Re-Analysis of 16S Amplicon Sequencing Data Reveals Soil Microbial Population Shifts in Rice Fields under Drought Condition

Rice (Oryza sativa. L) has been intensively studied to ensure a stable global supply of this commodity in the face of rapid global climate change. A critical factor that decreases crop yield is drought, which has been analyzed in various ways through many researches. Microbiome-based studies of rice investigate the symbiosis between rice and bacteria, which has been proposed as a way to overcome problems caused by drought. Several rice-associated metagenomic profiles obtained under drought conditions have been reported since the advent of next generation sequencing (NGS) technology. To elucidate the future diversity of plants and microorganisms and to promote sustainable agriculture, we reanalyzed 64 of the publicly available 16S amplicon sequencing data produced under drought condition. In the process of integrating data sets, however, we found an inconsistency that serves as a bottleneck for microbiome-based sustainability research. While this report provides clues about the composition of the microbiome under the drought conditions, the results are affected by differences in the location of the experiments, sampling conditions, and analysis protocols. Re-analysis of amplicon sequencing data of the soil microbiome in rice fields suggests that microbial composition shifts in response to drought condition and the presence of plants. Among the bacteria involved, the phylum Proteobacteria appears to play the most important role in the survival of rice under drought condition.

amplicon sequencing methods has accelerated analysis of this interaction.
Diverse microbes interact with plants (Lugtenberg and Kamilova 2009;Berg et al. 2014). Thus, the power of amplicon sequencing analysis in rice has been applied to better understand the relationship between plants and microbes in the underground areas, namely, the soil, rhizosphere, rhizoplane, and endosphere (Arjun and Harikrishnan 2011;Sessitsch et al. 2012;Edwards et al. 2015). Moreover, microbiome analyses have proven useful in understanding the effects of external factors such as abiotic stress and fertilizer limitation (Ahn et al. 2012;Ikeda et al. 2014;Rothenberg et al., 2016). Amplicon sequencing has also been applied to rice under drought conditions (Santos-Medellin et al. 2017;Reim et al. 2017). However, amplicon sequencing data from different studies vary due to geographical differences and the presence or absence of plant. To better understand the relationship between the microbiome and drought condition, which are closely linked to rice productivity, we reanalyzed publicly available amplicon sequencing data on the soil microbiome of rice under drought condition. We identify a candidate phylum that plays a role in the plant's response to the limits imposed by drought conditions on rice growth.
We collected 2061 microbiome sequence datasets from different plant species, sorted the data according to the workflow, and reanalyzed 64 soil microbiome sequences of rice (Fig. 1, Table S1). The microbiome data consists of data from soil, endosphere, rhizosphere, and rhizoplane (Table S1); however, the amount of microbiome data from the endosphere, rhizosphere, and rhizoplane is insufficient to allow rigorous comparison and re-analysis. Thus, we utilized only soil microbiome data for our analysis. We compared four different normalization methods: minmax, median, average, and total count, all of which showed similar patterns (Tables S2, S3, S4, and S5). Using normalized data from each sample, we selected the top 16 most abundant phyla (Table 1).
Analysis of a stacked bar plot and alpha diversity boxplot (Fig. 2, and Figure S1) indicates that the predominant members of the microbiome under drought conditions changed according to the presence or absence of plants. A two-way ANOVA test confirms the significance of this microbiome shift in response to plants (Table S6). In addition, we compared the relative abundance levels between members of the microbiome community in different environmental conditions (Table S7). Based on the shift pattern of phyla, this analysis yielded the following four clusters: phyla affected by both factors, those not affected by both factors, those affected by only the presence of plants, and those affected by only drought condition (Table S6). Among these clusters, Cyanobacteria and Epsilonbacteraeota were not affected by either factor. Meanwhile, Acidobacteria, Actinobacteria, Euryarchaeota, Gemmatimonadetes, and Planctomycetes were affected by only one factor, i.e., the presence of plants. To focus only on soil microbiomes affected by drought conditions, we analyzed the two clusters formed with drought: that affected only by drought and that affected by the presence of plants and drought. To facilitate the comparison between the results of analysis in each condition, we Internal Transcribed Spacer (ITS) and methanogenic mcrA data were first removed. Then, abiotic stress data other than drought stresstreated root microbiome data and non-crop plant data were excluded. Finally, by removing data sampled at different developmental stages, we reanalyzed amplicon sequence data of crop plants under drought stress and at similar developmental stages calculated the relative abundance ratio of drought to watered condition (from here "ratio").
