Plant specimens and their respective rhizosphere soils were collected during Antarctic Scientific Expedition no. 53 (ECA53; February 2017) to the South Shetland Islands of Antarctica, organized by Chilean Antarctic Institute (INACH). The plant specimens were taken from mantles of D. antarctica and C. quitensis located at the following coordinates: 62°59′53″S, 60°35′17″W and 62°24′7″S, 58°18′29″W, respectively. The plant specimens were randomly taken in a 10 m transect by using a clean spade to remove intact roots from soil. Collected plants and soils were placed within plastic bags, stored at 4°C, and transported on ice to the Applied Microbial Ecology Laboratory (EMALAB) at La Frontera University for microbiological analyses.

Endosphere samples from four plants of each species were processed as described by Barra et al. (2016). Plant tissues (roots and leaves) were washed and surface sterilized by repeated immersion in 70% (v/v) ethanol for 3 min, followed by 2.5% (v/v) sodium hypochlorite (NaOCl) for 5 min, and exhaustive rinsing with sterile distilled water (SDW). Portions of tissues (1–2 g) were aseptically cut, frozen in liquid nitrogen, macerated and homogenized with a mortar and pestle, and stored at −80°C until DNA extraction. In parallel, quadruplicate phyllosphere leaf samples were processed as described by Cid et al. (2017). Briefly, 1 g portions of leaves were cut (aerial parts), gently washed, and vortexed for 10 min in 10 ml sterile saline solution (0.85% NaCl). Leaves were removed, and the recovered liquid was centrifuged at 15,700 × g for 10 min to collect detached bacterial cells. Bacterial cells were suspended in 50 μl of SDW, and this suspension was subsequently frozen in liquid nitrogen and thawed at room temperature three times. Samples were centrifuged at 15,700 × g for 40 min, and the supernatant (~40 μl) was used as template DNA in PCR reactions. Rhizosphere soil samples from each plant specimen were processed, in quadruplicate, as described by Lagos et al. (2014). Briefly, soil aggregates were detached from roots by vigorous vortexing and collected in sterile polypropylene microtubes. Rhizosphere soils were gently mixed, and 1–2 g subsamples were stored at −80°C, and later subjected to DNA extraction.

The physicochemical properties of the rhizosphere soils were also determined as follow. The pH was measured in 1:2.5 soil/deionized water suspensions. Available phosphorus (P_Olsen) was extracted using 0.5 M Na-bicarbonate method and analyzed using the molybdate-blue method (Murphy and Riley, 1962). Organic matter contents were estimated by wet digestion (Walkley and Black, 1934). Exchangeable cations (K⁺, Ca²⁺, Mg²⁺, and Na⁺) were extracted with 1M CH₃COONH₄ at pH 7.0 and analyzed using flame atomic adsorption spectrophotometry (FAAS) (Warncke and Brown, 1998). Exchangeable aluminum (Al³⁺) was extracted with 1M KCl and analyzed by FAAS (Bertsch and Bloom, 1996).

DNA from the endosphere and phyllosphere samples was extracted by using Quick–DNA^TM Plant/seed Miniprep kits (Zymo Research, CA, USA). DNA from rhizosphere soil samples was extracted with PowerSoil® DNA isolation kit (Qiagen, MO BIO Laboratories, CA, USA), both kits were used according to manufacturer instructions.

The distribution and relative abundances of endophytic bacteria in root endospheres, leaf phyllospheres, and rhizosphere soils, was assessed by high throughput DNA sequencing (HTS) analyses as follow. The V4 hypervariable region of the 16S rRNA was amplified, for bacteria and archaea, by using primer set 515F (5′- GTG CCA GCM GCC GCG GTA A−3′) and 806R (5′- GGA CTA CHV GGG TWT CTA AT−3′). Sequencing was done by the University of Minnesota Genomics Center (UMGC, Minneapolis, MN, USA) using barcoded primers and the dual indexing method (Gohl et al., 2016). Amplicons were gel purified, pooled, and paired–end sequenced at a read length of 300 nt on the Illumina MiSeq platform (Illumina, Inc., San Diego, CA, USA).

