Microbiome diversity and zoonotic bacterial pathogen prevalence in Peromyscus mice from agricultural landscapes and synanthropic habitat

Rodents are key reservoirs of zoonotic pathogens and play an important role in disease transmission to humans. Importantly, anthropogenic land‐use change has been found to increase the abundance of rodents that thrive in human‐built environments (synanthropic rodents), particularly rodent reservoirs of zoonotic disease. Anthropogenic environments also affect the microbiome of synanthropic wildlife, influencing wildlife health and potentially introducing novel pathogens. Our objective was to examine the effect of agricultural development and synanthropic habitat on microbiome diversity and the prevalence of zoonotic bacterial pathogens in wild Peromyscus mice to better understand the role of these rodents in pathogen maintenance and transmission. We conducted 16S amplicon sequencing on faecal samples using long‐read nanopore sequencing technology to characterize the rodent microbiome. We compared microbiome diversity and composition between forest and synanthropic habitats in agricultural and undeveloped landscapes and screened for putative pathogenic bacteria. Microbiome richness, diversity, and evenness were higher in the agricultural landscape and synanthropic habitat compared to undeveloped‐forest habitat. Microbiome composition also differed significantly between agricultural and undeveloped landscapes and forest and synanthropic habitats. We detected overall low diversity and abundance of putative pathogenic bacteria, though putative pathogens were more likely to be found in mice from the agricultural landscape. Our findings show that landscape‐ and habitat‐level anthropogenic factors affect Peromyscus microbiomes and suggest that landscape‐level agricultural development may be important to predict zoonotic pathogen prevalence. Ultimately, understanding how anthropogenic land‐use change and synanthropy affect rodent microbiomes and pathogen prevalence is important to managing transmission of rodent‐borne zoonotic diseases to humans.


| INTRODUC TI ON
Rodents are an important source of zoonotic disease spillover, accounting for a greater diversity of zoonotic pathogens than any other mammalian Order (Han et al., 2016).While many factors have been proposed to contribute to this (e.g.fast-paced life history, Han et al., 2015; cyclic population fluctuations, Kallio et al., 2009), recent studies have suggested that the tendency of particular rodent species to occasionally or exclusively live in human-built environments (synanthropy) is likely a key factor (Ecke et al., 2022).
Anthropogenic land-use change, leading to habitat fragmentation and the intensification of agricultural development and urbanization, is the major driver of zoonotic pathogen spillover (Gottdenker et al., 2014).Indeed, urbanized areas have been found to have a significant, positive effect on the abundance of rodent hosts of zoonotic pathogens compared to areas of native vegetation (Mendoza et al., 2019).Shifts in rodent biodiversity in anthropogenic (i.e.human-built) environments could further increase zoonotic risk, as rodent hosts and non-host rodents show opposite responses to agricultural and urban habitats, with the abundance of host species increasing and non-host species decreasing compared to areas of minimally disturbed primary vegetation (Gibb et al., 2020).
However, spillover of zoonotic pathogens at the human-wildlife interface does not solely flow from wildlife into humans.
Synanthropic wildlife (including rodents) also show increased prevalence of human pathogens: Escherichia coli, Clostridioides difficile, Salmonella enterica in Norway rats in New York City, New York (Firth et al., 2014); antimicrobial-resistant E.coli in racoons in Chicago, Illinois (Worsley-Tonks et al., 2021); and Salmonella in urbanized white ibis in southern Florida (Hernandez et al., 2016), representing both a concern for wildlife health and a potential source for spillback into human populations.As such, while the relationships between land-use change, wildlife, humans, and zoonotic pathogen prevalence are still being explored, synanthropic wildlife represent both important reservoirs for zoonotic pathogens and drivers of pathogen maintenance through spillover and spillback in anthropogenic environments (Hassell et al., 2017).
Synanthropy has also been shown to impact the gut microbiome of wildlife.The gut microbiome plays a role in host health (Marchesi et al., 2016) and immune function (Schluter et al., 2020), and disruption of the normal microbiome, or dysbiosis, has been linked to various health conditions in wildlife, livestock, and domestic animals (Funosas et al., 2021;Monteiro & Faciola, 2020;Suchodolski, 2022).
Wildlife living in close proximity to humans commonly experience changes to the composition of their microbiome compared to counterparts in native habitat (Anders et al., 2022;Berlow et al., 2021).
However, access to novel food resources or unique assemblages of wildlife gathering in anthropogenic environments can also increase microbiome diversity in synanthropic wildlife compared to native habitat (Dillard et al., 2022;Phillips et al., 2018), suggesting that the responses of wildlife microbiomes to anthropogenic environments could be species-or context-specific.
The stressors present in anthropogenic environments (e.g.noise and light pollution, pesticides, crowding) can also act to destabilize the microbiome, resulting in more heterogeneous community composition compared to wildlife in native habitat (Zaneveld et al., 2017).Changes in microbiome composition from a balanced, healthy state towards dysbiosis can allow for the establishment of pathogenic bacteria and facilitate infection (Stevens et al., 2021).
Anthropogenic environments can also be a direct source of increased parasite prevalence and diversity in wildlife, via increased contact among animals in urbanized environments and increased site fidelity (Murray et al., 2019).Together, these factors provide a link between anthropogenic-induced microbiome shifts and increased pathogen prevalence in synanthropic wildlife (Murray et al., 2020).
Here, we compare microbiome diversity and the prevalence of zoonotic bacterial pathogens in Peromyscus mice from forest and synanthropic habitats in both agricultural developed ('agricultural') and undeveloped landscapes in Minnesota, USA (Figure 1).Our research questions were twofold: (1) How do agricultural landscapes and synanthropic habitats affect the microbiome diversity of Peromyscus mice? and (2) Is there a greater prevalence and diversity of zoonotic bacterial pathogens in agricultural landscapes and synanthropic habitats compared to undeveloped landscapes and forest habitats?
We expected the microbiome of Peromyscus to be altered by the surrounding landscape and specific habitat as landscapes and habitat influence the availability of food resources and exposure to humans and their pathogens.We predicted that microbiome diversity would be lower in the agricultural landscape and synanthropic habitats compared to the undeveloped landscape and forest habitats due to lower diversity of food resources (Amato et al., 2013;Fuirst et al., 2018).We predicted that the agricultural landscape would have a higher prevalence and diversity of pathogenic bacteria since the area is dominated by crop fields and human habitation and thus increased exposure to manure as fertilizer (Frentrup et al., 2021), wastewater and runoff (Graczyk et al., 2009;Ramey & Ahlstrom, 2019), and trash (Sugden et al., 2020), whereas we predicted that the undeveloped landscape would have lower pathogen prevalence because the surrounding area is largely forested with little anthropogenic development.Characterizing rodent microbiomes across landscape and habitat types is important for quantifying the 16S amplicon sequencing, microbiome, nanopore sequencing, Peromyscus, synanthropy, zoonoses risk of rodent-borne zoonotic pathogen spillover in anthropogenic environments and understanding how microbiome shifts associated with synanthropy may influence pathogen prevalence.

