Genotypic variation in Norway spruce correlates to fungal communities in vegetative buds

Abstract The taxonomically diverse phyllosphere fungi inhabit leaves of plants. Thus, apart from the fungi's dispersal capacities and environmental factors, the assembly of the phyllosphere community associated with a given host plant depends on factors encoded by the host's genome. The host genetic factors and their influence on the assembly of phyllosphere communities under natural conditions are poorly understood, especially in trees. Recent work indicates that Norway spruce (Picea abies) vegetative buds harbour active fungal communities, but these are hitherto largely uncharacterized. This study combines internal transcribed spacer sequencing of the fungal communities associated with dormant vegetative buds with a genome‐wide association study (GWAS) in 478 unrelated Norway spruce trees. The aim was to detect host loci associated with variation in the fungal communities across the population, and to identify loci correlating with the presence of specific, latent, pathogens. The fungal communities were dominated by known Norway spruce phyllosphere endophytes and pathogens. We identified six quantitative trait loci (QTLs) associated with the relative abundance of the dominating taxa (i.e., top 1% most abundant taxa). Three additional QTLs associated with colonization by the spruce needle cast pathogen Lirula macrospora or the cherry spruce rust (Thekopsora areolata) in asymptomatic tissues were detected. The identification of the nine QTLs shows that the genetic variation in Norway spruce influences the fungal community in dormant buds and that mechanisms underlying the assembly of the communities and the colonization of latent pathogens in trees may be uncovered by combining molecular identification of fungi with GWAS.


