Grapevine wood microbiome analysis identifies key fungal pathogens and potential interactions with the bacterial community implicated in grapevine trunk disease appearance

Grapevine trunk diseases (GTDs) is a disease complex caused by wood pathogenic fungi belonging to genera like Phaeomoniella, Phaeoacremonium, Fomitiporia, Eutypa and members of the family Botryosphaeriaceae. However, the co-occurrence of these fungi in symptomatic and asymptomatic vines at equivalent abundances has questioned their role in GTDs. Hence, we still lack a good understanding of the fungi involved in GTDs, their interactions and the factors controlling their assemblage in vines. We determined the fungal and bacterial microbiome in wood tissues of asymptomatic and symptomatic vines of three main Greek cultivars (Agiorgitiko, Xinomavro, Vidiano), each cultivated in geographically distinct viticultural zones, using amplicon sequencing. We noted that cultivar/biogeography (lumped factor) was the strongest determinant of the wood fungal microbiome (p < 0.001, 22.7%), while GTD symptoms condition had a weaker but still significant effect (p < 0.001, 3.5%), being prominent only in the cultivar Xinomavro. Several fungal Amplicon Sequence Variants (ASVs), reported as GTD-associated pathogens like Kalmusia variispora, Fomitiporia spp., and Phaemoniella chlamydosporα (most dominant in our study), were positively correlated with symptomatic vines in a cultivar/viticultural zone dependent manner. Random Forest analysis pointed to P. chlamydosporα, K. variispora, A. alternata and Cladosporium sp., as highly accurate predictors of symptomatic vines (0% error rate). The wood bacterial microbiome showed similar patterns, with biogeography/cultivar being the main determinant (p < 0.001, 25.5%) of its composition, followed by the GTD status of vines (p < 0.001, 5.2%). Differential abundance analysis revealed a universal positive correlation (p < 0.001) of Bacillus and Streptomyces ASVs with asymptomatic vines. Network analysis identified a significant negative co-occurrence network between these bacterial genera and Phaemoniella, Phaeoacrominum and Seimatosporium. These results point to a plant beneficial interaction between Bacillus/Streptomyces and GTD pathogens. Our study (a) provides evidence that GTD symptomatic plants support a wood fungal microbiome, showing cultivar and biogeography-dependent patterns, that could be used as a proxy to distinguish between healthy and diseased vines, (b) points to strong interactions between the bacterial and fungal wood microbiome in asymptomatic vines that should be further pursued in the quest for discovery of novel biocontrol agents.


Background
Grapevine trunk diseases (GTDs) are considered a major threat for viticulture [1]. However, the ban of sodium arsenate, the main chemical used for their control, along with changes in cultivation practices and the use of infected planting material resulted in a strong upsurge in the incidence of GTDs globally [2][3][4] with estimated annual economic losses summing to 1.132 million euros [5]. GTDs are considered a complex of diseases that could be distinguished into three major groups: Eutypa, Esca and Botryosphaeria diebacks [6]. Several fungi have been linked with GTDs like: (1) Eutypa lata and other Eutypa species associated with Eutypa dieback [7], (2) Phaeomoniella chlamydospora, Phaeoacremonium spp., and Fomitiporia mediterranea associated with esca decline [8,9] and (3) different genera of Botryosphaeriaceae (i.e. Neofusicoccum, Lasiodiplodia, Diplodia, Botryosphaeria) associated with the Botryosphaeria dieback [10][11][12]. GTDs appear mostly in mature vines (> 8 years old) causing central, black necrosis and/or white rot that eventually lead to vine stunting and in acute cases to apoplexy [13]. Foliar symptoms often appear years after infestation [6] and show transitory patterns of appearance [14]. Fungal pathogens have not been recovered yet from symptomatic leaves suggesting that leaf symptoms are associated with the production and translocation to leaves of (1) secondary metabolites produced by the wood pathogens [15][16][17][18] (2) phytotoxic by-products of wood decay or phytotoxic metabolites produced by the plant-fungi interaction [19].
Several studies have used culture-dependent approaches to isolate the causal agents of GTDs from symptomatic and comparatively from asymptomatic vines and/or from necrotic or soft rot tissues of vines [3,5,20]. These studies suggested that all fungi previously identified as causal agents of GTDs were found at equivalent abundance levels in symptomatic and asymptomatic vines, implying that they are either latent pathogens or members of the endophytic vine microbiome causing symptoms when edaphoclimatic factors and vine genotype are conducive [5]. Culture-dependent approaches and low-resolution molecular tools (i.e. Single-Strand Conformation Polymorphism) failed to unravel the full diversity of the endophytic fungal microbiome in vines [3,20,21]. This is due to the limited cultivability of most environmental microorganisms [22] and the bias of culture-dependent methods towards fast growing strains [23]. A few recent studies using amplicon sequencing provided deeper insights into the composition of wood fungal microbiome in GTD symptomatic and asymptomatic vines. For example, Bruez et al. [24] did not identify compositional differences in the communities colonizing non-necrotic tissues of symptomatic and asymptomatic vines. Differences became evident when necrotic tissues were studied comparatively. In those tissues F. mediterranea dominated the fungal community. Similar studies in Portugal and Australia did not identify major differences in the fungal microbiome in asymptomatic vs symptomatic vines [25,26]. A recent study by Bruez et al. [27] showed that sodium arsenate application in escadiseased grapevines resulted in significant changes in the wood fungal microbiome with the most striking result being the strong inhibition of F. mediterranea. All these studies focused on a single cultivar in a single vineyard and have not considered differences in the resistance of vine cultivars to GTDs [6,28] or the role of biogeography in the manifestation of symptoms in vineyards.
Besides fungi, the endophytic bacterial microbiome of grapevine and its role in GTDs is largely unexplored. Grapevines support a diverse bacterial community from roots to berries. The latter have been the focus of most vine microbiome studies [29][30][31]. Wood-associated bacteria might have a suppressive effect on GTDs. They could act as forerunners of fungal wood pathogens facilitating the infection of wood tissues or as late comers colonizing necrotic or white rot tissues. To date only a few studies have investigated comparatively the composition of the bacterial community in the trunk of asymptomatic and symptomatic vines. Bruez et al. [32] identified Bacillales, Enterobacteriales and Xanthomonadales as the most abundant bacterial orders colonizing the trunk of grapevines. However, they failed to identify consistent differences in the composition of the bacterial communities in asymptomatic and symptomatic vines. Similar results were obtained in recent studies using amplicon sequencing [24,26].
We tested the hypotheses that (1) the composition of the fungal wood microbiome differs between asymptomatic and symptomatic vines with GTD-associated fungi being the key drivers of this differentiation, (2) the composition of the bacterial community differs between asymptomatic and symptomatic vines and specific bacterial taxa would be potential interactors with GTD-associated fungi and (3) vine genotype, biogeography and vine GTD condition interactively determine the composition of the vine wood microbiome and the fungal wood pathobiome. These hypotheses were evaluated in asymptomatic and GTD-symptomatic vines from three Greek cultivars, each cultivated in geographically distinct viticultural zones, using amplicon sequencing of the 16S rRNA gene for bacteria and the ITS2 region for fungi. Multivariate analysis and machine learning approaches identified members of the wood fungal microbiome that could act as GTD predictors, and network analysis revealed microbial interactions relevant for the appearance of the disease.

