Microbial diversity in Pyrenean ice caves: extreme cryogenic environments as analogues for extraterrestrial habitability

Abstract

Pyrenean ice caves are the least studies cryogenic environments that preserve perennial ice. These caves host microbial communities adapted to extreme oligotrophy, low temperatures, and episodic water availability, making them valuable analogues for subsurface habitats on icy planetary bodies. This study investigated four ice caves in the Central Pyrenees (Devaux, Cotiella A294, Sarrios 1 and Somola SO-01) to (i) characterize bacterial and microeukaryotic assemblages, (ii) assess how ice origin, physicochemical gradients, and cave geology structure these communities, and (iii) evaluate their relevance as terrestrial analogs for cold oligotrophic ecosystems. Amplicon sequencing of 16S and 18S rRNA genes, combined with detailed chemical profiling, revealed marked cave–specific differences associated with pH, major ions, short–chain organics, and ice–formation processes. Liquid water samples contained distinct assemblages dominated by ultra–small Patesibacteria, whereas firn–derived and congelation ice hosted stratified or hydrologically entrained communities. Microeukaryotic diversity was highest in light–exposed ice from Cotiella A294 and lowest in Sarrios 1, where fungal taxa prevailed. Redundancy analyses identified acetate, nitrate, sulfate, pH, and trace metals as the environmental variables most strongly aligned with microbial gradients. These findings provide a data–driven characterization of microbial organization in Pyrenean ice caves and offer empirical baseline parameters to inform future studies of microbial persistence and biosignature formation in cold terrestrial and extraterrestrial environments.

Pyrenean ice caves host microbial communities adapted to extreme oligotrophy and cryogenic stress, whose chemically driven diversity informs survival strategies and astrobiological biomarkers relevant to icy planetary environments.

Introduction

Ice caves contain perennial ice formed through snow transformation (firn) or the freezing of infiltrating water (Perșoiu and Lauritzen 2018). These environments often occur in regions where the mean annual external air temperature exceeds 0°C, making them particularly vulnerable to future climatic warming (Kern and Perşoiu 2013). Acting as localized thermal anomalies, their preservation depends primarily on cave morphology and internal ventilation patterns (Perșoiu and Lauritzen 2018, Wind et al. 2022, Sancho et al. 2025). At high altitudes in mountain ranges, subsurface ice still persists under current climatic conditions due to mountain permafrost, typically restricted to areas with mean annual air temperatures at or below freezing (Boeckli et al. 2012). For example, discontinuous mountain permafrost in the European Alps generally occurs above 2600–3000 m a.s.l. (Boeckli et al. 2012), while in the Central Pyrenees, the presence of discontinuous permafrost is above 2750 m a.s.l. (Serrano et al. 2019, Rico et al. 2021), and ice caves has provided additional evidence of sporadic permafrost occurrences (e.g. (Bartolomé et al. 2023).

Ice caves in the Central Pyrenees are predominantly developed within calcareous formations dating to the Upper Cretaceous and Paleogene. Most of these caves are concentrated around Monte Perdido, the highest calcareous massif in the Pyrenees. Despite their cryological significance, dedicated studies of these systems have historically been scarce, with early work limited mainly to speleological descriptions (Devaux 1929, Casteret 1953, St. Pierre 2007). This reflects the limited scientific attention they received until recent years, when more comprehensive research efforts began to emerge (Sancho et al. 2012, 2018, Serrano et al. 2018, Leunda et al. 2019, 2025). Among the ice caves of the region, the best–studied from a geological perspective are Devaux cave in the Monte Perdido massif (Bartolomé et al. 2023) and the A294 cave in the Cotiella massif (Belmonte-Ribas et al. 2014, Sancho et al. 2018, Leunda et al. 2019, Ruiz-Blas et al. 2023).

Since the early twentieth century, the Pyrenees has experienced a temperature increase of ∼1.3°C (Observatorio Pirenaico de Cambio Global 2018). The most visible part of the cryosphere, such as Pyrenean glaciers, has decreased by 23.2%, and average thickness has declined by 6.3 m between 2011 and 2020 (Vidaller et al. 2021). Climate model projections indicate that by 2050, annual maximum temperatures in the Pyrenees will rise by 1°C–4°C compared to the 1986–2005 reference period under the Representative Concentration Pathway (RCP) 8.5 scenario (Van Vuuren et al. 2011, Amblar-Francés et al. 2020, Bilbao Barrenetxea and Faria 2022). Similar to glaciers, but less known, cave ice deposits within Pyrenean are undergoing significant retreat (Belmonte-Ribas et al. 2014, Bartolomé et al. 2016, Sancho et al. 2025).

Aside from the relevance of ice caves as archives of past climate variability (Leunda et al. 2019, Perşoiu et al. 2021, Sancho et al. 2025) and as sentinels of ongoing global warming (Perşoiu et al. 2021, Obleitner et al. 2024, Leunda et al. 2025, Sancho et al. 2025), ice caves have also emerged as key sites for microbiological research (Brad et al. 2018, Itcus et al. 2018, Mondini et al. 2018). However, microbial diversity in these unique environments remains underexplored, and the functional potential of their microbial communities is still poorly understood. Investigated ecosystems include European limestone ice caves (Margesin et al. 2012, Hillebrand-Voiculescu et al. 2015, Itcus et al. 2018), icy volcanic caves in Oregon (Popa et al. 2012) and Hawaii (Teehera et al. 2018), and subglacial caves near Mt. Erebus (Tebo et al. 2015).

