Several expeditions to Antarctica were performed to collect soil samples in sterile 50 ml plastic tubes; the tubes were sealed and shipped at -20°C to the Genetics Laboratory of the Faculty of Science at the Universidad de Chile. Once samples arrived at the laboratory, they were maintained at -80°C until processing. The collection dates were January 2010 for King George island samples (Carrasco et al., 2012); January 2014 for Deception, Snow, Dee, Livingstone, Greenwich, Robert, Nelson and Litchfield islands (Troncoso et al., 2017); November 2014 for Union Glacier (Barahona et al., 2016); and January 2015 for Lagotellerie island. The geographical distribution of islands and Union Glacier is shown in Figure 1, including detailed collection sites from Lagotellerie. The details of the collection sites of other islands and Union Glacier were previously described (Carrasco et al., 2012; Barahona et al., 2016; Troncoso et al., 2017).

A PowerSoil^® DNA Isolation Kit (MO BIO Laboratories Inc., Carlsbad, CA, United States) was used for direct DNA extraction from soil samples. The manufacturers’ instructions were followed, with only one modification, the disruption step, was performed using a Mini Beadbeater-16 cell disrupter (BioSpec Bartlesville, United States) instead of vortex agitation, as no PCR-amplicons were obtained in reactions using samples obtained through vortex agitation. The PCR reactions were performed using 1 μl of DNA sample (direct or 1/10 dilution), 5 units of Taq DNA polymerase, dNTP mix at 0.4 mM each, forward and reverse primers at 1 mM final each, PCR buffer and MgCl₂ 2 mM. Amplification was performed using a 2720 (Applied Biosystems) thermal cycler using the following protocol: initial denaturation at 94°C for 3 min; 35 cycles of denaturation at 94°C for 30 s, annealing at 50°C for 3 min, and extension at 72°C for 3 min; and a final extension step at 72°C for 10 min. The universal primers F63 (5′-GCA TAT CAA TAA GCG GAG GAA AAG-3′) and LR3 (5′-GGT CCG TGT TTC AAG ACG G-3′) were used; these primers have been described as specific for amplifying the fungal D1/D2 region of the large subunit ribosomal gene (LSU) (Fell et al., 2000). In all PCR reactions, the same forward primer (F63) was used, but the reverse primer (LR3) was specific for each soil sample as a 454 adapter, a specific barcode and a linker sequence were added. Sequencing was performed at OMICS Solutions (Santiago, Chile) using the Ion 314^TM Chip Kit v2 (Thermo Fisher) and Ion Torrent personal genome machine (PGM), according to the manufacturer’s instructions (Rothberg et al., 2011). Three independent runs were performed: (i) samples from King George Island; (ii) samples from Deception, Snow, Dee, Livingstone, Greenwich, Robert, Nelson and Litchfield islands; and iii) samples from Lagotellerie and Union Glacier. Sequence data were deposited at the NCBI: Bioprojects PRJNA395715 and PRJNA397058.

Sequences were analyzed using Geneious 10 software and its plugins. Chimera sequences were removed using UCHIME v4.2.40 (Edgar et al., 2011). Sequences were classified according to the collection site where they were obtained according to barcodes, and primer sequences were removed. Sequences were compared, and those having ≥ 97% identity were clustered into an operational taxonomic unit (OTU); sequences that were not grouped into OTU were classified as Unique Sequences (UnSeqs). For comparative purposes, the OTU were generated considering the data independently at three geographical levels: (i) “Site,” considering sequences from each collection site; (ii) “Locality,” considering sequences from all collection sites on an island or in Union Glacier; and iii) “All,” considering the sequences from all collection sites obtained in this work. The OTU and UnSeq number and length average are shown in Supplementary Figure S1. OTU with a sequence length of at least 300 nt were compared against the NCBI database using nucleotide BLAST (BLASTn). Sequences displaying BLAST-hits of at least 90% identity and 70% coverage were considered for tentative OTU annotation. In the first instance, the search was filtered excluding sequences from uncultured sources at the NCBI database; then, sequences not annotated this way were compared again against the NCBI database, but this time considering sequences from uncultured sources. The majority of OTU displaying BLAST-hits had sequence lengths between 500 and 550 nt and showed a high percentage of identity and coverage to its best BLAST-hit.

