The dynamic nature of pathogen invasion of plant tissues and the onset of changes in plant metabolic profiles require a time series approach to critically analyze inter-omics relationships [31–33]. By following FOM progression over time, we tracked the orchestration of metabolites and root-associated microbiome assembly in two genotypes of Arabidopsis with different FOM susceptibilities.
Microbiome profiling revealed that Col-0 and Ler-0 exhibited distinct microbial compositions in the different FOM treatments across DAIs. The loss in bOTU richness in most inoculated samples could be a direct effect of FOM niche displacement [34]. The significant effects of DAI, genotype, and FOM on bacterial and fungal beta diversities confirm previous studies [35–37]. In both genotypes, the effect of FOM inoculation on microbial communities was strongest at the early sampling times, which could be attributed to high FOM densities. Genotype effects on microbial communities increased with DAI, as also genotype x FOM treatment effects. Although the relative importance of genotype on microbial communities has been observed to decline over time [38], our results showed the opposite, mostly in bacterial communities. These genotype response trajectories on the microbiome could possibly be a cause of the differences in disease resistances, as predicted.
Indicator species represent microbes that are mostly affected by treatments and could therefore be important in host responses to FOM [39, 40]. The identified indicator OTUs and their dynamics at different DAIs revealed (i) a large diversity of the Arabidopsis root-associated microbiota that was distinctively affected by the physiological status of the host, and (ii) a time-dependent assembly of microbial communities as previously reported [35, 36, 41]. For example, Edward et al., (2018)[35] found a highly dynamic host-associated microbiota in field grown rice, which was affected by host developmental stage and plant age. Thirdly, our results demonstrated a genotype-specific assembly of microbiota [42]. Notably, the enrichment of indicator bOTU10 assigned to Pseudomonas protegens in inoculated samples of both genotypes suggest that Arabidopsis recruit microorganisms to enhance FOM resistance. P. protegens is a well-known plant growth-promoting bacterium with broad-spectrum antifungal activity against phytopathogens [43]. The genus Streptomyces was strongly enriched in Col-0. Streptomyces is known for its unparalleled synthesis of antibiotics [44], and could thus contribute to FOM resistance [44, 45]. Moreover, the enrichment of more diverse indicator taxa in Col-0 at 25 DAI further supports distinct genotype effects.
We assessed the overall robustness of Col-0 and Ler-0 microbial networks to FOM invasion. A combination of network variables is best for explaining microbial community resilience and robustness [15]. We found node degree, hub numbers, and link densities to be highest in Col-0, indicating robust microbial communities. In contrast, connectance was higher in Ler-0, a result that corroborates earlier studies where high connectance promoted Ralstonia solanacearum host colonization [23]. In our study, all the measured properties were highest in the networks of non-inoculated Col-0 and Ler-0, suggesting a network breakdown during FOM invasion. Node degree breakdown during FOM infection was higher (30.7%) in the susceptible Ler-0 compared to Col-0 (16.8%). Altogether, these results support our initial hypothesis of a stronger network resistance in Col-0 and that stronger networks are indicative of pathogen-resilient microbial communities.
We found both positive and negative intra- and inter-kingdom co-occurrences, except that we did not identify any negative fungi-fungi co-occurrences, plausibly due to weaker antagonistic interactions among fungi. These results support previous studies of cooperative and antagonistic microbial interactions among microbial kingdoms [16, 22]. The differences in co-occurrence network structures underscore the distinct interactions in inoculated and non-inoculated samples of Col-0 and Ler-0 and are also essential in explaining aspects of host invasion resistance [15, 22]. The microbial networks further highlighted how indicator species are affected during FOM invasion. Surprisingly, keystone species and indicator species were mostly observed in the networks of the inoculated Col-0, and we speculate that enrichment of these species could serve as an important factor in resilience towards pathogen invasion [46]. Taken together, these results largely support the hypothesis of distinct microbiome structuring in resistant and susceptible genotypes across different DAIs.