Such research is crucial for understanding the community structure and the potential functional responses of micro-organisms to temperature and ice chemistry (García-Descalzo et al. 2022). Ruiz-Blas provided a metagenomic and proteomic analysis of Cotiella A294 cave microbiome, revealing stratification driven by ice age and chemistry (Ruiz-Blas et al. 2023). Mondini et al. investigated Scarisoara cave in Romania, demonstrating taxonomic and functional shifts in active microbiota following simulated heat shock (Mondini et al. 2018). A study by Lange-Enyedi et al. in the Obstans ice cave (Carnic Alps) characterized psychrophilic bacteria capable of carbonate precipitation, highlighting cryogenic biogeochemical activity (Lange-Enyedi et al. 2024). Meanwhile, Seixas et al. examined an ice cave in Utah, USA, describing spatial differences in bacterial assemblages associated with ice, water, and cryogenic cave carbonates (Sancho et al. 2018). Investigations in wider karst contexts have also advanced understanding of subsurface microbiomes. Liu et al. revealed that nitrate, a common byproduct of acid rain, shapes microbial community structure and functional potential in karst caves, illustrating anthropogenic impacts on subterranean biospheres (Liu et al. 2023).

Beyond their cryological and microbiological significance, ice caves represent natural analogues for extraterrestrial environments (Alcazar et al. 2009). The stable, cold, and isolated conditions of these caves mimic features expected in subsurface habitats on icy planetary bodies such as Mars or the moons of Jupiter and Saturn (Garcia-Lopez and Cid 2017). Studies of microbial communities in Pyrenean ice caves shed light on their composition and inferred functional capacities in cold, isolated environments. Such research enhances our understanding of habitability in terrestrial cryospheric systems and supports the development of frameworks for detecting life in analogous extraterrestrial settings.

This study examines and compares four caves: the previously mentioned Devaux and Cotiella A294, as well as Sarrios 1 and the little–known Somola SO-01. The main objectives are to characterize and compare the microbial assemblages inhabiting these four perennial ice caves in the Pyrenees, to evaluate how cave geology, ice properties, and environmental factors shape community structure and functional potential, and to assess the extent to which these systems can serve as terrestrial analogues for life in icy planetary environments.

Materials and methods

Experimental workflow is summarized in Figure S1.

Study sites and sampling

Four perennial ice caves located in the Pyrenees were selected for this study (Fig. 1): i) Devaux and ii) Sarrios 1 in the Monte Perdido massif (Devaux 1929, Bartolomé et al. 2023), iii) A294 in the Cotiella massif (Belmonte-Ribas et al. 2014, Sancho et al. 2018, Leunda et al. 2019, Ruiz-Blas et al. 2023), and Somola SO-01 in Collarada massif (Espeleo Club Zaragoza, Grupo Litera Montaña 2015) (Fig. 1). Geographic coordinates, elevation, and geological context for each site are provided in Table 1. Sampling was conducted during the end summer/early fall seasons to ensure accessibility and minimize disturbance to fossil ice by seasonal ice formation. The ice surface was always cleaned prior to sampling in all cases. Ice samples were collected using a 3 cm stainless-steel auger coupled to a battery-powered drill. Sample characteristics are summarized in Table S1. Each sample was placed in sterile polypropylene containers, sealed, and transported under frozen conditions to the laboratory. All samples were transported at −20°C from the field to the laboratory at the Centro de Astrobiología (Madrid, Spain). Site-specific geological and climatic description of ice caves is described in Supplementary Material Section.

Figure 1.

Pyrenean ice cave locations and representative ice bodies.(A) Spatial distribution of ice caves in the Central and Central-Western Pyrenees. (B) Ice deposit within the Somola SO-01 cave, Collarada Massif. (C) Pristine ice body in a gallery of the Devaux cave, Monte Perdido Massif. (D) Ice body in the Sarrios 1 cave, Monte Perdido Massif. (E) Ice body in the A294 cave, Cotiella Massif. Photographs by Miguel Bartolomé and José Leunda.

Table 1.

Geographic locations of ices caves included in this research.

Cave	Massif	Altitude m a.s.l.	UTM coordinates
Sarrios 1	Monte Perdido	2721	30T 0743807 4 730 202
Devaux	Monte Perdido	2838	31T 0254375 4 730 574
Somola SO-01	Collarada	2547	30T 0709030 4 731 005
A294	Cotiella	2238	31T 0281079 4 710 145

Cotiella A294 cave

A294 ice cave (2238 m a.s.l.) is located in the Cotiella massif (Central Pyrenees, Spain), within a karst-periglacial environment developed in Upper Cretaceous limestones (Belmonte-Ribas et al. 2014) (Fig. S2). For consistency, the cave is referred to as Cotiella A294. A total of four ice samples were collected from this cave (COT1-COT4) (Ruiz-Blas et al. 2023).

Devaux cave

Devaux cave (2836 m a.s.l.) is situated on the northeastern wall of the Gavarnie Cirque (France), within the Monte Perdido massif, spanning the Parc national des Pyrénées (France) and the Parque Nacional de Ordesa y Monte Perdido (Spain) (Fig. S3). A total of six samples were collected, comprising four ice samples and two liquid water samples (drip water and cave–river water) (DEV1-DEV6) (Muñoz-Hisado et al. 2023).

