Deciphering bacterial and fungal endophyte communities in leaves of two maple trees with green islands

Green islands (the re-greening of senescent leaf tissues) are particularly evident on leaves infected with fungal pathogens. To date, there is only a limited number of studies investigating foliar endophytic microorganisms in phytopathogen-infected leaves. Here, we analysed bacterial and fungal endophyte communities in leaves without green islands (control leaves; CL), within green island areas (GLA) and the surrounding yellow leaf areas (YLA) of leaves with green islands of Acer campestre and A. platanoides. GLA samples of A. campestre and A. platanoides were dominated by Sawadaea polyfida and S. bicornis, respectively, suggesting that these fungi might be responsible for the green islands. We detected a higher fungal richness and diversity in CL compared to GLA samples of A. campestre. Leaf status (CL, GLA, YLA) significantly altered the composition of fungal communities of A. campestre. This was related to differences in fungal community composition between YLA and GLA samples. Site was the main driver of bacterial communities, suggesting that bacterial and fungal endophytes are shaped by different factors. Overall, we observed Acer species-specific responses of endophyte communities towards the presence of green islands and/or leaf type, which might be attributed to several fungi and bacteria specifically associated with one Acer species.

The relative abundances of most fungal species did not differ between both Acer species, with exception of S. bicornis, S. polyfida and S. tulasnei. However, several fungal species differed in their abundance with regard to leaf status. Fungi assigned to the family Mycosphaerellaceae were predominant in YLA samples of A. campestre (68.59%) and A. platanoides (68.56%). In contrast, the fungal species S. polyfida (59.18%) and S. bicornis (61.32%) dominated the GLA samples of A. campestre and A. platanoides, respectively. The fungal species M. harthensis (4.94%) and Helicogloea farinacea (3.01%) were found in YLA of A. platanoides and CL of A. campestre, respectively, but were detected in very low abundance (<0.1%) in YLA samples of A. campestre and in CL of A. platanoides, respectively.
Leaf status significantly affected fungal endophyte diversity in A. campestre but not in A. platanoides leaves. Changes in alpha diversity values were evaluated by linear mixed effects models with site as random factor and Acer species as well as leaf status (model 1) or presence of green islands (model 2). Site significantly affected bacterial alpha diversity values (Supplementary Material Table S6). Acer species did not influence bacterial richness and phylogenetic diversity ( Table 2, Supplementary Material Table S6), but we detected a significantly higher bacterial diversity in A. platanoides than in A. campestre (model 1: F 1,54 = 5.43, P = 0.04; model 2: F 1,55 = 5.53, P = 0.02). Neither the presence of green islands nor the leaf status influenced bacterial alpha diversity values.
In contrast, fungal richness (F 1,62 = 8.71, P = 0.004) and diversity (F 1,62 = 7.37, P = 0.01) were significantly higher in control leaves compared to leaves with green islands ( Table 2, Supplementary Material Table S6). Similar results were obtained when including the leaf status in the model (richness: F 2,60 = 5.06, P = 0.01; diversity: F 2,60 = 3.99, P = 0.02). Multiple comparison using Dunn's test with Benjamini-Hochberg corrected P-values revealed a significantly higher fungal richness (P adj. = 0.04) in CL compared to YLA samples. In addition, we observed a significantly lower fungal diversity in GLA (P adj = 0.048) and YLA (P adj = 0.03) than in CL samples. Site and Acer species did not influence fungal richness or diversity. The effect of leaf status on fungal community composition differs between the two Acer species.
