Interpretation of gut microbiota data in the ‘eye of the beholder’: A commentary and re‐evaluation of data from ‘Impacts of radiation exposure on the bacterial and fungal microbiome of small mammals in the Chernobyl Exclusion Zone’

Abstract Evidence that exposure to environmental pollutants can alter the gut microbiota composition of wildlife includes studies of rodents exposed to radionuclides. Antwis et al. (2021) used amplicon sequencing to characterise the gut microbiota of four species of rodent (Myodes glareolus, Apodemus agrarius, A. flavicollis and A. sylvaticus) inhabiting the Chernobyl Exclusion Zone (CEZ) to examine possible changes in gut bacteria (microbiota) and gut fungi (mycobiota) associated with exposure to radionuclides and whether the sample type (from caecum or faeces) affected the analysis. The conclusions derived from the analyses of gut mycobiota are based on data that represent a mixture of ingested fungi (e.g. edible macrofungi, polypores, lichens and ectomycorrhizae) and gut mycobiota (e.g. microfungi and yeasts), which mask the patterns of inter‐ and intraspecific variation in the authentic gut mycobiota. Implying that ‘faecal samples are not an accurate indicator of gut composition’ creates an unnecessary controversy about faecal sampling because the comparison of samples from the caecum and faeces confounds many other possible drivers (including different animals from different locations, sampled in different years) of variation in gut microbiota. It is relevant also that Antwis et al.'s (2021) data lack statistical power to detect an effect of exposure to radionuclides on the gut microbiota because (1) all of their samples of Apodemus mice had experienced a medium or high total absorbed dose rate and (2) they did not collect samples of bank voles (M. glareolus) from replicate contaminated and uncontaminated locations. Discussion of Antwis et al.'s (2021) analysis, especially the claims presented in the Abstract, is important to prevent controversy about the outcome of research on the biological impacts of wildlife inhabiting the CEZ.


| INTRODUC TI ON
Animal gut microbial communities provide essential services for their host, such as interacting with the host's immune system (Pickard et al., 2017;Round & Mazmanian, 2009), defending against pathogen invasion (Rosshart et al., 2019;Pickard et al., 2017) and processing dietary material to provide important metabolites (Morrison & Preston, 2016;Sonnenburg & Bäckhed, 2016). As disruption to the gut microbiota can affect the health of the host, there is much interest in identifying features of the host or its environment that can impact the gut microbiota community composition (Zaneveld et al., 2017). Infection by parasites or pathogens (Kreisinger et al., 2015;Sabey et al., 2021), the level of biodiversity or habitat disturbance (Barelli et al., 2020), changes in diet and season (Guo et al., 2021;Lavrinienko et al., 2020;Maurice et al., 2015) and exposure to pollutants (Brila et al., 2021) are associated with a change in the gut microbiota of wildlife.
Evidence that exposure to pollutants impacts the gut microbiota extends to rodents inhabiting areas contaminated by radionuclides (Lavrinienko et al., 2020;. Antwis et al. (2021) characterised the gut microbiota of four species of rodent inhabiting the Chernobyl Exclusion Zone (CEZ), a region surrounding the former nuclear power plant at Chernobyl, Ukraine, where substantial areas are contaminated by radionuclides. Samples were collected from bank voles (Myodes glareolus), where animals were caught from locations that represented a gradient of contamination, and from three species of mice (Apodemus agrarius, A. flavicollis and A. sylvaticus), where animals had experienced 'medium' (4-42 μGy/hr) or 'high' (>42 μGy/hr) absorbed dose rates. The CEZ was established to limit human exposure to radionuclides, but the wildlife inhabiting the CEZ provide the best-studied models of the biological impacts of exposure to environmental radionuclides (Beresford et al., 2016;Møller & Mousseau, 2006;Mousseau, 2021). Despite decades of research on wildlife within the CEZ, the effects of exposure to environmental radionuclides in wildlife remain a source of controversy (Beresford et al., 2016;Beresford, Horemans, et al., 2020;Kesäniemi et al., 2018;Møller & Mousseau, 2006;Mousseau, 2021). (2) creates an unnecessary controversy as it neglects to account for confounding effects of multiple drivers of variation in the gut microbiota in the comparison of samples from the caecum and faeces. We discuss statement (3) to highlight the need for greater clarity about the power of Antwis et al.'s (2021) study design to identify the effects of exposure to radionuclides on the gut microbiota of small rodents. Antwis et al. (2021) used amplicon sequencing to characterise the bacterial and fungal components of the gut microbiota in four species of rodent (bank vole M. glareolus, wood mouse Apodemus sylvaticus, striped field mouse A. agrarius and yellow-necked mouse A. flavicollis) inhabiting the CEZ, Ukraine. Samples were collected by live trapping from two areas: (1) a contaminated site in the Red Forest and adjacent area, and (2) an uncontaminated site about 10 km south-west of the Red Forest ( Figure 1). We re-analysed some of the amplicon sequence data used by Antwis et al. (2021) to examine the potential proportion of non-resident fungal sequence variants (SVs) in their data. As the data used in this article were downloaded from a public archive, we did not seek permissions for fieldwork or ethical approval for the work.