Populations associated with drought conditions regardless of the presence of plants include Chloroflexi, Nitrospirae, Rokubacteria, and Bacteroidetes. In terms of the ratio of the abundance under drought to that under watered conditions, Chloroflexi shows a value of 0.738. In case of Nitrospirae and Rokubacteria, the ratios are 0.389 and 0.455, respectively (Table S8). Lastly, Bacteroidetes shows the lowest ratio of 0.250 (Table S8). Taken together, the microbiome developing under stress conditions is characterized by statistically significant shifts in the population that do not involve the establishment of a phylum with a relative abundance ratio greater than 1. This indicates that population of all phyla described in this section decreased under drought conditions (Table S6,  Table S8, Figure S2). Therefore, we did not further explore this cluster to identify useful candidate phyla to assist rice in overcoming drought stress.
A recent study reporting on Nitrospirae focused mainly on the nitrogen cycle (Xue et al. 2016); knowledge about its relationship with drought is limited. Similarly, drought-related information on Rokubacteria is limited, especially since this phylum has been newly classified (Becraft et al. 2017). Bacteroidetes make up the largest portion of the microbiome of mammalian   Table 1 intestines, thus its characteristics have been revealed through human microbiome studies. However, the interactions between plants and Bacteroidetes is largely unknown. Literature searches on all the phyla in this cluster provided no clues on their functional significance in relation to drought. Thus, our search for candidate phyla that can help plants overcome drought condition focused on the cluster consisting of populations that are affected by the presence of plants and stress. The populations affected by both drought conditions and the presence of plants are Armatimonadetes, Verrucomicrobia, Firmicutes, and Proteobacteria. The Armatimonadetes ratios of drought to watered are 0.819 in the presence of plants and 0.219 in the absence of plant (Table  S7). The reduction in the relative abundance of Armatimonadetes is consistent with previous results in potato (Gumiere et al. 2019). Similarly, the Verrucomicrobia ratios of drought to watered are 0.896 in the presence of plants and 0.378 in the absence of plants (Table S7). In the presence of plants, the abundance levels of populations of Armatimonadetes and Verrucomicrobia were lower under drought conditions than those under watered conditions. However, their population levels increased when they were interacting with plants under drought conditions, compared to their levels without this interaction. Therefore, these phyla have potential significance in assisting plants to overcome drought stress.
In the case of Firmicutes, the ratios of drought to watered are 0.161 in the presence of plants and 2.935 in the absence of plants (Table S7). This result that shows the ratio increasing under drought conditions in the absence of plants and is consistent with previous findings that members of the Firmicutes community increase under drought conditions (Chodak et al. 2015). In the case of Proteobacteria, the ratios of drought to watered are 1.029 in the presence of plants and 0.484 in the absence of plants. Considering that the p-value of ANOVAtest is less than 0.05 and the ratio is greater than 1 in the presence of plants, Proteobacteria is selected as the phylum that is most affected by drought conditions in the presence of plants (Table S6, Table S7). In addition, Proteobacteria are noticeably reduced under drought conditions in the absence of plants, an observation that is consistent with previous research showing that the abundance of Proteobacteria is very low ratio in water-limited desert soil (Bu et al. 2018) (Table S7). Although we did not find any reports indicating that Proteobacteria populations may shift in response to drought conditions in the presence of plants, we nevertheless selected Proteobacteria as the best candidate phylum to assist plants in overcoming drought stress. Thus, we performed more detailed analysis on this phylum.
