Habitat conditions and phenological tree traits overrule the influence of tree genotype in the needle mycobiome–Picea glauca system at an arctic treeline ecotone
Summary
- Plant-associated mycobiomes in extreme habitats are understudied and poorly understood.
- We analysed Illumina-generated ITS1 sequences from the needle mycobiome of white spruce (Picea glauca) at the northern treeline in Alaska (USA). Sequences were obtained from the same DNA that was used for tree genotyping. In the present study, fungal metabarcoding and tree microsatellite data were compared for the first time.
- In general, neighbouring trees shared more fungal taxa with each other than trees growing in further distance. Mycobiomes correlated strongly with phenological host traits and local habitat characteristics contrasting a dense forest stand with an open treeline site. Genetic similarity between trees did not influence fungal composition and no significant correlation existed between needle mycobiome and tree genotype.
- Our results suggest the pronounced influence of local habitat conditions and phenotypic tree traits on needle-inhabiting fungi. By contrast, the tree genetic identity cannot be benchmarked as a dominant driver for needle-inhabiting mycobiomes, at least not for white spruce in this extreme environment.
Introduction
Among the various groups of plant-associated fungi, leaf-inhabiting fungal phytobiomes (mycobiomes) comprise an inconspicuous, yet highly diverse and dynamic component in almost all terrestrial ecosystems. These leaf and needle mycobiomes associate intimately with their plant hosts, and therefore have been studied intensively (Kennedy & Stajich, 2015; Vandenkoornhuyse et al., 2015). Several studies have shown high observed and expected density and diversity of such fungi in tropical and temperate forests (Jumpponen & Jones, 2009; Unterseher, 2011; Cordier et al., 2012a; Gazis & Chaverri, 2015), deserts (Sun et al., 2012; Unterseher et al., 2012), boreal and arctic environments (Higgins et al., 2007).
Most of our knowledge of the richness and composition of phyllosphere mycobiomes is based on pure-culturing of endophytes (Sieber, 2007; Gazis & Chaverri, 2015). Such studies have several drawbacks including the more frequent detection of fast-growing species, high time and labour costs, as well as the exclusion of biotrophic species. Combined, these culturing biases often result in an incomplete view of the resident communities (Unterseher, 2011). Fast-growing needle-inhabiting fungi have occasionally been interpreted as contaminants and discarded (Stefani & Bérubé, 2006), whereas later studies using high-throughput sequencing (HTS) have identified such fungi, for example members of the Zygomycota, as an active and abundant component of the needle mycobiome (Peršoh, 2013). A major advantage of HTS methods is the deep sampling and thus the ability to detect rare and uncultivable taxa (Rajala et al., 2014). So far, most available fungal HTS data have been generated with the disappearing 454 pyrosequencing technology (Toju et al., 2013; Blaalid et al., 2014; Tedersoo et al., 2014). Illumina platforms can provide unmatched sequencing depth (Smith & Peay, 2014), but read length is still limited to 2 × 300 bp in paired-end sequencing. Given that, on average, reverse Illumina reads are of weaker quality than forward reads and a comparable large overlap is needed for reliable contig making, Illumina data allow analysis of either the ITS1 or ITS2 regions of the ribosomal RNA gene repeats (Lindahl et al., 2013; Schmidt et al., 2013; Siddique & Unterseher, 2016).
Over the past few years, many abiotic and biotic factors influencing plant-associated mycobiomes have been observed with HTS data. The effect of changing climate is intensively debated (reviewed in Vandenkoornhuyse et al., 2015). Hagedorn et al. (2013) showed that long-term CO2 enrichment of alpine treeline soils had minor effects on the fungal biomass and the composition of fungal communities. By contrast, Streit et al. (2014) observed that warming altered microbial metabolic activity in treeline soils. Basic host attributes, such as species identity or tissue type act continuously on endophytes and other plant-associated fungi, leading to comparatively stable fungal core communities in the long term (Rajala et al., 2013; Bálint et al., 2015). By contrast, temporally variable parameters such as plant secondary metabolites, seasonality, environmental contaminants or multitrophic interactions may also have significant impact on the mycobiomes and may correlate with a generally high fungal community turnover, in particular of the rare or subdominant taxa (Korkama et al., 2006; Korkama-Rajala et al., 2008).