As we observed differences in the relative abundances of the predominant fungal species and bacterial genera with regard to Acer species and leaf status, we performed non-metric multi-dimensional scaling (nMDS) analyses based on Bray-Curtis and weighted UniFrac dissimilarities for bacteria and Bray-Curtis dissimilarities for fungi ( Supplementary Material Fig. S3). We observed that the majority of control leaf samples grouped together in the fungal but not in the bacterial dataset. The effect of site, leaf status and Acer species Only fungal and bacterial groups with an average abundance of 0.01% and 0.5%, respectively, are shown. Fungal and bacterial groups with an average abundance of less than 0.01% and 0.5%, respectively, are summarized as "others". Note that Allorhizobium-Neorhizobium-Pararhizobium-Rhizobium was shortened for visualization. Abbreviations: CL, control leaves without green islands; YLA, yellow leaf area; GLA, green island area; Site 1, Molkengrund; Site 2, Am Kehr; Site 3, Friedland.
on endophyte community composition was further analysed by permutational multivariate analysis of variance (PERMANOVA). The composition of fungal (Pseudo-F (2) = 3.38, P = 0.001) and bacterial communities (Bray-Curtis: Pseudo-F 2 = 2.55, P = 0.001; weighted UniFrac: Pseudo-F (2) = 2.9, P = 0.01) was significantly affected by site (Table 3, Supplementary Material Table S7). In contrast, Acer species had no significant effect on both bacterial and fungal community composition. Interestingly, green island presence had no effect on the composition of fungal and bacterial endophyte communities. Leaf status affected the composition of the fungal community (Pseudo-F (2) = 5.91, P = 0.04) only. However, multiple comparison based on Bray-Curtis distances revealed only marginally significant differences in fungal community composition between leaf states YLA and GLA (t = 4.11, P = 0.09, permutations = 192).  Table 2. Alpha diversity measures (mean ± standard deviation) for bacterial and fungal endophytes in leaves with and without green islands of A. campestre and A. platanoides collected at at three sites (Molkengrund, Am Kehr and Friedland) in Germany. Abbreviations: CL, control leaves without green islands; GIL, leaves with green islands; YLA, yellow leaf area; GLA, green island area; Ac, Acer campestre; Ap, Acer platanoides.
Alpha diversity values are represented by richness (number of observed zOTUs), diversity (Shannon diversity index H') and phylogenetic diversity (Faith's PD). Note that the Faith's PD was not calculated for fungi due to the high variability in the fungal ITS region. Abbreviation: NA, not available (standard deviation not available due to the low number of replicates). *Leaves with green islands (GIL) contained green (GLA) and yellow leaf area (YLA) samples. A,B Different superscript letters indicate significant differences (P ≤ 0.05) between the two Acer species (bacterial data) or between leaves with and without green islands (fungal data). a,b Different superscript letters indicate significant differences (P ≤ 0.05) between the three leaf states CL, YLA and GLA (fungal data). There was a marginally significant difference in fungal richness between CL and YLA samples of A. campestre (P = 0.053) and between CL and YLA samples of the entire dataset (P = 0.054). For further details, see Supplementary www.nature.com/scientificreports www.nature.com/scientificreports/ We were also interested in a separate analysis of endophyte community composition in leaves of each Acer species. Similar to the results for fungal diversity and richness, we found Acer species-specific responses of the fungal community composition towards leaf status (Fig. 3). GLA and CL samples of A. campestre grouped together with exception of one sample (Fig. 3a,b). On the other hand, no distinct clustering was observed for bacterial communities of A. campestre and A. platanoides (Fig. 3c,d) and for fungal communities of A. platanoides (Fig. 3a,b). These findings were supported by the results of the statistical analysis (PERMANOVA; Table 3). Leaf status had no effect on the community composition of bacterial endophytes in A. campestre and A. platanoides and fungal endophytes in A. platanoides, but significantly affected those of fungal endophytes in A. campestre (Pseudo-F (2) = 6.83, P = 0.02). Multiple comparison based on Bray-Curtis distances revealed significant differences in fungal community composition between leaf states YLA and GLA of A. campestre (t = 4.1, P = 0.048, permutations = 192).
Overall, we found different responses of bacterial and fungal endophytes towards site, presence of green islands, leaf status and Acer species (Table 4). In addition, the observed effects were not consistent among the Acer tree species, as fungal richness, diversity and community structure of A. campestre but not of A. platanoides differed among the leaf types and/or between leaves with and without green islands.
