Snow algae communities in Antarctica: metabolic and taxonomic composition

Field collections in Antarctica

Snow algae communities (Fig. 1) were collected in 6 × 50 ml sterile plastic sample tubes from layers of green and red dominant snow algal blooms at four locations in Ryder Bay, Antarctic Peninsula (Rothera Point, Anchorage Island, Léonie Island and Lagoon Island) in austral summer (January–February) 2015 (Supporting Information Table  S1). It was not possible to determine whether all blooms surveyed were successional stages or distinct assemblages. The layers of algae were 1–5 cm deep, but with much heterogeneity within a bloom, with only the algae layer sampled in that 1–5 cm depth. The algae were collected by filling a sterile 50 ml tube with snow, which was not compacted. Seven blooms were studied at Rothera Point (42 samples), nine at Anchorage Island (54 samples), five at Léonie Island (30 samples) and 10 at Lagoon Island (60 samples), resulting in 186 samples for subsequent analyses. Blooms lasted for at least 42 d with bloom areas ranging from c. 5 m² to > 2500 m². Single point photosynthetically active radiation (PAR) (Skye PAR Quantum Sensor, Skye Instruments Ltd, Llandrindod Wells, UK) received at the snow surface, together with temperature measurements (standard glass thermometer) at the snow surface and 5 cm depth were also recorded at the start of each sampling period, which lasted for 30–60 min. Samples were returned within 3 h of sampling to the Bonner Laboratory (Rothera Research Station, Ryder Bay, Antarctica), where they were melted in 4°C lit incubators (Sanyo, Osaka, Japan). Algal cell density was measured by adding 6 μl of snowmelt into Hycor Kova haemocytometer wells and counting the number of algal cells using bright field microscopy. Algal community dry cell mass was obtained by gravity filtration of 50 ml of melted snow through a preweighed dry filter (Whatman GF/C, 47 mm). Filters were dried at 80°C for at least 48 h before reweighing. Samples for FT‐IR analysis, which enabled measurements to be obtained as close as possible to the time of sampling ensuring minimal metabolic degradation, were processed by pelleting 2 ml of snowmelt (2000 g for 10 min at 4°C), discarding the supernatant and drying the pellet at 80°C for 24 h, followed by 24 h in a desiccator. The dried pellets were analysed on station using a Perkin‐Elmer Spectrum Two FT‐IR device, set to measure the absorbance intensity between wavenumbers 400 and 4000 cm⁻¹ and normalised against air. For the metabolite and genomic analysis, carried out at the Department of Plant Sciences, Cambridge, 10 ml of snowmelt was pelleted using centrifugation (2000 g for 10 min, 4°C), after which the supernatant was discarded and the remaining algal pellet was flash frozen in liquid nitrogen and stored at −80°C. Live algae were transported to the UK by adding 20 μl of snowmelt containing algae onto Tris‐acetate‐phosphate (TAP) agar slopes in 30 ml clear plastic tubes for growth under controlled conditions (4°C, 10 μmol m² s⁻¹ 12 : 12 h, light : dark cycle) at Cambridge and were imported under UK APHA/DEFRA licence number 119979/260872/0. The slopes and frozen samples were transferred to the UK by ship at 4°C or −80°C.

Pigment analysis

Total Chl and carotenoid concentrations were determined after extraction of pigments from cell pellets (from 1 ml snowmelt) with 1 ml dimethylformamide using the equations of Inskeep & Bloom (1985) and Wellburn (1994). Individual pigments were analysed by HPLC by first adding 1 ml of deionised water to resuspend the pelleted cells. The resuspended pellets were transferred to a 2 ml microfuge tube and repelleted (16 000 g for 10 min). The supernatant was removed and the remaining pellet was homogenised with glass beads and frozen in liquid nitrogen three times, after which 1 ml of dimethylformamide was added and sonicated for 30 min. Samples were repelleted and the supernatant was transferred to an HPLC glass vial, mixed with methanol (3 : 2) and stored at −80°C until analysis. Pigments were separated by HPLC (Surveyor system; Thermo Scientific, San Jose, CA, USA) as described by Remias & Lütz (2007) but using an injection volume of 50 μl, and were resolved on a Luna C18 column (250 × 2.0 mm; Phenomenex, Macclesfield, UK). Peaks were compared against standards (astaxanthin and astaxanthin esters, lutein, Chla and Chlb, and β‐carotene; all Sigma Aldrich) (Inskeep & Bloom, 1985; Wellburn, 1994).