Alaska is among the regions where temperatures have been rising faster than anywhere else during the last decades (ACIA, 2004). There is also growing scientific interest in identifying responses of fungal species and communities under such pronounced climate warming (Jumpponen & Brown, 2014; Bálint et al., 2015).
Here, we examined the needle mycobiome colonizing white spruce (Picea glauca) at the Alaskan northern treeline, the border between boreal forest and arctic tundra. We combined these investigations with ongoing research that focuses on assessing tree responses to climate change (e.g. Ohse et al., 2012). In the course of this project, trees were spatially mapped, measured and cored to document spatial and demographic data of the trees and to obtain information about their life history dynamics. Moreover, the trees were genotyped using microsatellites to reconstruct gene flow, reproduction and population dynamics under treeline conditions (see Eusemann et al., 2014, for methodology). A subset of those trees was analysed and the respective microsatellite data were included in the present study for mycobiome diversity analyses.
We hypothesized that mycobiome diversity and composition correlate with local environmental conditions. Specifically, we posited that sheltered, less exposed conditions would increase the diversity of the mycobiome and that community composition would reflect the increasing climatic harshness from a closed forest stand to the scattered isolated trees at the treeline. In addition, we evaluated the effects of host tree genotype on the mycobiome, an effect suggested in recent publications (Cordier et al., 2012b; Bálint et al., 2015).
Materials and Methods
Sampling sites and fieldwork
The study site was established in the Nutirwik Creek area in the Brooks Range of northern Alaska (67°56′N, 149°44′W, Fig. 1) at the northern treeline. Here, Picea glauca (Moench) Voss occurs at the northern margin of its range at a mean altitude of 900 m above sea level (asl) on south-facing slopes as the sole tree species undisturbed by direct human activities. An elevation gradient from 875 to 950 m asl within two neighbouring sites on a south-facing slope was sampled. Four environmental parameters and five phenological tree traits were recorded as described later (identifiers written in italicized text). Habitat type was defined as a cut-off between high-density and low-density stands. This bipartition can be viewed as a simplification of the real conditions, which show a gradual decrease of tree density uphill instead of a sharp dividing line between the two categories ‘forest’ and ‘treeline’ (Fig. 1). The exact position of 872 trees was recorded using an R3 differential GPS device (Trimble, Sunnyvale, CA, USA) with a resultant mean precision of 0.48 m in floating mode (Fig. 1), providing also data on elevation, which were then used to infer slope (inclination). For tree traits we measured height and stem diameter at breast height (DBH) using a Suunto PM-5 clinometer (Suunto, Finland) and a measuring tape. The total crown volume was estimated from lower crown diameter, height of the first living branches and tree height. Trees with a DBH exceeding 7 cm were cored for dendrochronological analysis (n > 450). Tree cores were mounted on wooden trays, sanded or cut using a microtome until cellular structures became visible, and measured for tree ring width using high-resolution scans of the cores and the program CDendro 7.8 (Cybis, Sweden). This allowed us to derive age and average tree ring width for each cored tree.
Asymptomatic needles of varying age and from three to five positions in the tree crown were sampled and immediately dried on silica gel for storage at ambient temperature until further processing. All habitat characteristics and phenological tree traits are available in Supporting Information Table S1.
DNA extraction and tree genotyping
Dry needles were powdered in a Retsch ball mill MM301 (Retsch, Germany). Approximately 70 mg of needle tissue was used for DNA extraction with the Invisorb Spin Plant Mini Kit (Stratec, Birkenfeld, Germany) following the manufacturer's protocol. DNA concentration was measured with NanoDrop Lite (Thermo Fisher Scientific, Waltham, MA, USA), adjusted to 5 ng μl−1 and used as template DNA for microsatellite analysis in three different multiplex reactions (Eusemann et al., 2014). As measures of genetic diversity and differentiation, as well as relatedness, we estimated the individual heterozygosity of all trees, genetic distance between all trees, and the fixation and differentiation indices FST and Dest between the two study sites. Further analysis of the genetic structure was performed using principal coordinates analysis (PCoA) and analysis of molecular variance (AMOVA). All population genetic statistics were performed using GenAlEx 6 (Peakall & Smouse, 2006). Genotype data are available in Table S1.