Sarrios 1 cave

Sarrios 1 (2721 m a.s.l.) is located on the southern flank of the Monte Perdido massif, at the base of the Sarrios ledge (Bernand and Van Thienen 1987) (Fig. S4). The ice body consists of a basal layer of transformed snow (firn; samples SAR40 and SAR85) overlain by a horizontal unconformity, above which the sequence is composed of congelation ice (samples SAR250, SAR400, SAR550, and SAR650). In total, six samples were collected from this cave.

Somola SO–01 cave

The Somola SO–01 ice cave (2547 m a.s.l.) lies on the southern face of Peña Nevera (Collarada Massif), at the boundary between a limestone cliff and a talus slope (Fig. 1). A total of five samples were obtained (SOM1-SOM5). Samples SOM1-SOM3 correspond to ice below the unconformity, whereas SOM4 and SOM5 represent ice above the unconformity (Fig. S5).

Sample processing

Ice samples were decontaminated following previously described procedures (Martinez-Alonso et al. 2019). Each section of the ice core was removed from −20°C storage and rinsed with 0.22 µm-filtered Milli-Q water (Millipore, USA). The ice was then allowed to thaw, and the meltwater was filtered separately through 0.22 µm pore-size filters using a vacuum pump inside a laminar flow hood previously disinfected with ethanol. Both filters and meltwater were used for DNA extraction and chemical analysis. To monitor potential laboratory contamination, sterile Milli-Q water was subjected to the same analytical processes. All laboratory steps were performed under aseptic conditions using a UV-irradiated laminar flow hood, ethanol-cleaned tools, and sterilized gloves.

Chemical composition of water and ice samples

Elemental analysis of Al, Ba, Br, C, Ca, Cd, Cl, Co, Cr, Cu, Fe, I, K, Li, Mg, Mn, Na, Ni, P, S, Sb, Se, Si, and Zn were analyzed by inductively coupled plasma-mass spectrometry (ICP-MS) on a Perkin Elmer ELAN9000 ICP-MS quadrupole spectrometer (Martinez-Alonso et al. 2019). Concentration of fluoride, lactate, acetate, formate, chloride, nitrate, phosphate, sulphate, sodium cation, ammonium cation, potassium cation, magnesium cation, and calcium cation were analyzed in each filtered sample by ion chromatography using suppressed conductivity detection in an 861 Advance Compact IC system (Metrohm AG, Herisau, Switzerland). Chromatograms were recorded using the Metrohm IC Net 2.3 SR4 software. The system was run in the isocratic mode with the columns Metrosep A Supp 7 250/4,0 and Metrosep C3 250/4,0 at 45°C. pH was measured using a Mettler Toledo FiveGo™ F2 pH/mV portable meter (accuracy ±0.01 pH, IP 67 waterproof).

DNA isolation and sequencing

Total DNA was isolated from 0.22 μm filters using DNeasy PowerWater Kit (Qiagen, Germany). DNA concentration was measured with a Qubit 1 fluorimeter using the high-sensitivity dsDNA quantitation assay (Thermo Fisher Scientific, USA). To assess microbial diversity, V3-V4 regions of the 16S rRNA gene were amplified by PCR using 341F/805R primers (CCTACGGGNGGCWGCAG/GACTACHVGGGTATCTAATC) (Herlemann et al. 2011). For microeukaryotes, V4-V5 regions of the 18S rRNA gene were targeted using 563F/1132R (GCCAGCAVCYGCGGTAAY/CCGTCAATTHCTTYAART) (Garcia-Lopez et al. 2022). Amplification products were sequenced using Illumina MiSeq platform with paired-end reads (2 × 250–2 × 300), to obtain ∼100 000 reads per sample. Sequencing was performed by Genomics Unit, Madrid Science Park Foundation, Spain. Control samples for contamination were processed using the same protocols. The resulting sequences differed markedly from those of the experimental samples, thereby validating the experimental conditions.

Amplicon sequencing workflow and analysis

Preprocessing and quality control

Raw FASTQ files were initially inspected using FastQC to evaluate sequence quality (Andrews,n.d.). Quality filtering and adapter removal were performed with Fastp, ensuring high-quality reads for downstream analysis. Subsequent steps were carried out within a QIIME2 environment v2024.10 (Bolyen et al. 2019). Primer sequences and Illumina universal adapters (AGATCGGAAGAG) were trimmed using Cutadapt via the QIIME2 plugin (Martin 2011). Sequence denoising, trimming, and paired-end merging were conducted with DADA2, applying optimized parameters to maximize the recovery of amplicon sequence variants (ASVs) while minimizing data loss during critical steps such as merging of 18S rRNA reads (Callahan et al. 2016). All selected parameters are detailed in Table S2. Whenever possible, parameters were optimized to minimize the inclusion of low-quality bases. Parameter optimization calculations were based on the theoretical length of each amplicon, minus the overlapping region. Mismatch error thresholds were refined through iterative optimization steps to maximize read-merging efficiency.

Filtering and rarefaction

Resulting ASVs were size-filtered to remove potential chimeric sequences that could bypass previous quality controls. For 16S rRNA, only amplicons longer than 350 bp were retained, while for 18S rRNA the threshold was set at 500 bp. To focus exclusively on microbial communities, ASVs classified as Phragmoplastophyta, Vertebrata, and Nematozoa were considered contaminants and removed.

