Background

Recognition that host-associated microbial communities (microbiomes) affect host health [1, 2], phenotypes [3, 4], and fitness [5, 6] has sparked interest in understanding the causes, consequences, and eco-evolutionary relevance of microbiome variation in nature [7,8,9,10]. The microbiome’s greater mutability (relative to the host genome) undergirds theorization that the microbiome plays an important evolutionary role in vertebrates [10,11,12]. Namely, it is hypothesized that the microbiome may help to buffer the adverse effects of novel environmental change by extending a host’s phenotypic range. In doing so, the microbiome might facilitate adaptive stop-gap solutions during host colonization to a new ecological niche. The gut microbiome is commonly invoked in such theorization, since the semi-permeable intestinal epithelium provides an intimate host-microbe interface through which microbiota can shape host behaviour [13], immunity [14], homeostasis [15], digestion [16], and dietary detoxification [17]—traits which ultimately define a species fundamental niche. For example, dietary detoxification by intestinal microbes may facilitate persistence of woodrat (Neotoma lepida) populations at the southern extent of their range, despite the northward expansion of creosote bush (Larrea tridentata) which produces a toxic resin and an armament of plant secondary compounds [17, 18]. Central to the hypothesis that microbiomes facilitate host population persistence is the prediction that among host populations, there exists genetic variation in traits that shape microbiome plasticity [11]. Many free-living study systems are intractable for empirically testing these predictions, partly because it is difficult to define what constitutes a truly ‘novel’ environmental change.

Urban ecosystems have achieved recognition in the field of evolutionary biology as opportunistic, mass-replicated, free-living ecological and evolutionary experiments [19, 20]—experiments which are useful for probing the host-microbiome relationship. Cities are among the fastest growing terrestrial ecosystems in the world [21] and expose wildlife to combinations of stressors and stimuli unlike anything observed in nature [22]. The novelty of urban environmental selective pressures is such that, globally, rates of phenotypic change among plants and animals show a distinct urban signature [23]. However, for many of these phenotypic changes, an underlying host genetic basis has not be strongly demonstrated [24]. The rapidity of phenotypic change of plants and animals in response to urbanization, and the nascence of cities on an evolutionary timescale, make urban environments a useful testing-ground for hypotheses that the microbiome can facilitate wildlife or—from a holobiont perspective—drive adaptation to a new ecological niche [11, 25, 26].

Despite widespread evidence for an effect of urbanization on the microbiomes of wild plants and animals, reported patterns are inconsistent across host species [27,28,29,30,31,32,33,34,35]. Inconsistencies of urban effects are likely attributable to; 1) city-specific variation (most studies to-date have focused on populations residing within a single city); 2) differences in the operational definition of urbanization; 3) differences in the how host species interact with the urban environment. For example, among vertebrates, consumption of human food resources is hypothesized to be the primary driver of gut microbiome variation [30, 32, 36, 37]. Dietary variation is likely an important contributor to urban gut microbiome variation for some wildlife, but not all species rely on human food subsidies to the same extent [38]. Alongside diet, patterns of intra- and inter-specific microbiota dispersal between hosts may help shape microbiome differences between environments [39,40,41,42]. Likewise, the physiological responses of hosts to environmental stimuli may be more important than the details of the stimuli itself [35, 43]. The capacity of the microbiome to facilitate wildlife expansion to a new ecological niche may therefore strongly depend on the outcome of gene x environment x microbiome interactions [44].

Building on past research, we adopt a multi-city perspective in studying the effect of urbanization on the faecal microbiome of eastern grey squirrels (Sciurus carolinensis)—a species with a prominent coat colour polymorphism which is both linked to urbanization and pleiotropically connected to gene x environment interactions among squirrels [45, 46]. Specifically, this colour polymorphism is the result of an incompletely dominant mutation in the gene which encodes the melanocortin-1 receptor of the proopiomelanortin system [47]; agouti: homozygous for the wild type allele, black: homozygous for the mutant allele, intermediate: heterozygous for the wild-type and mutant alleles (Fig. 1). Along replicate rural-urban gradients, the melanic phenotype has been reported to increase in frequency relative to the agouti wild-type [48, 49]. Fur and feather colour polymorphisms are frequently pleiotropically linked to differences in vertebrate physiology and behaviour, such as immunity, stress physiology, social behaviour, and grooming [50]. Variation in each of these traits is known to affect the host-associated microbiomes [35, 51, 52]. If dimensions of eastern grey squirrel physiology are pleiotropically linked to coat colour, then we predict microbiome differences between colour morphs, or colour morph differences in the microbiome’s response to urbanization, if colour pleiotropy linked traits are subject to gene x environment interactions.

Fig. 1
figure 1

Representative photos of the (left) agouti (zero copies of the melanism causing MC1R-Δ24 mutant allele), (centre) intermediate (one copy of the MC1R-Δ24 mutant allele), and (right) black (two copies of the MC1R-Δ24 mutant allele) eastern grey squirrel (Sciurus carolinensis) coat colour phenotypes

Full size image

Using a 16S rRNA gene amplicon sequencing approach, we characterized the faecal bacterial microbiome of 195 individual adult eastern grey squirrels spanning three cities (Guelph, Waterloo, Windsor) and three rural forests in southern Ontario (Canada). A single rural forest site was located a short distance outside each of the three focal cities, to mitigate the effect that spatial biases could have on environmental comparisons. To test for an effect of habitat heterogeneity within a single city (Guelph), we also sampled squirrels from three land classes presumed to possess disparate biotic and abiotic characteristics: 1) an urban university campus, 2) two suburban parks, and 3) two urban forest fragments.

