Introduction

Meltwater ponds are a common feature in summer in many ice-free areas in Antarctica, particularly along the coastal margins (Hawes et al. 2011). The McMurdo Ice Shelf (MIS) hosts a network of meltwater ponds and small lakes that cover about 1500 km of the Shelf, making it one of the most diverse wetlands in Antarctica (Howard-Williams et al. 1990). There are large variations in the physico-chemical conditions amongst these meltwater ponds, despite some being close to one another (de Mora et al. 1994; Sutherland 2009). Rich in both benthic microbial mat biomass and species diversity, these ponds form important biodiversity elements in otherwise barren landscapes on the MIS, hosting the largest concentration of non-marine organisms in the southern Ross Sea region (Vincent 2000; Quesada et al. 2008). The mats are dominated by cyanobacteria and, to a lesser extent, diatoms, and underpin much of the non-marine productivity in this area (Vincent 2000).

Many of these ponds are closed basins where water accrual occurs by melt of ice and snow whilst water loss is by ablation, sublimation and evaporation (Hawes et al. 2014). Water chemistry responds quickly to changes in the water balance, increasing salinity and ionic concentrations during dry periods and decreasing concentrations during wet periods (Verleyen et al. 2003). As a result, both water balance and resultant physico-chemical characteristics are sensitive to climatic variability, which subsequently impacts their aquatic ecosystems (Doran et al. 2008; Lyons et al. 2012; Hawes et al. 2014). The degree to which these changes occur both intra- and inter-annually is dependent on the size of the pond. Large and relatively deep (> 2 m) ponds with low conductivity tend to remain permanently ice-covered, except for a narrow moat melt area, even during the summer period. The relatively large volume of water, coupled with the permanent ice cover, is thought to act as a buffer against fluctuations in the physico-chemical environment (Hawes et al. 2014). In contrast, small, shallow ponds with a large surface area to volume ratio, are subject to freeze–thaw cycles and relatively higher water losses that leads to more extreme fluctuations in the physico-chemical environment (Sutherland 2009). These freeze–thaw cycles result in both chemical and thermal stratification of the water column, as brines, excluded during freezing, remain in the lower layer of the pond (Schmidt et al. 1991; Wait et al. 2006). The benthic microbial mats must, therefore, tolerate the chemical extremes imposed on them during winter and, for some, throughout long periods of stratification during summer (Wait et al. 2006).

It has been shown how flux in the water balance, as a result of climate change, can propagate to substantial changes in the habitat and biotic components of polar aquatic ecosystems (Hawes et al. 2014). The ponds on the MIS wetland would most likely be equally sensitive to such changes, but our ability to predict the effects of a changing climate are confounded by our poor understanding of the inherent variability of these ponds and their microbial communities in response to the physico-chemical environment. This study was conducted to better understand how microbial mat biomass and community composition is shaped across broad physico-chemical gradients and to identify key indicator species that may allow us to better predict the effects of climate change on the MIS wetlands.

Methods

Study site

The 24 meltwater ponds selected for this study (unofficial names listed in Table 1) were located on the MIS, approximately 200–400 m south of Bratina Island and were within a few hundred meters of each other but differed widely in water chemistry and microbial mat development and biomass (Fig. 1). The ponds were sampled during the austral summer of 2007–2008. Most of the 24 study ponds are known to have persisted, in some form, for at least 20 years, with 12 of the 24 ponds forming part of the long-term monitoring dataset of Hawes et al. (2014).

Table 1 List of environmental variables measured in each of the 24 study ponds
Full size table
Fig. 1
figure 1

Map showing location of the study ponds, Bratina Island, McMurdo Ice Shelf

Full size image

Overlying water-column environmental variables

Water column pH, temperature and conductivity were measured immediately above the surface of the microbial mats being sampled, using a portable meter (WP-90, TPS) calibrated in-field against standard solutions. Pond water overlying the microbial mats was collected into acid-washed Nalgene bottles and returned to the field laboratory for sub-sampling and filtering for subsequent analyses of nutrients, silica, alkalinity, dissolved organic carbon (DOC) and major ions. Samples were kept either cool or deep frozen, depending on their analytical requirements, and transported back to New Zealand for analyses. Nitrate (NO3–N), ammonium (NH4–N), dissolved reactive phosphorus (DRP), total dissolved nitrogen (TDN) and total dissolved phosphorus (TDP) were determined on a flow injection analyser (Lachat Quikchem FIA + 8000) after UV irradiation in alkaline or acidic conditions (Downes 2001). Major ions (Na, K, Ca, Mg, Cl, SO4), silica, alkalinity and DOC were analysed according to standardised methods described in APHA (1998).

