4 | Discussion
Accounting for spatial processes in our system was integral to explaining observed patterns of microbiome variation. Longitude explained variation in almost every microbiome diversity measure considered. Alpha diversity declined from west to east, and longitude was correlated with both Euclidean and weighted UniFrac measures of β-diversity. Unmeasured environmental variables may account for these patterns; however, plant communities representing the Sable Island horses’ primary forage were present in our analyses. Furthermore, if environmental selective pressures acting on the microbiome were spatially autocorrelated, we would have expected co-occurring horse microbiomes to be more phylogenetically similar, and distant horses to be more phylogenetically disparate, than expected by chance. Contrary to these expectations, we did not observe a positive correlation between βMNTDses and the distance separating horses. Further, the microbiomes of horses in the west and towards the longitudinal extremes of the island tended to be further from the population average. The significance of spatial terms in PERMANOVA analysis of Euclidean distances may therefore derive partly from differences in community variance rather than differences in average community structure.
Host genetics, and thus indirectly the physiological environment with which microbes directly interact, might explain some of the spatial variation in microbiome variance. Based on microsatellite data, Sable Island horse heterozygosity is higher in the east (Lucas, McLoughlin, Coltman, & Barber, 2009) which is where we also observed lower microbiome alpha diversity and beta-dispersion when compared to horses in the west. Evidence from captive and wild mammalian systems has shown microbiome alpha diversity to be negatively correlated with host heterozygosity (Grosser et al., 2019; Wadud Khan, Zac Stephens, Mohammed, Round, & Kubinak, 2019). Similarly, an effect of population-level heterozygosity has been reported on the bacterial microbiome of free-living bighorn sheep (Couch et al., 2020). The homozygosity implicit of inbred hosts might restrict their immunological complexity (Potts & Wakeland, 1993; Reid, Arcese, & Keller, 2003), thereby also restricting the dexterity with which host’s recruit and “leash” (see Foster, Schluter, Coyte, & Rakoff-Nahoum, 2017) their microbial communities, perhaps allowing for greater stochastic variation between individuals. Alternatively, Fst values suggest population sub-structuring between the east and west (Lucas et al., 2009), therefore genetic differences between horses might explain why the microbiome differs across the island’s length. For example, among free-living house mice genetic similarity along a latitudinal gradient was a better predictor of microbiome similarity than spatial proximity (Suzuki et al., 2019). Genetic variation among Sable Island horses expressed as phenotypic variation could therefore drive microbiome variance across the island’s longitude (Alberdi, Aizpurua, Bohmann, Zepeda-Mendoza, & Gilbert, 2016).
While we cannot rule out a role for host genetics, in the present absence of data informative for testing this, bacterial dispersal limitation between horses provides the most parsimonious explanation of observed patterns. For example, we observed an apparent positive correlation between the proximity of horses and the similarity of their microbiome in Euclidean space (independent of similarities in local environment). A similar relationship was observed with respect to weighted UniFrac distances, however no positive relationship was observed among phylogeny-informed null modeling approaches (βMNTDses). Assuming bacterial niche space and phylogeny are non-independent, these patterns suggest that the decrease in microbiome similarity with spatial separation was not due primarily to differences in selective pressures across space. Conversely, the positive relationship between spatial separation and RCbray values suggest dispersal limitation may occur over relatively short spatial scales. Evidence for dispersal limitation may be unsurprising given a zero-inflated ASV count table; of 3767 ASVs, only 2 were detected in all horses and only 441 were present in at least half of the horses.
In addition to a positive correlation with spatial separation, RCbray values were negatively correlated with average longitude, suggesting greater dispersal limitation among horses in the west than the east. This was unexpected since horse population density, which could facilitate bacterial dispersal between individuals, decreases from west to east (Marjamäki et al., 2013). However, while multiple above-ground ponds can be found in the west, horses in the east must crater through sand to access freshwater (Contasti et al., 2012). Horse-excavated wells are semi-permanent within a season and visited by multiple social bands but are only accessible to 1–2 horse at a time (Figure 2D). Prolonged occupancy of an area of social band overlap, and bottlenecked access to a communal consumable resource, could catalyze bacterial dispersal despite low population densities in the east. Similar host aggregation due to patchy resource distribution on urban landscapes facilitates disease transmission in wildlife (Bradley & Altizer, 2007); the same aggregative effect could as easily facilitate transmission of commensal and mutualistic microbiota.
