Bacillus drives functional states in synthetic plant root bacterial communities

Abstract

Background

Plant roots release root exudates to attract microbes that form root communities, which in turn promote plant health and growth. Root community assembly arises from millions of interactions between microbes and the plant, leading to robust and stable microbial networks. To manage the complexity of natural root microbiomes for research purposes, scientists have developed reductionist approaches using synthetic microbial inocula (SynComs). Recently, an increasing number of studies employed SynComs to investigate root microbiome assembly and dynamics under various conditions or with specific plant mutants. These studies have identified bacterial traits linked to root competence, but if and how these traits shape root microbiome dynamics across conditions is not well understood.

Results

To explore whether bacterial trait selection follows recurrent patterns, we conducted a meta-analysis of nine SynCom studies involving plant roots. Surprisingly, we observed that root communities frequently assemble into two distinct functional states. Further analysis revealed that these states are characterized by differences in the abundance of Bacilli. We propose that these Bacilli-associated functional states are driven by microbial interactions such as quorum sensing and biofilm formation and that host activities, including root exudation and immune responses, influence these functional states.

Conclusions

Whether natural root communities also organize into distinct functional states remains unclear, but the implications could be significant. Functional diversification within root communities may influence the effectiveness of plant-beneficial bioinoculants, particularly Bacilli-based inoculants. To optimize microbiome-driven plant benefits, a deeper understanding of the mechanisms underlying functional state differentiation in root microbiomes is needed.

Supplementary Information

The online version contains supplementary material available at 10.1186/s13059-025-03739-8.

Background

Assembly of the root microbial community, or microbiome, results from numerous interactions among soil micro-organisms and the plant root. Plants exude a complex array of organic compounds that attract soil microbes, which utilize these compounds for their growth. Consequently, these microbes travel to the rhizosphere and colonize the root surface [1]. The established root communities can, in turn, positively affect the plant by enhancing its health and growth [2–4]. Inoculations of plant-beneficial bacterial strains on plant roots consistently demonstrate positive effects on plant health and growth when cultivated in laboratory settings [5–10]. However, several studies reported that laboratory-based plant-beneficial effects disappear in agricultural field trials, where plants grow in complex natural microbiome environments [6, 11–14]. This highlights a significant challenge in developing plant-beneficial bioinoculants or leveraging the root microbiome for plant-beneficial purposes: our limited understanding of root microbiome dynamics and functioning [15]. For example, a bioinoculant might perform variably in different complex community contexts [16–19]. To overcome this problem, we need to better understand root microbiome assembly, its dynamics, and the root colonization strategies of microbes within. A comprehensive catalog of the microbial functions, and their interactions, present in the root microbiome can potentially reveal what and how plant–microbe interactions and/or microbe-microbe interactions contribute to root community assembly.

Currently, little is known about the microbial functions that exist in complex soil microbial communities. Sequencing microbial marker genes provides insight in the microbiome profile but cannot reveal the functional information of microbial genes and functions. Metagenome prediction tools such as PICRUSt2 and Tax4Fun have been developed to address this issue, though they have been found inaccurate for soil and root microbiome samples due to database limitations [20–22]. Sequencing the full metagenome of these samples can provide functional information, but even with high coverage, it is not possible to discover all microbial functions in the root microbiome [23]. Additionally, metagenome sequencing is expensive and is affected by host-derived contamination of the metagenomic reads [24], leading researchers to seek alternative methods. In the last 10 years, researchers have turned to synthetic community (SynCom) reconstitution experiments, which involve inoculating plants with small to large mixes of laboratory-cultured bacteria [25]. This method requires isolating microbial strains from the roots, cultivating pure cultures from these strains, and sequencing the strains to gain insight in their microbial functions [26–28]. Once these efforts have been completed, researchers have full control over the SynCom composition, allowing them to include or exclude bacterial strains based on taxonomy, functionality, or other criteria [25]. Additionally, SynCom reconstitution experiments enable causal observations, as microbiome compositional changes or plant phenotypic changes can be explained by underlying microbial genetics using genome-wide association studies, and experimentally validated using mutant strains or drop-out experiments [25].

In contrast to natural microbiome studies, SynCom-based root communities are assembled from a predefined mix of cultured bacteria, each introduced in equal proportions at the start. This controlled setup enables researchers to study root community assembly in a stepwise manner, uncovering fundamental patterns, similarities, and differences in community formation. Given that microbiome assembly is governed by deterministic principles [29, 30], we hypothesize that root communities exhibit functional similarity. If so, this could suggest that specific bacterial functions associated with root competence are consistently selected across different communities.

To test this hypothesis, we conducted a meta-analysis of nine studies that inoculated SynComs of varying complexity on different host plants, or that cultured SynComs with diverse plant-derived compounds. For each study, we quantified the abundances of SynCom members and analyzed their genomic sequences to identify bacterial functions associated with root colonization. Additionally, these studies examined root microbiomes across multiple plant genotypes and environmental conditions, providing insights into how plant traits and environmental factors influence community composition and bacterial functional dynamics. We show that root communities exhibit functional overlap despite being composed of different, taxonomically diverse bacterial strains. Remarkably, our analysis reveals that in each study root microbiomes consistently segregate into two distinct functional states, defined by the presence or absence of specific bacterial functions. We identify Bacillus strains as key drivers of this functional state separation and show that the plant’s immune system plays an important role in shifting the root microbiome toward a Bacillus-associated state or a non-Bacillus-associated functional state.