Microbial mats

Quantitative samples of microbial mats were collected from the 24 ponds at a fixed water depth of 30 cm, using a piston corer with a 2.84 cm2 cross-sectional area. In each pond, duplicate cores were collected from five sites and the microbial mat was carefully separated from the underlying sediment layer, and either stored frozen or preserved in (5% v/v) glutaraldehyde, for determination of biomass and algal community composition, respectively.

Biomass

On return to New Zealand, the frozen microbial mats were freeze-dried and ground to a fine powder. Each mat was weighed and sub-sampled, to known quantities, for organic matter content, chlorophyll a (chl-a) and particulate nutrient analyses. For organic matter, the freeze-dried sub-samples were weighed (to give total solids) then combusted at 450 °C for 4 h, cooled in a desiccator, and then re-weighed to determine ash content. Organic matter was estimated as the difference between the total solids and the ash content. Chl-a samples were re-hydrated in 90% acetone and the pigments extracted at 4 °C, in the dark, for 24 h. Following extraction, samples were purified by centrifugation then the absorption was measured at 675 and 750 nm on a spectrophotometer (Shimadzu UV-2550) before and after acidification (Marker 1980). Chl-a concentration was estimated using calibration standards prepared with purified chl-a (Sigma chemicals). Particulate nitrogen (PN) and particulate phosphorus (PP) were determined on a flow injection analyser (Lachat Quikchem FIA + 8000) following digestion.

Community composition

The preserved mat samples were disrupted and dispersed to form a homogenous sample. Sub-samples were viewed under an inverted microscope at magnifications up to × 400 and on a compound microscope using bright field illumination, at up to × 1000 magnification. Cyanobacteria comprised the majority (> 70%) of the photoautotrophic community in the mats. However, despite mechanical disruption of the mats, many trichomes remained tangled, particularly the fine, narrow trichomes, which meant that counting individual trichomes was not feasible. Instead, a score of relative percentage biovolume, based on observations of 20 fields of view (following the methods of Sutherland 2009), was used for cyanobacteria estimates for each of the five replicates from each pond. Cyanobacteria taxonomy was based primarily on morphological characteristics as described in Komárek and Anagnostidis (2005), and more specialised literature including Broady and Kibblewhite (1991), Komárek (2007) and Komárek et al. (2008). The authors acknowledge the limitations of identifying cyanobacteria based on morphological characteristics alone, which was taken into consideration during interpretation of the results.

The percentage occurrence of each diatom species was evaluated by counting 200 frustules in each of the five replicates from each pond. To ensure that diatom frustules from previous growth seasons were not included, only those frustules that contained intact chloroplasts were counted. For taxonomic purposes, diatom sub-samples were cleaned of organic content by oxidation in H2SO4 and H2O2 (Barber and Haworth 1981), settled onto a cover glass and mounted in Naphrax (Northern Biological Supplies, UK). Cleaned frustules were then examined on a compound microscope at × 1000 magnification with differential interference contrast. Taxonomic identification was based on Krammer and Lange-Bertalot (1991, 1997a,2000a, b) and more specialised literature including Cremer et al. (2004), Esposito et al. (2008), Kellogg and Kellogg (2002), Sabbe et al. (2003), Spaulding et al. (1997, 1999), Kohler et al. (2015) and taxon names were aligned with those reported by Spaulding et al. (2016).

Statistical and numerical analyses

Microbial mat biomass, chl-a, organic matter and particulate nutrients were examined using linear and non-linear regression analysis and Analysis of Variance (ANOVA). ANOVA was followed by post hoc analysis (Tukey’s HSD test) to establish significant differences (STATISTICA v. 12).