Total cellular lipids and FAMEs

Lipids were extracted using the chloroform/methanol/water method and triacylglycerides (TAGs), polar lipids and free fatty acids in the total lipid extract and total fatty acid methyl esters (FAMEs) were analysed by GC, as described in Davey et al. (2014).

Metabolite profiling

Soluble polar and nonpolar metabolites were extracted using the methanol/chloroform/water method as described in Davey et al. (2008). Compounds within the polar methanol/water phase were derivatised by N‐methyl‐N‐trimethylsilyl‐trifluoroacetamide (MSTFA) and trimethylsilyl (TMS) as described by Dunn et al. (2011) and subsequently separated and profiled by GC‐MS (Thermo Scientific Trace 1310 GC with ISQ LT MS, xcaliber v.2.2) with a ZB‐5MSi column (30 m, 0.25 mm ID, 0.25 μm film thickness; Phenomenex). The injection volume was 1 μl (splitless) with an injector temperature of 300°C, using helium as a carrier gas (constant flow rate of 1.0 ml min⁻¹). The following gradient was used: initial oven temperature 70°C; 130°C at 10°C min⁻¹; 230°C at 5°C min⁻¹; 310°C at 20°C min⁻¹; hold for 5 min. The MS conditions in the positive mode were: transfer line 310°C; ion source, 310°C; mass range 45–800 Da; dwell time 0.17 amu s⁻¹. GC‐MS spectra were aligned to an internal standard (phenyl‐β‐d‐glucopyranoside hydrate 98%, Davey et al., 2008) and processed using thermo tracefinder (v.3.1) and nist software (nist v.2.0 http://www.nist.gov/srd/nist1a.cfm) to aid identification based on molecular mass. Pathway analysis of the identified metabolites used metaboanalyst open source software (v.4, pathway analysis tool) using the Arabidopsis thaliana metabolic pathway library (Chong et al., 2018, www.metaboanalyst.ca). R script for the metaboanalyst software can be downloaded at https://github.com/xia-lab/MetaboAnalystR (Chong & Xia, 2018).

