Bioactive diterpenoids impact the composition of the root-associated microbiome in maize (Zea mays)

Plants deploy both primary and species-specific, specialized metabolites to communicate with other organisms and adapt to environmental challenges, including interactions with soil-dwelling microbial communities. However, the role of specialized metabolites in modulating plant-microbiome interactions often remains elusive. In this study, we report that maize (Zea mays) diterpenoid metabolites with known antifungal bioactivities also influence rhizosphere bacterial communities. Metabolite profiling showed that dolabralexins, antibiotic diterpenoids that are highly abundant in roots of some maize varieties, can be exuded from the roots. Comparative 16S rRNA gene sequencing determined the bacterial community composition of the maize mutant Zman2 (anther ear 2), which is deficient in dolabralexins and closely related bioactive kauralexin diterpenoids. The Zman2 rhizosphere microbiome differed significantly from the wild-type sibling with the most significant changes observed for Alphaproteobacteria of the order Sphingomonadales. Metabolomics analyses support that these differences are attributed to the diterpenoid deficiency of the Zman2 mutant, rather than other large-scale metabolome alterations. Together, these findings support physiological functions of maize diterpenoids beyond known chemical defenses, including the assembly of the rhizosphere microbiome.


Background
Extensive research in recent years has demonstrated the composition and importance of rhizosphere microbial communities to plant health and tness [1][2][3]. The cooperative partnership of microbes and plants has been attributed largely to an exudation of photosynthate sugars from the plant in exchange for nutrient supply and protection against biotic and abiotic stress, ultimately contributing to increased plant vigor and yield [4,5]. Despite extensive characterization of species-and tissue-speci c microbial communities and how these vary in their ecological and genetic contexts, our understanding of the mechanisms by which plants recruit and maintain root-associated microbial communities is still limited [5][6][7][8][9][10][11].
As the most economically important crop in the United States, maize (Zea mays) has been the subject of long-standing research to improve crop yield and stress resilience traits [12]. The "core" microbiome of maize has provided insights into the presence of speci c phyla in and near maize roots [6, 13,14]. A study of 27 maize inbred lines across developmental stages and geographical locations demonstrated maize genotype as a replicable factor in de ning the microbiome, with signi cant variation due to the different genetic backgrounds among inbred lines. Five "core" Operational Taxonomic Units (OTUs), all in the phyla Proteobacteria, were found to be present in all samples [14]. Drawing on these detailed insights into the maize microbiome, further research has ventured into determining the chemical signaling factors that determine microbiome composition and variation across genotypes. Specialized metabolites mediate various plant interactions with the environment and other organisms, and several major bioactive metabolite groups have been identi ed in maize, including diterpenoids, sesquiterpenoids, oxylipins, and benzoxazinoids [15][16][17][18][19][20]. Given their distinct structures, bioactivities and abundance across different maize tissues and developmental stages, maize specialized metabolites can be hypothesized to play a critical role in the plant communication with root-borne microbes and thus help determine the root microbiome. Recently, benzoxazinoids were shown to in uence the maize response to stress via altered rhizosphere microbial communities mediated by 6-methoxy-benzoxazolin-2-one (MBOA) [21]. Additional work on benzoxazinoids mutants showed changes to fungal and bacterial communities as a result of benzoxazinoid de ciency [22]. Furthermore, benzoxazinoid-de cient mutants were shown to feature dramatically altered root metabolomes, suggesting alterations in the microbiome may not be solely attributed to a lack of speci c benzoxazinoids, but rather global changes in root metabolites in mutant plants [23]. Alongside these ndings, there is a large degree of variation in the observed differences in alpha and beta diversity in benzoxazinoid mutants [21][22][23].
The diverse group of terpenoid metabolites also has shown to be critical in mediating above-and belowground interactions between plants and other organisms, including microbes [24]. Kauralexins and dolabralexins are two major diterpenoid groups in maize that have demonstrated or predicted roles in biotic and abiotic stress responses [16][17][18]25]. Kauralexins show stress-elicited accumulation in several tissues, including stems and scutellum, and mediate quantitative defenses against fungal pathogens such as species of Fusarium, Aspergillus and Cochliobolus, as well as insect pests including the European corn borer (Ostrinia nubilalis) [17,[26][27][28]. The more recently discovered group of dolabralexins shows pathogen-inducible accumulation predominantly in roots and, like kauralexins, have strong growth-inhibitory activity against Fusarium pathogens [18]. In addition to their defensive potential against biotic stressors, both kauralexin and dolabralexin production was shown to be increased in response to abiotic stress, such as drought or oxidative stress [16,18]. Kauralexins and dolabralexins derive from a common precursor, ent-copalyl diphosphate (ent-CDP), which is also shared with the gibberellin (GA) biosynthetic pathway critical for plant growth (Fig. 1) [17,29]. Two catalytically redundant diterpene synthase (diTPS) enzymes, ANTHER EAR 1 (ZmAn1) and ANTHER EAR 2 (ZmAn2) control ent-CDP formation in maize [29,30]. Genetic studies revealed that ZmAn1 is critical for GA biosynthesis, whereas ZmAn2 feeds ent-CDP into kauralexin and dolabralexin biosynthesis, thus enabling a pathway partition separating precursor ux toward general and specialized diterpenoid pathways [29][30][31] (Fig. 1). This pathway separation is supported by the phenotype of the Zman2 mutant, which features a loss of function in the Zman2 gene through a stable Ds insertion from the Activator (Ac) and Dissociation (Ds) system [16,29]. Zman2 has kauralexin and dolabralexin de ciency but normal GA levels [16]. Although it has a normal growth phenotype, Zman2 shows increased susceptibility to biotic and abiotic stress as compared to its WT sibling, consistent with the protective bioactivity of kauralexins and dolabralexins [16]. This phenotype, paired with the bioactivity of maize diterpenoids and their accumulation in roots in response to abiotic stress, suggests a possible role of these metabolites in broader plant-microbe interactions, including root microbiota.
To test this hypothesis, we investigated the microbial communities of the diterpenoid-de cient Zman2 mutant compared to its wild type (WT) sibling under both well-watered and drought conditions. Using sequencing of the microbial 16S rRNA gene, we detected and classi ed OTUs to analyze the structure and variation of bacterial communities in both the rhizosphere and endosphere. We subsequently provide evidence that dolabralexins can be exuded from the roots, and that the Zman2 mutant lacks other signi cant metabolic changes other than diterpenoid content, suggesting that diterpenoid de ciency is responsible for the altered microbiome.