| Study sites
Three major North American biomes intersect in Minnesota: the eastern deciduous forest, northern coniferous forest, and western prairie, providing diverse habitats and biological communities of hosts and pathogens.With respect to land-use, the state is dominated by agricultural cropland and forest with interspersed developed areas ranging from dense metropolitan areas to small, rural communities.Together, the biological and anthropogenic factors create a heterogeneous landscape of natural areas mixed with agricultural and urban developed landscapes where synanthropic rodents have many opportunities to overlap with humans.We focus our study on mice of the genus Peromyscus (i.e.Peromyscus leucopus and Peromyscus maniculatis) which are highly adaptable generalists that are common throughout Minnesota and can thrive in agricultural and urban settings as well as forests and grasslands.Importantly, Peromyscus mice are known reservoirs of zoonotic and foodborne pathogens (e.g.Borrelia, Campylobacter spp., E. coli, Giardia spp., hantavirus; Jahan et al., 2021).
For our study, we focused on two landscape types: native, contiguous forest with little permanent human habitation or agriculture (hereafter 'undeveloped landscape') and a mosaic of fragmented forest interspersed with crop fields and low-density housing (hereafter 'agricultural landscape').the state park is frequented by hikers and visitors, the surrounding landscape is contiguous forest with no agricultural development and very sparse permanent human habitation (Figure 1a).Cedar Creek is located in central Minnesota in the eastern deciduous forest and oak savanna biome.The landscape surrounding the reserve is dominated by agricultural development (e.g.pasture/ hay, cultivated crops), woody and herbaceous wetlands, and lowmedium intensity housing communities (Figure 1b).

| Rodent trapping
Two consecutive nights of rodent trapping were conducted at each study site (a 'trapping session') using 100 Sherman livecapture traps baited with oats.Traps were opened at dusk and checked at dawn the following morning.Traps were closed during the day between trap nights at a single site and were reopened at dusk for the second night.Captured Peromyscus mice were sampled and then released at the point of capture.Due to the difficulty in distinguishing P. leucopus and P. maniculatus -two species found across our study landscapes in Minnesota -based solely on morphologic features, we did not attempt to identify captured Peromyscus mice to the species level.Captured non-target (i.e.non-Peromyscus) species were released immediately and were not sampled.Longitudinal trapping was conducted at the agricultural landscape sites.Each site was sampled three times throughout the summer (June, July, and August 2019) with 3-4 weeks between trapping sessions.Captured Peromyscus mice were marked with a uniquely numbered metal ear tag to identify mice that had been previously sampled if they were recaptured during a subsequent trapping session.Due to financial and logistical limitations, only one trapping session was conducted at the undeveloped landscape sites.This single trapping session was conducted in July to align with a time of high rodent population density and provide a comparable sample size to the agricultural landscape.For each captured Peromyscus, a faecal sample was collected and body mass, sex, and reproductive status were recorded (reproductive individuals identified by the presence of scrotal testes for males or any of the following traits for females: perforate vagina, enlarged nipples, palpable embryos).Individuals captured a second time within a trapping session were not resampled and were promptly released at the point of capture.

