Successional changes in bacterial phyllosphere communities are plant-host species dependent

ABSTRACT

Phyllosphere microbial communities are increasingly experiencing intense pulse disturbance events such as drought. It is currently unknown how phyllosphere communities respond to such disturbances and if they are able to recover. We explored the stability of phyllosphere communities over time, in response to drought stress, and under recovery from drought on temperate forage grasses. Compositional or functional changes were observed during the disturbance period and whether communities returned to non-stressed levels following recovery. Here, we found that phyllosphere community composition shifts as a result of simulated drought but does not fully recover after irrigation is resumed and that the degree of community response to drought is host species dependent. However, while community composition had changed, we found a high level of functional stability (resistance) over time and in the water deficit treatment. Ecological modeling enabled us to understand community assembly processes over a growing season and to determine if they were disrupted during a disturbance event. Phyllosphere community succession was characterized by a strong level of ecological drift, but drought disturbance resulted in variable selection, or, in other words, communities were diverging due to differences in selective pressures. This successional divergence of communities with drought was unique for each host species. Understanding phyllosphere responses to environmental stresses is important as climate change-induced stresses are expected to reduce crop productivity and phyllosphere functioning.

IMPORTANCE

Leaf surface microbiomes have the potential to influence agricultural and ecosystem productivity. We assessed their stability by determining composition, functional resistance, and resilience. Resistance is the degree to which communities remain unchanged as a result of disturbance, and resilience is the ability of a community to recover to pre-disturbance conditions. By understanding the mechanisms of community assembly and how they relate to the resistance and resilience of microbial communities under common environmental stresses such as drought, we can better understand how communities will adapt to a changing environment and how we can promote healthy agricultural microbiomes. In this study, phyllosphere compositional stability was highly related to plant host species phylogeny and, to a lesser extent, known stress tolerances. Phyllosphere community assembly and stability are a result of complex interactions of ecological processes that are differentially imposed by host species.

INTRODUCTION

Microorganisms are an important source of biodiversity and are ubiquitous in nature. Across diverse environments, microbes contribute to ecosystem functionality through their involvement in biogeochemical cycling (1). Climate extremes, such as drought, heat waves, and flooding, are expected to increase in frequency and severity due to anthropogenic climate change (2). Because of the importance of microbial communities for ecosystem functioning, understanding how microbial community functional and compositional profiles respond to these climate disturbances will help us predict community and ecosystem responses to future climate conditions (3, 4).

Disturbances are events that either alter the environment thus affecting the inhabiting communities or are events that directly change a community through processes such as mortality or change in relative abundance of certain members (4). Depending on their duration, disturbances can be classified as pulses or presses (5). Pulses are distinct, short-term events, while presses are continuous, long-term events (4). How communities respond to disturbance is a measure of their stability, which can be broken down into two measurable states: resistance and resilience (4, 6). Resistance is the degree to which communities remain unchanged as a result of disturbance, and resilience is the ability of a community to recover to pre-disturbance conditions (4). Stability can be determined by looking at community composition, function, or the interrelatedness of the two metrics. Disturbance does not always result in the same degree of change in terms of community composition and function. For example, communities that display a high degree of functional redundancy could maintain a stable functional profile while their compositional profile might change, resulting in net neutral effects on ecosystem functioning (6).

One important but understudied microbial ecosystem is the phyllosphere, or the aerial surface of plants (7). Phyllosphere communities must withstand harsh, ever-changing conditions (such as UV radiation, desiccation, precipitation, and rapidly fluctuating temperatures, among others) and are subject to frequent and unpredictable dispersal events (e.g., insects, herbivores, wind, and precipitation). Despite these conditions, previous phyllosphere studies found patterns of seasonal succession and other predictable temporal changes in phyllosphere communities (8–10). Additionally, microbes in the phyllosphere promote host fitness and ecosystem functioning through effects such as phytohormone and nutrient production, increased stress tolerance, and protection against pathogens (11–15), but little work has been done to understand the stability of phyllosphere communities during and after environmental disturbances. Recent studies have estimated that plants and bacteria are the two leading sources of biomass on earth (16). Therefore, understanding how they interact has important implications for understanding ecosystem functionality in the face of changing climate.

By understanding how changes in community composition and functionality relate to ecological processes that drive community assembly over time, between environments, and during disturbance, we can develop a better and more holistic understanding of microbial community stability. Community assembly is the result of deterministic and stochastic processes shaping microbial communities (17). Deterministic selection comprises two processes: variable and homogenous selection. Variable selection occurs when turnover is greater between communities than is expected by chance resulting in significantly different communities. Homogenous selection happens when turnover occurs at a lower rate than expected by chance, resulting in more similar, or convergent, communities (17–20). Stochastic changes are composed of two dispersal processes: dispersal limitation and homogenizing dispersal. Dispersal limitation occurs when low levels of movement between communities cause them to drift apart, which can occur over space or time (17, 18, 20, 21). It is important to note that communities separated over time that are characterized by dispersal limitation are experiencing ecological drift and not an inability to mix (19). Homogenizing dispersal occurs when communities have a high dispersal rate resulting in similar community structures (18, 20). Finally, community assembly can be a result of multiple competing processes or weak selection and dispersal resulting in no dominant process. Under these conditions, community turnover is described as undominated (18, 20).

The goal of this study was to understand phyllosphere bacterial community resistance and resilience to water deficit (drought) in different grass host species during the course of a growing season. Since phyllosphere inhabitants are adapted to live in an extreme habitat, we hypothesized that phyllosphere communities would exhibit high levels of compositional and functional stability, but the degree of stability would be directly related to known levels of drought tolerance of the host species. Here, we investigated compositional stability by assessing changes in bacterial community structure, and to assess functional stability, we studied the rates of bacterial nitrogen fixation and changes in community growth rates (microbial biomass). These measurements were chosen for assessing functional stability because of their implications for plant and microbial community health. Nitrogen fixation in the phyllosphere is increasingly being explored for its input into the nitrogen cycle, its role in plant growth and development, and as a potential target for agricultural and ecosystem production (22, 23). Changes in microbial community biomass act as a control over community turnover, which is an important measure for understanding the rate of recovery after disturbance events (4). We further investigated community succession and stability by exploring community assembly processes using ecological null models. Due to the harsh nature of the phyllosphere, we expected microbial communities to be dominated by deterministic processes: homogenous selection within each host species under irrigated control conditions, but variable selection under disturbance conditions. We found that community stability and assembly processes are directly related to plant host phylogeny. Different degrees of stability were observed within each host species, which related to their dominant assembly processes. We focus on the phyllosphere of grasslands because this biome is both ecologically and agriculturally important on a global level and has the potential to help with climate change mitigation (24, 25). As an external stressor, we chose water deficit because grasslands around the world are expected to experience an increase in the frequency and severity of drought events as a result of climate change (26–29).

