Plant and soil samples were collected from eight sampling sites in three distinct regions representing different climate zones; Ny-Ålesund, high Arctic (3 sampling sites), Kilpisjärvi, low Arctic (3 sampling sites) and Mayrhofen, European Alps (2 sampling sites) (Figure 1A). Kilpisjärvi is at the northwestern Finland and located along the Fenno-Scandinavian border. Its flora is dominated by mountain birch forest in the valleys and by fell tundra at higher elevations. The annual mean temperature is about −2.2°C with plant growth season of ca. 90–100 days. Ny-Ålesund (Svalbard, Norway) is located on an isolated archipelago in the high Arctic; the land cover is dominated by glaciers and permafrost layers, and the mean annual temperature is −4°C. The soil temperatures have been reported to be below zero for more than 250 days per year ranging from −6° to −25°C (Coulson and Hodkinson, 1995). Most biological activity is restricted to less than 10% of the total land mass coupled with about 3 months of plant growing season. The sampling location in the Mayrhofen is located above the tree line south of Mayrhofen over the snow-covered mountains (altitude ca. 2400 m above sea level) in the alpine tundra of the European Alps. Coordinates and details of sampling sites are listed in Supplementary File S1.

Twelve replicates of bulk soil samples (the top 5 cm soil was removed and soil samples from 5–10 to 10–15 cm, corresponding to major root mass of target plant species, were both used for analysis) and six samples of both O. digyna and S. oppositifolia (as whole plants with adhering rhizosphere soils) were collected from all sites, except in site “Saana” (Kilpisjärvi) where only O. digyna plants were sampled and site “Cliff” (Mayrhofen) where we sampled only 6 bulk soil samples. Sampling was performed during summer 2012. All harvested plants were flowering at the time of the sampling. Rhizosphere and bulk soil samples were processed and stored as specified by Kumar et al. (2016). After removing rhizosphere soils, plant roots were thoroughly washed with water and surface sterilized by immersing the plant material into 3% sodium hypochlorite for 3 min and then subsequently in sterile double distilled water (3 × 90 s). 80–100 mg of root samples were weighed, snap frozen with liquid nitrogen and stored at −80°C for further DNA analysis.

Soil pH and soil organic matter (SOM) content were measured as described in Kumar et al. (2016), while available phosphorous (P) was measured based on Bray No 1 extraction method (Bray and Kurtz, 1945). All the soil chemical analyses were performed in duplicates (2 technical replicates) per sample, and with 4–8 biological replicates per site and sample type (Table 1).

Microbial DNA from soil samples were extracted following manufacturer's instruction using MoBio Power soil kit (MoBio, Carlsbad, CA, USA). For soil samples 0.5 g of soil was used instead of 0.25 g because of low microbial counts in our soils (data not shown). For isolation of endophyte samples, Invisorb Spin Plant Mini Kit (STRATEC Biomedical AG, Germany) was used in order to ensure prolonged stability of endophytic DNA in the plant derived samples. Frozen plant tissues were homogenized by bead beating for 45 s with 0.1 mm sterilized glass beads with FastPrep homogenizer (mpbio.com), followed by DNA extraction according to manufacturer's protocol.