Rarefaction curves were generated using the QIIME2 diversity alpha-rarefaction method, and rarefaction was performed with the subsample function from scikit-bio. Sampling depths were selected to retain all samples. The chosen values are provided in Table S2. Differences in ASV counts and Shannon diversity at the selected depth were tested using the Kruskal-Wallis test using the kruskal method from scipy.stats. When P < 0.05, Dunn’s post hoc test was applied for pairwise comparisons, and p-values were adjusted using the Benjamini–Hochberg false discovery rate (FDR) method.

Taxonomic assignment

Taxonomic annotations were obtained using the qiime feature-classifier classify-sklearn method from QIIME2. This approach requires region-specific, trained Naïve Bayes classifiers (Bokulich et al. 2018). Both classifiers (16S rRNA and 18S rRNA) were built from the SILVA_138.2_SSURef_NR99 database and curated using RESCRIPt (Robeson et al. 2021).

Beta diversity analyses

Ordination analyses were based on the Centered Log Ratio (CLR) transformation of rarefied tables, using 1 as a pseudo-count. Rarefaction was applied to minimize biases associated with differences in the total number of features (i.e. ASVs) across samples. The CLR transformation was performed using the skbio.stats.decomposition module.

Principal component analysis (PCA) was then applied to the transformed tables using sklearn.decomposition. Statistical significance of potential differences between sample groups, based on the distance matrix, was assessed using PERMANOVA using skbio.stats.distance.permanova package. The analysis included only variables without missing values and with at least two samples per group, using 9999 permutations. To evaluate how variation in microbial community composition could be explained by environmental variables, a Redundancy Analysis (RDA) was performed using skbio.stats.ordination.rda (parameters: scale_Y=False, scaling=2). Rarefied and CLR-transformed ASV and genus tables were used as the response matrix, while CLR-transformed environmental data served as the explanatory matrix.

Data analysis and multivariate statistical analysis

PCA was performed to explore patterns in the dataset and detect potential differences among samples. Prior to analysis, all numerical variables were standardized to z-scores to ensure comparability across different scales. Missing values were imputed using the mean of the corresponding variable to minimize bias and retain the maximum amount of information. For PCA involving categorical variables, these were converted into dummy numerical variables using one-hot encoding. PCA and subsequent visualization were conducted in QIIME2 v2024.10 using the q2-diversity plugin (Bolyen et al. 2019).

Results and discussion

Environmental context, chemical patterns, and ordination of samples

Water and ice chemistry showed substantial variation across caves (Table S1; Fig. S6). Overall pH ranged from slightly acidic to alkaline (6.9–9.1), with Sarrios 1 samples presenting near–neutral values and Devaux liquid samples exhibiting the highest pH. Across all sites, solute composition was dominated by calcium, sodium, chloride, and sulfate, although their concentrations differed markedly among caves. Liquid samples from Devaux contained the highest ionic loads, whereas Somola SO-01 samples showed lower total solute concentrations but consistent enrichment in Ca and Si. Sarrios 1 samples were characterized by comparatively high silicon and sulfur/sulfate values, and Cotiella A294 samples displayed moderate ion concentrations typical of transformation–derived ice. These contrasts established the main physicochemical gradients structuring subsequent ordinations.

PCA of environmental variables (Fig. 2) showed that sample separation was primarily driven by sample type (ice vs. liquid) and cave of origin. DEV1 and DEV2 formed a distinct cluster along PC1, consistent with their high solute concentrations. Ice samples grouped by cave along PC2, with Somola SO-01 forming the tightest cluster and Sarrios 1 showing broader variability. Thus, environmental heterogeneity was greatest in the Monte Perdido massif (Devaux, Sarrios 1) and lowest in Somola SO-01. A second PCA including categorical variables (Fig. S7) produced similar patterns, confirming the chemical distinctiveness of liquid samples and cave specific clustering among ice samples.

Figure 2.

PCA based on environmental variables. The ordination illustrates differences and similarities among samples according to chemical composition and categorical factors. Liquid samples are clearly separated along PC1, while ice-derived samples cluster by cave of origin, highlighting cave-specific environmental characteristics.

Liquid water samples, DEV1 and DEV2, contained higher ionic loads than ice samples, consistent with solute exclusion during freezing (DeGrandpre et al. 2021) and with mineral enrichment in flowing water within Devaux (Bartolomé et al. 2023). Among ice samples, Sarrios 1 showed greater variability, likely reflecting temperature fluctuations, variable hydrological inputs, and differences in ice–formation processes. Depth–stratified sampling in Sarrios 1 may also capture longer–term changes linked to atmospheric deposition and episodic meltwater infiltration over centuries (Moreno et al. 2021). In contrast, Somola SO-01 and Cotiella A294 exhibited more consistent chemical signatures, suggesting greater environmental stability.

Microbial community composition

Quality control retained a substantial number of ASVs in both datasets. For the 16S rRNA dataset, a total of 2 820 858 ASVs were recovered, with the lowest count observed in sample SAR550 (44 399 ASVs; Table S4). In the 18S rRNA dataset, 668 939 ASVs were detected, with a minimum of 859 ASVs in sample SOM2 (Table S5). After taxonomy-based filtering to remove non-target eukaryotic sequences (see Methods), the lowest retained ASV count was 725 (sample SOM2).

