Desiccation, but not shading, substantially shifts the microbial community composition of benthic microbial mats in meltwater ponds near Bratina Island, Antarctica

Abstract

The Bratina Island meltwater ponds (McMurdo Ice Shelf, Antarctica) host complex microbial communities in the form of benthic microbial mats. These ponds are long-lasting but can experience significant alterations in physicochemistry on an annual or decadal basis, with additional alterations expected due to climate change. Here, we explored the impact of two potential future climate scenarios on the 16S rRNA gene sequence-based microbial community composition of three meltwater pond mats: desiccation (due to decreased precipitation and/or increased evaporation) and shading (due to increased precipitation resulting in increased ice cover). In a multi-year manipulation experiment, we found that the mat bacterial community composition shifted substantially with desiccation in all ponds studied. Desiccation reduced overall diversity, with cyanobacteria mostly being lost and members of the Bacteroidota phylum increasing in relative abundance. Halotolerance was indicated as a potential trait allowing for survival upon desiccation. We found that the community composition of the mats was resilient to shading (in the form of 75% light reduction), with no significant changes in composition observed. Instead, we observed a strong coloration change in the mats in response to shading, which could be due to positive phototaxis of Oscillatoria trichomes in the shaded mats. These results help inform how the prokaryotic communities in these highly productive mats may respond to future climate scenarios.

Benthic microbial mat communities living in Antarctic meltwater ponds are highly sensitive to drying out, but are capable of coping with reduced sunlight due to phenotypic changes in the photosynthetic community members.

Introduction

In Antarctica, water availability is one of the strongest factors that limits microbial growth (Cary et al. 2010, Tiao et al. 2012, Niederberger et al. 2019). Antarctic meltwater ponds provide small, stable oases of water for life to flourish. On the McMurdo Ice Shelf near Bratina Island, a large number of these shallow ponds have formed due to the melting of snow and ice, and subsequent accumulation in depressions in the sediment-covered terrain (Wait et al. 2006). Environmental conditions differ substantially between ponds (Howard-Williams et al. 1990, Mora et al. 1994, Glasser et al. 2006) and within ponds on annual (Hawes, Safi, Sorrell et al. 2011, Webster-Brown et al. 2012) and decadal time scales (Hawes et al. 2014). Between- and within-pond differences are largely driven by evaporative salinisation (Mora et al. 1994, Glasser et al. 2006, Wait et al. 2006), along with freeze-melt cycles (Hawes, Safi, Sorrell et al. 2011).

The ponds are highly productive during the austral summer, when constant sunlight and warmer temperatures allow for photosynthesis to be conducted, primarily in the benthic microbial mats that dominate biomass in most ponds (Howard-Williams et al. 1990). The primary photosynthetic bacterial members of the mats are from cyanobacterial groups such as Phormidium, Oscillatoria, Lyngbya, and Nostoc, with the dominant member(s) in each pond varying depending on pond salinity (Jungblut et al. 2005, Sutherland 2009, Sutherland et al. 2020). The summer-long intensive photosynthetic activity results in near-universal high pH (>9) and, in some situations, supersaturating oxygen concentrations (Hawes, Safi, Webster-Brown et al. 2011). The growing season of the mats is curtailed by the onset of autumn, when lower light levels (due to less daily light exposure and increased snow and ice cover) and lower temperatures slow photosynthesis, resulting in significant changes to the prokaryotic communities in the water column (Hawes et al. 1999, Noell et al. 2026).

A large diversity of non-photosynthetic microbes associate with these mats, primarily from groups such as Bacteroidota and Pseudomonadota (formerly Proteobacteria), which play important roles in nutrient cycling in the ecosystem (Archer et al. 2015). These microbes are adapted to extreme conditions, including extremes in salinity, oxygen levels, and temperature (Noell et al. 2026). The mats serve as important reservoirs of microbial life overwinter and provide major contributions towards nitrogen and carbon fixation in Antarctic ecosystems (Fernández-Valiente et al. 2001, Archer et al. 2014).

Meltwater ponds in Antarctica are forecast to experience significant change in the coming decades due to climate change, potentially creating challenges for the maintenance of these mat communities (Kingslake et al. 2017, Wessem van et al. 2023). It is unknown exactly what kinds of stressors meltwater pond communities will face, but two possibilities are: (1) reduced snowfall, resulting in increased evaporation of ponds and, thus, increased desiccation of mats; (2) increased snowfall and ice cover, resulting in less light availability for mats. As the primary producers in these ponds, it is of significant interest to understand how vulnerable these cyanobacteria and associated communities may be to environmental perturbances.

It is well established that cyanobacterial mats from these ponds have a high tolerance for and recovery rate from desiccation, but this differs among different types of microbial mats (Quesada and Vincent 2012). Previous studies of similar environments (including meltwater ponds on the McMurdo Ice Shelf) have shown that when cyanobacterial mat communities are exposed to environmental stressors such as desiccation and light stress for relatively short periods (<30 days), they tend to recover quickly—sometimes within minutes of being restored to optimal conditions (Hawes et al. 1992, Chennu et al. 2015). However, it is unknown what effect long-term exposure to desiccation might have on the cyanobacteria present, as well as the rest of the associated prokaryotic community. Additionally, the long-term effect of decreased sunlight availability on the mats remains unknown; short-term shading experiments in the past have indicated significant conformational responses of the photosynthetic mat members (Vincent et al. 1993, Nadeau et al. 1999, Hawes et al. 2016).

