Résumé
Plant virus community ecology is still poorly understood. Specifically, the biotic and abiotic factors that shape viral communities in natural settings remain little studied. However, the swift advancement of sequencing technologies and metagenomics techniques has revolutionized our ability to detect and inventory plant viruses in both natural and agriculture-impacted ecosystems. Here, we have leveraged the latest viral metagenomics techniques to investigate the different factors structuring viral composition in Camargue grasslands, which are subjected to intense environmental constraints such as salinity, climate variability, and limited water access. Based on 42 sampling sites with nine replicates each, we have identified 251 viruses in 239 plant species. These inventories further enabled us to build community matrices describing plant and viral species in sites, as well as metadata describing the environment and plant characteristics. The complexity of the available data, in terms of uneven dimensionalities (few samples compared number of species/variables), variable types (binary, continuous, percentage) and variable diversity (e.g. soil physico-chemical characteristic and land-use) poses challenges for statistical analysis and interpretation. To address this, we categorized all available explanatory variables by the type of process type that might approximate, either environmental filtering or dispersal. We further clustered similar samples inside those categories and used a community detection model applied to networks: namely the Stochastic Block Model (SBM). Congruence analyses on the clusters selected by the SBM were conducted to link community structures to process approximation structure. These analyses were complemented with a selection of generalized linear models explaining viral richness. We show that virus composition is only explained at 19 % by environmental filtering and dispersal, suggesting that viral composition is probably mostly structured by ecological drift or unmeasured variables. On the other side viral richness was explained at 50% by plant cover heterogeneity, plant biomass and surrounding land-uses.