The microbiome of cryospheric ecosystems

The study addresses the lack of an integrative, global understanding of microbiome structure and function across diverse cryospheric ecosystems (snow, ice, glaciers, permafrost, and glacier-influenced waters) despite their rapid loss due to climate change. Prior work has revealed adaptations of psychrophiles and provided insights from specific habitats (e.g., permafrost), but a comprehensive, cross-ecosystem perspective was missing. The authors aim to assemble and reanalyze global 16S rRNA and metagenomic datasets to identify a distinct cryospheric microbiome, characterize its phylogenetic patterns and potential evolutionary history, and delineate its functional repertoire relevant to biogeochemical processes under cold, oligotrophic conditions.

The paper situates itself within research on microbial ecology of cryospheric habitats and psychrophilic adaptations, referencing classical microbiology and sequencing-based studies that revealed cold-adaptation mechanisms and metagenomic insights from specific ecosystems (e.g., permafrost soils). It notes reported prevalence of Alpha- and Gammaproteobacteria in cryosphere literature, the role of genome streamlining in cold adaptation, and links between horizontal gene transfer and cold-adaptation genes. It highlights the absence of a unified, cross-ecosystem catalogue and the underrepresentation of cryospheric sequences and taxa in public databases as key gaps.

• Datasets: Curated 695 cryospheric 16S rRNA gene samples (snow/ice, glacier-affected freshwaters, permafrost soils, coastal ocean under glacier influence) and 3552 non-cryospheric samples (temperate/tropical lakes and soils). Two standard primer pairs were analyzed separately: PP1 (341f–785r) and PP2 (515f–806r). Total reads: 241,502,708 paired reads; resulting ASVs: 530,254 (PP1) and 410,931 (PP2). Analyses were performed at genus level. Metagenomes: 34 cryospheric and 56 non-cryospheric metagenomes spanning glacier surfaces, ice-covered lakes, and Antarctic soils, totaling 2,427,818,072 paired reads and yielding 41,068,842 predicted gene sequences.
• 16S processing: Raw fastq from ENA; quality trimming (Trimmomatic); denoising (DADA2 in QIIME2); removal of chloroplast/mitochondria and non-bacterial sequences; samples with <5000 reads removed. Taxonomy assigned with QIIME2 feature-classifier against SILVA 138 nr99; samples with <75% ASVs assigned to phylum and <50% to genus removed.
• Machine learning classification: Logistic regression (scikit-learn) on presence–absence ASV tables (presence threshold 0.005 relative abundance), stratified 5-fold CV repeated 40 times; balanced accuracy, precision, recall computed with class imbalance correction. Odds ratios derived from model coefficients.
• Phylogenetics: Separate phylogenies for PP1 and PP2 ASVs using MAFFT (FFT-NS-2), TrimAl (gt 0.95), IQ-TREE (GTR, -fast). Calculated Sorensen’s phylogenetic index and β-MNTD across random subsamples (50 iterations, 50 cryospheric and 50 non-cryospheric samples per iteration) using Picante. α-MPD, α-MNTD, and α-PD computed and compared with linear models controlling for richness and dataset.
• Community structure: NMDS (k=2; Bray–Curtis; stress=0.206) and PERMANOVA (adonis2) to test ecosystem type and primer effects; pairwise comparisons with pairwise.adonis2 (Bonferroni correction). Core microbiome defined as genera with ≥0.1% abundance and ≥20% prevalence across cryosphere and present in at least one sample in each of four ecosystem types. Probability of presence modeled with logistic binomial regression (ecosystem and primer as fixed effects).
• Differential abundance: Genus-level aggregation; ANCOM v2.1 with W>0.7 and positive CLR mean difference; primer pair included as random effect; zero-cut=0.995 (genera present in ≥21 samples).
• Reference genomes analysis: Compiled 13,414 RefSeq bacterial genomes across 660 genera in dataset; retrieved genome sizes; predicted ORFs (Prodigal); codon usage and growth predictions (gRodon, coRdon); tested GC content differences (Wilcoxon with Holm correction) and amino acid enrichment (DESeq2 on amino acid counts).
• Metagenomic functional analysis: Assembly with MEGAHIT within IMP; ORFs predicted with modified Prokka/Prodigal; functional annotation via HMMs and KEGG orthologs (hmmsearch); read counting with featureCounts; DESeq2 for KO enrichment (FDR<0.01, log2FC>1); pathway reconstruction with KEGGdecoder; mapping enriched KOs to contig taxonomy (Kraken2, NCBI Taxonomy).
• Gene clustering of functional space: Clustered 41,068,842 predicted genes with MMseqs2 linclust (30% identity, 80% coverage); retained largest clusters with ≥30 sequences in ≥2 samples (n=12,125); generated consensus sequences (MAFFT, EMBOSS cons) and searched against UniProt TrEMBL; computed within-cluster identity (Clustal distmat); compared UniProt match rates and identities across KEGG-assigned, ambiguous, and unassigned clusters for cryosphere vs non-cryosphere vs shared.