Results

Functional overlap among rhizobacteria and root microbiomes

To explore similarities and differences in root community assembly based on microbial functions, we investigated the bacterial functional composition in nine publicly available SynCom datasets [31–39]. These studies applied complex SynComs (> 20 strains) to plant roots or root-derived compounds and characterized the resulting communities using amplicon-based sequencing. The SynComs originated from different soils, were investigated under varied environmental conditions or plant genotypes, and exhibited taxonomic diversity at multiple levels (Additional file 1: Table S1). The most pronounced taxonomic differences in the input inocula were observed between US-derived [27] and European-derived SynComs [26] (Finkel_2019, Finkel_2020, Lebeis_2015, and Qi_2021 versus Duran_2021, Hou_2021, Schandry_2021, Wippel_2021, and Wolinska_2021) (Additional file 2: Fig. S1). Despite substantial taxonomic variation, functional overlap in these input inocula was evident—particularly at higher taxonomic levels such as family level—as revealed by KEGG Ortholog (KO) profiles. Across all datasets, 55.5% of detected KOs (4267 out of 7688 KOs) were shared (Additional file 2: Fig. S2), highlighting a core functional repertoire among root-associated bacteria. This functional similarity already present in the input inocula suggests a degree of functional similarity in root microbiomes as well, regardless of geographic or taxonomic differences.

To assess whether functional overlap is also evident in SynCom-inoculated plant roots, we quantified SynCom members and extrapolated isolate abundances to KO abundances using the generated KO profiles. When excluding Lebeis_2015, we observed slightly higher functional similarity compared to genus similarity in 8 out of 9 studies (Fig. 1), as indicated by a marginally lower PERMANOVA R² value, suggesting reduced dissimilarity in compositions (R²: 0.24 for KO versus 0.26 for genus). Including Lebeis_2015 in the dataset, however, resulted in a higher similarity in genus composition than in functional composition (R²: 0.35 for KO versus 0.34 for genus). This discrepancy is driven by a relatively large fraction of unique KO functions (283) present in the SynCom of Lebeis_2015, distinguishing its functional composition from the other nine studies (Additional file 2: Fig. S2). In Lebeis et al. [34], SynCom members were selected based on their root competence and varied responses to the plant’s immune system, particularly focusing on salicylic acid (SA) signaling pathways. This selection approach differs from other SynCom studies that primarily choose bacteria based on their prevalence in the root microbiome. Despite these differences, PERMANOVA results indicate that root microbiome assembly varies between SynComs composed of different genera or originating from different soils when assessing functional compositional similarity.

Fig. 1.

Principal coordinate analysis (PCoA) on Bray–Curtis distances depicting the compositional differences among samples from the nine SynCom studies included in this meta-analysis. A PCoA based on Bray–Curtis distances of genus-level taxonomic profiles. B PCoA based on KO (KEGG Orthology) abundances. Each point represents an individual sample, with proximity indicating greater similarity in microbial composition

Root microbial communities assemble into two functional states

The limited overlap observed in the Bray–Curtis analysis of the functional composition across studies (Fig. 1) may stem from the metric’s sensitivity to conserved KOs present in multiple copies, such as ribosomal genes [40, 41]. To address this, we employed Jaccard distances, which evaluated compositional similarity based on the presence or absence of genera or KOs, rather than their abundance (Fig. 2). This approach revealed greater functional similarity compared to genus composition (R²: 0.37 for KO versus 0.43 for genus). The similarity became even more pronounced upon excluding the Lebeis_2015 dataset (R²: 0.24 for KO versus 0.35 for genus). These findings suggest that the bacterial genetic features associated with root competence are conserved across communities composed of different genera or derived from distinct soils.

Fig. 2.

Principal coordinate analysis (PCoA) plot based on Jaccard and unweighted UniFrac distances. Jaccard distances between samples show compositional similarity between studies on the genus (A) and functional (KO) level (B). Unweighted UniFrac distances between samples indicate differential similarity between studies compared to the Jaccard distances (C). Compositional differences between the functional states that are evident in B are shown for the KO Jaccard distances, genus Jaccard distances, and unweighted UniFrac distances (D). Significant differences between PERMANOVA R² values in D are assessed by a paired Wilcoxon test and average values are indicated by dotted lines

Remarkably, when examining Jaccard distances between functional root microbiomes within studies, we found that 7 out of 9 studies (all except Hou_2021 and Wolinska_2021) exhibit a division of samples into two distinct clusters or groups (Fig. 2B, Additional file 2: Fig. S3), a pattern not readily observed at the genus composition level (Fig. 2A). To assess whether these two groups differ statistically, we applied K-means clustering [42] and analyzed their compositional differences. Both KO and genus compositions differed significantly between the two groups, though the differences were more pronounced at the functional level (mean R²: 0.28 for genus versus 0.55 for KO) (Fig. 2D, Additional file 1: Table S2). Given the clear divergence in functional composition with relatively minor taxonomic differences, we designate these groups as functional states—distinct microbiome configurations characterized by a specific combination of functionalities. The fact that these functional states consistently emerge along the first two principal coordinate axes in nearly all studies (except for Schandry_2021; Additional file 2: Fig. S4) underscores their role as primary drivers of compositional differences. Although Hou_2021 and Wolinska_2021 do not exhibit a clear separation of groups (Fig. 3), they still show substantial dissimilarity between the two K-means clusters (PERMANOVA R²: 0.33 for Hou_2021 and 0.29 for Wolinska_2021). In conclusion, while it is widely recognized that deterministic principles guide root community assembly [29, 30, 43], our findings reveal that these principles might lead to multiple functional outcomes, particularly evident in SynCom studies.

