Microbiota entrapped in recently-formed ice: Paradana Ice Cave, Slovenia

Paradana is one of the biggest ice caves in Slovenia, with an estimated ice volume of 8,000 m3. Reflecting climatological conditions, the cave ice undergoes repeated freeze-thaw cycles and regular yearly deposition of fresh ice. Three distinct ice block samples, collected from the frozen lake in May 2016, were analysed to obtain data on ice physicochemical properties and the composition of associated microbiota. Isotopic composition of the ice samples (18O, 2H) and a local meteoric water line (LMWL) constructed for monthly precipitation at Postojna were used to estimate the isotopic composition of the water that formed the ice, which had high values of deuterium excess and low concentrations of chloride, sulphate and nitrate. The values of total organic carbon (1.93–3.95 mg/l) within the ice blocks fall within the range of those measured in karst streams. Total cell count in the ice was high and the proportion of cell viability increased along the depth gradient and ranged from 4.67 × 104 to 1.52 × 105 cells/ml and from 51.0 to 85.4%, respectively. Proteobacteria represented the core of the cave-ice microbiome (55.9–79.1%), and probably play an essential role in this ecosystem. Actinobacteria was the second most abundant phylum (12.0–31.4%), followed in abundance by Bacteroidetes (2.8–4.3%). Ice phylotypes recorded amounted to 442 genera, but only 43 genera had abundances greater than 0.5%. Most abundant were Pseudomonas, a well-known ice dweller, and Lysobacter, which previously was not reported in this context. Finally, two xanthophytes, Chloridella glacialis and Ellipsoidion perminimum, known from polar environments, were cultured from the ice. This indicates that the abundance and ecological role of phototrophs in such environments might be greater than previously deduced.


Results and discussion
Ice environment. Physicochemical analyses of individual ice blocks were conducted to observe eventual differences that could be attributed to spatially related gradual freezing-melting and fresh ice deposition, and to characterize the habitat that enables long-term survival of ice microbiota. All ice samples contained low concentrations of salts, indicating that they originated from recent clean snow. Concentrations of anions in the upper layers, Ice-1 and Ice-2, were similar. However, the bottom layer Ice-3 had distinctly higher electrical conductivity (EC), hardness and alkalinity, less nitrate, and more sulphate. This could indicate that this ice stratum includes a higher proportion of percolation water, which contains more ions than rain and snow as shown by the differences between the percolation water from the cave Planinska jama (that was used for preparing growth media) and the ice, as shown in Table 1. Total organic carbon (TOC) concentrations in the ice were in a range typical of karst streams 22 , and above the minimum values reported for surface streams, i.e. 0.1-36.6 mg/l 23 , indicating a significant input of organic matter for the underground ecosystem. TOC indicates an available in situ source of carbon for the ice microbiome. Nitrogen expressed as nitrate did not exhibit high values in ice samples (Table 1). In this respect, a parallel can be drawn with karst sediments, where microbes are commonly limited more by carbon and phosphorus than by nitrogen 24 .
Besides EC and temperature, pH and dissolved oxygen are additionaly two influential parametres that can affect the abundance and taxonomic structure of microbial communities. pH was found to drive the shift in the community structure not only in habitats such as freshwater, marine sediments or soils but also in cold habitats as Antarctic soils 25 . In the current samples, the pH effect on the microbial community structure is less evident because all the values are rather similar (Table 1). Cave ice habitats with incoming waterflow are probably not oxygen depleted; on the contrary, for example in Antarctic lakes, glacial meltwater inflow is responsible for oxygen supersaturation 26 .