Results
Dolabralexins are secreted outside of maize roots To examine a possible role of maize diterpenoids in plant-microbiome interactions, we utilized the Zman2 mutant genotype in comparison to its isogenic WT sibling [16,29]. Previous studies showed that 30-dayold WT maize plants signi cantly accumulate dolabralexins, and to a lesser extent kauralexins, in roots [16], whereas the Zman2 mutant genotype is almost completely devoid of these diterpenoids [16][17][18]. Despite the de ciency of both kauralexins and dolabralexins in Zman2 root tissue, mutant plants did not show an apparent phenotype under well-watered conditions ( Fig. 2A), consistent with prior reports describing largely unaltered root and shoot weight, developmental features, and GA and zealexin levels in the Zman2 mutant [16]. Zman2 plants were used as a control for analyzing root metabolite exudation due to the de ciency of diterpenoids, and subsequent analysis of the microbiome and metabolome.
To determine a possible ability of maize diterpenoids to affect the rhizosphere microbiome, we rst tested if diterpenoids can be exuded from maize roots. For this purpose, 38-day-old maize plants, both Zman2 and its WT sibling, were grown on soil then gently cleaned and subsequently suspended for 48 hours in nutrient water. After removing the plants, metabolites were extracted from the nutrient water using an equal volume of ethyl acetate and analyzed by LC-MS/MS against authentic metabolite standards. As a positive control, the benzoxazinoid 1,3-benzoxazol-2-one (BOA), known to be secreted from maize roots, was measured and used as a standard to detect BOA and other benzoxazinoids metabolites with similar mass spectra. Benzoxazinoids were found to be present in both Zman2 and WT plant exudates, while absent in nutrient water without plants, as expected (Fig. 2B). Of the known dolabralexins for which standards were available -epoxydolabradiene, epoxydolabranol, and trihydroxydolabrene -only trace amounts were detected in nutrient water after incubation of the Zman2 mutant (Fig. 2B), as expected based on the known mutant phenotype. Trihydroxydolabrene and epoxydolabradiene were both signi cantly enriched in the WT root exudate samples than Zman2 or nutrient water without plants (Fig. 2B). Epoxydolabranol was not detected in mutant nor WT plant exudates.

Maize Root Microbial Communities Are Distinct By Compartment
The Zman2 mutant genotype and its corresponding WT sibling serve as a tool to investigate the effect of diterpenoids, or the lack thereof, on the maize root microbial community [16,18]. Under abiotic stress conditions, while WT maize plants were shown to have increased levels of kauralexin and dolabralexin metabolites and a greater root/shoot ratio, Zman2 plants were shown to be more susceptible to stress conditions via earlier onset of leaf curling and a reduced root/shoot weight ratio [16]. Based on this knowledge and the growth conditions of previous research on the mutant genotype, we used one-monthold Zman2 and isogenic WT sibling plants to comparatively examine the impact of diterpenoid-de ciency on the maize root microbiome. Half of the plants were treated with drought stress in order to investigate if changes in the microbiome exist under ideal and/or stress conditions, and if these changes may suggest an in uence on the Zman2 ability to cope with abiotic stress. Representative plant images are shown in Fig. 2A.
Microbiomes of the rhizosphere (1-2 mm of soil outside the root) and endosphere (inside the root), the latter representing root samples after removal of rhizosphere and rhizoplane microbes through washing and sonication of the roots, were analyzed. Bulk soil without plants was used as a control to examine background soil microbial communities. The 16S rRNA gene (V4 region) was sequenced using Illumina MiSeq and sequences were clustered into operational taxonomic units (OTUs) using the QIIME pipeline and the Greengenes database [32]. After ltering to remove mitochondrial and chloroplast OTUs, 4,259 distinct OTUs remained. OTU counts were then normalized by relative abundance, which was used rather than rarefaction methods so as not to discard low abundance OTUs. In all, four factors were analyzed: compartment, genotype, water status, and the interaction between genotype and water status.
Consistent with previous research in maize and other plant species [7,33], the microbial communities of the two plant compartments and the bulk soil were all statistically distinct. The alpha diversity, as measured by the Shannon index, showed the greatest diversity of microbes in bulk soil, with reduced diversity in the rhizosphere and further reduction in the endosphere (Fig. 3). A permutational multivariate analysis of variance (PERMANOVA) was used to measure the diversity between samples (beta diversity) and showed that, when accounting for all factors, compartment accounts for 23% of the variation between samples (p < 0.001). This was con rmed by a principle coordinate analysis (PCoA), in which compartment was the greatest source of variation (Fig. 4). A total of 960 OTUs, in 21 phyla (out of 32 total phyla), were enriched in the rhizosphere as compared to the endosphere, whereas 82 OTUs in 9 phyla were enriched in the endosphere as compared to the rhizosphere, as determined using the DESeq2 package in R. Among the 10 most abundant phyla plotted for each sample type, some phyla were found to be enriched in both compartments, whereas the rhizosphere was predominantly enriched for OTUs in the phyla Actinobacteria, Acidobacteria, and Alphaproteobacteria (Fig. 5).
The endosphere did not demonstrate signi cant differences attributed to genotype, water treatment, nor their interaction, as determined by PERMANOVA and PCoA (Fig. 4B). No individual OTUs were signi cantly enriched or depleted in regards to sample type in the endosphere as analyzed by generalized linear models. Because there was no apparent difference, all further analysis focuses on the rhizosphere compartment only, and the bulk soil and endosphere samples were omitted from further analyses.