Metabarcoding

Frozen pellets (c. 1 cm³) of field‐collected algal communities from 10 ml snowmelt were allowed to thaw before being resuspended in 1 ml of RNase‐free water. After transferring to a clean 1.5 ml microfuge tube, the samples were ground with sterilised sand before adding another 1 ml of RNase‐free water and subsequent transfer to a 15 ml tube to which 3 ml of SDS‐EB buffer (2% SDS, 400 mM NaCl, 40 mM EDTA, 100 mM Tris‐HCl, pH 8.0) was added, followed by mixing by vortexing and shaking for 5 min at 4°C. Subsequently, 3 ml of chloroform was added, mixed gently by inversion and the whole suspension was centrifuged for 5 min at 2000 g and 4°C, resulting in a two‐phase separation. The top aqueous phase was transferred to a new 15 ml tube and two volumes of 100% chilled ethanol were added before incubating overnight at −20°C. The following day, the mix was spun at 6800 g at 0°C for 30 min. After carefully discharging the supernatant, the pellet was resuspended with 1 ml of ethanol (70%) and recovered in a clean microfuge tube before determining total RNA concentration and quality. Libraries of the fourth hypervariable (V4) domain of the 16S rRNA gene and ITS of the rRNA gene were produced using the NEXTflex ‘16S V4’ and ‘18S ITS’ Amplicon‐Seq Library Prep Kit and primers (BIOO Scientific, Austin, TX, USA), respectively. For consistency we hereafter use the term ‘ITS’ for the NEXTflex 18S ITS region. The microbial 16S rRNA gene forward primer (V4 Forward) sequence was: 5′‐GACGCTCTTCCGATCTTATGGTAATTGTGTGCCAGCMGCCGCGGTAA‐3′ and the reverse primer (V4 Reverse) sequence was: 5′‐TGTGCTCTTCCGATCTAGTCAGTCAGCCGGACTACHVGGGTWTCTAAT‐3′. The eukaryotic ITS forward primer (18S ITS Forward) sequence was: 5′‐CTCTTTCCCTACACGACGCTCTTCCGATCTTCCGTAGGTGAACCTGCGG‐3′ and the reverse primer (18S ITS Forward) was 5′‐ CTGGAGTTCAGACGTGTGCTCTTCCGATCTTCCTCCGCTTATTGATATGC‐3′. Samples were sequenced by Cambridge Genomic Services (Cambridge, UK) using an Illumina MiSeq v3 600‐Cycle Sequencer following the manufacturer's protocol and primers. Quality control analysis of the Illumina MiSeq paired‐end reads (2 × 300 bp) was performed using fastqc (https://www.bioinformatics.babraham.ac.uk/projects/fastqc/). Taxonomic analysis of 16S rRNA gene sequences was performed using qiime 2 release 2017.10 (Caporaso et al., 2010; https://qiime2.org). In brief, for each sample, demultiplexed paired‐end sequences were imported into qiime 2. Potential amplicon sequence errors were corrected with the qiime 2 implementation of dada2 (Callahan et al., 2016). To remove lower quality bases, reads were truncated at position 280 based on the fastqc reports during this step. Taxonomy was assigned to the sequences in the feature table generated by dada2 using silva release 128 as the 16S/ITS(18S) marker gene reference database (Quast et al., 2013), trimmed to the V4 region, bound by the 515F/806R primer pair used for amplification. Taxonomic analysis of the ITS sequence data was done as described for the 16S rRNA gene up to the taxonomic assignment step. Because of the lack of an ITS marker reference database representative of the species diversity of the environments investigated, we carried out sequence similarity searches using NCBI blast (Altschul et al., 1990) against release 134 of the European Nucleotide Archive (ENA). Taxonomic assignments were made manually, based on blast scores and the presence or absence of ambiguity of the taxonomic lineages reported by blast. For ambiguous blast hits (e.g. a similar score for unrelated taxa), the lowest common denominator was used for taxonomic assignment. Since there were so few operational taxonomic units (OTUs), and in the absence of an ITS database that is representative of the communities under investigation, we treated these results as exploratory. Sequence reads were submitted to the ENA Sequence Read Archive at the European Bioinformatics Institute (https://www.ebi.ac.uk/ena) and are available under accession number PRJEB23732.

Targeted 18S rRNA gene PCR for isolated snow algae species

Snow algae were isolated from a field sample (Lagoon Island) and grown axenically on TAP agar plates supplemented with ampicillin (50 μg ml⁻¹) and kasugamycin (50 μg ml⁻¹). Cultures were maintained under a 12 : 12, light : dark photoperiod at 4°C. PCR mixtures contained REDTaq ReadyMix PCR Reaction Mix (Sigma) with 10 μM of the 18S rRNA gene universal eukaryotic primers (forward sequence ‘SA Forward’: 5′‐CGGTAATYCCAGCTCCAATAGC‐3′, reverse sequence ‘SA Reverse’: 5′‐GTGCCCTTCCGTCAATTCC‐3′), with expected product size of 582–584 bp. Primers were adapted from Wang et al. (2014) where our forward primer was a 3′ section of the first primer in their Table 1, with the addition of nucleotides downstream of it so the primers would have similar annealing temperatures, avoiding a run of guanines and cytosines and to avoid the possibility of forming stable secondary structures or primer dimers. Our reverse primer is the reverse complement of a section of the second listed primer in their Table 1. The PCR cycle was 95°C for 5 min, 95°C for 20 s, 55°C for 20 s, 72°C for 1 min (35 cycles), and 68°C for 5 min. PCR products were extracted from the gel using a QIAquick Gel Extraction kit (Qiagen), following the manufacturer's instructions, and nucleotide sequencing (both directions using the above primers) was performed using Applied Biosystem sequencing platforms (Abi 3730xl genome analyser, 50 cm 96 capillary array) at Source BioScience (Cambridge, UK) and viewed using snapgene v.4.2.4. Nucleotide sequences were deposited at GenBank and are available under accession numbers MK330877–MK330880.