Species diversity, defined as the number of species for a given number of individuals, was measured as the Shannon diversity index (H) while species evenness, defined as the extent to which community abundance is dominated by a small number of species, was expressed as Pielou’s eveness index (J). Species diversity and evenness were calculated using Primer 7.

All environmental data, except pH, alkalinity, silica and NH4–N, had skewed distributions and were log transformed for further numerical analyses. A Pearson correlation matrix with Bonferroni-adjusted probabilities was performed (SYSTAT v. 13) to test for correlations between environmental variables. Based on these results, the following variables were excluded from further analysis: Ca, Mg, Na, K, Cl, SO4, DOC (all significantly (at least p < 0.05) correlated with conductivity). Both the diatom and cyanobacteria community datasets were square-root transformed to down-weigh the effects of dominant taxa (> 20% in many samples) (Laing and Smol 2000).

For both cyanobacteria and diatom community composition, a cluster analysis with a posteriori test for significance was undertaken using the SIMPROF procedure in the University of Plymouth’s programme PRIMER (Clarke and Warwick 2001). SIMPROF was considered appropriate for initially exploring the dataset to identify significant changes in the communities as there was no prior structuring of samples into groups and individual abundance was expressed in relative terms as a percentage of the total abundance for all species in each sample (Clarke and Warwick 2001). Dendrogram matrices with posterior confirmation by SIMPROF together with Bray Curtis similarities were constructed to represent the statistically significant community patterns across the meltwater ponds.

LINKTREE procedures were used to explore whether structures in the diatom and cyanobacterial assemblages between ponds could be explained by environmental variables. The LINKTREE procedure uses a constrained clustering technique to construct a hierarchical tree to identify the best subset of environmental variable thresholds that explain significant patterns based on assemblage data (Clarke et al. 2008). Differences in environmental variables are compared to the Bray Curtis similarity matrix for the diatom and cyanobacterial assemblages. The samples are first treated as one group, and then subdivided according to the variables that maximised the between-cluster variance. The degree of separation is indicated by A%, the higher the A%, the higher the variance between the separations and significant (5% level) thresholds of environmental variables are identified (Clarke et al. 2008).

Indicator value index (\({\text{IndVal}}_{ij}\)), defined as the degree to which a species is an indicator for a given cluster of ponds, was determined following the procedure of Dufrêne and Legendre (1997). For each species in a given cluster, the \({\text{IndVal}}_{ij}\) was computed from the two value specificity (\({A}_{ij})\) and fidelity (\({B}_{ij}\)). \({A}_{ij}\) is the mean abundance of species (\(i\)) in the sites of each group (\(j\)) compared to all groups in the study, whereas \({B}_{ij}\) is the relative frequency of occurrence of species (\(i)\) in the sites of each group \((j\)). \({\text{IndVal}}_{ij}\) is the indicator value of species (\(i)\) in each group (\(j)\) (Dufrêne and Legendre 1997). \({A}_{ij}, {B}_{ij}\) and \({\text{IndVal}}_{ij}\) were calculated as follows:

$$A_{ij} = \frac{{N_{{{\text{individuals}}_{ij} }} }}{{N_{{{\text{individuals}}_{i} }} }}$$
$$B_{ij} = \frac{{N_{{{\text{sites}}_{ij} }} }}{{N_{{{\text{sites}}_{j} }} }}$$
$${\text{IndVal}}_{ij} = A_{ij} \times B_{ij} \times 100$$

\({\text{IndVal}}_{ij}\) uses the mean abundance of a species in each cluster as this removes any effect of the number of ponds in various clusters, as well as the variance between ponds within a cluster (Dufrêne and Legendre 1997).

For individual species, LINKTREE procedure was used to explore whether the exclusion of an individual species from a pond could be explained by measured environmental variables. SIMPROF, Dendrogram plots, Bray Curtis similarity matrices and LINKTREE analyses were carried out in PRIMER, v. 7.