| INTRODUC TI ON
At any given time during their life cycle, trees are colonized by a wide range of microbes, including fungi. These fungi may reside both epiphytically, on the surface of the tissue, and endophytically, within the host tissue without causing any visible symptoms (Arnold, Henk, Eells, Lutzoni, & Vilgalys, 2007;Rajala et al., 2013;Rodriguez, White, Arnold, & Redman, 2009;Sieber, 2007). The epi-or endophytic fungi associated with a tree have varying ecological roles, such as being mutualistic symbionts (latent) pathogens or facultative saprotrophs (Rajala, Velmala, Vesala, Smolander, & Pennanen, 2014;Saikkonen, 2007). Pathogenic fungi cause damage to the host that can reduce host growth and fitness, such as by decreasing photosynthetic capacity, and causing premature leaf shed or lesion formation (Hanso & Drenkhan, 2012;Pan et al., 2018). Highly diverse fungal communities have been reported in studies on current-year Norway spruce needles, that is those that flushed during the sampling season (Menkis, Marciulynas, Gedminas, Lynikiene, & Povilaitiene, 2015;Nguyen, Boberg, Ihrmark, Stenström, & Stenlid, 2016). It has been shown that the needle endophyte community in Norway spruce (Picea abies L., Karst) varies among genotypes (Rajala et al., 2013(Rajala et al., , 2014. This would indicate that apart from the presence of fungi with the capacity to colonize the host, environmental and climatic factors (Eusemann et al., 2016;Menkis et al., 2015;Millberg, Boberg, & Stenlid, 2015;Moler & Aho, 2018;Nguyen et al., 2016;Rodriguez et al., 2009), the assembly of the phyllosphere community associated with a given tree will also depend on factors encoded by the host's genotype (Cordier, Robin, Capdevielle, Desprez-Loustau, & Vacher, 2012;Rajala et al., 2013Rajala et al., , 2014. Although it is hypothesized that the plant-inhabiting community is synergistically determined by environmental and host genetic factors (Horton et al., 2014;Rajala et al., 2013;Terhonen, Blumenstein, Kovalchuk, & Asiegbu, 2019), the role of specific host genetic variation under natural conditions is poorly understood. Reports on specific host genetic variation that affects the phyllosphere community composition come from annual plant species and deal primarily with bacterial colonization (Horton et al., 2014;Roman-Reyna et al., 2019;Wallace, Kremling, Kovar, & Buckler, 2018). In a study of 196 accessions of Arabidopsis thaliana, Horton et al. (2014) showed that both the presence/absence and abundance of fungal species in the associated phyllosphere communities are influenced by the Arabidopsis genotype, but for only the more abundant species in the study. Genome wide association study (GWAS) results have implicated a role of both defence responses and cell-wall integrity in the assembly of Arabidopsis phyllosphere fungal communities (Horton et al., 2014). However, hitherto there have been no reports on specific genetic variants affecting phyllosphere fungal community composition in perennial plant species, such as trees. These have a more complex architecture (relative to most annual plants) and are exposed to their changing biotic and abiotic environment over many consecutive growth cycles. These interactions between genetics, environment and time are likely to significantly influence the structure of the phyllosphere community of a host tree.
Conifers dominate the boreal forests in the Northern Hemisphere (Farjon & Page, 1999). Owing to their often large population sizes, outbreeding mating systems and efficient gene flow (wind pollination), conifers are characterized by high levels of heterozygosity and intraspecific diversity, which is reflected in low levels of genetic differentiation between populations (Savolainen, Pyhäjärvi, & Knürr, 2007). Norway spruce is one of the most important conifer species in Europe, both ecologically and economically. Together with Scots pine (Pinus sylvestris L.), it essentially makes up the continuous boreal forests of the continent. The fungal community composition of Norway spruce needles has been reported to change along a latitudinal gradient on a continental scale as well as between individual genotypes of Norway spruce within a stand (Nguyen et al., 2016;Rajala et al., 2013Rajala et al., , 2014, a pattern which may be considered consistent with horizontal transfer of the fungi, and possibly also with the high intraspecific diversity in conifers (Prunier, Verta, & MacKay, 2016).
The development of high-throughput sequencing (HTS) methods paved the way for the generation of the first draft assembly of the Norway spruce genome (Nystedt et al., 2013). The availability of the genome sequence for Norway spruce has opened new possibilities for the development of genetic markers to produce highly resolved genotypes of an individual tree (Vidalis et al., 2018) and to conduct GWAS . Similarly, sequencing of the ITS (internal transcribed spacer) region with HTS methods has allowed mycologists to describe fungal communities and community dynamics in various ecosystems in greater detail than before, advancing functional understanding of various ecological processes and phylogenetic relationships (Clemmensen et al., 2013;Kubartová, Ottosson, Dahlberg, & Stenlid, 2012;Rosling et al., 2011;Seena & Monroy, 2016;Tedersoo et al., 2014;Voříšková & Baldrian, 2012). The combination of ITS sequencing of phyllosphere fungi with the recently available genotyping resources in Norway spruce in an association study may provide insights in to how tree phyllosphere communities assemble, through the identification of specific fungal taxa and with genetic variants associated with general shifts, in the tree phyllosphere communities. For instance, the communities of seemingly healthy needles often include known (e.g., Lirula macrospora or Rhizosphaera kalkhoffii) or suspected (e.g., Phoma herbarum and Sydowia polyspora) needle pathogens (Menkis et al., 2015;Nguyen et al., 2016;Rajala et al., 2013Rajala et al., , 2014. Furthermore, in a recent metatranscriptomics study of Norway spruce tissues, similar frequencies of fungal transcripts were found both in needle and in bud samples (Delhomme et al., 2015), suggesting that vegetative buds as well as needles harbour active fungal communities, but the study provided no insights into the composition of the bud community.
Here we report the results of a study in the perennial conifer Norway spruce that combined ITS sequencing of the fungal communities associated with dormant buds with GWAS of 478 individuals to (a) describe the fungal community associated with dormant buds in a large population of unrelated trees and thus describe the abundance of possible latent pathogen colonizations of asymptomatic tissues, (b) suggest loci in the genome of Norway spruce that correlate with variation in the fungal communities across the studied population, and (c) identify loci in the genome of Norway spruce that correlate with the presence of specific, latent, pathogens.