Bacterial dispersal between horses undoubtedly occurs; however, it may be largely restricted to between individuals within the same, or closely interacting, social bands. Social band membership was correlated with both Euclidean and weighted UniFrac β-diversity; however, microbiome phylogenetic diversity (βMNTDses) was no more similar between members of the same band than between members of different bands when compared to null expectations, offering little support for homogenizing selection as the mechanism for the effect of band membership on the microbiome. RCbray values, which were lower between members of the same band than between horses of different bands, suggests bacterial dispersal limitation as a primary cause for the observed effect of social band. This interpretation is consistent with Antwis, Lea, Unwin, & Shultz (2018) who report an effect of band identity and inter-band connectivity on microbiome β-diversity among three large social bands of feral Welsh ponies. Similar differences in band connectivity might explain why, above and beyond parameterized environmental terms, distance from the population’s centre was correlated with Euclidean β-diversity and β-dispersion. No relationship was observed with respect to βMNTDses but RCbray values were positively correlated with average distance from the centre of the population. Horses on the edges of the population—those more poorly connected within the population’s microbiome meta-community (Miller et al., 2018)—might be vulnerable to erosion of microbiome diversity through microbial extinctions and exacerbated ecological drift. Together these results support recent suggestions of the importance of inter-host dispersal in maintaining the ‘social microbiome’ in free-living populations (Sarkar et al., 2020).
Phylogeny-informed measures of diversity were often better explained by local plant community composition than spatial terms. Horses with sandwort in their 150-m radius buffer had lower alpha diversity and differed in both phylogeny-informed (Euclidean) and phylogeny-independent (weighted UniFrac) β-diversity measures. The intuitive explanation is that plant community availability reflects dietary composition, and different dietary components differ in the functions required to metabolize component polysaccharides (David et al., 2014; Julliand & Grimm, 2017). However, among pairwise comparisons in which sandwort was present for at least one horse, microbiomes were no more phylogenetically disparate than expected by chance (higher βMNTDses). By comparison, the microbiomes of horses without access to sandwort tended to be more phylogenetically similar. Conversely, grassland and beach pea dominated habitat classes were negatively correlated with βMNTDses, while heath (only present where sandwort was absent) appeared positively correlated with βMNTDses. Therefore, phylogenetic patterns most consistent with homogenizing selection on the microbiome were observed when sandwort and heath were absent, but beach pea and marram grass were abundant; with reversed conditions, phylogenetic similarities did not deviate far from stochastic expectations.
Increased evidence for stochasticity may stem from the fact that sandwort, forbs and small graminoids—the primary horse forage in heathland habitat—possess lower neutral detergent fibre (NDF) when compared to beach pea and marram grass (personal communication K. Johnsen; Lee, 2018). NDF is a coarse measure of plant lignin, hemicellulose, and cellulose (Mongeau & Brassard, 1982); compounds which many herbivores are obligately reliant on their gastrointestinal microbiota to metabolize (Costa & Weese, 2012). The low NDF observed in sandwort and heathland forbs may alleviate the horses’ reliance on their intestinal microbiota, allowing them to directly absorb nutrients from a relatively labile diet. Loss of dietary complexity constrains fibrolytic and cellulolytic niche-space in the microbiome which can manifest as reductions in bacterial gene richness (Cotillard et al., 2013) or alpha diversity (Schnorr et al., 2014). Conversely, high fibre forage (e.g. marram grass and beach pea) can facilitate complex microbial symbioses in which different species specialize on metabolizing different biochemical compounds, and in doing so, create by-products to be absorbed by the host or further metabolized by other microbiota (Oliphant & Allen-Vercoe, 2019). The reduction in alpha diversity observed in horses with access to sandwort mirror the effects of low dietary fibre manipulations in domestic horses (Julliand & Grimm, 2017). When compared to with marram grass and beach pea, sandwort might represent a reduction in the carbon source complexity accessible to the microbiome, a property thought to have a stabilizing effect on the microbiome (Coyte et al., 2015). A diet containing sandwort might not select for different microbial functions, so much as fail to support fibrolytic niche-space supported by high fibre diets, leading to species extirpation and greater ecological drift within individual host microbiomes (Deehan & Walter, 2016). This could also explain the greater variability in weighted UniFrac β-diversity among horses with access to sandwort and the decrease in dispersion in response to beach pea availability. These results highlight how dietary derived microbiome variation might not always be the result of strong differential selective pressures between communities; the relationship between dietary complexity and ecological drift must also be considered (Adair & Douglas, 2017; Zhou & Ning, 2017).