To minimize biases in diversity analyses due to uneven sequencing depth, all count tables were rarefied to the lowest depth per dataset: 44 399 ASVs for 16S rRNA and 725 ASVs for filtered 18S rRNA. Rarefaction curves confirmed that these depths retained the majority of expected ASVs in the 16S rRNA dataset (Fig. S8A). For the 18S rRNA dataset, some samples, particularly those from Cotiella A294 cave, still exhibited a positive slope, suggesting that additional ASVs might remain undetected (Fig. S8B). Correlative curves using the Shannon index indicated that the selected depths captured the bulk of diversity in both datasets (Fig. S8B, S8D).

Diversity patterns across caves

Kruskal–Wallis tests revealed significant differences in bacterial ASV counts among caves (P < 0.05). However, pairwise comparisons using Dunn’s test with multiple-testing correction did not identify any specific cave pairs with statistically significant differences at the chosen threshold (Table S3). Rarefaction curves and boxplots (Fig. S9) suggest that Sarrios1 cave exhibited lower ASV counts than other caves, although Shannon diversity did not differ significantly among caves.

For microeukaryotic ASVs, Kruskal–Wallis tests again indicated significant differences (P < 0.05). Boxplots revealed a clear tendency for Cotiella A294 to have higher ASV counts than other caves (Fig. S9). Dunn’s post-hoc tests confirmed a significant difference between Cotiella A294 and Devaux (Table S3), while comparisons with Sarrios1 or Somola SO-01 were not significant but approached the threshold. Shannon diversity mirrored ASV count patterns: Cotiella A294 was the most diverse cave, whereas no significant differences were detected among the remaining caves.

Taxonomy of microbial communities

Bacterial communities

Bacterial communities were dominated by Actinomycetota, Pseudomonadota, Bacteroidota, Patesibacteria, and Bacillota, which together accounted for over 90% of ASVs (Fig. 3A; Table S4). This composition is consistent with other cryospheric environments (Tebo et al. 2015, Itcus et al. 2018, Paun et al. 2019, Ruiz-Blas et al. 2023). Strong variability among samples was observed: Patesibacteria dominated the liquid samples DEV1-DEV2, whereas Bacillota peaked in SAR550 and DEV4. Actinomycetota and Pseudomonadota remained broadly abundant except in DEV1-DEV2.

Figure 3.

Taxonomic composition of microbial communities across Pyrenean ice caves. (A) Relative abundance of dominant bacterial phyla. (B) Distribution of the 20 most abundant bacterial genera, highlighting cave-specific patterns. (C) Relative abundance of major microeukaryotic phyla. (D) The twenty most abundant microeukaryotic genera are represented, although genus-level resolution was only achieved for a subset of taxa. Plots illustrate taxonomic variability among caves and sample types.

At the genus level, 68.83% of ASVs were classified, a higher resolution than previously reported for ice caves (Itcus et al. 2018). The most abundant genera belonged to the dominant phyla, with Lysobacter widespread and Daejeonella, Polaromonas, Cryobacterium, and Aeromicrobium present across caves (Fig. 3B). Other genera such as Actimicrobium, Methylobacter, Oryzihumus, and Rhodoferax were restricted to specific sites. These taxa are commonly reported in glacier and ice-cave ecosystems (Margesin et al. 2012, Singh et al. 2014, Paun et al. 2019, Fair et al. 2024, Gladkov et al. 2024, Lange-Enyedi et al. 2024) and some may remain metabolically active in ancient ice (Paun et al. 2019). Liquid samples DEV1-DEV2 differed markedly, lacking most dominant genera and being enriched in Candidatus Nomurabacteria, consistent with their Patesibacteria-dominated profile.

Comparison with alpine microbial datasets suggests that several detected genera (e.g. Polaromonas, Pseudomonas, Arthrobacter, Cryobacterium, Sphingomonas) likely originate from aeolian deposition or meltwater inputs, as they are common in alpine soils, aerosols, and snowpacks (Margesin et al. 2012, Donhauser and Frey 2018, Fiore-Donno et al. 2024). At the same time, distinct cave–specific signatures indicate the persistence of autochthonous lineages shaped by local physicochemical gradients.

Microeukaryotic communities

Microeukaryotic communities were dominated by Ascomycota and Basidiomycota, followed by Chlorophyta, Ochrophyta, and Ciliophora (Fig. 3C; Table S5). Their distribution varied strongly among caves: Sarrios1 was dominated by Ascomycota and Basidiomycota, whereas Cotiella A294 showed more balanced compositions. Photosynthetic groups (Chlorophyta, Ochrophyta) were mainly detected in exposed or entrance-proximal samples. Other phyla, such as Chytridiomycota and Diatomea, were abundant only in specific sites (e.g. DEV1, DEV2).

Genus-level resolution was limited, with only 12 of the 20 most abundant groups classified (Fig. 3D). Cladosporium and Naganishia were the most widespread genera, though not universal.

Ecological interpretation

Microeukaryotes constituted a minor proportion of total ASVs, reinforcing the predominance of bacteria in cold microbial communities (Jungblut et al. 2012). Despite methodological limitations in directly comparing 16S rRNA and 18S rRNA datasets, the disparity in ASV counts supports this observation. The high taxonomic heterogeneity across caves, sample types, and environments, reflects interactions between hydrological regimes, cave morphology, exposure gradients, and potential in situ metabolic activity in ancient ice. The prevalence of certain microbial taxa in specific environments suggests adaptation to local conditions and niche differentiation. To fully understand these dynamics, further diversity analyses that include rare taxa and link taxonomic and functional traits will be essential.