Isotopically, the Ice-3 stratum was significantly lighter than the stratum represented by Ice-1 and Ice-2 ( Table 1). Correlation of δ 2 H and deuterium excess did not indicate any effect of kinetic fractionation during  27 . The ice lake in Paradana is presumably formed by the refreezing of water from melting snow accumulated during the winter months 20 , with some contribution of water dripping from the cave ceiling. November and December 2015 had only a few days with precipitation in Postojna (5 and 4, respectively). However, January and February 2016 had 12 and 20 days with precipitation and monthly totals were high, 152 mm and 312 mm, respectively. The air temperature data adjusted for the elevation difference between Postojna and the Trnovski gozd karst plateau (about 600 m) indicate that about one third of the precipitation in January and one half in February probably fell as snow. The rest was probably a mixture of solid and liquid precipitation, but heavy rains could have occurred as well (e.g. about 55.5 mm of precipitation was measured in Postojna on February 8-9, with mean daily air temperatures between 8 °C and 9 °C). Isotopic composition of precipitation varied significantly between and also during individual events. It is known that snow cover can preserve the isotopic composition of the original snowfalls for long periods 28 . However, individual snowfalls can mix at the entrance of the cave and the isotopic composition of snow accumulated in the cave can also be influenced by thaws caused by temporary increases of air temperature or rainfall. The isotopic composition of snowmelt water that eventually refreezes in the cave is therefore the result of many processes. Further research with better temporal and spatial resolution of samples and sampling of snowmelt water would be needed to improve knowledge on the dynamics and sources of ice formation.  Fig. S2). Therefore, the LMWL at Postojna could be the best representation of the isotopic composition of precipitation supplying water to the Paradana Ice Cave (after considering the elevation difference between the two sites, which is about 600 m). When analysed in more detail, results obtained using the approach described above (to calculate the isotopic composition of the water that formed the sampled ice) also revealed the sensitivity of the constructed LMWL, the length of data series and extreme values. This is illustrated by records of isotopically very light precipitation in November and December 2015 (δ 18 O − 17.6‰ and − 14.2 δ 18 O, respectively). Although such isotopically light precipitation occurred in just two of the 27 months of the observation period, the two values changed the LMWL intercept significantly. However, because they did occur, they cannot be disregarded in the LMWL construction. Daily precipitation data indicate that in both cases monthly values were influenced dominantly by precipitation that fell during just one day (precipitation on those days represented almost the entire monthly precipitation). The LMWL intercept at Postojna without those two months would be 8. It is worth noting that the δ 18 O values of Ice-1 and Ice-2 are higher than those reported for the Paradana Ice Cave by Carey et al. 20 . Deuterium excess is also significantly higher than the mean value reported for samples from different depths of ice by Carey et al. 20 . The difference in δ 18 O values could be related to different sampling sites. Carey et al. 20 sampled the wall ice, whereas the samples collected during this study represent the frozen lake. Investigation of the difference in deuterium levels would be especially interesting. It could point at the input (either by overland flow from the cave entrance or by percolation from the vadose zone) of water from the autumn/winter months, with precipitation from the Eastern Mediterranean air masses having particularly high d-excess (up to 22‰). The Western Mediterranean air masses have d-excess of about 14‰, whereas air masses from the Atlantic have values of only about 10‰ 29 . Late autumn to early winter precipitation in Slovenia (October to December) regularly exhibits high d-excess 27 . Unfortunately, the available data are insufficient to support analysis of the reason for high deuterium excess of the ice in detail. Study samples also display far lower concentrations of chloride, sulphate and nitrate than samples collected by Carey et al. 20 .

Concentration of microbes in cave ice.
