Fungal community assembly in drought-stressed sorghum shows stochasticity, selection, and universal ecological dynamics

Abstract

Community assembly of crop-associated fungi is thought to be strongly influenced by deterministic selection exerted by the plant host, rather than stochastic processes. Here we use a simple, sorghum system with abundant sampling to show that stochastic forces (drift or stochastic dispersal) act on fungal community assembly in leaves and roots early in host development and when sorghum is drought stressed, conditions when mycobiomes are small. Unexpectedly, we find no signal for stochasticity when drought stress is relieved, likely due to renewed selection by the host. In our experimental system, the host compartment exerts the strongest effects on mycobiome assembly, followed by the timing of plant development and lastly by plant genotype. Using a dissimilarity-overlap approach, we find a universality in the forces of community assembly of the mycobiomes of the different sorghum compartments and in functional guilds of fungi.

Fig. 1. Experimental design.

a Field layout of the 18 plots (16 × 8 m² each) in a random block design of three treatments (control, pre-flowering drought and post-flowering drought) and two sorghum cultivars (RTx430 and BTx642) with three replicates in a 76 × 59 m² field. The discovery that the effect of host genotype is negligible allowed us to use six replicates in most of our analyses. Each plot consisted of ten, 16 m long rows, each containing approximately 200 plants spaced 8 cm apart. b Irrigation scheme and sampling strategy. Irrigation for all treatments was identical until week 3 when pre-flowering drought was initiated. Irrigation for control and post-flowering drought was identical until week 10 when post-flowering drought was initiated. Soil samples were collected 20 cm from the plant stem to a depth of 15 cm with a soil corer, while leaf, rhizosphere and root samples were collected from plants extracted by shovel to a depth of approximately 20 cm.

Fig. 2. Structure of sorghum mycobiome.

a Principal coordinate (PCo) analysis of fungal community Bray–Curtis dissimilarity with permutational analysis of variance (PERM ANOVA) showing significant association of fungal community composition with, in order of importance, compartment, time point (TP), drought treatment and sorghum cultivar (***P < 0.001). Note that results of principal component (PC) analysis of Aitchison distance is presented in Supplementary Fig. 5. b Temporal change in relative abundance of fungal operational taxonomic units (OTUs) at each TP in the four compartments, three treatments and two sorghum cultivars. To avoid redundancy, pre-flowering treatment sampling began at the third week and post-flowering sampling began at the 8^th week. Source data are provided as a Source Data file.

Fig. 3. Temporal dynamics of the sorghum mycobiome.

a Succession is strong in leaf, root and rhizosphere, and weak in soil, based on Mantel testing of the correlation between temporal distance and Bray–Curtis community dissimilarity. b Turnover of fungal community composition demonstrated by Simpson dissimilarity among replicate plots and sampling times. Note strong fungal compositional turnover among weeks 1–9 in leaves followed by significantly less turnover for weeks 9–17. Similar strong turnover is seen among weeks 1–12 in roots, weeks 7–13 in rhizosphere, and weeks 9–13 in soil, but none is significantly different from other time points in these compartments. The Simpson metric differs from the Bray–Curtis and Jaccard metrics in that it is free from richness variance, as seen in our analysis of the sorghum root mycobiome (Supplementary Figs. 11–12) and our re-analysis (Supplementary Fig. 13) of the rice root microbiome of Edwards et al.. c Temporal change of individual OTU abundance shown by threshold indicator taxa analyses (TITAN). For each OTU, functional guild is noted by color and filled symbols show declining abundance (z−), and open symbols show increasing abundance (z+). Genus names of the OTUs in each guild can be found in Supplementary Fig. 10. Source data are provided as a Source Data file.