RESULTS

Plants show limited response to drought stress

In response to 10 weeks of drought stress, soil moisture significantly decreased for the plots of all three host species (P < 0.001 for all species) compared to irrigated treatments and fully recovered to irrigation control plot levels after 3 weeks of rewatering (Fig. 1A). Soil moisture content in tall fescue plots was significantly higher under drought than in both orchardgrass (P < 0.001) and ryegrass plots (P = 0.003), which were not significantly different from each other. Leaf relative water content (RWC), measured weekly to determine leaf water deficit, showed no significant drop within each host species as a result of drought (Fig. 1B). However, RWC was significantly lower in ryegrass host species than in tall fescue and orchardgrass (P < 0.001) (Fig. 1B). To further understand if plants were stressed, relative chlorophyll content was determined weekly (Fig. 1C). Only tall fescue was significantly and negatively impacted by the drought treatment (P = 0.018), while orchardgrass and ryegrass showed no differences between treatments. However, significantly different relative chlorophyll contents were found on the different host species (P < 0.001), and within each host species, relative chlorophyll levels showed significant changes over time (P < 0.001). Plant cellular membrane stability, determined by measuring electrolyte leakage, was not significantly affected by drought treatment on any of the three species (Fig. 1D) but was significantly higher during the first week of the experiment for each host species (P < 0.001). Furthermore, leaf proline content, measured as an additional plant stress indicator in week 10 (end of drought period) and week 13 (end of recovery period), showed no significant difference as a result of water reduction or time but was significantly higher in ryegrass compared to orchardgrass (P < 0.001) and tall fescue (P < 0.001) (Fig. 1E). Fine root biomass and its depth distribution, measured at the end of the drought period (week 10), showed no significant difference as a result of drought treatment or between species within any of the three sampling depths (Fig. 1F).

Fig 1.

Plant hosts showed few measured changes during the 10-week drought period despite a significant decrease in soil moisture in each of the three host species (A). Plant measurements taken throughout the experimental period include (B) leaf relative water content, (C) relative chlorophyll content, and (D) electrolyte leakage. (E) Proline content was measured (dry leaf mass) at the end of the drought period and the end of the recovery period. (F) Root mass was measured at three different depths at the end of the experimental period before rewatering occurred. Significant differences between treatments on an individual species are represented by asterisks, where *P < 0.05, **P < 0.001, and ***P < 0.001. Significant differences between host species are represented by lowercase letters. Letters that are the same represent no significant difference between host species.

Alpha diversity was not impacted by drought

Bacterial community richness [number of unique amplicon sequence variants (ASVs) observed within the phyllosphere communities] increased over the experimental period on all host plant species regardless of treatment (Fig. 2A). However, observed species richness was significantly different between the three host species, with ryegrass communities displaying the highest observed species richness and orchardgrass the lowest observed species richness. Similar trends in bacterial diversity were observed for both observed and estimated species richness (Fig. S1). Despite observed species richness increasing over the experimental period, community evenness (measured using Pielou’s evenness and defined as the number of each individual ASV) either showed no significant changes over time (orchardgrass and tall fescue) or showed a slight decrease over the experimental period (ryegrass) (Fig. 2B). Additionally, orchardgrass showed significantly lower community evenness compared to ryegrass and tall fescue, which showed no difference between one another. The corresponding relationships between species richness and evenness were related to microbial species dominance in the phyllosphere communities. Species dominance was higher in communities from orchardgrass and increased in all host species over time.

Fig 2.

Phyllosphere bacterial community diversity was measured for each host species and treatment over the experimental period using (A) observed species richness and (B) evenness. Community evenness on tall fescue hosts was the only alpha diversity metric to show a significant response to drought (P = 0.001). Differences in diversity levels between host species are represented by the lower-case letters at the top of each plot. Significant differences are represented by different letters, and no difference is represented by the same letter. Trends over time were determined using linear models. R² values are represented at the bottom of the plots for significant trends. N.S., not significant.

Throughout the experimental period and regardless of treatment, orchardgrass communities were dominated by bacteria from the class Gammaproteobacteria (75.8% SD 18.9); tall fescue communities were dominated by Alphaproteobacteria (43.7% SD 12.1); and ryegrass was dominated by both Gammaproteobacteria (30.1% SD 20.0) and Alphaproteobacteria (27.1% SD 11.8) (Fig. 3; Fig. S2–S4). While the dominance of Alphaproteobacteria and Gammaproteobacteria changed to different degrees within each of these host species, the relative abundance of Alphaproteobacteria increased over time on each host species, while the relative abundance of Gammaproteobacteria decreased on each host species. On orchardgrass, Alphaproteobacteria increased from 6.2% SD 5.0 to 15.2% SD 9.9 (P < 0.001), on tall fescue it increased from 23.4% SD 8.7 to 50.1% SD 8.6 (P < 0.001), and on ryegrass it increased from 23.1% SD 7.7 to 34.0% SD 9.7 (P < 0.001). However, Gammaproteobacteria decreased in relative abundance on orchardgrass from 81.2% SD 8.3 to 70.1% SD 19.0 (P < 0.001), on tall fescue it decreased from 36.2% SD 13.9 to 14.6% SD 10.1 (P < 0.001), and on ryegrass it decreased from 39.2% SD 16.0 to 15.4% SD 13.1 (P < 0.001). Alphaproteobacteria were significantly impacted by water deficit on orchardgrass hosts (P = 0.005) but not on tall fescue or ryegrass. Additionally, Gammaproteobacteria were not significantly impacted by water deficit on any of the three host species. Relative abundances and trends of bacterial classes were more similar on tall fescue and ryegrass compared to orchardgrass.

Fig 3.

Relative abundances of phyllosphere bacterial classes were significantly different between the plant host treatments and had different responses to drought. Communities from tall fescue and ryegrass hosts were found to be more similar to each other than they were to orchardgrass using hierarchical clustering using Euclidean distance and average linkage. The red triangle separates the drought period from the recovery period. Legend column A represents classes from orchardgrass hosts that were significantly different from ryegrass. Column B represents classes from orchardgrass hosts that were significantly different from tall fescue. Column C represents classes that were different between ryegrass and tall fescue. Black asterisks indicate differences in both control and drought treatments, green asterisks indicate differences between control only, and yellow asterisks indicate differences between drought treatments. Significance levels are assigned as P > 0.05 (not significant); *P ≤ 0.05; **P ≤ 0.01; and ***P ≤ 0.001.

Microbial community composition was differentially impacted by drought stress

Phyllosphere community composition was impacted by host species, time, and drought stress. The strongest driver of phyllosphere community structure was plant host species [R² = 0.59, P = 0.001; permutational analysis of variance (PERMANOVA) on Bray-Curtis distances] (Table 1) (Fig. S5). Sampling time (R² = 0.04, P = 0.001) and drought treatment (R² = 0.03, P = 0.001) were also significant drivers. PERMANOVAs were conducted for each week to understand how host species influenced community assembly throughout the growing season and to determine if and when drought affected bacterial community structure. Additionally, beta dispersion assessed inter-sample variation and whether or not any group had more compositional variance. At the start of the experiment, no treatment or species had significantly different dispersion (treatment: P = 0.59; host: P = 0.60); therefore, resistance was assessed based on when communities diverged from the control treatment compared within a given week. Due to successional changes over time, drought impact was determined based on when communities significantly diverged from the control community within that week and not when they diverged from the control community at the start of the experiment. While host species was always a significant driver of community differences, its contribution to community structure increased over the course of the growing season (Fig. 4; Table S3). In the first week of the experimental period, host species explained 32% of the variability in bacterial community structure compared to the last week of the experiment, in which it explained 41% of the observed variability (Fig. 4E).