The upper ice stratum represented by Ice-1 and Ice-2 had comparable microbial load expressed in total ATP concentration and total cell counts, whereas the Ice-3 block exhibited significantly higher values (Table 1). Interestingly, the total cell counts of microorganisms in the ice samples was similar (4.67 × 10 4 -15.15 × 10 4 ) to that recorded in the Pivka River (SW Slovenia) at the ponor connecting to the karst underground, i.e. 4.29 × 10 4 -12.38 × 10 4, 30 . A large proportion (51.0-85.4%) of entrapped microbes in the ice were viable, showing that they were able to survive ice formation and melting, or even several freezing-melting cycles. A relatively high cell viability can be linked to the availability of compatible solutes, indicated by correspondingly high TOC (Table 1). Not only do sugars and polyols increase microbial resistance to freezing, they can also be used inside the cell as carbon and nitrogen sources 31 . Higher concentration of salts in Ice-3 block was accompanied by the highest total cell counts and percentage of viable cells (Table 1). In ice from Scărişoara Cave total cell counts varied from 0.84 × 10 3 to 3.14 × 10 4 cells/ml with corresponding viability from 28.2 to 84.9%, but no correlation was observed between the ice age (0-13,000 years BP) or depth (0-25 m) and the total number of cells or viability 14 .
The media types used in this study differed in their ability to stimulate the growth of colonies. In general, nutrient-poor media and low temperatures resulted in higher colony counts in all samples. This phenomenon has www.nature.com/scientificreports/ been reported previously in cave microbiology, but was not correlated with phylogenetic diversity of microbes obtained on the growth media 32 . After 28 days of incubation, samples grown on the oligotrophic medium with percolation water (PWA) and cultivated at 10 °C produced the highest colony counts ( Table 2). In context this indicates that cave percolation water contains soluble compounds that are not present in tap water and which support the growth of cave-ice microorganisms. With respect to individual samples, the highest colony counts were found in the Ice-3 sample, i.e., 167.37‰ of all cell biomass, determined by flow cytometry (Table 2), and this sample also contained the highest concentration of nutrients (Table 1). Cultivable anaerobic bacteria and fungi were detected in all the ice samples (Table 2). Communities in the ice blocks differed in the representation of r-strategists, with their predominance in the Ice-1, and a big difference between Ice-1 and Ice-2, the two ice samples from the same stratum. Interestingly, a more-uniform community structure in terms of r-strategists was displayed in ice block Ice-2-Ice-3 (Table 1). R-strategists commonly dominate in uncrowded and unstable habitats where resources are temporarily abundant and available; with development of a community, r-strategists are gradually replaced by the slow-growing equilibrium K-strategists 33 .
Cultivation on different media showed that the ice contained metabolically diverse microorganisms, aerobic and anaerobic bacteria and fungi. Two species of yellow-green algae were also recovered in cultures from samples Ice-2 and Ice-3. The two cultivated species, Chloridella glacialis and Ellipsoidion perminimum (for identification see Supplementary Fig. S3), were also found in green ice from Antarctica 34 . It is known from results of previous studies that algae in ice can survive and even grow under such adverse conditions [34][35][36] . They can also be well adapted to low light and low water temperature; for example they can thrive under ice-and snow-cover where the available photosynthetic photon flux density is only around the photosynthetic compensation point 37 . In these terms, and particularly in ice caves with available light, algae and cyanobacteria should not be overlooked as an important part of the ice microbial community. Interestingly, in Himalayan-type glaciers, the algae-rich layers in ice cores were suggested as providing accurate boundary markers of annual layers 38 . It remains unclear whether algae can be applied similarly as boundary markers in cave ice. Their existence is already known from some caves, for example in Hungary in a small ice cave colonizing surfaces of the ice 39 , Romania in Scarişoara Ice Cave at the ice/water interface 40 and in New Mexico, USA, in Zuni Ice Cave giving the distinctive greenish patina of the layered ice 35 .