Surprisingly, we only found negative co-occurrences between fOTU1 and bOTUs, and these were observed mostly in non-inoculated Ler-0. Among these bOTUs several belonged to Proteobacteria and Actinobacteria, which are dominant taxa of the Arabidopsis microbiome [47]. In contrast to Col-0 (having the RFO gene), Ler-0 could possibly be deploying other mechanisms to resist pathogens by recruiting antagonistic microorganisms. This putative defense mechanism, however, was broken down by the abrupt invasion caused by FOM inoculation. Similarly, Snelders et al., [48] demonstrated the ability of the fungal pathogen Verticillium dahliae to manipulate tomato and cotton microbiomes by suppressing antagonistic bacteria. The ability of FOM to cause this disruption could be explained by disturbance via resource competition and secretion of antifungal compounds [16, 48].
Plant chemical compounds are strong modulators of microbial communities, and variations in their diversity and quantities can be used to predict resistance profiles of plants towards specific pathogens [49]. We observed a clear separation of metabolites (in OPLS-DA) at individual DAIs and these coincided with the marked assembly of host-associated microbiota, confirming that metabolites have a regulatory role in shaping the microbiome [5, 8]. Several metabolites were found in higher amounts in Ler-0, mostly at 5 and10 DAI, that were induced in response to FOM infection. The observed differences in iGLS, aGLS, camalexin, and phenylpropanoid concentrations in Col-0 and Ler-0 could contribute to the differential resistances observed in the two genotypes. For example, aGLSs were induced at higher levels in inoculated Col-0 compared to Ler-0 at 5 DAI, which could potentially affect microbial communities [7, 42]. aGLSs and their hydrolysis products have been reported to have higher effects on microorganisms compared to iGLSs [50]. Thus, we further predict that the high aGLS content could inhibit FOM in Col-0. We did not find any strong evidence of inhibitory effects on FOM of camalexin, which is in line with other studies [51, 52].
Changes in concentrations of a number of phenolic compounds, including cinnamic acid, coumarins, lignin precursors and lignans notably in the inoculated samples have been observed earlier [53, 54]. The elevated concentrations of a number of these phenols (sinapyl aldehyde, syringin, sinapyl alcohol, and coniferyl aldehyde) in infected samples support a previous study where soluble derivatives of the phenylpropanoid pathway increased defense in an Arabidopsis-Verticillium longisporum interaction [53].
The positive and negative correlations between metabolites and OTUs, including a number of indicator and keystone members, support the selective effects of metabolites on specific microbial members of root microbiomes [6–8, 33]. We further identified metabolite-OTU correlations that uniquely occurred in either Col-0 or Ler-0, underpinning the different resistances and recruitment of specific microbial taxa [55–58]. For instance, the iGLSs glucobrassicin, neoglucobarassicin, and 4-hydroglucobrassicin correlated with specific bOTUs in Col-0, while the aGLSs glucoiberin and sulforaphane correlated with bOTUs in only Ler-0. Genotypic variation in chemical diversity affected fungal and bacterial communities associated with plants [59]. Moreover, the observation that aGLS mostly correlated with bOTUs and iGLS mostly correlated with fOTUs is indicative of differential effects of GLSs on microbial communities. Similarly, phenolic compounds displayed genotype-specific relations as they were correlating with bOTUs only in Ler-0, supporting previous studies in which phenolic compounds were found to selectively inhibit F. oxysporum and Verticillium dahliae [33].
We found 4-methoxyglucobrassicin, indole-3-carbinol, and syringin had the strongest overall effects on bacterial communities, while neoglucobrassicin, camalexin, and coumaric acid highly affected fungal communities. Our findings consolidate these compounds as highly bioactive that could be of interest in future studies. Moreover, the results suggest a dominant role of iGLSs on the Arabidopsis root microbiota. GLS have been widely studied due to their well-known role in plant-microbe interactions, and Zeng et al., (2003)[60] reported 4-methoxyglucobrassicin as a growth stimulator of ectomycorrhizal fungi, while indole-3-carbinol is known for its broad antimicrobial activity against bacteria and yeasts [61]. The iGLSs 4-methoxyglucobrassicin and indole-3-carbinol displayed strong positive correlations with specific OTUs, while aGLSs, including glucoerucin, glucoraphanine, and sulforaphane, were strongly negatively correlating with indicator taxa such as Gammaproteobacteria and Bacteroidia. These results suggest that aGLSs display higher toxicity to major microbial members compared to iGLSs, as was also found in previous studies where the effect of aGLSs on Alternaria brassicicola was stronger than the effect of iGLSs [62]. The strong inhibition of microorganisms by aGLSs is attributed to degradation products such as isothiocyanates, thiocyanates, oxazolidinethiones, and nitriles produced from enzymatic cleavage by myrosinase [63–65].
