Deterministic colonization arises early during the transition of soil bacteria to the phyllosphere and is shaped by plant–microbe interactions

Abstract

Background

Upon seed germination, soil bacteria are activated to transition to the plant and eventually colonize mature tissues like leaves. These bacteria are poised to significantly influence plant health, but we know little about their colonization routes. We studied the mechanisms of the transition of soil bacteria to germinating plants and leaves using an in-planta isolation approach and by experimentally manipulating inoculation times. We then tested how plant-microbe-microbe interactions shape assembly mechanisms in natural soil communities by amending soil with a trackable, labeled strain of the opportunistic pathogen Pseudomonas viridiflava (Pv3D9).

Results

We identified 27 diverse genera of leaf-associated bacteria that could transition alone from a few cells near a germinating plant to mature leaves, suggesting that the soil-to-leaf transition is probably important for them in nature. Indeed, when plants were only inoculated by soil after the emergence of true leaves, less diverse bacteria transitioned to mature leaves via different colonization mechanisms than when plants germinated in the presence of soil microorganisms. In particular, deterministic processes drove the colonization of phylogenetic bins dominated by Pedobacter, Enterobacter, Stenotrophomonas, Janthinobacterium, Pseudomonas, and Chryseobacterium only in the natural soil-to-leaf transition. Host genotype and soil amendments with Pv3D9, both of which affect host physiology, had strong effects on mainly deterministic colonization.

Conclusions

Diverse bacteria transition from soil to leaves during natural colonization, resulting in characteristic diversity in healthy leaf microbiomes. The mechanisms of colonization are a mix of stochastic processes, which will be largely shaped by competition, and deterministic processes which are more responsive to factors that shape host physiology. In the chase toward targeted manipulation of microbiomes, identifying these mechanisms for a given host and environment provides important information. Developing targeted treatments, however, will require further dissection of the pathways by which host factors drive stochastic and deterministic transitions from soil to leaves.

Video Abstract

Background

Plant leaves are colonized by diverse microbes, including bacteria, fungi, and oomycetes. This microbiome plays a critical role in determining how plants deal with stresses like drought, disease, and herbivory [1, 2]. In particular, the bacterial component of the microbiome (the bacteriome) aids plant survival in the presence of detrimental soil microbes [3, 4] and can protect leaves against pathogen attack [5]. Soil is thought to be an important source of inoculum for all plant tissues, but only a small fraction of soil microbes will later make up a significant proportion of the plant microbiome [6]. This filtering is not surprising considering the many barriers to a successful transition of bacteria to germinating seedlings and ultimately mature leaves. Upon germination, a radicle and then cotyledons emerge into the vast microbial diversity contained in the soil [7]. Growth of true leaves follows only later. Thus, for soil bacteria to reach aerial leaf tissue, they must deal with edaphic (soil) factors [8], competitively reach host tissue and establish a niche [9, 10], and either grow together with the plant to establish in or on emerging leaves [11] or invade the tissues later. These early stages are likely very important determinants of the complex transition of bacteria from soil to leaves and ultimately of plant health. However, how they play out and the processes governing them are not well understood.

Stochasticity plays major roles in microbial community assembly in diverse environments [12, 13]. In the context of the plant, stochastic colonization indicates that a niche could be occupied by many different bacteria with similar fitness. In other words, such bacteria are mostly exchangeable. Given the massively diverse soil inoculum, stochasticity could dominate early immigration processes from a complex inoculum like soil to germinating seedlings. By contrast, bacteriome structure in mature leaves is determined at least partly by deterministic factors like host selection [14], and the interaction between host factors and the environment [6]. Indeed, while the abundance of many leaf taxa is correlated to their abundance in soil, suggesting stochastic dynamics, there is extensive variation in this trend that can indicate deterministic drivers [15]. Importantly, recently developed tools can use microbiome structure data to quantify the contribution of stochastic or deterministic processes. This can be done both on the level of whole microbiomes and on the level of individual taxa, to reveal taxa that may occupy specialized niches in hosts [16].

The importance of stochastic and deterministic processes in natural colonization may also be shaped by direct or indirect interactions between bacteria [17]. In plant leaves, early arriving pathogens can alter the immune state of the host and thereby facilitate the colonization of late comers [18,19,20]. Alternatively, early colonizers could occupy essential nutrients like iron [21] or other favorable niches [22]. Experiments simulating the invasion of mature leaves have shown that priority effects can strongly influence bacteriome assembly, with early colonizers persisting and influencing subsequent colonization outcomes [23]. During natural colonization from the soil, most bacterial taxa making the transition to leaves are likely to start from very low cell numbers, given the massive soil bacterial diversity and relatively low abundance in the soil of leaf-colonizing bacteria [15]. Whether or not the arrival of low-abundance soil bacteria can significantly affect colonization processes is unknown. However, this would be important because it may suggest ways to shape colonization mechanisms and community assembly trajectories.

Here, we experimentally dissected the transition of leaf-colonizing bacteria from soil to leaves to better understand how it shapes leaf bacteriomes and ultimately plant health. First, we tested whether and how diverse bacterial taxa can efficiently colonize leaves starting from extremely low levels near a germinating seedling, as a simple soil diversity model suggests they must do in soil. Next, we used three different Arabidopsis thaliana genotypes, including two wild genotypes isolated from Jena, Germany, and compared the assembly mechanisms of bacterial communities in mature leaves when plants germinated in inoculated soil vs. when inoculation occurs post-germination. Finally, we amended natural soil with a trackable opportunistic pathogenic bacterium to better understand how community assembly mechanisms are shaped by bacterial interactions. The results reveal new insights into the types of niches occupied by bacteria that transition from soil to leaves, which can help point the way toward their management.

Materials and methods

Plant materials

We utilized three A. thaliana wild-type genotypes: Col-0, NG2, and PB. Col-0 is a widely available model genotype of A. thaliana, while NG2 and PB are genotypes that were previously recovered from wild populations in Jena, Germany [14]. Near-isogenic lines of each were generated via multiple generations started from single seeds. NG2 is available from NASC (N2110865). PB is being made available there as well and is available upon request from the authors.