We sought to test the prediction that the colonization of cities by eastern grey squirrels drives convergence in the bacterial microbiome, and to identify the spatial scale at which this convergence occurs. To inferentially parse what (non-mutually exclusive) ecological processes (e.g., selection, dispersal, drift) might underlay inter-environmental microbiome variation, we used a phylogenetic null modelling approach [53,54,55]. We predicted that homogenizing selection should act on the microbiome within environments, but that evidence for disparate selective pressures acting on the microbiome would be observed between environments. To evaluate competing hypotheses that local (forest versus human built-environment) [27, 29] versus landscape (city versus rural) [31] level environmental variation is more important in shaping the microbiome, we contrasted variation in the microbiome explained by between- versus within-city habitat classifications (university campuses, suburban parks, urban forests, rural forests). Finally, to test for an effect of intra-specific physiological variation, we tested for interactions between coat colour phenotype and the bacterial microbiome’s response to urbanization.

Results

α-Diversity

We observed 1256 operational taxonomic units (OTUs) across 195 squirrel faecal samples, with an average of 379 OTUs ±4 SE per squirrel. The eastern grey squirrel microbiome was comprised primarily of Lachnospiraceae (37% ± 1% SE), Ruminococcaceae (19% ± 0.6% SE), Muribaculaceae (10% ± 0.5% SE), Prevotellaceae (8% ± 0.5% SE), Lactobacillaceae (7% ± 0.8% SE), and Bacteroidaceae (5% ± 0.5% SE). The community membership patterns we observed are consistent with previously published microbiome characterizations of eastern grey squirrels [35], and other tree squirrel and ground squirrel species [56, 57].

No difference in α-diversity was observed between sexes (β = 2.11 ± 8.12 SE, t = 0.26, p = 0.80). Agouti squirrels from city sites trended towards having fewer OTUs than agouti squirrels from rural sites, albeit this effect was marginally non-significant (β = 59.88 ± 31.17 SE, t = − 1.92, p = 0.06). However, we observed a colour phenotype x environment interaction, whereby α-diversity decreased with increases in MC1R mutation number (agouti: 0, intermediate: 1, black: 2) at rural sites (β = − 29.39 ± 12.26 SE, t = − 2.40, p = 0.02) but increased with melanism mutation number at city sites (β = 37.27 ± 13.86 SE, t = 2.69, p < 0.01; Fig. 2). Qualitatively similar results were observed with respect to Shannon diversity estimates (box-cox-transformed, λ = 4), wherein no effect of sex was observed (β = 6.83 ± 4.59 SE, t = 1.49, p = 0.14) and city agouti squirrel microbiomes had marginally lower Shannon diversity than agouti squirrels at rural sites (β = − 37.36 ± 18.50 SE, t = − 2.02, p = 0.05). A marginally non-significant negative correlation between melanism and Shannon evenness among rural sites (β = − 12.59 ± 7.00 SE, t = − 1.80, p = 0.07), but a positive correlation between Shannon diversity and melanism existed among city sites (β = 18.32 ± 7.90 SE, t = 2.32, p = 0.02).

Fig. 2
figure 2

Bacterial α-diversity in the squirrel microbiome is negatively correlated with melanism at rural sites (p = 0.02), but positively correlated with melanism among squirrels within city sites (p < 0.01). Solid horizontal lines within violins denote the interquartile range

Full size image

β-Diversity

OTU composition (Euclidean distance of centre log-transformed OTU counts) of the microbiome was correlated with environment (city versus rural forest; F = 6.96, R2 = 0.03, p < 0.01), within-city land class (urban campus, suburban park, urban forest, rural forest; F = 4.95, R2 = 0.05, p < 0.01), trapping site (F = 2.78, R2 = 0.08, p < 0.01; Supplementary Information: Fig. S1), and sex (F = 1.78, R2 = 0.01, p < 0.01; Supplementary Information: Fig. S2) based on PERMANOVA testing. Conversely, we observed no effect of coat colour phenotype (F = 1.01, R2 = 0.01, p = 0.38) and no interaction between environment and coat colour (F = 1.07, R2 = 0.01, p = 0.21). Based on ordination of the bacterial microbiome (Fig. 3) and hierarchical clustering analysis (Supplementary Information: Fig. S3), squirrels from suburban parks within the city of Guelph clustered most closely with those from university campuses (built-environment sites). Conversely, squirrels from urban forests in Guelph clustered most closely with squirrels from distant rural forests (forest sites).

Fig. 3
figure 3

The first three axes of a principal coordinate analysis ordination of Euclidean distances (centred log-ratio transformed OTU dataset) separating eastern grey squirrel microbiomes. Points coloured by land class and shaped by local environment type (forest versus built-environment). Solid lines denote 95% confidence ellipses around environment type

Full size image

Like OTU composition, phylogenetic structure of the squirrel microbiome (weighted UniFrac distance) was correlated with environment (F = 4.24, R2 = 0.02, p < 0.01), within-city land class (F = 5.05, R2 = 0.05, p < 0.01), trapping site (F = 3.26, R2 = 0.09, p < 0.01), and sex (F = 2.47, R2 = 0.01, p < 0.01). Coat colour phenotype was marginally non-significant (F = 1.33, R2 = 0.01, p = 0.06), and interestingly, we observed a significant interaction between environment type and coat colour phenotype (F = 1.49, R2 = 0.01, p = 0.02).

