An in house GFP transformant of Fusarium graminearum PH-1 (Tan et al., 2020) was grown on potato dextrose agar (PDA, Sigma-Aldrich) for 14 days at 21°C under a regime of 12 h of dark and 12 h of combined UV-C and UV-A light (2× TUV 8W T5 and 1× TL 8W BLB; Philips). Conidia were harvested by adding a sterile solution of phosphate buffered saline (PBS) with 0,01% (v/v) Tween 80 to the PDA plates and scrubbing the mycelium and spores with a sterile Drigalski spatula. Subsequently, the suspension was filtered over two layers of sterile Miracloth to remove the mycelium and separate the conidia. Finally, the conidiospores were counted with a haemocytometer and the spore suspension was diluted in PBS until a concentration of 1 x 10⁶ conidia/mL was reached.

Streptomyces rimosus subsp. rimosus LMG19352 was derived from the Belgian Coordinated Collections of Microorganisms (BCCM) (Ghent, Belgium). The strain was grown from its -80°C glycerol stock on tryptic soy agar (TSA) plates for 3 days at 28°C, after which a single colony was picked with a sterile inoculation loop and was grown in 5 mL tryptic soy broth (TSB) for 3 days at 28°C and 180 rpm. The bacterial suspension was then centrifuged at 4,000 rpm for 5 minutes and washed twice with PBS. Eventually, the optical density at 600 nm (OD₆₀₀) was measured with a spectrophotometer and set to a value of ±0.6.

Spring wheat (Triticum aestivum L.) cultivar Tybalt was grown under greenhouse conditions in 3-liter pots (15 cm diameter x 30 cm height) until anthesis (Zadoks Growth Scale (GS) 65). Twelve seeds were grown per pot in commercial potting soil (DCM Huis & Tuin) and mineral fertilizer was added to ensure optimal development. One spikelet per ear was used for co-inoculation with bacterial and fungal suspensions. Spikelets were inoculated with 10 µL of bacterial suspension through point inoculation 24 h before fungal inoculation (-1 dpi). One day after bacterial inoculation, the same spikelet was inoculated with 10 µL of F. graminearum spore suspension (0 dpi). An infected control, in which sterile PBS was used instead of bacterial suspensions, and a blank control, which was co-inoculated with sterile PBS, were included ( Figure 1 ). After inoculation, plants were put in a custom-built growth chamber at 21°C with a photoperiod of 16 h and kept at raised relative humidity for 3 days by spraying the plants and the walls of the growth chamber with water until run-off twice a day (approximately 100% relative humidity). Samples were taken at 0, 7 and 14 dpi by detaching the ears, visualizing them through multispectral imaging, and subsequently flash freeze them in liquid nitrogen. Samples were stored at -80°C for further analysis. All treatments and controls were included in 15 randomly distributed replicates.

Disease assessment was performed as described by Tan et al. (2020). In short, a custom built multispectral imaging platform was used (CropReporter, Phenovation B.V., Wageningen, The Netherlands) to visualize diverse physiological traits in real time, based on specific absorption, reflection and emission patterns at a resolution of 6 μm. A 6Mp-16 bit camera, including a filter wheel equipped with 12 optical interference filters, allows pixel-to-pixel capturing of RGB values, a diverse set of spectral indices, chlorophyll fluorescence (F_v/F_m), and GFP fluorescence. The effect of disease was assessed via the photosynthetic efficiency in the electron transport chain of photosystem II in a dark adapted state (F_v/F_m) (Baker, 2008). Additionally, GFP tagged F. graminearum PH-1 was visualized and quantified by measuring GFP fluorescence, enabling the assessment of fungal growth in planta. The GFP signal was corrected for autofluorescence of senescing ears, resulting in the corrected GFP value (cGFP).