In-planta isolation of leaf bacteria experiment

To obtain an overview of the diversity of bacteria that could realistically make a transition from soil to leaves, we developed an in-planta enrichment whereby bacteria from a natural mixed inoculum derived from wild A. thaliana leaves were given the chance to colonize a germinating seedling and successful ones were identified later in mature leaves. A test enrichment experiment was performed to optimize the procedure for recovery and sensitivity and that is described in the (Supplementary Methods and Results).

Production of inoculum from leaves

We sampled the leaves of wild plants in the spring of 2019. We collected 10 A. thaliana rosettes of different sizes from the wild populations NG2 and PB, located in Jena, Germany. They were washed three times with autoclaved ultrapure water to remove dirt, ground with a metal pestle, and suspended in PBS/S (PBS buffer + 0.02% Silwet L-77) by vortexing. After centrifugation (1 min, 19,000 × g), the inoculum was aliquoted, mixed with glycerol (final glycerol concentration: 21.5%), and stored at − 80 °C. For the experiment, the stock was diluted in PBS/S so that the inocula finally contained an average of 3.5 and 1.45 cells/seedling (for NG2 and PB, respectively—details on how the dilution level was obtained are in the Supplementary Methods).

Plant growth, inoculation, and identification of successful leaf colonizers

Seeds were surface sterilized with 2% bleach and 70% ethanol, washed three times with autoclaved ultrapure water, and vernalized for 4 days in 0.1% agarose at 4 °C in the dark. We sowed one seed in each well of a 24-well plate filled with 1 mL ½ MS + 0.2% sucrose + 1% agar. The plants were incubated in growth chambers (PolyKlima PK-520) at 24 °C/18 °C 16 h/8 h day/night cycle with 50% light intensity. Five microliters of leaf microbe extract (see bacteria inoculation density in the previous section) was inoculated as follows: NG2 or PB inocula were added directly onto the cotyledons of 3-day-old NG2, PB, or Col-0 seedlings. We also mock-inoculated control seedlings with PBS/S and did not observe contamination. Inoculated plants were grown for 2 weeks after which whole rosettes were harvested with flamed tweezers. After removing roots and any flower stems, they were transferred into deep 96-well plates and frozen at –80 °C until further processing. Next, we crushed and plated a dilution series of the rosettes to look for bacterial colonies. They were identified by DNA extraction and PCR amplification of the entire 16S rRNA gene region, followed by Sanger sequencing (see detailed supplementary methods).

Inoculation time experiment

Preparation of microboxes

In total, 18 microboxes each containing 9 flow pots [24] were prepared, six for each of the three A. thaliana genotypes (Col-0, NG2, and PB). After they were autoclaved, opening the microboxes for inoculation and sampling was performed in a laminar flow hood. On the first day of the experiment (ID0) the flow pots were flushed with 25 mL sterile MilliQ water to wet the substrate. Seeds had been surface-sterilized with 2% bleach and 70% ethanol and were stratified in the dark for 3 days in 0.1% agarose at 4 °C. Between 7 and 10 seeds were added to each pot, and these were later thinned to 1–3 plants per pot. More detailed box preparation instructions can be found in the supplementary methods.

Inoculations

For each plant genotype, two microboxes were randomly assigned to be inoculated with a natural soil microbial community on the first day of the experiment (ID0), two microboxes on day 7 of the experiment (ID7), one microbox on day 14 of the experiment (ID14), and one microbox was only inoculated heat-killed (double-autoclaved) version of the inoculum (HK). The three inoculation times are all early in development, corresponding to the seed (D0), cotyledon (D7), and first true leaf (D14) stages of plant growth (Fig. 1). Inoculation was performed by injecting a natural soil slurry + Murashige and Skoog (MS) medium mixture into the soil via a tube attachment (5 mL of inoculum) and by spraying from above with an airbrush system (three sprays for half a second each). To ensure that treatment effects were only due to living microorganisms and not, for example, immune priming by molecular patterns, all plants that did not receive live inoculum at each treatment time received the heat-killed inoculum. The boxes were closed and sealed with medical tape (Duchefa Biochemie). Detailed information on preparing the natural soil slurry and seed preparation can be found in the supplementary methods.

Fig. 1

Diagram of flow pot experiment 1, showing for each of three treatments and the control the time of inoculation with live inoculum (start of green arrows), heat-killed inoculum (yellow x), and the sampling times (SD)

Plant growth conditions, sampling, and processing

The prepared boxes with seeds were incubated in growth chambers (PolyKlima PK-520) at 20 °C/14 °C 16 h/8 h day/night cycle with 75% light intensity. Humidity was not controlled, but given the environment inside the closed boxes, was likely close to saturation. Plants were sampled on day 14 (SD14), day 21 (SD21), day 28 (SD28), and day 35 (SD35). At each sampling day except SD35, rosettes were collected with a flamed tweezer and placed in a screw cap tube with sterile beads (two 3-mm steel beads (WMF) and 0.2 g of 0.25–0.5 mm glass beads (Carl Roth)) and stored at − 80 °C until further processing (1 whole plant rosette = 1 sample). At SD35, only three leaves per plant were combined into one sample due to the large size of plants: an old, a middle-aged, and a young leaf. Because we wanted to focus on all leaf colonizers, we processed whole-leaf samples without separation of surface and endophytic compartments. DNA was extracted by bead-beating, followed by incubation for 10 min in a cetrimonium bromide (CTAB) buffer, direct precipitation in ethanol/sodium acetate then ethanol washing. The DNA was further purified using a magnetic bead-based cleanup. DNA extraction details can be found in the supplementary methods.

Pseudomonas viridiflava 3D9 soil amendment experiment

Methods for the experimental setup to initially test how P. viridiflava 3D9 (Pv3D9) affects plant phenotypes and physiology (immune response) in natural soil are found in the supplementary methods.