| Amplification, sequencing and analysis of phyllosphere fungal community
Healthy looking vegetative buds were sampled from 518 trees in the southern Swedish Norway spruce breeding archives, located at Ekebo and Maltesholm. The trees in the breeding archive were planted approximately 35 years ago with 7 m between each tree and 5 m between rows. Buds were collected by hand from exposed branches about 2 m above ground. The collected buds were placed into labelled Ziploc plastic bags. In the field, the samples were stored in sterox boxes filled with cooling blocks. After each day of field work the samples were transferred to -20°C for long-term storage.
Total genomic DNA was extracted from approximately five buds per tree using the Qiagen Plant DNA extraction kit (extraction details as described by Baison et al., 2019).
Each sample was amplified using unique barcode combinations. The gITS7-ITS4 amplicon amplifies the ITS2 region, which provides good species resolution capacity and can be sequenced throughout the entire length with available HTS technologies (Clemmensen et al., 2016;Ihrmark et al., 2012).
Prior to the amplification of the ITS2 region, DNA quantification was performed using the Qubit ds DNA Broad Range Assay Kit (ThermoFischer) and samples were diluted when needed ensuring sample concentration in the range of 1-10 ng/µl. PCR amplifications were done according to Clemmensen et al. (2016 The sequences generated were subjected to quality control and clustering in the scata NGS sequencing pipeline (http://scata.mykop at.slu.se). Quality filtering of the sequences included the removal of short sequences (<200 bp), sequences with low read quality (any base in the sequence that has a PHRED score < 10) and primer dimers; homopolymers were collapsed to 3 bp before clustering.
Sequences that were missing a tag or primer were excluded. The primer and sample tags were then removed from the sequence, but information on the sequence association with the sample was stored as metadata. The sequences were then clustered into different operational taxonomic units (OTUs) that essentially correspond to the species level by single-linkage clustering based on 98.5% similarity. The most common genotype (real read) for clusters was used to represent each taxon. For clusters containing two sequences, a consensus sequence was produced. The fungal taxa were taxonomically identified using the UNITE database version 7.2 (https :// unite.ut.ee/index.php) and the blastn algorithm. The criteria used for identification were as follows: sequence coverage > 80%, similarity to species level 98%-100% and similarity to genus level 94%-97%.
Sequences not matching these criteria were considered unidentified. To obtain further information on abundant yet unidentified sequences (i.e., suspected Norway spruce clusters), a secondary search was performed in GenBank (NCBI) using the blastn algorithm. After removal of singletons and nonfungal sequence reads, the relative abundances of each fungal cluster (OTU) in each sample were calculated. The data set was tested for PCR (polymerase chain recation) contamination by analysing the PCR blanks, and confounding factors such as variation between sequencing libraries, sites and site characteristics. From the original population samples, 473 trees growing at the Maltesholm site were selected for subsequent analysis.
The 1% most abundantly sequenced OTUs were analysed with multivariate ordination methods in Past 3.20 (Hammer, Harper, & Ryan, 2001). To identify the main drivers in the data set, a principal component analysis (PCA) was made and the first six eigenvectors from the PCA were analysed in the subsequent GWAS.
OTUs including more than 2% of the reads, corresponding to known conifer pathogens and with a presence in at least 35% of the Norway spruce samples, were selected for targeted GWAS of latent pathogens ( Figure S3). Separate files with either relative abundance or presence/absence data were prepared for each of the OTUs that met the criteria for GWAS of latent pathogens and were used in trait-association mapping.

| Norway spruce genotyping and SNP annotation
Generation and evaluation of Norway spruce exome capture is described elsewhere (Vidalis et al., 2018). In brief, 478 samples from a subset of 9,000 maternal trees on which sequence capture was performed using 40,018 previously evaluated diploid probes and samples, were sequenced to an average depth of 15×. Illumina sequencing compatible libraries were amplified with 14 cycles of PCR with the probes being hybridized to a pool comprising 500 ng of eight equimolarly combined libraries following Agilent's SureSelect Target Enrichment System (Agilent Technologies) protocol. These enriched libraries were then sequenced on an Illumina HiSeq 2500 using the 2 × 100-bp sequencing mode.
Read mapping and initial variant calling is described in detail by Baison et al. (2019). Basically, the sequence reads were aligned to the Norway spruce genome using the Burrows-Wheeler Aligner (BWA; Li & Durbin, 2010)  Only bi-allelic single nucleotide polymorphisms (SNPs) with a minor allele frequency (MAF) and "missingness" of <0.05 and >20%, respectively, were removed. For the selected set of trees a total of 178,101 SNPs passed the filtering and were used for downstream analysis. Annotation was performed using default parameters of sn-PEff 4 (Cingolani et al., 2012) and local Norway spruce genome annotated database. Ensembl general feature format (GTF, gene sets) information was utilized to build the Picea abies snPEff database.

| Trait association mapping
Loadings on the first six axes from the PCA of the relative abundance data of the 1% largest OTUs were used for the GWAS, with each individual axis explaining at least 5% of the variance in the detected phyllosphere communities. Subsequently, the relative abundance or presence/absence data for the OTUs (OTU_5, OTU_9, OTU_15 and OTU_19) on each host genotype were used for the trait-association mapping.
The statistical LASSO model as described by Li et al. (2014) was applied to the traits associated with the detected phyllosphere communities and phyllosphere pathogens.