Bacterial and fungal taxa associated with leaf status and Acer species. To identify bacterial and fungal taxa responsible for the observed differences among leaf status and Acer species, we performed an  Table 3. Summary of PERMANOVA results for the analysis of differences in endophyte community composition across the different factors (i.e., site, leaf status, Acer species and their interactions). Data were analysed with site as random effect and Acer species as well as leaf status (model 1) or presence of green islands (model 2) as fixed factors. Each Acer species was also tested separately with site as random and leaf status as fixed factor (models 3 and 4  Table S8). We detected a lower number of significantly associated bacterial (n = 72) than fungal (n = 106) zOTUs. In general, the number of significantly associated fungal and bacterial zOTUs was lower for A. platanoides than for A. campestre. The highest number of associated bacterial (n = 53) and fungal (n = 48) zOTUs was observed for CL samples of A. campestre.
Several zOTUs belonging to the Mycosphaerellaceae were significantly associated with CL samples of both Acer species. Four fungal zOTU of this family were significantly affiliated with leaf status YLA of A. campestre, while no fungal zOTU was associated with YLA samples of A. platanoides.
Six zOTUs belonging to S. polyfida and three zOTUs of S. tulasnei were significantly linked to leaf status GLA of A. campestre. Six zOTUs of S. bicornis, five zOTUs of S. polyfida and one zOTU belonging to S. tulasnei were significantly associated with leaf status GLA of A. platanoides. Three of these zOTUs were linked to leaf status GLA in both Acer species. In addition, we identified two bacterial zOTUs (Pseudomonas lutea and Acidovorax valerianellae), which were significantly associated with YLA samples of A. campestre. The three bacterial zOTUs assigned to Paracoccus sp. K3B-8, Sphingomonas rubra and Escherichia-Shigella were significantly affiliated with YLA samples of A. platanoides. Only two bacterial zOTUs were significantly associated with GLA samples of A. campestre (one unidentified Sphingomonas and one unidentified bacterium of Acetobacteraceae) and A. platanoides (one Sphingomonas sp. Leaf339 and one unidentified bacterium belonging to P3OB-42 of the Myxococcales).

All samples
Site

Discussion
Studies on fungal and bacterial endophytes and their interactions in non-healthy plant tissues are still lacking. In the current study, we assessed fungal and bacterial endophyte communities in leaves with and without green islands of A. campestre and A. platanoides. Consistent with previous research 29,30 , bacterial communities of both Acer species were dominated by Gammaproteobacteria. The predominant fungal classes in our study were Dothideomycetes and Leotiomycetes. Our results are only partly in line with previous studies on foliar fungal endophytes of trees [30][31][32][33] . We assume that the contrasting findings in endophyte communities observed in our and other studies are related to differences in sampling site and plant species investigated. Previous studies showed that sampling site 9,34-36 and plant species 4,6,36 are main drivers of foliar endophyte communities. This is only partly supported by our study. Acer tree species had no effect on fungal and bacterial alpha diversity values (except bacteria's Shannon diversity index). However, we cannot exclude that the differences observed for the bacterial diversity detected between the two Acer species are the result of an indirect sampling site effect as we did not collect A. platanoides leaves at site Friedland.
Neither leaf status nor the presence of green islands influenced bacterial community composition. In contrast, Ren et al. 24 observed differences in the composition of bacterial communities in needles of asymptomatic and symptomatic (Heterobasidion-rotten) Picea abies trees. In another study on bacterial endophyte communities in citrus roots, infection by Candidatus Liberibacter asiaticus had a strong effect on structure and composition of these communities 37 . This is in line with a study of Koskella et al. 38 who showed that tissue type (asymptomatic vs. symptomatic) significantly affected bacterial community composition in the bark of horse chestnut (Aesculus hippocastanum) trees.
