Divergence and convergence in epiphytic and endophytic phyllosphere bacterial communities of rice landraces

ABSTRACT

Phyllosphere-associated microbes can significantly alter host plant fitness, with distinct functions provided by bacteria inhabiting the epiphytic (external surface) vs endophytic niches (internal leaf tissue). Hence, it is important to understand the assembly and stability of these phyllosphere communities, especially in field conditions. Broadly, epiphytic communities should encounter more environmental fluctuations and frequent immigration, whereas endophytic microbiota should face stronger host selection. As a result, we expect greater variability in epiphytic than endophytic communities. We analyzed the structure and stability of leaf phyllosphere microbiota of four traditionally cultivated rice landraces and one commercial variety from northeast India grown in the field for 3 consecutive years, supplemented with opportunistic sampling of eight other landraces. Epiphytic and endophytic bacterial communities shared dominant core genera such as Methylobacterium and Sphingomonas. Consistent with an overall strong environmental effect, both communities varied more across sampling years than across host landraces. Seeds sampled from a focal landrace did not support vertical transmission of phyllosphere bacteria, suggesting that both types of communities are assembled anew each generation. Despite these points of convergence, epiphytic communities had distinct composition and significantly higher microbial load and were more rich, diverse, modular, and unstable than endophytic communities. Finally, focused sampling of one landrace across developmental stages showed that the divergence between the two types of communities arose primarily at the flowering stage. Thus, our results show both convergent and divergent patterns of community assembly and composition in distinct phyllosphere niches in rice, identifying key bacterial genera and host developmental stages that may aid agricultural interventions to increase rice yield.

IMPORTANCE

Phyllosphere (leaf-associated) microbes significantly impact plant fitness, making it crucial to understand how these communities are assembled and maintained. While many studies have analyzed epiphytic (surface) phyllosphere communities, we have a relatively poor understanding of endophytic communities which colonize the very distinct niche formed inside leaf tissues. We found that across several rice landraces, both communities are largely colonized by the same core genera, indicating divergence at the species level across the two leaf niches and highlighting the need to understand the mechanisms underlying this divergence. Surprisingly, both epiphytic and endophytic communities were only weakly shaped by the host landrace, with a much greater role for environmental factors that likely vary over time. Thus, microbiome-based agricultural interventions for increasing productivity could perhaps be generalized across rice varieties but would need to account for the temporal instability of the microbiota. Our results thus highlight the importance of data sets such as ours—with extensive sampling across landraces and years—for understanding phyllosphere microbiota and their applications in the field.

INTRODUCTION

The phyllosphere is the aerial part of a plant and is home to various microbes in distinct niches. For instance, the leaf phyllosphere has microbes residing either on the surface (epiphytes) or inside the leaf tissue (endophytes). These leaf-associated microbes (henceforth “phyllosphere microbes”) can greatly impact host plant fitness, ranging from nutrient acquisition to resistance against pathogens [reviewed in references (1–4)]. Hence, understanding the assembly and maintenance of phyllosphere microbiota is an important research focus. Broadly, the phyllosphere microbiome is shaped by a combination of processes: ecological drift, host selection, environmental factors, dispersal, and evolution (5). Though patterns of host genotype- or population-specific phyllosphere microbiota (3, 6–10) are consistent with a stronger host selection effect, in many plants, the phyllosphere microbiota is influenced by both host plant genotype and abiotic environmental factors such as light, temperature, humidity, and water (11–13). Recent work also suggests that the effects of host vs environment can be dynamic. In an evolution experiment with tomato plants inoculated with naturally derived phyllosphere communities, both host genotype and environment had strong effects on microbial community structure, but the host genotype effect weakened across passages (14). Typically, we expect strong host selection to reduce stochasticity in community assembly, whereas strong environmental effects are expected to introduce more variation across the microbiota of conspecific plants. Indeed, temporal (including seasonal) variation in phyllosphere microbiota is commonly observed across multiple plant species (13, 15–17). However, most prior studies have focused on the epiphytic community, with far fewer analyses of the impact of host vs environmental factors on the leaf endosphere.

The epiphytic and endophytic plant niches are quite distinct and impose different selection pressures on colonizing microbes (1). Bacteria in each niche thus exhibit different suites of traits, provide distinct benefits to the host, contribute differentially to global nutrient cycles, and respond in distinct ways to climate change (9, 10, 18). An interesting example comes from the pathogens Pseudomonas syringiae which transitions from an early epiphytic commensal phase to a later pathogenic endophytic stage, with associated changes in suites of key genes expressed in each stage (19). Colonization of epiphytic vs endophytic niches also involves different host vs environment effects and ecological processes governing community assembly. For instance, we expect that the endophytic community should experience stronger host selection, with a relatively weaker influence of external environmental factors. In contrast, we expect the epiphytic community to be more strongly affected by the environment because it is continually and directly exposed to a large influx of microbes from several sources such as soil, air, and water. Additionally, local dispersal of microbes from neighboring plants can contribute to the assembly of the phyllosphere epiphytic microbial community (20). Lastly, we also expect to see distinct communities and core membership across different plant niches (18). However, very few studies have directly tested for divergence in community assembly and maintenance of epiphytes vs endophytes. The few studies that have been conducted report variable results. In the endophytic microbial community associated with Acacia (13) and rice plants (21), a stronger host effect was observed than for epiphytes; host effects were similar for endophytic and epiphytic communities of olive trees (22); whereas in sugar maples, the host effect was greater in epiphytic than the endophytic phyllosphere community (23). Thus, the effect of host vs environment may not be easily generalizable, and it is unclear whether the variability arises due to differences in host species or differences and limitations in sampling and experimental design (e.g., sampling only one plant growth stage). Hence, a deeper understanding of differences between epiphytic and endophytic phyllosphere microbiota requires monitoring of the dynamics of both phyllosphere communities across the same host species and across time and plant growth stages.