Pv3D9-141 preparation

Pv3D9 was fluorescently labeled by integrating pMRE-Tn7-141 plasmid containing the mTurquoise2 gene as well as chloramphenicol and gentamicin resistance cassettes [25] to generate Pv3D9-141. It was grown in LB medium supplemented with gentamycin and chloramphenicol at 220 rpm and 28 °C to OD₆₀₀ of roughly 1.5, the cells were then harvested by centrifugation for 5 min at 2000 rcf, the supernatant was removed, and the culture was washed once with 10 mM MgCl₂. After centrifugation, the cultures were resuspended in 10 mM MgCl₂ before diluting into the MgCl₂ wetting solution.

Soil preparation and plant growth

Surface-sterilized A. thaliana NG2 seeds (2% bleach and 70% ethanol) were stratified in the dark for 3 days in 0.1% agarose at 4 °C. On the day of the experiment, a base soil mix was prepared by mixing 4 L Floraton 3 potting soil (Floragard), 2 L perlite (0–6 mm Perlite Perligran, Knob), and 25 g Substral Osmocote garden flower fertilizer (Celaflor). This was wetted with either 2 L of 10 mM MgCl₂ (soil “LS”) or with 2 L of MgCl₂ mixed with 200 g of dried garden soil (soil “LS + GS”). For the Pv3D9-141 treatments, cells were added to the MgCl₂ or MgCl₂ + soil slurry to target 1000 CFU/cm³ or 1,000,000 CFU/cm³ of soil. If the spermosphere (zone of influence) of a germinating seed (0.5 mm × 0.25 mm × 0.25 mm) reaches a distance into the surrounding soil equaling its own dimensions, then this works out to about 0.8 (“low”) and 800 (“moderate”) 3D9-141 CFUs in the spermosphere, respectively. Vernalized seeds were sown with 15 to 20 seeds per pot (5.5 cm upper diameter) and incubated at 193 μmol/(m²∙s) at 22 °C/18 °C and 16 h/8 h day/night conditions. To minimize batch and positional effects, the tray positions were switched randomly every 2 days.

Pv3D9-141 effects on shoot growth

Seven days after germination, seedling shoots were sampled by removing entire plants from the soil and cutting off the root with flamed scissors and tweezers. Because plants were very small, each replicate is the average of eight plants from one pot that were measured together. At the same time, the number of plants in one pot was thinned out to three.

Estimating bacterial and Pv3D9-141 loads in grown plants with CFU counting

To count bacterial loads in the leaves, 14 days after germination, six samples of two to three leaves each were collected and stored at –80 °C until they were processed. For total bacteria and 3D9-141 CFU counting, frozen leaf samples were crushed in 120 µL of 10 mM MgCl₂ using 3-mm diameter sterilized metal beads (WMF) in a mini bead-beater 96 (Biospec Products) at 1400 strokes/min. After homogenization, a dilution series in 10 mM MgCl₂ was drop-spotted onto three different media in duplicates and the plates were incubated at 28 °C and counts were made after 72 h (details on the media and their selection can be found in the supplementary methods). Additionally, a fluorescence stereo microscope equipped with a GFP filter (Zeiss 38 HE GFP—ex 450–490 nm/em 500–550 nm) was used to record the number of mTurquoise2 fluorescent colonies.

qPCR quantification of 3D9-141 leaf loads

For qPCR quantification of 3D9-141 and 16S rRNA gene amplicon sequencing, 14 days after germination, six samples of two to three leaves each were collected and stored at –80 °C until gDNA was extracted. DNA was extracted by bead-beating in a CTAB buffer, followed by a phenol–chloroform cleanup, precipitation in isopropanol, and ethanol washing. Details on DNA extraction can be found in the supplementary methods. For the qPCR, the single-copy mTurquoise2 gene integrated into the 3D9-141 genome was targeted. For the normalization of copy numbers to the host, we targeted the A. thaliana EF1-α gene [26]. Copy numbers were estimated with standard curves generated from a 1:8 dilution series of the linearized pMRE-Tn7-141 plasmid and a 1:8 dilution series of NG2 gDNA. Details on the primers, reaction components, controls, and thermocycler setup are provided in the supplementary methods.

Bacterial community characterization via 16S rRNA gene amplicon sequencing

Library preparation

The extracted gDNA was diluted 1:10 in 10 mM Tris–HCl (pH = 8) and used as a template for 16S rRNA gene amplification. The amplicon sequencing library preparation targeted the V3-V4 region of the 16S rRNA gene and used blocking oligonucleotides to avoid plant plastid amplification, as previously described [27] with a few modifications (see detailed steps in supplementary methods). The final library was spiked with 10% PhiX genomic DNA to ensure high enough sequence diversity and was loaded onto an Illumina MiSeq and sequenced for 2 × 300 cycles. Amplicon sequencing data was split on indices and adapter sequences were trimmed using Cutadapt 1.2.1 [28]. The quality of the reverse reads was not satisfactory, so we proceeded with high-quality forward reads only, which were clustered into amplicon sequencing variants (ASVs) using dada2 [29]. First, reads were trimmed and filtered to remove reads with more than three errors per 100 bp. Sequences were then dereplicated and denoised using the error rate information before calling amplicon sequence variants (ASVs). Chimeric ASVs were removed, and taxonomy was assigned using the latest Silva 16S rRNA gene database (v. 138.1) [30].