This study explores the resistance and resilience of bacterial and archaeal members of microbial mats to desiccation and reduced irradiance in meltwater ponds on the McMurdo Ice Shelf (MIS). We hypothesized that, faced with realistic application of these stressors, prokaryotic members of the mats would respond through shifts in community composition that would facilitate retention of critical community functions. To test this hypothesis, manipulation experiments were established in three ponds (New, AMS, and P70) on the MIS, and samples of microbial mat were collected from each treatment (control, desiccation, or shading) for the duration of the experiment (annually for two years), and the prokaryotic community characterised through sequencing of the V4-V5 region of the 16S rRNA gene, with functional inferences based on 16S rRNA gene databases. A recovery treatment was added to the shading experiment to see how well the mat community would rebound after the shading stress.

Materials and methods

Site description, sample collection and experimental design

This experiment explored the resilience and recovery of bacterial and archaeal members of microbial mat communities in response to long-term shading and desiccation in three ponds near Bratina Island from 2016 to 2018. Three treatments: shading, desiccation and control were established in each of the three ponds, AMS, Pond 70, and New (AMS and New are not official names) over the duration of the experiment (see Fig. 1 for location of the ponds and geographical setting). GPS coordinates for these ponds are: Pond 70, South 78.0147, East 165.5527; AMS, South 78.0150, East 165.5565; and New, South 78.0151, East 165.5583. These ponds were chosen for their comparable physicochemistry (pH between 8.5–9.5, conductivity between 2.5–15 mS/cm (Fernández-Valiente et al. 2001, Archer et al. 2014, Husain et al. 2025)) and cyanobacterial composition of the mats present (primarily non-heterocystous filamentous cyanobacteria (Fernández-Valiente et al. 2001)). This allows any changes and observations made throughout the experiment to generally be compared between the ponds.

Figure 1.

(A) Map of the Ross Sea region, including Ross Island, the Ross Sea, and the Ross / McMurdo Ice Shelves, with Bratina Island highlighted in the red box. Inset shows map of Antarctica with the Ross Sea region highlighted in the red box. Map of Antarctica, with region of interest in the Ross Ice Shelf / Ross Sea region in the black box. (B) Location of study sites: Ponds AMS, New and P70 on the McMurdo Ice Shelf. (C) Example of shading apparatus used in experiments. (D–F) Images of the three ponds in this study (top row), desiccation treatment mats (middle row), and shading sites after removal of shades (bottom row). For the bottom row, the former location of the shades is indicated by the boxes and text. (D) New Pond, (E) Pond 70, (F) AMS Pond. Maps in (A) are from the Antarctic Digital Database Map Viewer https://www.add.scar.org/, Open Source. All other images are courtesy Ian Hawes.

During year one of the experiment (T0), control samples were collected from each of the three ponds in triplicate. Cloth shades were installed in each pond in triplicate to replicate environments with reduced ice transparency (i.e. increased ice cover or snow accumulation on the ice). Light levels underneath the shades were measured using a LiCor 192 Underwater PAR sensor in year one of the experiment, with transmission measured at 25% (i.e. the shades blocked 75% of light). Blocks of microbial mat were cut out and placed on the pond edge to replicate the desiccation of the microbial mat as the pond water level dropped.

During year two of the experiment (T1), three treatments were sampled from each of the ponds: controls, shaded mats, and dried mats. The shades were relocated to a second spot in each pond to allow the microbial community to recover from the shading, acting as a pulsed stress mechanism. The light levels underneath the shades were measured prior to moving the shades, with transmission at 15% due to accumulation of material on top of the shades. The sample site from the previous year, from where shades were removed, became the recovery plots.

During year three of the experiment (T2), there were four variables to be sampled. These were the controls, desiccation, shade, and the new recovery plots for each pond. After this final year of the experiment, shade cloths and desiccation plots were removed to allow the microbial mat to return to its normal state. All samples were collected aseptically and preserved by freezing.

DNA extraction and quantification

DNA was extracted in duplicate from 0.5 g of each of the seventy-two samples, with extraction negative controls included. The DNA extraction protocol used was a modified CTAB/bead-beating method (Herbold et al. 2014). DNA was re-suspended in 20 µl sterile LO-TE and quantified using the Qubit Fluorometer high sensitivity dsDNA protocol on the Qubit 2.0 fluorometer (Thermo Fisher Scientific, Massachusetts, USA) and the Denovix DS-11 spectrophotometer/ fluorometer (Denovix Incorporated, Delaware, USA).

PCR and sequencing

The V4-V5 region of the 16S rRNA gene was amplified in triplicate reactions using the fusion IonXpress primers 515F (5′CCATCTCATCCCTGCGTGTCTCCGACTCAGXXXXXXXXXXXGATGTGCCAGCMGCCGCGGTAA’3) and 926R (5′CCACTACGCCTCCGCTTTCCTCTCTATGGGCAGTCGGTGATCCGYCAATTYMTTTRAGTTT ’3). Specific barcodes were used per sample to uniquely identify each sample (Comeau et al. 2017). Briefly, 20 μL PCR reactions each contained: 1 ng of total DNA, dNTPs (240 μM), MgCl2 (6 mM), bovine serum albumin (0.24 μM), forward and reverse fusion primers (0.2 μM), 1 U of Platinum Taq polymerase (Invitrogen Inc., Carlsbad, California) and PCR buffer (1.2X).