Rhizosphere Microbial Communities Are Distinct Under Different Watering Conditions
Drought was de ned in this study as a withholding of water for seven days, with a single watering on day four. For both WT and Zman2, drought-treated plants showed severe leaf curling, but were still green and not wilted over completely ( Fig. 2A). Volumetric water content (VWC) of the soil surrounding well-watered plants was 10.7 ± 1.9 percent, while drought plants was 6.9 ± 1.9 percent. When accounting for all rhizosphere samples, water status accounted for 12.6% of the observed microbiome variation (p < 0.001), and visibly separated sample types in a principal components analysis (Fig. 4C).
We next identi ed OTUs enriched or depleted across treatments using the R software package DESeq2. When including all plants, irrespective of genotype, drought-stressed plants were enriched for 51 OTUs, which were predominantly composed of taxa from the Phylum Actinobacteria (Fig. 6A, OTU abundances by sample type in Supplemental Fig. 1). Well-watered plants were enriched for 97 OTUs as compared to drought conditions, mostly represented by Bacteriodetes, Alphaproteobacteria, Gammaproteobacteria, and Betaproteobacteria (Fig. 6A, OTU abundances by sample type in Supplemental Fig. 2). In bulk soil samples, 21 OTUs, including Bacteroidetes and Verrucomicrobia, were enriched, whereas six OTUs, primarily Actinobacteria, were depleted under well-watered conditions (Fig. 6B). Of these, 21 signi cant OTUs were also enriched or depleted in the rhizosphere samples as compared to the bulk soil control samples, suggesting they are in uenced by drought stress regardless of any plant-microbe interactions ( Fig. 6C). Together, these results are consistent with the demonstrated contribution of water status to the rhizosphere microbiome composition [33][34][35], and suggest that environmental as well as host-controlled processes in uence the composition of rhizosphere-associated microbiomes under drought stress.
The Zman2 mutant features a distinct microbial community composition Next, the impact of genotype on the rhizosphere microbiome composition was assessed by investigating each water status separately for genotype effects by water treatment (GxD). Under well-watered conditions, signi cant differences in the microbiome composition were observed between WT and Zman2 plants, with genotype accounting for 5.8% of the variation for well-watered samples alone (p < 0.05). Zman2 plants harbor a more diverse microbiome as determined by a greater alpha diversity compared to the WT sibling (Fig. 3). Six OTUs were more abundant in WT plants, whereas none were enriched in Zman2 samples ( Fig. 7; OTU abundances by sample type plotted in Supplemental Fig. 3). Of the six OTUs, all were assigned to Alphaproteobacteria belonging to the order Sphingomonadales, three of which were assigned to the genus Sphingobium, whereas the remaining OTUs were unclassi ed at the genus level.
In contrast to well-watered conditions, there is much less variation in the rhizosphere microbiome of WT and Zman2 under drought conditions. Distance-based approaches analyzing beta diversity, analyzing variance between samples, showed them to be indistinguishable. In addition, no differences in the alpha diversity (variance within a sample) were observed between the two genotypes under drought conditions. However, a handful of individual OTUs are signi cantly differentially abundant between the two genotypes when determining individual OTUs that were signi cantly different: six were enriched in Zman2 (including Sphingomonadales and Enterobacteriales) and none were signi cantly enriched in WT ( Fig. 7; Supplemental Fig. 3).