• Distinct cryospheric microbiome: Logistic regression classified cryospheric vs non-cryospheric communities with high accuracy (balanced accuracy >0.96; high precision), consistent across primer pairs.
• Phylogenetic specificity and evolutionary signal: Cryospheric samples showed higher pairwise phylogenetic overlap than cryo–non-cryo or non-cryo–non-cryo pairs (Sorensen’s index; Wilcoxon, Holm-adjusted p<0.001). β-MNTD was lower between cryospheric samples than between cryo–non-cryo (p<0.001), implying higher niche similarity. α-MPD was larger in cryospheric communities (linear model, p<0.001), consistent with early radiation followed by constrained diversification.
• Differentially abundant taxa: Identified 589 cryospheric-enriched bacterial genera across 46 phyla; 34.8% Proteobacteria, 13.4% Bacteroidota. Highly enriched genera included Sphingomonas, Polaromonas, Rhodoferax, Brevundimonas, Acidiphilum (Alpha- and Gammaproteobacteria), and Hymenobacter, Ferruginibacter, Polaribacter (Bacteroidota). Additional contributions from Actinobacteria, Chloroflexi, Cyanobacteria, and some Firmicutes. Newly highlighted or previously unassociated genera included Oryzihumus and Pseudolabrys; numerous placeholder clades indicate many uncultured lineages.
• Community structure across ecosystems: Microbiomes differed significantly among cryospheric ecosystem types (PERMANOVA r^2=0.183, p<0.001; all pairwise p<0.001); minor primer effect (r^2=0.019, p<0.001). Defined a core cryospheric microbiome of 37 genera (e.g., Pseudomonas, Acinetobacter, Flavobacterium); core genera were disproportionately represented among cryospheric-enriched genera (Fisher’s exact p<0.001, odds ratio=6.93). Core genera constituted the highest relative abundance in ice/snow (23.05% PP1; 24.8% PP2) and the lowest in marine cryosphere (16.9% PP1; 13.3% PP2). Alpha-diversity (Shannon H) increased from snow/ice (H=2.86) to freshwater (H=2.99), marine (H=3.25), and terrestrial (H=3.67).
• Genomic traits: Cryospheric genera exhibited higher GC content than others (median difference 8.8%; Wilcoxon, Holm-adjusted p=0.0011) and enrichment of GC-rich amino acids (e.g., alanine, arginine, glycine). Average genome sizes aligned with reported psychrophile values.
• Functional repertoire: Of 17,191 KOs, 980 were significantly enriched in cryospheric metagenomes. Enriched functions were disproportionately associated with cryospheric taxa, especially core genera (e.g., Pseudomonas, Sphingomonas, Novosphingobium). Chemolithotrophic pathways (manganese/iron uptake, sulfur, nitrogen, hydrogen metabolism) were prominent, consistent with low organic carbon availability. Chitinase genes were common, potentially linked to permafrost carbon cycling and freezing adaptation. Genes for adhesion, motility, and secretion systems indicate biofilm formation as a key survival strategy.
• Uncharacterized functional space: 43.4% of protein-coding genes in cryospheric samples lacked KEGG KO annotations. Clustering 41,068,842 genes yielded 12,125 large clusters; cryosphere-specific clusters had markedly lower UniProt match rates and identities, especially among ambiguous and unassigned clusters (e.g., Unassigned clusters with UniProt matches: Cryosphere 17.65% vs Non-cryosphere 46.94%), revealing strong database underrepresentation of cryospheric functions and taxa.