| MATERIAL S AND ME THODS
Full details about the sample data are provided in the original publication (Antwis et al., 2021). Briefly, data were obtained from GenBank (PRJNA594002) and processed in QIIME2 v.2020.6 (Bolyen et al., 2019), using CUTADAPT (https://github.com/marce lm/cutadapt) to remove primer/adaptor sequences and DADA2 (Callahan et al., 2016) to denoise the data. Taxonomy for SVs was assigned using the SKLEARN machine learning taxonomy classifier (Bokulich et al., 2018) against the UNITE v.8 (Nilsson, Larsson, et al., 2019) reference database. Because there are no reliable data that identify all species of fungi that could be ingested (either by direct consumption, or by association with other components of the diet) by bank voles and Apodemus mice in the CEZ, we used informed filtering to identify likely non-resident fungal SVs (see Lavrinienko, Scholier, et al., 2021).
SVs were assigned to the major classes of microfungi using the information at the Microfungi Collections Consortium (www.micro fungi.org/table1). We further classified fungal SVs according to guild or growth form based on assignments made by FUNGUILD v.1.2 (Nguyen et al., 2015). SVs categorised as plant pathogens, epi-and endophytes, lichens, mycorrhizae and wood saprophytes were assumed to be part of the ingested, non-resident fungal material in the gut, as were SVs with large fruiting bodies/growth forms (e.g. as gasteroid, pezizoid, tremelloid, etc.), which left the remaining microfungi and yeasts (and taxa with unclear growth forms but which were assigned as animal pathogens or SVs that lacked information about guild, e.g. because taxonomic resolution was not assigned below Phylum level) as candidate resident gut mycobiota. Thus, we make a contrast between data that can represent dietary items (principally macrofungi and lichens, plant-associated fungal pathogens, mycorrhizae or endophytes) and the remaining data as a candidate resident gut mycobiota (many microfungi and yeasts, taxa associated with animals and poorly known fungi).
Data were imported into PHYLOSEQ (McMurdie & Holmes, 2013) for analyses in R v.4.0.5 (R Core Team, 2020). Fungal data were rarefied to an even depth of 5,000 (all SV data, that resulted in a loss of 11 samples and 1,020 SVs), 3,000 (possible resident gut fungi, with a loss of 27 samples and 980 SVs) or 1,000 (possible dietary and diet-associated fungi, with a loss of 21 samples and 710 SVs) reads per sample. We calculated alpha diversity (observed number of SVs) and beta diversity (Bray-Curtis dissimilarity) in phyloseq. Variation in alpha diversity was assessed using pairwise Wilcoxon rank-sum tests with Holm correction. The ADONIS2 function in vegan (Oksanen et al., 2020) was used to examine the amount of variation in beta diversity explained by features of the data, such as host species (bank vole, wood mouse, striped field mouse and yellow-necked mouse), sampling year (2017, 2018) and total absorbed dose rate of radiation (μGy/hr). BETADISP function in vegan (Oksanen et al., 2020) was used to determine whether there were significant differences in dispersion among groups of samples.  Lavrinienko, Mappes, et al. (2018, circles), in Lavrinienko et al. (2020, triangles) and in Antwis et al. (2021, squares). The figure was created using the ggmap (https:// github.com/dkahl e/ggmap) package in R  (Lavrinienko et al.) Uncontaminated (Antwis et al.) Contaminated (Lavrinienko et al.) Contaminated (Antwis et al. ) 51.