The LASSO model is:
where y i is the phenotypic value of an individual i (i = 1, …, n; n is the total number of individuals), α 0 is the population mean parameter, x ij is the genotypic value of individual i and marker j coded as 0, 1 and 2 for three marker genotypes AA, AB and BB, respectively, α j is the effect of marker j (i = 1, …, n; n is the total number of markers), and λ (>0) is a shrinkage tuning parameter. A fundamental idea of LASSO is to utilize the penalty function to shrink the SNP effects toward zero, and only keep a small number of important SNPs which are highly associated with the trait in the model.
The stability selection probability (SSP) of each SNP being selected by the model was applied to determine significant SNPs (Gao et al., 2014;Li & Sillanpää, 2015). For a marker to be declared significant, an SSP inclusion ratio (Frequency) was calculated for all selected SNPs for each trait and a minimum inclusion frequency of 0.56 was chosen as the most prudent cut-off. The set of markers with nonzero effects was recorded; after bootstrapping, this provides an approximation of a p-value. The SSP threshold used for defining significant SNPs was estimated as suggested by Meinshausen and Bühlmann (2010). The threshold is calculated based on the number of SNPs, number of individuals included in the subset and expected number of false positives. A lambda of 250 with 1,000 bootstraps and a false positive cutoff of five (5) were applied to the entire association analysis. Population structure was accounted for in all analyses by including the first five components from a PCA of the genotypic data as covariates in the LASSO model and the total variance explained by the five PCs was 19.3%. Finally, an adaptive LASSO approach (Zou, 2006) was used to determine the percentage of phenotypic variance (PVE; H 2 QT ) of all quantitative trait loci (QTLs). The analyses were all performed with glmnet in Rstudio, R version 3.4.0 (Team, 2015), and the codes used can be found at https ://github. com/Rosar ioGar ciaLa b/Norway-Spruce-Assoc iation-Mapping.
Information on putative candidate genes associated with the QTLs and the expression pattern of the candidate genes in the Norway spruce clone Z4006 were collected from the publicly available Norway spruce genome portal and P. abies exAtlas (https ://www.conge nie.org). The position of the detected QTLs in the Norway spruce genome was estimated by searching an ultradense genetic map (Bernhardsson et al., 2019) for markers derived from the same probes from which the SNP markers holding the QTLs originated.

| Norway spruce buds are colonized by wellknown phyllosphere endophytes and pathogens
After quality control, 676,375 ITS sequence reads remained. At 98.5% similarity, these sequences clustered into 4,899 OTUs, ex-  Cluster size is the number of reads that are associated with the OTU.
c Samples (%) is the frequency of the OTU presence in the 493 samples.
d Thetaxonomic assignment where "Reference" is the taxonomic assignment based on blastn searches in UNITE,"Accession" is the best hit, "score" and "E-value" are the blast score and E-values, and "%" is thepercentage identity in the alignment.
To reduce the number of variables for the subsequent GWAS analysis, the relative abundance data of the OTUs/species with the highest number of reads were used in a PCA to identify the most prominent drivers in the data set. The first two principal components (PCs) explained 14.9% and 12.8% of the total variance respectively in the phyllosphere community. The first and second PCs were strongly influenced by the abundance data of the undescribed ascomycete (OTU_4) and C. cygneicollum, while the second axis was also driven by C. herbarum, S. polyspora and OTU_10 (unclassified fungus; Figure 1a; Table 1). In addition to several of the taxa shaping the first two PCs, F. wieringae appeared to be an important driver on the third, and T. areolata on the fourth (Figure 1b; Table 1). PCs 5 and 6 explained 7.8% and 6.3% of the variance, respectively but the axes were shaped by mostly different OTUs than the first axis, such as A. pullulans, T. areolata, R. kalkhoffii, Ceramothyrium sp. and L. macrospora ( Figure 1c; Table 1).