Statistics

To determine whether the differences between green and red communities, or among bloom site locations, were statistically significant, t‐tests (Excel, Microsoft Office 2007) or two‐way ANOVA with Tukey's test (sigmaplot v.13.0, Systat Software Inc., Chicago, IL, USA) were performed. Multivariate analyses to test whether green and red communities could be discriminated based on their identified and unidentified metabolites (from FT‐IR fingerprints or GC‐MS profiling datasets) were performed using principal components analysis (PCA) (Paliy & Shankar, 2016) on unit‐variance scaled data (FT‐IR absorbance values or GC‐MS identified peak area units) within the simca‐P v14.1 PCA pipeline (Umetrics, Umeå, Sweden) to produce standard score scatter plots and ranked score contribution plots of how each variable (FT‐IR wavenumber or GC‐MS metabolite) contributed to clustering within the PCA score scatter plot.

Alpha diversity and beta diversity were measured using qiime 2's diversity analyses (q2‐diversity) plugin (version 2017.10; https://docs.qiime2.org/2017.10/tutorials/moving-pictures/). In brief, the core‐metrics‐phylogenetic method was applied, which first subsamples the counts from each sample in order to obtain an even sampling depth (43 828 for the 16S rRNA data and 69 for the ITS data). Alpha and beta diversity metrics (Faith's Phylogenetic Diversity and Pielou's Evenness) were subsequently computed (PERMANOVA). To determine whether the richness of the samples had been fully captured, alpha diversity rarefaction plots were calculated using the qiime diversity alpha‐rarefaction visualiser. Plots for 16S rRNA gene and ITS data approached saturation at depths of 43 828 and 69 sequences, respectively (Fig.  S1), suggesting that the majority of the diversity in the communities had been captured. Beta diversity nonmetric multidimensional scaling (NDMS) plot analysis of the raw read metabarcoding data was carried out using r script in vegan (Methods  S1; https://jonlefcheck.net/2012/10/24/nmds-tutorial-in-r/) with stress levels of 0.1–0.2, using the Bray–Curtis dissimilarity calculation (Paliy & Shankar, 2016). Given the limited access to some blooms, and to reduce the environmental impact, the sample numbers (n) varied (1–30) per site, and these sample numbers are given in the figure and table legends.

Algal community cell density and biomass

There were many fewer cells in the red algal communities than in the green communities, with mean (± SD) cell densities of 0.15 × 10⁶ cells per ml snowmelt (± 0.17 × 10⁶) vs 1.24 × 10⁶ cells per ml snowmelt (± 0.79 × 10⁶), respectively (Fig.  S2a). Similarly, the two communities differed in dry biomass (mg l⁻¹ of snowmelt), with, on average across all sites, 62% less biomass in the red‐dominated communities than in the green communities (Fig.  S2b). There were no significant differences (ANOVA, P > 0.05) in cell numbers or cell dry masses between sampling locations in the red communities. However, the green communities on Léonie Island had a greater number of cells (c. 2.2 × 10⁶ more cells per ml) and biomass (c. 2.9 g dry mass more per litre) of snowmelt than on the other islands (ANOVA, P ≤ 0.01).

FT‐IR metabolite fingerprinting

FT‐IR was used during the field campaign to analyse the metabolic composition of the snow algae communities as close to field conditions as possible (Fig. 2a). Based on PCA, there were specific FT‐IR wavenumber regions that were strongly associated with green (1489–1581, 1589–1664 cm⁻¹) or red (1002–1094, 1141–1144, 1732–1756, 2850–2856, 2911–2933 cm⁻¹) snow algae communities (Fig. 2b). These regions were associated with protein bands (amide I and II) – suggesting active growth – in the green communities, whereas in red communities the major features were associated with lipids, lipid esters and polysaccharides (Fig. 2b; Table  S2). There was no clustering of samples based on sample island location. Little variation across island locations suggests that the sampling regime was standardised and robust.