After isolating DNA from all six plant replicates from both plant species, four (rhizo- and endosphere) or eight (bulk soil) samples technically best samples (good DNA yield, good PCR amplification) were included in the next generation sequencing library construction.16S rRNA gene was amplified using primers 799f/1492r (Chelius and Triplett, 2001) and M13-1062f/1390r in a nested approach. The nested primers targeting the V6-V8 regions of 16s rRNA gene enable elimination of plant chloroplast 16S rRNA gene amplicons as well as separation of endophyte amplicons from plant mitochondrial amplicons by size fractionation (799f–1492r, Chelius and Triplett 2001) and produce an amplicon with high phylogenetic coverage and optimal size for IonTorrent sequencing (1062f–1390r). Primers 1062f (Ghyselinck et al., 2013) and 1390r (Zheng et al., 1996) were tagged with M13 sequences to enable sample barcoding as described below and in Mäki et al. (2016). Both reactions had 1 μl of sample DNA, 1x PCR buffer, 1 mg/ml of BSA, 0.2 mM dNTP's, 0.3 μM of each primer and 1250 U/ml GoTaq DNA Polymerase (Promega, WI, USA) in a 30 μl reaction volume. 5–10 and 25–30 ng of soil and endophyte DNA, respectively, was used in the first PCR, and 1 μl of 1:10 diluted amplicons (for bulk and rhizosphere soil samples) and 1 μl of amplicons (for endosphere samples) from the first PCR were used as a template for the second run. Amplifications for both PCR reactions were performed as follows: 3 min denaturation at 95°C followed by 35 cycles of denaturing, annealing, and extension at 95°C for 45 s, 54°C for 45 s, and 72°C for 1 min, respectively. Final extension was carried out at 72°C for 5 min. Prior to library production, the PCR protocol was optimized with regard to several primer pair combinations, PCR protocols and test of PCR blockers to minimize the strong interference of mitochondrial rRNA in O. digyna and S. oppositifolia. The above described protocol, using high coverage, minimal bias primer pairs, was shown to produce enough eubacterial (endophytic) amplicons with no observable decline in diversity (as detected by T-RFLP) for sequencing, while most alternatives lead to very low amplification levels endophytes and strong mitochondrial signal.

Sequence libraries were prepared by running a third PCR to attach the M-13 barcode system developed by Mäki et al. (2016). Amplicons from second PCR were diluted 1:5 and re-amplified using barcode attached M13 system as forward primer and 1390r-P1 with adaptor A as a reverse primer. PCR mix and conditions were similar as described above, with an exception of using 8 cycles for amplification. Amplified libraries were purified using Agencourt AMPure XP PCR purification system (Beckman Coulter, CA, USA). Purified samples were quantified with Qubit Fluorometer (Invitrogen, MA, USA) and an equivalent DNA quantity of each sample was pooled together. The pooled samples were then size fractionated (size selection range of 350–550 bp) using Pippin Prep (Sage Science, MA, USA) 2% Agarose gel cassette (Marker B) following the manufacturer's protocol. Size fractioned libraries were sequenced using Ion 314 chip kit V2 BC on Ion Torrent PGM (Life Technologies, CA, USA) in Biocenter Oulu, Finland.

The raw sequence reads were processed using QIIME (Caporaso et al., 2010) and UPARSE (Edgar, 2013) based on a 16S rRNA gene data analysis pipeline developed by Pylro et al. (2014) with slight modifications in quality filtering. Sequences were trimmed by removing sequences with low quality reads (Q score <25) and shorter base pair (<150) length. Furthermore, all the raw reads were trimmed (200 bp), aligned and clustered at 97% identity using USEARCH algorithm (Edgar, 2010). UCLUST algorithm along with Greengenes database (DeSantis et al., 2006) was used to assign taxonomies at 97% identity to the individual OTUs. In total, 426,135 high-quality reads (1468 reads-5331 reads per sample) were clustered into 985 OTUs. For alpha diversity analysis all the samples were rarefied (subsampled) to 1400 reads per sample. Shannon index and species richness were obtained using Univariate Diversity Indices (DIVERSE, PRIMER 6 (PRIMER-E Ltd.)). The differences in diversity indexes between the soil samples and their correlation with soil physico-chemical properties were determined using two-way ANOVA and Pearson correlation (SPSS Statistics, IBM). The significance of the differences between the soil samples were tested by Games-Howell post-hoc tests (two-way ANOVA).

To normalize the data for community structure and other analyses all the samples with more than the median reads were rarefied to the median (2780 reads), while the samples with less reads were used as such, as described in Cárcer et al. (2011). In addition, all the singletons and OTUs with less than 50 reads were removed before processing. The influence of sampling site, geographic region, plant compartment, and plant species on bacterial community structures, based on Bray-Curtis distance matrixes of square root transformed abundance data, were analyzed using permutational multivariate analysis of variance (PERMANOVA) and visualized by PCoA ordinations at the OTU level. Taxonomic groups (phyla or OTU) with strongest impact on significant differences between community structures were identified with SIMPER (Similarity Percentages–species contributions), all performed with PRIMER 6 software package with PERMANOVA+ add-on (primer-e.com).