Genotype By Environment Interaction Suggested By Microbial Communities
A genotype by environment (GxE) interaction is implied by the different responses of the Zman2 and WT genotypes to drought stress. This is re ected by the interaction of genotype:water status, which accounts for 5.3% of the variation in beta diversity in all rhizosphere samples (p < 0.05) and by the principal component analysis, in which Zman2 and WT are separate under well-watered conditions, but converge to have the same beta diversity under drought conditions (Fig. 4C). Analysis of the GxE interaction using log ratio tests of full and reduced models in DESeq2 determined that one Chitinophagaceae and one Sphingomonadales OTU were signi cantly impacted by this interaction (OTU abundances plotted by sample type in Supplemental Fig. 4).
Since the microbiome compositions between Zman2 and WT genotypes differed much more under wellwatered but not drought conditions, we analyzed both water statuses separately for each genotype to look at drought effects on each genotype (DxG). This analysis revealed that drought had a larger effect on WT, in which 70 OTU were enriched and seven were depleted under well-watered conditions as compared to drought stress, and drought-exposed plants had a greater alpha diversity. In comparison, for Zman2, only 12 OTU were enriched and six depleted in well-watered conditions, and drought treatment did not impact the alpha diversity.
Dolabralexins, but not other specialized metabolites, are more abundant in wild type plants To verify that differences in microbiome composition can be attributed to a de ciency in diterpenoids in the roots of Zman2 plants, metabolite pro ling using both targeted and untargeted LC-MS/MS analysis was performed on the same root samples used for microbial composition analysis. Targeted metabolite analysis of the major dolabralexin metabolite, trihydroxydolabrene (THD) con rmed via a standard that THD was near absent in the Zman2 mutant while present in WT under well-watered conditions (Fig. 8).
Epoxydolabranol was not found in either mutant nor WT plants (Supplemental Fig. 5), while epoxydolabradiene was found to be lowly abundant in both mutant and WT plants (Fig. 8), presumably because of their conversion to THD. Dolabralexins were not signi cantly enriched in WT versus Zman2 under drought conditions, due to notable variation between individual WT plants, but were still markedly trace in Zman2 drought plants (Fig. 8). Mirror plots demonstrate these identi cations, as well as their absence in the mutant plants ( Supplemental Fig. 5). This observation is consistent with previous research, showing low levels of dolabralexin and kauralexin metabolites in Zman2, predictably due to ent-CDP derived from ZmAn1 activity [16]. As a control, benzoxazinoid abundance was calculated using BOA as a standard, and found to be present in both Zman2 and WT roots without signi cant differences in abundance between well-watered and drought-stressed plants. Using BOA for LC-MS/MS generates multiple peaks with similar mass spectra due to various benzoxazinoids compounds, and their total area was analyzed here (Supplemental Fig. 5).
Parallel untargeted metabolomics analysis also did not indicate any signi cant variance in the metabolite pro les of the WT and Zman2 plants beyond the focal diterpenoids ( Fig. 9A and B). Two observed data points of signi cant difference represented the same plants in both positive and negative mode ( Fig. 9A and B). PERMANOVA analysis based on all dominant mass ions and corresponding speci c retention times demonstrated neither genotype nor water status signi cantly impacted the metabolome using either positive or negative ionization modes. This result was further corroborated by a PCoA (Fig. 9A and   B). Although the PERMANOVA and PCoA demonstrated no signi cant difference overall between sample types, a generalized linear model was used to identify individual metabolites that may be signi cantly enriched or depleted. A total of 111 metabolites were enriched in WT as compared to 85 enriched in Zman2 under well-watered conditions using positive ionization mode (Fig. 9C, Supplemental Table 1). Consistent with the targeted metabolite pro ling, THD was among the 111 enriched metabolites in WT roots. Annotation of the remaining metabolites with distinct pro les in WT and Zman2 roots by comparison to mass spectral databases identi ed signi cant matches for 53 compounds, representing 27% of the differentially abundant metabolites (Supplemental Table 1). Further metabolite annotation using the PACTOLUS method generated signi cant matches for an additional 113 compounds (Supplemental Table 1). Although a few metabolites were enriched or depleted in abundance between WT and Zman2 plants (Fig. 9C, Supplemental Table 1), no other alterations in the metabolite pro le were identi ed as signi cant, and the distinct abundance of dolabralexins was the dominant change in the overall metabolome composition of Zman2 and WT roots. The analysis was repeated for differences between the genotypes under drought conditions (in which the microbiomes are more similar), and less metabolites were signi cantly different (Fig. 9C, Supplemental Table 2). Similar to genotype, drought stress did not signi cantly affect the metabolite pro les of the WT or Zman2 roots ( Fig. 9A and B), as determined by PERMANOVA and PCoA. While the metabolite compositions variance between genotypes was not greater than the variance within the genotypes, there was a handful of metabolites that were determined to be signi cantly enriched or depleted, as determined by linear models. There was a signi cant enrichment of 99 metabolites in WT plants and 283 metabolites in Zman2 plants under drought stress as compared to well-watered plants (Fig. 9C, Supplemental Tables 3 and 4). By comparison, 36 and 88 metabolites were enriched under well-watered conditions in WT and Zman2 roots, respectively, as compared to drought (Fig. 9C, Supplemental Tables 3 and 4). While there are limited numbers of metabolites that change their abundance between drought and well-watered conditions, the overall metabolomes and variance between watering statuses remains statistically indistinguishable by PERMANOVA and PCoA.