Pigments composition in snow algae blooms

Crude solvent extractions and ultraviolet–visible spectrometry of green and red blooms shortly after field collection showed the presence of peaks indicative of Chla and b and astaxanthin (Fig.  S3). There were significant differences in the total Chl content of green‐ and red‐dominant snow algae communities, with more total Chl present in the green than in the red communities when expressed on a per unit dry mass (c. 83% more) and per volume snowmelt (c. 90% more) (Fig.  S4a,b). There was no difference in the concentrations of total carotenoid between the red and green blooms, when expressed either on a per unit dry mass basis (Fig.  S4c) or on a per litre of snowmelt (Fig.  S4d) basis. There was a single effect of location, with the Chl content of the Léonie Island green community having more Chl per unit of snowmelt than the other islands (P ≤ 0.05), which was probably due to the greater biomass per unit of snowmelt at that location. There were no other effects of location on total Chl or total carotenoid in the red communities.

Pigment composition was analysed in detail using HPLC. As there was minimal effect of location from the above pigment analyses, samples from the islands were pooled to provide an average composition over Ryder Bay for green and red communities (Table 1). Dominant pigments in the green community were Chla and b, β‐carotene and lutein, and in the red community were Chla and b and astaxanthin esters. The main differences between the communities were that the red communities had significantly less Chla and b, β‐carotene, lutein (t‐test, P ≤ 0.05) and xanthophyll and more astaxanthin‐like and astaxanthin esters, although these were not significant (Table  1). The ratio of Chla to Chlb was similar in both green (1.9 : 1) and red (2.2 : 1) communities.

Lipid profiling of green and red blooms

More detailed analyses using GC‐FID and GC‐MS were performed to qualify and quantify differences between communities identified from FT‐IR spectra. We first measured the overall glycerolipid composition of the samples (expressed as mg per unit dry cell mass). Unlike the other metabolites detected, there was little variation in the nonpolar (TAG) and polar lipid content between the green and red communities, nor between the different island locations. The exception to this was significantly (P ≤ 0.05) more neutral storage lipids (TAGs) per unit dry mass in the lipid extracts from the red community at Lagoon Island (Fig. 3a). A significantly higher concentration of free fatty acids was also measured in the red communities from Léonie Island (Fig. 3e). However, when data were expressed per litre of snowmelt, the polar membrane lipid content of the red community from Léonie Island was less than that of the green community (P ≤ 0.01; Fig. 3d), and the mean free fatty acid concentration in the red communities was lower compared to the green communities from all locations except Lagoon Island (P ≤ 0.05; Fig. 3f). There were no significant differences in lipid composition in either the red or the green communities between the sampling locations.

FAME analyses showed that the snow algae communities contained a range of saturated and unsaturated fatty acids from C14:0 to C22:6 and were rich in the saturated C16:0 and unsaturated C18:1(11) fatty acids (Table  S3). Overall, there were no statistically significant differences in the fatty acid profile between the green and red communities, but the trend was for greater amounts of C16:0, C18:1(11) and C18:1(9) fatty acids in the red community, and lower or similar amounts of all other fatty acids.

