus (Shuryak, 2020;Stark et al., 2017). Given that disruptions of the gut microbiota in early life have been shown to have persistent lifelong effects on microbiota and host performance (Knutie et al., 2017), such radiation escape can be particularly important during early life as A. argenteus raises young in tree cavities (Oka, 1992). Indeed, radiation dose rates even at 1 m height can be up to two-fold lower than at ground level (Kubota et al., 2015) because most radionuclides derived from the Fukushima accident were deposited in the topsoil and leaf litter (Hashimoto et al., 2013).
There are no estimates of the fraction of time this species spends on trees vs. the ground, yet given that we captured this species using live traps placed on the ground, they at least sometimes descend to the ground when foraging. As both A. argenteus and A. speciosus forage on contaminated dietary material, they exhibit similar radiation exposures from internal sources (Table 1). However, their internal dose typically accounts for less than 10% of the total absorbed radiation dose (Table 1) (Kubota et al., 2015). Hence, rather than any specific radioresistance (Shuryak, 2020), it is most likely that the tree-dwelling lifestyle of A. argenteus helps them to escape most external radiation exposure, resulting in a lack of association between their gut microbiota and soil radionuclide contamination (Figure 4). It is important to note that our estimates of external radiation dose for A. argenteus are extrapolated from experimental TLD data on soildwelling rodents (see Table S3; and Beresford et al., 2008;Chesser et al., 2000;Lavrinienko et al., 2020). Data from TLDs fitted on A. argenteus are required to quantify the effect of species-specific variation in space use on radiation exposure Shuryak, 2020). The finding of an apparent importance of host lifestyle offers an interesting counterpoint for future analyses into the effects of radiation or any other toxic exposure on a host and its associated microbiota.
Numerous studies have sought a reliable biomarker of exposure to radiation (Lourenço et al., 2016;Zhang & Steen, 2018). The Firmicutes to Bacteroidetes (F:B) ratio in the bank vole gut microbiota has a strong positive association with the level of radiation exposure . Although the Firmicutes and Bacteroidetes drive compositional changes in response to radiation exposure also in Apodemus mice (Figure 4a), the lack of a systematic increase in F:B ratio in relation to radiation exposure in all species ( Figure S10) indicates that the F:B ratio is not a general biomarker of exposure to radiation. This is perhaps not unexpected given the diversity of metabolic functions within the bacterial phylogeny (Lagkouvardos et al., 2019) and the probable species-or strainspecific responses to any changes in host physiology and/or diet that accompany exposure to radionuclides (Kesäniemi et al., 2019;. Indeed, exposure to radionuclides elicits both an increase and a decrease in the proportion of ASVs assigned to higher taxa, such as the families S24-7, Ruminococcacae and Lachnospiraceae (Figure 4a). Using a simple metric of microbiota community composition, such as F:B ratio, to define host condition thus masks underlying heterogeneity in bacterial function (Ridenhour et al., 2017). As such, this finding suggests the overall limited utility of such broad taxonomic microbiota measures as biomarkers in ecotoxicology (Tu et al., 2020).
Gut microbiota can influence host fitness (Alberdi et al., 2016) and it is an important next step to identify whether changes in the gut microbiota of mice exposed to radionuclide contamination impact host health. The bacterial clade that emerges from this study as particularly responsive to radiation exposure, the S24-7 family (other proposed names: Homeothermaceae or Muribaculaceae) (Figure 4a), represents an interesting target for such work. This could pose some challenges, as members of the S24-7 are difficult to culture and only recently have their genomes been characterized (Lagkouvardos et al., 2019;Ormerod et al., 2016). Genome-resolved metagenomic analysis of the S24-7 indicate that it is a speciose (~700 predicted species) and metabolically diverse family, whose members may utilize complex carbohydrates such as starch, hemicellulose and pectin, and also host-derived glycans (Lagkouvardos et al., 2019). Such functional capacity of the S24-7 is consistent with Apodemus mice diets and can be beneficial during seasonal fluctuations in resource (seeds) availability (Butet & Delettre, 2011). Metabolic diversity within the S24-7 family probably accounts for the lack of a unified response to radiation (Figure 4a), and similar differential responses of S24-7 to external stimuli have been reported in studies on captive rodents (van Leeuwen et al., 2020;Ridenhour et al., 2017;Smith et al., 2019). Future studies would need to combine manipulative field experiments with metagenomic analyses to quantify gut microbiota functional profiles and link compositional changes to potential changes in bacterial function in the gut microbiota of mice exposed to radiation.
Extensive sampling of multiple host species inhabiting both the Chernobyl and Fukushima accident sites provides a rare opportunity for a synthesis of how radiation exposure impacts rodent gut microbiota. Here we demonstrate that the effects of radiation exposure upon the gut microbiota are generally consistent across host genera (Apodemus and Myodes; see Figure S3; and also . Such general effects of radiation could result from (i) indirect impacts through changes in wider environment and/or available diet, or (ii) a direct impact on animal host and/or its microbiota.
Hypothesized indirect impacts such as dietary shift cannot fully explain the patterns observed in this study. Dietary shift is unlikely to occur concurrently in murine and arvicoline rodents that inherently differ in dietary ecology (Butet & Delettre, 2011;. Similarly, the likelihood that changes in mouse gut microbiota reflect comparable ecological impacts of radiation on generally distinct environments at both the Chernobyl and Fukushima study sites is rather low. Although radiation exposure may act directly as an environmental filter on certain individual taxa, we find no difference in gut microbiota richness between radionuclide contamination treatments in three of the four mouse species studied, and the bank vole (Figure 3; Figures S4 and S5; for voles see Figure S3a; and also . Hence, it is conceivable that exposure to radiation is associated with selection targeting either features of host physiology (metabolism, immunity; see Kesäniemi et al., 2019), which in turn affect microbiome assembly or specific gut microbiota profiles that provide the host with essential services. Whether changes in the gut microbiota of bank voles and mice living under chronic exposure to radiation facilitate or impede their animal hosts' overall responses to irradiation under natural conditions remains to be quantified.

ACK N OWLED G EM ENTS
We are grateful to Anders P. Møller, Gennadi Milinevsky, Igor

Chizhevsky, Maksym Ivanenko, Marian Moriguchi and Hidekatsu
Ouchi for help in organizing fieldwork. Sofia Sanchez provided assistance during material collection. We would like to thank the Finnish The funders had no role in study design, data collection and analysis, decision to publish or preparation of the manuscript.

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

DATA AVA I L A B I L I T Y S TAT E M E N T
The raw sequence data for this study have been deposited in the European Nucleotide Archive (ENA) at EMBL-EBI under accession no. PRJEB44039 (https://www.ebi.ac.uk/ena/brows er/view/PRJEB 44039). The data sets (qiime2 files), metadata and code for analyses described in this work are available at: https://github.com/alavr inien ko/hot-mice, with the final version also archived in Zenodo at: http:// doi.org/10.5281/zenodo.4670304 (Lavrinienko, 2021).