Another important reason to compare epiphytic vs endophytic communities is to better understand variation in core community members that may be most relevant for plant fitness. A large body of work has demonstrated important functions provided by core phyllosphere bacteria, including nitrogen fixation, resistance against pathogens, and drought tolerance (4, 24). Bacteria from the genera Methylobacterium and Sphingomonas from alpha-proteobacteria are well-known examples of beneficial symbionts for many plants (25–27), enhancing seed germination (28, 29), promoting plant growth (27, 30, 31), and providing resistance against pathogens (32, 33) Additionally, we recently demonstrated substantial yield benefits from a specific Methylobacterium strain on its host rice plants in field conditions, though the effect of Methylobacterium inoculation was not generalizable across host-bacterium pairs (34). The impacts of specific bacteria may differ not only across hosts but also in a community context and with environmental variation. For instance, inoculation of the phyllosphere microbiota on tomato plants confers resistance against the pathogen P. syringae, but this effect is contingent on available nutrient resources (35). More generally, the action of specific microbes can either facilitate or hinder the success of subsequent colonizers (36). However, in such studies, it is difficult to identify which community members provide the observed beneficial function. One way to begin approaching this problem is to identify sets of core microbes that consistently co-occur at high abundance despite temporal variation at the community level. Such co-occurring sets of taxa may indicate interacting partners that respond to host-imposed selection, and testing for unique co-occurring taxa across plant niches may thus allow the identification of niche-specific, putatively interacting bacteria.

We addressed these gaps using multiple rice landraces grown in the state of Manipur in northeast India. These local rice varieties have been selected for specific traits (such as yield, aroma, and grain color) for many generations, resulting in substantial phenotypic and genotypic diversity (37–40). We surmised that this host diversity might drive differential selection on bacterial taxa associated with each landrace, and we quantified phyllosphere community dynamics in leaf niches across time. We collected leaf samples from four rice landraces and one commercial high-yielding variety (HYV) across three consecutive growing seasons (Fig. S1A). Using 16S rRNA amplicon sequencing, we jointly analyzed leaf epiphytic and endophytic communities. Our work thus quantifies the impacts of various factors in shaping phyllosphere communities and provides a deeper understanding of host-plant-associated bacterial communities in the field.

MATERIALS AND METHODS

Sample collection

We collected leaf samples of flowering rice plants from two sites located ~20 km apart in Manipur in 2019, 2020, and 2021 (Table S1). We focused on four phenotypically distinct rice landraces that were consistently grown by the farmer during these years, i.e., Chakhao (CK), Phouren-mubi (PM), Phoungang (PN), and Tolenphou (TP) and one commercially developed high yielding rice variety (HYV) (Table S1). Note that different landraces were cultivated at each site (Table S1). For each landrace, we randomly chose 5–10 plants from the field, cut a 3–4 cm length of the flag leaf from each plant, and immediately immersed it in 1 mL of 1× Redford buffer (1M tris HCl, 0.5M EDTA, and 1.24% Triton) placed on ice, as described earlier (34) (Fig. S1A). In 2021, we similarly collected samples of leaves from different plant growth stages from landrace CK (Fig. S1B). Finally, in 2019 and 2020, we sampled seeds from landrace CK. After harvesting, seeds were transported in plastic bags to the laboratory, where they were stored at –80°C.

Determining phyllosphere and seed microbiota composition

We followed our previously described protocol to extract DNA for epiphytic bacterial communities (34). Briefly, leaf samples were processed immediately after transport to the laboratory for epiphytic DNA extraction. Tubes were placed horizontally on a vortex for 2 hours with intermittent shaking to dislodge surface bacteria, after which we removed the leaves. We centrifuged tubes at 10,000 rpm for 25 min and used the pellet for extracting DNA as per the Qiagen DNeasy blood and tissue kit instructions. Simultaneously, the removed leaves underwent two rounds of washing using sterile water, after which we confirmed the absence of leaf surface bacteria by plating on agar media. We stored washed leaves at –80°C for endophytic DNA extraction. These stored samples were washed once with PBS (phosphate buffer saline) and crushed using liquid nitrogen. We used the powdered leaf (~100 mg) for DNA extraction using the same kit as above, with the following modifications. We added 180 µL of lysis buffer and 20 µL proteinase K and incubated at 56°C overnight. To remove leaf debris, we used a brief centrifugation step and used the supernatant for subsequent steps as recommended in the kit instructions. We used 30 µL of nuclease-free water for eluting DNA. We note that 2019 samples were processed using a different kit (PowerSoil DNA isolation kit, Mo Bio). For seed microbiota, we extracted epiphytic DNA using three seeds/replicate/landrace as described above, in 500 µL Redford buffer. For seed endophytic bacteria, we removed the seed husk and surface-sterilized seeds using immersion in 70% ethanol. Sterilized seeds were crushed using liquid nitrogen, and 100 mg of tissue was used for DNA extraction.