TABLE 1.

Phyllosphere community structure was most impacted by host species and was also changing over time (week) and as a result of the drought treatment^a

	F value	R 2	P-value
Species	369.59	0.594	0.001***
Treatment	39.46	0.032	0.001***
Week	52.01	0.042	0.001***
Species × treatment	12.71	0.020	0.001***
Species × week	9.94	0.016	0.001***
Treatment × week	3.54	0.003	0.029*
Species × treatment × week	1.52	0.002	0.194
Residuals		0.290

^a The impact of each variable on community structure was determined using a PERMANOVA test on Bray-Curtis distance measures. Significance levels are assigned as P > 0.05 (not significant); *P ≤ 0.05; **P ≤ 0.01; and ***P ≤ 0.001.

Fig 4.

Phyllosphere bacterial communities from each host species became more distinct over time and were significantly impacted by drought stress. Non-metric multidimensional scaling (NMDS) ordination was plotted using Bray-Curtis distances for (A) the beginning of the experimental period (week 1), (B) early drought period (week 3), (C) late drought period (week 9), and (D) end of the recovery period (week 13). (E) PERMANOVAs were conducted for each corresponding period to understand community stability by determining when communities under disturbance showed signs of change and if they were able to recover from the disturbance event. Significance levels are assigned as P > 0.05 (not significant); *P ≤ 0.05; **P ≤ 0.01; and ***P ≤ 0.001.

Despite almost no differences observed in plant traits between irrigated control plants and drought-treated plants, phyllosphere communities showed significant responses to the drought treatment. Significant differences between treatments were first observed in week 3 (R² = 0.05, P = 0.031), further increasing during the drought period (week 10: R² = 0.11, P = 0.003). During the recovery period, treatment continued to have a significant but decreasing impact on bacterial community structure (week 13: R² = 0.07, P = 0.01), indicating communities were not fully able to recover from drought during the re-watering period (Fig. 4E). Recovery was assessed by comparing drought treatment to control treatment within a given week instead of to community composition at the start of the experiment because seasonal succession led to large shifts in the control communities over the growing season.

While communities within each plant species were influenced by time and drought, the degree to which each community was affected by drought and was able to recover was different between plant host species (Table 2). Ryegrass phyllosphere communities were most strongly affected by drought; they were the first to show changes as a result of drought treatment, and community composition remained significantly impacted during the recovery period. The next most susceptible phyllosphere communities were from tall fescue, which first showed differences in community structure as a result of drought stress in week 4 and remained significantly impacted throughout the entire recovery period. The more distantly related host species, orchardgrass (30), was the least affected by drought and only showed significant impact as a result of drought in weeks 7 and 10. These results indicate that phyllosphere communities showed varying degrees of stability (resistance and resilience), and this variation relates to the phylogeny and known drought tolerances of the hosts.

TABLE 2.

The effect of drought on phyllosphere community composition was different based on host species^a

	Orchardgrass			Ryegrass			Tall fescue
	F value	R 2	P-value	F value	R 2	P-value	F value	R 2	P-value
Week	18.065	0.121	0.001***	12.879	0.105	0.001***	13.995	0.096	0.001***
Treatment	7.753	0.052	0.001***	16.189	0.083	0.001***	11.092	0.076	0.001***
Week × treatment	1.291	0.009	0.202	3.555	0.023	0.001***	3.034	0.021	0.001***
Residuals		0.818			0.789			0.806

^a Ryegrass communities had the strongest response to drought and were the first to show changes as a result of treatment (week 3). Tall fescue communities were more stable than ryegrass communities showing a lower overall response to stress and showed changes as a result of stress later than ryegrass communities (week 4). Orchardgrass communities were the most stable, which was determined by the lowest overall impact of stress, the last to show significant changes as a result of drought (week 7), and orchardgrass was the only host species in which communities fully recovered to non-stressed composition. Bold values indicate the first week drought impact was observed on each host species. Significance levels are assigned as P > 0.05 (not significant); *P ≤ 0.05; **P ≤ 0.01; and ***P ≤ 0.001.

Phyllosphere community assembly is dominated by ecological drift

Ecological null modeling was used to assess community assembly and response to drought in three ways: (i) to understand changes within each week (irrigated control samples compared to drought samples each week) (Fig. 5), (ii) changes over the experimental period (each week compared to the first week separately for irrigated control and drought treatments) (Fig. S6), and (iii) investigate small-scale changes over time (each week compared to the previous week separately for irrigated control and drought treatments) (Fig. S7).

Fig 5.

Phyllosphere community assembly processes were different on each of the different host species, but all three hosts had high levels of undominated processes. Boxplots show the β-nearest taxon index (βNTI) values (A, C, and E) and RC_BC values (B, D, and F) comparing communities from the irrigated control treatment to communities from the drought treatment for each sampling week. Dashed red lines represent the significance cutoffs for βNTI (|βNTI| > 2) and RC_BC (|RC_BC| > 0.95). RC_BC values that fall between the dashed red lines represent undominated processes.

While all three host species displayed strong levels of undominated processes, indicating no individual assembly process can explain the observed differences, different degrees of deterministic and stochastic assembly processes were observed across plant species (Fig. S8). Phyllosphere community assembly on orchardgrass was marked by homogenizing dispersal and undominated processes, on ryegrass was influenced by variable selection, dispersal limitation, and undominated processes; and on tall fescue was dominated by dispersal limitation and undominated processes (Fig. 5).

When comparing irrigated to drought-stressed communities within a given week, homogenizing dispersal accounted for 32% of community assembly on orchardgrass hosts, suggesting species exchange between these communities (Fig. 6). On ryegrass, variable selection (23%) and dispersal limitation (47%) drove community assembly processes (Fig. 6). Additionally, deterministic processes began to increase in week 4 as differences between ryegrass communities resulting from the drought treatment intensified (Fig. 5C). On tall fescue hosts, community assembly under drought stress within a given week was characterized mostly by undominated processes, but dispersal limitation (22%) and variable selection (6%) did have moderate impacts on community assembly (Fig. 5E and F; Fig. 6).

Fig 6.

Treatment effects on phyllosphere community assembly processes for each of the host species. While undominated processes were prevalent on each host species, the defined processes varied. The percentage of contribution was calculated by dividing the number of significant pairwise comparisons for each process by the total number of pairwise comparisons.

When comparing assembly processes from each week to the start of the experimental period, orchardgrass was characterized by higher levels of homogenizing dispersal in the irrigated control treatment than in the drought treatment, thus indicating microbial communities on irrigated control plants are more similar over time than the communities on drought-stressed plants. However, ryegrass was dominated by dispersal limitation—40% of assembly in the irrigated control communities and 40.6% in the drought-exposed communities (Fig. S6B and D), demonstrating ecological drift occurred over time. Furthermore, variable selection was responsible for 20% of community assembly in the drought treatment compared to only 2% in the irrigated control treatment over time. Similarly, on tall fescue hosts, assembly processes were more influenced by dispersal limitation the longer the experiment progressed (Fig. S6C). Variable selection similarly influenced communities from irrigated control and drought-treated plants over the course of the experiment (15% and 13%, respectively) (Fig. S6D).