Fig. 3.

Contributions of pathways and genera to functional state separation. A The KOs that contribute to the functional state separation were taken from the sPLS-DA analysis and grouped into the pathway they function in. The mean contribution strength of the pathway was calculated from these KOs (x-axis). The y-axis indicates the number of studies in which KOs of these pathways were found to have a contributing effect. The size of the coordinate indicates how many KOs have been found in that pathway. B For each of the top five pathways in A, the proportion of the contributing KOs belonging to different bacterial families was averaged. High proportions for a family-pathway combination indicate that contributing KOs can almost exclusively be found for that bacterial family. C The isolates that contribute to functional state separation are grouped into genera displaying the mean contribution strength of the genus on the x-axis and the number of studies in which an isolate of the genus contributes to functional state separation on the y-axis. The size of the coordinate indicates how many isolates of each genus contributes to functional state separation, and the color indicates the phyla of the bacterial genera

The observation that these two functional states are less pronounced yet still detectable at the genus level suggests a potential phylogenetic influence. To investigate this, we analyzed compositional differences between the functional states using unweighted UniFrac distances, which consider both the presence or absence of taxa and their phylogenetic relationships (Fig. 2C, Additional file 2: Fig. S3). This analysis revealed clustering patterns like those observed at the genus level, with an unexpected overlap between Schandry_2021 and Lebeis_2015. Furthermore, the dissimilarity (R²) values between functional states were slightly higher when accounting for phylogeny (Fig. 2D, Additional file 1: Table S3), suggesting a modest phylogenetic signal associated with these distinct functional states.

Bacillus is the main driver of distinct functional states

We have shown that the root microbiome may organize itself into two functional states, a pattern observed across nearly all studies, suggesting a common underlying mechanism. This may involve the presence or absence of specific bacterial functional groups, potentially originating from similar taxonomic lineages. Given that functional composition consistently accounts for the largest variation between these states, we first examined the contribution of individual KOs to the state separation across studies. To identify KOs associated with each state, we performed sparse partial least squares discriminant analysis (sPLS-DA) on each study (Fig. 3A, Additional file 2: Fig. S5, Additional file 1: Table S3). This supervised machine learning method extracts key features that distinguish different classes within the data.

From the sPLS-DA, we identified 722 KOs including 669 unique ones, spanning 105 pathways that contribute to functional state separation across the nine studies. These KOs have a mean contribution of 0.082 (Additional file 1: Table S3) and a summed contribution of 1–23.2 per study. Interestingly, in three studies (Finkel_2020, Qi_2021, and Schandry_2021), the functional states are distinguished by only one KO each (Additional file 2: Fig. S5), which are involved in peptidoglycan biosynthesis, biofilm formation, and intermembrane transport, respectively. This suggests that, in these cases, the presence or absence of a single KO is sufficient to predict functional states with high accuracy. In contrast, the number of KOs that contribute to functional state separation in three other studies is more than 200 (Additional file 2: Fig. S5), highlighting the significant variability in how functional states are separated across studies. The contributions of individual KOs to these functional states are often minimal, resulting in only slight differences in their relative abundances between functional states. The large number of contributing KOs across studies suggests that phylogenetically related groups—whether comprising a few or many strains—tend to be present in one state and absent in the other. Since phylogenetically related bacteria tend to be functionally similar, they often share near-identical KO profiles. Therefore, the absence of these groups in a given state results in the concurrent loss of multiple KOs, each contributing incrementally to functional state separation. To better interpret these patterns, we focus on the pathways most frequently associated with functional state separation: peptidoglycan biosynthesis, quorum sensing, two-component systems, ABC transporters, and thiamine metabolism (Fig. 3A).

Across these pathways, 6.3 to 23.7% of KOs contribute to functional state separation (Additional file 1: Table S4). Surprisingly, the most influential KOs originate from the conserved peptidoglycan biosynthesis pathway, which exhibits both the highest proportion of contributing KOs (9 out of 38; 23.7%) and the highest mean contribution (0.22) (Fig. 3A). To further investigate their taxonomic origins, we filtered out widely distributed KOs, focusing instead on those that are more family-specific (Additional file 2: Fig. S6). Seven out of the nine peptidoglycan biosynthesis-related KOs meet this criterion, with six being restricted to Bacillaceae and Paenibacillaceae isolates (Additional file 2: Fig. S6). Many of the most influential KOs from the top five pathways are found in Bacillaceae isolates (Additional file 2: Fig. S6), with many being nearly exclusive to this family (Fig. 3B). This pattern is especially pronounced in the peptidoglycan biosynthesis, quorum sensing, and thiamine metabolism pathways. These findings suggests that Bacillaceae strains may be the primary contributors to functional state separation across the nine SynCom studies.

To confirm whether Bacillaceae strains are the main drivers of the functional state separation, we performed sPLS-DA on the isolates for each study (Fig. 3C, Additional file 2: Fig. S7, Additional file 1: Table S5). We identified 14 Bacillus strains that collectively contribute a mean of 0.31 to functional state separation in six studies, a substantial higher amount than any other bacterial genus (0.03–0.05).