We quantified the DNA concentration for all samples using Qubit and then amplified the V3–V4 hypervariable region of the 16S rRNA using standard Illumina primers (forward 5′-TCGTCGGCAGCGTCAGATGTGTATAAGAGACAGCCTACGGGNGGCWGCAG-3′; reverse 5′-GTCTCGTGGGCTCGGAGATGTGTATAAGAGACAGGACTACHVGGGTATCTAATCC) (41). We used polypeptide nucleic acid to block the amplification of plant mitochondrial and chloroplast DNA in our 16S rRNA PCR, as described previously (42). Finally, we carried out amplicon sequencing on the Illumina Miseq platform (300 × 2 paired-end).

qPCR to quantify the total bacterial load

We used qPCR to quantify the total bacterial load for 2021 samples of the four focal landraces (CK, PM, PN, and TP) for both epiphytes and endophytes and across different growth stages of landrace CK in the same year. We designed new primers (forward: 5’- CGGTAATACGGAGGGTGCAA-3′; reverse: 5’- TCCTCCAGATCTCTACGCAT −3’) to selectively amplify 200 bp of the bacterial 16S rRNA sequence. We used primers for the rice gene OsACTIN1 (forward: 5′-ATGAAGTGCGACGTGGATATTAG-3′; reverse: 5’GGGCGACCACCTTGATCTTC-3′) to estimate the amount of host DNA from endophytic samples and used it as an internal control for both epiphytic and endophytic samples (recall that these were paired analyses, conducted using the same leaf sample, as described above). We used 1 µL of DNA as a template for qPCR (BioRad CFX system) and followed the qPCR program: initial denaturation at 95°C for 10 min, 45 cycles of 95°C for 30 s, 75°C for 10 s, 55°C for 30 s, 72°C for 50 s, and finally at 65°C for 0.05 s and 95°C for 0.5 s. We tested five biological replicates/landrace/community type. For each biological replicate, we carried two technical replicate qPCR reactions and normalized Ct values for the target gene against OsACTIN1 to obtain the normalized total bacterial load as 2^−ΔCt, where ΔC_t = C_{t(16S rRNA gene)} – C_t(actin). This quantity approximates the amount of bacterial DNA per unit host DNA.

Statistical analysis

We analyzed and visualized data using R (43). To analyze amplicon sequencing data, we used the DADA2 workflow, using 100% sequence identity with the Silva reference database (training set v138.1) (44). After filtering out chloroplast and mitochondrial reads, we obtained an average of 50,000 reads (range 10,000–500,000) per epiphytic sample and an average of 29,000 reads (range 5,000–68,000) per endophytic sample (Fig. S2A and B). Our negative controls without template had an average of 98 reads, indicating low levels of reagent contamination. Rarefaction analysis of a subset of samples showed that ~20,000 reads/sample for epiphytic samples and ~10,000 reads/sample for endophytic samples leads to near-saturation of community richness (Fig. S2C). Several endophyte samples initially had low read depth, and this problem was not always solved by additional sequencing. Samples that had fewer reads than the threshold noted above were excluded from further analysis. Bacterial taxa with <1% of the total reads were combined into a single bin as “rare taxa” or “others,” only for visualization. For all statistical analyses, the full matrix of relative abundances of each taxon was used, unless specified otherwise.

To quantify the impact of different factors (e.g., landrace and sampling year) on community composition, we first calculated Bray-Curtis distances between the samples and carried out PERMANOVA (permutational analysis of variance) using the function “adonis2” in the package “vegan” (45). We used this distance matrix to carry out Canonical Analysis of Principal Coordinates based on Discriminant Analysis (CAPdiscrim) (46) using the R package “Biodiversity R” (47), permuting 1,000 times. Two of the dominant linear discriminants were plotted to visualize each cluster with ellipses reflecting 95% confidence intervals using the function “ordiellipse” in the R package “vegan” (45). We used the package “ggplot2” (48) for visualization. We used matrix2png for visualizing the heatmaps (49).

To compare the richness and Shannon diversity index for epiphytes vs endophytes, we carried out paired t-tests. To estimate the effect of landrace in a given year, we compared microbiota from all the landraces collected in that year. To estimate the effect of sampling year, we compared microbiota for the four focal landraces across all years. We also used ANOVA (Analysis of Variance) for testing the interaction effect of landrace, year, community type, or growth stage on both epiphytic and endophytic microbial communities. We used the R package NST (Normalized Stochasticity ratio) (50) to measure taxonomic beta diversity (β_{rc -} Raup-Crick Dissimilarity) as a measure of the degree of individual variation across community types in the focal rice landraces (51). This metric ranges from −1 to +1, with values of +1 indicating that communities are more dissimilar across individuals, and values of −1 indicating more similar communities across individual samples (51) (in our case, each sample was from a different individual plant).

For network analysis, we first used the relative abundance data set (four focal landraces for 3 years) to generate a correlation matrix by calculating all pairwise Spearman’s rank correlations for all bacterial genera (combining data for multiple ASVs (Amplicon Sequence Variant) assigned to a given genus) across all epiphytic as well as endophytic samples. We filtered out bacterial pairs where the absolute correlation coefficient was less than 0.5 and P > 0.05, retaining only significant pairs to construct the occurrence network of bacterial taxa. Finally, we used Gephi (52) for visualizing and calculating the modularity of the network.

RESULTS

Since epiphytic communities are more exposed to random environmental fluctuations as well as frequent immigration, we expect greater stochasticity in epiphytic community assembly and maintenance, with a stronger role of deterministic processes (such as host selection) in shaping the endophytic community. Accordingly, we outlined the following expectations regarding differences between epiphytic and endophytic communities:

1. Higher bacterial richness and diversity in epiphytic communities, in all landraces and years.

2. Weaker effect of host rice landrace and stronger effect of sampling year on epiphytic community composition.

3. Lower consistency and stability of epiphytic communities across landraces and years.

Epiphytic and endophytic phyllosphere communities converge on similar core genera in all host landraces