Preparation of fungal ITS amplicon libraries for Illumina sequencing
For the present fungal biodiversity analysis, 48 trees were selected randomly with the sole requirement of equal numbers of forest and treeline trees (Fig. 1c).
In the first tagged PCR (PCRTA) with custom-made primers (Siddique & Unterseher, 2016), the fungal ITS rDNA was amplified from the same genomic DNA used for tree genotyping. The primer pairs for PCRTA consisted of the commonly used ITS1F and ITS4 sequences concatenated with multiplexing barcode tags of varying length and half of the Illumina sequencing primer. The latter region served as binding site for the primer pairs of the second indexed library PCR (PCRIN), which contained further barcodes for multiplexing purposes and the Illumina-specific anchor sequences (Siddique & Unterseher, 2016).
The final amplicon libraries consisted of the full ITS1-5.8S-ITS2 rDNA in the centre, flanked by the ITS1F/ITS4 primer sequences, a first pair of sample-specific barcode sequences (here termed tags), the forward and reverse Illumina-sequencing primer (SP1/SP2), a second pair of sample-specific barcode sequences (here termed indices), and the terminating Illumina adapter sequences P5 (forward) and P7 (reverse). Thus, all reads from the same sample were defined by an unique combination of four barcodes; that is, forward index (five variants coded 501–505), forward tag (four variants, F1–F4), reverse tag (four variants, R1–R4) and reverse index (five variants, 701–705). Amplicons were pooled in equal DNA quantities, purified with Dynabeads Sequencing Clean Up Kit (Thermo Fisher Scientific) and transferred to the Genetics Sections, Biocenter of the LMU Munich, Germany. They were prepared for sequencing according to Siddique & Unterseher (2016).
In addition to the original samples, we retained four samples as negative controls, these showing no sign of positive amplification (no gel bands, no reproducible NanoDrop measurements). Further controls contained mixtures of known operational taxonomic unit (OTU) composition and abundance (measured as relative concentration of DNA from axenic fungal cultures). See Methods S1 for analysis and interpretation of those mock communities.
Analysis of Illumina sequences
Raw forward (R1) and reverse (R2) reads were demultiplexed by the Illumina sequencer software according to their index combinations (501-701, 501-702, etc.) and provided as FASTQ files that had Illumina adapters, indices and sequencing primers already removed. Reads in all fastq files thus started with forward and reverse tags, respectively.
Our bioinformatics workflow contained stringent quality filtering at phred scores of ≥ 30 for at least 75% of the read length and three consecutive low-quality base calls allowed before a read was truncated. A second demultiplexing step was applied to separate all tag combinations, followed by chimera checking, ITS trimming and the separation of ITS1 and ITS2 sequence data. Analyses of read quality with FastQC (freely available from http://www.bioinformatics.babraham.ac.uk/projects/fastqc/; accessed March 2016) showed overall lower base calls for R2 than for R1 reads. Consequently, a much higher number of R1 reads (covering ITS1) than R2 reads (ITS2) was retained after quality filtering. In addition, read quality of both R1 and R2 reads was too low in the 5.8S region and did not allow successful contig assembly of read pairs. We continued our workflow with ITS1 reads only, because a detailed comparison of ITS1- and ITS2-derived community data was beyond the scope of this study and would have inflated the methodic aspect of this paper. This issue is addressed only briefly in the Discussion section.