Results

Pond environmental variables

The water column environmental variables varied widely across ponds, with the most notable being conductivity (0.1–66 mS cm−1), TDN (18–18,430 mg m−3) and TDP (36–892 mg m−3, Table 1). Ponds with the lowest conductivity were also low in alkalinity, silica, and dissolved nitrogen, but had the highest pH. As conductivity increased, alkalinity, DOC, TDN and major ions typically increased to varying degrees (Table 1).

Microbial mat biomass and nutrients

Microbial mat chl-a concentrations ranged across an order of magnitude from 3.32 to 38.77 μg cm−2 across the 24 ponds, while total organic matter ranged from 2.63 to 39.68 mg cm−2 (Table 2). Chl-a was significantly higher (p < 0.001) in ponds Huey, Jax and K081, while total organic matter was significantly higher (p < 0.01) in ponds Huey, Jax, K081 and Maslin compared to all other ponds. There was a strong positive correlation between chl-a and organic matter (R2 = 0.8952; Fig. 2a) and between chl-a and microbial mat particulate phosphorus (R2 = 0.6433) and particulate nitrogen (R2 = 0.7497; Fig. 2b, c). For most ponds, the microbial mat N:P ratio was constrained between 10:1 and 15:1, while the N:P ratio of the overlying water was more variable, ranging from 3:1 to 30:1. The N:P ratio of the microbial mats bore no relationship to the N:P ratio of the overlying water (Fig. 2d). Both the mat biomass (chl-a and organic matter) and particulate nutrients (PP and PN) were not correlated with any other measured environmental variable.

Table 2 Microbial mat organic matter and chlorophyll a (Chl-a) for the 24 ponds
Full size table
Fig. 2
figure 2

Relationship between a organic matter and chl-a; b microbial mat particulate phosphorus and chl-a; c microbial mat particulate nitrogen and chl-a and; d nitrogen and phosphorus ratio of the microbial mats and the overlying pond water in the 24 study ponds

Full size image

Species diversity and evenness

For all 24 ponds, the microbial mats were dominated by filamentous cyanobacteria, with representatives from the orders Oscillatoriales and Nostocales comprising at least 80% of the total biovolume, while diatoms comprised of the remaining ~ 20%. Total species (S) ranged from 8 to 28, species diversity (H) ranged from 1.23 to 4.73 and species evenness (Jʹ) ranged from 0.546 to 0.858 (Table 3). There were no significant relationships between the total species, richness or evenness with any of the measured environmental variables.

Table 3 Total species, richness and evenness of the cyanobacteria and diatoms communities in 24 ponds
Full size table

Diatom community composition

A total of 31 diatom species were identified across all ponds (Table 4). Achnanthes taylorensis and Luticola muticopsis fo. reducta were the most frequently occurring species, recorded in 22 ponds. Nitzschia westiorum was recorded in 21 ponds, while Muelleria peraustralis and Fistulifera pelliculosa were both recorded in 20 ponds. Four diatom species (Luticola laeta, Luticola mutica, Nitzschia australocommutata and Pseudostaurosia brevistriata) were each recorded in only one pond and three species (Chamaepinnularia cymatopleura, Fistulifera pelliculosa and Navicula gregaria) were each recorded in only two ponds (Table 4).

Table 4 List of diatom and cyanobacteria species identified from microbial mats in 24 ponds, Bratina Island
Full size table

The SIMPROF and LINKTREE analysis showed significant (p < 0.01) structure in the diatom communities, revealing five distinct clusters that separated out along a conductivity gradient (Figs. 3, 4). The first split of the LINKTREE (D1) separated Salt pond from all other ponds due to significantly higher (p < 0.01) conductivity, TDN, TDP and NH4–H concentrations (Fig. 4, Table 1). The second significant (p < 0.01) split (D2) separated ponds with conductivity > 5 mS cm−1 (P70e, Brack, Huey, Upper and VXE6) from all other ponds (Fig. 4). The third LINKTREE split (D3) clustered ponds Isabella, Wednesday, Duet, Permanent Ice and Brionna together due to significantly lower (p < 0.01) conductivity and TDN compared to all other remaining ponds (Fig. 4). The final split separated the remaining ponds based on significant differences (p < 0.01) in conductivity with one cluster of ponds (D4; Foam, Dosli, Legin, Orange, Bud, Jax) within a conductivity range of 2–5 mS cm−1, while the other cluster of ponds (D5; Fresh, KO81, Vincent, Maslin, Casten, Fogghorne) were within a conductivity range of 0.4–1.5 mS cm−1 (Fig. 4, Table 1).