Five grams of each soil or sedimentary rock sample was mixed with 6 ml of sterile water and agitated in a vortex for 16 h at 22°C using an Intelli Mixer RM-2M (ELMI, Calabasas, CA, United States). After decantation of coarse particulate material, 200 μl aliquots of soil suspension were seeded onto YM agar plates (0.3% yeast extract, 0.3% malt extract, 0.5% peptone, 1.5% agar) supplemented with 1% glucose or 1% maltose and onto MYP agar plates (0.7% malt extract, 0.5% yeast extract, 2.5% peptone, 1.5% agar, pH 5.0). All plates were supplemented with 100 μg/ml of both ampicillin and chloramphenicol. Three plate replicates were incubated at 15°C and 22°C and periodically inspected. Once yeast-like colonies were observed, they were transferred to a new plate with fresh media. For yeast DNA extraction, a loopful of yeast from agar plates was suspended in 600 μl of TE buffer (Tris 25 mM, EDTA 10 mM, pH 8.0) and mixed with 100 μl of 0.5-mm diameter glass beads and 600 μl of phenol:chloroform:isoamyl alcohol (25:24:1, v/v). Cell disruption was performed using a Mini-Beadbeater-16 cell disrupter (BioSpec Products Inc., Bartlesville, OK, United States) for 30 s. Samples were centrifuged at 10,000 × g for 10 min, and the supernatant was transferred to a new tube where it was mixed with 1 ml cold ethanol and incubated for 2 h at -20°C. After centrifugation at 10,000 × g for 10 min, the obtained DNA pellet was washed with 70% ethanol, dried at 37°C for 10 min and suspended in 100 μl of TE buffer with 40 μg/ml of RNase A. The integrity and concentration of the DNA samples were checked by 1.5% agarose gel electrophoresis, using as a reference a 1-Kb DNA ladder (New England Biolabs, Ipswich, MA, United States). From DNA samples, the D1/D2 region of the 18S rDNA was PCR-amplified using the universal primers F63 and LR3 (Fell et al., 2000). Amplicons were resolved via 1.5% agarose gel electrophoresis and purified from gels (Boyle and Lew, 1995), and the nucleotide sequences were determined using the sequencing services of Macrogen Inc. (Seoul, Korea). Sequences were analyzed using Geneious Pro 10 software (Biomatters, Auckland, New Zealand), and taxonomic analysis was performed using the maximum likelihood method based on the Tamura–Nei model (Tamura and Nei, 1993) conducted with MEGA7 software (Kumar et al., 2016). Only sequences from yeast type species from the taxonomic revision of Basidiomycota were used as references (Wang et al., 2015a,b), which were downloaded from the NCBI database.

The number of OTU and UnSeq sequences generated were similar when considering all data, and the length of most of them ranged from 500 to 550 nt (Supplementary Figure S1A). This was similar at the Locality and Site levels (Supplementary Figures S1B,C), with the exception of sequences from King George Island, where a high number of sequences were obtained but with shorter average length. Considering the OTU generated with all data, the sequences were classified as “shared” or “exclusive” if they were found in at least two Localities or a single Locality, respectively. Except for King George Island, the Localities had more shared than exclusive OTU (Figure 2). Higher percentages of shared OTU were found in Nelson (89%), Robert (87%) and Dee (86%) islands. The Localities that shared the highest number of OTU were Deception and King George (101 OTU), Dee and King George (71 OTU), Robert and King George (37 OTU), and Dee and Deception (36 OTU). Considering the distance between these Localities—Deception and King George (154 km), Dee and King George (83 km), Robert and King George (64 km), and Dee and Deception (73 km)—there is no correlation between the geographic distance among Localities and “molecular” similarity. To test this preliminary observation, all Localities were compared and hierarchically grouped according to geographic distance and molecular similarity. As shown in Figure 3, molecular clustering grouped the islands of the South Shetland Archipelago, except Greenwich Island, and separated them from Litchfield and Lagotellerie islands and Union Glacier, in agreement with their geographical clustering. No correspondence was observed between molecular and geographic clustering among islands from the South Shetland Archipelago, e.g., at the molecular level, Nelson and Robert islands (being both the most geographically closer) grouped with Snow island but not Greenwich island, even though the latter is closer to them. Similar results were observed when the collection Sites in a Locality were analyzed. As an example, there was no coincidence between the molecular and geographic clustering of two groups of samples (M1–M4 and M5–M8) collected from different extremes of Deception island, which were separated by Neptune’s Bellows channel (a channel on the southeast side forming the entrance to Port Foster). Another example is the T2 and T29 sampling sites from King George Island, which grouped together in the molecular analysis even though they are geographically further away than other King George’s sampling sites; T2 is located at the Fildes Peninsula and T29 at the Lions Rump (45 km apart) (Figure 4). Similar results were observed when other sampling sites from other islands or sites at Union Glacier were analyzed (Supplementary Figure S2). Considering these results, a correspondence to molecular similarity can be observed that accounts for the latitude of Localities studied.