Based on the results of our multispectral imaging platform, the ten most representative wheat ear samples of each treatment were selected and randomly pooled per two. Per ear, the spikelet located two spikelets above the inoculated spikelet was selected ( Figure 1 ) and was ground with liquid nitrogen. A total DNA extraction was carried out using the DNeasy Plant Mini Kit (QIAGEN, The Netherlands) according to the manufacturer’s instructions with some slight adaptations. After washing the column twice with wash buffer AW2, the spin column was transferred to a new microcentrifuge tube and airdried for 1 min to make sure all the ethanol was evaporated. The volume of the elution buffer was adapted to 50 µL, and columns with elution buffer were incubated for 5 min at room temperature. Afterwards, samples were centrifuged for 1 min at 6,000 g and final steps were repeated once with the eluate. Concentration of extracted DNA was evaluated using the QuantiFluor^® dsDNA System and the Quantus™ Fluorometer from Promega (Madison, Wisconsin, USA). The DNA extracts were finally diluted with nuclease-free water to a concentration of 5 ng/μL, unless the concentration was lower than 10 ng/µL, then no further dilution was performed.

The bacterial V3-V4 region of the 16S rDNA was amplified by an amplicon PCR using the primer pair S-D-Bact-0341-b-S-17/S-D-Bact-0785-a-A-21 (Klindworth et al., 2013) ( Table 1 ), extended with Illumina specific adaptors. The fungal ITS2 region was amplified by an amplicon PCR using the forward primer fITS7bis (adapted from Ihrmark et al. (2012)) and reverse primer ITS4NGSr (adapted from Tedersoo et al. (2014)) ( Table 1 ), both extended with Illumina specific adaptors. Afterwards, an index PCR was performed to attach dual indices and Illumina sequencing adaptors to both the V3-V4 and the ITS2 fragments using the Nextera XT Index Kit (Illumina, San Diego, CA, USA). Identical PCR conditions for both PCRs were used as described by Debode et al. (2016) and PCR mastermixes for the amplicon and index PCR were prepared using KAPA HiFi HotStart ReadyMix (Kapa Biosystems, Wilmington, MA, USA) according to the manufacturer’s instructions to a total amount of, respectively, 25 µL and 50 μL per sample. Each PCR step was followed by a PCR clean-up using CleanNGS beads (CleanNA, Waddinxveen, The Netherlands). The cleaned PCR products were each time controlled for their quality via gel electrophoresis and final concentrations were measured using the QuantiFluor^® dsDNA System and the Quantus™ Fluorometer (Promega, Madison, Wisconsin, USA). Eventually, the barcoded libraries were diluted to 10 nM, equally pooled and sequenced on an Illumina MiSeq system (2 × 300 bp) by Admera (South Plainfield, NJ, USA).

To avoid amplification of mitochondrial and plastid 16S rDNA derived from the host plant during PCR, peptide nucleic acid (PNA) clamps were added to the PCR mastermix used for amplification of V3-V4 16S rDNA (PNABio, California, USA) ( Table 2 ). PNA is a synthetic DNA mimic in which the DNA deoxyribose phosphodiester backbone is replaced with a pseudopeptide backbone (Nielsen and Egholm, 1999). These PNA clamps selectively bind to a targeted region of the plant genome and inhibit amplification of this host DNA during PCR, since the PNA/DNA duplexes are thermally more stable than the DNA/DNA duplexes formed during primer annealing (Lundberg et al., 2013; Ørum et al., 1993).

Demultiplexing of the amplicon dataset was done by the sequencing provider. Cutadapt was used to look for primers and only keep those fragments where both primers were found at the beginning and at the end. Afterwards, primers and adapters were removed. Additionally, fragments were filtered by allowing maximum three errors, and fragments with N’s were excluded. Raw Illumina forward and reverse reads were merged using the program PEAR v.0.9.8 (Zhang et al., 2014). Dereplication, sorting, amplicon sequence variant (ASV) calling, chimera removal, and taxonomic assignment were done making use of the DADA2 algorithm v1.16 (Callahan et al., 2016). Briefly, all ambiguous bases were removed and the number of maximum expected errors for forward and reverse reads was set at three. After dereplication, reads were merged and an ASV table was built. The DADA2 AssignTaxonomy function was used to assign taxonomy to the resulting ASVs, with the SILVA database v138.1 (Quast et al., 2012) serving as a reference for V3-V4 16S rRNA gene sequences and the UNITE database v9.0 for fungal ASVs. The resulting count table was further filtered based on length and taxonomy. A length cut-off for the merged sequences for V3-V4 and ITS2 were set between 400-450 bp and 235-450 bp, respectively. Taxa unassigned to the kingdom of Bacteria, unidentified on phylum level, assigned to the class Chloroplast or family Mitochondria were excluded from the bacterial dataset. Taxa unassigned to the kingdom of Fungi and/or unidentified on class level were excluded from the fungal dataset. The filtered count tables were used for statistical analyses.