| DNA extraction
Faecal samples of up to 250 mg were stored without buffer or ethanol and frozen at −80°C immediately after sampling until DNA extraction.DNA was extracted using a QIAamp PowerFecal Pro kit (Qiagen, Hilden, Germany) following manufacturer instructions both manually and using a QIAcube robotic workstation (Qiagen, Hilden, Germany).DNA extracts were quantified using a Qubit 4 fluorometer (Thermofisher Scientific, Waltham, MA, USA) using the Qubit dsDNA BR Assay Kit (Thermofisher Scientific, Waltham, MA, USA) following the manufacturer's instructions.Samples with low DNA yield (<24 ng/μL, n = 16) were excluded from downstream analysis.

| Library prep and nanopore sequencing
The Oxford Nanopore Technologies (ONT) Rapid 16S Barcoding Kit (SQK-16S024 [utilizing 'Kit 9' chemistry], Oxford, UK) was used to prepare barcoded amplicon libraries for sequencing, largely following the manufacturer's protocol (methods described in detail in Jahan et al., 2021).In brief, all faecal DNA extracts were diluted in nuclease-free water to a concentration of 10-30 ng/μL.The full-length bacterial 16S rRNA gene region (1.6 kb) was amplified via PCR using specific primers and between 20 and 40 ng of DNA template, a long-range master mix (LongAmp Hot Start Taq, 2×; New England Biolabs, Ipswich, MA, USA), and sample-specific barcode identifier.PCR products were purified and prepared for sequencing through a series of magnetic bead wash steps (AMPure XP beads; Beckman Coulter Life Sciences, Indianapolis, IN, USA).Barcoded samples were pooled (pools of 21-24 samples) with ONT rapid sequencing adapter mixture into a final library for sequencing.Seven pooled libraries were sequenced on FLO-MIN106 MinION flow cells utilizing R9 sequencing chemistry (Oxford Nanopore Technologies, Oxford, UK), run for 24 h using the ONT MinKNOW GUI (v4.3.20;Oxford Nanopore Technologies, Oxford, UK).
Taxonomic abundance profiles were generated using Emu, an expectation-maximization algorithm designed specifically to account for the increased read length and error rate often associated with long-read data such as ONT-generated sequences (v3.4.4;Curry et al., 2022).Compared to conventional taxonomic identification algorithms, Emu is able to reduce the false positive rate of identification and accurately identify long reads to species level (Curry et al., 2022).Reads were mapped using the Emu default database settings: a combination of rrnDB v5.6 (Stoddard et al., 2015) and NCBI (National Center for Biotechnology Information) 16S RefSeq downloaded on 17 September 2020 Leary et al., 2016).The output of Emu is an estimated abundance (read count) of each identified species in a given sample.
Because read counts are estimated based on likelihood probabilities, outputted values are not necessarily integer counts.Values were rounded to the nearest integer for analysis.