Visualization of weighted UniFrac ordination plots suggest that the environment x phenotype interaction we observed might be partly the result of differences in β-dispersion between phenotypes across environments, rather than solely mean community dissimilarities (Fig. 4). This interpretation was supported by post-hoc permutation tests of multivariate homogeneity for phenotype-environment grouping beta-dispersion. Specifically, city agouti squirrels harboured microbiomes which were more phylogenetically variable than rural agouti squirrels (p = 0.05). No other phenotype-environment groups differed in beta-dispersion, but city intermediate squirrels trended towards displaying greater phylogenetic variability than rural agouti squirrels (p = 0.07).

Fig. 4
figure 4

Weighted UniFrac principal coordinate analysis ordination of the eastern grey squirrel microbiome separated by environment and coloured and shaped by phenotype with 95% confidence ellipses. Agouti squirrels display greater beta-dispersion (p = 0.05) within cities (right panel) than in rural environments (left panel)

Full size image

Null-modelling inference

Conventional β-diversity analyses are used to test whether community structure differs between groups but provide little evidence with which to infer why communities may differ. Further, patterns in β-diversity can be influenced by imbalances in α-diversity between communities [58]. Phylogeny-independent and phylogeny-weighted null modelling methods can be used to determine whether community composition (or between community dissimilarities) deviate from random expectations, given community α-diversity and pre-defined pool of γ-diversity (see Methods) [53, 55].

We tested for a phylogenetic signal among the bacteria observed in the squirrel microbiome with respect to their average relative abundance within forest and built-environment squirrel populations. We observed a significant positive signal over short phylogenetic distances (Supplementary Information: Fig. S4). This suggests that closely related bacterial OTUs tended to occupy a similar ecological niche, a result which supports the use of mean nearest taxon distance (MNTD), a measure of phylogenetic clustering/dispersion within communities [54].

Effect size standardized and box-cox transformed (λ = 0.4) absolute MNTDses values were closer to stochastic expectations (i.e., 0) among city agouti squirrels compared to rural agouti squirrels (β = − 1.28 ± 0.58 SE, t = − 2.22, p = 0.03); however, we observed an interaction whereby—among city squirrels—|MNTDses| increased with melanism mutant allele number (β = 0.51 ± 0.24 SE, t = 2.15, p = 0.03; Fig. 5A). Conversely, we observed no effect of sex (β = 0.17 ± 0.14 SE, t = 1.20, p = 0.23) or melanism among rural squirrels (β = − 0.29 ± 0.21 SE, t = − 1.38, p = 0.17).

Fig. 5
figure 5

An interaction between coat colour phenotype and environment affected (A) |mean nearest taxon distances| (box-cox transformed) (city versus rural), and (B) β mean nearest taxon distance (βMNTDses) (built-environment versus forest), such that city and built-environment agouti squirrels had values closer to stochastic expectations. Dotted horizontal lines denote values beyond two standard deviations from the null distribution. Solid horizontal lines within violins denote the interquartile range. Horizontal brackets and ‘*’ denote significant differences in βMNTDses variance homogeneity using permutation tests (p < 0.05)

Full size image

The same principles which underlie null modeling of within community phylogenetic structure can be applied to between community contrasts. Using this approach, phylogenetic distances between all taxa in one community are measured to their nearest related taxon in another community [53]. When effect-size is standardized by the standard deviation of the corresponding null distribution, these values provide an indication of whether any two communities are more (βMNTDses > 2), less (βMNTDses < 2), or no more (|βMNTDses| < 0) phylogenetically disparate than expected by chance [53]. If taxon niche-space is shallowly phylogenetically conserved (an assumption supported by the positive phylogenetic signal described above), deviations from null expectations are one indication that similar (or disparate) selective pressures act between communities. Based on a PERMANOVA test, microbiome βMNTDses values were not correlated with environment (F = 1.65, R2 = 0.01, p = 0.10), within-city land class (F = 0.42, R2 < 0.01, p = 0.96), trapping site (F = 1.27, R2 = 0.04, p = 0.16), sex (F = 1.16, R2 = 0.01, p = 0.33), or coat colour phenotype (F = 1.38, R2 = 0.01, p = 0.17); however, we observed an interaction between environment and coat colour phenotype (F = 1.57, R2 = 0.02, p = 0.05). Post-hoc permutation tests of multivariate homogeneity indicated that city agouti squirrels had smaller βMNTDses variances than either rural agouti squirrels (p = 0.04) or city black squirrels (p = 0.02). Similarly, city squirrels of the intermediate morph had smaller variances than city black squirrels (p = 0.02). Notably, the magnitude of βMNTDses values among city agouti squirrels and city intermediate morph squirrels, were more consistent with stochastic expectations than they were homogenizing selection. Conversely, black squirrels maintained comparably large βMNTDses values, regardless of their environment.

