3.4 | Ecological Null Modeling
We detected a positive signal over short phylogenetic distances in relation to sandwort availability (p < 0.05, r = 0.02; Figure S3, Supporting information). Of the physiological, environmental, and spatial terms considered, only parental status (p = 0.03) was associated with non-random patterns of phylogenetic dispersion. Namely mares with foals had higher |MNTDses| values (Table S6, Supporting information). Based on between-sample comparisons, communities were more often phylogenetically conserved (βMNTDses < 0) than they were phylogenetically disparate (βMNTDses> 0) but usually did not deviate in expected phylogenetic similarity from pairs of randomly assembled communities (|βMNTDses| < 2).
βMNTDses values were correlated with Julian date (R2 = 0.03, p < 0.01), sandwort presence (R2 = 0.07, p < 0.01), beach pea availability (R2 = 0.03,p = 0.02), heathland availability (R2 = 0.03, p = 0.01), and grassland availability (R2 = 0.02, p = 0.04; Table S7, Supporting information). In the absence of sandwort, βMNTDses values appeared to be negatively correlated with average grassland availability (Figure 5), but positively correlated with average heath availability (Figure S4A, Supporting information); conversely, βMNTDses values were greater when sandwort was present and appeared to be negatively correlated with average Julian date. In contrast, the absolute magnitude of phylogenetic deviation from stochastic expectations (|βMNTDses|) was correlated with beach pea availability (R2 = 0.05, p< 0.01), longitude (R2 = 0.05, p = 0.04), and parental status (R2 = 0.03, p= 0.03; Table S8, Supporting information). Specifically, |βMNTDses| values appeared to be positively correlated with beach pea availability (Figure S4B, Supporting information) as well as average longitude, and greater among mares with foals than mares without foals (Figure S4C, Supporting information). No significant correlation was observed between longitudinal separation and βMNTDses after controlling for sandwort presence (partial Mantel test: p = 0.68). Similarly, no effect of band membership on βMNTDses was observed (PERMANOVA: p = 0.45).
Approximately 14% of βMNTDses were beyond 2 standard deviations of the randomized null distributions. Of the remaining ~86% of pairwise comparisons, ~97% had corresponding RCbray values exceeding 0.95, which signals greater ASV turnover than expected under ecological drift alone, a pattern suggestive of dispersal limitation. Based on PERMANOVA analysis, RCbray values were correlated with longitude (R2 = 0.02, p < 0.01), distance from the centre of the population (R2= 0.02, p < 0.01), sandwort presence (R2 = 0.01, p = 0.04), beach pea availability (R2 = 0.01, p = 0.04), and Julian date. (R2 = 0.02, p < 0.01; Table S9, Supporting information). Additionally, in a univariate PERMANOVA, band membership was significantly correlated with RCbrayvalues (R2 = 0.65, p < 0.01). Furthermore, RCbray values were positively correlated with the distance separating horses even after controlling for βMNTDses values (partial Mantel test:rpearson = 0.17, p < 0.01; Figure 6A), but negatively correlated with average longitude and lower among members of the same band than between members of different bands (Figure 6B).