Results
The metaweb constructed for the French-Alps region using the taxa identified in 418 soil samples consisted of 55 trophic groups, three basal resources and a total of 383 potential interactions (Fig. S1). On average, local food webs were composed of 41 ± 4 trophic groups (and resources), with 206 ± 37 interactions. All 10 trophic classes were present in all local food webs, with a single exception where no group from the fungivore class was detected. The mean dissimilarity between local soil food webs was primarily influenced by changes in the dominance of trophic groups, rather than by changes in the presence or absence of groups (as indicated by increasing dissimilarity when increasing parameter q, which represents the weight of trophic group abundance and interaction, Fig.S2).
Overall, we found that food web composition varied across the environmental range covered in this study. The habitat type, particularly when comparing grassland and forest habitats, strongly structured the food web composition, both in terms of trophic group and interaction composition (Q1) (Fig.2, Fig.S3). The influence of habitat type on interaction composition was more pronounced compared to trophic group composition (Table S1). When examining shrubland areas, which represent ongoing shifts from grassland to forest, we found highly variable food web structures (Fig. 2). These structures exhibited similarities to both forest and grassland food webs, potentially due to the spatial proximity of shrubland sampling sites to forests or grasslands, or the specific stage of succession in the shrubland areas. By contrasting the average food webs of forests and grasslands, we identified compositional differences in the abundance of trophic groups and classes between the two habitats (Fig.3). Forest soil food webs were enriched in ectomycorrhizal fungi, litter and wood saprotrophic fungi, macro-detritivores like earthworms and diplopods at basal level, fungivores (particularly eudaphic collembola and protura), soil top predators including predatory mites, centipedes, predatory coleoptera and pseudoscorpions, as well as zooparasites. In contrast, grassland soil food webs had a higher proportion of decomposers and detritivores, particularly soil saprotrophic fungi, enchytraeids and coprophagous coleoptera (e.g., dung beetles), arbuscular mycorrhizal fungi and root endophytes, and most groups of plant phytoparasites and autotrophs. Grasslands also exhibited a higher proportion of bacterivores and omnivores compared to forests, notably bacterivorous nematodes, rotifers and epigeic collembola, at higher trophic levels. We further examined the ß-diversity of interactions between habitats and compared it to null expectations derived from randomising trophic group abundances while preserving trophic group diversity and network structure (see methods). The observed ß-diversity was significantly higher than expected by chance (Fig.4), indicating that trophic groups enriched in each habitat were not randomly distributed in the food webs, but instead formed pairs of trophically linked groups. This suggests that differences in trophic group abundance between habitats (Fig.3) are influenced by the structure of trophic links within the food web (Q2).
The composition of forest and grassland food webs was influenced by distinct ecological filters (Q3). In both habitats, predictors associated with these filters accounted for approximately 10% of the variance in soil food web dissimilarity (Fig.5, Fig.S4). In forests, spatial and environmental predictors played a more important role, particularly soil C/N ratio (related to organic matter degradability), pH, and frozen degree days (FDD) as indicators of abiotic stress (Fig.5a). Dissimilarities in trophic groups and interactions were more pronounced between sites separated by large spatial distances and remained consistent along the gradients of FDD and pH (Fig.5b). Changes in soil C/N ratio primarily influenced interaction dissimilarity, particularly at higher values of the C/N ratio gradient (>20).
In contrast, grassland soil food webs were predominantly driven by plant taxonomic and functional composition, as well as soil C/N ratio (Fig.5a). Soil C/N ratio and plant taxonomic/functional composition better explained interaction dissimilarity (Fig.5c), while plant taxonomic composition was more effective in predicting trophic group dissimilarity. Dissimilarities in trophic groups and interactions remained relatively constant along the plant dissimilarity gradient but were more pronounced in the lower part of the FDD gradient (> 40 cumulative degrees) and the upper part of the C/N ratio gradient (>15).