Bacterial community structure. Previous study of ice from the Paradana Ice Cave showed that it probably originates from local rainfall that reaches the cave as drip water after dissolving bedrock while percolating from the surface, and from snow that includes dust particles 20 . Thus, the largely impacted cave ice in Paradana has different sources, each bringing along a diverse and adaptable microbiota. 16S metagenomic analysis was conducted to describe the taxonomic composition of bacteria found in different ice blocks. Quality filtration of sequence readings gave a total number of 120,381 sequences in the three studied samples ( Table 3). The number of operational taxonomic units (OTUs) varied from 185 in Ice-2 to 304 in Ice-1. This pattern was in alignment with values of alpha diversity parameters: extrapolated richness (Chao1), abundance-based coverage estimator Table 2. Colony counts (colony-forming units-CFU/ml) and their proportion to total cell counts determined by flow cytometry (‰) at different cultivation conditions and media. In bold-highest value; RIDA RIDACOUNT total aerobic count; RIDA-Y&M RIDACOUNT yeast&mold rapid; an anaerobic, + presence, not countable. www.nature.com/scientificreports/ (ACE) and Shannon index ( Table 3). The rarefaction curves indicated that the diversity had been sampled sufficiently ( Supplementary Fig. S4). A Venn diagram of the distribution of 441 distinct OTUs found in the three studied samples is presented in Fig. 1. Observations showed that 119 OTUs (28.3%) occurred in all three samples and can be interpreted as "a core microbiome". Three of these OTUs dominated microbial communities in individual samples (relative abundance range 14.5-56.5%) and corresponded to the members of the genera Pseudomonas, Lysobacter, and Sphingomonas, as discussed below. These were followed in abundance by Polaromonas, Flavobacterium, Rhodoferax, Nocardioides, and Pseudonocardia (relative abundance range 3.3-6.9%). Another 35 OTUs had relative abundance above 0.5% and the remaining 76 OTUs had relative abundance below 0.5%. The unique OTUs probably contribute to the variability due to internal variations within the ice block caused by incoming snow or the freezing of percolation water. For example, samples Ice-2 and Ice-3 were cut from the same ice block in a vertical ice profile, but differed in their content of dark, particulate, organic inclusions.
Proteobacteria commonly represent a dominant group in various cold habitats where their abundance is associated with increased nitrogen loads and copiotrophic growth conditions [41][42][43] . Indeed, Proteobacteria are commonly associated with water influx into the karst underground [44][45][46] . The prevalence of Proteobacteria in the ice samples coincided with the decreasing trend of TOC concentration in samples Ice-2, Ice-3 and Ice-1 (Table 1). This notion was further supported by a low relative abundance of members of the class Acidobacteria (0.1-0.3%, Supplementary Fig. S5), whose members are mostly oligotrophs 47 . Because Proteobacteria form the core of the cave-ice communities studied, it is likely that they play an essential role in the functioning of this recently formed ice ecosystem.
The second most abundant phylum in the studied samples was Actinobacteria, a group associated with cave sediment communities 48,49 . Its relative abundance declined from 31.4% in sample Ice-1 to 26.7% in sample Ice-3 and 12.0% in sample Ice-2. This compares with the results from Scărişoara Cave, where the microbiome was dominated by Actinobacteria (38.5%) over Proteobacteria (33.5%), but Proteobacteria were assigned as the largest group of metabolically active microbes 14 .
In Paradana the relative abundance of Bacterioidetes phylotypes was comparably lower than that of the other two and was 4.3% in Ice-1, 3.2% in Ice-2, and 2.8% in Ice-3. Three other bacterial phyla represented > 1% of phylotypes in at least one sample and corresponded to Firmicutes, Cyanobacteria and Gemmatimonadetes. www.nature.com/scientificreports/ Phototrophic bacterial phylotypes belonging to Cyanobacteria were recovered from all three samples. They represented 1.3% of phylotypes in sample Ice-1, but only 0.6% and 0.3% in samples Ice-2 and Ice-3 respectively, from where algae, C. glacialis and E. perminimum, were obtained via cultivation. Phyla whose relative abundance was less than 1% were grouped together and classified as "Rare phyla". These phyla comprised 2.2%, 1.5% and 1.2% of Ice-1, Ice-2, and Ice-3, respectively. Their relative abundance is presented in Supplementary Fig. S5.