Fig. 3
figure 3

Dendrogram and Bray Curtis similarity matrices identifying significant changes in diatom community composition across the 24 ponds. Full names for diatom codes can be found in Table 4

Full size image
Fig. 4
figure 4

LINKTREE analysis of Bratina Island pond communities showing divisive clustering of ponds from diatom species composition, constrained by environmental variables. Dashed line (red) indicates significant change in environmental variables. Significant clusters are indicated by the surrounding solid blue line

Full size image

Indicator values calculated for the LINKTREE clusters identified two diatoms as indicator species for each of the cluster D1, D3, D4 and D5, and only one diatom as an indicator species for cluster D2 (Table 5). Of the nine indicator species only one diatom was found exclusively in one cluster with Pseudostaurosira brevistriata occurring only in Salt Pond (Tables 3, 5).

Table 5 Diatom and cyanobacteria indicator species for the clusters identified from LINKTREE analysis
Full size table

Cyanobacteria community composition

A total of 18 cyanobacteria morphotypes were identified across all ponds (Table 4). Nodularia harveyana and Phormidium autumnale were the most frequently occurring species, each recorded in 19 ponds (Table 4). Leptolyngbya fragilis, Leptolyngbya fritschiana and Oscillatoria sp. were the least frequently occurring species, recorded in two, three and one ponds, respectively (Table 4).

The SIMPROF and LINKTREE analysis showed significant (p < 0.01) structure in the cyanobacterial communities, revealing three distinct clusters (Figs. 5, 6). The first split of the LINKTREE (C1) separated Salt Pond from all other ponds due to significantly higher (p < 0.01) conductivity, TDN, TDP and NH4–H concentrations (Fig. 6, Table 1). The second significant (p < 0.01) split separated all remaining ponds into two distinct clusters that were defined as those ponds with NH4–N concentrations < 5 mg m−3 and those ponds with NH4–N concentrations > 5 mg m−3 (C2 and C3, respectively, Figs. 5, 6). Within these two significantly distinct clusters, the ponds were further separated by the combination of conductivity, TDN and pH for group C4 and by pH for group C5, but these cluster separations were not found to be statistically significant (Fig. 6).

Fig. 5
figure 5

Dendrogram and Bray Curtis similarity matrices identifying significant changes in cyanobacteria community composition across the 24 ponds. Full names for cyanobacteria codes can be found in Table 4

Full size image
Fig. 6
figure 6

LINKTREE analysis of Bratina Island pond communities showing divisive clustering of ponds from cyanobacteria species composition, constrained by environmental variables. Dashed line (red) indicates significant change in environmental variables. Significant clusters are enclosed within solid blue lines, non-significant but strong indicator species clusters are enclosed indicated by within dotted lines

Full size image

Indicator values calculated for the LINKTREE clusters identified three cyanobacteria morphotypes as indicators, one each for clusters C1, C4 and C5 (Table 5). Of the three indicator morphotypes, Leptolynbya erebi was found exclusively in all ponds associated with C4. These ponds were low in conductivity, dissolved inorganic nitrogen, DOC, TDN and high in pH (Table 1). The morphotype-type Oscillatoria sp. was found exclusively in Salt Pond, the only pond within cluster C1. There were no indicator morphotypes for those ponds within clusters C2 or C3 (Fig. 6).

Discussion

Microbial mats

Differences in microbial mat biomass across the 24 ponds could neither be explained by any measured environmental variable nor by combination of these variables. Ponds that were permanently ice covered (i.e. those with conductivity ≤ 0.2 mS cm−1) typically had the lowest organic matter and chl-a, but this was not always the case. Microbial mats in meltwater ponds and lakes in Antarctica have been shown to be perennial and comprising of accumulated net growth over many years (Hawes et al. 1993, 2013; Quesada et al. 2008). The accumulated mat biomass most likely reflects integration with long-term environmental variables, rather than the current environment (Hawes et al. 2014). In a regression tree analysis of long-term (> 20 years) benthic chl-a of 12 ponds at Bratina Island, including eight from the current study, Hawes et al. (2014) found that warm winter temperatures and nitrogen concentrations explained 60% variance in biomass, but they were not able to interpret these relationships further.