Mothur ver. 1.34.0 was used for most sequence analyses (Schloss et al., 2009). In brief, after trimming low-quality regions at the ends of reads, the paried-end sequencing reads were merged by Fastq-join software (Aronesty, 2013), maintaining an average quality score >33. Primer sequences were removed from reads and high quality sequencing reads were aligned on the basis of the Greengenes ver.13.8 (McDonald et al., 2012). The UCHIME software package was used to identify and remove probable chimeric sequences (Edgar et al., 2011). Non-microbiota (e.g., chloroplast and mitochondria) sequence reads were removed via QIIME (Caporaso et al., 2010), and data was rarefied to 14,000 sequence reads per sample set prior to statistical analysis. Raw sequencing data were deposited in the Sequence Read Archive (SRA) of NCBI under Accession Number PRJNA509213. For statistical analysis, the mothur program was also used to calculate alpha diversity indices, including Good's coverage, the Shannon index, and the Abundance-based Coverage Estimate (ACE). Principal coordinate analysis (PCoA) was used to ordinate the samples. Differences in beta diversity among the community were evaluated by analysis of similarity (ANOSIM) using Bray-Curtis dissimilarity matrices (Bray and Curtis, 1957; Clarke, 1993). Molecular variance (AMOVA) was used to measure differences in sample clustering (Excoffier et al., 1992). The VennDiagram package in R (https://www.r-project.org/) was used to identify shared OTUs of bacterial communities between plant niches (Chen and Boutros, 2011). Putative indicator OTUs in Antarctic vascular plants that were in association with the differentiation of plant niches were identified on the basis of the multipatt function using the indicspecies package in R (de Caceres and Legendre, 2009). The associations were further considered to be significant by using a false discovery rate (q < 0.1) (Strimmer, 2008). Visualization of the putative association of indicators with plant niches in the Antarctic plants was produced by using gplots package in R package (heatmap) and by iTOL (tree) (Letunic and Bork, 2016). FAPROTAX was used to predict potential functions among members of the microbial community in the different niches (rhizosphere, endospheres, and phyllosphere). Potential functions were determined via the default settings on the basis of taxonomic information of microbiota in Antarctic vascular plant (Louca et al., 2016).

Rare microorganisms were defined as those that were not found above 0.01% relative abundance in the rhizosphere, endosphere and phyllosphere samples from D. antarctica and C. quitensis. The co-occurrence network was constructed as described by Ma et al. (2016). Briefly, a Spearman correlation matrix was used to generate the co-occurrence network via the WGCNA package. The nodes indicate the OTUs, while the edges, which are connecting the nodes, represent correlations between OTUs. Prior to network construction, random matrix theory (RMT) was performed to identify the appropriate similarity of 0.82 as the threshold (Luo et al., 2006), and the P-values of correlations were defined by using the Benjamini and Hochberg false discovery rate (FDR) of < 0.05 (Benjamini et al., 2006). The igraph package was used to measure the network properties (Csardi and Nepusz, 2006), while Gephi was further used to achieve the network image and calculations of closeness centrality and betweenness centrality for each node (Bastian et al., 2009). In addition, the occurrence of putative keystone taxa, which play pivotal roles in the structure and functioning of the microbial community (Banerjee et al., 2018), was determined in each niche as follows (Berry and Widder, 2014): For the endosphere network, OTUs with degree > 8, closeness centrality > 0.18, and betweenness centrality < 0.05 were used to identify putative keystone taxa. For the phyllosphere network, OTUs with > 5 degrees, closeness centrality > 0.15, and betweenness centrality < 0.05 were selected as putative keystone taxa. For the rhizosphere network, OTUs with > 7 degrees, closeness centrality > 0.17, and betweenness centrality < 11 were chosen as putative keystone taxa.

The physicochemical properties of triplicate rhizosphere soils are summarized in Table 1. These analyses revealed differences between rhizosphere soils from both Antarctic plants. Rhizosphere soils from D. antarctica showed higher contents of available P, K, and organic matter compared with that from C. quitensis. In contrast, rhizosphere soils from C. quitensis had a greater higher cation exchange capacity (CEC) and Al saturation compared with those from D. antarctica. Despite these differences, both plant species had rhizosphere soils with similar pH_H2O values, 6.1 and 6.3 for D. antarctica and C. quitensis, respectively.