Green island presence had no effect on fungal endophyte community composition. Similarly, Abdelfattah et al. 27 found no significant differences in fungal composition in citrus leaves asymptomatic and symptomatic for greasy spot disease. They suggested that the putative pathogen responsible for this disease is present in asymptomatic leaves as latent infections. This conclusion is supported by a recent study on fungal diversity in ash leaves infected by the invasive ascomycete Hymenoscyphus fraxineus 39 . Here, symptoms of necrotic leaf lesions were observed after a long latent phase, although a continuous accumulation of pathogen biomass in ash leaves from the initiation of sporulation throughout the season was detected. Previous studies showed that several fungal endophytes are latent pathogens 40,41 and can switch their life strategy when the host/the plant tissues senesces 42,43 .
In the current study, we did not detect other symptoms or signs of fungal infections such as mildew patches. We hypothesize that such symptoms would be more prominent in senescent leaves.
We found significant differences in the fungal community composition between YLA and GLA samples of A. campestre. In addition, we observed a lower fungal richness and diversity in GLA samples of A. campestre but not of A. platanoides. Interestingly, alpha diversity values were not affected by both presence of green islands and leaf status. Previous studies investigating bacterial and fungal endophyte diversity in diseased and apparently healthy plant tissues found similar contrasting results 9,21,23,27 . For instance, no significant differences in fungal richness and diversity between healthy and dieback-diseased plants of five invasive species across Northern Australia were detected by Raghavendra et al. 44 . Contrarily to our results for bacterial alpha diversity values, Bogas et al. 21 observed a higher bacterial diversity in Paullina cupana plants associated with symptomatic anthracnose (caused by Colletotrichum spp.) compared to asymptomatic plants, indicating interactions between anthracnose and the bacterial endophytic communities investigated.
We speculate that interactions between fungal endophytes are responsible for the observed differences in fungal community composition and lower alpha diversity values in GLA samples. As endophytes are supposed to "guard their turf " from potential usurpers (see 43 ), fungal species occupying plant tissues should block other species from invading these areas. Yan et al. 45 have elegantly shown that fungal endophytes boost their diverse arsenal of secondary compounds aiming to fence their tissues from invasion by other colonizing species. On the other hand, as endophytic taxa have quite different life history traits resulting in functional complementarity, they may occupy even closely related niches within plant tissues 46 .
Another possible explanation for the observed differences between fungal communities in GLA and YLA samples of A. campestre is that the fungal endophytes differ in their niche preferences, as Mycosphaerellaceae and S. polyfida dominated YLA and GLA samples, respectively. This supports a previous study of Ernst et al. 47 on fungal endophytes of Phragmites australis. The authors observed that three unrelated species of the Ascomycota differed in their niche preferences (i.e., host organ, host habitat, and time of season) and concluded that spatio-temporal niche partitioning allowed for differential colonization of common reed by the fungal species investigated.
The high abundance of powdery mildew genus Sawadaea in GLA samples and the results of the indicator species analysis suggests that these fungi could be responsible for the green islands in the Acer leaves. Members belonging to this genus are obligate parasitic fungi of several maple species and have a limited host range 48 . However, other symptoms or signs of fungal infections supporting this conclusion were not detected. As a consequence, future studies over longer time periods are needed to identify the cause of the green islands observed in the current study.
Finally, we showed that fungal and bacterial communities responded in different ways towards the factors investigated, supporting the need for synergy between bacterial and fungal endophyte research efforts 2 . Similar different responses of fungal and bacterial endophytes towards abiotic and biotic factors have been observed in two previous studies 5,7 . The observed different responses in our and the above-mentioned studies might be related to differences in the lifestyle strategies of fungi and bacteria, which have been described previously 1 . Another possible explanation for the more pronounced response of fungal endophytes towards the presence of green islands is that the fungal pathogen competes with other fungi for the same ecological niche 49,50 . Negative associations between co-occurrence of fungal endophyte species were observed for Cirsium arvense by Gange et al. 36 . The authors suggested that these findings are related to the sequence of endophyte colonization. As a consequence, (2019) 9:14183 | https://doi.org/10.1038/s41598-019-50540-2 www.nature.com/scientificreports www.nature.com/scientificreports/ future studies should investigate endophyte communities and their interactions spanning the whole lifetime of leaves (see 39 ).