Parental status was more strongly correlated with measures of microbiome variance, rather than mean community structure. Specifically, mares with foals had microbiomes which were a) more diverse, b) marginally less variable in weighted UniFrac space, c) less randomly phylogenetically dispersed (greater |MNTDses|), and d) further from phylogenetic null expectations of random community assembly (higher |βMNTDses|) when compared to mares without foals. Effects of parturition and maternal status on microbiome alpha and β-diversity have been observed in livestock (Lima et al., 2015) and wildlife (Amato et al., 2014). Although, to our knowledge, a difference in β-dispersion between parental states has not previously been reported. Myriad changes to maternal physiology during pregnancy and parturition are likely partly responsible for microbiome differences during birth and child-rearing (Huang et al., 2019; Nuriel-Ohayon, Neuman, & Koren, 2016). In addition to these physiological changes, maternal care among mammals, especially lactation, saddles mothers with a heavy energetic burden (Dufour & Sauther, 2002; Scantlebury, Russell, McIlrat, Speakman, & Clutton-Brock, 2002). To meet higher energetic demands, hosts may become increasingly reliant on their microbiomes (Amato et al., 2014); especially in species, such as horses, which are obligately reliant on their gut microbiomes for nutrient uptake (Costa & Weese, 2012). Therefore, during periods of high energetic demand hosts might be forced to enforce stronger control on the microbiome to maximize metabolic efficiency. For example, in laboratory mice, postpartum dampening of bi-directionality in the host-microbiome relationship is evidenced by attenuated bacterial driven immunomodulation (Mu et al., 2019). We suggest that hosts facing a high energetic burden might keep their microbial constituents on a “tighter leash” than those with a lower energetic demand (Foster et al., 2017). Within host species, host physiological variation might in many cases act to facultatively constrain β-dispersion, rather than drive changes in β-diversity, although patterns of the former are often overlooked (Zaneveld et al., 2017). The reverse causal relationship could also explain the patterns observed, whereby a diverse microbiome under tight host control signals better host health and therefore greater likelihood of carrying a foal to term.
Overall, the bacterial microbiome of Sable Island horses is dominated by clades of fibrolytic taxa, including Ruminococcaceae, Lachnospiraceae, Prevotellaceae, and Fibrobacteraceae (Biddle, Stewart, Blanchard, & Leschine, 2013; Esquivel-Elizondo, Ilhan, Garcia-Peña, & Krajmalnik-Brown, 2017; Spain, Forsberg, & Krumholz, 2011). Spirochaetaceae and Kiritimatiellae are also present at modest relative abundances; however, their metabolic niches are currently less well characterized. These results are consistent with findings from domestic, feral, and wild horse systems (Antwis et al., 2018; Costa et al., 2015; Metcalf et al., 2017) and a comprehensive comparison of wild and domestic equid species (Edwards et al., 2020). Unlike previous studies, however, we detected no effect of age, likely because we constrained sampling to horses of at least 3 years of age, and the horse microbiome appears to reach maturation after ~1 year (Antwis et al., 2018; De La Torre et al., 2019; Metcalf et al., 2017).
We characterized the bacterial microbiome of 86 mares from the feral horse population of Sable Island (Nova Scotia, Canada) and contrasted the ability of spatiotemporal, physiological, and diet-linked environmental variables to explain microbiome variation. Phylogeny-independent measures of diversity were best explained by spatial variables while phylogeny-informed measures were generally better characterized by host physiology (parental status) and measures of local habitat heterogeneity; however, despite statistical significance, these variables explained only nominal variation in overall β-diversity. Only the longitudinal distance separating horses and social band membership explained what could be considered substantive variation, and yet, much of the variation in the Sable Island horse microbiome remained unexplained. In context, our results suggest a predominant importance of bacterial dispersal and ecological drift in shaping faecal microbiome variation among Sable Island horses. Our findings are relevant to the study of wildlife microbiome variation: clearly data on the spatial distribution of hosts should be collected, even at the within-population scale, alongside metrics of individual-based environmental variation. Further, when a response of the microbiome to environmental or physiological variation is observed, deterministic processes must not be assumed as the sole causal process.