The impact of S. rimosus subsp. rimosus LMG19352 on a F. graminearum infection was investigated on wheat ears. After 0, 7 and 14 dpi, FHB symptom development was measured via F_v/F_m and cGFP assessing plant health and active fungal biomass, respectively ( Figure 2 ). At 7 dpi, the fungal pathogen was still in its presymptomatic stage, as no significant differences across treatments were observed, while at 14 dpi clear symptom development was established.

At 14 dpi, the infected spikelet of wheat ears co-inoculated with BCA S. rimosus subsp. rimosus LMG19352 and F. graminearum resulted in a significant decrease in FHB symptom development, hallmarked by significantly higher F_v/F_m values as compared to the infected F. graminearum control ( Figure 2B ; p < 0.001). Additionally, a significant decrease in cGFP value was detected, indicating a lower biomass accumulation of F. graminearum at that timepoint ( Figure 2C ; p < 0.001). These observations both confirmed the biocontrol activity of strain LMG19352 as earlier observed by Tan et al. (2021). Additionally, upon comparison between the co-inoculated ears with the uninfected, LMG19352-treated control at 14 dpi, no significant differences were observed between the cGFP values, suggesting a complete loss of actively growing F. graminearum on the infected spikelet after 14 dpi (p = 0.995). Eventually, no phytotoxic effects were observed in wheat ears inoculated with strain LMG19352 alone.

Assessment of symptom development of the sampled spikelets, located two spikelets above the infection site ( Figure 1 ), resulted in a similar trend as observed for the infected spikelets, but did not result in any significant differences ( Figures 2D, E ). At 14 dpi, the co-inoculated treatment showed a higher variation and a lower median of F_v/F_m values in the sampled spikelets ( Figure 2D ) when compared to the results of the inoculated spikelets ( Figure 2B ). In addition, the opposite effect was observed for the median cGFP value ( Figures 2C, E ). These results both suggest that the biocontrol effect of strain S. rimosus subsp. rimosus LMG19352 diminishes when moving further away from the inoculation site.

The fungal and bacterial community structures on wheat ears inoculated with S. rimosus subsp. rimosus LMG19352 and/or F. graminearum were examined at 0, 7 and 14 dpi. In the original, unfiltered ASV tables, a total of 10,911,934 and 2,506,704 reads were detected and clustered into 6,481 and 6,063 ASVs for the fungal and bacterial dataset, respectively. After filtering the dataset based on ASV length and taxonomy, the fungal and bacterial datasets contained 1,445 and 759 ASVs, respectively. Chao1 richness, Pielou’s evenness and the Shannon diversity index were calculated for both filtered datasets ( Table 3 ). No significant differences in α-diversity measures across treatments were observed in the bacterial dataset. However, at 7 dpi, the mean chao1 richness was higher for treatment LMG19352 compared to the other treatments, due to an extreme value of 217 species counts in one of the replicates, although not significant. In contrast, analysis of the fungal dataset at 7 dpi showed that richness, evenness and diversity were significantly lower in treatment LMG19352 when compared to the blank control (p = 0.034, 0.021 and 0.01, respectively). Both fungal and bacterial differences in α-diversity were restored at 14 dpi. The F. graminearum infected control at 14 dpi showed a significantly lower richness, evenness and diversity in the fungal dataset compared to the blank (p = 0.008, 0.011 and 0.01, respectively). This trend was not yet observed at the presymptomatic timepoint (i.e. 7 dpi), indicating that the loss in diversity is associated with symptom development of F. graminearum. This assumption was supported by a significant positive correlation of the fungal richness, evenness, and diversity with the plant health status (F_v/F_m) in both infected and sampled spikelets. Additionally, fungal richness, evenness and diversity were negatively correlated with F. graminearum growth (cGFP) in both infected and sampled spikelets, however the correlation between fungal richness and cGFP was not significant ( Supplementary Figure 1 ), indicating that lower fungal diversity is associated with increased biomass of the inoculated F. graminearum. Eventually, co-inoculated samples displayed intermediate α-diversity values between the blank and infected control at 14 dpi, indicating a partial recovery of α-diversity towards the original state, similar to the trend observed for FHB symptoms.

