Community richness and structure at multiple thresholds of genetic similarity
Superficial layers tend to have higher richness than their corresponding deep soil layers across all four habitats, with significant richness differences between soil layers found for thermophilous woodland and pine forest (Fig. S1). After combining superficial and deep soil layers for all 52 sites, mean site richness (α diversity) within habitats ranged 55 - 73.5 for haplotypes, 38.5 - 49 for 3% clusters and 34.5 - 43 for 15% clusters (Fig. 2A). Differences in richness by sample among habitats were small and maintained across different clustering thresholds, and pointed to dry scrubland community samples as poorer (lower richness by site) compared to the other habitats (Fig. 2A).
Mean endemicity by site (proportion of lineages that occur exclusively in that site) ranged from 24.0% to 48.8% at the haplotype level, from 13.5% to 22.7% for 3% clusters, and from 6.8% to 15.4% for 15% clusters (Fig. 2B). Comparisons among habitats revealed that endemicity was significantly higher for dry scrubland communities than for laurel forest communities (Fig. 2B). Compositional dissimilarity among communities (β diversity, βsor) was high and was dominated by lineage turnover (βsim), rather than nestedness (βsne), with βsor values ranging 0.87-0.96 across all clustering levels and habitats. Dry scrubland communities showed the highest levels of compositional dissimilarity across the different clustering thresholds (Fig. 2C).
Total observed richness at the island scale (ɣ diversity) by habitat ranged from 534 - 588 haplotypes, 278 - 316 lineages at 3% and 194 - 255 lineages at 15% (Fig. 2C), while extrapolated values (Chao index) nearly doubled observed values (Fig. 2D). Differences in ɣ diversity among habitats were not consistent across different clustering thresholds, with thermophilous woodland showing the lowest number of haplotypes but the highest number of lineages at the 15% clustering threshold (Fig. 2C). Accumulation curves reveal no plateau in the accumulation of entities across samples for any habitat or genetic threshold, with the laurel forest showing the lowest rates of accumulation (Fig. 2D).
Comparisons with biodiversity measures obtained by Arribas et al.(2020) in forest and grassland sites in a continental setting revealed that richness by sample (α diversity) was lower in the samples of Tenerife compared with continental soils (Kruskal p < 0.001; Fig. S2). Comparisons of β diversity values restricted to a comparable spatial scale of 15 km resulted in significantly lower β diversity values in Tenerife for haplotypes (p < 0.001) but not for 3% OTUs (Fig. S2). Finally, ɣ diversity by sampling region, as estimated by the total number of haplotypes and OTUs recorded, was similar for the different habitats of Tenerife (534 – 588 haplotypes and from 278 - 316 3% OTUs) and the six continental settings in Arribas et al. (2020) (558 - 623 haplotypes, and 276 – 319 OTUs) (Fig. S2).
NMDS for the compositional dissimilarity of the communities of Tenerife soils showed habitat as a major driver of the ordination of samples, and accordingly, for all clustering levels, a significant proportion of variance (0.18 < r2 < 0.28; p < 0.001) was explained by the habitat factor (Fig. 3A). In addition, dry scrubland communities showed the highest dispersal, while the laurel forest communities were the least scattered (Fig. 3A).
Analyses of community similarity (1-pairwise β diversity) with spatial distance within each habitat revealed significant distance decay for all clustering levels in all habitats, except for dry scrubland (Fig. 3B). For laurel forest, pine forest, and thermophilous woodland, slopes of the exponential decay curves were very similar at all threshold levels, and assemblage similarity increased with each level (Fig. 3B). Genetic similarity showed a high and significant log-log correlation with the number of lineages (0.97 < r2 < 0.99, p < 0.001), initial similarity (0.92 < r2 < 0.99; p < 0.001), and mean similarity of communities (0.97 < r2 < 0.99; p < 0.001) (Table S5), as expected if community variation across hierarchical levels of similarity is described by a fractal geometry (Baselga et al., 2013, 2015).
A decrease in community similarity with environmental distance (Fig. S3) was only significant for the laurel forest and some clustering levels in the pine forest (Table S6). However, variance partitioning showed that variance uniquely explained by environmental distance (i.e. independently of the spatial distance) was lower (3.2% - 9.0% of explained variation at all levels) than the uniquely explained variance by the spatial distance (6.9% - 45.0% of explained variation).