Relationships between organic matter and chl-a concentrations have often been used to indicate either photoautotrophic or heterotrophic dominance within microbial mats and other periphyton communities (Dodds 2003). As organic matter increased so did chl-a biomass, suggesting that, as the microbial mat biomass increased, the proportion of photoautotrophs to heterotrophs remained constant. However, differences in the chl-a concentrations per individual cell between species, or in the same species in different ponds, or in different positions within the mat itself (as a result of differences in light penetration or shading within and between mats) could mask changes in the proportion of photoautotrophs to heterotrophs across the pond gradient in this study. Chl-a within individual cells has been shown to increase or decrease as a result in changes in environmental variables such as available light (Boston and Hill 1991; Greenwood and Rosemond 2005), dissolved inorganic carbon and pH (Sobrino et al. 2008; Sutherland et al. 2015), and major nutrients (Bachmann et al. 2002). Due to the complexity of the cyanobacterial matrix, it was not possible to accurately determine number of cells per unit volume of microbial mat and further refinement of methodologies would be needed to address this challenge.

The ratio of particulate N to particulate P (N:P) in the microbial mats was low (< 15) across the pond gradient. In benthic microalgal communities, a N:P ratio < 15 is often regarded as being indicative of nitrogen limitation (Hillebrand and Sommer 1999; Dodds 2003). N limitation can favour cyanobacterial morphotypes that are capable of nitrogen-fixation, such as Nodularia harveyana and Nostoc cf. commune that were present in many of the ponds in this study. However, mat N:P ratios in the present study may also have been artificially lowered due to either P precipitation onto the mat or through luxury uptake and storage of P by the cells (Portielje and Lijklema 1994; Noe et al. 2003). The mat particulate matter is comprised of cyanobacterial, diatom and bacterial biomass, detrital matter and precipitated mineral crystals (Sirova et al. 2006). The measured microbial mat P concentration would have included any precipitated or adsorbed P. In a study on periphyton mats from the Everglades, a substantial proportion of particulate P was Ca-bound, which was eventually incorporated into the organic matter of the mat (Noe et al. 2003; Scinto and Reddy 2003). Precipitation of P salts or P adsorption onto CaCO3 crystals can occur under conditions of elevated pH (> 8.9), high-dissolved mineral content and low CO2 partial pressure (Craggs et al. 1996, Rejmankova and Komarkova 2005, Borovec et al. 2010). In the present study, 71% of ponds had pH > 8.9 and elevated CaCO3 concentrations. We have previously observed 4–11% by dry weight of CaCO3 crystals in microbial mats from lakes and ponds in Antarctica (data not shown). While the CO2 partial pressure was not measured during this study, high pH in Antarctic ponds is thought to be related to photosynthesis-mediated dissolved inorganic carbon depletion of the water column (Webster-Brown et al. 2010; Hawes et al. 2014), suggesting carbon limitation.

Pond clustering

Distinct clustering of pond types, against measured environmental variables, for both the diatom and cyanobacterial communities, as well as the low number of indicator species, suggests that changes in community composition was driven more at the functional group level rather than at the species level. This suggests high functional redundancy (Hubbell 2005). Functional redundancy is the ecological theory that the environment selects for life forms of species that share similar roles in ecosystem functionality, rather than specific taxa (Naeem 2006). In the case of our study, this suggests that the diatoms and cyanobacteria that comprise the significantly distinct communities in the pond clusters have overlapping traits, or functional similarity within each pond cluster. However, further functional analyses based on biogeochemical and ecological attributes would be required to support this claim. At the time of publishing, there is insufficient data on these attributes for a large number of species in this study.