The fungal and bacterial community compositions of the spicosphere microbiome were investigated over time (0, 7 and 14 dpi), and across treatments on uninoculated wheat ears (blank), wheat ears inoculated with F. graminearum (Fg control), wheat ears inoculated with S. rimosus subsp. rimosus LMG19352 (LMG19352) and wheat ears co-inoculated with F. graminearum and S. rimosus subsp. rimosus LMG19352 (Fg + LMG19352).

Before filtering, the fungal abundance dataset contained 7 phyla, 33 classes, 82 orders, 201 families and 341 genera, while the bacterial abundance dataset contained 24 phyla, 46 classes, 107 orders, 152 families and 219 genera. The dataset was filtered to only retain the relevant phyla, classes, orders, families or genera with a minimal abundance of 0.05% in at least 5 samples, after which the fungal and bacterial dataset respectively consisted of 3 and 10 phyla, 11 and 15 classes, 25 and 24 orders, 32 and 34 families, 31 and 24 genera and 133 and 26 ASVs.

Cross-kingdom correlation networks were created for each treatment at each sampling timepoint ( Supplementary Figures 4 - 12 ). Multiple topological parameters of the correlation networks ( Table 4 ) revealed that they differ among treatments in complexity and stability.

The complexity of the networks was evaluated based on the number of nodes and edges, together with the mean degree and density. The number of nodes were relatively stable throughout the treatments and throughout time, confirming a stable microbial diversity as revealed earlier ( Table 3 ). However, a decrease in number of nodes was observed during the symptomatic stage of the infected control, which indicates that the microbial diversity decreases throughout time upon FHB symptom development. In the blank, the number of edges increased over time, indicating a more interconnected community over time. The combined inoculation of F. graminearum and LMG19352 showed a similar pattern to the blank, although more connections (i.e. edges) were observed at 14 dpi than in the blank control. Contrastingly, in wheat ears inoculated with Fusarium alone, the number of edges first increased and subsequently decreased over time, indicating a loss of complexity in the microbiome upon Fusarium and FHB symptom proliferation. The same pattern was observed in ears inoculated with LMG19352 alone due to the high number of edges at 7 dpi. Though, a higher number of connections than the blank control was observed at 14 dpi, which was not the case for the infected control samples. To conclude, besides the high number of edges in the correlation networks of treatment LMG19352 at 7 dpi and Fg+LMG19352 at 14 dpi, they also had a higher degree and density, suggesting those two networks are more complex when compared to the rest, while the correlation network of the infected control showed a lower complexity at the symptomatic stage.

Further analysis revealed that the correlation networks at 14 dpi of the infected control, LMG19352 and Fg+LMG19352 had a lower proportion of positive edges compared to the blank control. The lowest value was measured for the infected control, indicating the strongest competition among classes. Additionally, the clustering coefficient and assortativity were higher in the infected control at 14 dpi than in all other treatments while on the other hand, average path length was lowest. These results suggest an efficient network, i.e. a network wherein highly interconnected nodes tend to cluster together with other interconnected nodes and vice versa.

The networks of the blank control at 0 dpi, Fg control at 7 and 14 dpi, and Fg+LMG19352 at 7 dpi all had a modularity index slightly greater than 0.4, implying a modular structure of the network (Newman, 2006). Herein, nodes tend to correlate with other nodes within the same modules, but have sparse connections between nodes in different modules. Over time, the modularity tended to decrease, except for treatment LMG19352. The lowest values were observed for treatment LMG19352 at 7 dpi and Fg+LMG19352 at 14 dpi. This could indicate that interaction patterns in the network are more homogeneous, meaning there are fewer distinct groups of highly connected nodes. Instead, interactions are more evenly distributed across the network. On the other hand, low modularity in a biological network may also be a hallmark of a diseased or dysfunctional state, characterized by reduced robustness to damage or environmental stress. Though the latter hypothesis is considered unlikely, since both networks were evaluated as the most complex and previously shown results indicated a diverse and balanced microbiome for both treatments and timepoints, similar to the blank control ( Figure 3 ).