Discussion
The dynamic interrelations between plants and their species-speci c root microbiota directly in uence plant health and stress tolerance [4]. Despite the importance of these mutualistic relationships, the underlying chemical mechanisms coordinating inter-organismal interactions remain largely elusive. In particular, limited knowledge exists on how speci c metabolites, blends thereof, and the corresponding pathways impact plant-microbe interactions and microbial community assembly. For example, recent maize studies illustrated that mutant genotypes de cient in benzoxazinoid metabolites (speci cally MBOA) showed an altered stress response mediated by the in uence of MBOA on the below-ground microbial community, thus underscoring the importance of these metabolite-guided plant-microbe interactions on plant health [21][22][23]. In this study, we show that speci c groups of bioactive diterpenoids in maize, namely dolabralexins and/or kauralexins, also contribute to the assembly of the rhizosphere microbiome.
Although the underlying secretion mechanisms require further study, presence of dolabralexins in maize root exudates supports a possible role of these compounds in below-ground plant-microbe interactions (Fig. 2). Microbiome analysis of the root microbial communities showed no signi cant in uence of genotype or water status on endosphere communities using distance-based methods (Fig. 4). This observation differs from prior maize studies, which showed variation in the endosphere communities under well-watered or drought conditions [33]. Considering the diversity and dynamics of plant root microbiomes and metabolite blends, these contrasting ndings may be attributed to differences in plant age, soil, and extent of droughting [36]. It appears plausible that diterpenoids do not impact endophytic microbes due to the spatial separation of endophytic microbes that predominantly colonize the apoplast [37,38], whereas functionalized diterpenoids accumulate intracellularly as demonstrated in several plant species [39][40][41]. Substantiating previous research on root microbiomes in maize, the maize rhizosphere microbiome was signi cantly different between well-watered and drought stress conditions, regardless of genotype (DxG) (Fig. 4) [33,34]. Phyla known to be in uential in the plant response to drought, including Actinobacteria, were found enriched under drought conditions (Fig. 6) [33,35]. Our results showing distinct rhizosphere microbial communities of Zman2 and its WT sibling under well-watered conditions with a more diverse root microbiome alpha diversity in Zman2 (Fig. 3), provide evidence supporting a role of diterpenoids in the microbiome assembly by reducing the community diversity. This is further supported by the distinct beta diversity between Zman2 and WT under well-watered conditions (Fig. 4), with genotype accounting for 11% of the variation between samples. Notably, the signi cant differences between the two genotypes were de ned by only a few OTUs, most of which were assigned to the order Sphingomonadales (Fig. 7). Sphingomonads have been reported to degrade phenolic compounds and utilize them as carbon sources [42], and were among the OTUs displaying the greatest heritable variation (H 2 ) across maize lines of the NAM (Nested Association Mapping) diversity panel [14]. Considering the variation of dolabralexins across selected maize inbred lines [18], it can be speculated that not only phenolics, but also maize-speci c diterpenoids mediate the interaction with species of Sphingomonadales.
The overall increased microbiome diversity in the Zman2 mutant stands in contrast to previous research supporting that a greater microbial diversity promotes crop resistance to soil pathogens (Fig. 3) [43,44], given that previous work has shown Zman2 to be more susceptible to fungal pathogens [29]. Considering these ndings in association with the demonstrated anti-microbial activity of both kauralexins and dolabralexins [17,18], the relationship between the disease-preventative properties of dolabralexins and bacterial diversity remains more complex, and microbe-microbe interactions likely play an important role determining the overall microbial diversity phenotype.
Despite distinct alpha and beta diversities under well-watered conditions, the Zman2 and WT communities were not distinguishable in their diversity or composition under drought stress using distance-based approaches (Fig. 3). While the drought-stressed communities of WT and Zman2 were indistinguishable, their response to drought in terms of the changes in enrichment and depletion associated with differences with their respective well-watered samples was different. This is substantiated by the signi cant factor of the interaction between genotype and water status, a GxE interaction. While WT experienced signi cant changes in relative abundance of OTUs between wellwatered and drought conditions, Zman2 had fewer OTUs altered and no signi cant change in alpha diversity (Fig. 3), supporting the observation that Zman2 plants have a lower capacity for environmental adaptation [16,45]. The protective properties of kauralexins and dolabralexins have been previously described, in which Zman2 is more susceptible to drought stress [16]. This protective mechanism may be hypothesized to be a result of interactions with rhizosphere microbes, in which these diterpenoids contribute to a speci c community being selected for in the WT plant under well-watered conditions. This, in turn, may subsequently provide enhanced resilience under environmental stress conditions. However, under continued drought stress, other factors than diterpenoid bioactivity seemingly become pertinent in the assembly of the rhizosphere microbiome, leading to a convergence of the WT and Zman2 communities to become indistinguishable (Fig. 4). This is consistent with the fact that the levels of dolabralexins in WT plants remained largely unaltered in response to drought stress (Fig. 8). By contrast, earlier research showed drought-induced accumulation of kauralexins [16]. This discrepancy in inducible diterpenoid formation may result from differences in the selected maize genotype or plant cultivation and drought treatment conditions, considering the large variation of diterpenoid metabolism across maize genotypes and environmental stimuli [17,18].
Recent research into the effect of benzoxazinoids provides insight into the magnitude of these changes and their importance to plant health. For example, decreased MBOA abundance in a de ned benzoxazinoid maize mutant affected the plant stress response via the in uence of MBOA on the below-ground microbial community [21]. Unlike maize diterpenoids, which signi cantly changed alpha and beta diversity in the present study (Fig. 3), benzoxazinoids did not affect either the diversity or composition of microbial communities using distance-based approaches, with the exception of MBOA shown to affect select OTUs with a resulting impact on plant tness [21,46]. However, recent studies investigating the same genes targeted for mutation (Zmbx1, bx2, and bx6) did report, albeit minor, differences in the rhizosphere alpha and beta diversities by distance-based methods [22]. These contrasting results highlight a likely impact of plant age, soil type, environmental stimuli, and/or sampling on metabolitemicrobiome interactions. Moreover, the difference in the impact of benzoxazinoids and diterpenoids on the rhizosphere diversity and community composition point toward distinct functionalities of different metabolite classes in maize-microbiome interactions. Recent analysis of two rice mutants, Oscps2 and Oscps4, de cient in species-speci c diterpenoids showed indistinguishable rhizosphere microbiomes and found to be from their WT siblings, suggesting that maize diterpenoids are distinct in their ability to modify the rhizosphere microbiome [47].
The lack of changes in the metabolite pro les of WT and Zman2 roots beyond the expected dolabralexin de ciency in Zman2 are consistent with WT phenotype of Zman2 mutant plants under healthy conditions ( Fig. 9) [16], and underscores that the observed microbiome alterations are a result of dolabralexin and/or kauralexin bioactivity rather than other metabolic perturbations caused by the ZmAN2 loss of function.
Notably, a recent study investigating the root metabolome and microbiome in maize mutants de cient in the biosynthesis of selected benzoxazinoid compounds found the opposite effect, where mutant plants displayed signi cant metabolic changes across many pathways, which were predicted to have global effects on the microbiome, rather than changes in a speci c metabolite group alone [23]. These works underscore the importance of metabolite analysis to understand the broader metabolic implications of pathway mutations especially within complex, branching metabolite networks.