The ITS extractor (Bengtsson-Palme et al., 2013) was used for removing the conserved flanking regions and grouping of OTUs was done with Usearch (centroid-based complete-linkage clustering, Edgar, 2010) at 97% sequence similarity. All steps were accomplished using Qiime as the main analytical backbone (Navas-Molina et al., 2013). The Fastx toolkit (fastx_renamer, available at http://hannonlab.cshl.edu/fastx_toolkit/download.html, last accessed January 2016) and a Perl script to rename Fasta headers (Bálint et al., 2014) were used as small helper application. Two reference datasets for chimera checking (Nilsson et al., 2015) and taxon assignment (Kõljalg et al., 2013) were also used. All steps and commands are available in Methods S2. The raw Illumina reads are provided under the NCBI SRA accession SRX1287546.
A spreadsheet master data file was then generated from two major output files (the final OTU table and the representative sequence Fasta file; see Notes S1). These data were conservatively curated by removing all tentative nonfungal OTUs (those returning ‘no blast match’ after automatic taxon assignment), unique OTUs (OTUs occurring in only one sample regardless of the number of reads) and OTUs with < 5 reads over all samples (cf. Brown et al., 2015). This ITS1 dataset was used for the biodiversity analyses.
Mycobiome biodiversity
The entire curated dataset was used for the analysis of community composition, although there are persuasive arguments to focus on only the core assemblage, because a few core OTUs are generally thought to contribute disproportionately to community dynamics and ecosystem functions (Gibson et al., 1999). However, a thorough consideration of the various interpretations of core–satellite species hypothesis (Hanski, 1982; Magurran & Henderson, 2003) is beyond the scope of this paper. Furthermore, we feel that this would add an excessive theoretical component to this study. For application of the core–satellite hypothesis to molecular data, please refer to Galand et al. (2009), Unterseher et al. (2011) or Bálint et al. (2015).
Six diversity indicators – Fisher's alpha (richness), Shannon index (a combined index of richness and evenness) and the first four Hill numbers from Hill's series of diversity (Hill, 1973) – were used for the diversity analysis of the needle mycobiomes. We applied two different approaches to determine fungal habitat specialists. First, we evaluated significant linear models of single OTUs (at P < 0.05) to search for taxa whose abundance correlated with the continuous elevation gradient as a whole (e.g. high read abundance at low elevation, medium read abundance at medium elevations and low read abundance at high elevation). Second, we used Legendre's indval procedure (Dufrêne & Legendre, 1997) to determine ‘indicator’ OTUs for each of the variable classes ‘forest’ and ‘treeline’. In other words, the indicator OTUs were defined separately for these two habitats. Thus, forest indicator OTUs would have a significantly higher abundance in all forest trees compared with all treeline trees. Results from these two approaches were merged, because an OTU responding significantly to the elevation gradient will be detected with the GLM method, but not with the indval analysis. Neither indicator analysis may discriminate sufficiently between abundant and rare OTUs. For example, on the one hand, a rare OTU with 10 reads is represented as an indicator with the highest possible confidence, if all 10 reads were recovered from a single variable class (e.g. all from forest trees). On the other, an abundant OTU with 1000 reads will be less significant, if 900 reads were recovered from one variable class and 100 from another. To avoid random effects caused by rare key OTUs, only those with > 500 reads were considered as ‘true’ key players. Those key OTUs were manually assigned to known taxa using Insd (Nakamura et al., 2013) and Unite (Kõljalg et al., 2013), followed by assessment of their tentative biology and ecology.
Community composition was assessed by applying nonmetric multidimensional scaling (NMDS) and PCoA based on Bray–Curtis dissimilarities of square-root transformed read abundances. The distinctiveness of fungal assemblages in different subdatasets (i.e. forest vs treeline) was tested using a permutational multivariate analysis of variance (PERMANOVA/adonis; Oksanen, 2015) using the same distance metrics. In addition, we used generalized linear models (GLM) to test for significance of environmental responses to the multivariate community data (so-called multivariate or multispecies GLMs) (Wang et al., 2012; Bálint et al., 2015). The analyses of diversity and community composition were repeated using a rarefied dataset downsampled to the lowest number of reads per sample.
The entire biodiversity analysis was performed in R. All corresponding commands and all necessary input data files are available in Notes S2.
