The microbiome of cryospheric ecosystems

The dataset

We curated and explored 695 published 16S rRNA gene samples from cryospheric ecosystems (Methods section and Supplementary Table 7), including polar ice sheets, mountain glaciers and their proglacial lakes, permafrost soils and the coastal ocean under the influence of glacier runoff, and compared these to 3552 published 16S rRNA gene samples from non-cryospheric ecosystems, including temperate and tropical lakes and soils (Supplementary Table 7). This approach allowed us to identify and explore features specific to the cryospheric microbiome and compare it to other environmental microbiomes. However, we note a geographical bias towards polar regions in current publicly available repositories, and the paucity of alpine samples specifically highlights the need to further characterise these habitats given that they are among the most endangered cryospheric ecosystems globally. This bias is further compounded by the inconsistent methodologies applied across studies (e.g. primer pairs and sequencers used). To account for potential primer biases, we analysed two 16S rRNA primer pairs (Primer Pair 1, PP1: 341f-785r; Primer Pair 2, PP2: 515f-806r)^12,13 commonly used in amplicon high-throughput sequencing. In total, this dataset contains 241,502,708 paired sequence reads, resulting in 530,254 and 410,931 amplicon sequence variants (ASVs) for PP1 and PP2, respectively. Moreover, all taxonomic analyses were performed at the genus level, to account for the limitations of 16s rRNA amplicon data. To gain deeper insights into the functional space of the cryospheric microbiome, we compared 34 published metagenomes from cryospheric ecosystems with 56 metagenomes from similar but non-cryospheric ecosystems (Fig. 1A). Given the difficulty of obtaining high-quality metagenomes from cryospheric ecosystems, we restricted our analyses to glacier surfaces, ice-covered lakes, and Antarctic soils. Although our analyses were limited to samples where raw sequence data are available (Methods section), the breadth of habitats covered are representative of the most abundant cryospheric ecosystems, e.g., glacier ice, cryoconites, subglacial lakes and sea ice. On the other hand, several niches such as glacier snow, glacier-fed rivers/streams, and the full-breadth of permafrost may not entirely be represented due to data unavailability. We reanalysed all metagenomes using the same bioinformatic pipeline (IMP3; see Methods section) to avoid analytical biases. Overall, the metagenomic analyses from 2,427,818,072 paired reads yielded 41,068,842 gene sequences. Thus, we here present a catalogue representing a snapshot of the functional diversity in the cryospheric microbiome, integrating across diverse habitats. This represents what we believe to be the first global overview of the functional repertoire of the Earth’s cryosphere compared to other ecosystems.

A cryospheric microbiome

Given the communal constraints imposed by the harsh environment of cryospheric ecosystems (e.g., low temperature, oligotrophy), we expected them to harbour a specific microbiome. Accordingly, machine-learning classification (logistic regression models, Methods) based on community composition was able to differentiate between cryospheric and non-cryospheric microbiomes with high accuracy (balanced accuracy >0.96, Supplementary Table 1). Both primer pairs consistently yielded a high classification accuracy and especially a high precision. Interestingly, many of the discriminating cryospheric ASVs were spread widely across the bacterial tree of life (Fig. 1A and Supplementary Fig. 1).