Metabolic profiles of red and green snow algae communities

To provide further insight into the metabolic composition of the different snow algae communities, an untargeted metabolic profiling approach was used in which the extracts were derivatised by MSTFA and analysed by GC‐MS, and the peaks identified. PCAs showed distinct clustering of green and red communities (Fig. 4a). There was no clustering of samples based on island location in any of the principal components. The score contribution of metabolites (based on their molecular masses and comparisons with NIST MS libraries) and the metabolic pathways in which they are involved were ranked in order of importance for either green or red snow algae communities using the in silico MetaboAnalyst Pathway Analysis Tool. Metabolites involved in energy production and the TCA cycle and in nitrogen and amino acid metabolism, such as succinic acid and the amino acids asparagine and valine, were strongly associated with green snow algae communities (Fig. 4b; Tables  S4, S5). The most frequent metabolites associated with red communities were quite different from those in the green communities. They were largely associated with osmolytes (mannitol, xylitol) and the fatty acids heptadeconoic acid (an unsaturated C17 fatty acid) and dimethyl‐heptanoic acid (a C7 volatile acid).

Snow algae community composition

Bright‐field microscopy revealed mainly flagellated and nonflagellated green algal cells representing the vegetative stage in green blooms, and orange to red mature zygospores or large hypnozygotes in red blooms (Fig.  S3). Although green algal blooms in Ryder Bay were observed to become red over periods of about 30 d, given the remote location, which precluded many repeat samplings, it could not be irrefutably established whether the red forms of the cells were derived from the green vegetative cell forms, or if the red cells were a separate assembly that succeeded green blooms.

We therefore carried out metabarcoding analysis to investigate the species composition of the communities, via sequencing libraries of the V4 region of 16S rRNA gene and ITS region of each community and then by NMDS plot analysis of the 16S rRNA gene or ITS OTUs. Both read frequencies and percentage contributions were obtained for two major taxonomic levels (Kingdom/Phylum Level 2 to Genus Level 7). To test for associations between discrete metadata categories (green and red algae communities) and alpha diversity data, the community richness and evenness were calculated using Faith's Phylogenetic Diversity (a qualitative measure of community richness that incorporates phylogenetic relationships between the features) and Pielou's Evenness (a measure of community evenness). No significant differences in community richness or evenness were measured between green and red 16S rRNA gene and ITS sequence‐based communities (P ≥ 0.05, Kruskal–Wallis test) (Fig.  S5). For beta diversity, a PERMANOVA (Anderson, 2001) test (using qiime 2's beta‐group‐significance command) on unweighted UniFrac distances generated during the first diversity analysis step was used to test whether sequence reads from samples within a bloom type were more similar to each other than they were to samples from the other bloom type. Similar to alpha diversity, there was no significant dissimilarity between the green and red 16S rRNA gene and ITS communities (P ≥ 0.05) (Fig.  S6). Beta diversity NMDS plots of both 16S rRNA gene and ITS data sets also revealed close taxonomic similarities between the green and red communities (Fig.  S7).

Despite being the major observable organisms in the samples, the Chlorophyta contribute little to the overall diversity, which was instead dominated by fungi, other protists and bacteria (Fig. 5; Tables  S6–S8). However, of note is the difference in the Chlorophyta between the green and red blooms: in the green blooms OTUs whose closest hit in the databases were to Chloromonas, Chlamydomonas and Chlorella were detected in approximately equal measures, whereas in the red blooms only Chloromonas was identified, with the other OTUs being assigned to unknown Chlorophytes (Fig.  S8), indicating that the red community contains other, unidentified green algal species. Further investigation into the identity of the snow algae in the red blooms, based on preliminary analysis of their morphology, suggested that they could either be Chlamydomonas nivalis or Chloromonas nivalis. 18S rRNA gene PCR was thus performed on nucleotide extracts from red snow algae cultures that were isolated from a field sample (Lagoon Island) and grown axenically, during which the cells transformed from their red phase to a green phase (over 21 d). A blast search of the forward and reverse nucleotide sequences resulted in 98–99% similarity to Chloromonas sp. and Chlamydomonas sp., supporting our initial classification. However, a blast search against only Chlamydomonas nivalis or Chloromonas nivalis sequences resulted in just 92% similarity (Table  S9), indicating that these might be other species. At the class level, sequence reads from the 16S rRNA gene metabarcoding showed that the communities were dominated by Flavobacteria, Sphingobacteria and beta‐proteobacteria, in particular Flavobacterium, Pedobacter and Hymenobacter, respectively (Fig. 5a). The number of Sphingobacteria OTUs was statistically significantly lower (P < 0.05) and Chryseobacterium reads significantly higher (P < 0.01) in the red communities compared to the green communities (Tables  S7, S8).