Microbial diversity analysis

Downstream analysis was performed in R with phyloseq [31] and other packages listed below. Host-derived reads were removed by filtering any ASVs in the orders “Chloroplast” and “Ricketsialles” from the 16S ASV tables and data was filtered to remove samples with less than 500 reads. For further processing, the reads were either not grouped (ASV level) or grouped at the genus and order levels to provide different robust levels of resolution. Beta diversity was based on the Aitchison distance, and the “RDA” function was used to perform either unconstrained principal components analysis on the Aitchison distance matrix or a redundancy analysis on the matrix constrained by experimental factors, as stated in the text and figures. The statistical significance of factors was tested with an ADONIS test (a permutational analysis of variance) with 999 permutations. Beta dispersion (between-sample beta diversity) was also based on Aitchison distance with statistics as described in the figures. The alpha diversity metrics Chao1 (estimate total richness of community) and Shannon (index accounting for both richness and evenness) were used, and statistical analysis was based on a non-parametric Kruskal–Wallis test. In the inoculation time experimental results, to identify taxa whose abundance in mature leaves significantly varied with experimental variables, we used DESeq2 [32] on data from combined sampling days 21, 28, and 35. The log-transformed abundances of taxa identified by DESeq were plotted in boxplots for visual inspection with a p-value based on a non-parametric Kruskal–Wallis test. In the Pv3D9-141 amendment experiment, we used the RDA axes constrained by treatment and bi-plot data to identify the taxa that were most correlated to the Pv3D9-141 treatments. These taxa were selected for visualization in the biplot and in the boxplots.

For the analysis of processes driving colonization, we employed the tool “Infer Community Assembly Mechanisms by Phylogenetic bin-based null model analysis” (iCAMP 1.5.12) [16]. First, we generated a phylogenetic tree. The 16S rRNA sequences of the ASVs were aligned using the DECIPHER (2.20.0) function AlignSeqs and highly variable regions were masked using MaskAlignment. The masked alignment was then converted to the phangorn package (2.11.1) with the function phyDat and output as a fasta alignment with the function write.phyDat [33]. Finally, the alignment was fed to RAxML next generation (1.2.1) to generate a phylogenetic tree using the GTR + G + I model with default parameters (20 tree searches with 10 random and 10 parsimony-based starting trees) [34]. iCAMP was run by loading the OTU table, taxonomy information, and tree and generating a treatment table that linked the samples to the plant ecotype and inoculation day. For the environment table, the ecotype information was provided. To optimize parameters, first, the phylogenetic niche preference was assessed and then the within-bin phylogenetic signal was measured. We tested a range of values for ds and bin.size.limit and chose those where both the relative abundance of bins with significant phylogenetic signal (RAsig.adj) and the mean correlation coefficient across bins (meanR) showed a peak. For ds = 0.2, and bin.size.limit = 20, in addition, we made sure that the relative contribution of stochastic processes was comparable to the phylogenetic normalized stochasticity ratio, as suggested by the authors. iCAMP was then run with these parameters using “Confidence” as the significance index (a non-parametric, one-sided confidence score) and the beta mean pairwise distance (bMPD) as the metric for phylogenetic null model analysis. One thousand randomizations were used for confidence checking. Finally, bin-level statistics were produced using bootstrapping and were output to a text file.

Results

The capacity for leaf bacteria to transition from soil to leaves is widespread

Although soil is thought to be an important inoculum for all plant tissues, we questioned whether many leaf bacteria are likely to have the capacity to make the complex transition from soil to leaves. To answer this, we first estimated cell number distributions for strains in theoretical homogenous soils using a simple diversity model (see details in Supplementary Results). This analysis suggested that most potential leaf-colonizing bacterial strains will start the colonization process from just a few cells in the spermosphere (the area under the influence of a germinating seedling). Thus, we generated inocula from leaves from wild A. thaliana NG2 and PB populations, diluted them to about 3.5 and 1.45 cells/drop inoculation, respectively, and then inoculated each onto 120 axenic 3-day old NG2, PB, and Col-0 seedlings (360 plants per inoculum). Based on a test experiment, these dilutions were determined to result in sufficient colonized plants, most with only one bacterium per plant (Supplementary Results). Two weeks later, mature leaf material was collected, crushed, and plated on R2A media (Fig. 2A).

Fig. 2

Many leaf bacteria can reach and colonize leaves when only a few cells encounter a germinating seed. A The in-planta isolation method based on dilution-to-extinction of leaf-associated microbes from wild A. thaliana leaves and inoculation onto germinating axenic seedlings, followed by 2 weeks of growth and isolation and identification of colonizers. This was used to identify leaf bacteria that are likely to be efficient colonizers when highly dilute in soil. B Bacteria collected from wild NG2 or PB plants were diluted to 3.5 and 1.45 cells/seedling and inoculated onto germinating NG2, PB, or Col-0 plants. Bar chart showing taxonomy of 30 random colonizers from each treatment (31 for PB PB). Taxonomy is grouped at the order level, taxonomy at the genus level and associated statistics can be found in Figure S1 and Table S2. Detailed data on numbers of plants inoculated and colonized are found in Table S1

25.0% (NG2 ≥ PB) to 55.8% (PB ≥ NG2) (Table S1) of inoculated plants were colonized, suggesting that the inoculum level should result in mostly plants colonized by single bacteria. From 160 positive plants, 181 bacteria were isolated and identified. These were diverse, including 27 genera from three gram-positive (Bacillales, Corynebacterales, and Micrococcales) and seven gram-negative families (Burkholderiales, Caulobacterales, Hyphomicrobiales, Sphingomonadales, Moraxellales, Pseudomonadales, and Xanthomonadales) (Fig. 2B). One hundred forty plants (87.5%) had only one discernable colony. Pseudomonas, Janthinobacterium, and Sphingomonas isolates were significantly more often enriched from NG2 leaf extracts, while Clavibacter was more abundant from PB leaf extracts (Figure S1, Table S2). For bacteria detected often enough to statistically evaluate, there was little evidence that they preferred a particular genotype, with only Curtobacterium from PB slightly more frequently detected in PB plants than in other genotypes (χ²: p = 0.082, Table S2). In 20 plants (12.5%), more than one bacterium was co-isolated. For Clavibacter, Methylobacterium, and members of the Burkholderiales (Janthinobacterium, Variovorax, Acidovorax, and Rugamonas), this was significantly more often than expected (χ²: p < 0.05), suggesting they likely benefit from a partner (Table S3). Some of the Burkholderiales were only observed once, but as a group, the seven total Burkholderiales isolates (all Comamonadaceae and Oxalobacteriaceae) always were together with a co-colonizer, although this would be expected only once by chance alone (χ²: p = 2.56 × 10⁻¹²). Together, these findings confirm that diverse leaf-associated bacteria can efficiently transition to leaves from minute levels near a germinating seedling, suggesting that the soil-to-leaf pathway may be important for many taxa but some taxa may depend on the effects of others to make the transition.