Results
Analysis of phenological tree traits
Phenological tree traits differed significantly between the forest and treeline samples (Fig. 2a; PERMANOVA; F = 3.29, R2 = 0.084, P = 0.037, df = 1). Forward model selection revealed tree height and age as the two most important and significant variables explaining this division (height: R2 = 0.13, F = 6.66, P = 0.013; age: R2 = 0.15, F = 6.56, P = 0.025). The treeline trees were, on average, 10 yr younger than forest trees (120.1 vs 129.7 yr; t-test: t = 2.45, P = 0.021, df = 27) and 1.6 m shorter (7.2 vs 8.8 m; t-test: t = 2.08, P = 0.045, df = 35) (Fig. 2b,c).
Analysis of tree genotypes
PCoA of the genetic distances resulted in weak but significant separation of the two habitat types (PERMANOVA; F = 2.12, R2 = 0.056, P = 0.003, df = 1). Again, elevation was the most important variable (Fig. 3a; PERMANOVA; F = 1.61, R2 = 0.043, P = 0.029, df = 1). The population-wide indexes Dest and Fst, although comparatively low in value, also significantly distinguished genetic differentiation between forest and treeline sites (Dest = 0.100, Fst = 0.033, both at P = 0.002). Correlation tests of tree traits (phenotypic characters) and shared allele distances (tree genotypes) were statistically insignificant (Fig. 3b).
The mycobiome – basic sequencing results
The resulting sequence dataset contained 2009 OTUs from 399 219 quality-filtered ITS1 reads and 44 samples (Table S1). The four negative control samples yielded very low numbers of reads and were removed from the dataset. Due to the interoperability with the genotype dataset, another six samples were discarded. The removal of rare and unassigned OTUs according to the criteria described earlier reduced the dataset to 450 OTUs (a 78% decrease in OTU richness) and 387 451 ITS1 reads (a 3% decrease).
The average sequencing depth (reads per sample) was 8806 and ranged from 3350 to 15 900. Sequencing depth did not differ significantly between the habitat groups ‘forest’ (mean read number per sample = 8927) and ‘treeline’ (mean read number per sample = 8685; t-test: t = −0.27, P = 0.795) (Fig. S1). Among the 31 identified orders, the five most dominant were Capnodiales (53 OTUs), Pleosporales (48 OTUs), Dothideales (25 OTUs), Helotiales (18 OTUs) and Tremellales (13 OTUs). Of all identified OTUs, 86% belonged to Ascomycota and the remaining 14% were assigned to Basidiomycota. Analysis of mock samples showed that variation (noise) increased in the molecular community compared with the original setting, but that the original relations among samples were preserved after bioinformatics, data curation and transformation of the ITS1 reads (Methods S1).
Mycobiome diversity and composition in relation to site characteristics, phenological tree traits and tree genotype structure
The six diversity indexes did not differ significantly between the forest and treeline group, yet the needle mycobiome of forest trees possessed consistently higher diversity compared with the treeline trees (Fig. S2).
Sixty-six OTUs were identified initially to respond significantly to the habitat categories ‘forest’ and ‘treeline’. The indval approach identified 24 ‘specialists’ for forest and 29 for treeline. Among them, six OTUs had > 500 reads and were taxonomically annotated to their closest available relatives (Table 1; Fig. S3).
| Preferred isolation source | OTU ID | No. of reads | Blast annotationa | Unite SH | Unite SH annotaion |
|---|---|---|---|---|---|
| Forest site | 40 | 888 | Meristemomyces, Xenophacidiella | SH200725.07FU | Mycosphaerellaceae |
| 51 | 788 | Gyoerffiella, Lemonniera, Articulospora | SH184432.07FU | Gyoerffyella | |
| 154 | 672 | Venturia | SH219604.07FU | Venturia | |
| 93 | 598 | Constantinomyces, Catenulostroma | SH027297.07FU | Capnodiales | |
| Treeline site | 1 | 30 648 | Perusta, Aureobasidium | SH206394.07FU | Dothideomycetes |
| 36 | 944 | Pseudocercospora, Zasmidium, Xenomycosphaerella | SH194525.07FU | Zasmidium |
- a Blast search limited to sequences from type material. OTU, operational taxonomic unit.