Diatom communities

Conductivity gradients were correlated with the separation of five significantly distinct diatom community types across the pond gradient. Variability in conductivity is not mediated by biological processes but by changes in the net water balance (water input from melting snow and ice less the water loss from ablation and evaporation) of each pond. In some ponds, conductivity can also be influenced by wind-induced mixing of deeper, more saline waters which had formed as a consequence of freeze-concentration during winter (Wait et al. 2006; Hawes et al. 2011). As a result of seasonal changes in pond water balance and wind-mixing of saline waters, conductivity can vary by up to 15 times (Hawes et al. 2014).

Conductivity as a main environmental driver of diatom community composition is consistent with other floristic studies of lakes and ponds throughout the Antarctic region. In 56 lakes in the Larsemann Hills and Bølingen Islands (East Antarctica) the diatom community composition was strongly related to both conductivity and lake depth (a proxy to light climate; Sabbe et al. 2003). At Macquarie Island (sub-Antarctica), conductivity, pH, DRP, silicate and temperature were the main environmental variables explaining variance in diatom communities across 50 coastal and inland lakes (Saunders et al. 2009). Conductivity was also associated with diatom community composition in ponds and lakes at Vestfold Hills (Roberts and McMinn 1996), Windmill Islands (Roberts et al. 2001), Bunger Hills (Gibson et al. 2006) and Skarvsnes lakes (Ohtsuka et al. 2006).

In the present study, while some pond clusters exhibited a narrow conductivity range, e.g. cluster D3, others such as D2 and D4 had much broader ranges. Ponds within D3 were large and deep relative to the other ponds and, except for a narrow moat melt area, were permanently ice covered. Both a large volume of water and permanent ice cover is thought to act as a buffer to changes in conductivity (Hawes et al. 2014). For D2 and D4, broad conductivity ranges most likely reflect changes in conductivity that these ponds experience as a result of changes in water balance both within and between growth seasons. In small, shallow ponds, such as those in D2, volumetric changes due to evaporative losses over the growth season could result in substantial changes in pond water level and associated conductivity. Hawes et al. (1993) estimated a 15 cm decrease in water level for similar sized ponds on Bratina Island, while Sutherland (2009) estimated an average 132% increase in conductivity over the summer as a result of evaporative water loss. Similarly, inter-annual variation in conductivity within a pond reflects differences in the net water balance between years (Hawes et al. 2014).

In the present study, conductivity was also strongly related to several other environmental variables including Ca, Mg, Na, K, Cl, SO4 and DOC, while positive correlations between conductivity and dissolved organic nitrogen and dissolved organic phosphorus have also been reported in several of these ponds (Hawes et al. 2014). As is standard practice in statistical analysis, these variables were discarded from further analyses due to these strong correlations. Furthermore, while LINKTREE identifies the environmental variables that explain significant differences between assemblages, once a cluster is removed by this procedure, further interpretation of the importance of other variables in shaping that community may be lost (Calhoun et al. 2008). Therefore, it is difficult to determine if conductivity is truly the only main driver, if one or more of the correlated variables also drive diatom community change, or if conductivity is a surrogate for other environmental variables that were excluded from the analyses.

Cyanobacteria communities

While the present study identified three significantly distinct cyanobacterial community types, many of the cyanobacteria morphotypes were recorded across a wide ecological range. Cyanobacteria are regarded as being ecologically very plastic, meaning that they can adapt their physiological characteristics very rapidly to changing environmental conditions (Hagemann 2013). Antarctic pond microbial communities undergo markedly different physical and chemical conditions between summer and winter periods (Hawes et al. 2011). This is likely to have selected for cyanobacterial morphotypes that are well-adapted and tolerant of a wider spectrum of ecological conditions compared to temperate morphotypes (Komárek and Elster 2008). While it is plausible to suggest that the morphotypes identified in our ponds were adapted to a wide range of environmental conditions, and therefore, inhabited ponds across a broad gradient, it is equally plausible to suggest that different genotypes, with visually identical morphology, were present across the different pond clusters. Molecular phylogenetics, using 16s RNA, may help to clarify this, which would then enable better distinction of environmental drivers of cyanobacterial communities on the MIS.