| Data analysis
In the agricultural landscape where rodent sampling was conducted across three months, 18 individuals were captured and sampled in multiple months (20 samples total).To control for non-independence between repeat samples of the same individuals, only one sample per mouse (n = 140) was included in all the statistical analyses.We chose to include only the July sample for all recaptured mice to avoid introducing variation based on sampling month (all recaptured animals were sampled in July, but not in June or August) and to better align with the undeveloped landscape sampling (which was only conducted in July).
To enable direct comparisons between samples of heterogeneous sequencing depth, rarefaction was performed prior to all downstream analyses.All samples (n = 140) were rarefied to the number of reads of the least abundant sample (n = 74,517) using the 'vegan ' R package (v2.6.4;Oksanen et al., 2022).Specific details are outlined below.The full faecal microbiome was analysed at the sample level using measures of alpha and beta diversity (to quantify within-sample and between-sample bacterial diversity, respectively).
Alpha diversity indices included species richness (count of observed species), Shannon-Weiner diversity, Simpson diversity, and species evenness (Shannon diversity/ln(richness)). Shannon diversity and Simpson diversity make different assumptions about species evenness and how it contributes to diversity: Shannon diversity assumes all species are present and are randomly sampled while Simpson diversity gives more weight to common species.Calculating both indices can suggest how common or rare species may affect diversity estimates for different populations.Beta diversity was quantified using the Bray-Curtis dissimilarity index to compare bacterial microbiome community composition at the species level between all pairs of samples.As a subset of the full faecal microbiome, the presence of pathogenic bacteria (foodborne and zoonotic pathogens of concern for human, livestock, and domestic animal health) was quantified at the sample level and then analysed by landscape-habitat pairing.
For the alpha diversity analysis, bacterial read count data for all samples were rarefied and alpha diversity indices (richness, Shannon, Simpson, and evenness) were calculated at the sample level for each iteration of the rarefied count data using the 'vegan' R package.This was repeated for 100 iterations, and the average values of the alpha diversity indices for each sample were recorded.We investigated whether alpha diversity was affected by landscape or habitat type by developing a linear regression model for species richness and Shannon diversity and a beta regression model for Simpson diversity and species evenness.In all models, the response variable was the alpha diversity index and the explanatory variables were landscape type (agricultural or undeveloped), habitat type (forest or synanthropic), the interaction of landscape and habitat type, mouse sex, reproductive status (reproductive or non-reproductive), body mass, and sampling month (June, July, or August).Alpha diversity values were compared between landscape-habitat pairings using a Kruskal-Wallis test with a post hoc Dunn test with a Holm correction for multiple comparisons.
Beta diversity was visualized using non-metric multidimensional scaling (NMDS) ordination performed on the rarefied data (rarefied to 74,517 reads, n = 100 iterations) using the Bray-Curtis dissimilarity index in the 'vegan' R package.NMDS was first performed with two dimensions (k), and the k value was iteratively increased until the stress value was below 0.1.Nonparametric statistical analyses were performed on the rarefied distance matrices using the 'adonis2', 'anosim', 'betadisper', and 'permutest' functions also in the 'vegan' R package.Differential abundances of bacterial species were calculated and compared between landscape-habitat pairings using multigroup analysis of compositions of microbiomes with bias correction For the analysis of putative pathogenic bacteria, the Emuestimated read counts per mouse were rarefied and averaged across 100 iterations.These mean counts of each bacterial species per mouse were then used for downstream analysis.A list of 209 putative pathogenic bacteria species was generated using the PHI-base pathogen database (Urban et al., 2020;accessed S1).The species-level read count abundance data from the sequenced samples were filtered for reads assigned to the pathogen species on this list, and the read count per pathogen species was thresholded to at least 50 reads.Differences in putative pathogen species richness per mouse between landscape-habitat pairings were assessed by Kruskal-Wallis test.The effect of landscape, habitat, and their interaction on the probability of detecting putative pathogen reads in captured mice was examined using a generalized linear model (binomial family, logit link).The response variable was a binary indicating whether or not putative pathogen reads (≥50 reads per pathogen species) were detected for each sampled mouse.The patterns of putative pathogen read count per mouse, grouped by landscape-habitat pairing, were visualized using a heatmap.

| Rodent samples
We collected 176 faecal samples representing 153 unique Peromyscus mice.After DNA extraction, 16 samples with low DNA yields (<24 ng/ μL) were excluded from downstream analysis.We sequenced 140 faecal samples, each representing a unique mouse.In the agricultural landscape, faecal samples from 40 and 29 unique mice were collected in forest and synanthropic habitats, respectively, across three months of rodent trapping (Figure 1c).In the undeveloped landscape, faecal samples from 31 and 40 unique mice were collected from forest and synanthropic habitats, respectively, in July (Figure 1d).

| Nanopore sequencing summary
After quality filtering, over 29.2 million high-quality reads were retained (mean Q score 12.8 ± 0.30 SD; Q score of 12.8 corresponds to base call accuracy of 94.75%).The mean number of reads per sample was 208,873.1, though the number of reads per sample was highly variable (standard deviation: 76,322.1;range: 74,517-517,058 reads; Table 1).
The Emu algorithm identified 1212 unique bacterial species across the 140 faecal samples.The mean number of species per sample was 212 (standard deviation: ±57.0; range: 82-367).Rarefaction curves were plotted for all sequenced samples (n = 140).The asymptotic nature of these curves suggest that reasonable sequencing depth was achieved for all samples and all bacterial species observed were represented at the minimum sample size (Figure S1).

Table S2).
Mean observed species richness (Holm-corrected Dunn test, p < .0001),Shannon diversity (Holm-corrected Dunn test, p < .02),and species evenness (Holm-corrected Dunn test, p < .05)were lower in the undeveloped-forest habitat compared to all other landscape-habitat pairings (Figure 2; Table S3).Mean Simpson diversity was lower in the undeveloped-forest habitat compared to the agricultural-forest and agricultural-synanthropic habitats (Holmcorrected Dunn test, p < .005; Figure 2c).However, contrary to our hypotheses, there was no difference in species richness, diversity, or evenness between forest and synanthropic habitats in the agricultural landscape or between agricultural-synanthropic and undeveloped-synanthropic habitats.
Bacterial microbiome community composition at the species level was compared between all pairs of samples using the Bray- four landscape-habitat pairings suggested that the between-group dissimilarity in microbiome community composition was significantly greater than the within-group dissimilarity (p = .001).