Because squirrels from rural and urban forests clustered together in Euclidean-based β-diversity analyses, we additionally considered a PERMANOVA in which squirrel faecal samples from city forests were grouped with those from rural forests to test for an effect of urbanization at a finer spatial scale (forest versus built-environment). Again, we observed no effect of within city land class (F = 0.00, R2 = 0, p = 0.99), trapping site (F = 1.27, R2 = 0.04, p = 0.10), sex (F = 1.17, R2 = 0.01, p = 0.36), or coat melanism (F = 1.39, R2 = 0.01, p = 0.14); however, we did observe a significant effect of urbanization at this finer spatial scale (human built environment versus forest; F = 2.73, R2 = 0.01, p < 0.01) and an interaction between habitat-type and coat colour phenotype (F = 1.86, R2 = 0.02, p = 0.01). As above, post-hoc permutation tests of multivariate homogeneity indicated that built-environment agouti squirrels had smaller βMNTDses variances than forest agouti squirrels (p < 0.01), forest black squirrels (p < 0.01), or built-environment black squirrels (p < 0.01; Fig. 5B) and trended towards smaller variances than forest intermediate morph squirrels (p = 0.06). Built-environment intermediate squirrels likewise had smaller βMNTDses values than forest black squirrels (p = 0.04), and although marginally non-significant, trended towards having smaller values than either forest agouti squirrels (p = 0.07) and built-environment black squirrels (p = 0.09).

Raup-Crick bray (RCbray) values which—in contrast to βMNTDses measurements—indicate the magnitude of OTU turnover, independent of taxa phylogenetic relatedness. In the absence of βMNTDses deviations from stochastic expectations (i.e., |βMNTDses| < 2), RCbray values which are higher (> 0.95) or lower (< − 0.95) are suggestive of dispersal limitation and homogenizing dispersal, respectively. Conversely, values between these extremes are considered indicative of ecological drift. Of 18,943 pairwise comparisons between squirrels, 14,143 (75%) had |βMNTDses| < 2, and of these pairs, 13,851 (98%) had corresponding values of RCbray > 0.95. Environment (city versus rural; F = 1.21, R2 = 0.01, p < 0.01), within-city land class (F = 1.78, R2 = 0.02, p < 0.01), trapping site (F = 1.34, R2 = 0.04, p < 0.01), and sex (F = 1.28, R2 = 0.01, p < 0.01) were all correlated with RCbray values in a PERMANOVA test. But we observed no effect of phenotype (F = 1.05, R2 = 0.01, p = 0.20), and only a marginally significant interaction between environment and phenotype (F = 1.10, R2 = 0.01, p = 0.06). However, the latter result might be expected, given the significant interaction of environment and phenotype on βMNTDses values.

Post-hoc Kruskal-Wallis tests indicated that RCbray values were higher among comparisons made within trapping sites than between trapping sites (χ2 = 530.97, p < 0.01). Interestingly, RCbray values within city sites were smaller than those observed within rural sites (χ2 = 19.77, p < 0.01; Supplementary Information: Fig. S5A). RCbray values also differed between city and rural environments with respect to between site comparisons (χ2 = 19.99, p < 0.01). Specifically, RCbray values were smaller between city sites than between rural sites (p < 0.01) or between city and rural sites (p < 0.01). RCbray values also differed by sex both within sites (χ2 = 7.83, p = 0.02; Supplementary Information: Fig. S5B) and between sites (χ2 = 15.41, p < 0.01). Within sites, post-hoc dunn tests indicated that female-female pairs had lower RCbray values than male-male pairs (p = 0.02) or female-male pairs (p < 0.01). In comparisons made between sites, female-male pairs had larger RCbray values than either female-female (p = 0.01) or male-male pairs (p < 0.01).

Differential abundance testing

To generate a robust understanding of which taxa drove the apparent divide in the microbiome between squirrels in forest versus built-environments, we used two different, but complimentary, statistical approaches. The first, selection balance analysis, accounts for the compositional nature of most 16S amplicon datasets by seeking the most parsimonious log-ratio of taxa that delineate two groups [59]. Results of selection balance analyses are therefore derived from parsimony and discriminative sensitivity, rather than test statistics typical of frequentist approaches. The second approach, analysis of compositions of microbiomes with bias correction (ANCOM-BC), normalizes sequence counts by a process similar to centred-log ratio transformations and applies corrections to control for false discovery rates [60].

Selection balance cross-validation identified that the apparent divide in the microbiome between squirrels in the human built-environment (university campuses and suburban parks) versus forests (urban forest and rural forests) could be most parsimoniously discriminated by a log-ratio balance of 12 genera. Squirrels from the built-environment had a greater abundance of Odoribacter, Oscillospira, Sulfurimonas, Ruminiclostridium (sub-group 5), Coprococcus (sub-group 3), and unclassified genera within the class Mollicutes RF39, relative to Alistipes, Ruminiclostridium (sub-group 1), Rikenellaceae (sub-group RC9), Lachnospiraceae (sub-group UCG-010), Clostridiales (sub-group vadinBB60), and unclassified genera of Lachnospiraceae, when compared to forest squirrels (Supplementary Information: Fig. S6A). This balance was able to discriminate between built-environment and forest samples with high sensitivity (AUC-ROC = 0.975). Similarly, at the family level, a log-ratio of Mollicutes RF39: Rikenellaceae most parsimoniously delineated built-environment and forest squirrels (AUC-ROC = 0.808; Supplementary Information: Fig. S6B).