Early colonization of germinating seedlings shapes mature leaf bacteriomes

To study the importance of the soil-to-leaf transition of bacteria in a natural colonization of Arabidopsis thaliana, we controlled when plants of three genotypes were inoculated by a natural soil microbiota (day 0, D7, D14, or heat-killed). Whole leaf bacterial community structures were more strongly shaped by inoculation day than either sampling day (SD14, SD21, SD28, or SD35) or plant ecotype (Col-0, NG2, or PB) at all investigated taxonomic levels (ASV, genus, and order). In particular, the inoculation day effect was strongest in the Col-0 genotype (explaining up to 18% of variation at the order level) but was also significant in NG2 and PB (explaining about 10% of the variation in each) (Fig. 3A; Figures S2A and S3A; Table 1; Tables S4 and S5). The “bacteriome” of plants treated with heat-killed control inocula were much less beta- and alpha-diverse than plants treated with live inocula (Fig. 3A; Figures S2A, S3A, and S4) and generally clustered closer to plants inoculated at D14 than to other plants (Figure S4). In all genotypes, mature leaf bacteriomes (21 days post inoculation) that had been live-inoculated at D0 were more similar to those inoculated at D7 than to those inoculated at D14, especially at higher taxonomic levels (Fig. 3B; Figures S2B and S3B and between-sample beta diversity in Fig. S4). This is because many taxa were significantly enriched in early (D0 or D7)-inoculated plants compared to D14-inoculated plants (Figures S5–S7), leading to general patterns of higher alpha diversity and usually higher evenness in D0- and D7-inoculated plants (Fig. 3C; Figures S2C and S3C).

Fig. 3

Early-inoculated plants (before the emergence of true leaves) develop distinct and diverse bacteriomes in mature leaves. Data is based on genus-level taxonomy of bacteria associated with the mature leaves of A. thaliana after inoculation at day 0, 7, or 14 or with heat-killed inoculum (D0, D7, D14, HK). A RDA ordination of beta diversity (Aitchison distance) of all samples including multiple sampling times. The results of a PERMANOVA test for the significance of factors are in Table 1. B RDA ordination of beta diversity (Aitchison distance), including only live-inoculated samples collected 21 days post-inoculation. The results of a PERMANOVA test for the significance of factors are in Table 2. C Alpha diversity (Chao1 estimates of total diversity and Shannon Index) of samples collected 21 days post

Table 1 Significance of factors in describing the variation in bacterial communities between samples. Values highlighted in bold have a significant p-value

The effect of plant genotypes also depended on inoculation time. Specifically, plants inoculated at D0 or D7 had more taxa that were differentially enriched between the genotypes (43 and 57 ASVs, respectively) than plants inoculated at D14 (28 ASVs, Figure S8). Additionally, 15 ASVs were shared between D0- and D7-inoculated plants, while only 3 ASVs were shared between D0- and D14-inoculated plants. However, plant genotype explained more beta diversity in D14-inoculated plants (p < 0.01, with 26–28% of variation depending on taxonomic level) than in plants inoculated at D0 or D7 (11–17%, insignificant at some taxonomic levels) (Table 2; Tables S6 and S7) because some differentially enriched taxa in D14-inoculated plants were highly abundant, especially in the Col-0 genotype (FiguresS9–S10). Overall, these results show that leaf bacteriome assembly is shaped uniquely by the early transition of bacteria from soil to plant leaves, resulting in higher diversity and more taxa that are influenced by plant genotypes.

Table 2 Significance of factors in describing the variation in bacterial communities between samples. Values highlighted in bold have a significant p-value

Results are based on the Aitchison distance between samples collected at all time points at the genus level and includes plants treated with only heat-killed inoculum. Analysis was carried out with PERMANOVA with 1000 permutations.

Results are based on the Aitchison distance between samples collected at 21 DPI at the genus level and do not include the heat-killed inoculum. Analysis was carried out with PERMANOVA with 1000 permutations.

In the overall stochastic transition of bacteria to leaves, some taxa transition deterministically

To understand why plants that germinated in inoculated soil (inoculation D0) were different than those inoculated later (inoculation D14), we studied the mechanisms of assembly of leaf bacteria (stochastic vs. deterministic processes). To do so, we employed a phylogenetic bin-based null model analysis [16], which determines to what extent different colonization processes shape phylogenetically grouped bins of ASVs. Overall, stochasticity was driven by the colonization process itself, since inoculation of plants with heat-killed inoculum resulted in much less stochastic “bacteriomes” than in live-inoculated plants (Fig. 4A). In live-inoculated plants, stochastic processes dominated in all genotypes and inoculation times (> 60% contribution to assembly, Fig. 4A), with colonization slightly more stochastic in D0- than in D14-inoculated plants. Despite this dominance, deterministic processes were still important and were dominated by homogenous selection (HoS), making up ~ 15–35% relative importance depending on plant genotype (highest in Col-0, lowest in PB) and inoculation time (Figures S11A and S11B). These results were validated based on agreement with a second, independently calculated stochasticity metric, the phylogeny-based normalized stochasticity ratio [35] (Figure S11D).