Collection of samples
Wood samples were collected during July-August 2019 from vineyards located in three important and geographically distinct viticultural zones in north-west (Amyntaio), central-south (Nemea, Peloponnese) and southern Greece (Crete) (Additional file 1: Fig. S1). These regions are mainly cultivated with the Greek cultivars Xinomavro, Agiorgitiko and Vidiano, respectively. The three viticultural zones are characterized by different climatic conditions: (1) The Amyntaio region is humid (mean annual precipitation 611 mm) and cool (mean annual temperature 12.2 °C, max 18.4 °C and min 6.7 °C) (2) The Nemea region is equally humid (mean annual precipitation 611.3 mm) but warmer than Amyntaio (mean annual temperature 15.8 °C, max 22.5 °C and min 8.7 °C) (3) The Heraklion region in Crete is the driest (mean annual precipitation 464 mm) and warmest (mean annual temperature 17.8 °C, max 21.9 °C and min 14.0 °C) of the viticultural zones studied.
In each zone, three vineyards, with the age of grapevines varying from 10 to 15 years, were selected for collecting samples. Details on the management practices, age, pruning system and rootstocks used in the nine vineyards of the current study are given in Table 1. In each vineyard, three vines without any visible GTD symptom on leaves and wood, and three vines exhibiting typical foliar GTD symptoms (including leaf discoloration, tiger-stripe chlorosis or necrosis and plant wilting) in the last two years were selected and nominated as asymptomatic and symptomatic vines, respectively. The only exception was recorded for the cultivar Agiorgitiko where for technical reasons asymptomatic plants were sampled only from one of the three vineyards. Each one of the vines selected as symptomatic or asymptomatic were destructively sampled. Specifically, wood samples from the cordons and the trunk of the selected vines were removed using a sterile saw and transferred in the laboratory. Then, cordon and trunk samples were cut longitudinally to identify the extent of necrosis in these wood tissues. Wood samples from asymptomatic plants showing extensive necrotic areas were excluded from further sampling. Overall, samples from symptomatic plants were characterized by extensive necrotic areas, whereas samples from asymptomatic plants were largely free of necrotic areas. Based on this visible inspection, wood chips from necrotic tissues of symptomatic vines and wood chips from non-necrotic tissues of asymptomatic vines were collected using a sterile drill and stored at − 20 °C until further use. White rots were occasionally found adjoining necrotic areas and they were sampled along with necrotic tissues but they were not considered separately. A detailed list of the samples used in the study is given in Additional file 1: Table S1.