Sequencing resulted in an estimated 98 to 99% coverage of OTUs in the endosphere and phyllosphere samples from both Antarctic plants. Substantial, but significantly (Tukey's post–hoc test, p < 0.05) lower coverage (95 to 96%) was observed in the rhizosphere samples relative to those from the other plant niches (Table 2). Similarly, a significantly (p < 0.05) greater number of OTUs (define at a 97% similarity) was observed in rhizosphere samples (1,551 and 1,628 for D. antarctica and C. quitensis, respectively), compared with those found in other plant niches. In this sense, the numbers of OTUs observed in the endosphere and phyllospheres samples were lower in D. antarctica (434 and 522 OTUs, respectively) compared to those in C. quitensis (662 and 666 OTUs, respectively). Bacterial alpha diversity, revealed by the Shannon index, was significantly (p < 0.05) greater in rhizosphere samples (6.2 for both plant species) compared with endospheres (3.9 and 4.9 for D. Antarctica and C. quitensis, respectively) and phyllospheres (3.7 and 4.4 for D. Antarctica and C. quitensis, respectively). In addition, significantly lower ACE values (p < 0.05) were observed in the endosphere (525 and 884 for D. Antarctica and C. quitensis, respectively) and phyllosphere (865 and 1,312 for D. Antarctica and C. quitensis, respectively) samples compared with those in the rhizospheres (2,093 and 2,237 for D. Antarctica and C. quitensis, respectively).

Assignment of taxonomic affiliation to members of bacterial communities indicated that members of the phylum Proteobacteria were relatively abundant in both Antarctic plants, with values of 48.8 to 58.9, 60.6 to 75.4, and 35.7 to 36.1% for endosphere, phyllosphere, and rhizosphere samples, respectively (Figure 1). While the endosphere communities in both plant species were also colonized by relatively large numbers of Actinobacteria (22.6 to 26.1%), Bacteroidetes (9.3 to 12%), and Firmicutes (5.3 to 7.9%) phyla, the phyllosphere was co-dominated by members of the phyla Bacteroidetes (14.3 to 30.9%) and Actinobacteria (4.8 to 7.5%). In contrast, the rhizosphere was also co-dominated by members of the phyla Bacteroidetes (14 to 19.7%), Acidobacteria (11.7 to 12.8%), Actinobacteria (9.2 to 13.7%), and Verrucomicrobia (8.7 to 9.9%). With respect to minor taxa, those present at < 10% of communities, a higher presence of bacterial groups was observed in rhizosphere samples compared with those in other niches, and were mainly attributed to members of the phyla Armatimonadetes (0.8 to 1%), Chlorobi (0.4 to 0.7%), Saccharibacteria (formerly TM7) (0.5%), and Nitrospirae (0.3 to 0.6%) in the rhizosphere of both plant species.

At the family level, a greater relative abundance of taxa in the endosphere were attributed to Pseudomonadaceae (18.7 to 26.2%), followed by Enterobacteriaceae (5.3 to 13.8%), and Microbacteriaceae (7.2 to 7.4%) (Figure 2). A greater diversity of families was observed in the phyllosphere samples, with higher relative abundances of Pseudomonadaceae (25.5 to 33.3%) followed by Enterobacteriaceae (5.3 to 23.4%), Sphingobacteriaceae (6.5 to 10.7%), Oxalobacteriacea (7.7 to 9%), and Flavobacteriaceae (5.6 to 12.7%) families. Interestingly, Pseudomonadaceae were found as an abundant group in the endospheres and phyllospheres, but not in the rhizospheres. In contrast, the rhizosphere samples were dominated by members of the Chitinophagaceae (8.3 to 8.9%) followed by Chthoniobacteraceae (5.3 to 6.3%), Xanthomonadaceae (4.4 to 4.9%), and Comamonadaceae (3.8 to 4.8%) families.