The Proflex PCR system (Thermo Fisher Scientific) was used to amplify the 16S rRNA amplicons. The PCR protocol was: initial denaturation at 94 °C for 3 min, second denaturation at 94 °C for 45 sec, annealing at 50 °C for 1 min, extension at 72 °C for 1 min 30 sec and a final extension at 72 °C for 10 min. These conditions ran for a total of 30 cycles. To confirm amplification of the PCR product before sequencing proceeded, amplicon bands were visualised by gel electrophoresis with a 1% Agarose gel using the Alpha Innotech imaging system (Alpha Innotech, California, USA). PCR was also conducted on negative (extraction and reagent) and positive (extracted DNA from New Zealand soils) controls interspersed among the samples. Absence of substantial, amplifiable DNA in the negative controls was confirmed visually on the agarose gel.

Triplicate PCR products from each sample were pooled. A total of 25 µl of each sample was treated with Invitrogen SequalPrep Normalization (Thermo Fisher Scientific) to normalize the PCR products, remove DNA fragments smaller than 100 base pairs and to purify the samples. 2 µl of each PCR triplicate was pooled together to generate the sequence library. The sequence library was quantified using the Qubit 2.0 Fluorometer (Thermo Fisher Scientific) to determine the concentration of DNA in ng/µl. The number of amplifiable molecules was determined and treated with Ion PGM^TM Template IA 500 Kit (Thermo Fisher Scientific). Ion PGM^TM Hi-Q^TM View Sequencing Kit (Thermo Fisher Scientific) was then used to prepare the library for sequencing. The library was added to an Ion 318 Chip Kit v2 BC (Thermo Fisher Scientific). Samples were sequenced on an Ion Torrent PGM (Thermo Fisher Scientific), read length ∼300 bp.

DNA sequence processing

Raw data reads obtained for each sample from the Ion Torrent PGM (Thermo Fisher Scientific) were first trimmed of sequencing primers using cutadapt (Martin 2011), then processed using the DADA2 pipeline (v 1.16) in R. Quality profiles and error rates of the sequence reads were inspected. Sequences were trimmed and filtered following the recommended procedure. Sample inference was performed using the `pool=TRUE` option, as well as recommended error settings for Ion Torrent PGM data (`HOMOPOLYMER_GAP_PENALTY=-1, BAND_SIZE=32`). Chimeras were removed using the function `removeBimeraDenovo`, method “consensus.” Taxonomy was assigned to amplicon sequence variants (ASVs) using the DADA2 Bayesian classifier with the SILVA 16S rRNA gene database (138.2) (McLaren 2024), and species assignments made using exact matching (100% ID) with the Silva 16S species-assignment training fasta set (silva_v138.2_assignSpecies.fa.gz), as recommended (Edgar 2018). Sequences unassigned at the Domain level, assigned to Eukaryota, mitochondria, and chloroplast were removed using the following commands:

• physeq ← subset_taxa (physeq, !is.na(Kingdom) | (Kingdom !=“Eukaryota”))

• physeq ← subset_taxa (physeq, is.na(Order) | Order !=“Chloroplast”)

• physeq ← subset_taxa (physeq, is.na(Family) | Family != “Mitochondria”)`

ASVs with less than 10 reads in all samples were removed (low abundance ASVs, 90). The R-packages Decipher (Wright 2016) and phangorn (Schliep 2011) were used to generate an unrooted phylogenetic tree by the neighbor-joining method. The resulting set of ASVs in a phyloseq (v. 1.50.0) (McMurdie and Holmes 2013) object was used for subsequent analyses.

Statistical analyses

All plots were visualized using ggplot2 v. 4.0.0 (Wickham 2016) with figure refinement performed in Inkscape (https://www.inkscape.org/).

Prior to calculating alpha and beta diversity, samples were rarefied to an even depth (based on minimum sample read count, 7890 reads) using 1000 iterations without replacement, as recommended in (Schloss 2024). Two samples that were outliers in terms of having very low read counts (New_shade2_T1 and P70_control1_T2) were removed prior to rarefaction and alpha/beta diversity. Multiple rarefaction was performed using the `phyloseq_mult_raref_avg` function from the metagMisc R package, v. 0.6.0.9 (Mikryukov et al. 2025). This produced a relative abundance-normalized phyloseq object that was used only for alpha and beta diversity. Alpha diversity was assessed within vegan v. 2.6–8 (Oksanen et al. 2020) via the Shannon index. Unconstrained beta diversity of the microbial community structures was investigated by calculating an unweighted Unifrac distance matrix using phyloseq. This Unifrac distance matrix was used to perform a PERMDISP test (`betadisper` in the vegan package) to test for significant differences in dispersion between groups; for those with non-significant results, a PERMANOVA test (`permutest` in the vegan package) was then used to identify any significant differences in community composition between groups. This distance matrix was also used for constructing PCoA plots using plot_ordination from the phyloseq package.