The results of ANCOM-BC analyses were qualitatively like those of the selection balance analysis, with 17 genera observed to differ between forest and human built-environments. Squirrels from the built-environment had greater relative abundances of Mollicutes (sub-group RF39), Ruminiclostridium (sub-group 5), Prevotella, Parasutterella, Oxalobacter, unclassified genera of Eggerthellaceae, as well as genera of Ruminococcaceae (sub-group UCG-014, sub-group UCG-008) and Lachnospiraceae (Lachnoclostridium, Blautia, Eisenbergiella, Coprococcus sub-group 3). Conversely, forest squirrels had greater abundances of Alistipes, Clostridiales (sub-group vadinBB60), Lachnospiraceae (sub-group FCS020), Lachnospiraceae (sub-group NK4A136), and a collection of unclassified Lachnospiraceae genera (Supplementary Information: Table S1). Notably, these environmental differences appeared driven primarily by the agouti morph (Fig. 6). After repeating ANCOM-BC analyses upon datasets subset by colour phenotype, we observed that between forest and human built-environments, 14 genera differed among agouti squirrels, 4 genera differed among the intermediate morph, and 0 genera differed among black coated squirrels (Supplementary Information: Fig. S7).

Fig. 6
figure 6

A stacked barplot of bacterial genera indicated by ANCOM-BC analyses to significantly differ in relative abundance between forest and built-environments in the full dataset or among at least one coat colour phenotype. Facetted by coat colour phenotype and environment type

Full size image

Discussion

Urbanization clearly affected the eastern grey squirrel microbiome. This result is consistent with findings from birds [27, 29,30,31,32], reptiles [36], humans [61, 62], insects [63], plants [64], and wild mammals [35]. We further demonstrate that convergence occurs across cities, but also that substantive variation exists both between and within cities. For example, the variation explained by city-scale urbanization was comparable to the variation explained by land class heterogeneity within a single city (university campuses, suburban parks, urban forests). This is consistent with reports that fine-scale environmental variation can play a substantive role in shaping the microbiome [65, 66] and suggests the need for caution when trying to quantify an individual’s environment with simplified univariate terms, like percent impermeable landcover in the context of urban landscapes. For example, our urban forest and suburban park sites were characterised by similarly sized urban greenspaces, but squirrels from urban forests tended to cluster with squirrels from rural forests in ordination and hierarchical clustering plots, while those from suburban parks were more alike those from university campuses. Broad-scale urbanization (city versus rural) had no effect on βMNTDses, however, βMNTDses values were affected by whether squirrels were from local habitats characterized as forest or human built. These phylogenetic patterns in the microbiome suggest that environmental selective pressures act on the microbiome (perhaps indirectly through host physiology) at fine spatial scales. Therefore, although we report an effect of urbanization on the microbiome, habitat heterogeneity within cities makes a representative ‘urban microbiome’ for host species unlikely.

Anthropogenic food subsidies in cities are hypothesized to underlie microbiome variation among urban mammals [30, 37], as well as reports of obesity and hyperglycaemia [67,68,69]. Western diet-induced obesity in mice and pigs maintained on diets low in indigestible fibre and starch (as might be expected of a western diet) are strongly characterized by blooms of Mollicutes which appear to specialize on fermenting simple sugars [70,71,72]. Tellingly, we observed Mollicutes to be one of the strongest discriminating features separating built-environment squirrels from forest squirrels in both selection balance and ANCOM-BC testing.

ANCOM-BC tests further identified that built-environment squirrels harboured greater relative abundances of Parasutterella—sister genus to the very ecologically similar Sutterella [73]. Sutterella were observed in one study to be a strong feature characterizing the microbiome of wild red squirrels supplemented with peanut butter, a food higher in sugar and fat than natural components of the squirrel’s diet [56]. Rather than dietary fats, Parasutterella and Sutterella appear to specialize on the bile acids, which hosts produce to solubilize dietary fats [74].

These results are exemplative of a broader pattern, whereby city and built-environment squirrel microbiomes were generally characterized by a shift towards taxa which show evidence of metabolizing fats and host-derived products, especially bile acids. For example, a large collection of unclassified genera of Lachnospiraceae—a family comprised of primarily fibrolytic specialists and plant fibre fermenters [75]—was less abundant among built-environment squirrels compared to forest squirrels. Despite this, a handful of Lachnospiraceae genera were found in ANCOM-BC analyses to be more abundant in squirrels from the built-environment, but like Parasutterella and Sutterella, these exceptions are known to metabolize animal host derived products (Lachnoclostridium [76,77,78,79], Blautia [80,81,82], Eisenbergiella [83]). Genera within the family Eggerthellaceae and the genus Odoribacter were likewise built-environment associated—in ANCOM-BC and selection balance analyses, respectively—and likewise specialize on bile acids and other host derived products [76, 84,85,86,87]. In addition to metabolizing host products, many of these genera have been implicated in western diet-related metabolic and gastro-intestinal diseases in humans (Lachnoclostridium [76, 88], Blautia [89,90,91,92], Eisenbergiella [88], Eggerthellaceae [76, 86, 88, 93, 94], Odoribacter [87, 92, 95,96,97]). While diet may be the ultimate cause of the built-environment versus forest squirrel microbiome divide, host physiological responses to their diet may be a complimentary, if not more proximate, mechanism. This distinction is important, since a genetic basis to host physiological responses to their diet allows for evolution in the diet x microbiome relationship [35].