Before testing the specific hypotheses outlined above, we describe the broad structure and composition of the phyllosphere communities. As expected from prior work (2, 3, 53), the phyllosphere communities of all four traditional rice landraces were dominated by the genera Methylobacterium, Sphingomonas, Pantoea, Aureimonas, and Pseudomonas (Fig. 1). Remarkably, both epiphytic and endophytic communities showed the same dominant taxa, and this was also true for the high-yielding variety (Fig. S3A) and for the eight additional rice landraces that were collected sporadically (Fig. S3B through D). The only exception was the landrace Moirangphou (MP), where Methylobacterium was not dominant, and instead, Rhodococcus and Acinetobacter had higher abundance (Fig. S3B). We also observe that the “others” bin was much higher for CK and PM sampled in 2019 (Fig. 1A and B). For epiphytes, these “others” were largely composed of rare taxa, but for endophytes, ~26% and ~13% of the total microbiota were composed of unassigned taxa, for CK and PM, respectively. Of the dominant taxa, Methylobacterium was the most abundant genus (on average ~20%–40% of all reads in epiphytic communities and 10%–70% in endophytic communities; Fig. 1; Fig. S3). At a lower taxonomic level, four Methylobacterium ASVs (ASV1, ASV3, ASV6, and ASV12) accounted for ~10%–30% of the total Methylobacterium abundance as epiphytes and endophytes and may be potentially important strains for the rice plants (Fig. S4). These patterns were also observed in a core microbiome analysis that accounts for both prevalence (across samples) and abundance of community members (Fig. 2). Methylobacterium and Sphingomonas were both dominant and consistently present across the four focal landraces (~98% prevalence in samples with high relative abundance; Fig. 2). However, Pantoea and Aureimonas were relatively less prevalent and typically occurred at lower relative abundance (~50% prevalence across the samples with low relative abundance), in epiphytic and endophytic communities, respectively. Thus, at the level of core community membership, the epiphytic and endophytic communities were remarkably convergent across individual plants, as well as across years and host landraces.

Fig 1.

Dominant taxa in phyllosphere microbiota. Stacked bar plots show the relative abundance of bacterial ASVs (genus level) for each landrace sampled in each year. (A) CK. (B) PM. (C) PN. (D) TP. Sample size (number of replicate plants) is indicated in parentheses. Taxa with <1% reads, or that were not assigned to known bacteria in the database, were combined into “others.”

Fig 2.

Similar core genera comprise epiphytic and endophytic phyllosphere communities. Heatmaps show the prevalence of core bacterial genera in (A) epiphytic and (B) endophytic communities. Prevalence was calculated as the percentage of all samples where a given taxon was present with the indicated threshold of relative abundance. For instance, Methylobacterium is prevalent (found in most samples) across a wide range of relative abundance (0.01–0.1), whereas Pantoea is prevalent only when rare.

Phyllosphere epiphytic bacterial communities tend to be more diverse, rich, and abundant than endophytic bacterial communities

Using the microbiota data and focusing on the four traditionally cultivated rice landraces, we next tested our specific predictions about differences between epiphytic and endophytic communities. Overall, microbiota composition across individual plants from a given landrace in a given year tended to be similar (Fig. 1; Fig. S3). However, in all cases, epiphytic communities were significantly more dissimilar across individual plants, compared to endophytic communities (Fig. 3), supporting our prediction that the epiphytic community is more vulnerable to stochastic processes. Epiphytic bacterial communities of all landraces also tended to be more diverse than endophytic communities in all years (Fig. 4A; also see Fig. S5), though the difference was not always statistically significant (results of paired t-tests comparing epiphytic vs endophytic communities for a given landrace and year are shown in Fig. 4A and Fig. S5; Kruskal Wallis tests for Shannon diversity ~ community type in each year: 2019, P = 0.004, 2020, P = 1.2 × 10⁻⁵; 2021, P = 5 × 10⁻⁶). Similarly, epiphytic communities tended to have higher richness in all cases (Fig. 4B; Fig. S5). Again, the landrace MP was exceptional, showing an opposite pattern in 2020, with higher diversity and richness in endophytic communities (Fig. S5). Together, these results support our first prediction that epiphytic communities should have higher richness and diversity.

Fig 3.

Individual variation is higher in the epiphytic than the endophytic phyllosphere community. Boxplots show taxonomic beta diversity (β_rc) of bacterial communities of rice landraces across years. Asterisks indicate significant differences across epiphytic and endophytic niches (Wilcoxon test, P < 0.05). Sample sizes for the year 2019 range from four to five biological replicates/landrace; 2020 range from 7 to 10 biological replicates/landrace; 2021 range from 8 to 10 biological replicates/landrace (Table S1).

Fig 4.

Epiphytic microbial communities are more diverse and abundant than endophytic communities. Boxplots show (A) Shannon’s diversity and (B) the richness of bacterial communities of rice landraces across years. Asterisks indicate P < 0.05 (paired t-tests). Sample sizes for the year 2019 range from four to five biological replicates/landrace; 2020 range from 7 to 10 biological replicate/landrace; 2021 range from 8 to 10 biological replicates/landrace (Table S1). (C) Shannon diversity of communities as a function of the normalized total bacterial load. The right-hand panel shows the full data set that includes some outliers (squares); the left-hand plot represents a truncated y-axis to highlight variation in the bulk of the data set, excluding the potential outliers. Each point represents one biological replicate plant (n = 5/landrace), for each of the four focal rice landraces sampled in 2021. (D) Normalized total bacterial load (bacterial DNA/host DNA) for phyllosphere communities in a given landrace and year 2021 (asterisks indicate P < 0.05, paired Wilcoxon tests) (ANOVA, load ~ landrace × community, P _landrace = 0.005; P _community = 0.048, P _{landrace x community} = 0.03).