To identify differentially abundant taxa between treatments, the microeco package in R, v. 1.10.0 (Liu et al. 2021) was used. In particular, the linear discriminant analysis effect size (LEfSe) method (Segata et al. 2011) was used within microeco, which incorporates statistical significance as well as biological relevance in terms of effect size. To predict functional profiles for the ASVs, we used a functional prediction program (FAPROTAX) (Louca et al. 2016) within microeco that uses taxonomic matching to cultured genera with known functional traits. Thus, assignment to function is based on matches between assigned taxonomy of ASVs in our samples with taxonomy of genera in the database at the genus level, not on sequence similarity. The FAPROTAX database currently contains over 7600 functional annotations covering over 4600 taxa. We used microbiomeMarker v. 1.12.2 (Cao et al. 2022) to conduct LEfSe tests on predicted functional data.

Results

Sequencing results and pond comparison

In total, we collected 72 samples across all treatments (control, desiccation, shade, and recovery), time points (year zero for controls, years one and two for desiccation and shade, and year two for recovery), ponds (AMS, New, and Pond 70–P70), and replicates. After DNA extractions, we were able to successfully produce community 16S rRNA gene sequences from 59 samples, which resulted in less than three replicates for some pond/treatment/timepoint combinations due to low DNA yield (Table S1). In total, our 16S rRNA gene sequence data set had 2327 ASVs across 1 081 504 reads, after removing erroneous assignments and low abundance (<10 reads across all samples) ASVs (Table S2). Different treatments had roughly similar median read counts, although the number of reads within each treatment varied by sample (Fig. S1A-B). For this reason, we chose to use rarefaction to randomly bring read count to the same number for all samples for alpha and beta diversity analyses. Prior to rarefaction, the microbial communities in these samples were clearly diverse and not completely sampled (Fig. S1C).

In general, control samples from the ponds had comparable communities in terms of being dominated by photosynthetic Cyanobacteriota accompanied by diverse members of phyla such as Bacteroidota and classes Alphaproteobacteria and Gammaproteobacteria within Pseudomonadota phylum (Table S2). Despite the attempt to select ponds with similar mat communities by matching pH, conductance, ice cover, and mat appearance, the prokaryotic communities were significantly different from each other (PERMANOVA, R² = 0.29, p = 0.001; Table 1, Fig. S2A). The ponds were, however, similar in terms of alpha diversity (Fig. S2B). The taxonomic groups identified as being most likely to explain differences between the ponds using a Linear discriminant analysis Effect Size (LEfSe) test (Segata et al. 2011) indicated that AMS was enriched in Oscillatoria species, P70 was enriched in two genera from Bacteroidota, and New was enriched in distinct genera from the phyla Bacteroidota, Pseudomonadota (Alphaproteobacteria), and Planctomycetota (Fig. S2C).

Table 1.

Results from PERMDISP tests to determine if there were significant differences in data dispersion.

Experiment	Pond	PERMDISP F	PERMDISP p	PERM-ANOVA F	PERM-ANOVA R2	PERM-ANOVA p	n
Pond comparison	NA	1.70	0.21	3.25	0.29	0.001*	59
Desiccation	All	4.74	0.03*				33
Desiccation	AMS	0.34	0.78	1.94	0.22	0.01*	9
Desiccation	New	0.01	0.94	3.82	0.26	0.004*	13
Desiccation	P70	0.003	0.96	3.03	0.25	0.003*	11
Control vs. Shading	All	0.01	0.94	1.26	0.04	0.18	38
Control vs. Shading	AMS	0.001	0.92	0.95	0.11	0.46	10
Control vs. Shading	New	0.00 002	1.00	1.39	0.11	0.13	14
Control vs. Shading	P70	0.08	0.79	1.48	0.12	0.09	14
Control vs. recovery	All	3.67	0.06	1.21	0.05	0.20	27
Control vs. recovery	AMS	5.12	0.001*				6
Control vs. recovery	New	7.64	0.07	0.88	0.10	0.63	10
Control vs. recovery	P70	7.21	0.02*				11
Shade vs. recovery	All	2.83	0.11	1.45	0.06	0.09	24
Shade vs. recovery	AMS	0.86	0.36	1.49	0.20	0.11	8
Shade vs. recovery	New	15.46	0.05	1.05	0.17	0.28	7
Shade vs. recovery	P70	6.44	0.001*				9

Notes: If significant results were not returned from PERMDISP, PERMANOVA test results were used to determine significant differences in microbial communities between pond type, desiccation and control treatments, and shade and control treatments. Significant (p < 0.05) test results are highlighted in bold and with a *. n = number of samples.

We observed that many of the taxa observed to be significantly different between ponds fill similar ecological niches; thus, we asked whether there were any differences in community functional potential between ponds. While 16S rRNA gene sequence data alone cannot provide data on the genetic potential of these taxa, taxa-matching to databases of groups of organisms with experimentally verified functions using FAPROTAX (Louca et al. 2016, Sansupa et al. 2021) provides a tool to begin to predict community-wide functions and provide hypotheses to further examine with metagenomics or metatranscriptomics methods. This method has its limitations, as not all members of a genus necessarily share the same functions; nevertheless, it serves as a useful starting point. Using FAPROTAX, we did not identify any statistically significant (adjusted p > 0.05, Kruskal-Wallis test, Benjamini-Hochberg adjusted for multiple tests) differences in functional categories between ponds (data not shown). This preliminary result supports our observation that the three ponds, while taxonomically different, have similar functional profiles as predicted using FAPROTAX and supports our choice of the three ponds based on similar physicochemical profiles.