Among the 31 classes detected in this study, members of Gammaproteobacteria were most abundant and represented 20.1% (Ice-1), 45.3% (Ice-2) and 42.5% (Ice-3) of total detected phylotypes (Fig. 3A). This proteobacterial group was also most abundant in the ice from Scărişoara Cave 14 . Actinobacteria represented the second most abundant group of phylotypes, with its relative abundances declining from 30.8% in Ice-1 to 26.2% in Ice-3 and 11.7% in Ice-2. Other notably abundant classes were Alpha-and Betaproteobacteria, whose abundances ranged from 9.6 to 26.3% and from 6.9 to 12.3%, respectively.
Representatives of phylum Bacteroides, classes Flavobacteria, Sphingobacteria, Bacteroidia, and Cytophagia, represented 0.08-2.0% of detected phylotypes. Firmicutes were represented by classes Clostridia (1.0-1.8%) and Bacilli (0.6-2.1%), while the abundance of Gemmatimonadetes (0.3-1.8%), the only class of the phylum, was similar to that described in soil, around 2.0% 50 , and is probably a consequence of soil deposition in the ice. The relative abundances of the remaining 17 classes were less than 1.0%. These included Cyanobacteria, which were represented by the classes Nostocophycideae, Oscillatoriophycideae and Synechococcophycidae.
At the genus level, ice cave sequences from three samples corresponded to 442 genera, but only 43 genera had abundances greater than 0.5% in all samples (Fig. 3B). For comparison, 526 genera were identified in Scărişoara Cave 14 . The most abundant genera in the Paradana samples were Pseudomonas and Lysobacter (Gammaproteobacteria); however, their abundance varied within the samples.
Sphingomonas spp. represented the third most abundant phylotypes. Members of the family Sphingomonadaceae were commonly found in cave drips 58 . In Scărişoara Cave, members of Sphingomonas were particularly common in recently formed ice strata 14 . In Paradana the abundance of Sphingomonas phylotypes followed that of Lysobacter spp. and was 19.2% and 14.5% in Ice-1 and Ice-2, but only 1.1% in Ice-3.
It is possible that the above mentioned dominant bacteria interact with each other, but the nature of any interactions is not yet elucidated. For example, Lysobacter is commonly know for its antagonism towards other bacteria under nutrient-poor conditions, but in a direct assay of Lysobacter predation, the population of Pseudomonas fluorescens was not affected 59 . Furthermore, plant-colonizing Sphingomonas protect plants against pathogenic P. syringae 60,61 but there are reports that Sphingomonas can be outcompeted by fast growing Pseudomonas 62 .
It is important to mention specific genera that represented a minor portion of diversity. Polaromonas phylotypes are particularly abundant in polar and high-elevation environments owing to the dispersal of dormant cells by air currents 63 . In cave ice, Polaromonas phylotypes represented a minor proportion of the diversity. They were most abundant in Ice-3 (6.9%), and scarce in Ice-1 (1.1%) and Ice-2 (0.7%). Other notable genera whose abundance exceeded 0.5% were Flavobacterium (abundance range 2.7-3.6%), Nocardioides (abundance range

Conclusions
Study of recently formed ice in Paradana revealed no evidence of either uniform depositional conditions or uniform microbial representation, even in the same ice stratum, indicating compartmentalization within a larger ice block. Paradana, and caves having similar geological settings and climatological conditions, are subjected to repeating cycles of freezing-melting events, and ice deposition, that influence the microbiome. In general, ice as a habitat in karst caves at lower altitudes in temperate zones is currently prone to wasting. Soluble organic compounds transported by percolation water and melted snow mitigate osmotic stress, reduce the environmental extremity and support the dynamic nature of the cave-ice microbiome. Proteobacteria, particularly from the class Gammaproteobacteria, represent the core of the cave-ice microbiome in this recently formed ice ecosystem. Pseudomonas and Lysobacter were the most abundant genera identified in Paradana. The roles and abundance of phototrophs might also be significant, particularly in caves or parts of caves with available sunlight. It remains unconfirmed, however, whether specific microorganisms or concentrations of microorganisms can provide reliable boundary markers for regular episodes of ice formation in caves.