Looking at changes over shorter time periods by comparing consecutive weeks, orchardgrass had high levels of homogenizing dispersal (Fig. S7A), which indicates that communities were getting more different as time progressed. Ryegrass had high levels of dispersal limitation in the irrigated control community for 36% of assembly processes compared to only 14% in the drought treatment communities, suggesting drift had an increasing influence on community assembly from week to week in the irrigated control communities compared to the related communities in the drought treatment (Fig. S7B and D). Conversly, on tall fescue, community development was mostly a result of undominated processes (Fig. S7C). However, the variable selection had a greater influence on drought-stressed communities from week to week (4% and 15%, respectively) (Fig. S4D) suggesting an increase in selective pressure as drought conditions intensify.

Mantel tests were used to correlate results from the ecological null modeling over time and across host species. To understand the relationships between community assembly and time, sampling day was transformed into an Euclidian distance matrix, which was correlated to both the β-nearest taxon indices (βNTI) and RC_BC distance matrices. This was done separately for each host species and treatment. βNTI was correlated with time for communities from the drought treatment on ryegrass (R = 0.116, P < 0.001) and orchardgrass (R = 0.061, P = 0.05). RC_BC values were significantly correlated with time for every host species and treatment (Table S2). Phylogenetic host distances were used to correlate community assembly processes with host relatedness. βNTI values were correlated to phyllosphere communities from the irrigated control treatment (R = 0.06, P < 0.001) but not from the drought treatment. However, RC_BC values were significantly correlated to host phylogeny for both, irrigated control (R = 0.41, P < 0.001) and drought treatments (R = 0.39, P < 0.001).

Phyllosphere showed functional stability under drought

Microbial biomass of the phyllosphere communities was determined through direct cell count using epifluorescence microscopy (Fig. 7). Ryegrass had a significantly higher ratio of bacterial biomass per leaf mass than orchardgrass (P < 0.001) and tall fescue (P < 0.001). Microbial biomass was not affected by drought on any of the host species but was influenced by time. In tall fescue and orchardgrass, samples taken on the first day of the experiment had significantly higher microbial biomass levels than the subsequent sample days, which were stable from week 5 until week 13. Bacterial biomass on ryegrass hosts showed greater temporal variability and an overall decreasing trend over the course of the experiment, with only week 12 (P = 0.005) and week 13 (P = 0.009) showing a significant decrease relative to week 1.

Fig 7.

The effects of host species and treatment on phyllosphere bacterial biomass (cell numbers) over time. The total number of bacterial cells per leaf material from each host species was calculated by washing bacterial cells off the surface of leaves, direct counts using epifluorescence microscopy, and comparing to leaf areas using ImageJ. Each watering treatment is represented by five biological replicates for each host species on each sampling day. After the first sampling day, orchardgrass and tall fescue showed stable bacterial biomass, but ryegrass had more day-to-day variation.

The ability of phyllosphere bacteria (diazotrophs) to fix atmospheric nitrogen was determined by measuring the rate of nitrogen fixation using stable isotope tracing. Bacterial communities from ryegrass had a significantly higher rate of nitrogen fixation per gram of leaf material than was found for tall fescue (P < 0.001) and orchardgrass (P < 0.001) (Fig. 8A). Within each host species, no difference in nitrogen fixation was found as a result of treatment or time. To test if in situ nitrogen fixation rates were related to the potential for nitrogen fixation, nifH gene copy numbers were determined using qPCR. nifH abundance was not significantly affected as a result of treatment or time (Fig. 8B). However, gene abundances per gram of leaf material were different between all three host species. Tall fescue had the highest number of nifH genes, while orchardgrass had the lowest. The relationship between nitrogen fixation rates and nifH gene abundances was evaluated using linear regression, but this was not significant (R² = −0.007, P > 0.05). Additionally, nifH gene copy number and nitrogen fixation rate were compared to the dominant bacterial classes, but no significant correlation was found.

Fig 8.

Rates of nitrogen fixation (A) and abundances of the nifH genes (B) were significantly different between host species but were not affected by the drought treatment. Nitrogen fixation rates were measured using stable isotope tracing, exposing the bacterial communities to ¹⁵N₂. The abundance of nifH genes was measured using qPCR and standardized to the number of copies per gram of leaf material. Significant differences between host species, determined by linear models with Tukey’s Honest Significant Difference post hoc analysis, are represented by lowercase letters in the upper left corner of each plot. Different letters indicate statistical differences between hosts.

DISCUSSION

Several phyllosphere studies have shown that host species selection is an important driver of community assembly, but few studies have been conducted across grass species (8, 9, 31–35). Despite grass leaves being close to the soil, which could provide a constant source of microbial colonizers, phyllosphere communities in our study could be clearly differentiated based on host species identity. Perennial ryegrass and tall fescue are more related to each other than they are to orchardgrass (30), which was closely mirrored in microbial community similarities: they had similar diversity levels and were more similar to each other compositionally than they were to orchardgrass. Like in many phyllosphere studies, all microbial communities in our study were dominated by Proteobacteria (8, 32, 34, 36, 37). However, orchardgrass was dominated by Gammaproteobacteria, tall fescue was dominated by Alphaproteobacteria, and perennial ryegrass was dominated by both. In addition to compositional differences at a high taxonomic level (class) between host species, community succession throughout the growing season followed different patterns between the host species. From this, we conclude that successional changes in bacterial phyllosphere communities are plant-host species dependent. To further understand the differences in phyllosphere community assembly, we investigated if the response to a disturbance (drought) remains host species dependent or if a common disturbance drives communities to become more similar to each other.

Microbial community stability has many definitions. Here, we used the definition of Shade et al. (4) that stability is composed of resistance (the degree to which a community is unaffected by disturbance) and resilience (the rate at which a community returns to an undisturbed condition). We used drought as a disturbance event because it is a climatic stressor that grasslands are experiencing with increased frequency and severity (28, 29, 38). Community stability can be related to compositional or functional stability, which can exhibit similar trends under stress but does not always do so (6). Resistance was measured by determining if and when a community showed significant differences from the control community (alpha diversity, beta diversity, or functionality). Resilience was assessed by determining if communities were able to recover to a state indistinguishable from the control community at the same time point after soil rewetting. Because of the observed community succession over the course of the experiment, resistance and resilience were assessed in relation to irrigated control communities measured on the same day as opposed to comparing them to pre-disturbance communities.

