Insect herbivory reshapes a native leaf microbiome

Publication abstract:

Insect herbivory is pervasive in plant communities, but its impact on microbial plant colonizers is not well-studied in natural systems. By calibrating sequencing-based bacterial detection to absolute bacterial load, we find that the within-host abundance of most leaf microbiome (phyllosphere) taxa colonizing a native forb is amplified within leaves impacted by insect herbivory. Herbivore-associated bacterial amplification reflects community-wide compositional shifts towards lower ecological diversity, but the extent and direction of such compositional shifts can be interpreted only by quantifying absolute abundance. Experimentally eliciting anti-herbivore defenses reshaped within-host fitness ranks among Pseudomonas spp. field isolates and amplified a subset of putative P. syringae phytopathogens in a manner causally consistent with observed field-scale patterns. Herbivore damage was inversely correlated with plant reproductive success and was highly clustered across plants, which predicts tight co-clustering with putative phytopathogens across hosts. Insect herbivory may thus drive the epidemiology of plant-infecting bacteria as well as the structure of a native plant microbiome by generating variation in within-host bacterial fitness at multiple phylogenetic and spatial scales. This study emphasizes that 'non-focal' biotic interactions between hosts and other organisms in their ecological settings can be crucial drivers of the population and community dynamics of host-associated microbiomes.

Methods
This repository contains R data objects (.rds) and text summaries from each of the Bayesian regression models we ran to evaluate the relationship between insect herbivory and the abundance of various bacterial taxonomic groups from our field studies. Please refer to the main text and supplemental information of our manuscript for detailed information on model structure, assumptions, and interpretation. Each model object can be loaded in R and used to, among other things:

output summaries of coefficient posterior estimates
generate samples from the joint posterior of model coefficients
generate samples from the posterior predicted response variable
calculate additional information criteria (e.g., WAIC) and compare amongst models