Conclusions
Using a de ned pathway mutant, the present study supports a role of species-speci c bioactive maize diterpenoids in shaping the rhizosphere microbiome diversity and composition. These ndings expand our insight into diterpenoid functions in Poaceous crops beyond well-established anti-microbial and antifeedant bioactivities. Such deeper knowledge of the mechanisms underlying natural plant-microbe interactions will be critical for ultimately enabling broader agricultural applications.

Root exudate analysis
Seeds of Zman2 and its isogenic wild type sibling (both in the W22 background) were obtained from Dr. Eric Schmelz (UC San Diego). Seeds (n = 5 for each genotype) were planted and grown in potting soil in 1 gallon pots in a greenhouse and watered with nutrient water, containing calcium nitrate (0.6 g/L water), Growmore (0.3 g/L water), and magnesium sulfate (0.3 g/L water). After 38 days, plants were removed from pots and the roots were gently washed so as not to cause tissue damage. The plants were then placed in 2.8 L Erlenmeyer asks and suspended with tape such that only the roots were in the nutrient water. Flasks were wrapped in aluminum foil to prevent light stress to the roots and placed in a growth chamber (as detailed below). After 48 hours, plants were removed, and the nutrient water was ltered through a metal strainer to remove any possible tissue debris. Metabolites were extracted from the exudate water by adding 700 mL of ethyl acetate to 700 mL exudate water and leaving at 4˚C for 24 hours. The organic solvent layer was then separated and concentrated using a rotary evaporator for metabolite analysis. Nutrient water containing no plants was used as a control.

Plant growth conditions
Soil was collected at UC Davis a research eld site (coordinates 38.531152, -121.783182) by collecting approximately 6 inches of top soil using bleached shovels and collecting the soil in sterile bags. The eld site had grown maize for one year, and at the time of collection (November 19, 2016) was fallow and had the stover previously turned under after the summer harvest. The soil was then mixed in the sterile bags and distributed to 2.37 L pots that were sterilized using 3% bleach wash. Plants were grown in a growth chamber in the pots in order to control for all other environmental conditions. The growth chamber was set to a 16/8 hour day/night cycle, with a 26/22˚C day/night temperature cycle. Seeds of Zman2 and WT plants were sterilized in 3% (v/v) bleach for one hour, then washed ve times with deionized water, and planted approximately 3 cm deep in the pots with maize eld soil. Pots were distributed in the growth chamber in a block design to mitigate location effects. Treatment (drought vs well-watered) of Zman2 and WT plants was measured for six biological replicates each using bulk soil (no plants) as a control.
Pots were watered every other day with 175 mL of nutrient water (see contents in root secretion assay methods). After 35 days, half of the plants (or bulk soil) were subject to drought treatment. Here, plants were not watered for six consecutive days in which the well-watered plants continued the previous watering and drought plants were not watered. After 6 days of drought treatment, the drought plants were given 140 mL of nutrient water to ensure that the plants received su cient drought that the microbiome could respond to the treatment, but that the plants did not die. On the tenth day of drought treatment, all samples were harvested.

Microbiome sample collection
Sample collection and processing was adapted from Edwards, et al. 2015 [7]. In brief, plants were carefully removed from the soil, and gently shaken until ~ 2 mm of soil adhering to the root remained. n = 5 (bulk soil well-watered, bulk soil drought, rhizosphere drought Zman2, rhizosphere drought WT) or n = 6 (endosphere well-watered WT, drought WT, well-watered Zman2, drought Zman2; rhizosphere well-watered WT, well-watered Zman2); one plant from the drought Zman2 and drought WT sample types was removed from the rhizosphere sample groups due to inadequate PCR ampli cation (see details below). The roots were then transferred to a 50 mL falcon tube contained sterile phosphate buffer saline (PBS) and placed on ice. For analysis of the rhizosphere microbiome, roots were shaken using sterile forceps to remove the soil from the root surface and soil samples were stored at 4˚C until further processing the next day. For analysis of the endosphere microbiome, the above root samples were placed into fresh PBS buffer in a new 50 mL falcon tube and sonicated three 3 times for 10 s each, followed by placing the roots in fresh PBS buffer again to remove any rhizoplane microbes. Using these roots, ~ 4 cm sections of the primary root (beginning 2 cm below the root-shoot junction) was cut, placed in a new tube, frozen in liquid N 2 , and stored at -80˚C until further processing. Bulk soil samples from soil 2 cm below the surface were collected using a sterile scoop and stored in PBS buffer at 4˚C until sample processing the next day.
After collecting all root and soil samples, remaining soil from each pot was collected in a 50 mL falcon tube and weighed. After drying the soil samples for one week at 65˚C, the samples were weighed again to determine the volumetric water content (VWC) of each sample.