The notion that a part of the microbiome is specific to the cryosphere is also strongly supported by phylogenetic analyses of the 16 S rRNA gene amplicon dataset. First, we found higher pairwise phylogenetic overlap among cryospheric samples than among cryospheric/non-cryospheric or non-cryospheric samples (Sorensen’s index, Fig. 1C; Wilcoxon test, Holm adj. p < 0.001). This points towards a phylogenetic diversity that is specific to the cryosphere. Second, when we examined cross-sample nearest taxon distances (β-NTDs), we found that taxa in cryospheric samples have lower β-NTDs in other cryospheric samples than in non-cryospheric samples (Fig. 1D; Wilcoxon test, Holm adj. p < 0.001). This was less evident for non-cryospheric samples (Supplementary Table 2). Because phylogeny and functional similarity usually correlate at short phylogenetic distances¹⁴, this finding suggests higher niche similarity for cryospheric bacteria compared to their non-cryospheric equivalents. This evokes specific selective constraints of cryospheric environments acting on taxa across the entire bacterial tree of life. Interestingly, when we further examined radiation patterns, we found that taxa in a given cryospheric microbial community had on average larger phylogenetic distances (α-MPD) than their counterparts in a non-cryospheric community (linear model, p < 0.001). This could suggest early radiation events with subsequent “pruning” of phylogenetic diversity, which would explain the observed patterns¹⁵. However, we cannot exclude nor disentangle the action of contemporary evolutionary and assembly processes that can jointly shape community phylogenies. For example, transduction and genome plasticity have repeatedly been linked with cold adaptation in cryospheric bacteria. Moreover, horizontal gene transfer has also been shown to promote the diffusion of cold-adaptation genes¹⁶. Nevertheless, given the large number and breadth of included cryospheric ecosystems, we posit that the topologies of the inferred phylogenies are less prone to assembly processes. We rather interpret that the observed patterns are signs of early and constrained radiation in the cryospheric microbiome. Collectively, these results point to similar evolutionary trajectories in cryospheric microbiomes, probably owing to similar environmental conditions across various cryospheric ecosystems, over timescales, relevant for bacterial macroevolution.

The abundance of a given species in an ecosystem generally reflects its fitness and adaptive capacity to the respective environmental conditions. Therefore, we explored patterns of differential abundance (Methods section) and found 589 bacterial genera with higher abundances in cryospheric compared to non-cryospheric samples (Ancom, W statistic > 0.7, CLR mean difference > 0) that hereafter will be referred to as cryospheric genera. These genera were distributed widely across the bacterial tree of life and encompassed 46 different phyla. Despite this wide distribution, we found that 34.8% and 13.4% of the cryospheric genera were affiliated Proteobacteria and Bacteroidota, respectively (Fig. 2A). The relevance of Proteobacteria is in line with the high prevalence of Alpha– and Gammaproteobacteria typically reported in the cryospheric literature^4,17. Genera belonging to the Alpha- and Gammaproteobacteria classes displayed the highest differential abundance and included Sphingomonas, Polaromonas, Rhodoferax, Brevundimonas, and Acidiphilum (Fig. 2B) — some of them with taxa typically reported to be psychrophiles^10,18,19,20. Bacteroidota was the second most important phylum of the cryospheric microbiome with Hymenobacter, Ferruginibacter, and Polaribacter (for instance) as dominant genera, all of which are known from permafrost soils and ice ecosystems^21,22. Furthermore, as previously reported^23,24, the cryospheric genera included members of the Actinobacteria, Chloroflexi and Cyanobacteria phyla, alongside some Firmicutes. The former two are particularly common in supraglacial environments⁴, and Cyanobacteria are important components of cryoconite microbiomes²⁵. Our global analyses thus corroborate and extend previous reports on microbiome composition in distinct cryospheric ecosystems. Furthermore, our differential abundance analysis unveiled genera (e.g., Oryzihumus or Pseudolabrys) that have not been previously associated with the cryosphere (Fig. 2B). More importantly, many of the detected cryospheric genera only have placeholder names due to the lack of cultivated representatives (e.g., CL_500-29_marine_group, hgcl_clade, TRA3-20), underlining unique bacterial groups that are yet to be described. Collectively, these findings unveil an unexpectedly diverse and likely well-adapted microbiome specific to the cryosphere, and supports the notion of the cryosphere as a biome with its distinct association of microorganisms, alongside plants and animals¹⁷.