DNA extraction
DNA from the wood samples was extracted using the Cetyltetramethyl ammonium bromide (CTAB) method [33] with slight modifications. Briefly, 250 mg of the collected wood tissue were homogenized in a chilled pestle and mortar using liquid nitrogen. 1200 μl (600 μl in two doses) of 2X CTAB extraction buffer (100mMTris-HCl, 20 mM EDTA, 1.4 M NaCl, 2% CTAB, 0,5% v/v β-mecraproethanol) were added and homogenized. The samples were incubated at 65 °C for 30-60 min with occasional vortexing. Samples were centrifuged at 10,000 rpm for 10 min and the supernatant (~ 400 μl) was transferred to new tubes. 600 μl of chloroform:isoamyl alcohol (24:1) were added and mixed by inverting the tube several times. Samples were centrifuged at 13,000 rpm for 15 min. The aqueous phase was transferred into a new tube and 600 μl of chloroform were added. The tubes were centrifuged at 13,000 rpm for 10 min. Then, 1 volume of chilled isopropanol was added, followed by quick and gentle inversion and incubated at − 20 °C for 1 h. The DNA pellet was precipitated at 13,000 rpm for 20 min, washed twice with 70% ethanol and precipitated at 13,000 rpm for 3 min. The DNA pellet was then suspended in 50 μl of pre-heated TE-buffer. DNA integrity was initially assessed via agarose gel (0.8%) electrophoresis and DNA concentrations were determined using the Quant-iT kit with a Qubit Fluorometric device (Invitrogen, USA).

Amplicon sequencing analysis
The bacterial and fungal wood microbiome of vines was determined via amplicon sequencing of the V4 region of the 16S rRNA gene and the ITS2 region respectively, using a HiSeq, Illumina platform in Rapid Mode 2 × 250 paired-end in Admera Health company (New Jersey, USA). Amplification, facilitating sample-wise multiplexing during sequencing, was based on our in-house protocol described in detail in Vasileiadis et al. [34]. Briefly, extracted DNA was diluted to a final concentration of 1 ng μl −1 for fungi and 0.2 ng μl −1 for bacteria. We followed a two-step amplification process, which allowed indexing of our samples. In the first step, the bacterial 16S rRNA gene was amplified with the updated version of the primers 515f-806r [35] following the protocol of the earth microbiome project [36]. Regarding fungi, a first amplification step of the ITS2 region was performed with primers ITS7F-ITS4R [37,38]. The first amplification step included 28 cycles using the primers mentioned above, followed by a second PCR of 7 cycles using the same primers with sample-specific 5' overhangs. All PCR reactions were conducted with the use of Q5 ® High-Fidelity DNA Polymerase (NEB, Ipswich, Massachusetts, USA). Primers, PCR conditions and the reagents used are given in Additional file 1: Tables S2 and S3 respectively. Amplicon libraries were cleaned up with the use of NucleoMag ® NGS Clean-up and Size Select Kit (Macherey-Nagel, Duren, Germany) according to manufacturer instructions and they were sent for sequencing. Amplification of non-template controls gave no amplicons and their sequencing provided particularly low number of reads, hence they were not considered in the analysis.

Bioinformatic analysis
Sequence demultiplexing to their samples of origin was performed with flexbar v3.0.3 [39] not allowing any barcode mismatches. Sequences were then quality-trimmed/ filtered and checked for chimerical sequences with the dada2 v1.14.1 [40] package of the R software v.4.0.2 [41]. Amplicon sequence variants (ASVs) were assembled and taxonomically assigned using the Silva v.138 [42] and UNITE ITS v.8.2 (04.02.2020) [43] reference databases for bacteria and fungi, respectively. ASVs not classified at Kingdom and Phylum level were excluded from downstream analysis. Whereas we included in the analysis ASVs with 80% or higher taxonomy annotation bootstrap confidence in the domain/kingdom taxa as previously suggested [44]. ASV classifications are provided at the lowest taxonomical rank passing this threshold.