Conclusively, we discovered that Bacillaceae-associated KOs and Bacillus strains drive the functional state separation (Fig. 3). To support this, we computationally excluded Bacillus and its associated KOs from the dataset and reanalyzed the level of functional state separation (Additional file 1: Table S6). Except for Finkel_2019, we still find significant functional states in the absence of Bacillus, though the PERMANOVA R² values that explain the functional states decrease by 65.3–90.10% in seven studies, corroborating the importance of this genus in functional state separation.

In comparison with genus-level compositional differences, we observe that the functional states observed across the different studies (Fig. 2B) are not evident on the genus level (Fig. 2A, Additional file 2: Fig. S3), even though they appear to be driven by Bacillaceae strains (Fig. 3B, C). This suggests that the Bacillaceae clade may harbor many family-specific KOs, which collectively drive significant functional compositional differences. In turn, genus compositional differences are not greatly affected by Bacillus being a single entity. To confirm that Bacillus strains are indeed more present or abundant in one functional state compared to the other, we investigated the cumulative relative abundance of Bacillus strains in the functional states of each study (Additional file 2: Fig. S8, Additional file 1: Table S6). Apart from Qi_2021, which lacks Bacillus strains in the SynComs, all studies show a significantly higher proportion of Bacillus in one functional state compared to the other. Typically, one functional state is consistently depleted of Bacilli, whereas the other exhibits variable Bacilli abundance. Interestingly, compared to other bacterial taxa that contribute to functional state separation, Bacillus often appears at low abundances even within the Bacillus-associated functional state. In contrast, most other taxa—aside from a few exceptions—require higher abundances to influence functional state separation (Additional file 1: Table S6). These observations suggest that Bacillus may establish in the root microbiome through an all-or-nothing strategy, although the ecological significance of both Bacillus and this colonization pattern remains unclear.

Bacillus-associated states appear independent of the presence of other bacterial genera

The variable abundance of Bacillus in one functional state, paralleled to its absence or reduced presence in the other state, raises questions about the factors influencing the root microbiome’s functional state. One possible explanation is the effect of the other resident bacteria in the root microbiome, which may either positively or negatively affect Bacillus species or, on the other hand, thrive or decline because of the presence or absence of Bacillus. To test this hypothesis, we conducted a network analysis at the genus level across the nine studies to investigate which bacterial genera significantly correlate with Bacillus, positively or negatively. We focused on the top 10 most contributing bacterial genera to functional state separation (Additional file 1: Table S5) and examined how these top contributors are related to each other. Interestingly, we found Bacillus to be isolated from all other top contributors, with only weak but significant negative correlations with Acidovorax, Pseudomonas, and Flavobacterium, which positively co-occur (Additional file 2: Fig. S9). This finding aligns with current understanding that Pseudomonas and Bacillus primarily engage in competitive interactions within the root microbiome, evidenced by extensive co-culturing studies [44, 45]. The absence of positively correlating strains further supports the hypothesize that Bacillus establishes relatively independently of the presence or activity of other bacterial taxa. On the other hand, the high abundance of bacterial genera such as Pseudomonas might be more often associated with a non-Bacillus functional state. Further investigation is needed to better understand the specific interactions that lead to these distinct functional states in the root microbiome.

Functional states correlate negatively with complexity

Because these findings stem from SynCom studies that used communities of lower complexity than the natural root microbiome, we investigated whether the presence of two functional states depends on community complexity. Interestingly, we observed a significant negative correlation between SynCom complexity and the compositional separation of the functional states (Additional file 2: Fig. S10). This suggests that functional states are more pronounced in simpler SynComs, consistent with prior observations in the highly complex communities used by Hou_2021 and Wolinska_2021 (Additional file 2: Fig. S3), where functional state separation was less distinct. Whether this trend persists in natural, inherently more complex communities remains unclear and warrants further experimental investigation.

Bacillus strains benefit from plant’s defense systems and loss of receptors

In addition to examining the influence of bacterial genera and SynCom complexity on functional states, we investigated how plant genotype and environmental factors might shape these states. Both plant genotype and environmental conditions have been shown to significantly impact root microbiome composition [43], potentially steering it toward one functional state or another. The studies included in this analysis encompassed a variety of treatments and conditions across multiple plant genotypes. This diversity enabled us to assess whether specific treatment or environmental factors are associated with functional states. To identify such associations, we conducted binomial statistical tests for overrepresentation, as detailed in Fig. 4 and Additional file 2: Fig. S11.

Fig. 4.