Centrality measures, degree and betweenness, were used in quantitative representation of keystone species. Degree, i.e. the number of direct correlations to a node in the network, and betweenness centrality, i.e. the importance of a class as connector in the network, differ throughout sampling timepoints and across treatments ( Figure 5 ). High centrality values indicate that a microbial taxon plays a more crucial role in the network structure of a certain treatment at a given timepoint.

Some bacterial and fungal taxa exhibited notable changes in degree centrality over time and across treatments ( Figure 5A ). For instance, Planctomycetes and Chloroflexia appeared to have an increased degree centrality in the blank control at 7 and 14 dpi, respectively. This suggests that these groups become more interconnected and potentially important as the microbial communities change under these conditions. In the networks of the infected control (Fg control) at 7 and 14 dpi, maximum degree was lower compared to the other treatments with maximum values of 8 and 7, respectively. For treatment LMG19352, more taxa had a degree centrality value ≥10 at 7 dpi than at 14 dpi, while for the co-inoculated wheat ears the opposite trend was observed. However, the taxa with elevated degree centrality values were different between both treatments and timepoints.

The betweenness centrality values are relatively lower than the degree centrality values, indicating that few taxa serve as connectors between groups ( Figure 5B ). Moreover, in total 37% of all betweenness centrality values was zero. However, some bacterial and fungal classes exhibited high betweenness centrality values as well. For example, a value of 78 was determined for Planctomycetes in the blank control at 0 dpi. Additionally, Saccharimonadia displayed a value ≥ 80 in the co-inoculated network at the presymptomatic stage, while in the network of the infected control at that timepoint the highest betweenness centrality values (≥ 80) were shown for Actinobacteria and Agaricomycetes. These findings indicate that all these taxa act as key connectors within the microbial network of different treatments and different timepoints. Contrastingly, only low betweenness centrality values (≤ 4) were measured for all taxa in the network of the infected control during the symptomatic stage. However, it needs to be mentioned that more than half of the fungal classes did not show any significant correlations with other classes and thus, were excluded from the network.

When comparing both centrality measures, we were able to find three potential hub taxa out of three different networks. Over all treatments and timepoints, the highest degree was observed in the network of symptomatic, co-inoculated samples with a value of 15 for Parcubacteria. Within this network, the highest value for betweenness centrality was determined for Parcubacteria as well. These results imply their function as hub taxa within this network. Secondly, Planctomycetes stood out as potentially influential bacterial taxa in the blank control at 0 dpi, bridging the bacterial cluster with solely positive connections to the rest of the network ( Supplementary Figure 4 ). Thirdly, Chloroflexia acted as an important hub node in the blank control network at 14 dpi, having the highest degree and betweenness centrality values within that network. In all three networks, the potential hub taxa are located in the center of the network ( Supplementary Figures 4 , 6 , 12 ), confirming their important role in maintaining the stability of the network structure.

The symptomatic stage of FHB resulted in a significantly lower α-diversity in fungal communities from F. graminearum-treated wheat ears compared to nontreated ones, whereas the α-diversity in bacterial communities was not significantly altered. Additionally, topological and centrality features of the correlation network elicited a less complex network at the symptomatic stage of F. graminearum-infected wheat ears. Both a reduced α-diversity and a less complex network indicate a diseased status of the microbiome, as previously shown in other studies (Cardoni et al., 2023; Legrand et al., 2019; Rojas et al., 2020). In addition, when comparing flowering stage (i.e. 0 dpi) with mid-grain filling stage (i.e. 14 dpi) of F. graminearum inoculated ears, multiple significant shifts in the fungal spicosphere microbiome community were determined, whereas it was not at early grain filling stage (i.e. 7 dpi). Similarly, some significant changes were observed when comparing blank and infected control ears at mid-grain filling stage (i.e. 14 dpi) but not at early grain filling stage (i.e. 7 dpi). These results indicate that shifts in the fungal spicosphere microbiome are not directly caused by the inoculation with F. graminearum itself but rather by the development of FHB symptoms. These shifts may be triggered by increased production of mycotoxins such as deoxynivalenol (DON), which becomes apparent during symptom development as the mycotoxin is crucial for the spread and colonization of the fungus within the wheat rachis (Audenaert et al., 2014). DON interacts specifically with eukaryotic systems by forming three hydrogen bonds with the A-site of the 60S ribosomal subunit, inhibiting peptidyl transferase activity. This specificity could explain why the bacterial microbiome in the wheat spicosphere remained more stable than the fungal microbiome. Our hypothesis aligns with findings from Bakker and McCormick (2019), who reported a negative correlation between DON levels and microbial richness, evenness, and diversity, with more pronounced effects on fungal communities. However, they concluded that the microbiome did not independently influenced toxin accumulation beyond its effects on pathogen load. Future studies could elicit the impact of DON on shifts in the spicosphere microbiome by performing an additional mycotoxin analysis or by including wheat ears inoculated with DON.