Despite 10 weeks of no irrigation and a significant decrease in soil moisture, little to no drought-related effects were observed on the measured plant traits: leaf RWC, electrolyte leakage, chlorophyll, proline, and biomass. A meta-analysis conducted by Kröel-Dulay et al. (39) found that simulated drought field experiments underestimate plant responses to drought, which could explain the lack of plant response observed in this experiment. We expected to see deeper rooting in the drought-treated plots to account for the reduced water availability, but no difference in rooting between the treatments was observed. The established deep roots observed across all species were likely able to provide water to drought-treated plants and could therefore be another explanation of why drought impacts on the plant host were not observed. Despite almost no measurable effect detectable in plant hosts, microbial community composition was significantly affected by drought. All three hosts showed stable alpha diversity during disturbance but presented varying degrees of compositional stability. Host species phylogeny had a strong influence on phyllosphere community stability levels, with known drought tolerance of the host species exhibiting a lower effect. Ryegrass and tall fescue had more similar levels of microbial community stability compared to orchardgrass. Additionally, tall fescue communities were more stable than perennial ryegrass communities, which is consistent with the higher drought tolerance exhibited by tall fescue (40). However, orchardgrass communities showed much greater stability than tall fescue communities despite having a slightly lower drought tolerance (40). These differences could relate to the overall physiology and stress survival strategies of the hosts. Even though plants did not show visible signs of stress and our measurements showed minimal differences between the treatments, plants undergo biochemical changes before visual signs of stress are observed (41). Furthermore, orchardgrass and tall fescue have been shown to employ different drought resistance strategies. Tall fescue primarily relies on a deep root system, whereas orchardgrass has better water extraction efficiency from soil and exhibits drought tolerance in its tissues (42). These differences are seen to a small degree within our plant health measurements. Tall fescue maintained a higher soil moisture level than orchardgrass, which could be attributed to tall fescue plants accessing water at a much greater soil depth and orchardgrass extracting more water from the surface soil. Additionally, chlorophyll content was higher under drought treatment compared to the irrigated control treatment in tall fescue plots but not in orchardgrass plots. This is consistent with previous studies in tall fescue that found an increase in chlorophyll content under mild drought stress, which could result from decreasing cellular growth compared to chlorophyll turnover (43). A more in-depth comparison of plant traits and metabolic profiles under normal and stressed environments could provide further insight into the differences observed between microbial communities from different hosts and if these changes help mitigate negative effects associated with stress.

To understand what ecological processes may be underpinning compositional stability in the phyllosphere, ecological null modeling was used to determine the influence of deterministic and stochastic processes on community assembly under undisturbed (irrigated control) and disturbed conditions (drought). Since the phyllosphere is considered a harsh environment, we expected deterministic processes would dominate community assembly. Instead, we found high levels of stochastic processes and processes that could not be well defined as stochastic or deterministic (undominated assembly), with only low levels of deterministic assembly processes. Across host species, treatment, and sample day, dispersal limitation accounted for 44% of assembly processes, variable selection accounted for 11%, and homogenizing dispersal for 3%. Furthermore, within these quantifiable processes, different trends were observed within each host species, which related to their varied levels of community stability.

Ryegrass was marked by high levels of dispersal limitation over time and showed higher levels of variable selection between irrigated control and stressed samples than the other host species. Variable selection occurs when communities are more different than is expected by chance (44), and these changes between communities can result from selection pressures caused by ecological disturbances (4). The compositional instability observed on the drought-sensitive ryegrass hosts could result from disturbance driving deterministic selection processes (45). Despite our plant analysis revealing little effect on plant hosts as a result of disturbance, finer-scale responses may be taking place at the metabolic level resulting in different selection pressures. Additionally, when comparing communities to the first week, dispersal limitation resulted in 40% of irrigated control and 41% of disturbed assembly processes. This lack of species flow between sampling events makes communities progressively dissimilar over time, indicating microbial communities are unable to interact due to temporal separation. Furthermore, when communities are unable to mix, ecological drift over time causes phyllosphere communities on host plant species to become more different from one another (17, 18, 21).

Tall fescue had the highest levels of undominated processes when comparing control to disturbed samples within a given week and when comparing subsequent weeks. This could result from strong competing assembly processes, where both deterministic and stochastic processes are occurring and thus potentially canceling out the signal of each. When comparing communities to the first week in an undisturbed state, dispersal limitation accounted for 59% of assembly processes in the undisturbed control hosts, but only 25% in the disturbed hosts. This discrepancy between treatments relates to the changes in community evenness observed over time on tall fescue hosts. Evenness decreased over time in communities from tall fescue irrigated control plants as certain bacteria became more dominant over time, thus resulting in the loss of rare species, which potentially results from drift. However, evenness was stable under drought conditions, indicating the rare species were not lost, which could account for the lower levels of dispersal limitation that were observed.

Orchardgrass was the only host species with high levels of homogenizing dispersal. This was true both across time and when comparing undisturbed control and disturbed communities. Homogenizing dispersal suggests low rates of microbial community turnover as a result of dispersal mixing of the microbial communities (44). Additionally, orchardgrass had the lowest levels of dispersal limitation, indicating communities were not separated by space or time. The high species dominance and lower community richness observed on orchardgrass could explain why there were decreased levels of dispersal limitation and drift in orchardgrass communities, subsequently resulting in increased resistance and resilience.

Due to the harsh nature of the phyllosphere, we expected microbial communities from irrigated control plants to exhibit homogenous selection within each host species, and when subjected to disturbance expected to find variable selection as the dominant process as has been demonstrated in other systems (45–49). While we did observe some increase in variable selection under disturbance conditions, we found community assembly processes resulted mostly from dispersal limitation and undominated processes. Similar results were found in a serpentinizing aquifer system for which the authors concluded that the low diversity system dominated by a few taxa and many variable low abundance species made the system prone to ecological drift (21). Previous work has shown that environments exerting strong selection forces could result in low biomass and low diversity communities characterized by a few abundant taxa and many rare taxa, which are susceptible to large variations in growth and death rates, thus resulting in drift (17, 21, 50, 51). Since phyllosphere communities comprise a few abundant taxa and many rare taxa (52), we propose they are similarly prone to drift despite strong environmental selection pressures, but that levels of drift relate to degrees of stability.

Undominated processes, which were a major driver of assembly in the phyllosphere communities, frequently result from weak selection, weak dispersal, diversification, or drift (18, 20). Previous studies concluded that undominated processes can also result from counteracting forces that make it hard to quantify individual processes (19, 21). The extreme nature of the phyllosphere as a microbial habitat along with inconsistent dispersal events could result in counteracting deterministic and stochastic processes, which then led to undominated processes becoming dominant, as observed between host species, over time, and during environmental disturbance.

In addition to compositional stability, functional stability was determined for two different functions: bacterial growth and diazotrophic nitrogen fixation. Bacterial biomass (cell numbers), used as a proxy for bacterial cell growth, exhibited higher levels at the beginning of the experiment. Biomass of the phyllosphere community decreased and remained steady in orchardgrass and tall fescue but showed more variability over time in ryegrass. The high biomass on day 1 of the experiment corresponds with a higher level of electrolyte leakage in all host species. Electrolyte leakage measures membrane permeability and is thus an indicator of cell death; therefore, increased permeability leads to leaching of compounds, which could be utilized by phyllosphere microbes and thus account for the observed biomass differences (53). Despite the temporal differences, treatment had no effect on cellular biomass, indicating stable growth rates during disturbance. Growth rates are tied to resource use efficiency, which can have important implications for community stability. Microbes that are adapted to high nutrient environments (copiotrophs) are more resilient but have less resistance, while microbes that grow best at low nutrient levels and frequently exhibit slow growth (oligotrophs) have higher resource use efficiency and therefore have increased resistance (4, 54). This dynamic was observed in a study looking at the resistance and resilience of grassland soil communities in response to drought. The authors of this study found the slow-growing fungi were more stress resistant, while the faster-growing bacteria were more stress resilient (55). The phyllosphere communities in our study were dominated by Alphaproteobacteria and Gammaproteobacteria, which are frequently characterized as oligotrophs (54). These two classes of bacteria likely support community stability as they made up over half of the bacterial species found on each host species and showed no significant changes under drought stress.