Desiccation experiment

Across all three ponds, the desiccated mats appeared different to the control mats, having lost any photosynthetic coloration and turned grey/brown, with significant cracking evident (Fig. 1D-F, middle row). As expected based on this visual examination, desiccation significantly altered the mat bacterial communities, resulting in significant decreases in diversity (Fig. 2A), which remained steady across the two years of sampling (Fig. 2B). The diversity in the control treatments in all ponds in Year 2 was also reduced compared to Year 1, resulting in the difference in alpha diversity between control and desiccation treatments being non-significant in Year 2 (Fig. 2B).

Figure 2.

Examining the impacts of desiccation on the microbial communities of microbial mats from three different ponds across several time points. (A–B) Comparison of alpha diversity (as measured via Shannon Index) between the control and desiccation treatments, either (A) aggregated across all time points or (B) split by time point. The p-values given in (A–B) are the result of Wilcoxon signed-rank tests, which in (B) are adjusted for multiple tests (Benjamini-Hochberg). (C–E) Beta diversity measured with a Principal Coordinates Analysis (PCoA) using unweighted Unifrac distances, with relative abundance normalized read counts for ASVs, for (C) P70 Pond, (D) AMS Pond, (E) New Pond. The percentages next to the axes names indicate the amount of variation in the data that that axis explains. For all analyses in this figure, samples were rarefied to an even depth prior to calculation, which required removing two samples with very low read depth (see Methods).

Desiccation resulted in significantly different (PERMANOVA, p < 0.01, R² = 0.22–0.26) prokaryotic communities between the experimental and control mats in each of the ponds when considered separately (Table 1, Fig. 2C). We did not observe any separate clustering of prokaryotic communities by timepoint (Fig. 2C).

When aggregated across all timepoints and ponds, we observed that the desiccated mat communities were proportionately enriched in ASVs from the phylum Bacteroidota (class Bacteroidia) and the classes Bacilli and Gammaproteobacteria (Fig. 3A, Fig. S3A), and depleted in ASVs from the phyla Cyanobacteriota (class Cyanobacteriia), Verrucomicrobiota (class Verrucomicrobiia), and Planctomycetota (class Planctomycetes), as well as the class Alphaproteobacteria (Fig. 3A, Fig. S3A). These quantified changes in relative abundance at the phylum and class level were consistent across ponds, although the observed reduction in cyanobacteria was less consistent, with P70 retaining more cyanobacteria (Fig. 3C). Genera from Cyanobacteriota displayed reduced relative abundance in the desiccated mats across all three ponds (Fig. S4), the exception again being in P70, where Tychonema, Nodosilinea, and Dolichospermum all slightly increased in relative abundance in the desiccation treatments compared to the control (Fig. S4).

Figure 3.

(A–C) Community composition by relative abundance of microbial mat communities at the phylum level, (A) aggregated into control vs. desiccation treatments across all ponds and timepoints, (B) as in (A) but split out by timepoint, and (C) with all replicates, timepoints, and ponds split out. (D–E) Identification of microbial groups at the (D) Class or (E) Genus level that were significantly different between control and desiccation groups, irrespective of pond and timepoint, based on LDA score, calculated using the LEfSe method. Stars indicate significant difference between groups based on Benjamini-Hochberg adjusted p-values: *: p < 0.05, **: p < 0.01, ***: p < 0.001.

A LEfSe test to identify distinguishing taxa between treatments confirmed these initial observations and provided additional insights (Fig. 3D-E). In particular, we observed that the genera Psychrobacter (Gammaproteobacteria), Gillisia (Bacteroidota), Gelidibacter (Bacteroidota), and Planococcus (Bacilli) were significantly enriched in the desiccated mats (Fig. 3E). On the other hand, genera Geitlerinema (Cyanobacteriota), Rhodoferax (Betaproteobacteria), Cypionkella (Alphaproteobacteria), Hydrogenophaga (Betaproteobacteria), and Terrimicrobium (Verrucomicrobiota) were significantly depleted in desiccated mats.

To produce initial predictions for the question of whether these taxonomic differences in community composition between control and desiccation treatments resulted in changes in predicted functions in the communities, we used the FAPROTAX method, as described above. We observed that desiccated mat communities were significantly (p < 0.05, Wilcoxon signed-rank test, Benjamini-Hochberg adjustment) enriched in predicted heterotrophic functions, including chemoheterotrophy (anaerobic and aerobic) and aromatic compound degradation (Fig. 4). Control mats were significantly enriched in predicted functions relating to sulfur metabolism, including the general “respiration of sulfur compounds” category, sulfite respiration, and sulfate respiration (Fig. 4). Interestingly, we did observe a significant enrichment of predicted animal parasites or symbionts in the desiccated mats; there were 39 ASVs with this predicted function from the genera Candidatus Xiphinematobacter, Roseomonas, Uliginosibacterium, Psychrobacter, and Coxiella (Fig. 4).

Figure 4.