Curtis
An NMDS ordination plot calculated based on Bray-Curtis dissimilarity indices showed a high degree of overlap between samples from the four landscape-habitat pairings (Figure 4; see Figure S2 for comparison of NMDS output across k-values).Samples from the agricultural-synanthropic and undeveloped-forest habitats showed the greatest dissimilarity, while samples from agricultural-forest and undeveloped-synanthropic habitats were more similar.The variability among samples was high, but an analysis of multivariate homogeneity of group dispersion ('betadisper' and 'permutest' functions, 'vegan' R package) by landscape-habitat pairing showed no significant difference in variance between the groups (permutation test, p = .96),indicating that the differences in community composition were not only due to differences in sample variance.
A nonparametric PERMANOVA analysis was used to test the effects of landscape, habitat type, mouse sex, reproductive status, body mass, and sampling month on differences in microbiome community composition using the 'adonis2' function in the 'vegan' R package with the by = 'margin' option to determine the marginal effect of each parameter.There was a small but significant effect of landscape and habitat, suggesting that the microbiome of sampled mice was different between agricultural and undeveloped landscapes and between forest and synanthropic habitats (PERMANOVA, R 2 Landscape = .06,R 2 Habitat = .04,both p = .001;Table S4).Mouse reproductive status and body mass also had small, but significant effects (both p < .05).However, much of the variance in microbiome community composition was not explained by the modelled parameters (residual R 2 = .85).
We identified bacterial species associated with landscape-habitat pairing using multigroup differential abundance testing with bias correction (ANCOM-BC2).The agricultural-forest, agriculturalsynanthropic, and undeveloped-synanthropic habitats were compared to the undeveloped-forest habitat in pairwise fashion (Figure S3).In total, 38 species were found to be globally significant, though none of these were putative pathogens.Most of the differentially abundant species had a higher bias-corrected abundance in the more developed habitats compared to the undeveloped-forest habitat.

| Putative pathogen detection
Of the 209 putative pathogenic bacteria species screened for, 13 were identified in sampled mice (read count ≥50) and many of these are considered opportunistic pathogens in humans (Figure 5).At the population level, nine putative pathogen species were detected in the agricultural-forest habitat, six in the agricultural-synanthropic and undeveloped-synanthropic, and five in the undeveloped-forest habitat.At the individual level, putative pathogen species richness was low (0-2 species per mouse) and did not differ between landscape-habitat pairings (Kruskal-Wallis  2 (3) = 3.09, p = .38).
The probability of detecting reads of a putative bacterial pathogen was greater in mice from the agricultural landscape compared to those from the undeveloped landscape (Binomial GLM: Odds Ratio Landscape type = 4.25, p = .013,[95% CI: 1.44-14.61];Table S5).S5).