Statistical analysis
The ASV matrices of bacteria and fungi were used to assess the effects of vine GTD condition and the lumped factor biogeography/cultivar on the β-diversity of the wood microbiome. Non-metric multidimensional scaling (NMDS) and permutational multivariate analysis of variance (PERMANOVA) were used to assess β-diversity differences for bacterial and fungal communities between symptomatic and asymptomatic vines either collectively for all cultivars/viticultural zones or separately for each viticultural zone/cultivar. NMDS was performed with dissimilarity matrices using the Bray-Curtis algorithm, while the association of microbiome composition with the covariates of interest and the determination of the % of variance explained by those factors were accessed with PERMANOVA [45] using 999 permutations as implemented in the Vegan v2.5-7 [46] package of the R software. Differential abundance (DA) tests, using the Kruskal-Wallis and the Wilcoxon rank-sum post-hoc tests, were employed to identify bacterial and fungal taxa that were significantly associated with symptomatic and asymptomatic vines. P-value adjustment was conducted with false discovery rate (FDR) algorithm with implemented adjusted P-value cut-off values of 0.05. Heatmaps were produced to depict ASVs whose relative abundance was positively correlated with symptomatic or asymptomatic vines according to the DA tests.
The GTD-associated fungal genera and the fungal and bacterial genera that participate in 10% of the samples with at least 1% relative abundance were used in network analysis aiming to identify positive or negative co-occurrence networks. Spearman correlation tests were performed among microbial genera and the data obtained were used as adjacency matrix for the weighted network analysis performed with the igraph v1.2.6 [47] package of the R software. The minimum spanning tree algorithm was used for filtering the adjacency matrix choosing the shortest possible combination of connections among tree-nodes on the generated undirected networks [48]. Afterwards, the sub-community clusters were identified according to local densities. The force-directed Fruchterman-Reingold layout [49] was used for plotting the final network.
We further selected fungal ASVs to define potential predictors of the GTD status of grapevines via Random Forest analysis. The data matrix was initially subsampled with replacement to equal read numbers (rarefied) as suggested by Weiss et al. [50]. Then, a series of Random Forest supervised classification models were performed by the pime v0.1.0 R-software package [51], which relies on the randomForest v4.6.14 R-software package [52]. Briefly, low prevalence taxa were filtered out for a range of minimum prevalence values (from 5 to 65% at 5% increments) for each sample. Each Random Forest model was run with 1000 trees, using 66% of the samples for building the model and 33% of the samples as out of bag (OOB) samples for the model error assessment. The OBB rate for each prevalence level was calculated for the dataset and the best compromise between model error rate and the remaining dataset reads was selected. The Random Forest analysis was implemented on a dataset of 63 samples (18 from asymptomatic and 45 from symptomatic vines) and 1182 taxa. Overall, the 40% of minimum prevalence cutoff was selected as the best compromise between model error and ASV/sequence numbers for the model, corresponding to 118,161 sequences and 22 ASVs, with an internal OOB error rate 6.25%. Following the Random Forest-based pime approach, the Kruskal-Wallis non-parametric ANOVA equivalent test coupled with a post hoc Wilcoxon rank sum analysis was performed for further filtering out ASVs showing no significant differences between GTD sample groups. This reduced the dataset size to 10 ASVs and 50,649 sequences, which was re-tested for its sample classification accuracy with Random Forests according to the aforementioned parameters.
All data figures for the analysis were produced with the R package ggplot2 v3.3.3 [53]. The sequencing data were submitted to Sequence Read Archive of NCBI with bioproject accession number PRJNA738455. The full analysis code can be found in the Github repository in the link: https:// github. com/ Fotis bs/ Grape vine_ wood_ micro biome-. git

Fungal microbiome Composition of the wood fungal microbiome
In total 1,043,027 non chimeric quality fungal sequences (sequence numbers in the analyzed samples ranged from 1413 to 33,223) were obtained and they were assigned to 1378 ASVs. Rarefaction curves reached a plateau in all samples suggesting that our sequencing effort adequately covered the diversity of endophytic fungi in vines (Additional file 1: Fig. S2a) Fig. S3).

Factors structuring the wood fungal microbiome of vines
Preliminary analysis of the amplicon sequencing data suggested a non-significant effect of the type of plant tissue studied (cordon or trunk) on the fungal community and hence, it was not considered as an explanatory variable in our analysis. When samples from all viticultural zones/cultivars were considered, NMDS analysis showed a clear grouping (p < 0.01) of the samples according to viticultural zone/cultivar (Fig. 1a), but no clustering according to their GTD condition (symptomatic vs asymptomatic plants) (Fig. 1b). Still, PERMANOVA showed that GTD condition and viticultural zone/cultivar had a statistically significant effect and explained 3.5% (p < 0.001) and 22.7% (p < 0.001) of the dataset variance, respectively (Additional file 1: Table S4).
We further analyzed the differences in the β-diversity of the fungal microbiome imposed by GTDs separately for each genotype/viticultural zone. Regarding Xinomavro, NMDS analysis showed a clear separation of samples collected from GTD-asymptomatic and -symptomatic plants (Fig. 2b). Further PERMANOVA showed that GTD condition was the most significant factor (p < 0.001) explaining 13.9% of the variance, while vineyard topography (in the same viticultural zone) contributed 8.6% (p < 0.05) (Additional file 1: Table S4). In contrast, NMDS analysis for the cultivars Agiorgitiko and Vidiano showed a marginal (p < 0.05) or not clear separation (p > 0.05) respectively, of the symptomatic and asymptomatic samples (Fig. 2a, c).
PERMANOVA further indicated that the GTD condition of plants explained 8.4% and 4.2% of the variance for Agiorgitiko and Xinomavro respectively and had a marginally significant (p < 0.05) or non-significant (p > 0.05) effect on the fungal β-diversity. Still for these two cultivars, the vineyard location (within the same viticultural zone) was the most significant factor (p < 0.001) explaining 20.5% and 15% of the variance in each case (Additional file 1: Table S4).