Significant differences in NH4–N concentrations were correlated to the three distinct community clusters C1, C2 and C3. NH4–N availability has been shown to significantly influence the species structure of microalgal communities in other natural aquatic systems, e.g. dominance of cyanobacteria in Vancouver Lake (Canada), and in artificial high-rate algal ponds (Lee et al. 2015; Sutherland et al. 2017). NH4–N is a biologically mediated variable as its concentration is driven by biological processes. As there are no noteworthy penguin or other bird colonies at Bratina Island, guano is not considered to be a source of NH4–N input to the ponds and the microbial mats are most probably the main source of NH4–N. NH4–N is a commonly released metabolite (via ammonification) from the decomposition of organic matter in the microbial mats, especially under anaerobic conditions, and from bacterial N2 fixation (Paerl and Pinckney 1996). Reduction of nitrate by anaerobic heterotrophic bacteria is another potential source of NH4–N in the ponds. Nitrate accumulated in the soils and ice may leach out into the ponds, providing an additional input source (Michalski et al. 2005). NH4–N has long been considered to be the preferred form of nitrogen for algae and cyanobacteria, especially when nitrogen is limiting (Raven et al. 1992; Gilbert et al. 2016). This preference is considered to be due, at least in part, to the lower energy requirements for the cell, greater ease of transport across the cell membrane compared to NO3–N, and the repression effect of NH4–N on NO3–N uptake and assimilation (Gilbert et al. 2016).

Our data is inadequate for identification of the role that NH4–N has in structuring the three significantly distinct cyanobacterial community clusters. It is possible that while NH4–N concentrations were identified as the structuring environmental variable, it may only be serving as a proxy for other environmental processes within the mat that may be driving the cyanobacterial community.

Both the diatom and cyanobacteria communities each had a distinct cluster (D3 for diatoms and C4 for cyanobacteria) that were comprised of the same subset of ponds (Brionna, Duet, Isabella, Permanent Ice, Wednesday). These ponds were characterised by low conductivity, low nutrients (N, P and carbon) and high pH, and were permanently ice covered. The cyanobacterial community in these ponds was dominated by morphotypes of Leptolyngbya, which are comprised of very narrow (< 2 µm wide) filamentous species. These are often regarded as the building blocks of cyanobacterial mats, meaning that they form most of the biomass and structure of the mat (Rejmankova and Komarkova 2005; Komárek and Elster 2008), and are typically long-lived due to very low biochemical activity (Rejmankova and Komarkova 2005). Cluster C4 was separated from the other low NH4–N ponds in the LINKTREE analysis based on low conductivity, low total dissolved nitrogen and high pH, although this separation was not statistically significant (p = xxx). One morphotype, L. erebi, was dominant in all ponds within C4, but was absent outside this cluster. L. erebi has been previously recorded in small stagnant waters, pools, glacial pools and in cryoconite ponds on glaciers (Komárek 2007) but there is no published information on its growth rate or nutritional requirements.

The distribution of cyanobacterial morphotypes in ponds throughout Ross Island and southern Victoria Land has been thought to be driven by conductivity (e.g. Howard-Williams et al. 1990; Broady and Kibblewhite 1991). However, in the present study, with the exception of Salt Pond, and unlike the diatom community, conductivity did not appear to play a role. Similarly, no relationship was found between the distribution of cyanobacterial morphotypes and conductivity across 56 lakes in East Antarctica (Sabbe et al. 2003), while several phenotypes have been reported in two or more of Fresh, Orange and Salt ponds at Bratina Island (Jungblutt et al. 2005).

Conclusion

Our study has shown that diatom communities in meltwater ponds were more sensitive to change in environmental conditions than the cyanobacterial communities, despite the latter’s overall dominance across the studied ponds. Diatoms, are more likely to be sensitive to climate change and its impacts on Antarctic meltwater ponds than cyanobacteria. The recognition of distinct pond clusters based on diatom communities, and to a lesser extent on cyanobacteria, suggests that changes at a functional group level may be more important than changes at the level of individual species. Further understanding of diatom functional groups and the environmental parameters that drive them would provide us with the opportunity to hindcast past climates and water budgets within Antarctica using diatom frustules preserved in buried mats. However, the disconnect between microbial mat biomass and community composition currently prevents hindcasting past productivities in relation to environmental changes.