Fig. 4

Processes shaping bacterial colonization of leaves of the A. thalianagenotypes Col-0, NG2, and PB depend on both inoculation time and genotype. A The relative importance of stochastic processes in the colonization of the leaves of plants live-inoculated at ID0, ID7, ID14, or heat-killed inoculated. Stochasticity represents the sum of dispersal limitation, homogenous dispersal, and drift, but was consistently dominated by dispersal limitation. Deterministic processes (homologous and heterologous selection), as well as an independent measure of stochasticity (pNST) are shown in Figure S10. B. Assembly processes at the level of taxonomic bins, shown for Col-0, and trees for other genotypes are shown in Figure S11. The tree is annotated with, from inside to outside: colored dots representing the majority taxonomy of each bin, A barchart showing for plants inoculated at ID0 the relative contribution of each community assembly process for each bin (DL: dispersal limitation, DR: drift, HD: homogenizing dispersal, HeS: heterogeneous selection, HoS: homogeneous selection), and A heatmap showing the change in the relative importance of each assembly process for each bin between plants inoculated at ID0 and those inoculated at ID14 (yellow means a given process is more important at ID0, dark blue is more important at ID 14). The labeled taxa are those mentioned in the text that were selected deterministically at ID0 and stochastically at ID14 (homologous selection increased by at least 20% to reach > 50% total contribution in D0). The plant genotypes where this difference was observed are labeled

At the level of phylogenetic bins, colonization processes were usually dominated by dispersal limitation (DL), a stochastic process, and this was consistent across genotypes and inoculation times (Fig. 4B). Some taxa, however, were more strongly influenced by deterministic processes. Pseudomonadales were especially strongly and consistently shaped by deterministic HoS, in particular bin Pseudomonadales 41 (including only Pseudomonas ASVs) (Fig. 4B; Figure S12 and detailed bin data in Supplementaryfile 1). This determinism can help explain why many outlier ASVs that were unusually abundant for their frequency (fraction of samples where they were observed) were Pseudomonas (Figure S13). Several other bins also colonized deterministically, particularly in D0-inoculated plants (> 50% contribution from HoS—barchart in Fig. 4B). These were Sphingobacterales 19 and Enterobacterales 38 (in Col-0 and NG2, mostly Pedobacter and Enterobacter ASVs, respectively), Xanthomonadales 35 and Burkholderiales 17 (in Col-0, mostly Stenotrophomonas and Janthinobacterium, respectively), Pseudomonadales 40 and 41 (in Col-0 and PB, mostly Pseudomonas), and Flavobacteriales 20 (in PB, mostly Chryseobacterium). Notably, all of these (except Pseudomonadales 41 and 49 in NG2 and Col-0) were only deterministic in D0-inoculated plants but colonized stochastically in D14-inoculated plants (the difference between D0 and D14 was > 25%) (heatmap in Fig. 4B, Figure S12 and Supplementary file 1). For several of these same bins, HoS was more important at D0 than D14 in other genotypes as well but did not reach > 50% at D0 (Burkholderiales 17 and Flavobacteriales 20 in NG2 and Enterobacterales 38 in PB). Together, these results show that after seedling germination in soil, the transition from soil to mature leaves is a stochastic process for many colonizers. Most taxa that do make this transition deterministically, however, do so only during this complex transition and not when they are added later, helping to explain why plants colonized naturally from soil assemble distinct leaf bacteriomes.

Plant–microbe interactions shape the colonization of deterministic taxa

Determinic colonization indicates that a particular bacterium is not easily replaced, a sign that it could occupy a specialized niche. Thus, we reasoned that treatments that strongly affect plant phenotypes and physiology would most strongly affect deterministic colonizers. To test this, we selected Pseudomonas viridiflava 3D9, a strain isolated in the in-planta isolation experiment (Fig. 1B, treatment NG->NG). Pv3D9 is an opportunistic pathogen native to the NG2 genotype—just a few cells inoculated onto axenic NG2 seedlings are enough to nearly always kill the host plant (Figure S14). It was isolated from healthy plant tissue, however, and when we amend it to natural soil, we consistently observe that its presence actually rescues plant growth (Figure S15). This could be because opportunistic P. viridiflava strains interact with plant immunity [36], which in turn could protect against detrimental soil microbes. Consistent with this, we observe a slightly stronger immune response (ROS burst in response to the flg22 peptide) in some Pv3D9-treated plants (Figure S15). Thus, in a natural soil, Pv3D9 affects plant physiology and phenotypes, and so, we could use it to test the hypothesis that this should poise it to mainly affect deterministic colonizers.

As previously observed (Figure S15), plants grown in laboratory soil mixed with a natural garden soil slurry (LS + GS) showed about 35% decreased growth compared to laboratory soil alone (LS) (Fig. 5B). Pv3D9-141 (a labeled version of Pv3D9 detectable via qPCR and selectable markers [25]) reversed this, increasing the leaf fresh weight in a low amendment (~ 0.8 cells in the initial spermosphere) back to LS levels. In a moderate amendment (~ 800 cells in the initial spermosphere), shoot fresh weight was less than the low treatment, but still higher than mock levels (Fig. 5B). 3D9-141 reached mature leaves in the moderate treatment as measured by both qPCR (Fig. 5A) and resistant, fluorescent CFUs (Figure S16), but not in the low treatment. Based on total CFU counts, 3D9-141 made up roughly 1% of the leaf bacteriome and its median level in leaves was not significantly affected by the addition of GS. Thus, the opportunistic pathogen 3D9-141 can transition to plants and leaves depending on its inoculation level, and in natural soil, it always had a significant, positive effect on plant phenotypes.