Fungi of the wood microbiome associated with symptomatic and asymptomatic vines
We further attempted to define ASVs whose relative abundance significantly differs between asymptomatic and symptomatic plants. DA heatmaps were constructed collectively for all cultivars/viticultural zones and separately for each cultivar/viticultural zone. When all cultivars were considered together only three ASVs showed significant association with either asymptomatic or symptomatic vines (Fig. 3a). ASV102 assigned to the genus Alternaria, showed a consistently higher relative abundance in asymptomatic compared to symptomatic plants. Conversely, ASV027 and ASV068, assigned to Acremonium alternatum and Kalmusia variispora respectively, showed significantly higher relative abundance in the symptomatic vines. We expanded our analysis and looked separately in each cultivar/viticultural zone for fungal ASVs with differential abundance patterns in symptomatic vs asymptomatic plants. Regarding the cultivar Agiorgitiko, we identified 11 ASV (p < 0.05) that showed differential abundance between asymptomatic and symptomatic vines. Amongst them, ASVs belonging to Eutypa spp., Lophiostoma cynaroidis, Sporidiobolaceae, Malassezia restricta, Didymellaceae, Amphisphaeriaceae, Teloschistaceae, Seimatosporium vitis and Muriphaeosphaeria viburni were associated with asymptomatic plants, unlike AS0V52, assigned to the genus Fomitiporia, which was amongst the most abundant ASVs of the wood microbiome, and it was systematically associated with symptomatic plants (Fig. 3b). Regarding the cultivar Xinomavro, we identified a single ASV, assigned to genus Neosetophoma, that was associated with symptomatic vines (Fig. 3c). In contrast, ASVs belonging to the genera Alternaria, Curvularia and the family Nectriaceae were associated with asymptomatic vines. In the cultivar Vidiano, we identified ASVs belonging to P. chlamydospora, one of the most dominant ASVs in our dataset (ASV014), Acremonium sordidulum, Diaporthe spp., K. variispora and A. alternatum, all associated (p < 0.05) with symptomatic vines (Fig. 3d), compared to ASVs belonging to M. restricta and Catenulostroma spp., which showed a consistent association with asymptomatic vines.

Factors structuring the wood bacterial microbiome
In line with the fungal wood microbiome, we observed that the two plant parts sampled, cordons and trunk, supported similar bacterial communities and hence this factor was not considered further in our analysis. When samples from all viticultural zones/cultivars were considered, NMDS analysis showed a clear grouping (p < 0.01) of the samples according to viticultural zone/cultivar (Fig. 4a), whereas no clustering according to the GTD conditions of samples (symptomatic vs asymptomatic) was observed (Fig. 4b). Still PERMANOVA showed that GTD condition and cultivar/viticultural zone exerted a statistically significant effect (p < 0.001) on bacterial community structure explaining 5.2 and 25.5% of the dataset variance respectively (Additional file 1: Table S4).
We further analyzed our data separately for each viticultural zone/cultivar aiming to explore further the effect of GTD occurrence on the composition of the bacterial microbiome. Regarding the cultivar Xinomavro, NMDS analysis showed a clear separation of the samples from symptomatic and asymptomatic vines (p < 0.01) (Fig. 5b). PERMANOVA verified the significant effect of the GTD plant condition on the composition of the wood bacterial community (p < 0.001), which explained 19.7% of the variance, whereas vineyard topography (within the viticultural zone of Amyntaio) did not have a significant effect (p > 0.05) (Additional file 1: Table S4). Regarding the other two cultivars, we noted a weak clustering of the samples according to their GTD status only in the case of Vidiano (p < 0.05) (Fig. 5c). PERMANOVA analysis revealed that GTD condition and vineyard topography (at the viticulture zone scale) explained 5.3% (p > 0.05) and 13.6% (p < 0.05) of the variance respectively for Agiorgitiko, and 6% (p < 0.01) and 12.1% (p < 0.01) for Vidiano (Additional file 1: Table S4).

Wood microbiome bacteria associated with symptomatic and asymptomatic vines
DA heatmaps pointed to ASVs that were consistently associated with symptomatic or asymptomatic vines. When data from all viticultural zone/cultivars were considered, we identified several ASVs, mostly belonging to Rhizobiaceae, that showed consistently higher relative abundance in symptomatic vines (Fig. 6a). On the other hand, five Streptomyces, ten Bacillus, five Bacillaceae, and four Corynebacterium ASVs were consistently present in the asymptomatic vines. When the same analysis was conducted separately for each viticultural zone/cultivar, we identified four ASVs, all belonging to Bacillaceae, that were associated (p < 0.05) with asymptomatic vines of the cultivar Agiorgitiko (Fig. 6b). Similarly, we noted five Bacillaceae, eleven Bacillus and six Streptomyces ASVs that were systematically associated (p < 0.05) with asymptomatic vines of the cultivar Xinomavro (Fig. 6c). Regarding, Vidiano six ASVs belonging to Rhizobiaceae were systematically associated with symptomatic vines (Fig. 6d), whereas ASVs belonging to Sphingomonadaceae (2), Corynebacterium (4), Bacillaceae (2) and Acinetobacter (2) were associated with asymptomatic vines.