Two forest indicator OTUs (40 and 93) belonged to the order Capnodiales (Ascomycota). According to the taxonomic assignment, OTU 40 had an affinity to the family Mycosphaerellaceae (SH200725.07FU), whereas OTU 93 was not assigned further to a family, genus or species (SH027297.07FU). Blast search of corresponding ITS1 reads limited to sequences from type material, gave closest matches to the genera Meristemomyces (93% sequence similarity) and Xenophacidiella (92% similarity) for OTU 40, and to Constantinomyces (92% similarity) and Catenulostroma (90% similarity) for OTU 93. When excluding uncultured and environmental sequences only, high similarity with sequences from North American isolates became apparent for both OTUs (e.g. JQ759473 or HQ535864, FJ554069) Two further forest specialists were assigned to the genera Venturia (OTU 154) and Gyoerffiella (OTU 51) according to Unite's species hypothesis concept (Table 1). The two remaining OTUs significantly preferred treeline over forest trees (Fig. S3). Blast search against sequences from type specimens revealed a placement within Dothideomycetes (OTU 1, SH206394.07FU) with affinities to the genera Perusta, Celosporium and Rhizosphaera. The second treeline specialist (OTU 36) was assigned to the Mycosphaerellaceae (SH194525.07FU) with 97% similarity to a type sequence of Zasmidium (KC677913).
Mycobiome composition was significantly and positively correlated with the position of the sampled trees: the closer that two trees were to each other, the more similar were their needle mycobiomes (Fig. 4a). This interdependency was due mainly to the latitude (N–S) gradient (R2 = 0.109, P = 0.001) but not to the longitude (E–W) gradient of the trees (R2 = 0.038, P = 0.107). By contrast, mycobiome composition was not correlated with tree genotypes (Fig. 4b).
The pronounced effect of tree position and site parameters on needle fungi is further reflected in the results of multivariate analysis (Fig. 5). The treeline samples, displayed as brown squares and the forest samples, shown as green circles, hosted distinct fungal assemblages in both NMDS and PCoA (only PCoA is shown in Fig. 5). These mycobiome communities were correlated with elevation, latitude (northing), slope and tree age.
Discussion
Local habitat parameters and phenotypic tree traits influence the needle mycobiome
Based on the present data we confirm our hypothesis of a clear influence of local abiotic conditions on the needle mycobiome, but we cannot pinpoint a single major driver that is decisive for the observed fungal diversity signals. Similar interplay of a multitude of biotic and abiotic factors has been shown in recent studies of plant mycobiomes from different environments (Peay et al., 2010; Bálint et al., 2015; Matulich et al., 2015). In contrast to studies which were conducted at the scale of entire landscapes (see also Edman et al., 2004; Blaalid et al., 2014), samples analysed in this study were taken in close proximity to each other, with a maximum distance among the trees of 118 m E–W and 177 m N–S and over an elevational range of 75 m (875–950 m asl; gradient running N–S). On the one hand, we observed during several visits to this area that wind speed, desiccation rates and other small-scale climatic parameters (temperature, humidity, dew point, light intensity, soil melting) differ between and within both habitats, thus likely influencing fungal colonization and establishment (Nordén & Larsson, 2000; Hallenberg & Küffer, 2001). On the other hand, spore dispersal barriers are hardly plausible within this small investigation site (Geml et al., 2012; Blaalid et al., 2014). In addition, elevation (equalling a gradient in climatic harshness) and slope (possibly a proxy for water availability due to different runoff rates during snow melt) was significantly correlated with mycobiome diversity and composition.