The need to consider proximate host physiological mechanisms was exemplified by our observation that although urbanization had an apparent effect on the squirrel microbiome, these effects were partly mediated by squirrel coat colour phenotype. Coat melanism was negatively correlated with OTU richness in rural environments but positively correlated with OTU richness and Shannon diversity in the city. Black squirrels harboured microbiomes which were more phylogenetically dissimilar than predicted by null expectations but did not differ systematically between environments. When parsed by phenotype, no genera differed in abundance between black squirrels sampled in built-environments and forests, whereas 14 genera differed among agouti squirrels between habitats. Agouti squirrels also displayed greater phylogenetic variability in the microbiome within cities than black squirrels, and null modelling results suggest that this might be due to stochastic (rather than deterministic) processes (|MNTDses| and βMNTDses values closer to 0). If diet is indeed an ultimate cause for the effects of urbanization on the grey squirrel microbiome, then dietary effects might be mediated by physiological differences between squirrel colour morphs.

Fur and feather melanism—like that observed in grey squirrels—is often pleiotropically linked through the pro-opiomelanocortin system to myriad physiological pathways [50]. These pathways include baseline hypothalamic-pituitary-adrenal (HPA) physiology, HPA axis reactivity to stressors, and the immune system [98,99,100,101,102]—each of which is a dimension of host physiology known to affect wildlife microbiomes [35, 43, 103]. In eastern grey squirrels, the melanism-causing MC1R mutation [104] has been connected to behavioural and physiological differences [105, 106], most notably thermogenic physiology [45, 46]. Specifically, melanic squirrels show greater plasticity in their ability to adaptively lower their basal metabolic rate when exposed to sub-zero ambient temperatures. This gene x environment interaction may be the result of MC1R linked pleiotropy, however, more recent evidence suggests that the MC1R mutation in grey squirrels might have originated from introgression with eastern fox squirrels (Sciurus niger) [107]. Therefore, the MC1R mutation—and the greater physiological plasticity with which it appears correlated—might reflect more substantive underlying genetic differences between colour morphs.

Greater physiological plasticity among black squirrels might paradoxically explain why this morph tended to maintain similar microbiomes regardless of their external environment. Conversely, if agouti squirrels are physiologically inflexible, they may be incapable of acclimating to novel urban conditions or environmental stressors, and thereby lose, or relax, control of their resident microbiota. The resultant homeostatic disruption might explain the greater evidence for ecological drift within the city and built-environment squirrel microbiome. A physiological basis to the colour phenotype patterns we observed seems likely given past research [45, 46, 48, 50]. In testing the hypothesis that microbiome plasticity is adaptive for hosts in novel environments [11], we caution that it is important to parse whether observed variance in the microbiome is the result of stochastic processes (perhaps signalling a loss of host homeostatic control) or deterministic processes (possibly mediated by plasticity in host physiology); the outcome for host health and fitness may be very different depending on ecological process responsible for driving beta-dispersion in the microbiome [2].

Despite the emphasis we have placed on selection and ecological drift, patterns in bacterial dispersal likely also strongly contribute to the inter-environmental variation that we observed [41]. For example, bacterial dispersal limitation undoubtedly occurs between sampling locations—an interpretation supported by the smaller RCbray values observed within sites than between sites. This is unsurprising as spatial structure and social structure within [39, 108,109,110,111] and between [40, 65, 112, 113] mammalian populations has been demonstrated to affect the microbiome. More surprising, was our observation that RCbray values tended to be smaller within city sites than within rural sites—despite no effect of environment on βMNTDses values. This suggests that bacterial dispersal limitation might be stronger between squirrels within rural habitats versus between squirrels within city environments. Squirrel populations persist at greater densities on urban landscapes [114], which could facilitate greater microbial exchange via more frequent interactions with conspecifics [108]. Conspecific interactions are further catalyzed by spatial clustering of anthropogenic food sources (bird feeders, garbage cans, picnic areas etc.) which are known to increase rates of pathogen transmission in wildlife [115]; the same process could as easily facilitate greater exchange of commensal or mutualistic bacteria. Further, bacterial dispersal is not restricted to among hosts of the same species, but rather, are partly shaped by trophic interactions [40]. Urban food webs tend to have fewer species, and more interactions per species (i.e. greater connectivity [116]), which might help to promote the exchange of microbiota between the microbiomes of co-occurring colonizing wildlife. Greater connectivity and spatial overlap between con- and hetero-specific hosts within urban environments could facilitate more frequent microbial dispersal when compared to rural sites.

Environment dependent patterns of microbiota dispersal may likewise partly underlie beta-diversity differences in the microbiome between environments. We observed that RCbray values were smaller among pairwise comparisons made between city sites than comparisons made between rural sites or between city and rural sites. Substantive differences in abiotic and biotic factors between these environments ensure that urban and rural squirrels are very likely exposed to different pools of bacterial γ-diversity. For example, urbanization has been connected to predictable biodiversity loss and landscape homogenization [117], effects which might extend to the microbial communities. Furthermore, urban biological communities are strongly shaped by human socio-cultural factors [118]. Since the cities included in our study were built in a very similar socio-cultural context, the plant and animal species within these cities (and therefore the microbiota to which squirrels are exposed) might be more similar between cities than between rural forests. Furthermore, even when plant or animal species are found in both city and rural environments, the microbiota they transmit to squirrels may differ. For example, the phyllosphere microbiota of trees—with which squirrels closely associate—are themselves affected by urbanization [34]. Finally, the near-constant exchange of people and resources between cities facilitates gene flow and prevent the genetic isolation of urban wildlife populations [119, 120]. A similar mechanism might allow for greater bacterial dispersal between cities than between isolated rural forest fragments. Although speculative, it is important to consider an organism’s broader microbial milieu when studying host-microbe symbioses, rather than lay causality solely at the feet of host diet and physiology.