All the Ternary plots were made by calculating the mean relative abundances of OTUs per geographic region/compartment and with the function “ternaryplot” “vcd” (Meyer et al., 2015) from the R package. All other graphs (bar and scatter plots), also based on the mean relative abundances of taxa, were constructed using the R package “graphics.”

The highly conserved OTUs (core OTUs) were manually picked by selecting OTUs that were constantly observed (present in at least 3 out of 4 replicates per site) in the endosphere of either O. digyna and S. oppositifolia or both. To determine the distribution of core OTUs reads across different compartments, the averaged read count per compartment was calculated for each OTU, the averages were summed up and presented as the relative distribution of each of the 13 core OTUs per compartment.

Table 1 lists the soil characteristics in the three geographic regions: Mayrhofen (alpine), Kilpisjärvi (low-arctic) and Ny-Ålesund (high-arctic) (Figure 1A). The Ny-Ålesund [bulk] soils had significantly higher pH (two-way ANOVA, p < 0.05) and soil organic matter (SOM) values (two-way ANOVA, p < 0.01), and significantly lower levels of available phosphorus (two-way ANOVA, p < 0.05) than the Kilpisjärvi and Mayrhofen soils. The Kilpisjärvi soils had the lowest average pH values, but there were no significant differences in the other physico-chemical properties between the Kilpisjärvi and Mayrhofen bulk soils.

The species richness (SR) and α-diversity (Shannon index, SI) values of the bulk soil bacterial communities differed between the geographic regions. The Kilpisjärvi bulk soils (SR = 33.95, SI = 4.21) had significantly lower richness (two-way ANOVA, p < 0.01) and diversity (two-way ANOVA, p < 0.01) values than the Mayrhofen (SR = 41.04, SI = 4.61) and Ny-Ålesund bulk soils (SR = 42.55, SI = 4.92; Figure 2A). In contrast, there were no significant differences in the diversity levels of the rhizosphere soil or endosphere samples between the regions (Figure 2A).

There was a significant positive relationship of both SR and SI with soil pH (Pearson correlation [2-tailed], SR, p < 0.001, SI, p < 0.001) and a negative one with the levels of available phosphorus (P) (Pearson correlation [2-tailed], SR, p < 0.014, SI, p < 0.001). There was a significant positive correlation between SOM and SI, but not between SOM and SR (Figure 2B).

Collectively, the OTUs from all our samples fell into 21 bacterial phyla. Eight of these, i.e., Proteobacteria, Actinobacteria, Acidobacteria, candidate division AD3, Bacteroidetes, Firmicutes, Chloroflexi and Gemmatimonadetes, were prominent, collectively making up about 97% of the total microbiome. The remaining 13 phyla were present at less than 1% relative abundance each.

Bacterial community structures in the samples from the different regions were significantly different at phylum level (PERMANOVA F = 8.1155, p = 0.001). SIMPER analyses confirmed that Proteobacteria, Acidobacteria, AD3 and Actinobacteria were the main phyla contributing to the overall dissimilarities between the regions (Table 2). Proteobacteria were relatively more abundant in the alpine (Mayrhofen; average relative abundance 57%) than in the arctic regions (46% in Kilpisjärvi and 43% in Ny-Ålesund). The phylum Acidobacteria and the candidate division AD3 were observed in higher average relative abundances in Kilpisjärvi than in the other two regions (Figures 3A,B). The AD3 candidate division had reduced diversity (Figure 3A), with a single abundant OTU (OTU 10) dominating the Kilpisjärvi bulk soil samples, representing about 25% of the total bulk soil community. The Ny-Ålesund samples were enriched with Actinobacteria (Figure 3A), with average relative abundances in Kilpisjärvi = 14%, Mayrhofen = 14% and Ny-Ålesund = 23%.

The increased relative abundances of Proteobacteria in Mayrhofen, Acidobacteria and AD3 in Kilpisjärvi and Actinobacteria in Ny-Ålesund were also consistent in the communities in the different compartments (bulk soil, rhizosphere soil, endosphere) (Supplementary File S2), with the exception of AD3, which was present at very low abundances in the endosphere (<0.9%) in all three regions (Table 2). Additionally, the relative abundances of Firmicutes in the endosphere samples increased with increasing latitude, being lowest in Mayrhofen and highest in Ny-Ålesund.