Microbial community distribution

Ordination patterns in bacteria

Beta diversity analysis indicated clear compositional differentiation among the ice caves (Fig. 4A). Liquid samples from Devaux (DEV1-DEV2) separated strongly along PC1, while the remaining Devaux samples clustered with Sarrios1. Cotiella A294 occupied an intermediate position, and Somola SO-01 was most distinct along PC2. Five ASVs: Daejeonella (two ASVs), Methylobacter, Actimicrobium, and one unclassified Vicinamibacterales ASV, contributed most to sample separation (Table S6). These ASVs were absent from Devaux and enriched mainly in Somola SO-01, whereas Devaux samples were characterized by Parcubacteria, Candidatus Yanofskybacteria, and Candidatus Woesebacteria, consistent with the prevalence of Candidatus Patesibacteria in oligotrophic waters (Herrmann et al. 2019, Ruiz-González et al. 2025).

Figure 4.

Beta diversity ordinations of microbial communities across Pyrenean ice caves. (A) PCA based on bacterial ASVs showing cave-specific clustering and separation of liquid samples. (B) PCA at bacterial genus level highlighting differences among caves. (C) PCA based on microeukaryotic ASVs revealing two major clusters along PC1. (D) PCA at microeukaryotic genus level showing similar patterns to ASV-based ordination. Ordinations illustrate compositional variability driven by cave identity and sample type.

Genus-level ordination also revealed cave-specific clustering (Fig. 4B). Methylobacter and Actimicrobium were influential genera, abundant in Cotiella A294 and Somola SO-01 and absent from Devaux. Differences between ASV- and genus-level patterns highlighted pronounced microdiversity, with shifts such as Somola SO-01 aligning with Cotiella A294 at the genus level (Auladell et al. 2022). Overall, bacterial communities showed strong cave-specific signatures, with Devaux enriched in Candidatus Patesibacteria and Cotiella A294 and Somola SO-01 enriched in Methylobacter and Actimicrobium, consistent with patterns from snowpacks and glacier forefields (Lutz et al. 2016, Auladell et al. 2022).

Ordination patterns in microeukaryotes

Microeukaryotic ordinations also showed clear cave–specific structure (Fig. 4C and D). Cotiella A294 consistently separated along PC1, driven by ASVs assigned to Koliella and Chrysophyceae, with the latter also present in Somola SO-01. Sarrios1 was dominated by Leotiomycetes (ASV5), and Somola SO-01 by Naganishia (ASV3), also detected in SAR250. Genus-level patterns mirrored ASV-level results, reflecting lower microeukaryotic microdiversity (Fig. 4D). Chrysophyceae remained a major contributor to Cotiella A294, consistent with oligotrophic alpine and polar waters (Izaguirre et al. 2021). Leotiomycetes, Debaryomyces, and Naganishia in Sarrios1 and Somola SO-01 matched fungal assemblages reported in glacial and permafrost environments (Margesin et al. 2012, Bourquin et al. 2022). The presence of Naganishia, including taxa related to N. vishniacii, aligned with its known adaptation to extreme polar conditions (Onofri et al. 2019, Nizovoy et al. 2021). PERMANOVA confirmed significant differences among caves, massifs, and ice origins (P = 0.0001), with cave identity emerging as the main structuring factor (Table S7).

Environmental interaction

Following previous analyses, the influence of environmental variables on microbial community structure was investigated. The PCA performed exclusively on the chemical composition of samples revealed a distribution that did not correspond to any of the four CLR-based ordinations of microbial communities. This result is coherent, as it indicates that microbial composition is not exclusively determined by chemical factors but is also influenced by geological, historical, or even stochastic processes. To explicitly evaluate the relationship between environmental gradients and microbial community structure, redundancy analyses (RDA) were performed (Fig. 5).

Figure 5.

RDA showing the influence of environmental variables on microbial community structure. (A) RDA based on bacterial ASVs. (B) RDA at bacterial genus level. (C) RDA based on microeukaryotic ASVs.(D) RDA at microeukaryotic genus level. Ordinations illustrate the alignment of chemical gradients (e.g. acetate, nitrate, trace metals) with patterns of microbial variation, while confirming that community composition is shaped by multiple ecological factors beyond chemistry alone.

For bacteria, RDA ordinations based on ASVs and genera reproduced the spatial patterns observed in CLR-PCA (Fig. 5A and B). In microeukaryotes, RDA configurations mirrored the corresponding CLR ordinations (Fig. 5C and D), reflecting the arbitrary orientation of RDA axes. In all cases, relative distances and clustering among samples remained consistent with CLR-based results, and environmental gradients aligned with microbial variation, although the variables contributing most strongly differed among analyses.