Hub microbes are important for maintaining network structure and function [22]. Interestingly, all the identified hub OTUs were also indicator species in our analysis, reaffirming their importance in the microbiome. Most of these indicator/hub OTUs were bOTUs and were mostly found in Ler-0, thus corroborating our hypothesis that inter-omics interactions depend on both host genotype and host infection status. The indicator/hub OTUs were affected negatively or positively by specific metabolites and these were further observed to be physiologically dependent. Specifically, we observed unique metabolite-indicator/hub OTU correlations, mostly positive in non-inoculated Ler-0 and negative in inoculated Ler-0, suggesting that the host utilizes its metabolites to selectively recruit specific microbes under different physiological conditions [66]. The indicator/hubs bOTU259 (Niastella) and bOTU538 (Solirubrobacteraceae) had mostly negative and positive correlations, respectively, with metabolites in Ler-0. While Solirubrobacteraceae has been found to suppress common scab disease of potatoes [67], the Niastella is reported to improve soil health and promote root growth [68, 69]. In non-inoculated Ler-0, the indicator/hub OTUs Paenibacillus (bOTU281), Nocardioides (bOTU12), Xanthobacteraceae (bOTU4664), and Methyloligellaceae (bOTU27) exclusively correlated positively with metabolites. Interestingly, these taxa are considered ecologically important, either involved in nitrogen fixation or acting as antagonists against pathogens. For example, the plant growth-promoting Paenibacillus polymyxa induces host defense responses against F. oxysporum [70, 71]. Altogether, these results support our hypotheses of genotype specific metabolome-microbiome interactions primarily due to the differential FOM resistances. Specifically, we speculate that Ler-0, in the lack of the RFO resistance gene, is recruiting antagonistic microorganisms to prevent infection by synthesizing certain metabolites. However, after inoculation with FOM this pathogen resilient network is broken down.
The observation that cinnamic acid and glucoerucin had the highest numbers of positive and negative correlations with indicator/hub OTUs suggest their bioactivity. Both inhibitory and chemoattracting effects of cinnamic acid has been demonstrated [72, 73]. Glucoerucin is toxic against Xanthomonas campestris, Pseudomonas syringae [74] Pythium irregulare and Rhizoctonia solani [75]. Sulforaphane negatively correlated with Niastella (bOTU259) and Leptothrix (bOTU70) in inoculated Ler-0 and correlated positively with Nocardioides (bOTU12) and Duganella (bOTU61) in non-inoculated Ler-0. Sulforaphane is known for its selective effects towards Bacillus subtilis, Pseudomonas aeruginosa, and Candida albicans [76]. Indole-3-carbinol correlated positively with several indicator/hub OTUs in non-inoculated Ler-0. These findings further support differential effects of GLS and their hydrolysis products [77] on host-associated microbiomes [13, 14]. While sinapyl-alcohol showed both positive and negative correlations with several taxa, its glucoside syringin, as well as pinoresinol and fraxin, only correlated positively with indicator/hub taxa. Importantly, the results of distinctive interactions, for instance, sinapyl-alcohol negatively correlating with the family Chitinophagaceae in inoculated Ler-0 and positively with the family Xanthobacteraceae, are indicative of host-dependent effects of these compound on the microbiota. The potential of other phenylpropanoids, for example coumarins, to differentially inhibit both beneficial and pathogenic microorganisms has been reported [8, 78]. Together, these findings suggest an active role of phenypropanoids in shaping microbiota and should thus be prioritized in future studies.
Although most metabolite-OTU correlations were identified in Ler-0, an important observation in Col-0 was the positive correlation of syringin with the beneficial Streptomyces (bOTU4). 4-methoxyglucobrassicin only correlated with fungal indicator/hub OTUs in Col-0, indicative of a higher effect on fungal communities [60]. In addition, the phytohormone SA and the phenols sinapic acid and sinapyl alcohol showed negative correlations with a high number of indicator OTUs, suggestive of their modulating effect on the microbiota. Contrasting correlations between indicator/hub OTUs and phytohormones were observed, suggesting distinctive effects of these hormones on microbial communities.