Did community diversity and richness reflect total bacterial load, i.e., success of colonization of the phyllosphere? We tested this using samples collected in 2021. We did not find a correlation between bacterial load and either Shannon diversity (epiphytes, Spearman’s rho = 0.33, P = 0.14; endophytes, Spearman’s rho = 0.35, P = 0.12; Fig. 4C) or richness (Fig. S6), suggesting that community diversity and richness do not necessarily correspond with bacterial colonization success. For instance, despite substantially larger microbial loads in the landraces TP and PN, bacterial diversity was similar across landraces in 2021, suggesting that the same taxa were colonizing more successfully in these two landraces. However, bacterial loads did vary substantially across landraces, with PN showing the highest loads for the epiphytic community—an order of magnitude higher than other landraces (Fig. 4D). Additionally, epiphytic communities had higher total loads than endophytes (Fig. 4D), potentially reflecting greater exposure to the environment and weaker selection from the host. Thus, patterns of variation in microbiota community composition (diversity and richness), as well as load, broadly support our first prediction of greater stochasticity and environmental impacts on epiphytic communities.

Sampling year has stronger effects than the host landrace on both epiphytic and endophytic communities

Next, we tested our second prediction that host rice landrace identity should have a stronger impact on endophytic compared to epiphytic bacterial communities, whereas sampling year should have a stronger effect on epiphytic communities. Overall, both landrace and sampling year influenced the richness and diversity of both sets of communities. We found significant impacts of landrace and marginally significant effects of landrace × year interaction for both epiphyte and endophyte community richness, but a significant effect of sampling year only for endophytes (ANOVA, richness ~ landrace × year: for epiphytes, P _landrace = 2.8 × 10⁻⁹, P _year = 0.15, P _{landrace × year} = 0.047; for endophytes, P _landrace = 2.9 × 10⁻⁵, P _year <2 × 10⁻¹⁶, P _{landrace × year} = 0.05). In the case of community diversity, we found significant variation only across years but not across landraces for epiphytes (ANOVA, Shannon diversity index ~ landrace × year: P _landrace = 0.18, P _year = 1.3 × 10⁻⁶, P _{landrace × year} = 0.2), whereas for endophytes, both landrace and year, had significant impacts (ANOVA, Shannon diversity index ~ landrace × year, P _landrace = 0.03, P _year = 1.3 × 10⁻¹⁰, P _{landrace × year} = 0.6). Overall, both landrace identity and sampling year influenced endophytes more consistently than epiphytes, suggesting that epiphytic diversity and richness may also vary as a function of other unknown factors.

We then tested the impact of landrace, field site, and sampling year on community composition. Note that different landraces were cultivated at each site, such that we cannot disentangle the effect of site with landrace identity. We visualized the effect of landrace using linear discriminant analysis, which showed clustering of both epiphytic and endophytic microbial communities across years (Fig. S7). As expected, sampling year had a slightly stronger impact on epiphytic communities compared to endophytes, explaining 20% and 18% of variation, respectively (PERMANOVA; Table 1). Similarly, as expected, host landrace had a slightly higher effect on endophytic communities (4% in epiphytic communities and 6% in endophytic communities; Table 1). Field site had a relatively low and similar impact on both communities (4% variation explained in both cases; Table 1). On including two additional landraces sampled in 2 of the 3 years and the high-yielding variety, the effect of the host landrace on both epiphytic and endophytic communities became stronger (8% and 11%, respectively; Table S2A). However, this was primarily driven by MP, which had a very distinct community composition, as described above (Fig. S3). Within a given year, the effect of host landrace was generally much higher, ranging from 17% to 26% for epiphytic communities and 22%–38% for endophytic communities (Table S2A). However, overall, the sampling year had a much stronger effect than either landrace or site on the composition of both types of communities (Table 1). This was true even when we removed samples from 2019 (Table S2B) that were processed using a different DNA extraction kit (see Materials and Methods). Together, these results provide some support for our second prediction, though the magnitude of differences between epiphytic and endophytic communities was low.

TABLE 1.

Effect of landrace, field site, year, and community type on phyllosphere microbial communities^a

Community	Year	Variable(s)	R 2	P
Epiphytic	2019	Landrace	0.18	0.04
		Site	0.19	0.0009
	2020	Landrace	0.15	0.0009
		Site	0.11	0.0009
	2021	Landrace	0.12	0.0009
		Site	0.11	0.0009
	All	Landrace	0.04	0.0009
		Site	0.04	0.0009
		Year	0.20	0.0009
		Landrace × year	0.07	0.0009
		Site × year	0.06	0.0009
Endophytic	2019	Landrace	0.21	0.0009
		Site	0.17	0.0009
	2020	Landrace	0.13	0.0009
		Site	0.08	0.0009
	2021	Landrace	0.19	0.0009
		Site	0.09	0.0009
	All	Landrace	0.06	0.0009
		Site	0.04	0.0009
		Year	0.18	0.0009
		Landrace × year	0.08	0.0009
		Site × year	0.04	0.0009
Combining all years, landraces, and communities		Landrace	0.03	0.0009
		Site	0.02	0.0009
		Year	0.11	0.0009
		Community type	0.06	0.0009
		Landrace × year	0.04	0.0009
		Site × year	0.03	0.0009
		Landrace × community type	0.02	0.0009
		Site × community type	0.01	0.0009
		Year × community type	0.04	0.0009
		Landrace × year × community type	0.036	0.0009
		Site × year × community type	0.014	0.0009

^a Table shows the output of PERMANOVA analysis of communities from the four focal landraces and all landraces. All P-values shown are significant.