Fig. 5

Bacteria that deterministically transition from soil to mature leaves are influenced by other colonizers. Throughout, green corresponds to the mock treatment (no 3D9-141), light blue to the low 3D9-141 soil amendment, and dark blue to the moderate 3D9-141 soil amendment. APseudomonas viridiflava 3D9-141 abundance in mature leaf tissue based on qPCR measurement of copies of a vector-inserted gene for the mTurquoise2 protein normalized to the host plant elongation factor 1. B The fresh weight of A. thaliana shoots 7-days post-germination. C, D Alpha diversity of bacterial genera in mature leaf tissues (14 days post-germination) assessed by Chao1 metric (total estimated diversity) and Shannon index (diversity considering evenness of distribution of taxa). In A–D, a Kruskal–Wallis test was performed on the data overall, and when it was < 0.05, a Wilcoxon test was used to evaluate differences between groups. Signficant p-values are shown. E Principal components analysis of Aitchison distance between samples, constrained for treatment, showing biplot arrows for the top 3% of bacteria that correlate to the two axes. Circles correspond to LS and triangles to LS + GS, as in the boxplots. F Boxplots showing the abundance of the bacteria identified using the biplot analysis in E with p-value result of a Kruskal–Wallis test for significant differences between groups

Next, we tested how 3D9-141 influenced the colonization of other bacteria in whole-leaf samples. Based on CFU counts, there was no clear effect of the soil treatment or 3D9-141 on total leaf bacterial loads (Figure S16). The addition of garden soil did not significantly increase the bacterial diversity in mature leaves, but it did increase the evenness of the community (Fig. 5C, D). 3D9-141 appeared to slightly affect the medians of both alpha diversity and evenness, but it was not significant with conservative non-parametric statistics. It did, however, have a significant effect on bacterial community structure, with the effect of the soil type and 3D9-141 treatments each independently explaining about 6–10% of variation depending on the taxonomic level (Table 3). Generally, low and moderate 3D9-141 treatments clustered separately, driven by distinct changes to several taxa. Pseudomonas itself was only strongly correlated to treatments at the ASV level, while other taxa were clearly affected by treatments as well (Fig. 5E; Figures S17 and S18). Two of the four genera most strongly affected by 3D9-141 were affected similarly in both soils: Bacillus (Bacillales), whose abundance always increased with 3D9-141 treatment and Pedobacter (Sphingobacterales), which tended to be suppressed by 3D9-141. On the other hand, Chryseobacterium (Flavobacterales) was strongly suppressed by 3D9-141 at both low and moderate levels mainly in LS, while Janthinobacterium (Burkholderiales) was much higher abundance in LS only in the moderate treatment, when 3D9-141 reached leaves, but was hardly detected at all in LS + GS (Fig. 5F). Similar patterns at the ASV and order level underscore that effects were highly taxa-specific and depended on both soil and 3D9-141 inoculation level (Figures S17 and S18). Notably, three of four genera affected by 3D9-141 (Chryseobacterium, Janthinobacterium, and Pedobacter) were also strongly shaped by homogeneous selection in the natural colonization of NG2 plants (Fig. 4). Thus, taxa that behave deterministically in natural colonization appear to also be those that are most responsive to the effects of soil amendment with 3D9-141.

Table 3 Significance of factors in describing the variation in bacterial communities between samples. Values highlighted in bold have a significant p-value

Discussion

The transition of bacteria from soil to phyllosphere is an important colonization route

Bacteriomes that develop in mature leaves after seedling germination have important implications for host plant health [37]. The goal of this work was to gain better insight into how the well-known effects of host and environmental factors [6] shape natural colonization processes in these important communities. Leaf bacteriomes are composed of cells that must arrive in one of two ways: transitioning following seed germination, either from soil or from cells attached to or in seeds that were transmitted vertically from parent plants, or later arrival and “invasion” of already emerged leaves. Transitioning from soil is thought to be a very important route since leaf bacteria are typically mostly composed of a subset of the taxa found in soils [6]. The importance of the “soil route” is also recognized in commercial applications, where seed impregnation with biological treatments is meant to improve the colonization success of beneficial microorganisms [38]. However, this route is complex: bacteria must find and reach germinating seedlings and ultimately multiply on or in the growing plant to eventually establish in true leaves that only emerge later. Additionally, our simple model of soil bacterial diversity illustrated the complication that there will usually just be a few chances for success since most strains will start from just a handful of cells in the vicinity of the seedling. To our knowledge, there is little known about which bacteria efficiently manage this complex task in a natural context, how they do so, and what influences them. However, we here experimentally showed that many bacteria can and do manage this transition and that it leaves distinct signatures in the leaf bacteriome.

Natural colonization results in uniquely diverse leaf bacteriomes

In the transition to leaves from a diverse natural soil, we found that most bacterial colonizers exhibited stochastic assembly patterns. In neutral theory, stochasticity indicates that the differences in fitness between taxa are not significant, making their distribution largely subject to neutral processes including dispersal limitation, homogenizing dispersal, and drift [16]. Thus, while many bacteria can reach and colonize leaves, most apparently do so with similar fitness. This can enable simultaneous transitioning from the soil of many taxa that are essentially exchangeable and can help explain the unique and high diversity of taxa in naturally colonized plant leaves (inoculation day 0 in this study). It should be noted that stochasticity in this context refers only to niche fitness differences among leaf-colonizing bacteria and does not imply that leaf colonization is just a random selection of environmental bacteria. On the contrary, leaf bacterial diversity is usually far less than soil bacterial diversity, indicating that “phyllosphere-associated bacteria” are different than other soil bacteria that do not make the transition [39]. Supporting this, in our 1-on-1 plant-bacteria enrichment experiment, a wide diversity of typical leaf bacteria could start from just a few cells near a germinating seedling and weeks later reach detectable levels in mature leaves. Thus, these taxa are apparently well-adapted to thrive on primary resources from host plants and to deal with the harsh conditions in the leaf environment, even if they do so stochastically. Host physiology including plant immunity, as well as primary resources that stochastic taxa depend on including plant exudates, will be variable across host species and environments [40]. Thus, an approach like what we have used here can help identify these efficient colonizers. This would be important for applications, to know which taxa will thrive when given an advantage, for example, via seed impregnation or soil amendments.