| Trait association-mapping
Using the loadings on the first six PCs of the PCA (Figure 1) for GWAS, six QTLs associated with the fungal community that determine the PCs were detected (  (Table 3) Table 2). None of these QTLs was shared with the QTLs identified with the 1% most abundant OTUs; these traits were associated on independent SNPs, and appeared to be located at different positions in the genome (Table S4)

| D ISCUSS I ON
Foliar fungi of conifers are commonly predicted to be horizontally transmitted and the phyllosphere community of a given tree would be expected to represent a sample of the fungi present in the environment (Millberg et al., 2015;Rodriguez et al., 2009;Terhonen et al., 2019). In angiosperm tree species, the genetic distance between host tree individuals may have a critical impact on assembly of the foliar fungal community (Ahlholm, Helander, Henriksson, Metzler, & Saikkonen, 2002;Cordier et al., 2012). The impact of the tree genotype is unclear in conifers where some studies report an effect of the host genotype on community composition (Rajala et al., 2013), and others find no such correlation (Eusemann et al., 2016). In this study, combining ITS sequencing and GWAS techniques, we provide further evidence that host genotype influences the composition of the phyllosphere community in the significant marker trait-association mapping of the variation in communities associated with dormant Norway spruce buds.  (Menkis et al., 2015;Nguyen et al., 2016;Rajala et al., 2013Rajala et al., , 2014. The frequencies and abundances of phyllosphere fungi differ depending on the species present in the local environment (Nguyen et al., 2016;Rodriguez et al., 2009), physiological status of the host tree (Menkis et al., 2015;Rajala et al., 2014), tissue sampled and sample handling. The current study is an illustration of this, as the fraction of yeast-like fungi was higher here compared to earlier studies. The presence of a large fraction of yeast species may be due to the use of nonsurface-sterilized explants for DNA extraction, unlike in some previous studies (Nguyen et al., 2016;Rajala et al., 2013). This indicates a need for a similar marker trait association approach with both surface sterilized and nonsterilized explants and perhaps with the amplification and sequencing of the ITS2 region from several single buds from each tree instead of from a pooled sample. However, extraction and community sequencing of nonsurface-sterilized material as in our study, and the study by Menkis et al. (2015), indicates that the Norway spruce phyllosphere may have quite abundant, yet undescribed and possibly specific, epiphytes as illustrated by OTU_4. This unidentified taxon was one of the most common fungi in the phyllosphere community in the current study and the study by Menkis et al. (2015). It dominated on the first two PCA axes together with the basidiomycete yeast Curvibasidium cygneicollum, the common outdoor mould Cladosporium herbarum, to some extent the weak pathogen S. polyspora, and Filobasidium wieringae, suggesting that these axes may be driven primarily by highly frequent and highly abundant phylloplane fungi. It has previously been reported that tree genotypes can influence the occurrence of leaf epiphytes (Bálint et al., 2013;Cordier et al., 2012), but no marker-trait associations were detected with the two first PCs in this study. Possibly a much larger number of trees would be needed to pick up any association with the community on dormant buds. Other PCs were driven more strongly by well-known needle and wood colonizers such as S. polyspora, the biotrophic cherry spruce rust fungus Thekopsora areolata (primarily PC3 and PC4) and L. macrospora (PC5 and PC6).
Several, probable latent pathogens were identified among the most abundant species: T. areolata, R. kalkhoffii, L. macrospora, TA B L E 2 Significant association in the GWA study The trait upon which the marker associates, PC3-PC5 indicate the associations with loadings on the respective PC, and L. macrospora and T. areolata specify associations with the presence/absence data of these fungi among the samples. b The SNP name consists of the contig (MA_number) and SNP position on the contig. For example, the first SNP MA_24477_24501 was located on contig MA_24477 at position 24,501 bp. c Allelic variation associated with the SNP.
d Stability selection probability inclusion ratios for markers declared significant.
e Phenotypic variance explained (only values larger than 1.0% are displayed).

