Compositional shifts in root-associated bacterial and archaeal microbiota track the plant life cycle in field-grown rice

Abstract

Bacterial communities associated with roots impact the health and nutrition of the host plant. The dynamics of these microbial assemblies over the plant life cycle are, however, not well understood. Here, we use dense temporal sampling of 1,510 samples from root spatial compartments to characterize the bacterial and archaeal components of the root-associated microbiota of field grown rice (Oryza sativa) over the course of 3 consecutive growing seasons, as well as 2 sites in diverse geographic regions. The root microbiota was found to be highly dynamic during the vegetative phase of plant growth and then stabilized compositionally for the remainder of the life cycle. Bacterial and archaeal taxa conserved between field sites were defined as predictive features of rice plant age by modeling using a random forest approach. The age-prediction models revealed that drought-stressed plants have developmentally immature microbiota compared to unstressed plants. Further, by using genotypes with varying developmental rates, we show that shifts in the microbiome are correlated with rates of developmental transitions rather than age alone, such that different microbiota compositions reflect juvenile and adult life stages. These results suggest a model for successional dynamics of the root-associated microbiota over the plant life cycle.

Fig 1. The root-associated microbiota stabilizes after 8–9 weeks after germination.

(A) Principal coordinates analysis (PCoA) of Bray-Curtis dissimilarity between samples colored by root compartment. The numerical values used to construct this figure can be found in S1 Data. (B) The same plot as in panel A but colored by the field location. (C) The same analysis as in panels A and B but now showing principal coordinate (PCo) 1 versus PCo3, and the points are colored by the age of the plants from which the samples were taken. Hollow points represent bulk soil samples. (D) Heatmaps showing mean pairwise z-scores for similarity, computed as (1 − Bray-Curtis dissimilarity), between time points in each compartment for the 2014 California samples.

Fig 2. Shifts in the microbiota over time are associated with increasing and decreasing phyla.

(A) Bar plots of the top 11 phyla abundances over the course of the seasons in each compartment. Each bar represents 1 sample that was taken throughout the course of the growing season. The bars are ordered by the age of the plant as indicated by the colored points beneath each bar. Both the 2014 and 2015 data were used for this graph. The numerical values used to construct panel A can be found in S2 Data. (B) Beta regression coefficient estimates for microbial phyla that are either increasing (above 0) or decreasing (below 0) in relative abundance from the outside of the root to the inside of the root. (C) Beta regression coefficient estimates for microbial phyla that are increasing (above 0) or decreasing (below 0) in relative abundance over the course of the seasons in each compartment. All regression coefficients used to construct panels B and C can be found in S3 Data.

Fig 3. Random forest model detects taxa that are accurately predictive of plant age.

(A) The result of predicting plant age using the sparse random forest (RF) models for the 2014 and 2015 season. Each point represents a predicted age value. Solid circles represent predictions from training data, while hollow points represent predictions from test data. The numerical values used to construct panel A can be found in S4 Data. (B) Abundance profiles for the age-discriminant operational taxonomic units (OTUs) in rhizosphere and endosphere compartments over the course of the California 2014 growing season. OTUs are ordered along the y-axis by timing of peak abundance. The orders of the OTUs on the y-axis are not shared between rhizosphere and endosphere despite both the models for each compartment sharing a subset of age-discriminant taxa. OTUs are colored by their classification as early, late, or complex colonizers over the season (see color scale in panel C). See S3 Table for order of OTUs. There were 18 OTUs with decreasing patterns, 7 OTUs with complex dynamic patters, and 60 OTUs with increasing dynamic relative abundances in the rhizosphere. In the endosphere, there were 22 OTUs with decreasing relative abundances, 7 OTUs with complex patterns, and 56 with increasing relative abundances over the season. Slope estimates used to classify OTUs as having early, late, or complex patterns can be found in S5 Data. (C) Mean total abundance for the age-discriminant taxa across sites and compartments. The numerical values used to construct panel C can be found in S7 Data.

Fig 4. Drought exposure is associated with immature development of the endosphere microbiota.

(A) Microbiota age predictions for rhizosphere and endosphere samples from well-watered and drought-exposed plants. The horizontal line represents day 49, the age at which the plants were sampled. Age predictions for the drought-treated and well-watered control microbiota samples can be found in S8 Data. (B) Abundance of age-discriminant operational taxonomic units (OTUs) in the endosphere samples of well-watered and drought-exposed plants. Color scales are shared between panels A and B. Early, late, and complex colonizers are the same 85 OTUs used in the age-predicting random forest (RF) models (Fig 3B). *** P < 0.001, ** P < 0.01, * P < 0.05. Numerical values used to construct panel B can be found in S10 Data.

Fig 5. Rhizocompartments become more similar between field sites as a function of plant age.

(A) Pairwise distances between each site within each common time point and each common compartment. The numerical values used to construct this plot can be found in S11 Data. (B) Mean total relative abundance of the site-specific operational taxonomic units (OTUs) within each time point. All OTUs found to be differentially abundant between sites within each common time point can be found in S12 Data. The numerical values used to construct panel B can be found in S13 Data.

Fig 6. Varieties with different developmental rates have skewed microbiota progressions.

(A) The developmental stage of the tested varieties as a function of plant age. We staged each variety using descriptors previously described [24]. R0 corresponds to the panicle initiation stage. It is important to note that M401 and M206 had nearly identical times to panicle initiation but afterwards diverged in time to heading. Data used to construct panel A can be found in S16 Data. (B) Principal coordinates analysis (PCoA) of the 2016 data indicating that root-associated compartment and plant age are major determinants of microbiota structure. The numerical values used to construct panel B can be found in S14 Data. (C) Linear slope estimates for the principal coordinate (PCo) 2 in panel B as a function of plant age for rhizosphere and endosphere compartments of each variety. Separate linear models were constructed for data prior to 84 days (corresponding to the sixth collection time point, the point at which all varieties had at least entered the panicle initiation stage) and for data after this time point. Letters under each point indicate statistical significance. Statistical tests were constrained to individual compartments and times of the season (prior to 84 days and after 84 days). The values used to construct panel C can be found in S15 Data. (D) The predicted developmental stage of the 2016 data as predicted by the stage-discriminant sparse random forest (RF) models. Developmental stage predictions used to construct panel D can be found in S16 Data. (E) Total relative abundance estimates for early and late colonizing stage-discriminant taxa between each cultivar and compartment. The values used to construct panel E can be found in S17 Data.