Comparison of predicted functionality (using FAPROTAX database within microeco) of microbial communities between control and desiccation treatments. Only predicted functions that were significantly different between treatments are shown. Provided p-values are the result of a Wilcoxon signed-rank test with Benjamini-Hochberg adjustment for multiple tests.

Shading and recovery experiment

As with the desiccation treatment, we observed that shaded mats clearly looked different than control mats across all three ponds (Fig. 1D-F, bottom row). In particular, the control mats were orange-red in colour, while the shaded mats were distinctly green-yellow. However, we did not observe any significant differences in prokaryotic community composition between control and shaded mats, either in the PERMANOVA results (Table 1), alpha diversity (Fig. 5A), or beta diversity (Fig. 5C). There was also no significant difference in alpha diversity between control and shade treatments for both timepoints (Fig. 5B). The shade and control samples clustered closely together within each pond, indicating highly similar prokaryotic communities between treatments (Fig. 5C). We did observe some separation of samples by time point, but within each time point, sample and shade samples were tightly clustered. By extension, there were also no significant differences between control and recovery or shade and recovery treatments in PERMANOVA, alpha diversity, or beta diversity (Table 1, Fig. 5).

Figure 5.

Examining the impacts of shading and subsequent recovery on the microbial communities of microbial mats from three different ponds across several time points. (A–B) Comparison of alpha diversity (as measured via Shannon Index) between the control and desiccation treatments, either (A) aggregated across all time points or (B) split by time point. The p-values given in (A) are the result of Wilcoxon signed-rank tests, Benjamini-Hochberg adjusted for multiple tests. (C) Beta diversity measured with a Principal Coordinates Analysis (PCoA) using Unifrac distances, with relative abundance normalized read counts for ASVs. The percentages next to the axes names indicate the amount of variation in the data that that axis explains. For all analyses in this figure, samples were rarefied to an even depth prior to calculation, which required removing two samples with very low read depth (see Methods).

When examining bacterial communities at the phylum and class level, we again saw no obvious differences in community composition between control, shading, and recovery treatments (Fig. 6). Again, there were some differences between ponds, but in general, we did not observe any bacterial groups that were different between treatments. This was confirmed when a LEfSe test did not return any taxa at any taxonomic level (including species and ASV) that were able to distinguish (p < 0.05 with Benjamini-Hochberg adjustment) between control, shading, and/or recovery treatments (data not shown). Using the FAPROTAX method, as described above, to make initial predictions about differences in functional potential between treatments, we did not return any predicted functions that were significantly different (p < 0.05, Wilcoxon signed-rank test, Benjamini-Hochberg adjustment) between the control and either shaded or recovery mats (data not shown).

Figure 6.

Community composition by relative abundance of microbial mat communities at the phylum level, (A) aggregated into control vs. shade vs. recovery treatments across all ponds and timepoints, (B) as in (A) but split out by timepoint, and (C) with all replicates, timepoints, and ponds split out.

Discussion

Microbial mats are ubiquitous in Antarctic freshwater habitats, including in the meltwater ponds on the McMurdo Ice Shelf near Bratina Island. There, past research has shown, year on year persistence has been attributed to tolerance of seasonal and annual variability, coupled with a general lack of disturbance and seasonally limited, temperature-limited metabolism that makes species turnover necessarily slow (Hawes, Safi, Sorrell et al. 2011, Hawes, Safi, Webster-Brown et al. 2011, Safi et al. 2012, Webster-Brown et al. 2012), though these components have not all been well demonstrated. It is, however, clear that persistent, directional environmental change induces species turnover in these mat communities, even where function is retained, e.g. (Sutherland et al. 2020). Here we report on the response of the prokaryotic members of microbial mats from the McMurdo Ice shelf to environmental challenges that they currently face due to environmental variability, and that may increase in severity and directionality under a changing climate.

Our results indicate that, in the three ponds studied, long-term (≥ 1 year) desiccation was an extreme challenge to the maintenance of mat community composition and resulted in substantial change in relative abundance over timescales of one year. On the other hand, the bacterial community composition of mats seemed unaffected by shading (75% reduction in light availability), although the community did respond through what appeared to be a phenotypic change (coloration, most likely due to trichome migration (Vincent et al. 1993)). Thus, we partially supported our hypothesis that prokaryotic members of microbial mats would respond to stress through shifts in community composition that would facilitate retention of critical community functions. We argue for a hierarchy of responses to disturbance in these mats, with shading eliciting what appeared to be a phenotypic change that did not significantly impact community composition, preserved function (i.e. there were no significant differences—Wilcoxon signed-rank test p > 0.05, Benjamini-Hochberg adjusted for multiple tests—between shading and control mats in any predicted functional categories based on our preliminary results), and was rapidly reversible. On the other hand, desiccation elicited a compositional change, partly preserving potential function, but that will require recolonisation to return original function and structure to the mats. Below, after exploring our overall observations about the communities in these ponds, we discuss the implications of our findings considering our hypotheses around desiccation and shading.

