Spatial scaling of soil microbial co-occurrence networks in a fragmented landscape

Abstract

Habitat loss has been a primary threat to biodiversity. However, species do not function in isolation but often associate with each other and form complex networks. Thus, revealing how the network complexity and stability scale with habitat area will give us more insights into the effects of habitat loss on ecosystems. In this study, we explored the relationships between the island area and the network complexity and stability of soil microbes. We found that the complexity and stability of soil microbial co-occurrence networks scale positively with island area, indicating that habitat loss will potentially simplify and destabilize soil microbial networks.

Figure 1

Influences of island area on the structural properties of soil bacterial and fungal networks. (A) Bacteria; (B) Fungi. Network properties include node numbers (n), total links (L), average node degree (average K), connectance, relative modularity (RM), and the number of keystone OTUs. Island area is log₁₀–transformed. Line in each panel was fitted using linear regression with a 95% confidence interval. The adjusted R ² and P values are shown.

Figure 2

The relative importance of bacterial/fungal richness and different soil properties in driving the variations of soil bacterial and fungal network properties. (A) Bacteria; (B) Fungi. Random forest was used to determine the variable importance. R ² (pseudo) represents the total variance that could be explained by environmental factors. The richness of bacteria (bacRichness) and fungi (fungiRichness) was calculated as the total OTU number on each island. Soil properties represent the mean value of all samples on each island. OTU, operational taxonomic units; TN, total N; TOC, total organic carbon; TP, total P.

Figure 3

Relationships of network stability with island area and multiple network properties. (A, B) Network robustness was calculated as the remained proportion of taxa after randomly removing 50% of the taxa from each empirical network; network vulnerability was calculated as the maximum node vulnerability in each empirical network. Line in each panel was fitted using linear regression with a 95% confidence interval. The adjusted R ² and P values are shown. (C, D) Spearman correlations (*P < 0.05, **P < 0.01, ***P < 0.001) of network robustness and vulnerability with different network properties, including n, L, average K, connectance, RM, and the number of keystone OTUs.