In the bacterial ASV-based RDA, acetate, ammonium, chloride, sulfur, and chromium were the variables most strongly associated with community gradients. Acetate and ammonium were positively associated with an ASV assigned to Methylobacter, consistent with its ability to use acetate under oligotrophic conditions (Schneider et al. 2012). Chloride, sulfur, and chromium were negatively correlated with two ASVs from Cotiella A294 (order Vicinamibacterales and genus Daejeonella), whereas a second Daejeonella ASV showed a weaker relationship with these ions. At the genus level, gradients were primarily associated with acetate, Cr, Li, NO₃⁻, and S. Acetate correlated positively with Actimicrobium, predominant in Cotiella A294, and with Methylobacter, while Candidatus Woesebacteria and Candidatus Yanofskybacteria were negatively associated with acetate and positively with K and NO₃⁻, enriched in liquid samples (Fig. S6). The strong negative correlation between Methylobacter and chromium at the genus level contrasted with the ASV-level pattern, highlighting intra-genus heterogeneity.

For microeukaryotic ASVs, the variables most strongly associated with gradients were pH, Si, formate, Mg, and nitrate (Fig. 5C). Most gradients, higher in liquid samples, were negatively correlated with the most relevant ASVs, except for an unclassified Leotiomycetes ASV, which was positively associated with formate. Genus-level ordinations showed similar patterns, with Mg, pH, nitrate, Li, and Ca as the main variables. As in CLR ordinations, taxonomic resolution had only a marginal effect on microeukaryotic RDA results.

Ecological interpretations, in particular those associated with specific taxa should be approached with caution. Although these interpretations are supported by statistical analyses and are consistent with the published literature, they require further detailed evaluation for confirmation. Moreover, these relationships represent an initial approximation to the ecological dynamics of a poorly studied ecosystem. Although the sample size in this study is adequate given the difficulty of sample collection, larger-scale studies are required to confirm these trends. Unconstrained ordination of environmental variables did not reproduce the structure observed in microbial CLR-based ordinations, indicating that overall chemical similarity among samples is not directly translated into similarity in microbial community composition. However, redundancy analysis revealed that a subset of environmental gradients is aligned with the dominant patterns of microbial variation. This suggests that, while environmental variables do not fully determine microbial community structure, they contribute to shaping compositional gradients in combination with other ecological factors.

Similar associations between microbial composition and chemical gradients have been reported in oligotrophic aquatic systems and cryospheric environments. For example, acetate and ammonium have been identified as key drivers for methylotrophic bacteria in Antarctic lakes and glacier forefields, where organic carbon availability is extremely limited (Schneider et al. 2012, Lutz et al. 2016). The correlation of chromium and sulfur with specific bacterial taxa reflects patterns observed in subglacial habitats, where trace metals influence microbial metabolism and community assembly (Margesin et al. 2012). For microeukaryotes, the association of Leotiomycetes with formate and pH parallels findings from permafrost soils and ice caves, where fungi exploit simple organic acids and tolerate pH fluctuations (Bourquin et al. 2022).

Ice origin and microbial signal preservation

The origin of the ice appears to influence both the structure and the preservation of microbial signatures within the studied caves. In Cotiella A294, the ice body consists of stratified firn formed through the progressive transformation of deposited snow. This process incorporates micro-organisms and atmospheric particles incrementally, producing vertically layered archives in which microbial assemblages may reflect long–term atmospheric deposition, seasonal inputs, and snowpack microbial communities characteristic of high–mountain environments. In contrast, Devaux contains congelation ice formed by episodic flooding and subsequent freezing of infiltrating water, a process that can introduce micro-organisms mobilized from surface soils, meltwater runoff, or internal hydrological pathways. Flood–derived congelation ice is therefore expected to preserve more mixed and hydrologically transported microbial inputs and may exhibit lower stratigraphic continuity than firn–derived ice.

These contrasting formation mechanisms help explain differences observed in microbial heterogeneity and environmental associations among caves. Firn–derived ice, with its gradual accumulation and limited liquid phase, can preserve stratified microbial layers that track historical atmospheric and cryospheric inputs. In contrast, congelation ice can incorporate micro-organisms during transient liquid phases, enabling the entrainment of ultra–small cells and solutes, and potentially homogenizing microbial signatures across events. Clarifying these distinctions emphasizes that ice origin provides important context for evaluating whether microbial communities in perennial ice represent primarily allochthonous deposition archives, hydrologically mobilized assemblages, or mixtures shaped by both depositional and freezing processes.

Astrobiological relevance

The environmental gradients identified in these ice caves provide a framework for generating hypotheses about how microbial communities may respond to physicochemical constraints in icy extraterrestrial environments (Table S8). Acetate and formate showed coherent patterns of covariation with specific bacterial and fungal groups, indicating that these short–chain organics may structure microbial assemblages under cold, oligotrophic conditions. Although metabolic activity cannot be inferred from correlations alone, the observed co–occurrence identifies compounds and taxa that frequently arise together in cryogenic environments, providing baseline substrate-community relationships that are useful when defining potential habitability parameters for icy worlds (Wynn-Williams et al. 2001). These associations highlight chemical features that could serve as indicators of microhabitats where carbon availability is sustained in frozen settings.

Mineral-microbe associations were also evident. Variations in chromium- and sulfur–related ion concentrations corresponded with shifts in community composition, illustrating the influence of local mineralogy on microbial structure. Given that these elements are common constituents of planetary crusts (Vance et al. 2016), their role in shaping microbial patterns in the caves offers a terrestrial reference point for assessing how mineralogical contexts may constrain communities in Martian or icy–moon environments. Associations between fungal taxa, pH and formate suggest that eukaryotic groups respond to specific geochemical regimes, contributing additional criteria for identifying environmental niches relevant to polyextremophilic lineages documented in other cryogenic analogs (Postberg et al. 2018, Onofri et al. 2019).