Bacterial community structures were clearly different in the different compartments at the phylum level (PERMANOVA pseudo-F = 64.371, p = 0.001; Table 2). These differences were mainly driven by strong relative enrichment of Firmicutes in the endosphere-derived sequence data sets, compared to their very low abundances in the bulk and rhizosphere soils (Figure 4, Table 2). The relative abundances of Proteobacteria and Bacteroidetes increased progressively from bulk to rhizosphere soil to the endosphere, with a concomitant decrease in those of candidate division AD3, Gemmatimonadetes and Chloroflexi, which collectively constituted <4% of endosphere communities (Table 2, Figure 4B). This trend was similar in all three geographic regions.

The divergence of the endosphere communities from the soil communities was also evident at the class level. For example, OTUs representing the actinobacterial class Thermoleophilia were abundant in the bulk and rhizosphere soil communities, whereas these were rare in the endosphere communities. The latter were dominated by the class Actinobacteria (Class, order and family level analyses in Supplementary File S2).

The diversity values of the endosphere communities, analyzed at the OTU level, were significantly lower than those of the bulk or rhizosphere soil communities (Figure 2A). The rhizosphere soils had the highest diversity values, but the differences between rhizosphere and bulk soils were not significant (two-way ANOVA, p > 0.05, Figure 2A). However, we observed no such differences between the plant species, as O. digyna and S. oppositifolia had similar SR and SI indices in the rhizo- as well as the endosphere communities.

Also, the community structures of the endosphere bacterial communities differed clearly from those of the bulk and rhizosphere soil ones across all three regions, as demonstrated by PCoA (Figure 5A). A separate analysis of the soil-derived samples revealed that the bulk soil communities diverged from the rhizosphere soil ones in the three regions (Figure 5B). This was supported by PERMANOVA, where compartment was identified as a significant and strong driver of the differences between bacterial community structures (Pseudo-F = 30.962, p < 0.001, Table 3). Pair-wise analyses of the community structures supported the PCoA analyses, with significant (p = 0.001) differences between the endosphere and bulk soil, endosphere and rhizosphere and rhizosphere and bulk soil communities, and t-values of 6.339, 6.39, and 3.134, respectively.

In addition to compartment, sampling site (pseudo-F = 4.0646, p < 0.001) and region (Pseudo-F = 6.5495, p < 0.001) both had significant effects on the bacterial community structures, although these factors had less impact than compartment (Table 3). PERMANOVA performed on each of the compartments separately revealed that region and sampling site had the greatest influence on the structure of bulk soil communities (Pseudo-F = 9.1503, p < 0.001 and Pseudo-F = 7.6707, p < 0.001, respectively) with their influence decreasing for the rhizosphere (Pseudo-F = 5.9962, p ≤ 0.001 and Pseudo-F = 5.0728, p ≤ 0.001, respectively) and endosphere (Pseudo-F = 2.7877, p ≤ 0.001 and Pseudo-F = 2.1418, p < 0.001, respectively; Table 3). Interestingly, region shaped community structures more than sampling site for all compartments, indicating an impact of bioclimatic conditions (Table 3). Thus, in further analyses, we focused on comparing communities from the different regions and plant species.

PERMANOVA identified both plant species and geographic region as significant drivers of the community structures of the rhizosphere soil communities (PERMANOVA p < 0.01), but region (Pseudo-F = 5.8857) had more impact on the differences than plant species (Pseudo-F = 2.9879; Table 3). In contrast, while plant species, region and their interaction all had significant impact on endosphere community structures (PERMANOVA, p < 0.01), plant species had stronger impact on the differences between the communities (Pseudo-F = 4.0332) than region or interaction between these factors (Pseudo-F = 2.9678 and Pseudo-F = 1.6249, respectively; Table 3). The endosphere communities from all three regions, being relatively similar to each other, tended to diverge based on plant species (O. digyna or S. oppositifolia) on the first two axes in the PCoA ordination (Figure 5C), while we did not observe plant species specific clustering in the PCoA of the corresponding rhizosphere communities (data not shown).