To identify specific populations within Proteobacteria that respond to the presence of plants and drought, we analyzed the microbiome at the family level. The outcome of the analysis is presented in Fig. 3. The plot integrates 30 of the most abundant families in each sample. The following 13 families show a statistically meaningful Fig. 3 Relative abundance of the 30 most abundant Proteobacteria families from soil samples differentiated according to presence of drought stress or plants. Data points represent the abundance of each family relative to the total abundance of all families in each sample. The 30 most abundant families in each treatment were integrated, resulting in the identification of 31 families (one group consists of unassigned families). Each color represents a microbe in the following classes: Alphaproteobacteria (blue); Deltaproteobacteria (red); and Gammaproteobacteria (green). Bacteria not belonging to these classes are clustered as "Others." Grids are set according to the application of stress (drought (+)/watered (−)) and presence of plant (with (+)/without (−)). Detailed information of each sample is presented in Table 1 cluster shift (P < 0.05) in response to both drought and the presence of plants: Xanthobacteraceae, Acetobacteraceae, Beijerinckiaceae, Geobacteraceae, Syntrophobacteraceae, Syntrophaceae, SC-I-84, Xanthomonadaceae, Methylomonaceae, Rhodocyclaceae, Methylococcaceae, Steroidobacteraceae, and Hydrogenophilaceae (Table S9). Among these families, Xanthobacteraceae, Acetobacteraceae, Beijerinckiaceae, SC-I-84, Steroidobacteraceae, and Hydrogenophilaceae share similar trends as that of Proteobacteria (Table S10). Xanthobacteraceae is known to interact with plants and is involved in carbon and nitrogen cycling (Oren 2014;Wang et al. 2016). Meanwhile, members of Acetobacteraceae include plant growth promoting bacteria and well-known nitrogen-fixing bacteria (Saravanan et al. 2008;Reis and Teixeira 2015). In summary, we suggest six families belonging to Proteobacteria as candidates that assist plants cope with drought conditions.
Based on the results of our re-analysis, we constructed a model illustrating the statistically significant rice soil  (Table S6). The x-axis represents the ratio of drought to watered condition and y-axis indicates eight phyla significantly affected by drought. b The WordCloud refers to 13 families within the Proteobacteria that shift significantly as a cluster in the presence of both drought stress and plants (Table S9). Six families have statistically significant p-values and are presented in bigger size of letters with different color; these exhibit the same pattern as Proteobacteria in response to drought conditions in the presence of plants. The word sizes indicate the relative abundance ratio under drought conditions microbiome shift of four phyla in response to drought and the presence of plants; and four other phyla that are affected only by drought conditions. Figure 4 summarizes the abundance ratios of eight phyla affected by drought in the presence or absence of plants (Fig. 4a). Of these, Armatimonadetes, Verrucomicrobia, and Proteobacteria interacting with plants showed increased populations under drought conditions, while those of Firmicutes decreased. The Proteobacteria, in particular, includes more abundant populations under drought conditions than under watered conditions in the presence of plants, and was thus selected for the further analysis to identify six families that we propose as candidates that are worth further study (Fig. 4b). To help researchers in this field, we have analyzed the taxonomy of amplicons from reference studies that were excluded from our study due to insufficient amount of data ( Figures S3, S4, and Table S11).
To compare the results of our analysis to those of other crops, we analyzed sorghum (Sorghum bicolor L.) data under drought ( Figure S5). In this case, Proteobacteria did not display statistically significant population shifts. Thus, while additional experiments with Proteobacteria are recommended in rice, we cannot extend this recommendation to sorghum.
In conclusion, re-analysis of amplicon sequence data identifies several microbial shifts under drought conditions that depend on the presence of plants. We propose Proteobacteria as a suitable target for studies aiming to promote crop growth, in the same manner that previous studies have shown that inoculation of Actinobacteria enhances plant growth (Yandigeri et al. 2012;Hamedi and Mohammadipanah 2015). Because the composition of the microbiome depends largely on complex environmental variables and the genotype of the plant, new perspectives will be discovered as new microbiome data are generated.