In choosing the three ponds for this study, we looked for ponds that were similar in environmental conditions based on previous studies (Fernández-Valiente et al. 2001, Archer et al. 2014, Husain et al. 2025) (and thus would be likely to have similar prokaryotic communities), which would allow for results between ponds to be compared. All three communities were dominated by ASVs from the bacterial phyla Cyanobacteriota, Bacteroidota, and Pseudomonadota, as has been previously observed for these ecosystems (Sjöling and Cowan 2003, Jungblut and Neilan 2010, Archer et al. 2015). While the bacterial communities between ponds were significantly different from each other, our preliminary predictions of functional potential indicated that the ponds have similar functional profiles, with predicted functionality not significantly different in any functional category between ponds (adjusted p > 0.05, Kruskal-Wallis test, Benjamini-Hochberg adjusted for multiple tests). Overall, we saw highly comparable responses to our experimental treatments across all three ponds, providing stronger confidence in the validity of our results.

Our results showed quite consistent treatment effect on bacterial mat community across the time. For instance, in the shading experiment, although the different time points tended to cluster away from each other in the beta diversity analysis, the shading and recovery treatments stayed with the controls (Fig. 5C). As an additional example, the alpha diversity in the desiccation treatments did not change from years 1 to 2, despite the alpha diversity in the control mats increasing in year 1 and then returning to year 0 levels in year 2 (Fig. 2B). This increase in diversity in year 1 in the control mats is likely due to inter-annual variability in the pond environment (Archer et al. 2016).

Desiccation strongly shifts mat prokaryotic community composition

We found strong support for our hypothesis that the mat communities would experience a shift in community composition with the mortality of vulnerable taxa due to desiccation. We observed a shift away from a photosynthesis-powered mat community, supporting a rich community of predicted chemoheterotrophs and sulfur-metabolizing bacteria, towards a less diverse community that appears, based on preliminary predictions of functional potential, to be primarily chemoheterotrophic (Table 1, Fig. 3, Fig. 4). Interestingly, the loss of photosynthetic components of the community upon desiccation extended beyond Cyanobacteriota to also include the purple nonsulfur bacterium Rhodoferax (Madigan et al. 2000) (Fig. 3E). Our results indicate that most cyanobacteria in these mats are vulnerable to desiccation, which is somewhat surprising given that cyanobacteria are generally known for being resistant to harsh environments (Potts 1999). The low snowfall in this area means that the mats would be exposed to intense solar radiation in the desiccated state, which may be particularly stressful for pigmented, photooxidising organisms. Another confounding factor is that the desiccated mats were removed from the pond and thus exposed to wind erosion, which was not the case for the shading treatment. We also caution that our results are indicative only of decreased relative abundance in the community; it is possible that some of these cyanobacteria remained viable after desiccation in resting states that were resistant to DNA extraction (Hawes et al. 1992). Previous research has shown that microbial mats dominated by cyanobacteria from the Oscillatoriales order (Phormidium) are less tolerant to desiccation than mats dominated by Nostoc species, with the mechanism unknown (Hawes et al. 1992).

It is not immediately clear why certain predicted heterotrophic bacteria were vulnerable to desiccation while others were not. In the absence of physicochemical measurements and genomic data, it is challenging to generate hypotheses, especially given that species within a genus can vary with regards to given functional and growth traits. However, based on the growth characteristics of isolates from the genera identified as most differentially abundant between desiccation and control mats, we hypothesize that halotolerance may be an important characteristic for surviving desiccation.

The four genera that were most enriched in the desiccated mats compared to control mats were: Psychrobacter, Gillisia, Planococcus, and Gelidibacter. All species within the genera Psychrobacter and Gillisia are characterized by being halotolerant to halophilic (Bowman 2006, Trappen 2015). Planococcus antarcticus, the species of Planococcus identified in our data set, was isolated from lakes in the Dry Valleys in Antarctica and was found to be halotolerant (Reddy et al. 2002). Gelidibacter algens, the species of Gelidibacter identified in our data set, was isolated from a sea ice core from Ellis Fjord, Vestfold Hills, Antarctica, and is halotolerant (growth at 1.5X seawater, ∼5% NaCl) (Bowman et al. 1997). On the other hand, Cypionkella (formerly Pseudorhodobacter), Rhodoferax, Terrimicrobium, Hydrogenophaga, and Geitlerinema (a cyanobacterial species) were most affected by desiccation. Rhodoferax saidenbachensis, the species of Rhodoferax in our data set, and related species, have no tolerance for NaCl (Madigan et al. 2000, Kaden et al. 2014). All species of Terrimicrobium are strict anaerobes and not tolerant for NaCl (Qiu et al. 2014). Hydrogenophaga taeniospiralis, the species of Hydrogenophaga identified in our data set, is incapable of growth at NaCl concentrations above 0.5% (WILLEMS et al. 1989). On the other hand, Cypionkella contains species from diverse habitats with diverse salt tolerances (Lee YM et al. 2016), and no species assignment could be made for the Cypionkella species in our data set. Further exploration is needed to validate this hypothesis, potentially using metagenomic and/or metatranscriptomic data.

Desiccation is very likely to increase the salt concentrations of the mats, especially given that the measured conductivity near the bottom of these ponds is 2–15 mS/cm (Wait et al. 2006, Husain et al. 2025), indicating salinity levels were likely greater than those of freshwater in the mats prior to desiccation. Salinity has previously been found to have a strong, significant effect on the microbial communities inhabiting the soils of the McMurdo Dry Valleys, located not far from Bratina Island (Zeglin et al. 2011, Horn Van et al. 2014, George et al. 2021). Similarly, desiccation has previously been found to negatively impact the microbial diversity of mats elsewhere globally (Rothrock and Garcia‐Pichel 2005), as well as in Dry Valleys soils (Cary et al. 2010, Lee KC et al. 2018).