Metabolic composition differs between green and red snow algae blooms

From direct field analyses, we have shown that there are substantial differences in biomass and cell densities between green and red blooms. Additionally, the initial FT‐IR untargeted metabolic profiling revealed that wavenumbers associated with protein/amino acids were more abundant in the green blooms, whereas lipid and carbohydrate chemistry predominated in the red blooms. Such differences in FT‐IR spectra between the green and red communities were similar to those detected by single‐celled synchrotron‐based IR spectroscopy of Arctic snow algae communities by Lutz et al. (2015), who found that green communities were dominated by functional groups associated with proteins and red communities by lipids.

Astaxanthin increases and Chl content decreases in red blooms

Snow surface pigmentation is a fundamental marker for identifying and classifying snow algae communities during field campaigns and in research based on satellite images (Fretwell et al., 2011). However, it cannot be assumed that pigment composition does not vary between blooms, especially as the species composition of the green and red communities have not been previously described. We found that between green and red communities, although the composition of the pigments was similar, the concentrations of each pigment was not. Our pigment data for the Ryder Bay snow algal communities were consistent with those of other snow algal blooms around the world. For example, Remias et al. (2010, 2013) and Lutz et al. (2015, 2016, 2017) reported higher concentrations of Chl and xanthophyll cycle‐related compounds in green blooms, but, unlike Remias et al. (2010), we were unable to detect α‐tocopherol (vitamin E) in the cells. The detected carotenoids potentially play a key role in energy dissipation in chloroplasts under high light conditions (Demmig‐Adams & Adams, 1996; Remias et al., 2010). Higher concentrations of astaxanthin esters in red blooms have also been reported elsewhere (Remias & Lütz, 2007; Lutz et al., 2015), the production of which can be dependent on developmental stage (Holzinger et al., 2016) or upon environmental stresses, in particular light intensity and nutrient deficiency (Remias et al., 2005, 2010; Lutz et al., 2014; Minhas et al., 2016).

Concentrations of free fatty acids vary, but those of glycerolipids largely do not, between green and red blooms

Our analyses with both untargeted and targeted metabolomic profiling approaches (Bundy et al., 2009) reveal many differences and similarities between the green and red bloom communities and confirmed the field findings from the FT‐IR data. Lipid profiling showed that only the free fatty acid concentrations differed between bloom types (with higher concentrations in the red communities), and that the glycerolipid and fatty acid composition was similar between blooms. Such profiles are characteristic of other snow algae blooms, in which the high degree of fatty acid saturation is hypothesised to be related to membrane stability at low temperatures (Bidigare & Ondrusek, 1993; Spijkerman et al., 2012; Leya, 2013). Specifically, this is characteristic of Chlamydomonas nivalis (Řezanka et al., 2014), with Lukeš et al. (2014) relating the membrane lipid composition to a broad thermal tolerance in terms of growth, electron transport and oxygen evolution, compared to the temperate species Chlamydomonas reinhardtii.

Distinct metabolic profiles of red and green snow algae communities

GC‐MS profiling revealed that the dominant metabolites and metabolic pathways in green blooms were associated with nitrogen and amino acid metabolism, and in red blooms with osmolyte and fatty acid metabolism. High on the list of compounds associated with green communities were lysine and its precursor aminoadipic acid, the latter importantly being a precursor for penicillin synthesis in fungi that produce α‐aminoadipate (Fazius et al., 2012). Also dominant in the green blooms was calystegine, an alkaloid involved in plant–bacterial communication, including with Pseudomonas (a reported bacterial genus in our 16S sequencing), which is reported to catabolise it (Goldmann et al., 1996). Glycerol, sugar alcohols and other low‐molecular‐weight carbohydrates have been reported previously in red snow algae, with their function associated with osmotic acclimation (Eggert & Karsten, 2010). Remias et al. (2013) also detected high concentrations of glycerol and sugar alcohols in red blooms in Antarctica. By contrast, studies of snow algal communities in the High Arctic by Lutz et al. (2015) found that compounds related to purine and tryptophan metabolism were more abundant in green communities than in red communities, and were considered important in the increased growth rates of green blooms. Although we detected metabolites in these pathways, they were not identified as key determinants for either green or red blooms. This suggests that the community composition and exudates could be functionally different between Arctic and Antarctic sites. Such metabolic differences could be due to acclimation effects or true adaptation to the local environment, although more sites would need to be studied before these hypotheses can be fully tested.