In addition to the importance of local site parameters in shaping needle mycobiomes, various studies have confirmed a strong interplay of phyllosphere fungi with their host plants and significant correlations with phenotypic and physiological plant traits (Kembel & Mueller, 2014; Rajala et al., 2014; Tian et al., 2014). Similarly, our analyses revealed a significant correlation of fungal communities with the combined phenotypic differences of the trees and with tree age and height as strong single determinants. Treeline trees were, on average, younger and smaller than those in the forest, and appeared more healthy as they contained lower amounts of dead and broken branches in their tree crowns (pers. obs.). In an earlier study, Vitasse et al. (2009) identified a significant sensitivity of leaf phenology to temperature, whereas a few years earlier Niinemets (2002) observed clear effects of tree age and height on photosynthesis rates in Picea abies and Pinus sylvestris. The latter study also revealed correlations between needle functions and both tree size and age, which are generally attributed to greater water limitations and different conductivity in taller and older trees. The effect of needle age and seasonality on photosynthesis rates of mature black spruce trees (Picea mariana) was confirmed in a recent study (Jensen et al., 2015). Our own ongoing isotope analyses suggest a general strong influence of water limitations on tree responses at the Brooks Range treeline ecotone (unpublished data; but see Gutschick & BassiriRad, 2003).
It should be mentioned briefly that preparation of amplicon libraries, bioinformatic sequence processing and biodiversity statistics are not always free of biases (Gihring et al., 2012; Warton et al., 2012; Salter et al., 2014). The current workflow therefore consisted of several quality filtering steps (Siddique & Unterseher, 2016) and negative controls, the analysis of a mock community, and a stringent removal of unassigned and rare OTUs in order to prepare a dataset of maximal reliability for subsequent biodiversity analyses. Recent manual inspection of another dataset stemming from the same Illumina run indicated a large number of artificial OTUs among the rare and unidentified ones (69.2% with erratic Blast results), whereas all other OTUs were trustworthily identified as existing fungi (only 3.8% with erratic Blast results, data not shown).
Habitat specialists
The most abundant OTU of the entire dataset (OTU1, Table 1) was also one of those OTUs that was a significant indicator of habitat type. It was detected predominantly in treeline trees and showed high sequence similarities to endophytes from other conifers in Northern America (Sokolski et al., 2007; Larkin et al., 2012) with uncertain position within the order Dothideales.
Two additional abundant OTUs belonged to the same family (Mycosphaerellaceae). Leaf-inhabiting endophytes assigned to Mycosphaerella were also identified during a study of leaf mycobiomes of Populus balsamifera from Alaska, USA (Bálint et al., 2015). In their study, this genus was the most abundant in southern populations and decreased after relocation of the plants to higher latitudes. Bálint et al. (2015) interpreted their results as an indication of pathogen release after host relocation, but also issued the problems of assigning functionality based on sequence data alone in this diverse genus (Videira et al., 2015). Furthermore, the observed changes in mycobiome composition were caused by human-assisted rapid relocation (Bálint et al., 2015) and probably cannot be compared easily with mycobiome dynamics caused by natural slow range shifts of plants as a response of climate warming (Chen et al., 2011).
Tree genotype had no measurable effect on the needle mycobiome
There is a growing body of literature suggesting a marked influence of genotypic identity of the host plant on phyllosphere micro- and mycobiomes. In their review, Whipps et al. (2008) suggested ‘that within plant species, genotype has a key role in determining colonization and establishment of microbial communities in the phyllosphere’. However, studies cited in Whipps et al. (2008) were partly contradictory. More recently, this view was picked up again to explain observed differences in fungal phyllosphere communities of conspecific hosts (e.g. for needle/leaf endophytes: Cordier et al., 2012b; Rajala et al., 2013; Bálint et al., 2013, 2015; for mycorrhizal fungi: Velmala et al., 2013). Cordier et al. (2012b) found a significant correlation between the genetic distance of beech trees (Fagus sylvatica, based on eight microsatellite loci from nine trees) and the corresponding leaf mycobiomes. Bálint et al. (2015) found that leaf mycobiomes still correlated with population structure 2 yr after relocation of genetically distinct subpopulations of balsam poplar (P. balsamifera) in a common garden experiment. However, no effect was found on the core community. Only rare OTUs showed a genotype specific response (Bálint et al., 2013).