Overrepresentation analysis of genotypes and treatments in non-Bacillus or Bacillus functional states. Each point represents a unique genotype or treatment from one of the studies, color-coded by study for clarity. The x-axis denotes the proportion of samples within the Bacillus functional state, while the y-axis displays the negative logarithm (base 10) of the p value from a binomial test, indicating the statistical significance of the association. A higher position on the y-axis signifies a stronger statistical association between the genotype or treatment and the functional state

Our analysis revealed numerous significant associations between plant genotypes, treatments, and functional states within the root microbiome. For instance, bacterial communities were more likely to align with a Bacillus-associated functional state when the plant’s salicylic acid defense system was overexpressed (e.g., in cpr1 and cpr5 mutants) or when receptors for microbe-associated molecular patterns (MAMPs) were knocked out (e.g., bak1bkk1cerk1, fls2, nfr5, and apex mutants) (Additional file 1: Table S7). Conversely, impairing the plant’s defense system (e.g., in deps, cyp79b2b3, and myc2 mutants) was associated with a shift toward a non-Bacillus-associated functional state (Additional file 1: Table S7). These findings suggest a complex interplay between plant immune responses and Bacillus strains, with different components of the immune system influencing these bacteria in varied ways. We hypothesize that locally induced immunity via MAMP-triggered immunity (MTI) may suppress Bacillus strains, while systemic activation of plant immunity through mechanisms like induced systemic resistance (ISR) or systemic acquired resistance (SAR) may stimulate their presence. Additional findings show factors such as benzoxazinoid APO and low pH levels to negatively impact Bacillus strains (Fig. 4). Also, in Duran_2021, the presence of a fungal pathogen can lead to either Bacillus- or non-Bacillus-associated functional states (Additional file 2: Fig. S11), whereas in the absence of the pathogen the Bacillus-associated state consistently emerges for this specific SynCom composition. Due to the lack of data on fungal abundance, we could not assess whether direct interactions between the fungal pathogen and Bacillus underlie this state separation. Overall, these observations suggest that, aside from low pH, most environmental changes have limited impact on functional state outcomes. However, comprehensive experimental validation is necessary to confirm these hypotheses.

Discussion

SynCom reconstitution experiments have become increasingly popular for investigating the root microbiome [25, 46–48]. These experimental set-ups provide insights into microbiome dynamics on a reduced scale while offering information on microbial functions within the root microbiome [25]. However, the way root communities are established in SynCom reconstitution experiments differs from natural root communities. In natural soil, root communities develop from soil communities that vary spatiotemporally [49–51]. In contrast, SynCom experiments involve pooling bacterial species in proportionally similar abundances and inoculating them with plants to investigate root community establishment [52, 53]. The here-described meta-analysis demonstrates that inoculating taxonomically diverse bacterial strains results in functional overlap across studies, suggesting a common functional similarity in bacterial root colonization strategies. This overlap, however, is not clearly visible on the abundance level of KOs, which may have multiple explanations. One possibility is that SynCom members exhibit a degree of soil specificity in their functional profiles. Alternatively, the selection of rhizobacteria in some studies may not accurately represent the root microbiome, leading to differences in observed microbiomes. The way compositional differences are assessed—such as by using Bray–Curtis distances—may also influence results. KOs present in nearly all isolates, representing conserved bacterial functions, and those highly abundant within genomes tend to outweigh the effects of rare KOs in compositional similarity analyses. However, root-competence-associated functions can include both conserved and rare KOs [27, 53].

Interestingly, we identified functional dissimilarity in the root microbiome that separates samples into two distinct states: a Bacillus-associated state and a non-Bacillus-associated state. These Bacillus-driven functional states appear on the two largest principal coordinate axes in nearly all studies, implying that Bacillus has a substantial effect on shaping the root microbiome’s functional composition. In Schandry_2021, the assigned functional states appear to be defined by the first and third principal coordinate axes instead (Additional file 2: Fig. S4), suggesting that in this study there may be multiple primary drivers of functional compositional differences that are not defined by the first two principal coordinate axes. In consideration that this is the only study in which SynComs are inoculated with root exudates instead of inoculated with the plant, the plant’s presence might be the factor that leads to more defined functional states. The presence of multiple, dynamic, but stable root communities suggests the existence of plant enterotypes or rhizotypes—distinct compositional states in the root microbiome [54]. In the human gut, enterotypes reach robust equilibria that can shift with prolonged changes in dietary patterns [55]. Similarly, different enterotypes can be found in various human cultures due to dietary differences, analogous to how different rhizotypes may occur among different crop species [2]. Since these distinct functional states are not found at the microbiome level where bacterial abundances are considered like they are in enterotypes, we believe the term “functional state” to be more appropriate than “rhizotype.”

The association between Bacillus species and the two functional states observed in root communities is complex and not yet fully understood. Both conserved and Bacillaceae-specific functions contribute to functional state separation, but their roles in determining the root microbiome’s overall composition remain speculative. Bacillus species employ quorum sensing and biofilm formation as key strategies for root colonization [56]. Quorum sensing enables bacteria like Bacillus to coordinate collective behaviors, including biofilm formation, which enhance their ability to establish stable communities on plant roots. An in-depth analysis of biofilm formation within the Bacillus clade revealed considerable variation among biofilm formation strategies, underscoring its importance for Bacillus strain fitness [57–59]. Notably, Bacillus strains appear to cooperate in biofilm formation, which can be more effective than individual strategies [60]. We hypothesize that successful Bacillus biofilm establishment may promote a Bacillus-associated functional state. Conversely, failure to establish a biofilm could lead to exclusion from the root microbiome.

This root colonization strategy, along with the observation that Bacillus is not necessarily the most abundant genus on the root, supports the hypothesis that Bacillus abundance is highly spatiotemporally organized, potentially through localized biofilm formation. To investigate this spatial and temporal organization within the root microbiome, microfluidic and microscopy-based approaches could be employed. Previous studies using such methods have shown that Bacillus subtilis colonizes roots through pulsating crowd movements, preferentially targets the root elongation zone, and can exclude other soil bacteria following biofilm establishment [61, 62]. These findings offer preliminary insights into the mechanisms that may underlie the formation of the Bacillus-associated functional state.