In our preliminary functional potential prediction data, the loss of sulfur metabolism in the desiccation treatment is somewhat challenging to explain. Generally, sulfur respiration similar to that predicted in these mats thrives in anoxic environments generated inside photosynthetic mats upon heterotrophic consumption of organic matter produced via photosynthesis by associated organoheterotrophs (Risatti et al. 1994, Krekeler et al. 1998). However, it is unlikely that the loss of anoxic environment was the cause of the loss of sulfur respiration in the desiccated mats, since anaerobic chemoheterotrophy in general increased in the desiccation treatment, indicating that at least some parts of the desiccated mats remained anaerobic (Fig. 4). Several possibilities to reconcile these two results include: the sulfur-respiring microbes in the mats were also reliant on the production of sulfur-containing compounds from photosynthetic components of the mat; the sulfur-respiring microbes were incapable of migrating into the newly anaerobic portion of the mat to continue sulfur-respiration activity; or, one or both of these functional predictions based on 16S rRNA gene sequence similarity are incorrect.

Mat prokaryotic community composition is resilient to shading

We had hypothesized that the microbial mats exposed to shading would experience shifts in prokaryotic community composition, but that the functional potential of these communities would remain similar. Surprisingly, we did not observe any significant changes to bacterial community composition induced by shading, indicating that the bacterial composition of the mat community is not vulnerable to a full year of 75% reduction in light availability (Fig. 5, Table 1). We did, however, observe visual changes to the colour of the shaded mats, from red-orange in the control to yellow-green in the shaded mats. These mats have previously been observed to be orange red at the surface of the mats, due to the presence of photoprotective carotenoids (Vincent et al. 1993), with that coloration changing to green within hours of the mats being shaded (Vincent et al. 1993, Nadeau et al. 1999). This change in coloration with shading was shown to be due to the vertical migration of phycocyanin-rich, carotenoid-poor Oscillatoria sp. trichomes up to the surface. Under full sunlight, these were located deep within the mats at their optimum ∼5% irradiance depth, where this shade adapted taxon is less vulnerable to photodamage (Nadeau et al. 1999). Our results indicate that the response of mats to prolonged shading is highly similar to that observed after only a few hours of shading, with a physiological rather than compositional response facilitating resilience.

Annually, these photosynthetic mats completely cease photosynthetic activity with the onset of autumn and accompanying increase in ice thickness and decrease in sunlight availability, and photosynthesis continues for some time after ice has formed (Hawes, Safi, Webster-Brown et al. 2011). Thus, cyanobacteria in these ponds are accustomed to annual variations in light levels, which may be why we did not observe any changes in abundance of the cyanobacterial populations. Further experiments with increased measurement of environmental variables and metabolic processes may help resolve the physiological responses of these mats. In any case, it appears that the heterotrophic members of the mat community were also resilient to 12 months of shading. It is possible that the bacterial members of the mat communities in the shading experiment did experience significant shifts in response to shading but at the gene expression level; future studies incorporating metatranscriptomic data could examine this possibility.

Conclusions

The microbial mats of the Bratina Island meltwater ponds experience several environmental challenges on an annual basis, with additional stresses likely to be experienced due to climate change. In this study, we have provided evidence for a hierarchy of responses in these mats: conformational change in response to shading (vertical migration of Oscillatoria trichomes) and compositional change in response to desiccation (significant shift in community composition upon desiccation). In particular, multiple community members were found to be vulnerable to long-term desiccation, with halotolerance being implicated as a trait that may enhance survival during desiccation. Two significant weaknesses in our conclusions relate to the use of indicative metabolic functions from 16S rRNA gene databases, and that we were able only to target DNA, which does not in itself indicate viable organisms. Metagenomic and metatranscriptomic work exploring the genetic and expressed potential of the microbial communities in the ponds would be needed to confirm these preliminary results. This study lays the groundwork for future work exploring how these communities respond to the potential long-term challenges associated with climate change.

Funding

None declared.

Author Contributions

Stephen E. Noell (Formal Analysis, Investigation, Software, Validation, Visualization, Writing—original draft, Writing—review & editing), Elizabeth C Colville (Formal Analysis, Methodology, Visualization, Writing—original draft), Alexis Marshall (Conceptualization, Formal Analysis, Methodology, Software, Supervision, Visualization, Writing—original draft, Writing—review & editing), Ian Hawes (Conceptualization, Funding acquisition, Investigation, Project administration, Resources, Supervision, Validation, Writing—original draft, Writing—review & editing), and Ian R McDonald (Conceptualization, Methodology, Project administration, Resources, Supervision, Writing—original draft, Writing—review & editing)

Conflicts of interest

None declared.

Data availability

All R scripts used to analyse the data are available on Github at https://github.com/ThermophileResearchUnit/Bratina-island-manipulation-16S. All sequence data is available in GenBank. 16S rRNA gene sequence data has been deposited as accessions PX888075–PX889830.