On the basis of the above analyses, we found partial support for our hypothesis that plant species strongly shape the plant-associated bacterial communities, as this factor emerged as the major (albeit not the only) significant driver of the endosphere bacterial community structures over multiple sites and several regions (climate zones). Plant species also had a small, but significant impact on the rhizosphere community structures, but these were mainly determined by geographic factors.

Remarkably, the majority of the endosphere bacterial taxa was present in both plant species, but in different relative abundances. A total of 841 OTUs was found in the endosphere samples, comprising 152,050 reads. A vast majority, i.e., 612 OTUs (149,422 reads, 98.3% of all endosphere reads), was shared between the two plant species (Figure 6A), and many of these OTUs were consistently enriched along plant species. For example, OTUs representing Sphingobacteriales (Sphingobacteriia, Bacteroidetes), Burkholderiales (β-proteobacteria) and Bradyrhizobiaceae were enriched in the O. digyna samples, while OTUs in the Clostridiales, along with Actinobacteria, and Acidimicrobiia as well as several OTUs representing Myxococcales and Saprospirales were relatively more abundant in S. oppositifolia across the three climate zones (Figures 6B,C, Supplementary File S2). These were also identified as the main OTUs responsible for plant species specific community structures in the SIMPER analysis (Table 4). In addition to the shared bacterial taxa, 162 OTUs (1749 reads, 1.1% of the total endosphere reads) and 57 OTUs (879 reads, 0.6%) were observed only in O. digyna and S. oppositifolia, respectively (Figure 6A).

We examined the bacterial taxa that were highly conserved (belonging to the “tight” core) in the O. digyna or S. oppositifolia endospheres using as a criterion “OTUs present in at least three out of four endosphere samples per plant species across all sampling sites and regions.” Thirteen such OTUs were found, of which five, representing Bradyrhizobium (2 OTUs), Rhodoplanes (α-Proteobacteria), Janthinobacterium (β-Proteobacteria) and Planococcaceae (Firmicutes), were consistently present in both plant species (Table 5). Additionally, eight OTUs were consistently present in just one of the plant species. Thus O. digyna specific core OTUs belonged to Comamonadaceae (β-Proteobacteria) and Enterobacteriaceae (γ-Proteobacteria), whereas S. oppositifolia specific core OTUs belonged to Micromonosporaceae, Micrococcaceae (Actinobacteria), Bradyrhizobiaceae (α-Proteobacteria) and unidentified β-Proteobacteria (Table 5). Collectively, these (highly conserved) core OTUs accounted for 38% of the total reads in the endosphere communities. Significantly, eleven of these core OTUs (all except OTUs 171 and 429, Table 5) were among the main drivers of the divergence of the endosphere communities of the two plant species (Table 4). They also explained the differences between the endosphere and the soil bacterial communities, and those between the endosphere communities in the different geographical regions (Table 4). Of the 13 core OTUs, 11 were predominantly present in the plant associated compartments, as over 75% of their reads were detected in the endosphere, and over 80% in the endo- or rhizosphere (Figure 7).

In this study, we examined the bacterial communities in three regions spanning over 3000 km in distance, i.e., Mayrhofen (alpine), Kilpisjärvi (low-arctic) and Ny-Ålesund (high-arctic). In these three regions, the climatic conditions are clearly different. The highest bacterial species richness and diversity values in the bulk soils were found in the Ny-Ålesund samples, which was consistent with data by Chu et al. (2011) and Neufeld and Mohn (2005) who also detected highest bacterial diversities in high northern latitudes. However, our data stand in contrast to those from Yergeau et al. (2007), who reported decreasing bacterial diversities in Antarctic soils with increasing latitude toward the south pole. We found a clear positive correlation of bacterial diversity with soil pH and SOM, and a negative correlation with the level of available P, agreeing with studies that put forth soil pH as a major driver of bacterial diversity (Fierer and Jackson, 2006; Lauber et al., 2009; Rousk et al., 2010; Fierer and Lennon, 2011; Shi et al., 2015). Soil nutritional status and available P have also been shown to significantly impact bacterial diversity (Siciliano et al., 2014).