A few technical and biological limitations of this study must be considered. We cannot exclude the possibility that at least part of the bacteria and fungi detected in our study are not actual endophytes but saprotrophs originating from leaf surfaces or the soil, although we collected freshly fallen leaves only. In addition, more leaf samples collected from different sites over a longer sampling period should be analyzed to better understand the causal agents for the green islands and the multiple interactions of endophyte communities. Future studies should also perform epidemiological analyses to support the conclusion that the Sawadaea spp. are responsible for the green islands observed in the Acer leaves.

conclusion
Studies on fungal and bacterial endophytes and their interactions in non-healthy plant tissues are still scarce. The current study provides first insights into the diversity and community composition of fungal and bacterial endophytes in leaves with and without green islands of two maple tree species. The high abundance of S. polyfida and S. bicornis in GLA samples of A. campestre and A. platanoides, respectively, suggests that these fungi could be responsible for the green islands in the Acer leaves. We cannot, however, exclude that other microorganisms or the host species are responsible for the formation of green islands. Fungal endophyte communities appeared to be more affected by leaf status and/or presence of green islands than bacterial endophytes. However, the effects of green island presence or leaf status on fungal endophytes were Acer species-dependent: only fungal endophytes in leaves of A. campestre were affected by these factors. As a consequence, more research on endophyte communities in leaves with and without green islands of Acer trees collected at different sites over a longer time period as well as competition experiments with S. polyfida and S. bicornis are needed to unravel the diverse plant-microbiome-pathogen interactions in leaves with green islands.

Materials and Methods
Study site and sampling. Leaves with and without green islands (Fig. 1) of at least four A. platanoides and A. campestre trees per site and Acer tree species were collected at two sites in Göttingen (Molkengrund, E 9°57′39″/N 51°32′15″; Am Kehr, E 9°58′46″/N 51°31′36″) at 1 st November 2015 (Table 1). Every single leaf was collected directly under a single tree canopy (number of trees per site and Acer species: ≥4; Table 1). Only those leaves were sampled, which were clearly fallen from the tree in the last hours (i.e., only those leaves, which lay loosely on top of the litter layer). As there were only few A. campestre leaves without green islands, control leaves of approximately ten A. campestre trees were also collected in a forest next to Friedland (E 9°56′17″/N 51°25′13″). Each leaf was separately placed in a sterile bag and transported to the laboratory on ice within a few hours. The leaves were stored at −20 °C.

Surface sterilization of plant material and DNA extraction.
Leaves were surface-sterilized by serial washing in 70% ethanol for 2 min, 2% sodium hypochlorite for 3 min and 70% ethanol for 1 min, followed by two times immersion in sterile, distilled water for 45 s and once in sterile, diethylpyrocarbonate (DEPC)-treated water. Surface-sterilized leaves were subsequently dried on sterile tissue paper. Fresh solutions and separate, sterile collection tubes were used for each leaf to avoid cross-contaminations. The effectiveness of the applied sterilization process was controlled as described previously 51 . Briefly, aliquots of the water used in the final wash step were subjected to PCR targeting the bacterial 16S rRNA gene and the fungal ITS2 region. No PCR products were detected. In addition, water from the same aliquots was plated on common laboratory media plates. No growth of fungi or bacteria was observed after incubation in the dark at 25 °C for at least one week. These results confirmed that the surface sterilization of all Acer leaves was successful in eliminating cultivable as well as non-cultivable epiphytic fungi and bacteria as well as potential DNA traces from the leaf surfaces.
