Environmental legacy effects impact maize growth and microbiome assembly under drought stress

Background and Aims As the climate changes, plants and their associated microbiomes face greater water limitation and increased frequency of drought. Historical environmental patterns can leave a legacy effect on soil and root-associated microbiomes, but the impact of this conditioning on future drought performance is poorly understood. Precipitation gradients provide a means to assess these legacy effects. Methods We collected soil microbiomes from four native prairies across a steep precipitation gradient in Kansas, USA. Seedlings of two Zea mays genotypes were inoculated with each soil microbiome in a factorial drought experiment. We investigated plant phenotypic and root microbiome responses to drought and modeled relationships between plant growth metrics and climatic conditions from the soil microbiome origin sites. Results Drought caused plants to accumulate shoot mass more slowly and achieve greater root/shoot mass ratios. Drought restructured the bacterial root-associated microbiome via depletion of Pseudomonadota and enrichment of Actinomycetota, whereas the fungal microbiome was largely unaffected. An environmental legacy effect on prairie soil microbiomes influenced plants’ drought responses: counterintuitively, prairie soil inocula from historically wetter locations increased shoot biomass under drought more than inocula from historically drier prairie soils. Conclusion We demonstrated links between soil microbiome legacy effects and plant performance under drought, suggesting that future drying climates may condition soils to negatively impact plant performance.


46
Increasing drought frequency due to global climate change will impact both soil and plant-  respectively. ITS reads were processed using DADA2 in an identical manner, except reads were 209 not truncated to preserve biologically relevant length variation. Forward and reverse 16S and ITS 210 reads were then used separately to learn the error rate prior to dereplication of reads into Amplicon 211 Sequence Variants (ASVs) and merging of pair-end reads. Taxonomy was assigned to the resulting unassigned to a kingdom. Chao1 richness and diversity indices (Shannon and Inverse Simpson) 217 were calculated before the removal of ASVs not observed ≥25 times in ≥5 samples and samples    and Ziegler 2017) implementation of the random forest algorithm was used to assess how 249 accurately roots could be classified as droughted or well-watered based on microbiome community 250 data (bacterial and fungal, separately). Samples were randomly split into a training set (80%) and 251 a test set (20%), with the test set withheld from the model, and used to determine the final model's 252 performance. Hyperparameters were optimized using a grid search over minimum node size (1, 5, 253 10) and the number of features available at each node (10-100% of the ASVs). For each 254 combination in the grid, classifier accuracy was used to assess performance on out-of-bag samples 255 with 10-fold cross validation. Final predictions were made with the trained model on the withheld 256 test set and visualized using confusion matrices. The contribution of each ASV to classification 257 accuracy was assessed using permuted importance: mean decrease in accuracy (MDA) values were 258 calculated for each ASV across the 10-fold cross validation. The benefits of this strategy are twofold, in that we can assess if drought induced a predictable shift in the microbiome as a whole, 260 as well as identify individual ASVs with the greatest differential abundance as a result of drought.

261
Connecting plant growth responses to microbial abundances 262 Linear models were used to test associations between taxon abundances and plant growth.

263
To focus on the phenotypic variation that could not be explained by Genotype, Inoculum, Block 264 and Batch effects, we extracted residuals from a model that contained these terms and used them conditions. We choose family and order as the taxonomic level for testing bacteria and fungi, 270 respectively as conservative measure to limit the number of tests, p-values were adjusted by term 271 to account for multiple testing.

273
Soil and inocula community composition differs across collection sites 274 We assessed the microbial community composition of soils and their derived inocula from 275 the six collection sites (n=8 samples per site). Bacterial and fungal alpha diversity of soils differed 276 across collection sites (Tables S1 and S2); bacterial diversity was highest in the westernmost   Plant responses to soil microbiomes vary across genotype and drought treatment 295 We found no evidence that soil inocula impacted emergence rates (p-value = 0.37, 296 Pearson's 2 test;  Figure 3A-B; p-value=<0.01) and increased root/shoot ratio ( Figure 3C; 300 p-value<0.01). These traits also differed between genotypes, with Mo17 investing more into root 301 growth than B73 ( Figure S7; p-value=<0.01). Shoot mass rate was the only trait that responded to 302 soil inoculum (Table 1; p-value=<0.01); this effect was driven by the 'TLIP' inoculum, which 303 slowed the accumulation of root mass.  accuracy is shown, for additional classifier statistics see Table S8 356 357 In contrast, drought treatment did not alter relative abundances of any fungal genera, nor 358 overall root fungal community profiles ( Figure 5A). Variance in fungal community composition 359 was mostly due to initial soil inoculum source ( Figure S10; Drought and maize genotype each explained <1% of the variance in fungal community 363 composition ( Figure 5B and S11, respectively). Thus, while root-associated bacterial community 364 composition responded to drought treatment, we observed no trend for fungal community 365 composition.  shown, for additional classifier statistics see Table S9.

378
Identification of bacterial and fungal ASVs responding to drought 379 Our random forest models showed greater accuracy in predicting the drought treatment of relative to well-watered plants (10.8% decrease in relative abundance). In contrast, we did not 388 detect any fungal ASVs that were differentially abundant in response to drought treatment, nor any 389 that contributed to random forest classifier accuracy with MDA value >0.5% ( Figure 5C and S14-  We identified bacterial and fungal taxa correlated with promotion or suppression of plant 418 growth. Overall, 13 bacterial families showed a significant positive (6/13) or negative (7/13) effect 419 on shoot mass accumulation rate (Table 4). These represented only a few phyla (6 Actinomycetota,  abundance and drought treatment, these are presented in Figure S11. 437 438   needs, perhaps allowing resources to be reallocated to functions that are beneficial to plants. 512 Another possibility is that soils from historically wetter sites generally possess greater microbial Water is a critical and limiting resource for plant growth as well as a functional 567 microbiome. As maize is cultivated widely across the United States, production will have to 568 contend with more frequent episodes of drought as the global climate warms. Engineering plant 569 microbiomes to assist in plant resilience under drought remains an ambitious, but worthwhile goal.

570
Our work illustrates the differential responses of root and shoot growth allocation to drought in 571 two well-studied maize genotypes and shows that bacterial, but not fungal, microbiota undergo 572 restructuring with drought, with decreases in the relative abundance of drought-intolerant taxa. In      Table 1 for full ANOVA table).   table in Table S8.  table in Table S9.