| Composition of the community of fungi detected in rodent guts
By classifying probable ingested fungi as macrofungi, plant pathogens, endophytic species and taxa associated with decaying wood, we separated the fungal SVs in rodent gut samples into 2,608 putative gut residents and 1,566 possible non-residents. This filtering procedure indicates that about a third of the read data could be derived from ingested fungi ( Figure 2), with about 10%-15% of the reads assigned to macrofungi (e.g. Agaricomycetes).
The inclusion of all fungal SV data affects the analyses. Although interspecific differences in alpha diversity are apparent in unfiltered and filtered datasets, with bank voles having significantly fewer fungal SVs than species of Apodemus (pairwise Wilcoxon rank-sum test with Holm correction, p < 0.05 for all comparisons; Table 1 Table 2). Moreover, significant (p = 0.001) interspecific differences in the amount of dispersion were detected in the analyses based on the entire dataset and the suspected non-resident fungi, but not in the subset of possible resident gut fungi (Table 2, Figure 4).

F I G U R E 2
Proportions of fungal classes identified in the gut and faecal samples from four species of rodent, separated by their possible resident (mycobiota) or non-resident (ingested) status in the host's gastrointestinal tract

| Composition of the community of fungi detected in rodent guts
With our understanding of the processes that affect wild animal gut fungi (gut mycobiota) limited (Huseyin et al., 2017;Kong & Morris, 2017;Nilsson, Anslan, et al., 2019), it is important to carefully consider the possible sources of fungal material in samples from the animal gut (Lavrinienko, Scholier, et al., 2021). As amplicon sequencing enumerates all types of DNA (for the target region and taxon, such as a partial region of the 16S rRNA for bacteria or the ITS for fungi; Knight et al., 2018;Lavrinienko, Jernfors, et al., 2021) within a sample, the resulting SVs are derived from the authentic gut residents and any non-resident (ingested) material. A combination of (1) comparatively few fungal cells (compared with bacterial cells) in the vertebrate gut (Qin et al., 2010, Iliev et al., 2012 and (2) ingestion of fungi by many animals, for example by consumption of macrofungi or lichens (Abt & Bock, 1998;Fogel & Trappe, 1978) or intake of fungal plant pathogens, commensals/symbionts or the microfungi in fermenting or decaying material, raises the potential that amplicon sequencing-based studies of gut mycobiota will contain a substantial amount of non-resident gut fungi (Lavrinienko, Scholier, et al., 2021).

| Differences between microbiota detected in samples from faeces and the caecum
The extent to which sample type or laboratory procedures impact amplicon sequence data is an important issue for microbiota research Panek et al., 2018).  from samples (5) that differ in absorbed dose rates. As habitat, host genetics and season, etc. associate with variation in gut microbiota (Bonder et al., 2016;Lavrinienko et al., 2020;Li et al., 2019;Maurice et al., 2015;Park et al., 2019), the roughly 1% of variation in beta diversity attributed to 'sample type' could support the conclusion that sample type itself has little impact on microbiota composition in these data. Antwis et al. (2021) Antwis et al. (2021) note that associations between gut microbiota composition were not robust when the analyses were controlled for geographic distance, with sampling site explaining some variation in bacterial beta diversity. Hence, they conclude that 'any variation in microbiome composition arising from proximity to the Chernobyl Nuclear Power Plant is more likely a habitat effect than a result of radiation exposure'. Because bank voles can disperse several kilometres per year (White et al., 2012), Antwis et al.'s (2021) animals with high absorbed dose rates represent samples from one cohort of animals within a single contaminated area (the Red Forest locality) rather than from separate cohorts inhabiting the different contaminated sites within the CEZ (see Figure 1). A lack of replication of sites with similar dose rate categories is a curious design for a study of radiation effects on wildlife, especially when the focal contaminated area is the Red Forest as this location is argued to comprise poor habitat (Beresford, Scott, et al., 2020). Studies of radiation effects that do not employ a replicated study-site design confound the treatment (radiation exposure) with location and thus are somewhat destined to support the idea that exposure to radionuclides has no detrimental biological impacts: a lack of statistical effect can be interpreted that radiation exposure has little biological impact, while any apparent biological impacts can be dismissed as location-specific effects (e.g. the poor-quality habitat in the Red Forest) rather than exposure to radionuclides! Only by collecting samples from replicate contaminated and uncontaminated areas can location-specific effects be partitioned from impacts associated with radionuclide exposure Kesäniemi, Jernfors, et al., 2019;. The additional discussion points raised by Antwis et al. (2021) that 'Other studies of radiation effects in CEZ wildlife, including the microbiome studies of Lavrinienko et al. ..., also have their most contaminated sampling sites within the Red Forest...', and 'Any study that uses the Red Forest as a location for radiation effect studies on wildlife needs to consider the historical impacts of radiation and other stressors (e.g. wildfires) on this area...'