Five to ten fragments of approximately 10 mm diameter were cut of eight (A. platanoides) and nine (A. campestre) leaves without green islands (control, CL) using sterile scalpels. In addition, five to ten leaf fragments of around 10 mm diameter were cut from green islands as well as from the surrounding yellow tissues from twelve A. platanoides and twelve A. campestre leaves. We chose only leaves without any other symptoms of diseases, such as lesions. All fragments of the same leaf were pooled based on their leaf status (CL, YLA and GLA) and stored at −20 °C until DNA extraction. Total DNA of pooled leaf fragments was extracted employing the peqGOLD Plant DNA Mini kit (Peqlab, Erlangen, Germany; now VWR) according to the manufacturer's instructions with two modifications described previously 52 . In addition, all leaf fragments were incubated in lysis buffer at 32 °C for 12 hours and subsequently homogenized using ethanol-sterilized pestles. Concentration of DNA extracts was quantified using a NanoDrop ND-1000 spectrophotometer (NanoDrop Technologies, Wilmington, USA). In total, DNA of 65 Acer leaf samples was subjected to PCR targeting the bacterial 16S rRNA gene and the fungal ITS region (Supplementary Material Table S1).

Amplification of the 16S rRNA gene. Bacterial endophyte communities in A. platanoides and
A. campestre leaves were assessed by a nested PCR approach targeting the 16S rRNA gene. For details of the first PCR reaction mixture and the thermal cycling scheme see Wemheuer et al. 53 . In brief, the primers 799f (5′-AACMGGATTAGATACCCKG-3′) 54 and 1492R (5′-GCYTACCTTGTTACGACTT-3′) 55 were used in the first PCR to suppress co-amplification of chloroplast-derived 16S rRNA genes. Bacteria-specific bands were purified and quantified as described in Wemheuer and Wemheuer 51 . Purified PCR products were subjected to nested PCR.
The V6−V8 region of the 16S rRNA gene was amplified with primers 968F and 1401R 56 containing MiSeq adaptors (underlined) (MiSeq-968F 5′-TCGTCGGCAGCGTCAGATGTGTATAAGAGACAGAA CGCGAAGAACCTTAC-3′; MiSeq-1401R 5′-GTCTCGTGGGCTCGGAGATGTGTATAAGAGACAGCGGT GTGTACAAGACCC-3′) as described previously 51 with one modification: 0.5 U of Phusion high fidelity DNA polymerase (Thermo Scientific, Waltham, MA, USA) was used. Three independent PCRs were performed per sample. Obtained PCR products were controlled for appropriate size, pooled in equal amounts and purified using the peqGOLD Gel Extraction kit (Peqlab). Quantification of the PCR products was performed using the Quant-iT dsDNA HS assay kit and a Qubit fluorometer (Thermo Scientific) as recommended by the manufacturer. Purified PCR products were barcoded using the Nextera XT-Index kit (Illumina, San Diego, USA) and the Kapa HIFI Hot Start polymerase (Kapa Biosystems, Wilmington, USA). The Göttingen Genomics Laboratory determined the sequences of the partial 16S rRNA genes employing the MiSeq Sequencing platform and the MiSeq Reagent Kit v3 (2 × 300 cycles) as recommended by the manufacturer (Illumina).

Amplification of the ITS region. Fungal endophyte communities in leaves of A. platanoides and
A. campestre leaves were assessed by a nested PCR approach targeting the ITS region as described previously 51 . In the first PCR, the primers ITS1-F_KYO2 (5′-TAGAGGAAGTAAAAGTCGTAA-3′) 57 and ITS4 (5′-TCCTCCGCTTATTGATATGC-3′) 58 were used to suppress co-amplification of plant-derived ITS regions. Obtained PCR products were purified and quantified as described for the bacterial PCR products. The ITS2 region was subsequently amplified using approximately 50 ng product of the first PCR and the primers ITS3_KYO2 57 and ITS4 58 containing the MiSeq adaptors (underlined): MiSeq-ITS3_KYO2 (5′-TCGTCGGCAGCGTCAGATGTGTATAAGAGACAGGATGAAGAACGYAGYRAA-3′) and MiSeq-ITS4 (5′-GTCTCGTGGGCTCGGAGATGTGTATAAGAGACAGTCCTCCGCTTATTGATATGC -3′) as described previously 51 . Three independent PCRs were performed per sample. Obtained PCR products were pooled in equal amounts, purified and quantified as described for bacterial PCR products. Purified PCR products were barcoded using the Nextera XT-Index kit (Illumina) and the Kapa HIFI Hot Start polymerase (Kapa Biosystems). The Göttingen Genomics Laboratory determined the sequences of the ITS2 region employing the MiSeq Sequencing platform and the MiSeq Reagent Kit v3 (2 × 300 cycles) as recommended by the manufacturer (Illumina).