Nitrogen fixation was measured within this experiment to first assess if this process was occurring in these phyllosphere communities and then to understand whether it was a stable process relative to drought stress. Nitrogen fixation has been extensively studied in the rhizosphere and in leaf litter, but few studies have directly measured it as a process in the phyllosphere and none on grass hosts (15, 56–58). Bentley and Carpenter (56) found that phyllosphere microbes could account for 10%–25% of the nitrogen content found in their study species, indicating phyllosphere microbes are likely an important and underrecognized global source of reactive nitrogen. Developing a better understanding of nitrogen fixation in the phyllosphere and determining if it is a stable process have important implications for the use of phyllosphere microbes in sustainable agriculture (59). In our study, we found that nitrogen fixation occurs at a stable rate over time and under drought (i.e., being functionally stable) but occurs at different rates depending on host species. To further understand nitrogen fixation in the phyllosphere, we compared the rates of nitrogen fixation measured by ¹⁵N₂ tracing to the abundance of nifH genes to determine how fixation potential relates to in situ activity. Previous studies in leaf litter found nifH gene abundances to be closely related to nitrogen fixation rates in free-living bacteria (60). However, in our study system, nifH abundances did not relate to nitrogen fixation rates, even though both rate and abundance were species specific and showed no significant variation over time or as a result of disturbance, again showing the high functional stability of nitrogen fixers in the phyllosphere based on nifH abundance. Further studies investigating leaf uptake of phyllosphere available nitrogen and the microbes involved in nitrogen fixation will be important for understanding the phyllosphere’s contribution to the nitrogen cycle and whether phyllosphere microbes (diazotrophs) can be used as a biofertilizer to replace chemical nitrogen fertilizers.

Conclusion

In conclusion, phyllosphere compositional stability in this study was highly related to plant host species phylogeny and, to a lesser extent, known plant stress tolerances. However, under disturbance, all host species showed functional stability for both nitrogen fixation rate and bacterial biomass, which we concluded was due to no significant difference between treatments for each host species. These discrepancies could be a result of functional redundancy within the community or could relate to the nature of phyllosphere communities being prone to ecological drift. Phyllosphere communities are composed of a few stable dominant taxa and many rare taxa that are constantly in flux. We propose that the stable, dominant taxa are responsible for most of the measurable functionality of the community, which would explain why we see variability within and between days but not in response to disturbance. Phyllosphere community assembly and stability are a result of complex interactions of ecological processes that are differentially influenced by host species. Given the vast expanse of phyllosphere ecosystems and the climate stress agricultural systems are facing, it is important to continue to understand assembly processes in relation to community functionality and to understand how these functions relate to host species growth, metabolism, and stress response.

MATERIALS AND METHODS

Experimental design

The experiments were conducted at the University of Massachusetts Research and Education Farm (South Deerfield, MA, USA) in the summer of 2020. The soil was characterized as a silty loam soil (28.8% sand, 64.3% silt, and 7.0% clay), and soil analysis was conducted in the University of Massachusetts Amherst Soil and Plant Nutrient Testing Laboratory (Table S1). Seeds of three native temperate grass species, Festuca arundinacea (endophyte-free tall fescue cultivar “Cowgirl”), Dactylis glomerata (orchardgrass cultivar “Echelon”), and Lolium perenne (perennial ryegrass cultivar “Sierra”) were acquired from the Albert Lea Seed Company (Albert Lea, MN, USA). These species were chosen because they are common grasses found in North American forage systems and because their drought response has been extensively studied and characterized. In field studies, tall fescue frequently displays the highest level of drought tolerance, while ryegrass has the lowest drought tolerance (42, 61, 62). Grasses were established by seed in July 2019. Seeds were planted at a rate of 11, 10, and 15 kg ha^-1 for tall fescue, orchardgrass, and perennial ryegrass, respectively, into 10 plots of 2 × 3 m each per grass species resulting in a total number of 30 plots. The central 1.5 × 2.5 m area within each plot was selected as the study area to limit edge effects. Between plots, a 3 m wide border of Poa pratensis was planted to maintain distinct treatment plots and to limit the encroachment of other species into test plots. In the following year (May–June 2020), plots were mowed weekly to a height of 20 cm to stimulate growth prior to the experimental period, and in June 2020 plots were fertilized with 10-10-10 fertilizer (NutrienAg, Loveland, CO, USA). All borders were mowed weekly during the experimental period to a height of 2.5 cm.

The experiment consisted of three species divided into two experimental groups (irrigated control and drought stressed), for a total of six treatments arranged in a randomized design with five replicates for each combination of species and watering level. The drought experimental group had two separate treatment periods: a drought period followed by an irrigation period for recovery. Therefore, three treatments are discussed: irrigated control, drought, and recovery. To simulate drought exposure, the drought treatment plots were covered with rain exclusion shelters consisting of modified 102 m² greenhouse cold frames (Growspan, CT, USA). Each cold frame was covered with Thermal AC Greenhouse film (Greenhouse Megastore, CT, USA) chosen for its high light transmission, allowing more of the UV spectrum to pass through the film than conventional greenhouse films, and for its drip control, which prevented condensates from dripping onto the experimental plots. Each greenhouse frame was modified so that greenhouse film could be rolled up and down the sides and ends. This design allowed for maximal airflow through the plots and prevented heat from being trapped under the shelters. Drought treatments began on 13 July 2020 and lasted for 10 weeks until the rainout shelters were removed on 19 September 2020. Plots were rewatered for 3 weeks to allow for plant recovery and to test for microbial community recovery.

Drought stress for plants was created by establishing a controlled, repeatable experimental setup. Plants in the drought-stressed treatment received no water during the 10-week drought period when drought treatment plots were covered with rain exclusion shelters. This extreme drought is equivalent to a 20-year drought event in this area (63). The adjacent, positive control experiment was set up with constant soil moisture conditions to create the contrast of a drought-free growth environment for the plants. In addition to any seasonal rain, irrigated control plants were given supplemental water as needed to maintain soil moisture at above 25% volumetric soil moisture content. Soil moisture was measured using a MiniTrase TDR with a 15-cm stainless steel buriable probe (Soilmoisture Equipment Corp., Goleta, CA, USA). Soil moisture levels in the drought treatment plots were reduced to approximately 15%–18% by the end of the drought period and returned to the original >25% during the recovery period. Plant physiological and bacterial parameters were measured every week throughout the 13-week experimental period and are described in more detail below.