In contrast to earlier studies, we used the same DNA extracts for both microsatellite analysis of white spruce and fungal high-throughput sequencing. On the one hand, we found a division of the two habitat types for both tree genotypes and needle mycobiomes. There was a weak but significant genotypic separation of forest and treeline spruce populations as well as distinct forest and treeline mycobiomes. On the other hand, direct comparison of mycobiome composition and genetic distance between trees as performed by Cordier et al. (2012b) clearly failed in identifying significant correlations. Insufficient statistical support was also observed for correlation of individual heterozygosity of the trees and mycobiome composition.
Critical inquiry is certainly advisable before signals in mycobiome community ecology can be explained convincingly via host plant genotypes. Additionally, fingerprinting data cannot be used alone to explain leaf physiology and production of secondary metabolites (Bailey et al., 2005). Plant biochemistry is founded on the organism's genetic potential, but its implementation is to a large extent related to phenotypic plasticity as a response to changing environmental situations (Kramer, 1995; Trewavas, 2005; Valladares et al., 2014) or microbial colonization (Tian et al., 2014).
The genotypic and age differences between forest and treeline trees observed in this study and in ongoing parentage analyses within these sites (P. Eusemann & M. Schnittler, unpublished) are in accordance with the scenario of a comparatively young treeline moving uphill. The differences observed here (lower number of effective alleles, lower values for expected and observed heterozygosity) can be explained by preferential uphill seed migration from lower seed sources, thereby realising the concept of a sink population, with an even further reduced gene pool caused by the skewed proportion of potentially reproducing adult trees to actually reproducing ones (ongoing analysis, unpublished data). Given the small geographical distance and the only marginal genetic differentiation between the two sites, it seems plausible to interpret this observation as a result of the reproduction dynamics of an actively advancing treeline rather than a sign of restrictions in gene flow between the two sites.
Based on the available data, we postulate that tree phenology and habitat/microhabitat conditions (structural, physiological and biochemical differences) have profound consequences for the needle mycobiome of P. glauca in such extreme environments. Conversely, the importance of tree genotype as a defining driver of this particular phytobiome has to be questioned.
Conclusions
Montane habitats such as that in our study area are usually controlled by two primary factors: cold winter temperatures and water balance (Neilson, 1993). Because many plant species are already at the extreme margin of their environmental tolerance in such habitats (Gosz & Sharpe, 1989; Risser, 1995), they are expected to react strongest to the anticipated global climate change resulting from global warming (Holtmeier & Broll, 2005; Tinner & Kaltenrieder, 2005).
The effect of changing climate on plant-associated mycobiomes and entire phytobiomes has been debated intensively (reviewed in Vandenkoornhuyse et al., 2015), with partly contradicting results and predictions, depending on the targeted variables (CO2, Hagedorn et al., 2013; temperature, Streit et al., 2014; snow, Barbeito et al., 2013), organisms (microbes, Deslippe et al., 2012; plant, Elmendorf et al., 2012; protist, Geisen et al., 2014) and micro-environments (soil, Deslippe et al., 2012; leaf and roots, Fujimura et al., 2008). This underlines the complexity of interactions and responses, as well as the difficulties of an appropriate choice of variables, technologies and analytical methods.
Acknowledgements
M.U. acknowledges financial support from the German Science Foundation – DFG (UN262/9-1), H.R.N. acknowledges financial support from the Swedish Research Council of Environment, Agricultural Sciences, and Spatial Planning (FORMAS, 215-2011-498). A.B. was supported by the Virtual Institute of Integrated Climate and Landscape Evolution Analysis – ICLEA (grant no. VH-VI-415); M.W. by DFG Wi 2680/8-1 and P.E. by DFG EU 132/1-1. Data evaluation was greatly enhanced by the recently established DFG Research Training Group RESPONSE (RTG 2010). Andreas Brachmann (Munich) and Derek Peršoh (Bochum) were responsible for the design of Illumina primers, and contributed to library preparation and sequencing.
Author contributions
M.U. designed and performed the mycological research. P.E. and M.S. designed and performed the genotyping analysis with contributions by D.G.W. P.E., M.S., A.B. and M.W. were responsible for fieldwork and evaluation of environmental data, H.R.N., A.J. and M.B.D. analysed mycobiome data. All authors contributed to discussion of results and manuscript writing.