DNA Extraction
All DNA was extracted using the MoBio PowerSoil DNA extraction kit and eluted in 50 µL of DEPC-treated water. The rhizosphere samples were concentrated by pipetting 1 mL of the rhizosphere soil in PBS into a 2 mL tube and centrifuged for 30 s at 10,000 x g. The supernatant was discarded and the soil was used for DNA extraction. The endosphere samples were homogenized and ground in liquid N 2 for DNA extraction with the MoBio Powersoil DNA kit.
16S rRNA gene ampli cation, quantitation, and sequencing The V4 region (515 to 806 bp of the 16S rRNA gene) of the 16S rRNA gene was ampli ed according to Edwards et al. 2018 [48]. In brief, PCR was performed using Qiagen HotStart HiFidelity polymerase with the following parameters for each mix: 6.25 µL water, 2.5 µL buffer, 1.25 µL of 10 µM forward primer, 1.25 µL of 10 µM reverse primer, 0.25 µL HotStart polymerase, and 1 µL of DNA. Speci c primer pairs, containing unique 12 bp barcode adaptors on each end of the forward and reverse primers were used for each reaction. Samples without DNA were used as negative controls. A touchdown PCR program was used with the following parameters: 95 ˚C for 5 min; 7 cycles of 95 ˚C for 45 sec, 65 ˚C for 1 min decreasing at 2 ˚C per cycle, and 72 ˚C for 90 sec; 30 cycles of 95 ˚C for 45 sec, 50 ˚C for 30 sec, 72 ˚C for 90 sec; a nal extension of 72 ˚C for 5 min; and samples were held at 4 ˚C. Only samples producing single amplicon bands as veri ed by agarose gel electrophoresis were considered for further analysis, resulting in the loss of one biological replicate from each of the wild type drought rhizosphere sample group and one from the ZmAn2 drought rhizosphere sample group.
Amplicons were puri ed to remove primers using AmPure XP beads (Beckman Coulter). Here, beads were added to each PCR reaction, incubated at room temperature for 5 min, and placed on a magnet for 2 min to separate the beads. After removal of the supernatant, the beads were washed with 70% ethanol twice. The ethanol was then allowed to evaporate and the beads were resuspended in 50 µL water, mixed well, and placed again on a magnet to remove the supernatant containing the desired PCR products. DNA concentrations were measured using a Qubit, and pooled to reach samples of equimolar concentrations.
The pooled samples were cleaned as described above, separated by agarose gel electrophoresis and the 400 bp amplicons were extracted using a NucleoSpin Gel and PCR Clean-up kit (Macherey-Nagel).
Libraries were made and sequencing was performed at the UC Davis Genome Center using 250 × 250 paired end, dual index Illumina MiSeq sequencing.

Sequence analysis
Sequences were analyzed as previously described by Edwards et al. 2014 [7]. In brief, sequences were demultiplexed based on individual barcodes using a custom R script, and assembled into single sequences using PandasEq. Sequences were then clustered into OTUs with the NINJA-OPS pipeline using 97% pairwise sequence identity referenced against the Greengenes 16S rRNA sequence database (version 13_8) [32].
In total, 1,802,959 high-quality sequences were obtained with a median read count of 31,204 per sample, and a range of 1,699 − 87,795 (All data is available in Sequence Read Archive, Sequence Read Archive repository, BioProject ID PRJNA600272 [https://www.ncbi.nlm.nih.gov/sra/PRJNA600272]). Using the QIIME pipeline, reads were clustered based on 97% sequence identity into operational taxonomic units (OTUs) and were annotated using QIIME and the Greengenes database [32], resulting in 7,181 microbial OTUs. Chloroplast and mitochondrial OTUs represented 65 OTUs and were removed, along with lowabundance OTUs (less than 5% of the total sample), leaving 4,258 total OTUs. OTU counts were then normalized by relative abundance, which was used rather than rarefaction methods so as not to discard low abundance OTUs.
All statistical analysis of the OTU table generated by QIIME [32] were analyzed using custom R scripts (version 3.6.1) [49]. Alpha-diversity was measured using the "Shannon" method in the R package vegan [50]. Principle coordinate analysis (PCoA) were conducted using unconstrained principles and Bray distances in the R package vegan [50]. PERMANOVA (permutational multivariate analyses of variance) analysis was performed using the R package vegan function adonis to measure beta-diversity [50]. The DESeq2 package [51] was used to analyze differentially abundant phyla, that were collapsed OTU counts, and individual OTUs, along with the biobroom package [52]. Plots were visualized using the ggplot2 in the tidyverse package [53]. All scripts generated in this study have been deposited to GitHub (https://github.com/kmurphy61/maizemicrobiome).