The frequent presence of Bacillus in the root microbiome suggests that evolutionary pressures have shaped its adaptation to a root-associated lifestyle. A long-standing relationship with the plant may have contributed to the emergence of distinct functional states characterized by its presence. Other ubiquitous bacterial taxa in the root microbiome, like Rhizobium, Pseudomonas, or Streptomyces, also suggest long-term plant associations, yet we do not consistently observe functional states characterized by these taxa. This does not preclude their involvement in functional state separation. Bacillus, however, appears to provide a unique set of KOs, enabling clear differentiation of functional states. The apparent functional divergence of Bacillus from other ubiquitous root inhabitants may thus explain the emergence of Bacillus-associated functional states. Experimental validation using SynCom drop-out studies targeting individual taxa could help determine whether other bacteria also drive shifts in functional composition and contribute to the formation of distinct root microbiome states.

Its ubiquity in the root microbiome does not necessarily imply a commensal lifestyle, as many Bacillus strains are known for their plant-beneficial properties, including pathogen suppression and growth promotion. Notable examples include Bacillus amyloliquefaciens, B. subtilis, and Bacillus velezensis [6, 63–65]. However, field trials have shown inconsistent success with these bioinoculants [6]. We hypothesize that this variability may be partly explained by the existence of both Bacillus-associated and non-Bacillus-associated functional states in natural root communities. These functional states could affect the efficacy of Bacillus-based bioinoculants in agricultural settings, underscoring the importance of identifying the factors that govern Bacillus-associated functional state formation.

The plant’s immune system is a key factor in shaping its root communities [66]. Activation of the immune system through mechanisms like MAMP-triggered immunity (MTI), systemic acquired resistance (SAR), or induced systemic resistance (ISR) can lead to various responses, such as reinforcement of cell walls, release of reactive oxygen species (ROS), and production of antimicrobial compounds [3, 67–77]. These immune responses can affect root communities [34, 78–80]. Our findings suggest that the plant immune system may affect the functional state of the root community. The exact mechanisms driving this effect are, however, unclear, as both local and systemic components of the plant immune system appear to influence the functional state in different ways.

When the plant immune system is downregulated, knocked down, or knocked out, we might expect a reduction in Bacillus populations due to the absence of defense responses that typically negatively affect competing gram-negative bacteria. These bacteria are reportedly more susceptible to diverse plant compounds, like salicylic acid, flavonoids, camalexins, and ROS [81–83], while gram-positive bacteria like Bacillus generally benefit from a thicker peptidoglycan cell wall, offering them better protection against such compounds [83]. Bacillus strains may exhibit greater tolerance to plant-derived antimicrobial metabolites compared to some other bacteria. However, they might still be adversely affected by root microbes that have also adapted to these metabolites. Conversely, systemic plant defenses appear to positively influence Bacillus strains. Given that the Bacillus genus harbors multiple plant-beneficial bacteria, they may positively respond to a plants’ cry for help following aboveground infection and subsequent activation of systemic immune responses [6, 63–65]. However, it remains unclear whether Bacillus species are predominantly plant-beneficial, commensal or pathogenic, as research on this genus may be subject to bias. In addition, SynCom studies that study the plant root microbiome, including most of the studies here, mostly involve the model plant Arabidopsis thaliana (Additional file 1: Table S1). Thus, the functional states that we reveal here might possess a degree of Arabidopsis specificity. Additionally, our findings suggest that lignins, which have been reported to negatively affect Bacillus and other gram-positive species [84, 85], may explain the presence of a dominant Bacillus-associated functional state in the Arabidopsis root microbiome of a lignin-deficient myb15 mutant. On the other hand, several studies report the existence of lignin-degrading Bacillus strains [86, 87], which suggests that some Bacillus strains might actually favor lignin deposition in the root, leading to a Bacillus-associated functional state. In this particular study, Wippel_2021, a genetic screen of the Bacillus strains used in the SynCom might reveal whether these are able to degrade lignin and whether they might prefer lignin deposition in the root or not.

In addition to examining microbe-microbe and plant–microbe interactions, we evaluated the impact of various abiotic and biotic factors on functional states. Schandry et al. [38] demonstrated that gram-negative bacteria exhibit higher tolerance to the benzoxazinoid APO compared to gram-positive bacteria, a result we corroborate (Fig. 4). Notably, all APO samples were associated with the non-Bacillus-associated state. Duran et al. [37] reported no significant difference in Bacillus abundances in the root microbiome with or without a fungal pathogen, although Bacillus levels were slightly higher in the absence of the pathogen, a trend supported by our analysis. Conversely, Bacillus bacteria are known to associate with beneficial fungi [88–95] and pathogenic fungi [6, 63, 64, 90]. However, experimental validation is required to confirm positive interactions with beneficial fungi and negative interactions with pathogenic fungi. Additionally, inorganic compounds may also influence the functional state as we observed Bacillus depletion when plants were grown at pH 5.5, consistent with studies showing that Bacillus growth is severely inhibited at low pH [96–98]. Overall, our findings are in line with previous research on environmental perturbations leading to a Bacillus or a non-Bacillus functional state. The relationship between perturbation and functional state may be both direct, e.g., through a low pH or the presence of the benzoxazinoid APO, or indirect, e.g., through the plant’s immune response that may or may not change the community composition. In consideration that distinct bacterial taxa harbor different sets of KOs within their genome, a change in functional composition also means a taxonomic compositional change. These taxonomic compositional changes, however, may be less visible compared to functional compositional changes, especially when the changes are on the genus or species level.