Finally, using a full model, we tested the effect of landrace, year, and community type on community composition, observing that sampling year explained the highest amount of variation in the data set (11%), with weaker effects of landrace (3%), field site (2%), and community type (6%), and significant but weak interactions between variables (Table 1). Together with the high variation in communities across years, these results suggest that both types of phyllosphere communities are more strongly structured by environmental factors rather than host identity, with relatively weak differential effects across epiphytic vs endophytic communities. Nonetheless, the differences were generally in the predicted direction, with epiphytic communities being more variable.

No evidence for vertical transmission of leaf phyllosphere communities

Next, we tested whether the relatively weak effect of host landrace on the phyllosphere microbiota could arise from inconsistent vertical transmission of microbial communities across generations. For this analysis, we used seeds of the landrace CK, finding that the seed epiphytic and endophytic communities both varied significantly across years, with a higher impact of sampling year on the epiphytes (epiphytes, PERMANOVA, seed_{2019 vs 2020}, R² = 0.34, P = 0.0009; endophytes, PERMANOVA, seed_{2019 vs 2020,} R² = 0.18, P = 0.0009; Fig. 5). In contrast, the leaf microbiota were more stable, and many of the bacterial genera detected in the seed epiphytic or endophytic communities did not colonize the leaf epiphytic or endophytic niches in the next generation (Fig. 5). Indeed, sampling year had a larger impact on community structure than plant niche or community type [PERMANOVA, effect of niche (leaf vs seed): R² = 0.08, P = 0.0009; effect of community type: R² = 0.03, P = 0.0009; effect of year: R² = 0.13, P = 0.0009; niche × community type: R² = 0.02, P = 0.004; niche × year: R² = 0.04, P = 0.0009; community type × year, R² = 0.04, P = 0.0009, niche × community type × year: R² = 0.015, P = 0.0009]. Together, these results support the idea that weak vertical transmission via seeds contributes to stronger environmental effects on rice phyllosphere microbiota.

Fig 5.

Leaf and seed communities vary across generations of CK plants. Stacked bar plots show the relative abundance of bacterial ASVs (genus level) across the leaf and seeds of CK as (A) epiphytes and (B) endophytes. Sample size (number of replicate plants) is indicated in parentheses. Taxa with <1% reads, or that were not assigned to known bacteria in the database, were combined into “others.”

Epiphytic bacterial communities have co-occurrence networks with more positive correlations, indicative of lower stability

Lastly, to test our third prediction about the relative stability and consistency of epiphytic vs endophytic communities, we analyzed the network of co-occurrence of taxa. We estimated the correlation coefficients for the relative abundance of each pair of bacterial genera across landraces (in a given year) or across years (for a given landrace), retaining only significant correlation coefficients with absolute values greater than 0.5 (see Materials and Methods). We used these correlations to generate a co-occurrence network. Prior simulations suggest that a mixed set of correlations (i.e., some positive and some negative) increase network stability, compared to networks with only positive relationships (54). Epiphytic communities showed almost exclusively positive relationships among taxa in all 3 years, whereas endophytic communities were more balanced (Fig. 6A). This pattern held true even when we considered all landraces for a given year (Fig. S8), or if we excluded the 2019 epiphytic community of CK, which had a very high proportion of rare taxa (Fig. S9). Finally, we also compared co-occurrence networks across each of our four focal landraces, combining data across sampling years. The networks varied substantially across landraces, but we again observed higher proportions of positive relationships in epiphytic microbiota, except in the case of landrace TP, where epiphytic and endophytic community structures were similar (Fig. S10). Together, these results support our hypothesis that endophytic communities are more stable than phyllosphere epiphytic communities, across years as well as across host landraces.

Fig 6.

Epiphytic bacterial communities have more complex co-occurrence networks. (A) Networks indicate co-occurrence relationships between bacterial genera across the 3 sampling years, with each node representing a genus. Networks were constructed using data pooled for all four focal landraces (see Materials and Methods section for details). Node size and color both indicate the degree of connectivity, with larger node sizes and darker colors indicating nodes with more connections. The edge colors blue and red represent positive and negative relationships between nodes. Dominant and core bacterial genera are indicated in the networks. Bar plots show the proportion of pairwise relationships that are positive vs negative (blue vs red, respectively), in both types of communities (asterisks indicate significant differences, P < 0.05, χ² test). (B) Modularity score of epiphytic and endophytic bacterial communities of the four focal landraces across sampling years (also see Fig. S11). Numbers in parentheses indicate the total number of modules (sub-communities) detected.

Another important feature of these networks is modularity, which quantifies the number of sub-communities (modules) detected within each phyllosphere community type, normalizing for network size. We expected that the more variable and taxon-rich epiphytic communities would show higher modularity in the co-occurrence network, whereas we would see simpler, less modular networks for endophytic communities. However, epiphytic communities had higher modularity than endophytes only in 2019, though the total number of modules was higher in epiphytic communities for both 2019 and 2021 (Fig. 6B; Fig. S11). Combining data across sampling years, modularity scores were higher for epiphytic communities in only two of the focal landraces, CK and PM (Fig. S10). Similarly, combining data for all landraces, we found higher network modularity in epiphytic samples for 2019 and 2020 but not in 2021 (Fig. S12). Together, these results suggest that network modularity is highly variable across years and across landraces, even though the total number of modules is generally higher for epiphytic microbiota.

The overall variability in the co-occurrence networks is exemplified by focusing on the genus Methylobacterium, the most dominant genus in our data set, which was significantly correlated with different bacterial genera across years, in both types of communities (Fig. S13). However, Methylobacterium did have strong and consistent positive associations with the genera Aureimonas and Deinococcus in endophytic and epiphytic communities, respectively (Fig. S13).