| DISCUSS ION
Our objective was to examine the effects of agricultural landscapes and synanthropic habitats on microbiome diversity and the prevalence of zoonotic bacterial pathogens in Peromyscus mice in Minnesota.We found that landscape-habitat pairing affected microbiome richness and diversity, but species evenness was only affected by landscape.Overall, mean alpha diversity (richness, Shannon and Simpson diversity, evenness) was higher in the agricultural landscape and synanthropic habitat compared to undeveloped-forest habitat.
Species-level microbiome community composition was significantly different between landscapes (agricultural vs. undeveloped) and habitat types (forest vs. synanthropic), but microbiome variation was similar across all four landscape-habitat pairings.We detected reads for putative pathogenic bacteria across the four habitats, though the abundance and diversity of putative pathogens was generally low.
However, the prevalence of putative pathogens was higher in mice from the agricultural landscape than in the undeveloped landscape.
Microbiome diversity and richness (alpha diversity) were impacted by landscape-habitat pairing.We found higher bacterial species richness, Shannon and Simpson diversity, and species evenness in the habitats with more anthropogenic development: the agricultural landscape and undeveloped-synanthropic habitat, compared to the undeveloped-forest habitat.Increased microbiome diversity in wildlife from anthropogenic versus native environments has been documented in other wildlife systems, aligning with our findings (birds, Berlow et al., 2021;mesocarnivores, Sugden et al., 2020).Nonetheless, this finding is surprising as it is contrary to our hypotheses and the common assumption that alpha diversity is higher in more natural environments than anthropogenic ones (Diaz et al., 2023;Fackelmann et al., 2021;Lobato-Bailón et al., 2023).As such, the directionality of alpha diversity shifts in the microbiome of wildlife in anthropogenic versus native environments is likely affected by multiple species-and system-specific factors, and these patterns warrant further investigation to disentangle.
Microbiome community composition (beta diversity) varied between mice from the undeveloped and agricultural landscapes and between forest and synanthropic habitats.Shifts in microbiome composition are commonly found in wildlife species when comparing individuals in native and anthropogenic environments (Diaz et al., 2023;Murray et al., 2020).These shifts could be attributed to dietary shifts based on habitat type and food availability, particularly in anthropogenic environments (Anders et al., 2022).Microbiome shifts in anthropogenic environments can also signal plasticity in the gut microbiome that may be allowing individuals to adapt to novel environments and new food sources (Littleford-Colquhoun et al., 2019).This could be particularly important for synanthropic generalist species such as Peromyscus to continue to thrive in humanbuilt environments.In future studies, stable isotope analysis and microbiome functional analysis could provide additional insights into the diets of synanthropic and forest mice and the role that their microbiome is playing in biological processes.Such information would likely inform the microbiome composition observed in our data, as the PERMANOVA modelling approach utilized herein indicated a high degree of unexplained microbiome composition variability that was not accounted for by landscape or habitat type.
Despite the changes in microbiome composition, microbiome heterogeneity was similar between landscape-habitat pairings.This was surprising as stressors such as those associated with land-use change and anthropogenic environments are often observed to increase variation in microbiome composition (Zaneveld et al., 2017), a pattern which has been found across wildlife taxa including frogs (Jiménez et al., 2020), rodents (Heni et al., 2023), birds (Obrochta et al., 2022), and bats (Ingala et al., 2019).Given the synanthropic nature of Peromyscus mice, it could be that the environments we sampled were still within the range of their natural tolerance, and thus we did not detect an increase in microbiome variation.Studying Peromyscus along a gradient of native to agricultural to urban environments could better test the limits of their environmental tolerances and the effects on microbiome variation (e.g.Diaz et al., 2023).
Accurate detection and taxonomic assignment of reads is a key assumption for community diversity and metagenomic analyses.
Species richness and diversity estimates can be sensitive to the presence of rare species.The Emu algorithm has a built-in abundance threshold of 10 reads for large samples (over 1000 reads) to control against long tails of low-abundance species which are an artefact of the probabilistic expectation-maximization model (Curry et al., 2022).As a result, Emu has a limited ability to detect rare species, and thus our estimates of species richness and diversity are likely underestimations of the true community composition.However, Emu's strength is that it was specifically designed for taxonomic identification of long-read sequence data.
The Emu pipeline helps to correct errors and improve the accuracy of nanopore 16S amplicon sequencing through the expectationmaximization algorithm and has been shown to outperform algorithms designed for short-read (i.e.Illumina) data when classifying 16S nanopore sequences (Curry et al., 2022).Because we were most interested in the species-level identification of reads for the detection of putative pathogenic bacteria, we chose to prioritize accurate taxonomic assignment over the ability to detect rare species and more accurately estimate species richness and diversity.
Furthermore, nanopore sequencing provides a key advancement over short-read microbiome sequencing in that species-level identification is possible and accurate.In future research, we see great utility for taxonomic assignment algorithms like Emu designed specifically for long-read nanopore sequences and expect these novel methods to continue to improve the ability to accurately characterize and study species-level microbiome composition.
Indeed, updated nanopore chemistry and flow cells are able to outperform Illumina sequencing with less noise and higher accuracy, specifically for species-level classification of 16S amplicon sequencing of gut microbiota (Szoboszlay et al., 2023).
We detected 16S sequences of putative pathogenic bacteria species in samples from all four landscape-habitat pairings.Overall, the abundance and diversity of putative pathogenic bacteria was low in all sampled habitats; we detected 1-2 putative pathogens per mouse and zero putative pathogens in many mice, and many putative pathogens were detected at low read counts.Even still, the prevalence of putative pathogens was higher in mice from the agricultural landscape than mice from the undeveloped landscape, while there was no difference between forest and synanthropic habitats.Thus, the landscape surrounding the sites in the agricultural landscape: crop fields, pastures, and low-density housing, could represent a source of infection for these pathogens.Peromyscus are known to forage in crop fields as well as forest habitats, so it is likely that exposure to the surrounding agricultural landscape increased the prevalence of putative pathogens observed herein.By contrast, the sites in the undeveloped landscape were contained in a state park and the forest continues uninterrupted beyond the park boundary with little agricultural development, limiting sources of pathogen exposure.
Despite the landscape-level differences, the prevalence of putative pathogens was similar between forest and synanthropic sites within a landscape, suggesting similar levels of pathogen exposure for mice between these two habitats.The sampling design could be a limiting factor preventing the detection of strong differences between habitat types within a landscape; sites in our study may have been too close together to differentiate forest and synanthropic habitats.Another explanation is that the synanthropic habitats sampled were all at the interface of forest and human-habitated areas.
Potentially, the synanthropic mice only occasionally visit the human structures where they were trapped (maintenance garages and storage areas, cabins and tent platforms, etc.) and predominantly reside in the nearby forest.Frequent movement of mice between native vegetation and synanthropic habitats could also account for similar putative pathogen exposure within a landscape type.As such, synanthropic behaviour may be less informative of the abundance of potential zoonotic bacterial pathogens in Peromyscus than the surrounding landscape type.
It is also important to clarify that, while we can be confident in accurate taxonomic assignment of the bacterial species detected in the sampled mice, the detection of putative pathogen species does not guarantee zoonotic potential.Many of these putative pathogens are ubiquitous in the environment and can be commonly found in soil.Thus, the reads we detected could indicate bacteria picked up from the environment, either ingested while foraging or via grooming.Some of these bacteria are also commensal in the human and mammalian gut and may only be opportunistic pathogens or only certain serotypes possess virulence factors capable of infecting humans.We did not assess pathogen viability or infectivity as this requires more in-depth genotyping or lab cultures that were outside the scope of this research.Therefore, we cannot say whether the putative pathogens we detected were living or would be capable of colonizing wild rodents or humans.As such, the low read counts (read count <200)  et al., 2021).In agricultural settings, manure used as fertilizer may serve as a source of environmental contamination for faecal-oral pathogens such as C. difficile (Frentrup et al., 2021) which could provide a transmission route to rodents and other wildlife.
Our findings also support previous work done in Minnesota using nanopore sequencing to identify pathogenic bacteria in synanthropic rodents.Jahan et al. (2021)