Finally, the presence of Bacillus-associated functional states in natural soils remains uncertain. Our analysis shows a negative correlation between the presence of functional states and SynCom complexity. In SynComs with moderate complexity, functional states are less distinct compared to those in medium or low complexity settings, although differential Bacillus abundances are still observable. This negative correlation may be influenced by technical biases, as more complex communities tend to exhibit higher functional diversity, potentially diminishing the impact of Bacillus-derived functions on the overall community. Exploring Bacillus populations in natural rhizospheres through publicly available datasets could shed light on whether Bacillus strains drive functional state separation in natural root communities. Xu et al. [96], for example, have shown the distribution of B. subtilis across a variety of different natural soils, with pH being the main driver of Bacillus abundances that are in accordance with our findings. This suggests that plant root microbiomes might exist in multiple functional states in natural soils as well, though more experimental evidence would be required to confirm this.

Conclusions

In this study, we demonstrate that root microbiomes inoculated with synthetic communities assemble into two distinct functional states, distinguished by the presence or absence of Bacillus. Functions driving the Bacillus-associated state are linked to plant–microbe and microbe-microbe interactions, including quorum sensing and biofilm formation. We further show that abiotic and biotic factors may influence these states, illustrated by statistical associations of the plant’s local or systemic immune system and the non-Bacillus- and Bacillus-associated states, respectively. The identification of multiple Bacillus-driven functional states enhances our fundamental understanding of root microbiome dynamics and assembly. Recognizing the role of functional state-defining—and driving—bacterial taxa may improve bioinoculant development by addressing how these states affect bioinoculant establishment. Moreover, root microbiomes could potentially be engineered to favor specific functional states that support plant-beneficial microbes. Although we acknowledge that many findings from this meta-analysis require experimental validation, the observed effects of these functional states on microbial functions at the presence/absence level underscore the value of SynCom approaches for studying root microbiomes.

Methods

Overview of studies

To conduct a meta-analysis of published SynCom studies, the amplicon reads of nine different studies were downloaded from NCBI on August 3, 2022 (BioProject numbers shown in Table 1, extensive description in Additional file 1: Table S1). Together, these studies involve three different plant species and over 20 different plant genotypes [99–125]. Additionally, the genomes and the genomic reads of the SynCom isolates that were used in all studies were downloaded from either At-SPHERE (www.at-sphere.com), NCBI (BioProject: PRJNA297942 (used in Duran_2021, Hou_2021, Schandry_2021, Wippel_2021, and Wolinska_2021 [35–39]), Lj-SPHERE (www.at-sphere.com) (used in Wippel_2021 [39]), or from the DOE Joint Genome institute (GOLD study ID: Gs0019718 (used in Finkel et al., Finkel et al., Lebeis et al., and Qi et al. [31–34]). These SynCom studies were selected based on the following criteria: they involved SynComs inoculated on plant roots or exposed to root exudates, consisted of moderate to high complex communities, and had both microbiome and genomic data available. In these studies, the SynComs were generally designed to represent the natural root community, with the following exceptions: Lebeis_2015 included bacterial strains that were differentially abundant in the roots of immunocompromised Arabidopsis genotypes; Qi_2021 designed three SynComs based on their ability to inhibit root growth through auxin degradation [33, 126]; Schandry_2021 designed SynComs based on tolerance or sensitivity to benzoxazinoids (BXs); and Wippel_2021 combined both Arabidopsis- and Lotus-derived rhizobacteria into SynComs to study host specificity.

Table 1.

Datasets included in this meta-analysis and corresponding BioProject IDs

Study	BioProject number
Duran_2021	PRJEB43139
Finkel_2019	PRJNA531340
Finkel_2020	PRJNA543313
Hou_2021	PRJEB40980
Lebeis_2015	PRJEB9723
Qi_2021	PRJNA720397
Schandry_2021	PRJEB42439
Wippel_2021	PRJEB37695
Wolinska_2021	PRJEB45520

Isolate quantification

We quantified SynCom isolate abundances using SyFi (SynCom fingerprinting), which leverages assembled genomes and raw genomic sequencing reads to generate strain-specific 16S rRNA amplicon fingerprints [127]. These fingerprints serve as references for pseudoalignment-based quantification of metagenomic amplicon reads. Amplicon reads were first error-corrected, chimera-filtered, and trimmed using DADA2 (version 1.26.0) with default settings, adjusting trimming lengths based on read quality [128] (Additional file 1: Table S8). SyFi then retrieved strain-specific 16S rRNA V3–V4 or V5–V7 fingerprints for each study (Additional file 1: Table S9) and performed pseudoalignment with Salmon (version 1.4.0), applying a minimum sequence identity score threshold of 0.95 to balance sensitivity and specificity [127, 129]. Absolute read counts were normalized by strain-specific 16S rRNA copy numbers, generating an isolate count table.