Fig. 4. Stochasticity and ecological drift in community assembly.

a–d Stochasticity is detected from values of two indices, β-Nearest Taxon Index (βNTI) and Raup-Crick Index (RCI). Where |βNTI| < 2 and |RCI| < 0.95 (red bars), compositional variance is most likely due to stochasticity, which in this range is independent from dispersal and selection. Note that stochasticity is dominant in early time points in leaves and roots. Selection is favored as the force for community assembly where βNTI ≤ −2 and RCI ≤ −0.95 (black). Here, communities are more clustered than expected by chance both phylogenetically and ecologically, likely due to homogenous selection. Where |βNTI| < 2 and RCI ≤ −0.95 (blue), communities are ecologically more clustered than by expected by chance but phylogenetically stochastic. In this case, community composition can be a result of either homogenous selection or homogenous dispersal and, of the two choices, homogenous selection is preferred due to the observed strong selection by host compartment and time, as demonstrated in Fig. 2A and Supplementary Data 4. (e–g) Stochasticity and community size. Ecological drift can be strengthened when the stochastic component of compositional variance is negatively correlated with community size. Here, we demonstrate this correlation when community size is estimated in three different ways: e fungal rDNA, internal transcribed spacer (ITS2) reads as a percentage of total ITS2 reads amplified from fungal plus sorghum host DNA, f fungal 18S rDNA amount assessed by real-time or quantitative PCR amplification and g fungal RNA as a percentage of fungal plus plant reads found in the transcriptomes of sorghum leaves and roots. h Stochasticity and richness. Ecological drift is suggested by a positive correlation between fungal richness and the stochastic component of compositional variance, as shown here. Source data are provided as a Source Data file.

Fig. 5. Drought responses of sorghum mycobiome.

a Pre-flowering drought effect on OTU abundance and fungal community composition. Note the strongest effects (R²) on root and rhizosphere. b Post-flowering drought effect on OTU abundance and fungal community composition. Note strong effects (R²) on all compartments except soil. OTUs above a false discovery threshold [green horizontal line with P < 0.00005 (≈0.05/1070 OTUs) or −log (P) = 10] show significant bias between drought and control. The symbol size corresponds to OTU abundance (log transformed) and color corresponds to fungus functional guild. Genera of the significant OTUs can be found at Supplementary Fig. 28. The R² is the difference in fungal community composition between control and drought treatments, as determined by permutational analysis of variance (PERM ANOVA). c Plant pathogenic fungal OTUs significantly affected by drought. In the compartments showing the strongest drought effects, root and rhizosphere, under post-flowering drought, plant pathogenic fungal OTUs become significantly more abundant than control, but under pre-flowering drought the pathogens are never more abundant than controls. d Boxplot showing OTU richness of fungal communities was significantly affected by drought in rhizosphere and roots. Note decreased richness under pre- but not under post-flowering drought in the rhizosphere and, oppositely, decreased fungal richness under post- but not pre-flowering drought in roots. e Delay of fungal community development by drought. Random Forest modeling of fungal community age shows that both pre- and post-flowering drought delayed the development of fungal communities to a similar extent. Random Forest modeling was used because pre- and post-flowering droughts are inherently temporal partitioned (Fig. 3), making it improper to simply compare the temporally variable, community compositional variance. Source data are provided as a Source Data file.

Fig. 6. Universality of ecological dynamics of the sorghum mycobiome.

a–h Dissimilarity-overlap curves (DOC) for a all samples of all fungi, b leaf samples of weeks 1–7 for all fungi, c root samples of weeks 1–3 for all fungi, d root, rhizosphere and soil samples for all weeks of arbuscular mycorrhizal (AM) fungi, e all samples of endophytic fungi, f all samples of plant pathogenic fungi, g all samples of yeasts, and h all samples of saprotrophic fungi. For DOCs, the dissimilarity-overlap curve is in red, the distribution density of sample pair overlap is in gray, and the point at which a negative DOC is first observed is marked by a vertical blue line (chosen by median of 1000 bootstraps). The fraction of negative slope (Fns) is simply the fraction of data points in the interval where the DOC has a negative slope and significance is as follows: *P < 0.05, **P < 0.01, ***P < 0.001. i Co-occurring network of fungal guilds of sorghum Mycobiome. Each dot represents a fungal OTU, and each edge represents a positive correlation with R > 0.8 (P ≪ 0.001). The different color represented the functional guilds of fungi. The dot size corresponds to OTU abundance (log transformed). Note for the 12th cluster, co-occurrence may be exaggerated among the large number of rare OTUs due to the dominance of zeros for rare OTUs in most samples. Source data are provided as a Source Data file.