Materials and methods
Site description. The   Heat-map analysis of the relative abundance of members of cave-ice prokaryotic communities at class (A) and genus (B) levels in Ice-1, Ice-2 and Ice-3. Phylotypes whose relative abundances at class level were < 0.1% and phylotypes whose relative abundances at genus level were < 0.05% were amalgamated and classified as "Rare taxa". www.nature.com/scientificreports/ teau averages 1700-3200 mm, with its maximum in October and November, and with snow-cover present from November until the end of April. The mean annual temperature on the plateau at 1200 m a.s.l. is around 4.5 °C. The entrance to Paradana is at the bottom of a large, 50 m-deep doline, where snow accumulates during the winter and persists until summer due to cold-air circulation from the cave. This creates favourable conditions for temperature and vegetation-zone inversion 65 . The cave is more than 7,000 m long and 850 m deep. Because of the distinctive funnel-shaped morphology of the cave entrance, snow, soil, leaves, and tree branches slide into the cave. Permanent ice appears at a depth of about 35 m inside the cave. Most of the cave-ice forms in late winter from water that percolates through the ceiling and meltwater from the snow that accumulates in the entrance area. Water flows for some distance along the chilled passage floor before it freezes. Snowmelt produces additional water that partly inundates the passages and creates a frozen lake. This lake was used as a source of ice for cooling and food preservation until the beginning of the twentieth century. The commercial exploitation of ice was documented as early as 1867. Based on available documentation, the ice volume around 1917 was comparatively greater than currently. In the period from the 1950s to 1977 the cave entrance was completely ice-filled, and in 1978 the ice melted and further exploration of the cave became possible. During the following decade, the ice body changed its volume notably. Since 2000, the ice block has predominantly been shrinking in volume and surface area 66 . Relative mild winters in recent years accelerated the ice melt trend markedly.
The floor of the Velika ledena dvorana cave chamber contains a huge ice body lying over collapsed boulders (Fig. 4). Here, the ice is at least 12 m thick, with an estimated volume of about 8,000 m 3 . The cave's continuation beyond Velika ledena dvorana leads to smaller passages, followed by vertical pits and ramps. In this part of the cave, ice is present only to a depth of about − 200 m.
Both the formation and the preservation of ice have been attributed to the effects of air circulation 66,67 . The temperature of the Trnovski gozd karst massif and of the air inside its caves is relatively stable of about 4-5 °C. In winter, when the external temperature falls below 4 °C, an airflow is created leading to transfer of cold air from the surface into the cave. This air re-emerges from the karst massif at higher altitude (about 1400 m a.s.l.) where numerous patches of melting snow-have been observed. Initially the snowmelt is due to the effect of displaced air. The cold air flows through a system of interconnected cavities within the Golaki ridge. The height difference www.nature.com/scientificreports/ of about 270 m is great enough to create a "chimney effect" 68 . In the summer, the air temperature at the surface rises above the temperature within the cave system and the airflow reverses its direction. It moves from the lower parts of Paradana, passes the ice body, is cooled to about 1 °C, and exits the cave. Both winter and summer temperature regimes and flow directions are stable. During the summer months, cold airflow creates and maintains a large, stable, volume of cold air in the doline that holds the cave entrance.