Metabolite extraction
The remaining roots (~ 1 g fresh weight) used for endosphere microbiome analysis (see microbiome sample collection) were homogenized and ground in liquid nitrogen. Because of availability of tissue, the number of plant samples were reduced for metabolite extraction as compared to microbiome DNA extraction; n = 4 (WT well-watered, Zman2 well-watered, Zman2 drought) or n = 3 (WT drought). Samples were then placed in a 2 mL glass vial and metabolites extracted by incubation in 2 mL methanol overnight at 4°C with gentle rocking. Samples were centrifuged for 10 min at 4000 x g and the methanol phase transferred to a new vial using a glass pipette, air-dried, and resuspended in either 100 or 200 µL methanol.

Metabolite analysis
For metabolite analysis by liquid chromatography tandem mass spectrometry (LC-MS/MS), samples were spiked with 4 µM internal standard mixture of deuterium-labeled lipids (Cat# 110899, 857463P, 861809O, 110922, 110922, 110921, 110918, 110579, 110544, Avanti Polar Lipids, Inc) and 1 µg/mL ABMBA (2-Amino-3-bromo-5-methylbenzoic acid, Sigma). UHPLC reverse phase chromatography was performed using an Agilent 1290 LC coupled with a QExactive Orbitrap MS (QE = 139) (Thermo Scienti c, San Jose, CA). Chromatography was performed using a C18 column (Agilent ZORBAX Eclipse Plus C18, Rapid Resolution HD, 2.1 × 50 mm, 1.8 µm) at a ow rate of 0.4 mL/min and injection volume varied from 0.9 to 3.5 µL to normalize against sample dry weight. Samples were run on the C18 column at 60 ºC equilibrated with 100% buffer A (100% LC-MS water w/ 0.1% formic acid) for 1 min, following by a linear dilution of buffer A down to 0% with buffer B (100% acetonitrile w/ 0.1% formic acid) over 7 min, and followed by isocratic elution in 100% buffer B for 1.5 min. Full MS spectra were collected ranging from m/z 80 − 2,000 at 60,000 to 70,000 resolution in both positive and negative mode, with MS/MS fragmentation data acquisition using an average of stepped 10-20-40 and 20-50-60 eV collision energies at 17,500 resolution. For targeted analysis, product identi cation by comparison to standards was performed where authentic standards were available.
For untargeted analysis, exact mass and retention time coupled with MS/MS fragmentation spectra were used to identify compounds. Features -high intensity signals narrowly contained at a given retention time and m/z -were detected using the MZMine software v 2.24 (http://dx.doi.org/10.1093/bioinformatics/btk039). Further custom Python scripts were used to identify minimum and maximum retention time bounds for each peak, lter peaks that were not at least 3 times higher in a sample compared to all injection blanks, and lter peaks that did not have at least one MS/MS spectrum. Complete feature lists are available on GitHub (https://github.com/kmurphy61/maizemicrobiome.git). Features that showed a signi cantly different abundance (peak area) using linear models, calculated using custom R scripts [49]. Signi cantly different features were annotated using custom Python scripts. A wide range of chemical standards previously added to a mass spectral library were used to annotate metabolite identi cations based on matching m/z better than 10 parts per million (ppm), retention time better than 0.1 minutes, and matching fragmentation patterns better than a score of 0.7. The remaining compounds that met these criteria were added to a Metabolite Atlas (http://dx.doi.org/10.3390/metabo5030431) for positive and negative ionization modes. All scripts are available on GitHub (https://github.com/kmurphy61/maizemicrobiome.git). Additionally, enriched features were analyzed using PACTOLUS (https://github.com/biorack/pactolus). Signi cantly different features by sample type are available in Supplemental Tables 1 (well-watered), 2 (drought), 3 (WT), and 4 (Zman2).

Figure 3
Alpha diversity of each sample type, as measured by the Shannon's H index. Letters represent signi cantly different measurements determined using PERMANOVA, p ≤ 0.05. n= 5 (bulk soil wellwatered, bulk soil drought, rhizosphere drought Zman2, rhizosphere drought WT) or n = 6 (endosphere well-watered WT, drought WT, well-watered Zman2, drought Zman2; rhizosphere well-watered WT, wellwatered Zman2).  Distribution of the ten most abundant phyla for each sample type. Bars represent the relative abundance of all OTUs within each of the top 10 most abundant phyla. n= 5 (bulk soil well watered, bulk soil drought, rhizosphere drought Zman2, rhizosphere drought WT) or n = 6 (endosphere well-watered WT, drought WT, well-watered Zman2, drought Zman2; rhizosphere well-watered WT, well-watered Zman2). Signi cantly different OTUs detected using linear methods and their respective Phyla in the (A) rhizosphere, (B) bulk soil control, or (C) both the rhizosphere and bulk soil control, irrespective of genotype. n = 5 (bulk soil well-watered, bulk soil drought) or n = 11 (rhizosphere well-watered, rhizosphere drought). Differentially expressed OTUs in the rhizosphere, as measured by a linear model for genotype in (A) wellwatered or (B) drought conditions. Overlap is non-signi cantly different OTUs, at the level of p ≤ 0.05. n = 6 (well-watered) or n = 5 (drought) per genotype. one-tailed t-tests were used for the remaining metabolites, since they are predicted to be enriched in WT and de cient in Zman2. n = 4 (WT well-watered, Zman2 well-watered, Zman2 drought) or n = 3 (WT drought). Error bars represent standard error.