With respect to compartment, the endosphere bacterial communities were significantly less diverse than those in the corresponding soils. However, in contrast with studies from other soils (Smalla et al., 2001; Kowalchuk et al., 2002; Inceoğlu et al., 2011), where rhizosphere soil communities have been reported to be less diverse and rich than bulk soil ones, we observed a trend toward higher richness and diversity in the rhizosphere than in the corresponding bulk soils, although these differences were not statistically significant. This trend was similar to findings in a previous study from the Kilpisjärvi site, where the rhizosphere samples had highest richness and diversity (Kumar et al., 2016). Miniaci et al. (2007) and Coleman-Derr et al. (2016), studying low-SOM glacier forefield or desert soils, respectively, also observed higher bacterial diversity and richness values in the rhizospheres than in the corresponding bulk soils. Further, Yergeau et al. (2007) found that, although soil bacterial diversities in unvegetated Antarctic fell-field soils decreased with increasing (southern) latitude, those from vegetated sites did not. This suggests that a plant-incited “protective or nutritional” effect on bacterial communities becomes increasingly more important in soils in which conditions are challenging.

In this study, we detected only few “endemic” bacterial OTUs, as the great majority of the bacterial taxa was found in all three, geographically distant, regions. However, these taxa were present in very different relative abundances, leading to region-driven community structures. Roughly, proteobacterial taxa decreased and Gram-positive ones increased toward the north, with Acidobacteria and candidate division AD3 being enriched in the Kilpisjärvi samples. This clear progressive change in bacterial community structures hints at specific effects of the shifting local conditions on the aforementioned taxa. Thus, habitat filtering rather than [long-distance] dispersal impacts the bacterial community compositions across the three cold climate sites.

The dominance of Proteobacteria in the bulk soil samples from Mayrhofen was consistent with findings by Margesin et al. (2009) in alpine soils. Moreover, corroborating earlier studies (Männistö et al., 2007, 2013), the high abundance of Acidobacteria was likely linked to the low pH in the Kilpisjärvi soils (Chu et al., 2010; Griffiths et al., 2011). Also, the high abundance of candidate division AD3 in Kilpisjärvi (Figure 5B) was consistent with similar findings for the Mitchell peninsula in Antarctica (Ji et al., 2015) and low-nutrient sandy soils (Zhou et al., 2003). However, earlier studies by Männistö et al. (2007, 2013) have not detected candidate division AD3 in high-SOM Kilpisjärvi soils. We here assume that the candidate division AD3 members that were found are well adapted to the [low SOM/ low nutrient] soils. Alternatively, their absence from the previous data sets might be due to different 16S rRNA targeting primers used in the different studies.

A striking observation was that both O. digyna and S. oppositifolia sampled in any of the three regions exhibited quite similar endosphere bacterial communities. We previously observed compartmental influence between bulk and rhizosphere soils of O. digyna and S. oppositifolia (Kumar et al., 2016), and so extended this to the endospheres that were addressed in the current study. Clearly, even though the bulk soil bacterial communities were influenced by region and sampling site, which may relate to soil edaphic factors, the plant endospheres shared similar bacterial endophytes across the three regions. This points to a strong and specific filtering effect of the two pioneering plants that were studied, allowing similar bacteria to colonize plants from the widely divergent soils in different regions.

As a token of the plant-incited filtering effect, members of the Proteobacteria, Actinobacteria, Bacteroidetes and Firmicutes dominated the endosphere bacterial communities. Several other studies, performed with both agricultural and wild plants, also reported these four taxa to be dominant in several endospheres, with Proteobacteria being the most dominant one (Coleman-Derr et al., 2016; Santoyo et al., 2016; Zhao et al., 2016). Other taxa, including Acidobacteria, candidate division AD3, Chloroflexi and Gemmatimonadetes, were virtually absent from the endosphere. A general underrepresentation of Acidobacteria in the endosphere has also been observed in other systems (Edwards et al., 2015; Zarraonaindia et al., 2015; Coleman-Derr et al., 2016). The enrichment of Firmicutes in the endosphere samples in this study was mainly ascribed to the raised abundance of OTUs belonging to the Clostridia (in particular OTU 21; genus Clostridium). Possibly, such organisms might have been selected for their capacities to fix nitrogen in the cold and often water-logged soils (Rosenblueth and Martínez-Romero, 2006) in the permafrost-impacted Arctic sites. This hypothesis is supported by our [unpublished] observations, that nifH gene libraries prepared from the same plants as used in the current study are dominated by Clostridium-type genes in the (high) Arctic. Although Clostridium has been described as a strictly anaerobic genus, members of this genus have been shown to fix nitrogen in rice roots (Minamisawa et al., 2004), and survive in the potato endosphere in aerobic conditions (Shabuer et al., 2015). Interestingly, nifH genes of Clostridium spp. have been reported to be frequent in soil samples from the Canadian high Arctic (Deslippe and Egger, 2006). Similarly, in our study, plants were sampled in early growing season, when these started flowering and snow was melting in most sampling sites.