Sampling. The exact micro-locations for ice sampling were selected specifically, away from the margins of the large ice body, to ensure comparable ice strata in the samples and to avoid material such as branches, stones and other detritus that are scattered in the ice block. Two ice blocks, Ice-1 and Ice-2-Ice-3, approximately 1.5 m apart, were cut with a chain saw from the frozen lake in Velika ledena dvorana on 09 May 2016 (Fig. 4). At the sampling site there was very weak indirect natural illumination (< 1 lx, < 0.01 µmol photons/m 2 s, < 0.1 W/ m 2 , LI-1500, LI-COR). Ice-2 was cut aseptically from the ice block Ice-2-Ice-3 to obtain the same ice stratum (approximately 15 cm thick) as in Ice-1. Compared to Ice-1 and Ice-2, the Ice-3 block was less transparent, with more dark inclusions of particulate organic material, and it was formed before the Ice1-Ice2 stratum.
Monthly composite precipitation samples were collected in Postojna between August 2015 and October 2017 (45°46′31.55"N 14°12′47.64"E, 553 m a.s.l., about 38 km from Paradana) and analysed for stable isotopes in order to construct the local meteoric water line (LMWL). Additionally, mean daily air temperature and daily precipitation were measured at Postojna. LMWLs calculated for other precipitation stations in Slovenia 27, 69 were also examined.
Physicochemical analyses. Contact surfaces of the ice blocks were sterilized using a gas burner before melting in the laboratory. Complete ice melt occurred after 48 h of incubation at room temperature, with the blocks yielding 3.5-5.0 litres of water. All ice samples included layers containing dark inclusions of organic matter, which affected the transparency of both the ice and the water after melting. The water was sub-divided aseptically to provide aliquots for subsequent analyses: 500 ml was used for chemical analyses, 5 ml for cultivation of microbes, total ATP concentration and flow cytometry, 5 ml for isotopic analyses, and 1.5 litre of sample water was used to extract each total community DNA. DNA was isolated from filters after membrane filtration of water through 0.22 µm pore-size filters.
Alkalinity ), and total organic carbon (SM 5310 TOC) were determined following standard methods 70 . Electrical conductivity (EC) and pH were measured using a portable WTW MultiLine P4 meter.
Stable isotopes of oxygen and hydrogen were measured by Picarro 2120i and 2130i water isotope analysers using previously described protocols 71,72 . The samples were analysed in triplicates and seven injections were used per sample. The values were expressed in the conventional δ notation (‰) with respect to Vienna Standard Mean Ocean Water. The analytical accuracy was better than 0.2‰ for δ 18 O and better than 1.0‰, for δ 2 H. The modified intersection method was used to obtain the isotopic composition of the precipitation that formed the cave ice 73 . Correlation of δ 2 H and deuterium excess d (d = δ 2 H-8δ 18 O) in water from ice was first checked to exclude kinetic fractionation during water freezing. Then, intersection of the freezing line determined by stable isotopes in samples Ice-1 to Ice-3 and the LMWL constructed for precipitation at Postojna provided the probable isotopic composition of the water that formed the sampled cave ice.

Concentration of microorganisms and isolation of phototrophs.
Microbial load in the samples was measured as total adenosine triphosphate (ATP) (AquaSnapTotal, Hygiena, USA), and expressed in Relative Light Units (RLU). Total cell counts and cell viability were determined by flow cytometry. The Cell Viability Kit with Liquid Counting Beads (BD Biosciences) was used according to the manufacturer's instructions on a FACS Canto Flow Cytometer (Becton Dickinson) equipped with three lasers. At least 30,000 events were acquired. Data analysis was carried out using BD FACSDiva 6.1.2 software (Becton Dickinson) 74 .
To estimate bacterial r/K growth strategy, water samples were inoculated onto R2A agar plates and incubated at 20 °C. Microbes that grew after 3 days were classified as r-strategists (fast growing opportunistic species at low population densities), and microbes that grew within an additional 4-7 days as K-strategists 75 .