| Associations between radiation exposure and gut microbiota composition
are somewhat misleading as they neglect to consider the consistent patterns associated with radionuclide exposure in gut microbiota samples from replicate contaminated and uncontaminated locations Lavrinienko et al., 2020).
Outside this use of language, however, several features of Antwis et al.'s (2021) study design and interpretation of data warrant more discussion.
Demonstrating that the inclusion of non-resident fungal SV data impacts amplicon sequencing-based analyses of 'gut mycobiota' has important implications for studies of wildlife gut mycobiota (Lavrinienko, Scholier, et al., 2021). A consequence of not identifying the probable resident fraction of the gut mycobiota is to promote misunderstanding about the drivers of variation in wildlife gut mycobiota (e.g. proposing macrofungi as biomarkers of a gut microbial response to radiation exposure). The inclusion of all fungal SVs can alter the pattern of interspecific differences in 'gut mycobiota' due to variation in ingested fungal material among host species. As the ecologies and traits of many fungi are poorly known, it can be challenging to define the authentic gut mycobiota, especially for animals with a catholic diet like the bank vole. With information about the species of fungi eaten and/or present in dietary items, it could be possible to use informatic tools (e.g. SourceTracker, Knights et al., 2011) to identify the ingested fungi. Without such data, however, the analyses of wildlife gut mycobiota will depend on the filtering decisions; for example, one may decide to focus on the macrofungal fraction (rather than plant-associated fungi, such as endophytes, ectomycorrhizae and/or pathogens) and/or omit SVs with poor taxonomic resolution (e.g. those not assigned beyond Phylum level) (see e.g. Lavrinienko, Scholier, et al., 2021). Understanding whether laboratory protocols and sample type impact study conclusions is an important topic in microbiota research (Ingala et al., 2018;Knight et al., 2018;Videvall et al., 2018;Zhou et al., 2020). An apparent problem with interpreting some field studies on organisms inhabiting the CEZ is derived from authors overemphasising statistically significant relationships that have little explanatory power (Beresford, Scott, et al., 2020). Given that many (2) that they examined the effects of absorbed dose rates of 20 μGy/ hr and above on Apodemus mice but not the impacts of radiation exposure on gut microbiota. By contrast, with a sample of animals from contaminated and uncontaminated areas, Antwis et al.'s (2021) caecum samples from bank voles identified an association between absorbed dose rate and some taxa within the gut microbiota.
It is relevant also that the studies by  when attempting to replicate a microbiota study. Without an equivalent study design, or clear explanation of why changes to the protocol were made, important drivers(s) of apparent differences among studies could be overlooked.
A comprehensive discussion of the statements presented by Antwis et al. (2021) in their abstract is important because the outcomes of studies of the wildlife inhabiting the CEZ can be used to assess the risks of radiation exposure and formulate policy. Independent examinations of the biological impacts of radionuclide exposure are needed to form robust conclusions, but these studies are informative only when the comparison incorporates an appropriate study design: failure to do so will only cloud our understanding of the biological impacts of exposure to environmental radionuclides. For the reasons outlined above, Antwis et al. (2021) made some strong assertions that are hard to reconcile with their, or indeed others', data and study design(s). One unfortunate result of this attempt to stimulate debate is yet another controversy that does not appear to be justified.

ACK N OWLED G EM ENTS
The authors are grateful for access to computing facilities provided by CSC-FINLAND (www.csc.fi). Funding for research in the Chernobyl Exclusion Zone and Fukushima was provided by the Finnish Academy

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

AUTH O R S ' CO NTR I B UTI O N S
All authors conceived the ideas; P.C.W. analysed the data; A.L. and P.C.W. led the writing of the manuscript, with all authors making critical contributions to the drafts and giving their final approval for publication.