O. digyna and S. oppositifolia, the target plant species in this study, are both perennial herbs with similar habitat requirements, producing tap root systems of similar size and depth; the plants often grow at close proximity to each other. However, they are taxonomically quite distant (Soltis et al., 2000; Wikström et al., 2001) and have differing mycorrhizal associations. O. digyna is non-mycorrhizal, whereas S. oppositifolia is endomycorrhizal, which is likely to have strong impact on its nutrient acquisition efficiency.

Despite these differences, the endosphere communities of these two plants were strikingly similar. While we did find an effect of plant species on the endosphere community structures (Table 3, Figure 6C), the plants shared a core microbiome, dominated by Burkholderiales, Actinomycetales and Rhizobiales, across plants in the three arcto-alpine climatic regions. Of these, Actinomycetales and Burkholderiales have been reported as components of the core root microbiome of, e.g., A. thaliana (Schlaeppi et al., 2014). Rhizobiales are known plant symbionts with nitrogen fixing abilities, while Burkholderiales are well known for their biodegradative capacities and antagonistic properties toward multiple soil-borne fungal pathogens (Benítez and McSpadden Gardener, 2009; Chebotar et al., 2015). In our study, the core microbiome OTUs representing Burkholderiales, especially Comamonadaceae and Oxalobacteraceae, were relatively more abundant in O. digyna. We have repeatedly isolated bacteria from O. digyna vegetative tissues with very high sequence homologies to the above core OTUs ((Nissinen et al., 2012); unpublished). Further, we have isolated or detected (in clone libraries) bacteria in O. digyna seeds with 100% (16S rRNA gene based) identity to six of the core OTUs (OTUs 2 and 16 representing Rhizobiales, OTUs 8, 13, and 35 (Burkholderiales) and OTU 15 (Actinomycetales) (unpublished data). Core OTUs related to similar strains from seeds were highly enriched (Figure 7) in the endosphere or rhizosphere soils. Part of these core organisms could thus be seed-transmitted and colonize the rhizo- and endosphere of developing seedlings, as previously described by Puente et al. (2009) in desert cacti. This indicates the potential importance of such seed-transmitted endophytes in pioneer plants. Horizontal transmission of a set of endophytes has also been observed by Hardoim et al. (2012) and Johnston-Monje and Raizada (2011).

In addition, these core OTUs were among the primary drivers of region, compartment or host plant species differences among the bacterial communities. The higher relative abundances of Clostridia in Ny-Ålesund and Rhizobia in Mayrhofen in the endosphere communities is one such example, as discussed above.

In summary, we here report that, on the basis of data obtained with two plant species, host plant-specific endophytic communities can be acquired despite a distance of over 3000 km and differences in climate and chemistry between soils. These plant species-specific assemblages are formed from a shared core set of bacteria, most of which are strongly enriched in the endosphere. We surmised that plant-driven selection processes play a role, possibly concomitant with a highly efficient adaptation and fitness of these bacteria in the plant environment. Some of the core OTUs could even be seed-inherited, explaining their tight association with the host plant. Very closely-related endophytic taxa have previously been found to be shared by plants from other cold climates (Nissinen et al., 2012; Carrell and Frank, 2015; Poosakkannu et al., 2015), indicating the ecological tightness of [efficient] establishment of specific bacteria in arcto-alpine plants.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. The reviewer MA and handling Editor declared their shared affiliation, and the handling Editor states that the process nevertheless met the standards of a fair and objective review.