To isolate phototrophs, one ml of water was inoculated in 50 ml of BG11 liquid medium (Promega) and incubated in a laboratory with moderate natural illumination and at room temperature. After 33 weeks, algae were identified by examination under a Nikon Eclipse TE 300 phase-contrast microscope equipped with a digital camera and Nikon software NIS Elements. Cells were identified according to the specialized key of Kol and Flint 34 . For cell-size estimation, 30 randomly selected cells were measured, and the average size and standard deviation were calculated (Supplementary Fig. S3). www.nature.com/scientificreports/ Total community DNA extraction, 16S amplicon sequencing, and bioinformatic analysis. Water (1.5 l) from melted ice blocks was filtered through two 0.22 μm pore-size filters (47 mm in diameter, Millipore). Filters were used to isolate total community DNA using a MoBio PowerWater DNA Isolation kit. DNA purity and concentration were determined spectrophotometrically using standard absorbance at 260 nm and 280 nm on a Perkin Elmer Lambda 25 UV-VIS Spectrometer equipped with UV CellTray (Hellma Analytics). The 260/280 absorbance ratios ranged from 1.63 to 1.74. The best DNA results from the duplicates, in terms of quantity and purity (A 260 /A 280 ), were used for subsequent sequencing. Amplification of 16S rDNA, library preparation and IonTorrent sequencing were performed at Omega, Ljubljana (Slovenia). Each sample was sequenced once. Ion 16S Metagenomics Kit (ThermoFisher Scientific, Waltham, MA, USA) was used to amplify V2-4-8 and V3-6, 7-9 regions of 16S rRNA in two separate PCR reactions consisting of initial denaturation at 95 °C for 10 min, 25 cycles of denaturation at 95 °C for 30 s, annealing at 58 °C for 20 s, and polymerisation at 72 °C for 20 s, followed by a final extension step at 72 °C for 7 min.
Amplified DNA was purified using Agencourt AMPure XP (Beckman Coulter, Brea, CA, USA). Purified PCR products were quantified on LabChip GX Touch using HT DNA High Sensitivity Reagent Kit (both PerkinElmer, Waltham, MA, USA). Fifty nanograms of amplicons were processed to make the DNA library using Ion Plus Fragment Library Kit and barcode adapters Ion Xpress Barcode Adapters Kit (both ThermoFisher Scientific, Waltham, MA, USA). Size distribution and number of processed libraries were evaluated with the LabChip GX instrument (PerkinElmer, Waltham, MA, USA). Each sample was adjusted to 20 picomolar concentration. Equal volumes of all samples were combined and processed in emulsion PCR and enrichment steps with One-Touch 2 and One-Touch ES systems using the Ion PGM Hi-QT View OT2 Kit (all ThermoFisher Scientific, Waltham, MA, USA). Finally, the libraries were sequenced in a 316 v2 chip on the Ion Personal Genome Machine (PGM) using the Ion PGM Hi-QT View Sequencing Kit following the manufacturer's protocol (all ThermoFisher Scientific Waltham, MA, USA).
Raw reads were analysed using cloud service Ion Reporter Software v.5.10, which implements QIIME's opensource bioinformatics pipeline 76 . Primer sequences were removed from the reads, and reads shorter than 150 bp were removed from the dataset. The sequences were aligned against the Curated Greengenes 16S reference library v13.5. Minimum alignment coverage was set to 90% and the read abundance filter set to 10 reads. Genus-level OTUs were determined at 97% sequence identity and species-level OTUs at 99% sequence identity. Rarefaction curves were determined based on Chao1 metrics. Vegan Community Ecology 77 and Bioconductor 78 packages were used within an R free-software environment 79 .
to calculate the alpha-diversity measures Chao1, ACE and Shannon index, to produce heatmaps, and to generate a Venn diagram.

Data availability:
raw sequencing reads were deposited in the NCBI Sequence Read Archive under the BioProject accession number PRJNA635015 corresponding to BioSample accession numbers of SAMN15015198 (sample Ice-1), SAMN15015199 (sample Ice-2) and SAMN15015200 (sample Ice-3).