PCoA analyses showed a clear separation between microbiota in the rhizosphere and other plant niches in both Antarctic plants (Figure 3), but these differences were not seen between plant species. Our analysis also revealed that 24.2% (1,109 of 4,587) and 16.2% (678 of 4,181) of OTUs were shared between the three plant niches of D. antarctica and C. quitensis, respectively (Figure 4). In contrast, 75.8% (3,478 of 4,587) and 83.2% (3,503 of 4,181) of the OTUs were not shared and they were exclusively found in the individual plant niches of D. antarctica and C. quitensis, respectively. The greatest number of unique, not shared, sequences were found in rhizosphere samples, with 2,489 and 2,293 OTUs for D. antarctica and C. quitensis, respectively. The detailed distribution of shared and unique OTUs among bacterial communities in the plant niches are also shown in Table S1.

Presumptive microbial functional groups in each plant niches are shown in Figure 5. Independent of niches and species, the major functions were attributed to heterotrophy (30.5 to 44.3%) and aerobic heterotrophy (25.8 to 36.3%) (Figure 5A). Fermentation (3.4 to 9.7%) and nitrate reduction functions (1.8 to 5.2%) were also assigned in all niches. When minor functional groups were analyzed, a greater abundance of functional assignments were observed in rhizospheres, compared to those from the endosphere and phyllospheres (Figure 5B). A greater abundance of functions related to nitrogen cycling was observed in the rhizosphere of D. antarctica and C. quitensis. In contrast, in the phyllosphere samples the functions were mainly attributed to degradation of aliphatic and aromatic hydrocarbons. Lastly, samples from the endosphere also showed functions related to nitrogen cycling and hydrocarbon degradation, and dark oxidation of sulfur compounds.

Indicator analyses, based on taxonomic assignments from the genus to phylum levels, was used to investigate the association between taxon abundance and plant niches. Detailed information on the average abundance of each bacterium in each niche, their maximum association, the significance of associations (p-value), and the false discovery rate correction value (q) can be found in Figure 6 and Figure S1 and Table S2. Overall, our analyses identified 256 OTUs that were significantly associated with various plant niches. Of these, however, only 84 taxa could be classified to the genus-level. While these 84 taxa were distributed among 12 phyla, most belonged to the Proteobacteria, Actinobacteria, and Firmicutes. Notably, Pseudomonas, which had the greatest abundance among the three plant niches, could be used as putative indicator taxa in the endosphere, where it was significantly associated with this niche. Moreover, Clavibacter was also greatly associated (R=0.91) with the endosphere. In contrast, Novosphingobium was the best putative indicator bacterium representing the phyllospheres, although it had the highest association (R=0.86) among the three niches. In contrast, Dactylosporangium (R=0.95) and Bradyrhizobium (R=0.65) were the best putative indicators of the rhizosphere niche in Antarctic plants (R=0.95), and.

Due to the differences in microbial community structure and microbiota composition across the three plant niches, we further investigated the bacterial network and putative keystone taxa for each niche separately (Figure 7 and Tables S3–S5). Network analysis of the endosphere included 842 nodes (e.g., OTUs) and 1,062 edges, indicative of the association between OTUs. Results in Figure 7A show five putative keystone taxa in the endosphere network: Microbacteriaceae (0.05%), Pseudomonadaceae (0.6%), two Lactobacillaceae (0.14% and 0.06%), and Corynebacteriaceae (0.02%). Most notably, the family Microbacteriaceae was the most critical keystone taxon, bridging the maximum number of nodes and associations in the endosphere network. The family Pseudomonadaceae, which had greater relative abundance among these five putative keystone taxa identified, was coordinated with Pseudomonas as an putative indicator bacterium at the genus-level in the endosphere (Figures 6, 7A and Figure S1). In contrast, network analysis of the phyllosphere identified only 567 nodes and 386 edges. Impressively, although the phyllosphere network had fewer nodes and edges, 23 putative keystone taxa were identified (Figure 7B). The majority of the associations in the phyllosphere were from the families Chitinophagaceae (4 out of 23) and Sphigomonadaceae (3 out of 23). In the case of the phyllosphere, Sphigomonadaceae, in association with Novosphingobium, at the genus-level, was the best putative indicator in the phyllosphere samples (R = 0.86). In contrast, and perhaps expected due to its high diversity, the rhizosphere network consisted of 1,392 nodes and 2,682 edges, by far the largest of the three compartmental niches. Despite its large size, however, this highly complex rhizosphere network only had 7 putative keystone taxa (Figure 7C). Moreover, among these seven putative keystone taxa, only Rhodospirillaceae was identified at the family-level.