| CON CLUS IONS
The data presented herein provide a glimpse into the microbiome of Peromyscus mice in anthropogenic and native environments of

DATA AVA I L A B I L I T Y S TAT E M E N T
The full R code and data files used for analysis are available on GitHub (https:// github.com/ jmist rick/ MNpeRo) and archived with Zenodo (https:// doi.org/ 10. 5281/ zenodo.10655837).Field data for all trapped rodents, including metadata for all sequenced faecal samples, are archived with Dryad (https:// doi.org/ 10. 5061/ dryad.7m0cf xq2m) (Mistrick, Weinberg, et al., 2024).All raw data from ONT MinION sequencing runs described here are accessioned in FASTQ format with the National Center for Biotechnology Information Sequence Read Archive under BioProject ID PRJNA1068550 (Mistrick, Kipp, et al., 2024).

B EN EFIT-S H A R I N G S TATEM ENT
Benefits Generated: This research was supported by the Itasca Biological Station American Indian Fund whose goal is to foster sci-

F
Rodent sampling locations and sample size summary.Sampling was conducted at two locations in Minnesota, USA, representing undeveloped and agricultural landscapes.Study sites are outlined with white boxes (a, b).Sample size (total number of faecal samples) in forest and synanthropic habitats is shown for each landscape (c, d).Sampling was conducted once in the undeveloped landscape and three times in the agricultural landscape.Total number of samples per landscape-habitat pairing is noted first with samples per month in parentheses below.Maps and land cover classification legend from the National Land Class Database (NLCD) 2019 (Dewitz, 2021).Figure created with BioRe nder.com.
All rodent trapping and handling methods were reviewed and approved by the University of Minnesota Institutional Animal Care and Use Committee (protocol no.1903-36892A) and conducted under Minnesota Department of Natural Resources (MN-DNR) Division of Fish and Wildlife Special Permit No. 28440.
on 13 February 2023, plant-specific pathogens removed); resources from the U.S. Centers for Disease Control and Prevention on 'foodborne germs and illnesses' (CDC, 2022); and foodborne and mastitis-causing pathogens screened for by Jahan et al. (2021) (for full list of pathogens, see Table on observed species richness (F 1,131 = 15.28,β = 66.24, p < .001,95% CI [32.72, 99.76]), Shannon diversity (F 1,131 = 4.21, β = 0.55, p = .042,95% CI [0.02, 1.07]), and Simpson diversity indices (β = 0.56, p = .039,95% CI [0.03, 1.10]; Table dissimilarity index based on rarefied count data.A nonparametric analysis of similarities test ('anosim' function, 'vegan' R package) comparing dissimilarity indices between samples from the F I G U R E 2 Alpha diversity for all unique mouse faecal samples (n = 140) in agricultural and undeveloped landscapes and in forest and synanthropic habitats according to (a) observed species richness, (b) Shannon diversity index, (c) Simpson diversity index, and (d) species evenness.The reported Kruskal-Wallis test statistic compares each metric across landscapehabitat pairings; significant pairwise comparisons (Holm-corrected Dunn test) are shown with brackets.F I G U R E 3 Relative abundance of bacteria phyla per sample (n = 140) by landscape-habitat pairing showing phyla present at ≥1% relative abundance.Phyla observed at <1% relative abundance were grouped in a single category 'Other'.The microbiome of sampled mice was dominated by three phyla: Bacteroidetes, Firmicutes, and Proteobacteria.
Over 40% (30/69) of mice in the agricultural landscape had reads for at least one putative pathogen compared to approximately 15% (11/71) in the undeveloped landscape.There was no effect of F I G U R E 4 Non-metric multidimensional scaling ordination on microbiome community composition by Bray-Curtis dissimilarity index.Points represent individual samples, coloured by landscape-habitat pairing.Ellipses denote the 95% confidence level for a multivariate t-distribution of the data points per group (centroids marked with larger points).Stress value: 0.086 (k = 4).

