Results
Molecular diet analysis Of the 41 faecal samples collected at 200m and 2000m elevations, we were able to amplify and sequence prey DNA from 33 samples in total, from five bird species at low-elevations and nine different bird species at mid-elevations (Supplementary table S2) and all 25 weaver ant food samples. We were also able to amplify and sequence prey DNA from 36 additional bird faecal samples at higher elevations in Sikkim and 2 additional samples at 1200m elevation. Even though the primer set we used was supposed to be invertebrate-specific, it amplified some vertebrate taxa such as Squamata as well (Supplementary table S3). We recovered 1331 amplicon sequence variants (ASVs) from the dada2 pipeline. Of these, 1072 sequences yielded BLAST matches which reduced to 980 sequences after filtering for bacteria and contaminants (162 from birds at the 200m elevation, 224 from birds at the 2000m elevation, 325 from weaver ants and 333 from birds at 1200m, 2300m, 2700m and 3200m elevations; note that this is greater than 980 because of overlap in ASVs among these groups).
The most frequent orders in the bird diet at all elevations were Lepidoptera and Coleoptera (Fig. S3 & Supplementary table S3). At 200m, Lepidoptera was detected in 94% and Coleoptera in 88% of the samples. Molecular diet analyses confirmed that Lepidoptera (69%) and Coleoptera (48%) are common in weaver ant diet, in addition to Blattodea, Hemiptera and Hymenoptera (all 61%). Each of the orders present in weaver ant diet was also found in bird diets (Fig. 2 & Fig. S4, Supplementary tables S3-S6) but birds had consumed animals in nine additional orders. These included larger animals such as centipedes (order Scolopendromorpha, 33%), annelids (order Haplotaxida, 11%), molluscs (class Gastropoda, 5.5%) and lizards (order Squamata, 5.5%) as well as some small arthropods such as springtails (order Entomobryomorpha, 11%), stoneflies (Plecoptera, 22%), booklice (Psocoptera, 5%), thrips (Thysanoptera, 5%) and earwigs (Dermaptera, 11%). At elevations between 2000m and 3200m (where weaver ants are absent), we identified all but one of the 18 orders present at the low elevation, plus three more [lacewings (order Neuroptera, 44%), nemertean worms (order Monostilifera, 6%) and mayflies (order Ephemeroptera, 6%)].
Our EcoSimR results showed that the diets of birds at low elevations and weaver ants overlapped significantly more than expected from random resource utilization at all taxonomic levels, albeit not at the ASV level. Diet of birds from low and mid-elevations overlapped significantly at order, family and ASV levels. Diet of birds from mid-elevations and the diet of low elevation weaver ants (weaver ants are absent at mid-elevation) did not overlap significantly at finer taxonomic scales (Supplementary Table S7). Lower diet overlap at finer taxonomic scales could be a function of sampling as indicated by the absence of asymptote in accumulation curves at finer taxonomic levels (Fig. S5) or the difference in arthropod species across the elevational gradient.
Effect of weaver ants on arthropod abundance We found no significant difference between number of arthropods on trees with or without weaver ants (Fig. S6a). However, the number of insects belonging to orders Coleoptera and Lepidoptera, the two most common orders in bird diet at all elevations, was 1.7 x higher on trees without weaver ants than on trees with weaver ants (Fig. 3a). Leaf damage estimated from 10 leaves on each tree was significantly greater on trees without weaver ants than on trees with weaver ants (Fig. 3b) and showed a non-significant trend in the same direction on the clipped branches from these trees (Fig. S6b).
Weaver ant removal and exclusion experiment One month after weaver ant removals and exclusion, the numbers of arthropods(excluding ants and homopterans) had increased from 4.67 ± 0.95 SE to 12.73 ± 1.26 SE, N = 15, while controls showed no significant change (before: 7.73 ± 1.06 SE, after one month: 9.8 ± 1.51 SE, N = 15). Overall, the difference between the change over time in treatment and control trees was significant (paired t-test, N = 15, P = 0.018) (Fig. 4). In the following year, the increase in number of arthropods on treatment trees was greater than control trees, but the difference was not statistically significant (Figs. S7, S8). Some of the experimental trees had been recolonized by ants over the course of the year, which might contribute to the reduced effect (Figs. S7, S8).
In the experimental treatments, two taxonomic orders of insects were responsible for the increase on experimental trees with respect to controls (Fig. 5a). Lepidoptera increased significantly in abundance (Lepidoptera: treatment 1.0 ± 0.5 SE, control -0.5 ± 0.3 SE, N=15, paired t-test P = 0.01) and Coleoptera showed a large increase that was close to significance (treatment 1.3 ± 0.9 SE, control -0.3 ± 0.4 SE, P = 0.13). These orders are also the most frequent components of bird diets at all elevations (Fig. 5b & Fig. S3). On the other hand, Hemiptera decreased after weaver ant exclusion (paired t-test N=15; treatment -8.8 ± 5.5 SE, control 4.3 ± 4.9 SE, P = 0.09). This decrease is expected given that it contains the suborder Homoptera and weaver ants form mutualistic associations with Homopterans (Peng and Christian 2005; Crozier et al. 2009).