Gut microbiome composition is associated with spatial structuring and social interactions in semi-feral Welsh Mountain ponies

Background Microbiome composition is linked to host functional traits including metabolism and immune function. Drivers of microbiome composition are increasingly well-characterised; however, evidence of group-level microbiome convergence is limited and may represent a multi-level trait (i.e. across individuals and groups), whereby heritable phenotypes are influenced by social interactions. Here, we investigate the influence of spatial structuring and social interactions on the gut microbiome composition of Welsh mountain ponies. Results We show that semi-feral ponies exhibit variation in microbiome composition according to band (group) membership, in addition to considerable within-individual variation. Spatial structuring was also identified within bands, suggesting that despite communal living, social behaviours still influence microbiome composition. Indeed, we show that specific interactions (i.e. mother-offspring and stallion-mare) lead to more similar microbiomes, further supporting the notion that individuals influence the microbiome composition of one another and ultimately the group. Foals exhibited different microbiome composition to sub-adults and adults, most likely related to differences in diet. Conclusions We provide novel evidence that microbiome composition is structured at multiple levels within populations of social mammals and thus may form a unit on which selection can act. High levels of within-individual variation in microbiome composition, combined with the potential for social interactions to influence microbiome composition, suggest the direction of microbiome selection may be influenced by the individual members present in the group. Although the functional implications of this require further research, these results lend support to the idea that multi-level selection can act on microbiomes. Electronic supplementary material The online version of this article (10.1186/s40168-018-0593-2) contains supplementary material, which is available to authorized users.


50
All metazoan species harbour complex communities of microorganisms referred to as host 51 microbiomes. The host plus its microbiome complement can be considered as a distinct biological 52 entity, the holobiont, with a complementary genome, the hologenome [1]. Although the concept of a

5
Here we determine how the gut microbiome of semi-feral ponies from Snowdonia National Park is 110 influenced by spatial structuring, social interactions and kin relationships. Using social network 111 analysis combined with 16S rRNA gene amplicon sequencing of faecal samples, we test the following 112 hypotheses; i) there will be within-individual variation in microbiome composition, but this will not be as 113 large as between-individual variation; ii) mares will have more similar microbiomes to band stallions 114 than to other mares in their band; iii) mares will have more similar microbiomes to their own offspring 115 than to other juveniles in the band; iv) band, life-stage and sex will influence microbiome composition; 116 v) band-level variation in microbiome composition will be driven by spatial structuring (i.e. social

125
The population is essentially unmanaged aside from an annual roundup event in November, during 126 which individuals are herded onto adjacent farmland for one to two days for population management 127 purposes. Individuals can be identified using their age-sex classification and a photographic database 128 that depicts coat colour, face and leg markings and ear tags/notches. For this study, we collected data

139
We approximated the distance in metres between individuals. All individuals less than ~100 m apart 140 and moving as a cohesive unit were considered to be associated with each other [40]. Association 141 matrices were constructed for each day of sampling; individuals that were close together (< 15 m) or 142 interacted were given a score of 2, other individuals (i.e. those 15 -100 m apart) were given a score 143 of 1 and more than 100 m apart scored 0. Using these association scores, an overall weighted  (Table 1 and Table S1) (Table S1). Samples were stored and transported in cool bags to the University of

163
Manchester the same day and frozen at -80°C prior to DNA extraction.

219
To obtain the "average microbiome" for an individual, we merged raw sample data within an individual 220 using the merge_samples function in phyloseq (using "fun=mean") [49]. To determine whether there 221 was greater microbiome variation between bands than within bands, we calculated Jensen-Shannon

222
Divergence (JSD) values between individuals, using data from their average microbiome, as described 9 above. We used a one-way ANOVA with Tukey's posthoc analysis to compare JSD distances within 224 and between bands, and visualised the data using a box plot.

225
We produced NMDS plots in phyloseq using the Bray-Curtis distance matrix to visualise differences in 226 beta diversity according to band and life-stage. We used an adonis analysis to test for significant 227 effects of band, life-stage and sex on total microbiome community composition. We then calculated 228 the core microbiome as described above and repeated the adonis analysis for this core community.

229
Additionally, we agglomerated the core taxa to genus level and visualised the core microbiome as a 230 heat map to give a representation of the bacterial taxa present.

231
To identify differences in microbiome composition between foals (which at approximately 5-8 months

254
An adonis analysis showed pony ID had a significant effect on total microbiome composition (p < 255 0.001; Table 2), with 52.6% of the variation in the microbiome attributable to individual variation (Fig.   256 1). We obtained similar results for the adonis with the core microbiome (p < 0.001; Table 2

262
There were significant effects of band and life-stage on total microbiome composition, but not sex 263 (

270
Consistent with the spatial distribution of the bands (Fig. 4a), microbiome composition of individuals in

271
Valley differed considerably to those in Aber and Marsh, which are more similar to each other but still 272 display some degree of separation (Fig. 3a). There was a significant difference in JSD metric values 273 within and between the bands (F5,424 = 6.557, p < 0.001), and the Tukey posthoc indicated that within-274 band variation for Valley was significantly lower than the variation within the other two bands, and 275 significantly lower than between-band variation for all three combinations (Figure 2b). In addition to 276 this band-level differentiation of microbiomes, there was a significant correlation between social 277 network tie weight (i.e. spatial distribution) and microbiome composition (τ = -0.11, p < 0.001) within 278 bands, such that individuals that associate more have more similar microbiomes (Fig 4b and 4c).

11
The microbiome of foals was considerably different to that of sub-adults and adults, whereas these 280 latter two groups were very similar to one-another (Fig 3b and S2).

12
Band membership also predicted microbiome composition, with ~14% of the total microbiome 308 variation and ~19% of the core microbiome variation explained by this factor. That the microbiome 309 composition of ponies belonging to Aber and Marsh are more similar to one another than Valley may 310 be driven by both spatial structuring and diet, given that the home ranges of these bands overlap. The 311 home ranges of Aber and Marsh are also somewhat different to that of Valley in terms of elevation, 312 slope and soil moisture; these are more low-lying and marshier in comparison to the steeper, more 313 well drained and exposed slopes that characterised the home range of Valley during the study period.

314
The type and quality of grasses or forage across the study area (approximately 5km 2 ) also vary 315 according to habitat type and thus, diet quality may be driving the observed differences in bands. In

319
The microbiome of Equidae is highly susceptible to changes in diet with consequences for nutrient

331
Although it may be difficult to dissociate between the influence of shared living and diet on microbiome 332 composition between bands, spatial structuring was also identified within bands, suggesting that

385
Here we show that semi-feral ponies exhibit variation in microbiome composition between bands, 386 which may relate to social, dietary and environmental factors. In addition, due to the high level of 387 within-individual variation, the direction of selection may be influenced by the individual members 388 present in the group. Spatial structuring was also identified within bands, suggesting that despite 389 communal living, social behaviours still influence microbiome composition. We identify two such 390 interactions; mother-offspring and stallion-mare, that lead to more similar microbiomes, indicating that 391 individuals influence the microbiome composition of one another and ultimately, the group. Thus, we 392 provide novel evidence that microbiome composition is structured at multiple levels within populations.

393
The functional implications of this requires further research.     Table 1 605 Demographic data for each band used in this study. Table 2 608 Statistical outputs for microbiome adonis analyses.