At early grain filling stage (i.e. 7 dpi), significant changes in the spicosphere structure and composition of fungal communities were observed after applying S. rimosus subsp. rimosus LMG19352. The BCA-inoculated wheat ears showed a significant decrease in fungal richness, evenness and diversity when compared to the blank control. This reduction could largely be attributed to a high number of counts for a specific fungal genus, Cladosporium (class of Dothideomycetes), which is naturally present on wheat ears, as confirmed by its consistent isolation across sampling sites in a study conducted in Germany (Schiro et al., 2019). This fungal genus may have benefited from the inoculation with S. rimosus subsp. rimosus LMG19352 and outcompeted other fungal genera, resulting in a reduced richness, evenness and diversity. These results were supported by a significant negative correlation of Dothideomycetes with both Sordariomycetes and Leotiomycetes, as displayed in the correlation network. Similarly, Rojas et al. (2020) found that Cladosporium (class of Dothideomycetes) negatively correlated with Fusarium spp. (class of Sordariomycetes) on wheat ears, linking it to spikes that remained healthy (Rojas et al., 2020). Although Cladosporium herbarum is known to cause Cladosporium Ear Rot disease on maize, some other species have been explored as mycoparasitic BCAs on wheat against several fungal diseases such as Blumeria graminis (Zhu et al., 2024) and Puccinia striiformis f. sp. tritici (Zhang et al., 2022). Notably, in wheat ears treated with S. rimosus subsp. rimosus LMG19352 alone, the relative abundance of Blumeria was lower, suggesting that Cladosporium’s increased abundance may have contributed to its biocontrol activity and that the application of our BCA had an indirect effect on the biocontrol of Blumeria, although the pathogenic species was still omnipresent without causing disease symptoms. In future research, a higher-resolution technique with sequence information of longer reads, able to identify taxa at species or even strain level, could be used to determine which species of Cladosporium was observed (Knief, 2014).

At mid-grain filling stage (i.e. 14 dpi), most shifts in the fungal spicosphere microbiome were partially restored towards its original state at the non-inoculated stage (i.e. 0 dpi). Additionally, when comparing blank and LMG19352-treated wheat ears at mid-grain filling stage (i.e. 14 dpi), no significant differences could be observed between both treatments. These results imply limited long-term effects on non-target microorganisms after applying BCA S. rimosus subsp. rimosus LMG19352 on wheat ears in the absence of a F. graminearum infection. These results can provide a starting point for extended greenhouse trials or field trials by means of validation.