processing of bacterial and fungal datasets. Obtained sequencing data were initially quality filtered with the Trimmomatic tool version 0.36 52 . Low quality reads were truncated if the quality dropped below 15 in a sliding window of 4 bp. Subsequently, all reads shorter than 100 bp and orphan reads were removed. Remaining sequences were merged, quality-filtered and further processed with USEARCH version 10.0.240 59 . Filtering included the removal of reads shorter than 400 bp or longer than 470 bp (bacteria) or shorter than 220 bp and longer than 460 bp (fungi) as well as the removal of low-quality reads (expected error >1) and reads with more than one ambiguous base.
Processed sequences of all samples were combined, and redundant sequences were subsequently removed. Obtained unique sequences were denoised and clustered in zOTUs (i.e., sequences with 100% similarity) with the unoise3 algorithm 60 implemented in USEARCH 59 . Chimeric sequences were removed denovo using the UCHIME algorithm during clustering 61 . Subsequently, remaining chimeric sequences were removed using UCHIME 62 in reference mode with the SILVA SSU Ref NR 99 132 database 63 as reference data set for bacteria and the QIIME release of the UNITE database version 7.2 64 for fungi. To assign taxonomy of bacteria and fungi, unique and chimera-free sequences were classified by BLAST alignment 65 against the SILVA 63 and the UNITE 64 database, repectively. Taxonomy was assigned by stand-alone BLAST using the best-hit approach with an e-value cutoff of 1e-20. The taxonomy of the best hit identified during BLAST search is affiliated to the respective zOTU. Combined sequences of all samples were mapped on the final set of unique sequences to calculate the evenness and abundance of each zOTU in all samples. All non-bacterial or non-fungal zOTUs were removed based on their taxonomic classification in the respective database. In addition, all zOTUs consisting of one single sequence (singletons) were removed prior to statistical analysis. Final zOTUs tables for bacteria and fungi are provided as Supplementary Material Tables S2, S3, respectively. Sequence characteristics for bacterial and fungal dataset are  provided as Supplementary Material Tables S4, S5, respectively. Data analysis. All data analyses were conducted in R version 3.5.0 66 or the PRIMER 6.0 software with the PERMANOVA + add-on (PRIMER-E, Plymouth Marine Laboratory, UK). Differences were considered as statistically significant with P ≤ 0.05 and as marginally significant with P ≤ 0.1. Bacterial and fungal communities were analysed separately. Alpha diversity indices (richness, Shannon index of diversity, Faith's phylogenetic diversity, Chao1 and Michaelis-Menten Fit) were calculated using the R-packages vegan 2.5-2 67 , picante version 1.7 61 and drc version 3.0-1 68 . The zOTU tables were rarefied to 8,061 (bacteria) or 11,932 (fungi) sequences per sample prior to alpha diversity analysis using the rrarefy function in vegan 67 .