Since SyFi requires genomic reads of SynCom strains, which were unavailable for some studies (Supplementary data 1), we conducted a parallel ASV-based analysis. Here, DADA2-filtered reads were processed in Qiime2 (version 2019.7) to infer ASVs [130], which were then matched against the public SynCom genome sequences at 99% sequence identity using VSEARCH [131]. The resulting isolate counts were normalized using SyFi’s 16S rRNA copy number estimates.

We compared SyFi-pseudoaligned read counts with Qiime2-VSEARCH matched reads (Additional file 1: Table S10). Unless Qiime2-VSEARCH captured substantially more reads, we prioritized SyFi’s isolate count dataset. In summary, the Qiime2-VSEARCH method was used only for Lebeis_2015 and Finkel_2020.

Functional composition

To study bacterial functions in the root microbiome, we first annotated SynCom genomes using PROKKA (version 1.14.6) and eggNOG-mapper (version 2.1.4–2) [132, 133]. From these annotations, we extracted KEGG orthologs (KOs), quantified their occurrences per SynCom member, and compiled a genomic KO data table (Supplementary data 4) [134].

To estimate KO abundances in the microbiome, we applied PICRUSt2, a metagenome prediction algorithm that matches ASVs to bacterial genomes and infers their functional composition table [21]. Instead of relying on database-matched genomes, we used our genomic KO table and microbiome isolate count table as input for PICRUSt2. This approach ensured that functional predictions were directly linked to SynCom members rather than inferred from unrelated reference genomes. The 16S rRNA copy number normalization step in PICRUSt2 was omitted, as the microbiome isolate count table was already normalized for 16S rRNA copy number.

Diversity analyses

To investigate the similarity of the taxonomic and functional compositions of microbiome samples across and within studies, the Jaccard, Bray–Curtis, and unweighted UniFrac distances between samples were calculated. The Jaccard distances indicate compositional similarity based on the presence/absence of features, while the Bray–Curtis distances indicate compositional similarity based on feature abundances. The unweighted UniFrac distances, in turn, indicate compositional similarity based on presence/absence of taxa and their evolutionary relationships. For the Jaccard distances, the genus and KO count tables were converted to relative abundance tables and processed in Qiime2 (version 2019.7) [130], using the beta-diversity plugin with the Jaccard metric to calculate the between-sample Jaccard distances. To calculate Bray–Curtis distances (dissimilarities), we first added a pseudocount to the genus and KO abundance tables before normalizing it to relative abundances, as the Qiime2 version used (version 2019.7) did not support Bray–Curtis calculations with zeros. The normalized table was then imported into Qiime2 and processed with the beta-diversity plugin to compute Bray–Curtis dissimilarities [130]. The unweighted UniFrac distances were processed in Qiime2 as well, using the relative abundance isolate table without a pseudocount and the isolates’ 16S rRNA sequences as input.

To assess the significance of compositional differences between sample groups, we conducted PERMANOVA analyses using adonis2 from the vegan R package (version 2.6–4) [135] on Jaccard, Bray–Curtis, and unweighted UniFrac distance matrices. PERMANOVA quantifies the proportion of compositional variation explained by metadata variables such as study, SynCom, or root compartment.

Sparse partial least squares discriminant analysis

To identify bacterial KOs or isolates that drive sample group separation in the PCoA space of Jaccard distances, we performed sparse partial least squares discriminant analysis (hereafter sPLS-DA) using mixOmics (version 6.24.0) in R [136]. sPLS-DA is a supervised classification method that prioritizes features distinguishing two classes. We first optimized hyperparameters through a grid search, selecting the optimal number of principal components, dataset subsets, iterations, and subset sizes. Principal components correspond to PCoA dimensions, while dataset subsets define sample grouping and partitioning strategy. The tuned hyperparameters were then used to run sPLS-DA, generating mean contribution scores for each KO or isolate, highlighting their role in group separation.

Network analysis

Co-occurrence of bacterial genera across the nine studies was assessed using network analysis. Total abundance values were CLR-transformed and analyzed with FastSpar (SparCC) (version 1.0.0) [137]. Median correlations were computed from 1000 iterations, and significance was evaluated using 1000 bootstraps.

Authors' contributions

G.S. and R.d.J. conceptualized the conducted meta-analysis. G.S. gathered and investigated the data, conducted the bioinformatic analyses, visualised the results, and wrote the manuscript. R.d.J. oversaw project administration, supervision, and revised the manuscript. All authors read and approved the final manuscript.

Funding

This study was supported by the Novo Nordisk Foundation Grant no. NNF19SA0059362.

Data availability

The supplementary tables and datasets are available at the following Zenodo link: 10.5281/zenodo.14982431 [138]. The Zenodo repository includes an R script that generates all figures and supplemental figures in the manuscript and performs the sPLS-DA analysis. A zip file containing all necessary files to run the R script is included as well. Data S11, containing the Jaccard, Bray-Curtis and Unweighted UniFrac distance matrices, can also be found in this repository.

An overview of the data can be found in Additional file 3: Data S1, which includes BioProject IDs of the nine studies, SRA IDs of the SynCom microbiome samples , IMG IDs and Gold IDs of the DOE Joint Genome institute-derived bacterial genomes , and BioProject IDs of the At-Sphere-derived bacterial genomes . Also included in this data overview are the BioProject IDs and SRA IDs of the WGS data related to both the DOE Joint Genome institute- and At-Sphere-derived bacterial genomes.

Declarations

Ethics approval and consent to participate

Not applicable.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.