A 54 and 14% of fungal sequence BLAST-hits belonged to the Ascomycota and Basidiomycota phyla, respectively (Figure 5A), of which almost half (47%) corresponded to lichenized fungi. If only BLAST-hits from yeasts sources were considered, a 68% were basidiomycetes. The higher number of OTU sequences with BLAST-hits obtained at the species level were from King George island and Union Glacier, in which the major species corresponded to Circinaria fruticulosa and Xanthophyllomyces dendrorhous, respectively (Figure 5B). The fungal BLAST-hits were distributed in 87 genera and 123 species, of which the highest numbers were found for the genera Xanthophyllomyces, Circinaria, Verticillium, Malassezia and Verrucaria (Supplementary Figure S3A) and the species X. dendrorhous, C. fruticulosa, Verticillium sp., Malassezia restricta, Pseudogymnoascus sp. and Thelebolus globosus. The Sites with the highest number of fungal BLAST-hits were Lagotellerie I7, Union Glacier T9, Union Glacier T7, Dee M14, Dee M15, Union Glacier T13 and Deception M1 (Supplementary Figure S3B). Considering the number of different putative fungal species, the highest number was found at King George (73 species), Deception (33 species), Dee (28 species) and Lagotellerie (22 species) Islands. The most ubiquitous species and Localities with the highest number of BLAST-hits obtained were X. dendrorhous (10 localities/Union Glacier), Verticillium sp. (7 localities/Union Glacier), C. fruticulosa (7 localities/Lagotellerie), M. restricta (6 localities/Union Glacier), T. globosus (6 localities/Lagotellerie) and Epibryon diaphanum (6 localities/Dee) (Figure 6). Primers were designed based on the OTU sequences of X. dendrorhous and Malassezia species, for which a high number of BLAST-hits were obtained, especially from King George Island and Union Glacier, to perform PCR reactions with newly purified DNA from soil samples from different localities using new DNA extraction kits, reagents, and enzymes; these were performed under a laminar flow chamber. Amplicons of the expected lengths were obtained from the DNA samples of these different localities, which were not obtained in controls with no DNA (Supplementary Table S1), supporting the results obtained in the amplicon-metagenome analysis.

We previously isolated and identified yeast species from the same soil and sedimentary rock samples analyzed in this work (Carrasco et al., 2012; Contreras et al., 2015; Barahona et al., 2016; Troncoso et al., 2017). However, yeasts highly represented in this amplicon-metagenomic study such as Xanthophyllomyces and Malassezia species from the Lagotellerie island and Union Glacier have not been isolated. Therefore, we renewed our efforts to cultivate yeasts from samples from these two localities using different culture conditions. The yeasts isolated were identified according to the identity (≥99.5%; (Kurtzman and Fell, 2006)) of their D1D2 rDNA region sequence to the closest known yeast species and taxonomic position (Figure 7 and Supplementary Table S2). In this way, we were able to isolate and identify yeast species belonging to the genera Cystobasidium, Debaryomyces, Goffeauzyma, Sporidiobolus, Vishniacozyma and Holtermanniella, which were previously identified by us (Carrasco et al., 2012; Barahona et al., 2016; Troncoso et al., 2017) and yeasts of genera Lecanicillium, Naganishia and Tilletiopsis that have not been previously isolated from these localities.

Taking into account the BLAST-hits, obtained for UnSeq sequences, with higher identity, coverage and ranged from 500 to 550 nt, a 57% corresponded to fungal sources, most of which belonged to the phyla Ascomycota (44%) and Basidiomycota (15%) (Supplementary Figure S5A). The Localities with the highest number of fungal BLAST-hits were King George Island and Union Glacier (Supplementary Figure S5B), and the Sites were Union Glacier T9, Greenwich M18 and Union Glacier T7. The most ubiquitous species were Verticillium sp., X. dendrorhous and E. diaphanum, which were found in 6, 6 and 5 Localities, respectively.

Thirty-six fungal genera were found in the analysis of both OTU and UnSeq sequences, while 47 and 46 genera were found only in OTU and UnSeq, respectively (Supplementary Figure S6A). Compared to the cultivable fungi described from Antarctica, only 24 fungal genera were detected in the OTU analysis, a number that increased to 36 if results from UnSeq were also considered (Supplementary Figure S6A). If the comparison is restricted to viable Antarctic yeast species described in the literature, only two and one were found in OTU and UnSeq analysis, respectively (Supplementary Figure S6B). Considering the cultivable yeasts isolated from the same samples by our group, only Vishniacozyma victoriae and Goffeauzyma gastrica were also found in the OTU and UnSeq analysis, respectively.

For phylogenetic analysis, only the fungal OTU having sequence length ≥ 500 nt were considered together with their closest BLAST-hit, which were clustered into 11 classes (Figure 8). Each class contained sequences reported from countries on several continents, with the exception of one class that included uncultured fungal sequences reported mainly from northeast San Diego County, CA, United States and other localities in North America. The most represented fungal classes were Lecanoromycetes (sequences from 36 countries, mainly from Sweden) and Eurotiomycetes (sequences from 27 countries, mainly from the United Kingdom). The Agaricomycetes (sequences from 9 countries) and Microbotryomycetes (sequences from 12 countries, mainly from the United States) were less represented fungal classes. In relation to non-fungal eukaryotes, the most represented classes were Bryopsida, Ascochloris and Bacillariophyceae (Supplementary Figure S7). Unlike fungal sources, the geographic origin for most non-fungal BLAST-hits sequences was not available. However, the few sequences with this information available come from several continents. The global distribution of fungal BLAST-hits and OTU sequences is shown in Figure 9. Most uncultured fungal sources were reported in the United States and Canada, and when only cultured fungi were considered, most of them were reported in the United States, China, and Nordic countries.