Diversity was calculated using diversity function in vegan 67 . Sample coverage was estimated using the Michaelis-Menten Fit calculated in R. For this purpose, richness and rarefaction curves were calculated using the specnumber and the rarecurve function, respectively, in the picante package 61 . The Michaelis-Menten Fit was subsequently calculated from generated rarefaction curves using the MM2 model within the drc package 68 . The alignment and the phylogenetic tree necessary for the calculation of bacterial Faith's phylogenetic diversity were constructed in MUSCLE v3.8.425 69 using the neighbour-joining algorithm. All alpha diversity indices were calculated 10 times. The average of all iterations was used for further statistical analyses. Final tables containing bacterial and fungal alpha diversity values are provided as Supplementary Material Tables S4, S5, respectively. Alpha diversity data were tested for normal distribution with shapiro function and for homogeneity of variance with leveneTest function using the R package car version 3.0-3 70 . Because the distribution of microbial diversity, phylogenetic diversity and richness significantly differed from a normal distribution, changes in alpha diversity values were evaluated by linear mixed effects models using the R packages lmerTest version 3.0-1 71 and MuMIn version 1.42.1 72 . As fixed effects, we entered Acer species and leaf status or presence of green islands. We also tested for significant interactions between site and Acer species and between leaf status and Acer species.
The random factor considered in each model was site. The best model was chosen according to lowest Akaike information criterion (AIC). The sample size was relatively small in comparison to the number of estimated parameters. As a consequence, AICc was used for model selection in the R package MuMIn 72 . Visual inspection of residual plots did not reveal any obvious deviations from homoscedasticity or normality.
The final model was calculated using the function lmer provided within lmerTest with restricted maximum likelihood. P-values were obtained by likelihood ratio tests of the full model with the effect in question against the model without the effect in question. Significance levels for fixed factors and their interaction are based on F-values, calculated by a type III analysis of variance with Satterthwaite approximation for degrees of freedom within the R package lmerTest 71 . Statistically and marginally significant results of fixed factors were followed up with Dunn's test for multiple comparisons with Benjamini-Hochberg correction using the R-package FSA 73 . For the random factor site, an ANOVA-like table with likelihood ratio test statistics was generated using the ranova function in the R package lmerTest 71 .
Overall patterns of the entire bacterial and fungal community structure were analysed by permutational multivariate analysis of variance (PERMANOVA; Typ III) with 999 random permutations using the PRIMER 6.0 software with the PERMANOVA + add-on (PRIMER-E, Plymouth Marine Laboratory, UK). The same factors as described for alpha diversity analysis were included in the design. Four different dissimilarity measures were tested for the bacterial dataset: Bray-Curtis, Jaccard, and weighted as well as unweighted UniFrac. For the fungal dataset, only Bray-Curtis and Jaccard were tested due to the high variability of the ITS region which is inappropriate for phylogenetic analyses. The matrices were calculated in R using the vegdist and GUniFrac functions within the R packages vegan and GUniFrac 74 , respectively. The same bacterial tree used for Faith's phylogenetic diversity was used for the calculation of UniFrac distances. Pre-analyses revealed that Bray-Curtis and/or weighted UniFrac dissimilarities displayed a higher environmental sensitivity based on the higher coefficients of determination (data not shown). As a consequence, only results for these distance measures are shown here. Differences in community structure between the three leaf states (CL, YLA and GLA) were tested using the PRIMER 6.0 software. Differences in community structure were visualized using the metaMDS function in the vegan package 67 .
To identify potential indicators as well as highly associated zOTUs for each Acer species with regard to leaf status, multipattern analyses were applied. For this purpose, the multipatt function from the indicspecies package 28 was used. The biserial coefficients (R) with a particular leaf status or Acer species were corrected for unequal sample size using the function r.g 75 . The analysis was performed for each Acer species individually (Supplementary  Material Table S8). To enhance reliability of the indicator analysis, only fungal and bacterial zOTUs found in at least five samples of either A. campestre or A. plantanoides were considered. Associated fungal and bacterial zOTUS of leaf states GLA and YLA were visualized using Cytoscape version 3.6.1 76 .

Data Availability
Sequence data were deposited in the sequence read archive (SRA) of the National Center for Biotechnology Information (NCBI) under accession number SUB4399606 (bacteria) and SUB4399627 (fungi).