Plant health measurements

Plant physiological parameters were measured on a weekly basis as a measure of plant health in response to irrigated control, drought, and recovery treatments. Plant water status was monitored based on leaf relative water content (64). Leaf cellular membrane stability was determined based on an electrolyte leakage assay (64). Leaves were removed from plants, rinsed with distilled water, and placed into tubes containing 50 mL of distilled water. The tubes were placed on a shaker for 8 hours, and then the initial electrical conductivity (Ci) was measured using a conductivity meter (Orion 3-Star Conductivity Meter, Thermo Scientific, Beverly, MA, USA). The tubes were then placed in an autoclave to kill the leaf tissues and placed on a shaker for 24 hours before the final electrical conductivity (Cf) was measured. Electrolyte leakage was calculated as Ci/Cf × 100. Relative chlorophyll content was measured using a FieldScout CM 1000 chlorophyll meter (Spectrum Technologies, Inc., Aurora, IL, USA).

Proline is an amino acid found in plants that is synthesized and accumulated as a compatible solute and osmoprotectant (65). Proline accumulates under various stresses including water deficit, low temperature, and UV radiation. To help plants survive these stress conditions, proline can act as an osmolyte involved in osmotic adjustment and can help stabilize sub-cellular structures (66). Proline levels in leaves were determined using the ninhydrin-based colorimetric assay as modified by Abraham et al., using samples that were flash frozen in liquid nitrogen and lyophilized (67). Proline concentrations were determined for samples collected at the end of the drought period (week 10) and at the end of the recovery period (week 13).

Plant growth was measured nondestructively every 2 weeks. A quadrat on four support legs at each corner (to ensure the grass blades were not touched) was used to estimate the area of coverage within a plot. Additionally, the average height of grass plants was measured using the grazing stick method to measure height at several locations within each plot by placing the stick vertical to the ground and recording the average height of the plants at that spot. At the end of the drought period, above-ground biomass was destructively harvested from one half of each plot, while the other half of the plot remained untouched for the recovery period of the experiment. Above-ground plant material was cut to 1.5 cm above the soil surface to determine fresh and dry plant biomass. Soil cores were harvested from 18 plots (three from each species and treatment) to assess the allocation of fine root biomass as a physiological response to drought stress. From within each plot, three root samples were taken with a soil corer to a depth of 60 cm. The samples were divided into three depth intervals: top (0–20 cm), middle (20–40 cm), and bottom (40–60 cm). After the soil was gently washed from the roots, the roots were dried at 70°C, and the mass of dry root samples was measured.

Bacterial community sampling

Each week, phyllosphere bacterial community DNA was extracted using the Nucleospin Plant II Extraction Kit (Machery-Nagel, Düren, Germany) following a modified protocol. Three whole leaves of each plant were aseptically removed and placed into a 50 mL conical tube with 1.5 mL of NucleoSpin Type-B beads and 1.6 mL of Buffer PL1. Tubes were vortexed horizontally for 5 min at room temperature. The lysate was incubated for 60 min at 65°C, placed in a NucleoSpin Filter tube, and centrifuged for 2 min at 11,000 × g. The filtrate was added to 1.6 mL Buffer PC and the extraction was continued following the manufacturer’s protocol. Extracted DNA samples then underwent two subsequent PCR amplifications to attach Illumina adaptor sequences and barcodes. The first PCR step used chloroplast excluding primers 799F (5′ACA CTG ACG ACA TGG TTC TACA AAC MGG ATT AGA TAC CCKG-3′) and 1115R (5′TAC GGT AGC AGA GAC TTG GTCT AGG GTT GCG CTC GTTG-3′), targeting the V5–V6 region of the 16S rRNA gene (13). The underlined portion is the linker sequence followed by the primer sequence. The second PCR step uses the linker sequences to attach Access Array Barcodes (Fluidigm, San Francisco, CA, USA) (68). Pooled amplicons were separated into two sequencing runs that shared 10% of the same samples. The quality was assessed using the Agilent 2100 Bioanalyzer DNA High Sensitivity assay, Qubit (dsDNA High Sensitivity assay) and a library qPCR assay (NEB, Ipswich, MA, USA). The pooled libraries were spiked with ~20% PhiX control library (Illumina) to balance the low nucleotide base diversity. Each pool was sequenced on an Illumina MiSeq Platform, using the v2-500 cycle kit, with 251 bp paired-end sequencing chemistry using the Illumina recommended protocol at the Genomics Resource Laboratory (University of Massachusetts-Amherst).

Sequence analysis

Using the QIIME2 (69) pipeline, paired-end reads were demultiplexed, merged, trimmed to 315 base pairs, and binned inferring amplicon sequence variants. Taxonomic identities were assigned using the naïve Bayes sklearn classifier trained with the 799F/1115R region of the Silva 138 database (70).

The data from the greenhouse study contained 8,292 ASVs from 403 samples. After filtering out sequences from chloroplasts and mitochondria, all samples were rarefied to 1,500 reads, which resulted in the loss of 17 samples due to low read depth. The remaining 386 samples ranged from 1,559 to 28,181 reads/sample for a total of 3,417,569 quality reads. Because of the large number of samples, sequencing had to be performed in two separate runs. Fifteen samples overlapped between the two different sequencing runs to ensure no sequencing bias was introduced. The samples were assessed to determine if alpha diversity values (Shannon, Pielou’s evenness, Chao1, and observed richness) and community composition were the same between the sequencing runs (data not shown). After confirming that no differences existed between the results of the two sequencing runs, sequences were combined, and further analysis was performed with the merged data set.

Analysis of community function

Bacterial biomass sampling

The mass of three leaves placed in 50 mL conical tubes was measured before 10 mL of phosphate-buffered saline (PBS) was added, incubated at room temperature for 1 hour, and then vortexed horizontally (Vortex-adaptor, Qiagen, Germantown, MD, USA) at full speed for 10 min. The PBS leaf wash was collected, and samples were then fixed with fresh paraformaldehyde at a final concentration of 3.7% and stored at 4°C. Samples were filtered onto black polycarbonate membrane filters (pore size: 0.2 µm, 25 mm diameter) (Steriltech Corporation, Kent, WA, USA), stained using 0.1% acridine orange for 3 min, and analyzed with epifluorescence microscopy by counting 20 fields using SimplePCI (Hamamatsu, Japan) (71, 72).

Nitrogen fixation rate

To determine the rate of biological nitrogen fixation, stable isotope tracing was conducted at six different time points. Three samples were taken during the drought period (weeks 6, 7, and 10) and one each week during the recovery period (weeks 11, 12, and 13). Nitrogen fixation rates were determined by measuring the incorporation of the stable isotope ¹⁵N from ¹⁵N₂ into the leaf tissue. Leaf disks of known area were incubated in an artificial atmosphere containing 80% ¹⁵N₂ (Sigma-Aldrich, USA) and 20% O₂ (Airgas, USA) for 48 hours under ambient light conditions and ambient temperature. Corresponding control samples were incubated in a normal atmosphere to determine the natural ¹⁵N abundance background. After incubation, the leaf disks were dried at 70°C, weighed, finely ground, and 1–2 mg of plant powder weighed into tin capsules. Samples were sent to the SILVER lab of the University of Vienna to determine ¹⁵N incorporation using a continuous-flow isotope ratio mass spectrometer, consisting of an EA-Isolink elemental analyzer, a Conflo IV interface, and a Delta V Advantage isotope ratio mass spectrometer. Nitrogen fixation rates were determined using the following equation where N_leaf is foliar N concentration (mg g⁻¹), M_r is the molecular weight of ¹⁵N, and t is incubation time (15):

$${\displaystyle {\mathrm{N}}_2-\text{Fix }={\mathit{N}}_{\text{leaf }}\times \left({\mathrm{a}\mathrm{t}\mathrm{\%}}^{15}{\mathit{N}}_{\text{sample}}-{\mathrm{a}\mathrm{t}\mathrm{\%}}^{15}{\mathit{N}}_{\text{control }}\right)/100\times {10}^3/{\mathit{M}}_{\mathit{r}}/\mathit{t}}$$