Lastly, we unexpectedly observed evidence that some of the sex effects among squirrels might derive partially from patterns in microbiota dispersal. Namely, RCbray values were smaller between females than between sexes or between males, despite no effect of sex on βMNTDses values. These differences could derive from behavioural differences in bacterial transmission, or physiological differences which affect colonization success, as suggested by researchers who characterized a similar pattern of female-biased bacterial transmission among co-housed common marmosets (Callithrix jacchus [121];). Among North American red squirrels, inter-individual bacterial dispersal appears to occur primarily through the maternal line [56]; therefore, a pattern of lower OTU turnover might have been observed between females because bacterial dispersal occurs both from a female’s parents and to a female’s offspring. By contrast, males may not directly contribute microbiota to their offspring. These familial-structured bacterial dispersal patterns are likely continually reinforced among related female grey squirrels, which show a greater propensity for social grooming and nest sharing when compared to males [122]. These results are consistent with patterns in the microbiome of black howler monkeys (Alouatta pigra) in which social bonds are strongest among female-female dyads [123]. Interestingly, the opposite patterns are observed among semi-feral welsh ponies (Equus ferus caballus), in which males show greater centrality in both social and (inferred) microbiota dispersal networks [109]. Similarly, bacterial transmission within social networks of wild wood mice (Apodemus sylvaticus) is most strongly driven by males, despite no difference in social association strength between sexes [52]. Therefore, while dietary and physiological differences between hosts affect microbiota colonization success and abundances, organismal behaviour and variation in social structure shape microbiota metacommunities, and determine which microbiota are given the opportunity to colonize a new host [13, 41, 110].

Conclusions

Urbanization affected the eastern grey squirrel microbiome at landscape and intra-city spatial scales. Characteristics of the urban squirrel microbiome echo those reported in humans and laboratory mice maintained on high-fat, low-fibre, westernized diets; however, the response of the squirrel microbiome to urbanization was coat colour phenotype dependent. Namely, the inter-environmental differences we observed were driven primarily by the agouti morph, however, we did not observe evidence that this pattern was the result of divergent selective pressures. Rather, the city agouti squirrel microbiome displayed phylogenetic patterns more consistent with stochastic processes than black squirrels. Given the putative importance of a westernized diet in shaping the urban wildlife microbiome, colour polymorphic grey squirrels provide an interesting free-living system with which to study the gene x diet x microbiome interactions. Further research and a full holo-omics approach (integrated host and microbiome genomics, transcriptomics, proteomics, and metabolomics [26]) are required to probe the causes and fitness consequences of metagenomic plasticity and understand its importance for host species as they endeavour to colonize a new ecological niche.

Methods

Study sites

Sampling was divided among three site pairs throughout southern Ontario. Site pairs were comprised of one city and one nearby rural deciduous forest outside of city limits. Each city and rural site within a pair were trapped consecutively to limit the effect of seasonal confounds. Cities are inherently environmentally heterogeneous, therefore, to standardize our efforts, we limited sampling to urban university campuses as our representative city sites. Those sites were the University of Guelph (43°31′52.45″N 80°13′36.70″W, n = 29 squirrels), University of Waterloo (43°28′17.68″N 80°32′36.03″W, n = 30), and University of Windsor (42°18′23.77″N 83°4′1.92″W, n = 15), which were paired with forests at Starkey Hill Conservation Area (43°32′37.99″N 80°9′15.82″W, n = 20), the rare Charitable Research Reserve (43°22′57.72″N 80°20′54.43″W, n = 26), and the Kopegaron Woods Conservation Area (42° 4′38.52″N 82°29′34.04″W, n = 2), respectively. To understand how habitat heterogeneity within cities can affect the microbiome, we also sampled other urban land classes within the City of Guelph: two suburban parks (Exhibition Park, n = 29; Royal City Park, n = 13) and two urban forests (University of Guelph Dairy Bush, n = 11; Arboretum, n = 10)—small forest fragments entirely enveloped by an urban landscape.

Capture protocol

All trapping occurred in southern Ontario between May 9th and September 12th using tomahawk Model 102 traps (Tomahawk Live Trap Co., WI, USA) baited with peanuts. Traps were set between 6:00–16:00 and checked in 1 h intervals. Upon capture, we transferred squirrels to a cloth bag and checked in 15 min intervals for faecal pellets. Using single-use sterilized toothpicks, we collected fresh faecal pellets which we stored on ice in the field before transfer to -20 °C for storage within 4 h of collection. After faecal pellet collection, we transferred squirrels to a handling bag, recorded sex, and marked squirrels with unique alpha-numeric ear-tags before release to allow for identification of recaptures and prevent resampling of individuals.