Divergence in phyllosphere communities arises during the flowering stage of rice

Overall, the results described above support our prediction that endophytic communities of rice landraces are less variable and more robust than epiphytic communities. Does this difference arise early in the plant’s life cycle or emerge later during growth? To address this, in the last year of our fieldwork (2021), we collected leaf samples across different growth stages for one of the rice landraces in our study (CK) and analyzed changes in the microbiota as the plants developed. The epiphytic community composition differed significantly across growth stages, with several late-colonizing genera (e.g., Muribacter, Rodentibacter, Rothia, Streptobacillus, and Streptococcus) being lost or reduced in proportion as plants matured (Fig. 7A). In contrast, the composition of the endophytic community was more consistent, with 70%–80% of the microbiota composed of the genus Methylobacterium across growth stages, except in the panicle stage (with ~28% non-assigned and ~29% rare bacterial taxa, combined into the “other” category; Fig. 7A). Nonetheless, host plant growth stage had a slightly weaker effect in endophytic communities (effect of growth stage, endophytic community: R² = 0.29, P < 0.001, epiphytic community: R² = 0.23, P < 0.001; Fig. S15A), supported by a PERMANOVA analysis of full community composition showing a significant but weak growth stage × community type interaction (growth stage effect, R² = 0.14, P < 0.001; community type, R² = 0.14, P < 0.001; growth stage × community type, R² = 0.07, P < 0.001).

Fig 7.

Phyllosphere microbiota vary across growth stages of the host rice plant (CK). (A) Community composition of each plant was sampled at different host growth stages. Taxa with <1% reads, or that were not assigned to known bacterial genera in the database, were combined into “others.” (B) Shannon diversity index of epiphytic vs endophytic communities at different growth stages. n = 10 replicate plants for each growth stage; asterisks represent significant pairwise differences between community types at the specific growth stage (paired t-test, P < 0.05). (C) Total bacterial load in each plant sampled at specific growth stages (n = 5), normalized to the total leaf tissue as measured from the endophytic sample (the same leaf sample was used to extract both community types, as described in the Materials and Methods). Error bars show SE.

In the earlier stages of host growth (late vegetative and panicle formation), endophytic communities tended to have higher diversity and richness than epiphytes, but this pattern flipped at the flowering stage [pairwise tests shown in Fig. 7B; Kruskal Wallis tests, richness ~ stage (epiphytes), P = 0.0009; richness ~ stage (endophytes), P = 6.3 × 10⁻⁷; diversity ~ stage (epiphytes), P = 0.004; diversity ~ stage (endophytes), P = 2.02 × 10⁻⁶; Fig. 7B; Fig. S14]. Thus, by the time we sampled microbiota in the experiments described earlier (flowering stage), the diversity and richness patterns had already reversed and matched our predictions. Overall, the growth stage and its interaction with community type significantly influenced community richness (ANOVA, richness ~ stage × community; P _stage = 4.18 × 10⁻⁹;P _community = 0.8; P _{stage × community} = 4.16 × 10⁻⁵), while all factors significantly affected community diversity (ANOVA, diversity ~ stage × community; P _stage = 1.31 × 10⁻¹⁰; P _community = 0.000195; P _{stage × community} = 9.56 × 10⁻⁹). The decreasing diversity and richness of the endophytic community across growth stages suggests that many of the bacterial taxa are filtered out over time, potentially due to host selection.

We then asked whether the identified core bacterial taxa (shown in previous sections) were abundant throughout the growth stages or colonized at a specific growth stage of the rice plant. Many of the core bacterial genera (Aureimonas, Methylobacterium, Pantoea, Pseudomonas, and Sphingomonas) were most dominant at the flowering stage (Fig. S15B). However, the total bacterial load in leaf samples (relative to leaf tissue) did not increase at the flowering stage but instead peaked at the harvesting stage (Fig. 7C). Reflecting our previous results for 2021 communities, the total load was not correlated with either diversity or richness of either type of community (Fig. S16).

DISCUSSION

Our field study with paired sampling of epiphytic and endophytic phyllosphere communities highlights key factors governing community assembly and uncovers points of convergence as well as divergence in the microbiota of the two distinct leaf niches. We observed broad convergence in patterns of variation of both types of communities across time as well as host landraces. Very low individual variation across plants within a landrace, combined with significant differences across landraces cultivated in the same field in close proximity, suggest that the landrace effect likely reflects deterministic variation. This effect may reflect local adaptation to host-imposed divergent selection, e.g., as a function of host genotypes (55), and/or mediated by differences in host phenotypes such as leaf physical traits (36), nutrient availability (56), or production of hormones such as cytokinins (57) and ethylene (6). Further work is required to test these possibilities and to quantify the role of stochastic vs deterministic processes.

Despite the broad similarities in factors governing microbiota community composition and core community members, we observed several points of divergence between communities inhabiting the two leaf niches, beginning with distinct community composition. Epiphytic communities have higher richness and diversity, and are more variable across individual plants, as expected given their greater exposure to the environment. Interestingly, epiphytic communities also have higher bacterial loads. A network analysis revealed more modules in epiphytic communities, with greater positive associations between taxa. Previous work suggests that such communities, with more positive associations between members, are temporally more unstable (54). Our results therefore add to the growing body of work comparing epiphytic vs endophytic communities of different host plants, suggesting that their distinct roles and consequences (discussed in the Introduction) may be attributed to differences in community composition, stability, and individual variation.