Clear differences in beta–diversity among caves indicate that each site captures distinct facets of microbial organization under cold and nutrient–limited conditions. The strong separation of Devaux liquid samples, combined with the presence of ultra–small lineages such as Candidatus Patesibacteria, aligns with ecological patterns described for low–biomass, low–energy systems (Herrmann et al. 2019, Ruiz-González et al. 2025). These communities illustrate how micro-organisms partition chemically heterogeneous cryoenvironments, providing empirical examples of microbial structures that may emerge in brine–rich microhabitats pertinent to astrobiological exploration.

In Somola SO-01 cave, psychrotolerant fungi and yeasts such as Naganishia and Debaryomyces dominated the assemblages, consistent with environments influenced by freeze-thaw dynamics and limited nutrient availability. These groups, previously noted in extreme cryospheric settings (Postberg et al. 2018, Onofri et al. 2019, Nizovoy et al. 2021), offer insight into eukaryotic strategies that persist under cold stress and intermittent hydration, features relevant to the characterization of terrestrial analog sites used to frame astrobiological hypotheses.

Cotiella A294 samples were characterized by phototrophic and mixotrophic microeukaryotes such as Koliella and Chrysophyceae. While these patterns reflect light–exposed microhabitats within the cave system, sustained photosynthesis is unlikely in the subsurface environments of icy moons such as Europa. For such analogs, the relevance of Cotiella A294 lies not in photosynthesis itself, but in the presence of organisms capable of persisting under low–energy conditions and utilizing alternative metabolic pathways. In the context of Europa, potential analogies relate more closely to chemolithotrophic or radiolytically supported metabolisms, which rely on oxidants and reduced compounds generated by ice-rock interactions or radiolysis rather than light–dependent processes.

Sarrios 1 samples exhibited high abundances of Leotiomycetes and anaerobic–tolerant yeasts, reflecting microbial structuring influenced by freeze-thaw cycles and limited organic matter. These patterns mirror ecological processes described for fungi in cold, low–nutrient environments (Margesin et al. 2012, Onofri et al. 2019), offering additional examples of microbial groups that persist under physicochemical constraints comparable to those considered in analog studies of Mars and icy moons.

Differences between ASV- and genus–level patterns in bacterial communities, compared to their stronger alignment in microeukaryotes, highlight the role of taxonomic resolution in interpreting microbial structure. These patterns demonstrate the value of multi–level analyses when identifying community features that may persist across environmentally diverse cryogenic settings.

Overall, the four Pyrenean ice caves provide a set of environmentally distinct cold ecosystems in which microbial communities respond to well-defined chemical and physical gradients. These results identify ecological variables, community patterns and substrate associations that can inform the selection of microbial targets, environmental parameters and biosignature types for future astrobiological studies, offering data–driven constraints rather than speculative models for life in icy extraterrestrial environments.

Conclusion

This study set out to characterize the microbial diversity of four perennial ice caves in the Central Pyrenees, evaluate how environmental and geochemical factors structure these communities, and assess the relevance of these systems as terrestrial analogs for icy extraterrestrial environments. The results demonstrate that each cave hosts a distinct assemblage shaped by local physicochemical conditions, reflecting the combined influence of cave morphology, hydrological regime, and geochemical gradients on community organization.

By integrating microbial, chemical, and environmental datasets, the study identifies a set of variables, such as short–chain organics, trace ions, pH regimes, and ice–formation processes, that consistently align with differences in microbial composition across sites. These factors represent key ecological dimensions that may similarly influence microbial distribution and functional potential in cold, oligotrophic environments beyond Earth. The identification of chemically structured microbial assemblages across multiple cave systems provides comparative evidence of how micro-organisms partition cryogenic habitats under spatially heterogeneous conditions.

In an astrobiological context, the Pyrenean ice caves examined here offer complementary natural laboratories for evaluating how microbial communities may respond to low temperatures, limited nutrients, and mineralogical constraints. The environmental gradients and community patterns documented in this work supply empirically grounded parameters that can inform future analog studies, experimental simulations, and the development of biosignature–focused investigations relevant to icy planetary bodies.

Overall, the study establishes a baseline framework for understanding how microbial communities are shaped within perennial ice environments and highlights the value of multi–site cryospheric research for refining models of habitability and biosignature persistence in frozen extraterrestrial settings.

Conflicts of interest

None declared.

Funding

The work was financially supported by the Spanish Ministry of Science and Innovation/State Agency of Research MCIN/AEI/10.13039/501100011033 (reference PID2023-146641NB-C21) and National Parks Autonomous Agency (OAPN) (reference 2552/2020). Eva Garcia-Lopez is recipient of a Fellowship (PTA2022-021737-I) by the Spanish Ministry of Science and Innovation/State Agency of Research MCIN/AEI/10.13039/501100011033. Fernando Puente-Sánchez was supported by grant 2022-04801 from the Swedish Research Council.

Data availability

The raw sequence generated for this study can be found in the NCBI Short Read Archive (SRA) under the accession numbers PRJNA663780 (Cotiella A294), PRJNA1099162 (Devaux), PRJNA962306 (Sarrios1), and PRJNA1392111 (Somola SO-01).