Compositional patterns across cryospheric ecosystems

We next explored how microbial community composition varies across cryospheric ecosystems. Using similarity analyses, we found that the microbiome composition differed significantly between cryospheric ecosystem types (PERMANOVA, r² = 0.183, p < 0.001; pairwise.adonis, p < 0.001 for all pairwise comparisons) (Fig. 3A and Supplementary Table 4). Most conspicuous was the segregation of snow/ice and marine communities, bracketing freshwater and terrestrial cryospheric communities. We also found a significant but relatively small effect of the primer pair (PERMANOVA, r² = 0.019, p < 0.001) that could be attributable to primer bias, or inherent differences related to sampling. To further assess these distributions, we explored prevalence patterns to identify a core microbiome across cryospheric ecosystems (Fig. 3B). We found 37 bacterial genera, including Pseudomonas, Acinetobacter, and Flavobacterium, for instance, to constitute the core microbiome. The disproportionate representation of these core genera in the above-identified cryospheric genera (Fisher’s exact test, p < 0.001, odds ratio = 6.93) underlines their high abundance in cryospheric ecosystems (Supplementary Fig. 2). It also shows the prevalence and abundance of some cryospheric genera, indicating their potential relevance for ecosystem processes.

Additionally, analysing the relative abundance of the core cryospheric genera for each primer pair and cryospheric ecosystem types, we found that ice and snow microbiomes were associated with the highest proportions of core genera (23.05% and 24.8% for PP1 and PP2, respectively) (Fig. 3D). In contrast, the marine cryospheric microbiome is only marginally composed of these genera (16.9% and 13.3% for PP1 and PP2, respectively). This pattern is in line with our unconstrained ordination analysis (Fig. 3B) and suggests that snow and ice represent endmember cryospheric systems, while the cryospheric component of the microbiome dissipates in downstream freshwaters, soils and the coastal ocean. Furthermore, the alpha-diversity was higher in terrestrial (Shannon H = 3.67), marine (H = 3.25) and freshwater (H = 2.99) ecosystems than in snow and ice (H = 2.86), corresponding to increasing contributions of ancillary taxa to their microbiomes (Supplementary Table 5). These differences in diversity are likely attributable to environmental gradients characterised by more diverse energy sources and niches, such as when moving from snow and ice to aquatic and soil ecosystems. Our analyses revealed compositional patterns of the cryospheric microbiome suggesting that snow and ice ecosystems including supraglacial habitats (e.g., mountain glaciers, ice sheets, snow and cryoconites) may serve as a source of cold-adapted bacterial diversity, upon losing which the downstream diversity may become threatened as well.

Functional potential of the cryospheric microbiome

The adaptive and survival strategies of microorganisms to the extreme environmental conditions of the cryosphere have received substantial attention over the last years^26,27,28. For example, genomic insights from bacterial cultures have revealed mechanisms of thermal adaptation linked to bulk genomic features, such as GC content and genome size²⁹. Moreover, genome streamlining has been shown to be a relevant evolutionary force in the cryosphere²⁸. Therefore, we analysed the GC content and genome size of 13,414 reference genome sequences from the NCBI Refseq genomes database³⁰ to investigate shared properties of cryospheric genera, and to provide a framework to contrast future cryospheric metagenomic results. By comparing these reference genomes representing 660 bacterial genera present in our taxonomic analyses (29.8% of which are cryospheric genera according to our differential abundance analysis), we found that the cryospheric genera had a significantly higher GC content (Supplementary Fig. 3B; Wilcoxon test, Holm adj. p = 0.0011, median difference = 8.8%) compared to the other genera; a pattern also supported by an enrichment in sequences that encode GC-rich amino acids (e.g., Alanine, Arginine, Glycine) (Supplementary Fig. 3A and Supplementary Table 6). Therefore, our findings suggest that cryospheric genera indeed share an elevated GC content³¹, in line with reports on cold-adapted Synechococcus (SynAce01)³² and Actinobacteria³³. We also report that the average genome size of cryospheric genera is closely bracketed by published values for psychrophilic bacteria³⁴.