Another important difference across epiphytic and endophytic communities is the relative impact of environment vs host. As expected, focusing on our four main landraces, we find a stronger impact of sampling year on the epiphytic than the endophytic community. We also expected a stronger impact of host rice landrace and more predictable community members in the endophytic community. However, the variation explained by host rice landrace is only slightly higher in endophytic communities, and the landrace effect is weaker than the effect of sampling year on both communities, even when combining data from all landraces and years. Why do we observe a stronger effect of sampling year compared to host landrace? One possibility is that temporal variation in abiotic conditions was extremely high in our sampling years, overwhelming the baseline effects of host genotype. Another cause may be the lack of vertical transmission of phyllosphere communities across generations via seeds, in turn leading to a larger impact of environmental factors (which likely vary across years). Third, despite large phenotypic differences, our host landraces may be physiologically very similar and may therefore impose similar filters on their microbiota. We find some support for this speculation; when we included two landraces with very distinct microbiota in our analysis (MP and MR), the impact of host landrace increased (compare landrace effect in Table 1 vs Table S2; Fig. S3). Finally, our microbiota analysis is relatively coarse-grained, and prior work shows substantial hidden diversity across strains of a given species (58). Thus, more detailed taxonomic and genomic analysis of the dominant microbial taxa may reveal stronger host landrace effects if we could distinguish between functionally different strains or niche-specialist species. Further analysis is necessary to distinguish between these possibilities.

We also observed distinct patterns of variation in the two types of phyllosphere communities across plant growth stages. Focusing on Chakhao rice, we find that the growth stage has a higher impact on the endophytic than epiphytic community, with the latter showing a more consistent richness and diversity over time. Similar results were previously reported in maize plants (59). Prior work with Arabidopsis thaliana also suggests greater stochasticity and dispersal limitation in epiphytic communities at the early stages and later deterministic host selection in microbiota (60, 61). Thus, our results perhaps reflect broader patterns across plant species. Microbiota variation across growth stages may be driven by developmental shifts in host secondary metabolite profiles (62), or priority effects, where the presence of specific taxa at early stages alters subsequent community composition. For example, inoculating rice seeds with the endophyte Xanthomonas sacchari altered the stem and rhizosphere endophytic community significantly at later stages (63). Such priority effects may be especially important in leaf endophytic phyllosphere communities, leaving leaf epiphytic communities more prone to the effects of immigration.

Finally, the total bacterial loads in epiphytic and endophytic niches also tended to vary, with significantly higher loads in epiphytic communities of two landraces (Fig. 3C). These results mirror prior work with Arabidopsis showing that colony-forming units in epiphytes are ~100-fold higher in the endophytic community (4). The total microbial load for both community types also varied significantly across landraces, potentially arising due to differences in leaf architecture or metabolic profiles across landraces. For instance, Salmonella enterica colonization varies with the abundance of type 1 trichomes on tomato leaves (64). Strikingly, the total microbial load in our study was not correlated with microbiota diversity or richness for either community type, and this was also true across plant developmental stages. This was surprising because we generally expect that colonization by highly beneficial core taxa should increase the microbial load, driving a negative relationship between diversity and load. We speculate that the lack of such a relationship in our samples reflects shared functional traits across several microbes, such that a high colonization rate may be achieved by any of a set of taxonomically distinct but functionally similar bacteria. Another possibility is that plants may select for groups of interacting bacteria rather than single species or strains, allowing for high diversity to be maintained along with high microbial loads. This possibility is consistent with the strongly modular network architecture observed in our data set. Yet another possibility is that high microbial loads reflect transient pathogen infection rather than colonization by commensals or symbionts. However, this is unlikely in our case because we did not find known pathogens such as Xanthomonas oryzae in any of our communities. Finally, we note that for logistical simplicity, we used the 16S rRNA gene to measure absolute bacterial load, but this is not ideal because of copy number variation in this gene across bacteria. Further studies with single-copy genes would provide more accurate estimates of total bacterial load (65, 66).

Supporting our prior finding that Methylobacterium can be beneficial for rice hosts (34), we find the genus Methylobacterium to be one of the most abundant and consistent core bacteria for both phyllosphere community types in all landraces. Methylobacterium is a widespread plant-associated bacterial genus with the ability to utilize methanol released by plants as a carbon source, in turn promoting plant growth via several mechanisms (26, 67). However, the strength of the association between the host plant and Methylobacterium is variable (34, 68). This variability could potentially arise if several community members together influence host plant fitness. For instance, apart from Methylobacterium, we identified several genera such as Sphingomonas, Pantoea, Aureimonas, and Pseudomonas as core bacterial taxa in both the phyllosphere community types—similar to recent studies with wild and cultivated rice (53, 69)—some of which (Sphingomonas and Pantoea) are also known to promote plant growth (27, 70). We observed strong positive and consistent co-occurrence of Methylobacterium with Aureimonas and Deinococcus across years, landraces, and community types (Fig. S10). We speculate that these correlations may indicate cooperative interactions that need experimental validation. Interestingly, we find that nearly all of these genera are also consistent seed endophytes across generations (as seen in prior studies, e.g., 71, 72), suggesting that co-inoculating rice with several co-occurring core taxa may be a fruitful avenue for further research aimed at increasing rice yield or growth.

In summary, our work identifies and quantifies several factors that shape the rice phyllosphere microbiota, finding points of both convergence and divergence in the assembly and maintenance of bacterial communities in two distinct leaf niches. We hope that further studies determine if and how the identified core bacterial taxa impact their host rice landrace and investigate potential agricultural applications.

DATA AVAILABILITY

Raw sequence data are available in the Sequence Read Archive (SRA) of the National Center for Biotechnology Information (NCBI) under Bioproject accession number PRJNA1158963. All data supporting the findings and codes are available on Figshare (http://doi.org/10.6084/m9.figshare.26139961).

SUPPLEMENTAL MATERIAL

The following material is available online at https://doi.org/10.1128/msphere.00765-24.

Supplemental tables and figures. msphere.00765-24-s0001.docx.

Tables S1 and S2; Fig. S1-S16.

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.