Sequencing and bioinformatics

Using QIAamp DNA Stool Mini Kits (Qiagen, Hilden Germany), we extracted bacterial DNA from 0.2 g subsamples of faecal samples, which in other rodents are indicative of bacterial communities in the large intestines [124]. Extracts underwent triplicate PCR amplification of the 16S rRNA gene v4 region (515F-806R modified primers; [125]) at MetaGenomBio Inc. (Toronto Canada) alongside PCR negative controls. Triplicate PCR products were then pooled prior to sequencing on an Illumina MiSeq using v2 chemistry (250 bp read). We then processed pair-end reads in mothur using a standardized amplicon processing pipeline [126], aligned sequences to the silva v132 reference database [127], and classified operational taxonomic units (OTUs) clustered using OptiClust [128] based on a 97% similarity threshold. To remove potential contaminants or sequencing errors, we discarded all non-bacterial OTUs and OTUs which were not represented by at least 1 read in 5% of the samples prior to analysis [113]. In total, we generated 3,826,931 merged bacterial amplicon sequences, with an average sequencing depth of ~ 20,000 reads/sample (min = 9724, max = 60,891), after assembly, quality control, and filtering. Samples were rarefied to a depth of 9724 reads (Supplementary Information: Fig. S5), except for ANCOM-BC analysis and analyses pertaining to Euclidean measures of β-diversity which were calculated from raw count # centred log ratio transformed OTU datasets [129]. A relaxed neighbour-joining method was used to construct a phylogenetic tree using the mothur implementation of clearcut [126, 130]. We sequenced extraction kit negative controls which contained swabs of the bags which were used to cover the tomahawk traps. However, negative controls showed nominal amplification and most negative control OTUs were removed during the filtering procedures described above. The remaining OTUs were present at low abundance in negative controls but were abundant among biological samples and therefore likely the result of index hopping.

Diversity testing

We tested whether microbiome α-diversity (# of OTUs and Shannon diversity) differed with sex, coat melanism mutation number (agouti = 0, intermediate = 1, black = 2), environment (city versus rural), and an interaction between coat melanism and environment using linear mixed effects models in the R package ‘lmerTest’ [131], treating sampling location as a random effect. To determine whether bacterial microbiome β-diversity differed between city and rural eastern grey squirrel populations, we used permutational multivariate analyses of variance (PERMANOVA) tests of β-diversity in Euclidean and weighted UniFrac space. PERMANOVAs were parameterized by environment type (city versus rural), land class (campus, suburban park, urban forest, rural forest), sampling location, sex, coat colour phenotype, and an interaction between phenotype and environment type, in that order. To test for differences in environment-phenotype groupings, we used permutation tests of multivariate homogeneity, with β-dispersion calculated relative to the group centroid with a bias adjustment for small sample sizes.

Community null modeling

To determine whether observed patterns of microbiome diversity deviated from stochastic expectations, we used a null modelling approach [53, 54, 132]. These methods are predicated on species’ ecological niche-spaces being non-independent of their phylogeny [55], a notion broadly supported via analysis of functional traits derived from publicly archived microbial genome sequences [133, 134]. We tested for a positive phylogenetic signal in urban associated niche-space using the phyloCorrelogram() function from the R package ‘phylosignal’ [135]. OTU niche space was estimated as the deviation of an OTU from a 1:1 line of mean OTU relative abundance in city versus rural habitats (Supplementary Information: Fig. S4).

In brief, we calculated mean nearest taxon distance (MNTD) and β mean nearest taxon distance (βMNTD), which are among community and between community measures of phylogenetic dissimilarity, respectively. For all analyses, these measures were effect size standardized by the mean and standard deviation of MNTD or βMNTD null distributions created through stochastically assembled communities which possessed the same α-diversity as the focal observed communities [53, 54]. Null distributions were created by randomly shuffling taxa names and relative abundances across the system’s γ-diversity phylogenetic tree (MNTD: 9999 iterations, βMNTD: 999 iterations) using the function ses.mntd() from the package ‘picante’ [136] and ses.comdistnt() from the package ‘MicEco’, respectively [137]. Additionally, we calculated RCbray values, a phylogeny-independent measure of between community OTU turnover [54, 58]. In brief, observed Bray-Curtis values were compared to 9999 probabilistically assembled community pairs of the same α-diversity as observed community pairs [54].

A linear mixed effects model was used to test for an effect of sex, environment, and colour phenotype, as well as a phenotype x environment interaction on |MNTD|, which we used as a measure of among community deviation from stochastic expectations. Again, sampling location was included in the mixed model as a random effect. To determine whether the phylogenetic structure between communities were more similar (βMNTDses < 0) or more disparate (βMNTDses > 0) than expected by chance alone (|βMNTDses| > 2), we used a PERMANOVA (parameterized as in the β-diversity analyses described above). As above, post-hoc permutation tests of multivariate homogeneity were used to parse significant relationships. Dispersion in βMNTDses values were calculated relative to the group centroid with a bias adjustment for small sample sizes [138], using the ‘vegan’ function betadisper() [139].

RCbray values were analyzed using a PERMANOVA, as described above. To make within and between group comparisons, we used post-hoc Kruskal-Wallis and dunn tests to further parse results into within versus between trapping sites. We used an ultrametric phylogenetic tree transformed using the chronos() function in the R package ‘ape’ (λ = 1 [140];) for all phylogeny-weighted null modelling. Unless specified, all other analyses were completed in R (v. 3.5.1) using the R package ‘phyloseq’ [141].

Differential abundance testing

To identify how the bacterial microbiome differed between squirrels in forests versus the built-environment, we performed selection balance analyses after binning OTUs to genus and family [59]. Secondarily, we performed ANCOM-BC differential abundance tests and evaluated the agreement between these opposing statistical approaches [60].