The analyses demonstrate that cryospheric ecosystems harbor a distinct and diverse bacterial microbiome spanning much of the bacterial tree of life, yet showing strong phylogenetic overlap and lower β-MNTD among cryosphere samples, indicating similar selective pressures and niche constraints. The larger α-MPD in cryosphere communities is consistent with early radiation and subsequent constrained diversification, suggesting a long evolutionary history under cold, oligotrophic, and UV-exposed conditions. Taxonomic enrichment of Proteobacteria and Bacteroidota aligns with prior observations, but the identification of numerous uncultured clades and newly associated genera expands the known taxonomic breadth of cryospheric life. Community structure varies among cryospheric habitats, with ice/snow acting as endmember systems containing the highest proportion of core cryospheric genera, and diversity increasing downstream into terrestrial, freshwater, and marine environments. Functionally, cryosphere metagenomes are enriched in chemolithotrophy, chitin degradation, and biofilm-associated traits, reflecting strategies to cope with low organic carbon and environmental stressors. The substantial fraction of unannotated genes and lower similarity to UniProt among cryosphere-specific clusters highlights a large, poorly characterized functional space and the need for improved representation in reference databases. These results collectively address the central question by providing a global, data-driven reference for the structure, evolution, and functional potential of the cryospheric microbiome and emphasize its vulnerability and significance in a warming world.

This study delivers a global catalogue of cryospheric microbiomes, revealing a distinct, diverse, and functionally specialized bacterial community shaped by sustained evolutionary pressures, with evidence for early and constrained radiation. It identifies key taxonomic groups enriched in cryospheres, defines a cross-ecosystem core microbiome, and characterizes a functional repertoire emphasizing chemolithotrophy, chitin degradation, and biofilm formation. The work exposes a vast uncharacterized functional space due to underrepresentation of cryospheric sequences in current databases. Future research should prioritize cultivation and genomic characterization of uncultured cryospheric taxa, targeted exploration of functional gene space (including database expansion), longitudinal monitoring of cryosphere-to-downstream microbial transfers, and assessments of how ongoing cryosphere loss will impact ecosystem processes and biogeochemical cycles.

• Geographical and habitat sampling biases: Overrepresentation of polar regions and paucity of alpine samples; limited representation of some niches (e.g., glacier snow, glacier-fed rivers/streams, full breadth of permafrost) due to data unavailability.
• Methodological heterogeneity across studies: Different primer pairs, sequencing platforms, and protocols; although two common primer pairs were analyzed and primer effects assessed, residual biases may remain.
• Metagenomic scope constraints: High-quality metagenomes were limited to glacier surfaces, ice-covered lakes, and Antarctic soils; other cryospheric habitats were not included.
• Amplicon resolution limits: Analyses performed at genus level due to 16S rRNA constraints; species-level dynamics and functions may be unresolved.
• Database limitations: Underrepresentation of cryospheric taxa and genes in reference databases (SILVA, KEGG, UniProt) affects annotation rates and identity metrics, leaving a large fraction of functions uncharacterized.
• Observational meta-analysis: Inability to disentangle contemporary ecological assembly from historical evolutionary processes completely; lack of uniform environmental metadata across studies may limit mechanistic inference.