Snow algae community composition differs between bloom types but that of associated fungi and bacteria does not

The algal cells in our study were structurally similar to snow algae cells described in the Arctic (Svalbard) and North America (Hoham et al., 1983; Müller et al., 1998), with some red cells appearing morphologically similar to Haematoccus pluvialis (Wayama et al., 2013). Metabarcoding revealed that communities were dominated by an unknown Alveolata (SAR), a yeast in the genus Cryptococcus (Tremellaceae) and chytrids in the Rhizophydiales (Fungi) (Fig. 5). The Rhizophydiales are zoosporic fungi common in wet, cold habitats, and have been widely reported in other snow algae blooms (Schmidt et al., 2012; Naff et al., 2013; Brown et al., 2015; Comeau et al., 2016; Seto & Degawa, 2018). Within the Chlorophyta, we were able to detect Chlamydomonas, Chlorella, uncultured Chloromonas and two taxa assigned as unknown Chlorophyceae and unknown Chlorophyta. Whether the Antarctic community contains endemic species requires further study over a wider area (De Wever et al., 2009; Petz et al., 2007). In this context, Remias et al. (2013), studying snow algal communities from locations north (Goudier Island and Paradise Harbour, 64°S) of our location (c. 68°S), determined species matching our metabarcoding OTUs for the red blooms (Chloromonas), implying that this taxon may be distributed widely, at least along the Antarctic Peninsula. In a similar study in continental Antarctica (Yatude Valley, Langhovde at 69°S), Fujii et al. (2010) also identified a similar community of Chlorella, Chlamydomonas and Chloromonas as well as other green algae (Raphydonema and Koliella) and, as here, a range of yeasts. Although the detailed composition of bacterial, protist and fungal communities in snow algal blooms may be different across the globe, it is becoming apparent that snow algae communities have similar wide functional and taxonomic structures. The dominance of fungi, specifically yeasts and chytrids, in the communities studied here is of particular note, with fungi also having been identified as important components of Arctic snow packs (Maccario et al., 2014). Bacteria and fungi can utilise simple and complex organic compounds within the snow pack (e.g. Pseudomonas, See‐Too et al. (2016)) and numerous studies have reported bacterial genera, such as Polarmonas, Flavobacteria and Sphingobacteria, living in close association with Chloromonas and Chlamydomonadaceae (Hoham & Duval, 2001; Komárek & Nedbalová, 2007; Hisakawa et al., 2015; Lutz et al., 2015, 2016, 2017; Hamilton & Havig, 2017). There were also a large number of OTUs that could not be assigned to a genus or species, implying that the communities in the Antarctic snow packs are yet to be fully identified, characterised and incorporated into public databases such as Silva (Quast et al., 2013). Additionally, the diversity of the genetic, metabolic and growth phenotypes of a wider spatial range of populations need to be determined, to assess their resilience to ongoing environmental changes and to predict future shifts in their ranges. Whether the structure of these communities will be sufficiently resilient to withstand abiotic parameters that are outside the normal range of their niche, as a result of climate change, remains to be tested by resolving the phenotypic plasticity of each species (Hoham, 1975; Morgan‐Kiss et al., 2006; Convey et al., 2014; Rengefors et al., 2015). Overall, such studies will contribute to understanding the functional and taxonomic diversity of polar microbial ecosystems (Keeling et al., 2014; Cavicchioli, 2015) and the contribution of snow algae to polar ecosystems and global carbon budgets.