At least two factors could explain the stochastic recruitment of many or most leaf bacteria. For one, many leaf colonizers could use similar resources and have similar fitness so that none drastically outcompete others. Resource overlap, however, was previously found to be a relatively poor predictor of colonizer success on leaves [41]. On the other hand, resource overlap is usually calculated from resource preferences from metabolic models or in vitro experiments because of the difficulty of measuring the activity of bacteria in-planta. Indeed, we still know very little about niches during host colonization [42] and it is possible that there is extensive overlap among commensal bacteria. A second possibility is that space is limited and occupied on a “first-come-first-served” basis. This is supported by various studies showing that bacteria colonizing healthy leaves are limited spatially, for example, to epidermal cracks [43] or microscopic droplets [44] and that colonizers that establish there tend to be stable to subsequent invasion [23]. Regardless, the process of transitioning from soil to mature leaves is complex and may involve bacteria transitioning between multiple niches before reaching leaves. Thus, to understand and manipulate it, improved techniques to understand niches and niche diversity of non-pathogenic bacteria in-planta will be required.

The emergence of determinism in the midst of stochasticity

Our results suggest that most leaf-colonizing bacteria transitioned from soil to leaves stochastically. However, a few taxa showed strong signs of determinism (specifically homogeneous selection). It is not surprising that several Pseudomonas bins consistently colonized deterministically, given the well-described adaptation of Pseudomonas to A. thaliana colonization [45]. On the other hand, bins dominated by Pedobacter, Enterobacter, Stenotrophomonas, Janthinobacterium, and Chryseobacterium also colonized deterministically. In contrast to stochastic selection, which occurs when multiple taxa can occupy a niche with similar fitness (i.e., are exchangeable), homogenous selection implies that the taxa likely have more specialized niches [16], which likely also are shaped by plant exudates. Supporting this, we previously found that several taxa, including Burkholderiales taxa in the families Comamonadaceae and Oxalobacteraceae (like Janthinobacterium), are strongly enriched in some A. thaliana genotypes due to aliphatic glucosinolates, specialized metabolites of A. thaliana. We were able to identify a specialized metabolic niche whereby only certain bacteria with myrosinase enzymes (including an Enterobacterales) can metabolize glucosinolates as a carbon source, and specific cross-feeding interactions support the enrichment of other taxa (Pseudomonas and perhaps Janthinobacterium) [14]. A similar phenomenon was recently observed in maize, where specialized lactonases enable specific bacteria to grow on benzoxazinoids, leading to rhizosphere enrichment [46]. Given the wide diversity of specialized metabolites that plants produce [47], we speculate that these can be the basis of many specialized niches that underly deterministic recruitment in plants. However, new approaches are needed to understand what is available to colonizers and how bacteria take advantage of them. The approach we use here to identify deterministic taxa may be a good way to identify those whose colonization niche should be studied more closely.

Deterministic processes in natural colonization rely on the soil-to-leaf transition

Notably, here we found that except for Pseudomonas, all the taxa that were deterministic in the transition from soil to leaves colonized stochastically when soil inoculum was applied directly onto leaves (inoculation day 14). It is currently unclear why deterministic niche establishment in leaves would be dependent on at least some part of the complex transition from soil. However, several explanations are plausible. First, germinating seedlings exude a broad mix of compounds into the soil matrix that shifts in composition over time [48]. Exudates can serve as signals that cause soil bacteria to alter their phenotypes to promote host colonization [49]. Therefore, exudates secreted early on could serve as a signal to activate the metabolic pathways required to establish a specialized niche. Second, plants use molecular recognition machinery to react specifically to different types of microbes [50], where responses include altered secondary metabolism. Thus, the makeup of the soil microbiome at early colonization timepoints, by altering plant physiology and chemistry, can lead to altered patterns of deterministic colonization. This is consistent with our finding that even very low-level natural soil amendment with Pv3D9, an opportunistic pathogen, promoted plant growth, likely indicating that it altered the plant’s immune status. These changes to plant physiology had either direct effects or effects that rippled into the leaf bacteriome, but these were almost strictly limited to deterministically colonizing taxa, suggesting that processes they depend on had been altered.

A third possibility could be that important interactions between deterministic taxa developed during early colonization. Our results strongly suggest that deterministic colonization of leaves by some taxa arises out of inter-bacterial interactions. In the in-planta isolation experiments, Oxalobacteraceae and Comamonadaceae bacteria were always only found with a partner bacterium. Consistently, our previous work showed that Oxalobacteraceae and Comamonadaceae colonization in A. thaliana leaves was linked to aliphatic glucosinolates, but in vitro tested isolates probably relied on another bacterium that could metabolize the compound [14]. Similarly, in an enrichment of bacteria from peppers on the secondary metabolite capsaicin, two partners (a Pseudomonas and the Comamonadaceae Variovorax) fully relied on one another to use it as the sole carbon and nitrogen source [51]. More broadly, Comamonadaceae and Oxalobacteraceae have been highlighted as being strongly positively correlated to a range of bacteria in wild A. thaliana, including links to both Sphingobacteriaceae and Flavobacteriacae (which include Pedobacter and Chryseobacterium, taxa found here to colonize deterministically with Janthinobacterium) [52, 53]. Thus, we hypothesize that these bacteria enter into diverse interactions with other bacteria that define their niche and promote deterministic colonization. Regardless, once taxa are identified that colonize deterministically in a given environment, an important next step will be elucidating the molecular underpinning of the colonization mechanisms to realize targeted manipulations.

Conclusions

Understanding how plant-associated bacteria assemble in tissues like leaves is a critical step toward harnessing their potential for sustainable plant protection. While soil clearly is an important inoculum of bacteria for leaf tissues, the complex transition of soil bacteria to leaves following seedling germination has largely remained a black box. This work demonstrated that the early stages following germination are critical for the establishment of diverse bacteriomes in mature leaves. This is because of the unique mix of stochastic and deterministic mechanisms that arises during early colonization to shape individual bacterial taxa. Deterministic mechanisms especially are shaped by both host genotype and host-microbe interactions. A more detailed understanding of how these factors play out will still require characterizing the pathway of transition in or on plants. Thus, characterizing the mechanisms shaping bacterial taxa offers a way forward for targeted manipulation and optimization of leaf-associated bacteriomes.