Abstract
BACKGROUND: Hairy root disease (HRD), caused by rhizogenic Agrobacterium strains, is a significant disease threat to modern hydroponic greenhouses, which can result in up to 15% loss in yield. Our prior research has suggested increased alpha diversity after infection in hydroponic tomato root-associated microbiota. However, a more detailed investigation of how root-associated microbial components (MCs; clusters of weighted bacterial features) respond to disease and the underlying mechanisms remains lacking. To address this gap, we applied Latent Dirichlet Allocation (LDA) to analyze MCs from 12 Belgian commercial hydroponic tomato greenhouses. Using high-throughput amplicon sequencing of the 16S rRNA locus, three locations along each greenhouse irrigation system (beginning, middle, and end) were sampled at 5 time points throughout the 2018 growing season. RESULTS: In this study, we used LDA to identify root-associated MCs and gained insights into temporal changes and new health statuses. First, we observed a structured temporal pattern from the early stage (ES; sampling time points 1 and 2) through a transitional stage (TS; sampling time point 3) to the late stage (LS; sampling time points 4 and 5), showing different MC trajectories by health status. Second, MC4 (characterised by Paenibacillus spp.) was pronounced for healthy greenhouses in the ES, MC7 (characterised by rhizogenic Agrobacterium spp., Devosia and Limnobacter amplicon sequence variants (ASVs)) was pronounced for pre-symptomatic status, while MC0 (characterized by Comamonadaceae spp. ASVs) was indicative of an intermediate state between healthy and infected conditions. Furthermore, the ratio between Paenibacillus ASV and rhizogenic Agrobacterium ASV can be used as a biomarker to assess greenhouse health status in both ES and LS. CONCLUSION: We investigated hydroponic tomato root-associated MCs responses to HRD using LDA, which revealed different MC trajectories in terms of plant health. Our study advances knowledge of hairy root disease regarding the mechanisms that can improve plant health monitoring in greenhouses and biocontrol strategies. From a computational perspective, we demonstrate how to apply LDA-a powerful analytical tool-to understudied subfields through visual analytics.