Network analysis of the wood microbiome
Network analysis within the fungal microbiome across all cultivars revealed a significant negative co-occurrence pattern between Alternaria and the two GTD-linked fungal genera Phaeomoniella and Phaeoacremonium in the asymptomatic but not in the symptomatic vines (Additional file 1: Fig. S6). Network analysis between the GTD-associated fungi and the bacterial microbiome in all cultivars revealed a significant negative co-occurrence pattern between the GTD-relevant fungal genera Phaeomoniella and Seimatosporium with Bacillus, and of Phaeoacremonium with Streptomyces, only in the asymptomatic grapevines (Additional file 1: Fig. S6).

Random forest analysis of the fungal wood microbiome
To further explore our fungal dataset for possible marker ASVs for asymptomatic and symptomatic samples, a Random Forest supervised learning approach was used. This analysis showed that the fungal microbiome had strong discriminative power and pointed to 10 ASVs that could be considered as predictors of the GTD status of the vines. Hence ASVs belonging to Cladosporium spp., (ASV004, p < 0.05, most dominant), P. chlamydosporα (ASV014, p < 0.05, dominant), K. variispora (ASV068, p < 0.001), A. alternatum (ASV027, p < 0.01) and Lentitheciaceae (ASV154, p < 0.01) were all predictors of symptomatic plants (Fig. 7). Whereas ASVs belonging to Fusarium solani (ASV132, p < 0.038), Alternaria spp., (ASV102 p < 0.0 and ASV007 p < 0.001), Curvularia spp., (ASV039, p < 0.02) and Nectriaceae (ASV025, p < 0.001) were highlighted from the model as identifiers of asymptomatic vines. These ASVs had an overall (for both symptomatic and asymptomatic samples) OOB error rate of 7.94% (Additional file 1: Fig. S7). When using the five ASV identifiers for symptomatic samples we noted a high classification accuracy with OOB error rates of 0% suggesting that all 45 symptomatic samples were classified correctly. In contrast, we observed a low classification accuracy when using the ASV indicators of asymptomatic plants with an OOB error rate of 27%.

Discussion
We determined the composition of the wood fungal microbiome in symptomatic and asymptomatic vines from three different Greek cultivars, each located in a geographically distinct viticultural zone, aiming to identify consistent drivers of the disease. In this quest we used amplicon sequencing approaches and relevant statistical and machine learning tools to identify potential biotic drivers of GTDs. We further expanded our study to the bacterial wood microbiome expecting to identify interactors with the fungal microbiome.

