Community composition and indicator species

MOTU community composition differed significantly between treatments. The NMDS analyses retrieved final stress values of 0.21 and 0.23 for the ordinations according to COI and 16S MOTUs, respectively (Figure 5). For both markers, the first axis of the NMDS ordination (Axis 1) identified a gradient between control trees and one- and ten-years-excluded farms, with a clear segregation of the latter (long-term exclosures) with respect to the other two. Accordingly, correlation analyses showed that the main factor conditioning the community differences for COI and 16S markers was Treatment (r2 = 0.36; p = 0.001 and r2 = 0.32; p = 0.001 respectively) and, only for the 16S marker, the vegetation structure. In the latter, both PCA factors were significantly correlated with species composition, being the first one related with long-term exclosures, which also have a high cover of tall grass (PC1: r2 = 0.16; p = 0.002), and the second with control and one-year excluded trees in which shorter, medium-sized grass is more prevalent (r2 = 0.09; p = 0.039) (Figure 5B).
The species retention rate curves (Figure 6) showed that the long-term-exclusion trees were not only the most different compared with the other treatments, but also the most different among themselves, followed by the controls and the one-year-exclusion trees. This pattern was consistent with both datasets, albeit COI (Figure 6A) detected less species in common between samples than 16S (Figure 6B). As an example, out of the 24 trees in each of the three treatments, with COI, there was not a single MOTU that was present in more than 5 trees in the long-term exclosures (Figure 6A). The trend in both markers indicates that the trees under the one-year-exclusion regime were colonized by mostly the same arthropod species in many cases, while the communities in the long-tern-exclusion trees had diverged towards different species assemblages. In all cases, the models better fit to an exponential than to a power-law regression, indicating that community assembly within the treatments did not respond to niche differentiation, but rather to stochastic processes (McGeoch et al., 2019). When looking at the species retention rate curves on each farm separately, the pattern was largely maintained in relation to Treatment, although with some exceptions (Figure S3).
With regards to the indicator species analysis (Table 2, Figure 7), we identified 15 indicator MOTUs with COI and 17 with 16S. From the 15 MOTUs identified as indicators with COI, one corresponded to the control group, eight to one-year exclusion, and six to the long-term exclosures (Table 2, Figure 7A, Table S7). With 16S, on the other hand, we found no indicator MOTU for the control group, while eight were found for the one-year-exclusion group, and nine for the trees in which livestock had been excluded in the long-term (Table 2, Figure 7B, Table S7). Most MOTUs identified as indicators for the abandoned farms were herbivores and some predators, while the ones from the one-year excluded trees included a mix of herbivores (mostly), saprophagous, predators and parasitoids. In general, indicator species for one-year exclusion communities can be found abundantly in communities from other treatments, while indicator species for ten-year exclusion communities are almost exclusively found in these communities (Figure 7). Interestingly, with 16S we identified the aphid Sitobion avenaeas an indicator for the one-year-exclusion communities, and with COI we identified its predator, the ladybug Ryzhobius sp. , as an indicator for the same communities. In most cases (13 out of 15 for COI, and 11 out of 17 for 16S), the MOTUs identified as indicators for any treatment are also among the most abundant MOTUs in their respective datasets (Figures S3-4).