TA B L E 3
Candidate Norway spruce gene models associated with the community composition and pathogen presence Botrytis cinerea, P. herbarum and S. polyspora (Hennon, 1990;Horst, 2013;Kaitera, 2013;Pan et al., 2018;Rajala et al., 2013). With the exception of T. areolata, these species have been reported from Norway spruce phyllosphere communities previously (Menkis et al., 2015;Nguyen et al., 2016;Rajala et al., 2013Rajala et al., , 2014. The latent pathogens were, however, much more frequent in our study, found in up to 82% of the samples. It is possible that a surface sterilization treatment would have reduced the frequency of these taxa, but S. polyspora, L. macrospora, T. areolata and R. kalkhoffii were relatively highly abundant (File S3). Thus, our interpretation is that these four pathogens had colonized asymptomatic buds, acting as latent pathogens.
The interactions between trees and phyllosphere fungi are diverse and not fully understood. It has been reported that endophytic fungi are more diverse and abundant than pathogens within the phyllosphere community (Rodriguez et al., 2009;Terhonen et al., 2019), something that is reflected also in our study. Endophytic fungi in the phyllosphere have the potential to both enhance and reduce tree growth and fitness through various mechanisms (Rodriguez et al., 2009;Terhonen et al., 2019), one of them being the capacity of phyllosphere pathogens to act as both repressors and enablers of disease (Ridout & Newcombe, 2015).  (Park, Seo, & Chua, 2014;Peng et al., 2008) and to regulate plant immune responses (Yaeno & Iba, 2008 (Hejnowicz & Obarska, 1995). The sister chromatid cohesion proteins PDS5s, the Arabidopsis orthologues of MA_104333886g0010 which is associated with PC4, are involved in mitosis, and depletion of this protein lead to severe effects on development, among other traits (Pradillo et al., 2015).
Notably, MA_104333886g0010 is most highly expressed in the early phases of bud flush, which is characterized by high mitotic activity and also by changes in both morphology and starch and tannin distribution (Hejnowicz & Obarska, 1995 (Hannerz, Sonesson, & Ekberg, 1999). The function and expression patterns of each of the identified Norway spruce candidate genes, and perhaps the role of phenology, in the interaction between phyllosphere fungi and Norway spruce, will need to be tested in future experiments.
One of the candidate genes associated with the presence/absence of L. macrospora in the communities, MA_10432519g0010, has similarity to the HEC/Ndc80p family proteins. These proteins are part of the kinetochore complex, which provides an attachment site for spindle fibres in the centromere of chromosomes, and thus are important for cell division (Shin, Jeong, Park, Kim, & Lee, 2018). Disruption of the kinetochore complex may affect, for example, morphogenesis (Du & Dawe, 2007;Lermontova et al., 2011;Shin et al., 2018). It may be difficult to reconcile a potential role of MA_10432519g0010 in development and morphogenesis with control of L. macrospora colonization as disease symptoms are commonly seen on second-year needles (Butin, 1995). However, it has been suggested that L. macrospora infects flushing tissues in the spring with a latency period of about 1 year until symptoms are visible (Hennon, 1990), which would connect the peak transcriptional activity of the candidate gene with the crucial infection phase.
MA_10g0010, which harbours an SNP associated with the presence of T. areolata in the phyllosphere fungal community is a previously undescribed member of the class III peroxidase family.
The gene model shows an expression pattern strongly associated with differentiating and lignifying tissues in the P. abies exAtlas, particularly with young female cones. Clearly the candidate gene is active in tissues susceptible to the pathogen, and basidiospores of T. areolata are thought to infect Norway spruce cones and young shoots in the spring (Hietala, Solheim, & Fossdal, 2007;Kuporevich & Transhel, 1957). Thus, it is not far-fetched to imagine that MA_10g0010 may be associated with processes in these tissues that control their vulnerability to colonization, possibly through cell-wall enforcement (Elfstrand et al., 2001;Fagerstedt, Kukkola, Koistinen, Takahashi, & Marjamaa, 2010;Kärkönen, Warinowski, Teeri, Simola, & Fry, 2009;Marjamaa et al., 2006) or processes controlling the timing of bud break and bud flush; overexpression of the class III peroxidase SPI2 in Norway spruce plants led to significant delays of these process compared to in wild-type Norway spruce plants (Clapham, Häggman, Elfstrand, Aronen, & Arnold, 2003;Elfstrand et al., 2001).
Combining molecular identification and barcoding of phyllosphere fungi with GWAS provides insight into host-encoded factors affecting the assembly of phyllosphere communities and the colonization of needle pathogens. Taken together, GWAS results suggest that processes in the morphogenesis and flush of the Norway spruce shoot exert a strong influence on the dominant players in the phyllosphere community detected in dormant buds.

ACK N OWLED G EM ENTS
Financial support was received from the Swedish Foundation for Garcia@slu.se for genotype information