Fungal microbiome
The wood fungal microbiome of all studied cultivars was dominated by Ascomycota, while Basidiomycota constituted a substantially smaller fraction with Agaricomycetes being the most prevalent amongst them. Similar compositional patterns, denoting a clear dominance of Ascomycota in the wood microbiome of vines, have been reported in most previous studies [24,25,31]. An interesting pattern was the consistent and significant (p < 0.05) increase in the contribution of Agaricomycetes in symptomatic vines in all studied cultivars. This observation might be associated with the capacity of members of this fungal taxon to induce white rots in wood tissues, including Fomitiporia spp., a key colonizer of white rot tissues in vine exhibiting GTD symptoms [1]. Recently, Paolinelli et al. [54] suggested that the increase in the Basidiomycota/Ascomycota ratio observed in symptomatic over asymptomatic vines could be used as an early warning for the establishment of GTDs in vines.
We first explored the factors shaping the wood fungal microbiome and identified viticultural zone and/or cultivar, considered as a lump factor based on our experimental setup, as the main determinant of the fungal community. In contrast, we noted a weak differentiation of the wood fungal microbiome between symptomatic and asymptomatic samples. To date several studies have explored the effect of biogeography, cultivar and vintage on the grapevine microbiome focusing on above-ground (berries, leaves) [29,55,56] and/or below-ground plant associated compartments (soil, rhizosphere, roots) [30,57], while the trunk microbiome has attracted little attention. These studies have identified biogeography as a main determinant of the non-random assemblage of the fungal vine microbiome, especially when microbiomes are looked at the macro-geographical scale, whereas at regional scale (micro-geographical scale) other confounding factors like cultivar and vintage become important [29]. The wood fungal microbiome, as determined in our study, seems to share assemblage mechanisms with the other grapevine compartments. Although we could not distinguish between cultivar and biogeography, we speculate that the latter has possibly a stronger effect in shaping the fungal wood microbiome considering the wide geographical scale of our study. The limited dissimilarity of the fungal microbiome between asymptomatic and symptomatic vines, when the viticultural zone/cultivar factor was not considered, is in line with the previous studies, which, following similar sampling strategies, did not identify significant differences between symptomatic and asymptomatic vines [3,20,25,26].
We further determined the differences in the fungal wood microbiome separately for each cultivar/viticultural zone, taking under consideration the variable climatic conditions that prevail in the studied viticultural zones and the different sensitivity of the studied cultivars to GTDs. Indeed, Agiorgitiko and Vidiano are more  Fig. 7 Random Forest heatmap showing rarefied ASVs whose max relative abundance showed a significant positive correlation (p < 0.05*, p < 0.01**, p < 0.001***) with the wood samples from vines with (symptomatic) and without (asymptomatic) GTD symptoms susceptible to GTDs compared to the more resistant Xinomavro [28]. We noted that the three cultivars exhibited different responses in their assembly of the fungal microbiome in symptomatic and asymptomatic vines. Hence, Xinomavro was the only cultivar that assembled clearly distinct endophytic fungal communities in symptomatic vs asymptomatic plants. This result may be due to the clear visual distinction between potentially healthy and diseased plants, which is facilitated by the higher resistance of the cultivar Xinomavro to GTDs. Still, we could not exclude the contribution of the local climatic conditions, that varied in the three viticultural zones studied (see Methods), on the differential response of the studied cultivars to the assemblage of the wood microbiome in asymptomatic and symptomatic vines.
DA analysis identified members of the wood fungal microbiome that were specifically associated with asymptomatic or symptomatic vines. Considering the whole dataset, Acremonium alternatum, one of the most dominant ASV in the dataset overall, and K. variispora, were identified as indicators of symptomatic plants. The former is a common saprotrophic endophyte, previously isolated as grapevine endophyte [5,58], not exhibiting plant pathogenic traits. K. variispora has been characterized as a common wood fungus of vine, that causes irregular discoloration of vascular tissues, necrotic spots, carcinomas, and grapevine decline [20]. In Greece, this fungus has been described as a causal agent of wood necrosis and fruit rot on apples [59]. The fungus produces two polyketide massarilactones that were phytotoxic, reproducing grapevine decline symptoms on plant leaves [18]. Conversely, we identified an Alternaria ASV that was consistently associated with asymptomatic plants regardless of cultivar/viticultural zones. Fungi of this genus are common members of the cultivable grapevine wood microbiome [5,25]. Their consistent association with asymptomatic vines suggests a role in suppression of GTDs, further reinforced by the negative co-occurrence network pattern of Alternaria with hallmark GTD fungal genera like Phaemomoniella and Phaeoacromonium. In line with this Del Frari et al. [58] showed that a consortium composed of four endophytic fungi, including A. alternata, isolated from asymptomatic vines was able to decrease P. chlamydospora infection in planta. When looking at each cultivar separately, we noted different GTD-associated pathogens to be consistently present in symptomatic plants like Fomitiporia spp., in Agiorgitiko, P. chlamydospora, K. variispora and Diaporthe spp., in Vidiano. Bruez et al. [24] identified Fomitiporia as the single member of the wood microbiome that was consistently more abundant in white rot wood tissues of visibly healthy and diseased plants. They suggested that it is the key for the onset of the white rot syndrome collaborating with primary vascular tissue colonizers like P. chlamydospora, and Phaeoacremonium spp.; both of these fungi were prevalent in our study, and the latter dominated in the symptomatic Xinomavro vines. The key role of Fomitiporia, along with the two trachemycotic fungi (P. chlamydospore and Phaeoacremonium sp.) and other confounding factors like climatic conditions and vineyard age, in leaf symptoms manifestation was recently highlighted in two comprehensive reviews by Morreti et al. [60] and Del Frari et al. [19].
A range of wood endophytic fungi previously reported as members of the wood grapevine microbiome like L. cynaroidis [24] and M. restricta [25,58], but also fungi that have been reported as wood pathogens involved in GTDs like Eutypa spp., and S. vitis in Xinomavro [61,62], were associated with asymptomatic vines. The consistent presence of these pathogens in asymptomatic plants is not surprising, and previous studies have also reported the common occurrence of GTD-related pathogens in asymptomatic plants [3,5].
A Random Forest supervised learning model was used to identify members of the fungal wood microbiome that could be used as predictors of GTD symptomatic samples. This approach pointed to P. chlamydospora and K. variispora, both being members of the GTD-related pathobiome, along with A. alternatum and Cladosporium spp., not previously reported as wood pathogens in grapevine, as highly accurate predictors of symptomatic plants. Wood microbiome analysis of grapevines based on these predictors could be used as an early warning for the appearance of GTDs. Further testing with a wider dataset will verify the utility of such an approach in controlling the spread of GTDs in global vineyards.