Co-inoculating wheat ears with BCA S. rimosus subsp. rimosus LMG19352 and F. graminearum resulted in a significant decrease of FHB symptoms on infected spikelets during the symptomatic stage when compared to the infected control. These results are in agreement with earlier findings of Tan et al. (2021), who attributed this inhibitory effect to direct antagonism. Additionally, the fungal and bacterial community structures of co-inoculated ears were not significantly different from both the infected or blank control but acted intermediate, similar to findings of Alukumbura et al. (2022) who inoculated wheat ears with T. gamsii T6085 as BCA against naturally occurring FHB and could also not distinguish significant differences in fungal and bacterial community structures between BCA-treated and infected control plants. In addition, the fungal relative abundances of co-inoculated ears remained more stable over time when compared to the infected control ears. However, the relative abundance of Sordariomycetes (class of F. graminearum) in co-inoculated ears was intermediate and not significantly different from the blank and infected control, which is in accordance with F_v/F_m and cGFP values of the sampled spikelets and results of the HCPC. A plausible explanation for the intermediate effect on F. graminearum inhibition is the variable efficacy of the BCA, as often reported as a challenge for other BCAs. Combining BCAs among themselves or with chemical fungicides could help to reduce their variability and inconsistency (Ons et al., 2020; Xu et al., 2011). Alternatively, fluctuations in the microbiome could have influenced the BCA efficacy as well. For example, the efficacy of a BCA could be enhanced by the presence of a helper strain in the endogenous microbiome – an organism which does not have biocontrol properties itself but facilitates or enhances biocontrol properties of the BCA through, e.g. better BCA establishment, enhanced survival on the host, or improved metabolite production (Massart et al., 2015). In our study, Leotiomycetes might have acted as such a helper class since their relative abundance was significantly lower in cluster 6 of the HCPC, which includes samples with the highest abundances of Sordariomycetes. Additionally, their relative abundance was significantly increased in cluster 4 of the HCPC, containing three out of five replicates of co-inoculated ears at the symptomatic stage with reduced abundance of Sordariomycetes, potentially increasing the biocontrol potential of our BCA.

Together, these results indicate that BCA S. rimosus subsp. rimosus LMG19352 tempers shifts in the spicosphere microbiome induced by FHB symptoms, promoting a diverse and balanced microbiome with reduced abundance of F. graminearum. This suggests its role as BCA with a relatively novel mode of action, as previously described by Erlacher et al. (2014). Similarly, they demonstrated that the rhizosphere and phyllosphere microbiome of lettuce shifts as a consequence to pathogen attack but that BCA Bacillus amyloliquefaciens FZB42 can compensate these effects. Additionally, previous research has shown that S. rimosus subsp. rimosus LMG19352 is able to diminish the production of the mycotoxin DON, and to mitigate another important mycotoxin, zearalenone (ZEN), in vitro and in planta (De Troyer et al., 2024; Tan et al., 2021), supporting our hypothesis that shifts in the spicosphere microbiome are triggered by the production of mycotoxins. However, it remains unclear whether these changes are directly driven by S. rimosus subsp. rimosus LMG19352 since no elevated abundance in Actinobacteria was detected over time in the sampled spikelets of co-inoculated ears, which supports previous findings regarding the limited motility of Streptomyces spp (Muok et al., 2021). In future studies, tagging the inoculated bacterial strain with a fluorescent marker, e.g. RFP, could provide insights in its capacity to colonize wheat ears and could clarify whether the inoculation with S. rimosus subsp. rimosus LMG19352 directly influences the abundance of Actinobacteria in the sampled spikelets. Alternatively, applying the BCA via spraying may address this challenge of limited motility and may provide an effective alternative approach, since it has been demonstrated that S. rimosus subsp. rimosus LMG19352 could persist at the inoculated spikelet of wheat ears for at least 25 days (De Troyer et al., 2024).

Overall, the results of this study are promising, although further scientific research is required before S. rimosus subsp. rimosus LMG19352 can be implemented in the field as a BCA against FHB. Field trials are essential to fully assess the efficacy of S. rimosus subsp. rimosus LMG19352, as BCAs often show reduced performance under field conditions. For instance, Colombo et al. (2019) demonstrated that the biocontrol effect of Streptomyces sp. DEF09 was lower in a field study when compared to a greenhouse experiment. Future field studies should also consider incorporating more biological replicates, since within-treatment variability is likely to increase under field conditions, where environmental factors such as temperature and relative humidity cannot be controlled. Additionally, the spicosphere microbiome in field studies should be investigated as it will be more representative to the native microbiome compared to pot experiments. However, the wheat microbiome is influenced by multiple factors as reviewed by Kavamura et al. (2021), including anthropogenic factors (e.g. fertilizer use), soil properties (e.g. soil type), environmental conditions (e.g. abiotic stress), and host factors (e.g. growth stage), which complicates generalization of results across studies. Nonetheless, the results of this study together with all the features demonstrated in previous research on the biocontrol activity of S. rimosus subsp. rimosus LMG19352 against F. graminearum and its mycotoxins (De Troyer et al., 2024; Tan et al., 2021), make it a suitable candidate for further investigations as a BCA against FHB.