Bacterial DNA samples corresponding to each stable isotope probing timepoint were used to determine the absolute quantity of nitrogen-fixing bacteria per leaf area for each of the grass species and treatment using qPCR of the nifH gene. All samples were amplified in triplicate using the PolF (5′ TGC GAY CCS AAR GCB GAC TC 3′) and PolR primers (5′ ATS GCC ATC ATY TCR CCG GA 3′) (73) in a reaction containing Luna qPCR Master Mix (NEB, Ipswich, MA, USA), 0.75 µM forward primer, 0.75 µM reverse primer, 1 µL DNA, and molecular-grade water. The qPCR reaction was carried out using 5 min of initial denaturation at 95°C, 40 cycles of 95°C for 30 s, 60°C for 30 s, and 72°C for 30 s, followed by a final 10 min elongation step at 95°C. The abundance of the nifH gene was normalized to the number of nifH genes per gram of leaf mass.

Statistical analyses

Generalized linear mixed models were used to independently assess changes in bacterial biomass, alpha diversity [Shannon diversity index, observed species richness (number of ASVs), estimated species richness (Chao1), and Faith’s phylogenetic diversity], nifH abundance, and nitrogen fixation rates. The fixed effects used in the models were treatment, host species, and sampling time; sample ID was incorporated as a random effect to account for resampling over time. All models used gamma distribution with a log link since the data consisted of continuous variables bounded at zero. Models were run using the lme4 package (74) in R language, and the effects of each variable were determined using Tukey’s HSD (Tukey’s Honest Significant Differences) tests. Differences in the relative abundance of bacterial community groups were determined using analysis of variance (ANOVA) followed by Tukey’s HSD tests as implemented in R language. Hierarchical clustering using Euclidean distance and average linkage was performed on relative abundance data at the class level using the hclust function from the Vegan package in R (75). Nitrogen fixation rates and nifH abundance were correlated to dominant classes of bacteria using Pearson correlation and linear regression with the ggpubr (76) package in R.

To understand how phyllosphere bacterial communities on different host species were changing over time, whether they were impacted by drought stress, and if they were subsequently able to recover from drought, we conducted a permutational analysis of variance using Bray-Curtis distances and used beta dispersion to assess inter-sample variability within host species, treatments, and over time. Results were visualized using non-metric multidimensional scaling (NMDS). PERMANOVA, beta dispersion, and NMDS were conducted using the vegan package (75) and visualized using ggplot2 in R language (77). Paired PERMANOVAs were conducted by sub-setting host species and/or sampling day. P values from ANOVA and PERMANOVA analyses were adjusted for multiple comparisons using the Benjamini and Hochberg method (78).

Ecological null modeling

Ecological null modeling was performed to explore community assembly processes in the phyllosphere over time, on different host species, and how these processes are influenced by water deficit and irrigation during the drought and recovery periods. This was done using the Stegen framework (18) by calculating β-nearest taxon indices and Raup-Crick values (Bray-Curtis) (RC_BC) to understand contributions from deterministic and stochastic processes, respectively. Using the approach by Danczak et al. (19), β-mean nearest taxon distance (βMNTD) was first calculated for every pairwise comparison and for 999 random distributions using the comdistnt function in the picante package in R (79). βNTI was then calculated by comparing the observed βMNTD values to the randomization values. Once βNTI was calculated, pairwise comparisons between communities were made to understand phylogenetic turnover and to distinguish between stochastic and deterministic processes. If |βNTI| was greater than 2, differences in the communities were a result of deterministic processes. If, however, |βNTI| was less than 2, communities underwent a second null modeling step using RC_BC to determine if community assembly was a result of stochastic processes. βNTI is initially used to determine if assembly processes are deterministic and then further to differentiate deterministic processes into variable and homogenous selection processes (Table 3).

TABLE 3.

Definition of terms and associated outcomes from the ecological models

Term	Definition	Process	Results
Variable selection	Microbial populations diverge due to selective pressures, i.e., differences in physical conditions.	Deterministic	βNTI > 2
Homogeneous selection	Microbial populations converge due to selective pressures causing similar community compositions, i.e., similarities in physical conditions.	Deterministic	βNTI < −2
Dispersal limitation	Microbial populations separated by time or space are unable to mix resulting in ecological drift driving community divergence.	Stochastic	|βNTI| < 2; RCBC < 0.95
Homogenizing dispersal	Microbial populations are freely interacting resulting in microbial exchange between communities. Mixing between populations results in more similar communities between samples.	Stochastic	|βNTI| < 2; RCBC < −0.95
Undominated	Occurs when a single assembly process is unable to explain the variation because both deterministic and stochastic processes are influencing assembly.	Both	|βNTI| < 2; |RCBC| < 0.95

Deterministic and stochastic processes can be further classified. If βNTI > 2, community assembly is a result of variable selection or when differences in communities are more different than is expected by random chance. Alternatively, if βNTI < −2, communities are significantly more similar than is expected by random chance, resulting from homogenous selection. When βNTI did not indicate strong deterministic assembly processes (|βNTI| < 2), RC_BC was calculated to determine how dispersal and drift influenced community assemblages. RC_BC was calculated by creating null communities based on the microbial communities from 9,999 permutations for each pairwise comparison, which were then used to calculate a null distribution of Bray-Curtis values and compared to the observed Bray-Curtis values. The results are normalized from +1 to −1 and represent deviations of the observed values compared to the null values. RC_BC > 0.95 suggests community assembly is influenced by dispersal limitation, which occurs when communities are significantly different from each other due to their inability to mix over time and/or space, which results in drift. RC_BC < −0.95 indicates homogenous dispersal in which communities are similar to each other due to their ability to mix in their environment. However, if |RC_BC| < 0.95, differences in communities are considered undominated, or, in other words, cannot be explained by a single process. Since RC_BC can only be used to distinguish stochastic processes, samples with |βNTI| > 2 were not included in the RC_BC calculation.

DATA AVAILABILITY

The 16S rRNA gene sequences were deposited in the NCBI Sequence Read Archive (SRA) under BioProject ID PRJNA1013382.

SUPPLEMENTAL MATERIAL

The following material is available online at https://doi.org/10.1128/aem.01750-23.

Supplemental Tables S1-S3, Supplemental Figures S1-S8. aem.01750-23-s0001.docx.

Additional experimental details shown in 3 supplemental tables and 8 supplemental figures.

ASM does not own the copyrights to Supplemental Material that may be linked to, or accessed through, an article. The authors have granted ASM a non-exclusive, world-wide license to publish the Supplemental Material files. Please contact the corresponding author directly for reuse.