Bacterial microbiome
The bacterial wood microbiome was overly dominated by Proteobacteria, and mostly α-and γ-proteobacteria, followed by Actinobacteria and Bacteroidetes, in line with previous studies [24,31,32]. Amongst those phyla, members of the families Sphingomonadaceae and Bacillaceae were the most prevalent. Previous culture-dependent and culture-independent efforts to record the grapevine wood bacteriome have identified Bacillaceae and Sphingomonadaceae amongst the most abundant members of this community [24,63]. The bacterial community shared similar assembly mechanisms with the fungal community, and it was strongly driven by biogeography/ cultivar as a lumped factor, unlike GTD status which did not show a high discriminatory power. Previous studies looking at the vine microbiome suggested that biogeography was the most important factor shaping bacterial communities in berries [29], vineyard soil [64,65] and the bark of grapevines [66]. The few studies that have explored the composition of the bacterial wood microbiome identified limited differences between symptomatic and asymptomatic vines [24,32], in accordance with our findings. However, in line with the fungal microbiome, we noted cultivar-/ viticultural zone-specific differentiation between symptomatic and asymptomatic vines in the cultivar Xinomavro. It is not clear what may be the driving factors for the distinct assemblages of both fungal and bacterial communities in the Xinomavro cultivar compared with the other two, more susceptible to GTD cultivars. On-going analysis of the wood microbiome of this cultivar in symptomatic and asymptomatic vines across viticultural zones will shed further light into this hypothesis.
When looking for bacterial drivers of the distinction between asymptomatic and symptomatic plants, we identified a strong enrichment of Bacillus (and widely of members of the family Bacillaceae) and Streptomyces in the former. This enrichment was particularly prominent in the cultivars Xinomavro and Agiorgitiko, while in Vidiano we noted, in addition to Bacillus, an enrichment of Corynebacterium. Bruez et al. [32] identified Bacillus spp. as major components of the wood bacterial community in vines; however, they did not observe a significant differentiation in their abundance between symptomatic and asymptomatic plants. This was further confirmed in a follow up study on the same vineyard using high throughput sequencing [24]. In the same study Streptomyces showed a significantly higher abundance in asymptomatic plants (7% vs < 0.1%), in line with our findings. The consistent prevalence of Bacillus and Streptomyces in the asymptomatic plants and their negative co-occurrence patterns with the GTD pathogens Phaeoacremonium, Phaeomoniella and Seimatosporium in asymptomatic vines could be associated with the capacity of members of these bacterial genera to act as biocontrol agents and/or plant growth promoters in several plants including grapevines [67,68]. Indeed Rezgui et al. [69] and Haidar et al. [70] reported a high suppressive activity of endophytic Bacillus strains isolated from vines against GTD-related pathogens like P. chlamydospora, Lasidiodiplodia pseudotheobromae, Neofusicoccum parvum and Schizophyllum commune, acting through the production of VOCs, antibiotics or by inducing systematic resistance in the plant host. Alvarez-Perez et al. [71] isolated an endophytic Streptomyces from asymptomatic grapevines that suppressed P. chlamydospora and Phaeoacremonium minimum in planta and in vitro. Similarly, Trotel-Aziz et al. [72] showed that a Bacillus subtilis strain, isolated from grapevine rhizosphere, was able to attenuate Botryosphaeria dieback by triggering host immune responses and detoxifying phytotoxins produced by N. parvum, mechanisms further verified by genomic analysis of the studied strain [73]. On-going high-throughput isolation effort will explore this rich endophytic bacterial community of asymptomatic plants for novel potential bacterial control agents against GTDs.

Conclusions
We studied the wood microbiome of grapevines from three Greek cultivars located in three geographically distinct viticultural zones following a systematic sampling approach distinguishing between GTD-symptomatic and asymptomatic plants. Geographical location and/ or cultivar were the main determinants of both fungal and bacterial microbiome, unlike the GTD status of grapevines that differentiated between asymptomatic and symptomatic plants only for the most resistant to GTDs cultivar Xinomavro, suggesting complex interactions between cultivar/biogeography and the wood fungal pathobiome. P. chlamydospora, K. variispora, Fomitiporia spp., Diaporthe spp., all previously proposed as causal agents of GTDs, along with Acremonium sp. (non GTD-linked fungus) were consistently associated with symptomatic vines in a cultivar-dependent manner. In addition, the first two of these pathogens, along with Cladosporium spp. and A. alternatum were identified, via Random Forest analysis, as highly accurate predictors of symptomatic vines. The bacterial microbiome in asymptomatic plants was dominated by Bacillus and Streptomyces, which showed a negative co-occurrence pattern with Phaeomoniella, Phaeoacremonium and Seimatosporium. This suggested a role of in the suppression of GTDs, a hypothesis which warrants verification through isolation and phenotypic characterization of these bacteria. Overall, we provide evidence that GTD symptomatic plants support a wood fungal microbiome showing cultivar and/or biogeography-dependent patterns. Specific members of this fungal wood microbiome could be used as a proxy to distinguish between healthy and diseased vines. Further tests with similar datasets from different viticultural regions or at large geographical scale would verify the universal utility of this approach. Potential interactions between the bacterial and fungal wood microbiome in GTD-free vines should be further pursued in the quest for discovery of novel grapevine-accustomed biocontrol agents.