F
Heatmap of read counts of putative pathogenic bacteria species per mouse in each landscape-habitat pairing (count threshold ≥50 reads).The horizontal axis represents samples from an individual mouse.Warmer colours indicate higher read abundance (natural log scale).Potentially opportunistic human pathogens are noted with a star (*).
we detected could indicate that the bacteria were only transient and not actively colonizing the gut.Nonetheless, we did detect relatively high read counts of several putative pathogens (Citrobacter freundii, Enterococcus gallinarum, Morganella morganii, and Serratia marcescens) which indicate the potential for Peromyscus mice to be reservoirs for zoonotic pathogens and can inform future studies that characterize the pathogenicity of these bacteria in wild rodents.Many of the putative pathogens most frequently detected in this study (Enterococcus gallinarum, Citrobacter freundii, Clostridioides difficile, Morganella morganii, and Streptococcus sanguinis) have been previously detected in wildlife associated with anthropogenic environments.Urban coyotes consuming high levels of anthropogenic food have been found to be associated with high levels of Enterococcus and Streptococcus (Sugden et al., 2020).Urban rodents and those living on or near farms have been documented to be hosts of Clostridioides difficile, including antimicrobial-resistant strains (reviewed in Weese, 2020), and short-tailed shrews living on or near farms have been documented to carry Morganella morganii (Jahan pointed to the role that farms play in the increased abundance of putative pathogenic bacteria in rodents.Our work expands upon the foundation set by Jahan et al. by investigating less disturbed environments to understand the abundance and diversity of zoonotic bacterial pathogens in rodents in undeveloped and agricultural (cropland) landscapes.The diversity of putative pathogenic genera found in Peromyscus mice generally aligns between our studies: Jahan et al. similarly identified putative pathogenic genera including Clostridium, Enterococcus, and Streptococcus circulating in synanthropic rodents on Minnesota farms.Interestingly, Jahan et al. found lower abundance of putative pathogenic genera in Peromyscus mice compared to other rodent species trapped on farms including Mus musculus, Microtus pennsylvanicus, and Rattus norvegicus.While our study did not include other rodent species, the limited abundance of putative pathogenic bacteria found in Peromyscus herein corroborates the findings of Jahan et al. and could indicate lower exposure for Peromyscus compared to other synanthropic rodents.
Minnesota.By sampling from populations in agricultural and undeveloped landscapes and in forest and synanthropic habitats, we find that anthropogenic factors at the landscape and habitat levels influence microbiome diversity and community composition in wild Peromyscus.Our findings suggest that Peromyscus are occasional hosts of putative pathogenic bacteria, albeit at low abundance and diversity, and that agricultural development may play a role in increasing the prevalence of potential zoonotic bacterial pathogens in these rodents.Importantly, even where transmission risk seems low, infection in wildlife populations could represent sources of novel pathogenic strains, bridge hosts linking environmental contamination back to human or livestock infection, or vectors to translocate pathogens across the landscape.As such, this research underscores the importance of investigating zoonotic pathogen prevalence in synanthropic rodents and other wildlife to better characterize their potential as reservoirs and vectors for pathogen spillover at the human-wildlife interface.reported results.The data is available through Dryad at https:// doi.org/ 10. 5061/ dryad.7m0cf xq2m and through the NCBI SRA under BioProject ID PRJNA1068550.Code is available through Zenodo at https:// doi.org/ 10. 5281/ zenodo.10655837.
entific growth and collaboration between local students from area high schools and the researchers at the Itasca Biological Station and Laboratories.A recent high school graduate (C.C.Adams) collaborated on the field research and is included as a co-author.As described above, all data resulting from this research have been publicly shared via appropriate research databases.
Summary statistics of 16S nanopore sequencing data of mouse faecal sample DNA (after quality filtering) by landscape, habitat type, and sampling month.
Note: Mean and standard deviation are reported for number (N) of reads per sample (reported in units of thousands of reads), number of basepairs (BP) per sample (reported in units of millions of basepairs [Mb]), and read quality (Q) score.Individual sampling months in the agricultural landscape shown in italics, rows shaded in grey.Mean values across all three months shown in bold.