Next, using a gene-centric approach, we explored the functional space of the cryospheric metagenomes dataset. Out of 17,191 KEGG orthologues (KO), 980 KO were significantly enriched in cryospheric samples. Cryospheric genera and particularly cryospheric core members (e.g., Pseudomonas, Sphingomonas and Novosphingobium) disproportionately accounted for these gene families (Fig. 4A). Our analysis highlighted the relevance of chemolithotrophic pathways (e.g., manganese and iron uptake, sulfur, nitrogen and hydrogen metabolism), complementing earlier reports on these particular functional attributes of cryospheric ecosystems (Fig. 4B)^24,35,36. The apparent relevance of chemolithotrophic pathways is likely attributable to a relative scarcity of organic carbon in cryospheric ecosystems. Interestingly, we consistently identified chitinase genes, which are involved in permafrost carbon cycling, but may also be an adaptation to freezing³⁷. Finally, genes involved in adhesion, motility and various secretion systems collectively point to biofilm formation as an important strategy for life in cryospheric ecosystems³⁸, which are often characterised by extended periods of oligotrophy and elevated UV-radiation.

Our cross-ecosystem metagenomic analyses not only shed light on potential functions of the cryospheric microbiome across ecosystems, but also unveiled a large uncharacterised functional space with 43.4% of the protein coding genes in cryospheric samples unannotated to a KEGG orthologous group. While this does not seem unusual for environmental metagenomes in general³, it is notable that we may lose this functional potential as the cryosphere vanishes. In order to shed light on this uncharacterised functional space, we clustered 41,068,842 gene sequences based on a 30% sequence similarity and 80% sequence coverage threshold, subsequently mapping representative sequences of the largest clusters (>29 sequences in at least 2 samples, n = 12,125) to the UniProt TrEMBL database (Fig. 4C). While the KEGG assigned clusters overall had a high percentage of sequences that matched genes in the UniProt database (Table 1), we found that cryosphere specific sequences show a large decrease in the clusters assigned to multiple KEGG (i.e., ambiguous) and even more in the ones containing exclusively unassigned sequences, compared to non-cryospheric environmental metagenomes. In addition to the low percentage of gene sequences matching UniProt sequences, we found that the cryosphere specific clusters that align to the database show a largely decreased identity with the matching sequence (Supplementary Fig. 4). These findings underline the lack of representation of cryospheric sequences in current gene sequences databases, potentially linked to the specificity of certain taxa to the cryosphere, and/or functions. Finally, the large nucleotide similarity within these clusters (Supplementary Fig. 4) suggests that these are conserved functions of particular importance to microbial life under cryospheric constraints, and corroborates the notion of specific lineages of closely related taxa to dominate microbial life in the cryosphere. Aside from being uncharacterised, 170 of the unassigned gene clusters were only detected in cryospheric metagenomes and could thus represent unknown gene families of importance to understanding the adaptation of bacteria to these extreme ecosystems.

Collectively, our insights both at the taxonomic and functional level reveal key microbiome features that are exclusive to cryospheric ecosystems. Although entire taxonomic lineages are not unique to cryospheric ecosystems, it is evident that specific species and potentially strains are novel and adapted to these environments. Similarly, the emergent functional properties clearly demonstrate the exclusivity of functions, especially those that are yet to be characterised or that can be classified based on existing databases, within the cryosphere. On the contrary, we find that in both the taxonomic and functional complements, several taxa and functions are shared with non-cryospheric ecosystems. This is expected since the underlying genomic content supporting the taxonomic and functional annotations are shared between the cryospheric and non-cryospheric ecosystems. This is evident based on the >50% identity among the shared gene clusters that had matching identities in the KEGG database (Table 1).

Here we present what we believe is the first global data-driven approach to unravel specific features of the cryospheric microbiome. Our meta-analysis revealed diverse, distinct and functionally specific bacterial communities that appear to have been shaped by sustained evolutionary forces, suggesting an ancient origin of this biodiversity. While our study highlights key taxonomic groups such as Proteobacteria and Bacteroidota, our findings also disclose the importance of yet-uncultured bacteria and an uncharacterised genetic repertoire. In light of the threatened nature of the cryosphere, targeted efforts to unravel the phylogenetic and genomic underpinnings of bacterial adaptation to cryospheric ecosystems, including prospecting for cold-adapted biomolecules as well as the cultivation of cryospheric bacteria, are urgently required.

Source: Ecology - nature.com